Loading...
1/*
2* Copyright (c) 2001 The Regents of the University of Michigan.
3* All rights reserved.
4*
5* Kendrick Smith <kmsmith@umich.edu>
6* Andy Adamson <kandros@umich.edu>
7*
8* Redistribution and use in source and binary forms, with or without
9* modification, are permitted provided that the following conditions
10* are met:
11*
12* 1. Redistributions of source code must retain the above copyright
13* notice, this list of conditions and the following disclaimer.
14* 2. Redistributions in binary form must reproduce the above copyright
15* notice, this list of conditions and the following disclaimer in the
16* documentation and/or other materials provided with the distribution.
17* 3. Neither the name of the University nor the names of its
18* contributors may be used to endorse or promote products derived
19* from this software without specific prior written permission.
20*
21* THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24* DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30* NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32*
33*/
34
35#include <linux/file.h>
36#include <linux/fs.h>
37#include <linux/slab.h>
38#include <linux/namei.h>
39#include <linux/swap.h>
40#include <linux/pagemap.h>
41#include <linux/ratelimit.h>
42#include <linux/sunrpc/svcauth_gss.h>
43#include <linux/sunrpc/addr.h>
44#include <linux/jhash.h>
45#include "xdr4.h"
46#include "xdr4cb.h"
47#include "vfs.h"
48#include "current_stateid.h"
49
50#include "netns.h"
51#include "pnfs.h"
52
53#define NFSDDBG_FACILITY NFSDDBG_PROC
54
55#define all_ones {{~0,~0},~0}
56static const stateid_t one_stateid = {
57 .si_generation = ~0,
58 .si_opaque = all_ones,
59};
60static const stateid_t zero_stateid = {
61 /* all fields zero */
62};
63static const stateid_t currentstateid = {
64 .si_generation = 1,
65};
66
67static u64 current_sessionid = 1;
68
69#define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
70#define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
71#define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
72
73/* forward declarations */
74static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
75static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
76
77/* Locking: */
78
79/*
80 * Currently used for the del_recall_lru and file hash table. In an
81 * effort to decrease the scope of the client_mutex, this spinlock may
82 * eventually cover more:
83 */
84static DEFINE_SPINLOCK(state_lock);
85
86/*
87 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
88 * the refcount on the open stateid to drop.
89 */
90static DECLARE_WAIT_QUEUE_HEAD(close_wq);
91
92static struct kmem_cache *openowner_slab;
93static struct kmem_cache *lockowner_slab;
94static struct kmem_cache *file_slab;
95static struct kmem_cache *stateid_slab;
96static struct kmem_cache *deleg_slab;
97static struct kmem_cache *odstate_slab;
98
99static void free_session(struct nfsd4_session *);
100
101static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
102
103static bool is_session_dead(struct nfsd4_session *ses)
104{
105 return ses->se_flags & NFS4_SESSION_DEAD;
106}
107
108static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
109{
110 if (atomic_read(&ses->se_ref) > ref_held_by_me)
111 return nfserr_jukebox;
112 ses->se_flags |= NFS4_SESSION_DEAD;
113 return nfs_ok;
114}
115
116static bool is_client_expired(struct nfs4_client *clp)
117{
118 return clp->cl_time == 0;
119}
120
121static __be32 get_client_locked(struct nfs4_client *clp)
122{
123 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
124
125 lockdep_assert_held(&nn->client_lock);
126
127 if (is_client_expired(clp))
128 return nfserr_expired;
129 atomic_inc(&clp->cl_refcount);
130 return nfs_ok;
131}
132
133/* must be called under the client_lock */
134static inline void
135renew_client_locked(struct nfs4_client *clp)
136{
137 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
138
139 if (is_client_expired(clp)) {
140 WARN_ON(1);
141 printk("%s: client (clientid %08x/%08x) already expired\n",
142 __func__,
143 clp->cl_clientid.cl_boot,
144 clp->cl_clientid.cl_id);
145 return;
146 }
147
148 dprintk("renewing client (clientid %08x/%08x)\n",
149 clp->cl_clientid.cl_boot,
150 clp->cl_clientid.cl_id);
151 list_move_tail(&clp->cl_lru, &nn->client_lru);
152 clp->cl_time = get_seconds();
153}
154
155static void put_client_renew_locked(struct nfs4_client *clp)
156{
157 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
158
159 lockdep_assert_held(&nn->client_lock);
160
161 if (!atomic_dec_and_test(&clp->cl_refcount))
162 return;
163 if (!is_client_expired(clp))
164 renew_client_locked(clp);
165}
166
167static void put_client_renew(struct nfs4_client *clp)
168{
169 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
170
171 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
172 return;
173 if (!is_client_expired(clp))
174 renew_client_locked(clp);
175 spin_unlock(&nn->client_lock);
176}
177
178static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
179{
180 __be32 status;
181
182 if (is_session_dead(ses))
183 return nfserr_badsession;
184 status = get_client_locked(ses->se_client);
185 if (status)
186 return status;
187 atomic_inc(&ses->se_ref);
188 return nfs_ok;
189}
190
191static void nfsd4_put_session_locked(struct nfsd4_session *ses)
192{
193 struct nfs4_client *clp = ses->se_client;
194 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
195
196 lockdep_assert_held(&nn->client_lock);
197
198 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
199 free_session(ses);
200 put_client_renew_locked(clp);
201}
202
203static void nfsd4_put_session(struct nfsd4_session *ses)
204{
205 struct nfs4_client *clp = ses->se_client;
206 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
207
208 spin_lock(&nn->client_lock);
209 nfsd4_put_session_locked(ses);
210 spin_unlock(&nn->client_lock);
211}
212
213static inline struct nfs4_stateowner *
214nfs4_get_stateowner(struct nfs4_stateowner *sop)
215{
216 atomic_inc(&sop->so_count);
217 return sop;
218}
219
220static int
221same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
222{
223 return (sop->so_owner.len == owner->len) &&
224 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
225}
226
227static struct nfs4_openowner *
228find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
229 struct nfs4_client *clp)
230{
231 struct nfs4_stateowner *so;
232
233 lockdep_assert_held(&clp->cl_lock);
234
235 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
236 so_strhash) {
237 if (!so->so_is_open_owner)
238 continue;
239 if (same_owner_str(so, &open->op_owner))
240 return openowner(nfs4_get_stateowner(so));
241 }
242 return NULL;
243}
244
245static struct nfs4_openowner *
246find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
247 struct nfs4_client *clp)
248{
249 struct nfs4_openowner *oo;
250
251 spin_lock(&clp->cl_lock);
252 oo = find_openstateowner_str_locked(hashval, open, clp);
253 spin_unlock(&clp->cl_lock);
254 return oo;
255}
256
257static inline u32
258opaque_hashval(const void *ptr, int nbytes)
259{
260 unsigned char *cptr = (unsigned char *) ptr;
261
262 u32 x = 0;
263 while (nbytes--) {
264 x *= 37;
265 x += *cptr++;
266 }
267 return x;
268}
269
270static void nfsd4_free_file_rcu(struct rcu_head *rcu)
271{
272 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
273
274 kmem_cache_free(file_slab, fp);
275}
276
277void
278put_nfs4_file(struct nfs4_file *fi)
279{
280 might_lock(&state_lock);
281
282 if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
283 hlist_del_rcu(&fi->fi_hash);
284 spin_unlock(&state_lock);
285 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
286 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
287 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
288 }
289}
290
291static struct file *
292__nfs4_get_fd(struct nfs4_file *f, int oflag)
293{
294 if (f->fi_fds[oflag])
295 return get_file(f->fi_fds[oflag]);
296 return NULL;
297}
298
299static struct file *
300find_writeable_file_locked(struct nfs4_file *f)
301{
302 struct file *ret;
303
304 lockdep_assert_held(&f->fi_lock);
305
306 ret = __nfs4_get_fd(f, O_WRONLY);
307 if (!ret)
308 ret = __nfs4_get_fd(f, O_RDWR);
309 return ret;
310}
311
312static struct file *
313find_writeable_file(struct nfs4_file *f)
314{
315 struct file *ret;
316
317 spin_lock(&f->fi_lock);
318 ret = find_writeable_file_locked(f);
319 spin_unlock(&f->fi_lock);
320
321 return ret;
322}
323
324static struct file *find_readable_file_locked(struct nfs4_file *f)
325{
326 struct file *ret;
327
328 lockdep_assert_held(&f->fi_lock);
329
330 ret = __nfs4_get_fd(f, O_RDONLY);
331 if (!ret)
332 ret = __nfs4_get_fd(f, O_RDWR);
333 return ret;
334}
335
336static struct file *
337find_readable_file(struct nfs4_file *f)
338{
339 struct file *ret;
340
341 spin_lock(&f->fi_lock);
342 ret = find_readable_file_locked(f);
343 spin_unlock(&f->fi_lock);
344
345 return ret;
346}
347
348struct file *
349find_any_file(struct nfs4_file *f)
350{
351 struct file *ret;
352
353 spin_lock(&f->fi_lock);
354 ret = __nfs4_get_fd(f, O_RDWR);
355 if (!ret) {
356 ret = __nfs4_get_fd(f, O_WRONLY);
357 if (!ret)
358 ret = __nfs4_get_fd(f, O_RDONLY);
359 }
360 spin_unlock(&f->fi_lock);
361 return ret;
362}
363
364static atomic_long_t num_delegations;
365unsigned long max_delegations;
366
367/*
368 * Open owner state (share locks)
369 */
370
371/* hash tables for lock and open owners */
372#define OWNER_HASH_BITS 8
373#define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
374#define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
375
376static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
377{
378 unsigned int ret;
379
380 ret = opaque_hashval(ownername->data, ownername->len);
381 return ret & OWNER_HASH_MASK;
382}
383
384/* hash table for nfs4_file */
385#define FILE_HASH_BITS 8
386#define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
387
388static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
389{
390 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
391}
392
393static unsigned int file_hashval(struct knfsd_fh *fh)
394{
395 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
396}
397
398static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
399
400static void
401__nfs4_file_get_access(struct nfs4_file *fp, u32 access)
402{
403 lockdep_assert_held(&fp->fi_lock);
404
405 if (access & NFS4_SHARE_ACCESS_WRITE)
406 atomic_inc(&fp->fi_access[O_WRONLY]);
407 if (access & NFS4_SHARE_ACCESS_READ)
408 atomic_inc(&fp->fi_access[O_RDONLY]);
409}
410
411static __be32
412nfs4_file_get_access(struct nfs4_file *fp, u32 access)
413{
414 lockdep_assert_held(&fp->fi_lock);
415
416 /* Does this access mode make sense? */
417 if (access & ~NFS4_SHARE_ACCESS_BOTH)
418 return nfserr_inval;
419
420 /* Does it conflict with a deny mode already set? */
421 if ((access & fp->fi_share_deny) != 0)
422 return nfserr_share_denied;
423
424 __nfs4_file_get_access(fp, access);
425 return nfs_ok;
426}
427
428static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
429{
430 /* Common case is that there is no deny mode. */
431 if (deny) {
432 /* Does this deny mode make sense? */
433 if (deny & ~NFS4_SHARE_DENY_BOTH)
434 return nfserr_inval;
435
436 if ((deny & NFS4_SHARE_DENY_READ) &&
437 atomic_read(&fp->fi_access[O_RDONLY]))
438 return nfserr_share_denied;
439
440 if ((deny & NFS4_SHARE_DENY_WRITE) &&
441 atomic_read(&fp->fi_access[O_WRONLY]))
442 return nfserr_share_denied;
443 }
444 return nfs_ok;
445}
446
447static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
448{
449 might_lock(&fp->fi_lock);
450
451 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
452 struct file *f1 = NULL;
453 struct file *f2 = NULL;
454
455 swap(f1, fp->fi_fds[oflag]);
456 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
457 swap(f2, fp->fi_fds[O_RDWR]);
458 spin_unlock(&fp->fi_lock);
459 if (f1)
460 fput(f1);
461 if (f2)
462 fput(f2);
463 }
464}
465
466static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
467{
468 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
469
470 if (access & NFS4_SHARE_ACCESS_WRITE)
471 __nfs4_file_put_access(fp, O_WRONLY);
472 if (access & NFS4_SHARE_ACCESS_READ)
473 __nfs4_file_put_access(fp, O_RDONLY);
474}
475
476/*
477 * Allocate a new open/delegation state counter. This is needed for
478 * pNFS for proper return on close semantics.
479 *
480 * Note that we only allocate it for pNFS-enabled exports, otherwise
481 * all pointers to struct nfs4_clnt_odstate are always NULL.
482 */
483static struct nfs4_clnt_odstate *
484alloc_clnt_odstate(struct nfs4_client *clp)
485{
486 struct nfs4_clnt_odstate *co;
487
488 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
489 if (co) {
490 co->co_client = clp;
491 atomic_set(&co->co_odcount, 1);
492 }
493 return co;
494}
495
496static void
497hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
498{
499 struct nfs4_file *fp = co->co_file;
500
501 lockdep_assert_held(&fp->fi_lock);
502 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
503}
504
505static inline void
506get_clnt_odstate(struct nfs4_clnt_odstate *co)
507{
508 if (co)
509 atomic_inc(&co->co_odcount);
510}
511
512static void
513put_clnt_odstate(struct nfs4_clnt_odstate *co)
514{
515 struct nfs4_file *fp;
516
517 if (!co)
518 return;
519
520 fp = co->co_file;
521 if (atomic_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
522 list_del(&co->co_perfile);
523 spin_unlock(&fp->fi_lock);
524
525 nfsd4_return_all_file_layouts(co->co_client, fp);
526 kmem_cache_free(odstate_slab, co);
527 }
528}
529
530static struct nfs4_clnt_odstate *
531find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
532{
533 struct nfs4_clnt_odstate *co;
534 struct nfs4_client *cl;
535
536 if (!new)
537 return NULL;
538
539 cl = new->co_client;
540
541 spin_lock(&fp->fi_lock);
542 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
543 if (co->co_client == cl) {
544 get_clnt_odstate(co);
545 goto out;
546 }
547 }
548 co = new;
549 co->co_file = fp;
550 hash_clnt_odstate_locked(new);
551out:
552 spin_unlock(&fp->fi_lock);
553 return co;
554}
555
556struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
557 struct kmem_cache *slab)
558{
559 struct nfs4_stid *stid;
560 int new_id;
561
562 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
563 if (!stid)
564 return NULL;
565
566 idr_preload(GFP_KERNEL);
567 spin_lock(&cl->cl_lock);
568 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
569 spin_unlock(&cl->cl_lock);
570 idr_preload_end();
571 if (new_id < 0)
572 goto out_free;
573 stid->sc_client = cl;
574 stid->sc_stateid.si_opaque.so_id = new_id;
575 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
576 /* Will be incremented before return to client: */
577 atomic_set(&stid->sc_count, 1);
578 spin_lock_init(&stid->sc_lock);
579
580 /*
581 * It shouldn't be a problem to reuse an opaque stateid value.
582 * I don't think it is for 4.1. But with 4.0 I worry that, for
583 * example, a stray write retransmission could be accepted by
584 * the server when it should have been rejected. Therefore,
585 * adopt a trick from the sctp code to attempt to maximize the
586 * amount of time until an id is reused, by ensuring they always
587 * "increase" (mod INT_MAX):
588 */
589 return stid;
590out_free:
591 kmem_cache_free(slab, stid);
592 return NULL;
593}
594
595static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
596{
597 struct nfs4_stid *stid;
598 struct nfs4_ol_stateid *stp;
599
600 stid = nfs4_alloc_stid(clp, stateid_slab);
601 if (!stid)
602 return NULL;
603
604 stp = openlockstateid(stid);
605 stp->st_stid.sc_free = nfs4_free_ol_stateid;
606 return stp;
607}
608
609static void nfs4_free_deleg(struct nfs4_stid *stid)
610{
611 kmem_cache_free(deleg_slab, stid);
612 atomic_long_dec(&num_delegations);
613}
614
615/*
616 * When we recall a delegation, we should be careful not to hand it
617 * out again straight away.
618 * To ensure this we keep a pair of bloom filters ('new' and 'old')
619 * in which the filehandles of recalled delegations are "stored".
620 * If a filehandle appear in either filter, a delegation is blocked.
621 * When a delegation is recalled, the filehandle is stored in the "new"
622 * filter.
623 * Every 30 seconds we swap the filters and clear the "new" one,
624 * unless both are empty of course.
625 *
626 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
627 * low 3 bytes as hash-table indices.
628 *
629 * 'blocked_delegations_lock', which is always taken in block_delegations(),
630 * is used to manage concurrent access. Testing does not need the lock
631 * except when swapping the two filters.
632 */
633static DEFINE_SPINLOCK(blocked_delegations_lock);
634static struct bloom_pair {
635 int entries, old_entries;
636 time_t swap_time;
637 int new; /* index into 'set' */
638 DECLARE_BITMAP(set[2], 256);
639} blocked_delegations;
640
641static int delegation_blocked(struct knfsd_fh *fh)
642{
643 u32 hash;
644 struct bloom_pair *bd = &blocked_delegations;
645
646 if (bd->entries == 0)
647 return 0;
648 if (seconds_since_boot() - bd->swap_time > 30) {
649 spin_lock(&blocked_delegations_lock);
650 if (seconds_since_boot() - bd->swap_time > 30) {
651 bd->entries -= bd->old_entries;
652 bd->old_entries = bd->entries;
653 memset(bd->set[bd->new], 0,
654 sizeof(bd->set[0]));
655 bd->new = 1-bd->new;
656 bd->swap_time = seconds_since_boot();
657 }
658 spin_unlock(&blocked_delegations_lock);
659 }
660 hash = jhash(&fh->fh_base, fh->fh_size, 0);
661 if (test_bit(hash&255, bd->set[0]) &&
662 test_bit((hash>>8)&255, bd->set[0]) &&
663 test_bit((hash>>16)&255, bd->set[0]))
664 return 1;
665
666 if (test_bit(hash&255, bd->set[1]) &&
667 test_bit((hash>>8)&255, bd->set[1]) &&
668 test_bit((hash>>16)&255, bd->set[1]))
669 return 1;
670
671 return 0;
672}
673
674static void block_delegations(struct knfsd_fh *fh)
675{
676 u32 hash;
677 struct bloom_pair *bd = &blocked_delegations;
678
679 hash = jhash(&fh->fh_base, fh->fh_size, 0);
680
681 spin_lock(&blocked_delegations_lock);
682 __set_bit(hash&255, bd->set[bd->new]);
683 __set_bit((hash>>8)&255, bd->set[bd->new]);
684 __set_bit((hash>>16)&255, bd->set[bd->new]);
685 if (bd->entries == 0)
686 bd->swap_time = seconds_since_boot();
687 bd->entries += 1;
688 spin_unlock(&blocked_delegations_lock);
689}
690
691static struct nfs4_delegation *
692alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
693 struct nfs4_clnt_odstate *odstate)
694{
695 struct nfs4_delegation *dp;
696 long n;
697
698 dprintk("NFSD alloc_init_deleg\n");
699 n = atomic_long_inc_return(&num_delegations);
700 if (n < 0 || n > max_delegations)
701 goto out_dec;
702 if (delegation_blocked(¤t_fh->fh_handle))
703 goto out_dec;
704 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
705 if (dp == NULL)
706 goto out_dec;
707
708 dp->dl_stid.sc_free = nfs4_free_deleg;
709 /*
710 * delegation seqid's are never incremented. The 4.1 special
711 * meaning of seqid 0 isn't meaningful, really, but let's avoid
712 * 0 anyway just for consistency and use 1:
713 */
714 dp->dl_stid.sc_stateid.si_generation = 1;
715 INIT_LIST_HEAD(&dp->dl_perfile);
716 INIT_LIST_HEAD(&dp->dl_perclnt);
717 INIT_LIST_HEAD(&dp->dl_recall_lru);
718 dp->dl_clnt_odstate = odstate;
719 get_clnt_odstate(odstate);
720 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
721 dp->dl_retries = 1;
722 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
723 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
724 return dp;
725out_dec:
726 atomic_long_dec(&num_delegations);
727 return NULL;
728}
729
730void
731nfs4_put_stid(struct nfs4_stid *s)
732{
733 struct nfs4_file *fp = s->sc_file;
734 struct nfs4_client *clp = s->sc_client;
735
736 might_lock(&clp->cl_lock);
737
738 if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
739 wake_up_all(&close_wq);
740 return;
741 }
742 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
743 spin_unlock(&clp->cl_lock);
744 s->sc_free(s);
745 if (fp)
746 put_nfs4_file(fp);
747}
748
749void
750nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
751{
752 stateid_t *src = &stid->sc_stateid;
753
754 spin_lock(&stid->sc_lock);
755 if (unlikely(++src->si_generation == 0))
756 src->si_generation = 1;
757 memcpy(dst, src, sizeof(*dst));
758 spin_unlock(&stid->sc_lock);
759}
760
761static void nfs4_put_deleg_lease(struct nfs4_file *fp)
762{
763 struct file *filp = NULL;
764
765 spin_lock(&fp->fi_lock);
766 if (fp->fi_deleg_file && --fp->fi_delegees == 0)
767 swap(filp, fp->fi_deleg_file);
768 spin_unlock(&fp->fi_lock);
769
770 if (filp) {
771 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
772 fput(filp);
773 }
774}
775
776void nfs4_unhash_stid(struct nfs4_stid *s)
777{
778 s->sc_type = 0;
779}
780
781/**
782 * nfs4_get_existing_delegation - Discover if this delegation already exists
783 * @clp: a pointer to the nfs4_client we're granting a delegation to
784 * @fp: a pointer to the nfs4_file we're granting a delegation on
785 *
786 * Return:
787 * On success: NULL if an existing delegation was not found.
788 *
789 * On error: -EAGAIN if one was previously granted to this nfs4_client
790 * for this nfs4_file.
791 *
792 */
793
794static int
795nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
796{
797 struct nfs4_delegation *searchdp = NULL;
798 struct nfs4_client *searchclp = NULL;
799
800 lockdep_assert_held(&state_lock);
801 lockdep_assert_held(&fp->fi_lock);
802
803 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
804 searchclp = searchdp->dl_stid.sc_client;
805 if (clp == searchclp) {
806 return -EAGAIN;
807 }
808 }
809 return 0;
810}
811
812/**
813 * hash_delegation_locked - Add a delegation to the appropriate lists
814 * @dp: a pointer to the nfs4_delegation we are adding.
815 * @fp: a pointer to the nfs4_file we're granting a delegation on
816 *
817 * Return:
818 * On success: NULL if the delegation was successfully hashed.
819 *
820 * On error: -EAGAIN if one was previously granted to this
821 * nfs4_client for this nfs4_file. Delegation is not hashed.
822 *
823 */
824
825static int
826hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
827{
828 int status;
829 struct nfs4_client *clp = dp->dl_stid.sc_client;
830
831 lockdep_assert_held(&state_lock);
832 lockdep_assert_held(&fp->fi_lock);
833
834 status = nfs4_get_existing_delegation(clp, fp);
835 if (status)
836 return status;
837 ++fp->fi_delegees;
838 atomic_inc(&dp->dl_stid.sc_count);
839 dp->dl_stid.sc_type = NFS4_DELEG_STID;
840 list_add(&dp->dl_perfile, &fp->fi_delegations);
841 list_add(&dp->dl_perclnt, &clp->cl_delegations);
842 return 0;
843}
844
845static bool
846unhash_delegation_locked(struct nfs4_delegation *dp)
847{
848 struct nfs4_file *fp = dp->dl_stid.sc_file;
849
850 lockdep_assert_held(&state_lock);
851
852 if (list_empty(&dp->dl_perfile))
853 return false;
854
855 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
856 /* Ensure that deleg break won't try to requeue it */
857 ++dp->dl_time;
858 spin_lock(&fp->fi_lock);
859 list_del_init(&dp->dl_perclnt);
860 list_del_init(&dp->dl_recall_lru);
861 list_del_init(&dp->dl_perfile);
862 spin_unlock(&fp->fi_lock);
863 return true;
864}
865
866static void destroy_delegation(struct nfs4_delegation *dp)
867{
868 bool unhashed;
869
870 spin_lock(&state_lock);
871 unhashed = unhash_delegation_locked(dp);
872 spin_unlock(&state_lock);
873 if (unhashed) {
874 put_clnt_odstate(dp->dl_clnt_odstate);
875 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
876 nfs4_put_stid(&dp->dl_stid);
877 }
878}
879
880static void revoke_delegation(struct nfs4_delegation *dp)
881{
882 struct nfs4_client *clp = dp->dl_stid.sc_client;
883
884 WARN_ON(!list_empty(&dp->dl_recall_lru));
885
886 put_clnt_odstate(dp->dl_clnt_odstate);
887 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
888
889 if (clp->cl_minorversion == 0)
890 nfs4_put_stid(&dp->dl_stid);
891 else {
892 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
893 spin_lock(&clp->cl_lock);
894 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
895 spin_unlock(&clp->cl_lock);
896 }
897}
898
899/*
900 * SETCLIENTID state
901 */
902
903static unsigned int clientid_hashval(u32 id)
904{
905 return id & CLIENT_HASH_MASK;
906}
907
908static unsigned int clientstr_hashval(const char *name)
909{
910 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
911}
912
913/*
914 * We store the NONE, READ, WRITE, and BOTH bits separately in the
915 * st_{access,deny}_bmap field of the stateid, in order to track not
916 * only what share bits are currently in force, but also what
917 * combinations of share bits previous opens have used. This allows us
918 * to enforce the recommendation of rfc 3530 14.2.19 that the server
919 * return an error if the client attempt to downgrade to a combination
920 * of share bits not explicable by closing some of its previous opens.
921 *
922 * XXX: This enforcement is actually incomplete, since we don't keep
923 * track of access/deny bit combinations; so, e.g., we allow:
924 *
925 * OPEN allow read, deny write
926 * OPEN allow both, deny none
927 * DOWNGRADE allow read, deny none
928 *
929 * which we should reject.
930 */
931static unsigned int
932bmap_to_share_mode(unsigned long bmap) {
933 int i;
934 unsigned int access = 0;
935
936 for (i = 1; i < 4; i++) {
937 if (test_bit(i, &bmap))
938 access |= i;
939 }
940 return access;
941}
942
943/* set share access for a given stateid */
944static inline void
945set_access(u32 access, struct nfs4_ol_stateid *stp)
946{
947 unsigned char mask = 1 << access;
948
949 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
950 stp->st_access_bmap |= mask;
951}
952
953/* clear share access for a given stateid */
954static inline void
955clear_access(u32 access, struct nfs4_ol_stateid *stp)
956{
957 unsigned char mask = 1 << access;
958
959 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
960 stp->st_access_bmap &= ~mask;
961}
962
963/* test whether a given stateid has access */
964static inline bool
965test_access(u32 access, struct nfs4_ol_stateid *stp)
966{
967 unsigned char mask = 1 << access;
968
969 return (bool)(stp->st_access_bmap & mask);
970}
971
972/* set share deny for a given stateid */
973static inline void
974set_deny(u32 deny, struct nfs4_ol_stateid *stp)
975{
976 unsigned char mask = 1 << deny;
977
978 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
979 stp->st_deny_bmap |= mask;
980}
981
982/* clear share deny for a given stateid */
983static inline void
984clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
985{
986 unsigned char mask = 1 << deny;
987
988 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
989 stp->st_deny_bmap &= ~mask;
990}
991
992/* test whether a given stateid is denying specific access */
993static inline bool
994test_deny(u32 deny, struct nfs4_ol_stateid *stp)
995{
996 unsigned char mask = 1 << deny;
997
998 return (bool)(stp->st_deny_bmap & mask);
999}
1000
1001static int nfs4_access_to_omode(u32 access)
1002{
1003 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1004 case NFS4_SHARE_ACCESS_READ:
1005 return O_RDONLY;
1006 case NFS4_SHARE_ACCESS_WRITE:
1007 return O_WRONLY;
1008 case NFS4_SHARE_ACCESS_BOTH:
1009 return O_RDWR;
1010 }
1011 WARN_ON_ONCE(1);
1012 return O_RDONLY;
1013}
1014
1015/*
1016 * A stateid that had a deny mode associated with it is being released
1017 * or downgraded. Recalculate the deny mode on the file.
1018 */
1019static void
1020recalculate_deny_mode(struct nfs4_file *fp)
1021{
1022 struct nfs4_ol_stateid *stp;
1023
1024 spin_lock(&fp->fi_lock);
1025 fp->fi_share_deny = 0;
1026 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1027 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1028 spin_unlock(&fp->fi_lock);
1029}
1030
1031static void
1032reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1033{
1034 int i;
1035 bool change = false;
1036
1037 for (i = 1; i < 4; i++) {
1038 if ((i & deny) != i) {
1039 change = true;
1040 clear_deny(i, stp);
1041 }
1042 }
1043
1044 /* Recalculate per-file deny mode if there was a change */
1045 if (change)
1046 recalculate_deny_mode(stp->st_stid.sc_file);
1047}
1048
1049/* release all access and file references for a given stateid */
1050static void
1051release_all_access(struct nfs4_ol_stateid *stp)
1052{
1053 int i;
1054 struct nfs4_file *fp = stp->st_stid.sc_file;
1055
1056 if (fp && stp->st_deny_bmap != 0)
1057 recalculate_deny_mode(fp);
1058
1059 for (i = 1; i < 4; i++) {
1060 if (test_access(i, stp))
1061 nfs4_file_put_access(stp->st_stid.sc_file, i);
1062 clear_access(i, stp);
1063 }
1064}
1065
1066static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1067{
1068 kfree(sop->so_owner.data);
1069 sop->so_ops->so_free(sop);
1070}
1071
1072static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1073{
1074 struct nfs4_client *clp = sop->so_client;
1075
1076 might_lock(&clp->cl_lock);
1077
1078 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1079 return;
1080 sop->so_ops->so_unhash(sop);
1081 spin_unlock(&clp->cl_lock);
1082 nfs4_free_stateowner(sop);
1083}
1084
1085static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1086{
1087 struct nfs4_file *fp = stp->st_stid.sc_file;
1088
1089 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1090
1091 if (list_empty(&stp->st_perfile))
1092 return false;
1093
1094 spin_lock(&fp->fi_lock);
1095 list_del_init(&stp->st_perfile);
1096 spin_unlock(&fp->fi_lock);
1097 list_del(&stp->st_perstateowner);
1098 return true;
1099}
1100
1101static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1102{
1103 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1104
1105 put_clnt_odstate(stp->st_clnt_odstate);
1106 release_all_access(stp);
1107 if (stp->st_stateowner)
1108 nfs4_put_stateowner(stp->st_stateowner);
1109 kmem_cache_free(stateid_slab, stid);
1110}
1111
1112static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1113{
1114 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1115 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1116 struct file *file;
1117
1118 file = find_any_file(stp->st_stid.sc_file);
1119 if (file)
1120 filp_close(file, (fl_owner_t)lo);
1121 nfs4_free_ol_stateid(stid);
1122}
1123
1124/*
1125 * Put the persistent reference to an already unhashed generic stateid, while
1126 * holding the cl_lock. If it's the last reference, then put it onto the
1127 * reaplist for later destruction.
1128 */
1129static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1130 struct list_head *reaplist)
1131{
1132 struct nfs4_stid *s = &stp->st_stid;
1133 struct nfs4_client *clp = s->sc_client;
1134
1135 lockdep_assert_held(&clp->cl_lock);
1136
1137 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1138
1139 if (!atomic_dec_and_test(&s->sc_count)) {
1140 wake_up_all(&close_wq);
1141 return;
1142 }
1143
1144 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1145 list_add(&stp->st_locks, reaplist);
1146}
1147
1148static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1149{
1150 struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1151
1152 lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1153
1154 list_del_init(&stp->st_locks);
1155 nfs4_unhash_stid(&stp->st_stid);
1156 return unhash_ol_stateid(stp);
1157}
1158
1159static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1160{
1161 struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1162 bool unhashed;
1163
1164 spin_lock(&oo->oo_owner.so_client->cl_lock);
1165 unhashed = unhash_lock_stateid(stp);
1166 spin_unlock(&oo->oo_owner.so_client->cl_lock);
1167 if (unhashed)
1168 nfs4_put_stid(&stp->st_stid);
1169}
1170
1171static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1172{
1173 struct nfs4_client *clp = lo->lo_owner.so_client;
1174
1175 lockdep_assert_held(&clp->cl_lock);
1176
1177 list_del_init(&lo->lo_owner.so_strhash);
1178}
1179
1180/*
1181 * Free a list of generic stateids that were collected earlier after being
1182 * fully unhashed.
1183 */
1184static void
1185free_ol_stateid_reaplist(struct list_head *reaplist)
1186{
1187 struct nfs4_ol_stateid *stp;
1188 struct nfs4_file *fp;
1189
1190 might_sleep();
1191
1192 while (!list_empty(reaplist)) {
1193 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1194 st_locks);
1195 list_del(&stp->st_locks);
1196 fp = stp->st_stid.sc_file;
1197 stp->st_stid.sc_free(&stp->st_stid);
1198 if (fp)
1199 put_nfs4_file(fp);
1200 }
1201}
1202
1203static void release_lockowner(struct nfs4_lockowner *lo)
1204{
1205 struct nfs4_client *clp = lo->lo_owner.so_client;
1206 struct nfs4_ol_stateid *stp;
1207 struct list_head reaplist;
1208
1209 INIT_LIST_HEAD(&reaplist);
1210
1211 spin_lock(&clp->cl_lock);
1212 unhash_lockowner_locked(lo);
1213 while (!list_empty(&lo->lo_owner.so_stateids)) {
1214 stp = list_first_entry(&lo->lo_owner.so_stateids,
1215 struct nfs4_ol_stateid, st_perstateowner);
1216 WARN_ON(!unhash_lock_stateid(stp));
1217 put_ol_stateid_locked(stp, &reaplist);
1218 }
1219 spin_unlock(&clp->cl_lock);
1220 free_ol_stateid_reaplist(&reaplist);
1221 nfs4_put_stateowner(&lo->lo_owner);
1222}
1223
1224static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1225 struct list_head *reaplist)
1226{
1227 struct nfs4_ol_stateid *stp;
1228
1229 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1230
1231 while (!list_empty(&open_stp->st_locks)) {
1232 stp = list_entry(open_stp->st_locks.next,
1233 struct nfs4_ol_stateid, st_locks);
1234 WARN_ON(!unhash_lock_stateid(stp));
1235 put_ol_stateid_locked(stp, reaplist);
1236 }
1237}
1238
1239static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1240 struct list_head *reaplist)
1241{
1242 bool unhashed;
1243
1244 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1245
1246 unhashed = unhash_ol_stateid(stp);
1247 release_open_stateid_locks(stp, reaplist);
1248 return unhashed;
1249}
1250
1251static void release_open_stateid(struct nfs4_ol_stateid *stp)
1252{
1253 LIST_HEAD(reaplist);
1254
1255 spin_lock(&stp->st_stid.sc_client->cl_lock);
1256 if (unhash_open_stateid(stp, &reaplist))
1257 put_ol_stateid_locked(stp, &reaplist);
1258 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1259 free_ol_stateid_reaplist(&reaplist);
1260}
1261
1262static void unhash_openowner_locked(struct nfs4_openowner *oo)
1263{
1264 struct nfs4_client *clp = oo->oo_owner.so_client;
1265
1266 lockdep_assert_held(&clp->cl_lock);
1267
1268 list_del_init(&oo->oo_owner.so_strhash);
1269 list_del_init(&oo->oo_perclient);
1270}
1271
1272static void release_last_closed_stateid(struct nfs4_openowner *oo)
1273{
1274 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1275 nfsd_net_id);
1276 struct nfs4_ol_stateid *s;
1277
1278 spin_lock(&nn->client_lock);
1279 s = oo->oo_last_closed_stid;
1280 if (s) {
1281 list_del_init(&oo->oo_close_lru);
1282 oo->oo_last_closed_stid = NULL;
1283 }
1284 spin_unlock(&nn->client_lock);
1285 if (s)
1286 nfs4_put_stid(&s->st_stid);
1287}
1288
1289static void release_openowner(struct nfs4_openowner *oo)
1290{
1291 struct nfs4_ol_stateid *stp;
1292 struct nfs4_client *clp = oo->oo_owner.so_client;
1293 struct list_head reaplist;
1294
1295 INIT_LIST_HEAD(&reaplist);
1296
1297 spin_lock(&clp->cl_lock);
1298 unhash_openowner_locked(oo);
1299 while (!list_empty(&oo->oo_owner.so_stateids)) {
1300 stp = list_first_entry(&oo->oo_owner.so_stateids,
1301 struct nfs4_ol_stateid, st_perstateowner);
1302 if (unhash_open_stateid(stp, &reaplist))
1303 put_ol_stateid_locked(stp, &reaplist);
1304 }
1305 spin_unlock(&clp->cl_lock);
1306 free_ol_stateid_reaplist(&reaplist);
1307 release_last_closed_stateid(oo);
1308 nfs4_put_stateowner(&oo->oo_owner);
1309}
1310
1311static inline int
1312hash_sessionid(struct nfs4_sessionid *sessionid)
1313{
1314 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1315
1316 return sid->sequence % SESSION_HASH_SIZE;
1317}
1318
1319#ifdef CONFIG_SUNRPC_DEBUG
1320static inline void
1321dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1322{
1323 u32 *ptr = (u32 *)(&sessionid->data[0]);
1324 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1325}
1326#else
1327static inline void
1328dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1329{
1330}
1331#endif
1332
1333/*
1334 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1335 * won't be used for replay.
1336 */
1337void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1338{
1339 struct nfs4_stateowner *so = cstate->replay_owner;
1340
1341 if (nfserr == nfserr_replay_me)
1342 return;
1343
1344 if (!seqid_mutating_err(ntohl(nfserr))) {
1345 nfsd4_cstate_clear_replay(cstate);
1346 return;
1347 }
1348 if (!so)
1349 return;
1350 if (so->so_is_open_owner)
1351 release_last_closed_stateid(openowner(so));
1352 so->so_seqid++;
1353 return;
1354}
1355
1356static void
1357gen_sessionid(struct nfsd4_session *ses)
1358{
1359 struct nfs4_client *clp = ses->se_client;
1360 struct nfsd4_sessionid *sid;
1361
1362 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1363 sid->clientid = clp->cl_clientid;
1364 sid->sequence = current_sessionid++;
1365 sid->reserved = 0;
1366}
1367
1368/*
1369 * The protocol defines ca_maxresponssize_cached to include the size of
1370 * the rpc header, but all we need to cache is the data starting after
1371 * the end of the initial SEQUENCE operation--the rest we regenerate
1372 * each time. Therefore we can advertise a ca_maxresponssize_cached
1373 * value that is the number of bytes in our cache plus a few additional
1374 * bytes. In order to stay on the safe side, and not promise more than
1375 * we can cache, those additional bytes must be the minimum possible: 24
1376 * bytes of rpc header (xid through accept state, with AUTH_NULL
1377 * verifier), 12 for the compound header (with zero-length tag), and 44
1378 * for the SEQUENCE op response:
1379 */
1380#define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1381
1382static void
1383free_session_slots(struct nfsd4_session *ses)
1384{
1385 int i;
1386
1387 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1388 kfree(ses->se_slots[i]);
1389}
1390
1391/*
1392 * We don't actually need to cache the rpc and session headers, so we
1393 * can allocate a little less for each slot:
1394 */
1395static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1396{
1397 u32 size;
1398
1399 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1400 size = 0;
1401 else
1402 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1403 return size + sizeof(struct nfsd4_slot);
1404}
1405
1406/*
1407 * XXX: If we run out of reserved DRC memory we could (up to a point)
1408 * re-negotiate active sessions and reduce their slot usage to make
1409 * room for new connections. For now we just fail the create session.
1410 */
1411static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1412{
1413 u32 slotsize = slot_bytes(ca);
1414 u32 num = ca->maxreqs;
1415 int avail;
1416
1417 spin_lock(&nfsd_drc_lock);
1418 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1419 nfsd_drc_max_mem - nfsd_drc_mem_used);
1420 num = min_t(int, num, avail / slotsize);
1421 nfsd_drc_mem_used += num * slotsize;
1422 spin_unlock(&nfsd_drc_lock);
1423
1424 return num;
1425}
1426
1427static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1428{
1429 int slotsize = slot_bytes(ca);
1430
1431 spin_lock(&nfsd_drc_lock);
1432 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1433 spin_unlock(&nfsd_drc_lock);
1434}
1435
1436static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1437 struct nfsd4_channel_attrs *battrs)
1438{
1439 int numslots = fattrs->maxreqs;
1440 int slotsize = slot_bytes(fattrs);
1441 struct nfsd4_session *new;
1442 int mem, i;
1443
1444 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1445 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1446 mem = numslots * sizeof(struct nfsd4_slot *);
1447
1448 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1449 if (!new)
1450 return NULL;
1451 /* allocate each struct nfsd4_slot and data cache in one piece */
1452 for (i = 0; i < numslots; i++) {
1453 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1454 if (!new->se_slots[i])
1455 goto out_free;
1456 }
1457
1458 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1459 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1460
1461 return new;
1462out_free:
1463 while (i--)
1464 kfree(new->se_slots[i]);
1465 kfree(new);
1466 return NULL;
1467}
1468
1469static void free_conn(struct nfsd4_conn *c)
1470{
1471 svc_xprt_put(c->cn_xprt);
1472 kfree(c);
1473}
1474
1475static void nfsd4_conn_lost(struct svc_xpt_user *u)
1476{
1477 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1478 struct nfs4_client *clp = c->cn_session->se_client;
1479
1480 spin_lock(&clp->cl_lock);
1481 if (!list_empty(&c->cn_persession)) {
1482 list_del(&c->cn_persession);
1483 free_conn(c);
1484 }
1485 nfsd4_probe_callback(clp);
1486 spin_unlock(&clp->cl_lock);
1487}
1488
1489static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1490{
1491 struct nfsd4_conn *conn;
1492
1493 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1494 if (!conn)
1495 return NULL;
1496 svc_xprt_get(rqstp->rq_xprt);
1497 conn->cn_xprt = rqstp->rq_xprt;
1498 conn->cn_flags = flags;
1499 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1500 return conn;
1501}
1502
1503static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1504{
1505 conn->cn_session = ses;
1506 list_add(&conn->cn_persession, &ses->se_conns);
1507}
1508
1509static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1510{
1511 struct nfs4_client *clp = ses->se_client;
1512
1513 spin_lock(&clp->cl_lock);
1514 __nfsd4_hash_conn(conn, ses);
1515 spin_unlock(&clp->cl_lock);
1516}
1517
1518static int nfsd4_register_conn(struct nfsd4_conn *conn)
1519{
1520 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1521 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1522}
1523
1524static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1525{
1526 int ret;
1527
1528 nfsd4_hash_conn(conn, ses);
1529 ret = nfsd4_register_conn(conn);
1530 if (ret)
1531 /* oops; xprt is already down: */
1532 nfsd4_conn_lost(&conn->cn_xpt_user);
1533 /* We may have gained or lost a callback channel: */
1534 nfsd4_probe_callback_sync(ses->se_client);
1535}
1536
1537static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1538{
1539 u32 dir = NFS4_CDFC4_FORE;
1540
1541 if (cses->flags & SESSION4_BACK_CHAN)
1542 dir |= NFS4_CDFC4_BACK;
1543 return alloc_conn(rqstp, dir);
1544}
1545
1546/* must be called under client_lock */
1547static void nfsd4_del_conns(struct nfsd4_session *s)
1548{
1549 struct nfs4_client *clp = s->se_client;
1550 struct nfsd4_conn *c;
1551
1552 spin_lock(&clp->cl_lock);
1553 while (!list_empty(&s->se_conns)) {
1554 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1555 list_del_init(&c->cn_persession);
1556 spin_unlock(&clp->cl_lock);
1557
1558 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1559 free_conn(c);
1560
1561 spin_lock(&clp->cl_lock);
1562 }
1563 spin_unlock(&clp->cl_lock);
1564}
1565
1566static void __free_session(struct nfsd4_session *ses)
1567{
1568 free_session_slots(ses);
1569 kfree(ses);
1570}
1571
1572static void free_session(struct nfsd4_session *ses)
1573{
1574 nfsd4_del_conns(ses);
1575 nfsd4_put_drc_mem(&ses->se_fchannel);
1576 __free_session(ses);
1577}
1578
1579static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1580{
1581 int idx;
1582 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1583
1584 new->se_client = clp;
1585 gen_sessionid(new);
1586
1587 INIT_LIST_HEAD(&new->se_conns);
1588
1589 new->se_cb_seq_nr = 1;
1590 new->se_flags = cses->flags;
1591 new->se_cb_prog = cses->callback_prog;
1592 new->se_cb_sec = cses->cb_sec;
1593 atomic_set(&new->se_ref, 0);
1594 idx = hash_sessionid(&new->se_sessionid);
1595 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1596 spin_lock(&clp->cl_lock);
1597 list_add(&new->se_perclnt, &clp->cl_sessions);
1598 spin_unlock(&clp->cl_lock);
1599
1600 {
1601 struct sockaddr *sa = svc_addr(rqstp);
1602 /*
1603 * This is a little silly; with sessions there's no real
1604 * use for the callback address. Use the peer address
1605 * as a reasonable default for now, but consider fixing
1606 * the rpc client not to require an address in the
1607 * future:
1608 */
1609 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1610 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1611 }
1612}
1613
1614/* caller must hold client_lock */
1615static struct nfsd4_session *
1616__find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1617{
1618 struct nfsd4_session *elem;
1619 int idx;
1620 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1621
1622 lockdep_assert_held(&nn->client_lock);
1623
1624 dump_sessionid(__func__, sessionid);
1625 idx = hash_sessionid(sessionid);
1626 /* Search in the appropriate list */
1627 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1628 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1629 NFS4_MAX_SESSIONID_LEN)) {
1630 return elem;
1631 }
1632 }
1633
1634 dprintk("%s: session not found\n", __func__);
1635 return NULL;
1636}
1637
1638static struct nfsd4_session *
1639find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1640 __be32 *ret)
1641{
1642 struct nfsd4_session *session;
1643 __be32 status = nfserr_badsession;
1644
1645 session = __find_in_sessionid_hashtbl(sessionid, net);
1646 if (!session)
1647 goto out;
1648 status = nfsd4_get_session_locked(session);
1649 if (status)
1650 session = NULL;
1651out:
1652 *ret = status;
1653 return session;
1654}
1655
1656/* caller must hold client_lock */
1657static void
1658unhash_session(struct nfsd4_session *ses)
1659{
1660 struct nfs4_client *clp = ses->se_client;
1661 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1662
1663 lockdep_assert_held(&nn->client_lock);
1664
1665 list_del(&ses->se_hash);
1666 spin_lock(&ses->se_client->cl_lock);
1667 list_del(&ses->se_perclnt);
1668 spin_unlock(&ses->se_client->cl_lock);
1669}
1670
1671/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1672static int
1673STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1674{
1675 /*
1676 * We're assuming the clid was not given out from a boot
1677 * precisely 2^32 (about 136 years) before this one. That seems
1678 * a safe assumption:
1679 */
1680 if (clid->cl_boot == (u32)nn->boot_time)
1681 return 0;
1682 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1683 clid->cl_boot, clid->cl_id, nn->boot_time);
1684 return 1;
1685}
1686
1687/*
1688 * XXX Should we use a slab cache ?
1689 * This type of memory management is somewhat inefficient, but we use it
1690 * anyway since SETCLIENTID is not a common operation.
1691 */
1692static struct nfs4_client *alloc_client(struct xdr_netobj name)
1693{
1694 struct nfs4_client *clp;
1695 int i;
1696
1697 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1698 if (clp == NULL)
1699 return NULL;
1700 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1701 if (clp->cl_name.data == NULL)
1702 goto err_no_name;
1703 clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1704 OWNER_HASH_SIZE, GFP_KERNEL);
1705 if (!clp->cl_ownerstr_hashtbl)
1706 goto err_no_hashtbl;
1707 for (i = 0; i < OWNER_HASH_SIZE; i++)
1708 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1709 clp->cl_name.len = name.len;
1710 INIT_LIST_HEAD(&clp->cl_sessions);
1711 idr_init(&clp->cl_stateids);
1712 atomic_set(&clp->cl_refcount, 0);
1713 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1714 INIT_LIST_HEAD(&clp->cl_idhash);
1715 INIT_LIST_HEAD(&clp->cl_openowners);
1716 INIT_LIST_HEAD(&clp->cl_delegations);
1717 INIT_LIST_HEAD(&clp->cl_lru);
1718 INIT_LIST_HEAD(&clp->cl_revoked);
1719#ifdef CONFIG_NFSD_PNFS
1720 INIT_LIST_HEAD(&clp->cl_lo_states);
1721#endif
1722 spin_lock_init(&clp->cl_lock);
1723 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1724 return clp;
1725err_no_hashtbl:
1726 kfree(clp->cl_name.data);
1727err_no_name:
1728 kfree(clp);
1729 return NULL;
1730}
1731
1732static void
1733free_client(struct nfs4_client *clp)
1734{
1735 while (!list_empty(&clp->cl_sessions)) {
1736 struct nfsd4_session *ses;
1737 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1738 se_perclnt);
1739 list_del(&ses->se_perclnt);
1740 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1741 free_session(ses);
1742 }
1743 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1744 free_svc_cred(&clp->cl_cred);
1745 kfree(clp->cl_ownerstr_hashtbl);
1746 kfree(clp->cl_name.data);
1747 idr_destroy(&clp->cl_stateids);
1748 kfree(clp);
1749}
1750
1751/* must be called under the client_lock */
1752static void
1753unhash_client_locked(struct nfs4_client *clp)
1754{
1755 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1756 struct nfsd4_session *ses;
1757
1758 lockdep_assert_held(&nn->client_lock);
1759
1760 /* Mark the client as expired! */
1761 clp->cl_time = 0;
1762 /* Make it invisible */
1763 if (!list_empty(&clp->cl_idhash)) {
1764 list_del_init(&clp->cl_idhash);
1765 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1766 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1767 else
1768 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1769 }
1770 list_del_init(&clp->cl_lru);
1771 spin_lock(&clp->cl_lock);
1772 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1773 list_del_init(&ses->se_hash);
1774 spin_unlock(&clp->cl_lock);
1775}
1776
1777static void
1778unhash_client(struct nfs4_client *clp)
1779{
1780 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1781
1782 spin_lock(&nn->client_lock);
1783 unhash_client_locked(clp);
1784 spin_unlock(&nn->client_lock);
1785}
1786
1787static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1788{
1789 if (atomic_read(&clp->cl_refcount))
1790 return nfserr_jukebox;
1791 unhash_client_locked(clp);
1792 return nfs_ok;
1793}
1794
1795static void
1796__destroy_client(struct nfs4_client *clp)
1797{
1798 struct nfs4_openowner *oo;
1799 struct nfs4_delegation *dp;
1800 struct list_head reaplist;
1801
1802 INIT_LIST_HEAD(&reaplist);
1803 spin_lock(&state_lock);
1804 while (!list_empty(&clp->cl_delegations)) {
1805 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1806 WARN_ON(!unhash_delegation_locked(dp));
1807 list_add(&dp->dl_recall_lru, &reaplist);
1808 }
1809 spin_unlock(&state_lock);
1810 while (!list_empty(&reaplist)) {
1811 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1812 list_del_init(&dp->dl_recall_lru);
1813 put_clnt_odstate(dp->dl_clnt_odstate);
1814 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1815 nfs4_put_stid(&dp->dl_stid);
1816 }
1817 while (!list_empty(&clp->cl_revoked)) {
1818 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1819 list_del_init(&dp->dl_recall_lru);
1820 nfs4_put_stid(&dp->dl_stid);
1821 }
1822 while (!list_empty(&clp->cl_openowners)) {
1823 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1824 nfs4_get_stateowner(&oo->oo_owner);
1825 release_openowner(oo);
1826 }
1827 nfsd4_return_all_client_layouts(clp);
1828 nfsd4_shutdown_callback(clp);
1829 if (clp->cl_cb_conn.cb_xprt)
1830 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1831 free_client(clp);
1832}
1833
1834static void
1835destroy_client(struct nfs4_client *clp)
1836{
1837 unhash_client(clp);
1838 __destroy_client(clp);
1839}
1840
1841static void expire_client(struct nfs4_client *clp)
1842{
1843 unhash_client(clp);
1844 nfsd4_client_record_remove(clp);
1845 __destroy_client(clp);
1846}
1847
1848static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1849{
1850 memcpy(target->cl_verifier.data, source->data,
1851 sizeof(target->cl_verifier.data));
1852}
1853
1854static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1855{
1856 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1857 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1858}
1859
1860int strdup_if_nonnull(char **target, char *source)
1861{
1862 if (source) {
1863 *target = kstrdup(source, GFP_KERNEL);
1864 if (!*target)
1865 return -ENOMEM;
1866 } else
1867 *target = NULL;
1868 return 0;
1869}
1870
1871static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1872{
1873 int ret;
1874
1875 ret = strdup_if_nonnull(&target->cr_principal, source->cr_principal);
1876 if (ret)
1877 return ret;
1878 ret = strdup_if_nonnull(&target->cr_raw_principal,
1879 source->cr_raw_principal);
1880 if (ret)
1881 return ret;
1882 target->cr_flavor = source->cr_flavor;
1883 target->cr_uid = source->cr_uid;
1884 target->cr_gid = source->cr_gid;
1885 target->cr_group_info = source->cr_group_info;
1886 get_group_info(target->cr_group_info);
1887 target->cr_gss_mech = source->cr_gss_mech;
1888 if (source->cr_gss_mech)
1889 gss_mech_get(source->cr_gss_mech);
1890 return 0;
1891}
1892
1893static int
1894compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1895{
1896 if (o1->len < o2->len)
1897 return -1;
1898 if (o1->len > o2->len)
1899 return 1;
1900 return memcmp(o1->data, o2->data, o1->len);
1901}
1902
1903static int same_name(const char *n1, const char *n2)
1904{
1905 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1906}
1907
1908static int
1909same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1910{
1911 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1912}
1913
1914static int
1915same_clid(clientid_t *cl1, clientid_t *cl2)
1916{
1917 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1918}
1919
1920static bool groups_equal(struct group_info *g1, struct group_info *g2)
1921{
1922 int i;
1923
1924 if (g1->ngroups != g2->ngroups)
1925 return false;
1926 for (i=0; i<g1->ngroups; i++)
1927 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1928 return false;
1929 return true;
1930}
1931
1932/*
1933 * RFC 3530 language requires clid_inuse be returned when the
1934 * "principal" associated with a requests differs from that previously
1935 * used. We use uid, gid's, and gss principal string as our best
1936 * approximation. We also don't want to allow non-gss use of a client
1937 * established using gss: in theory cr_principal should catch that
1938 * change, but in practice cr_principal can be null even in the gss case
1939 * since gssd doesn't always pass down a principal string.
1940 */
1941static bool is_gss_cred(struct svc_cred *cr)
1942{
1943 /* Is cr_flavor one of the gss "pseudoflavors"?: */
1944 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1945}
1946
1947
1948static bool
1949same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1950{
1951 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1952 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1953 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1954 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1955 return false;
1956 if (cr1->cr_principal == cr2->cr_principal)
1957 return true;
1958 if (!cr1->cr_principal || !cr2->cr_principal)
1959 return false;
1960 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1961}
1962
1963static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1964{
1965 struct svc_cred *cr = &rqstp->rq_cred;
1966 u32 service;
1967
1968 if (!cr->cr_gss_mech)
1969 return false;
1970 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1971 return service == RPC_GSS_SVC_INTEGRITY ||
1972 service == RPC_GSS_SVC_PRIVACY;
1973}
1974
1975static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1976{
1977 struct svc_cred *cr = &rqstp->rq_cred;
1978
1979 if (!cl->cl_mach_cred)
1980 return true;
1981 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1982 return false;
1983 if (!svc_rqst_integrity_protected(rqstp))
1984 return false;
1985 if (cl->cl_cred.cr_raw_principal)
1986 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
1987 cr->cr_raw_principal);
1988 if (!cr->cr_principal)
1989 return false;
1990 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1991}
1992
1993static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1994{
1995 __be32 verf[2];
1996
1997 /*
1998 * This is opaque to client, so no need to byte-swap. Use
1999 * __force to keep sparse happy
2000 */
2001 verf[0] = (__force __be32)get_seconds();
2002 verf[1] = (__force __be32)nn->clverifier_counter++;
2003 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2004}
2005
2006static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2007{
2008 clp->cl_clientid.cl_boot = nn->boot_time;
2009 clp->cl_clientid.cl_id = nn->clientid_counter++;
2010 gen_confirm(clp, nn);
2011}
2012
2013static struct nfs4_stid *
2014find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2015{
2016 struct nfs4_stid *ret;
2017
2018 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2019 if (!ret || !ret->sc_type)
2020 return NULL;
2021 return ret;
2022}
2023
2024static struct nfs4_stid *
2025find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2026{
2027 struct nfs4_stid *s;
2028
2029 spin_lock(&cl->cl_lock);
2030 s = find_stateid_locked(cl, t);
2031 if (s != NULL) {
2032 if (typemask & s->sc_type)
2033 atomic_inc(&s->sc_count);
2034 else
2035 s = NULL;
2036 }
2037 spin_unlock(&cl->cl_lock);
2038 return s;
2039}
2040
2041static struct nfs4_client *create_client(struct xdr_netobj name,
2042 struct svc_rqst *rqstp, nfs4_verifier *verf)
2043{
2044 struct nfs4_client *clp;
2045 struct sockaddr *sa = svc_addr(rqstp);
2046 int ret;
2047 struct net *net = SVC_NET(rqstp);
2048
2049 clp = alloc_client(name);
2050 if (clp == NULL)
2051 return NULL;
2052
2053 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2054 if (ret) {
2055 free_client(clp);
2056 return NULL;
2057 }
2058 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2059 clp->cl_time = get_seconds();
2060 clear_bit(0, &clp->cl_cb_slot_busy);
2061 copy_verf(clp, verf);
2062 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2063 clp->cl_cb_session = NULL;
2064 clp->net = net;
2065 return clp;
2066}
2067
2068static void
2069add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2070{
2071 struct rb_node **new = &(root->rb_node), *parent = NULL;
2072 struct nfs4_client *clp;
2073
2074 while (*new) {
2075 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2076 parent = *new;
2077
2078 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2079 new = &((*new)->rb_left);
2080 else
2081 new = &((*new)->rb_right);
2082 }
2083
2084 rb_link_node(&new_clp->cl_namenode, parent, new);
2085 rb_insert_color(&new_clp->cl_namenode, root);
2086}
2087
2088static struct nfs4_client *
2089find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2090{
2091 int cmp;
2092 struct rb_node *node = root->rb_node;
2093 struct nfs4_client *clp;
2094
2095 while (node) {
2096 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2097 cmp = compare_blob(&clp->cl_name, name);
2098 if (cmp > 0)
2099 node = node->rb_left;
2100 else if (cmp < 0)
2101 node = node->rb_right;
2102 else
2103 return clp;
2104 }
2105 return NULL;
2106}
2107
2108static void
2109add_to_unconfirmed(struct nfs4_client *clp)
2110{
2111 unsigned int idhashval;
2112 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2113
2114 lockdep_assert_held(&nn->client_lock);
2115
2116 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2117 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2118 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2119 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2120 renew_client_locked(clp);
2121}
2122
2123static void
2124move_to_confirmed(struct nfs4_client *clp)
2125{
2126 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2127 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2128
2129 lockdep_assert_held(&nn->client_lock);
2130
2131 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2132 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2133 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2134 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2135 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2136 renew_client_locked(clp);
2137}
2138
2139static struct nfs4_client *
2140find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2141{
2142 struct nfs4_client *clp;
2143 unsigned int idhashval = clientid_hashval(clid->cl_id);
2144
2145 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2146 if (same_clid(&clp->cl_clientid, clid)) {
2147 if ((bool)clp->cl_minorversion != sessions)
2148 return NULL;
2149 renew_client_locked(clp);
2150 return clp;
2151 }
2152 }
2153 return NULL;
2154}
2155
2156static struct nfs4_client *
2157find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2158{
2159 struct list_head *tbl = nn->conf_id_hashtbl;
2160
2161 lockdep_assert_held(&nn->client_lock);
2162 return find_client_in_id_table(tbl, clid, sessions);
2163}
2164
2165static struct nfs4_client *
2166find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2167{
2168 struct list_head *tbl = nn->unconf_id_hashtbl;
2169
2170 lockdep_assert_held(&nn->client_lock);
2171 return find_client_in_id_table(tbl, clid, sessions);
2172}
2173
2174static bool clp_used_exchangeid(struct nfs4_client *clp)
2175{
2176 return clp->cl_exchange_flags != 0;
2177}
2178
2179static struct nfs4_client *
2180find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2181{
2182 lockdep_assert_held(&nn->client_lock);
2183 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2184}
2185
2186static struct nfs4_client *
2187find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2188{
2189 lockdep_assert_held(&nn->client_lock);
2190 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2191}
2192
2193static void
2194gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2195{
2196 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2197 struct sockaddr *sa = svc_addr(rqstp);
2198 u32 scopeid = rpc_get_scope_id(sa);
2199 unsigned short expected_family;
2200
2201 /* Currently, we only support tcp and tcp6 for the callback channel */
2202 if (se->se_callback_netid_len == 3 &&
2203 !memcmp(se->se_callback_netid_val, "tcp", 3))
2204 expected_family = AF_INET;
2205 else if (se->se_callback_netid_len == 4 &&
2206 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2207 expected_family = AF_INET6;
2208 else
2209 goto out_err;
2210
2211 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2212 se->se_callback_addr_len,
2213 (struct sockaddr *)&conn->cb_addr,
2214 sizeof(conn->cb_addr));
2215
2216 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2217 goto out_err;
2218
2219 if (conn->cb_addr.ss_family == AF_INET6)
2220 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2221
2222 conn->cb_prog = se->se_callback_prog;
2223 conn->cb_ident = se->se_callback_ident;
2224 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2225 return;
2226out_err:
2227 conn->cb_addr.ss_family = AF_UNSPEC;
2228 conn->cb_addrlen = 0;
2229 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2230 "will not receive delegations\n",
2231 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2232
2233 return;
2234}
2235
2236/*
2237 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2238 */
2239static void
2240nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2241{
2242 struct xdr_buf *buf = resp->xdr.buf;
2243 struct nfsd4_slot *slot = resp->cstate.slot;
2244 unsigned int base;
2245
2246 dprintk("--> %s slot %p\n", __func__, slot);
2247
2248 slot->sl_opcnt = resp->opcnt;
2249 slot->sl_status = resp->cstate.status;
2250
2251 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2252 if (nfsd4_not_cached(resp)) {
2253 slot->sl_datalen = 0;
2254 return;
2255 }
2256 base = resp->cstate.data_offset;
2257 slot->sl_datalen = buf->len - base;
2258 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2259 WARN(1, "%s: sessions DRC could not cache compound\n",
2260 __func__);
2261 return;
2262}
2263
2264/*
2265 * Encode the replay sequence operation from the slot values.
2266 * If cachethis is FALSE encode the uncached rep error on the next
2267 * operation which sets resp->p and increments resp->opcnt for
2268 * nfs4svc_encode_compoundres.
2269 *
2270 */
2271static __be32
2272nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2273 struct nfsd4_compoundres *resp)
2274{
2275 struct nfsd4_op *op;
2276 struct nfsd4_slot *slot = resp->cstate.slot;
2277
2278 /* Encode the replayed sequence operation */
2279 op = &args->ops[resp->opcnt - 1];
2280 nfsd4_encode_operation(resp, op);
2281
2282 /* Return nfserr_retry_uncached_rep in next operation. */
2283 if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2284 op = &args->ops[resp->opcnt++];
2285 op->status = nfserr_retry_uncached_rep;
2286 nfsd4_encode_operation(resp, op);
2287 }
2288 return op->status;
2289}
2290
2291/*
2292 * The sequence operation is not cached because we can use the slot and
2293 * session values.
2294 */
2295static __be32
2296nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2297 struct nfsd4_sequence *seq)
2298{
2299 struct nfsd4_slot *slot = resp->cstate.slot;
2300 struct xdr_stream *xdr = &resp->xdr;
2301 __be32 *p;
2302 __be32 status;
2303
2304 dprintk("--> %s slot %p\n", __func__, slot);
2305
2306 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2307 if (status)
2308 return status;
2309
2310 p = xdr_reserve_space(xdr, slot->sl_datalen);
2311 if (!p) {
2312 WARN_ON_ONCE(1);
2313 return nfserr_serverfault;
2314 }
2315 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2316 xdr_commit_encode(xdr);
2317
2318 resp->opcnt = slot->sl_opcnt;
2319 return slot->sl_status;
2320}
2321
2322/*
2323 * Set the exchange_id flags returned by the server.
2324 */
2325static void
2326nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2327{
2328#ifdef CONFIG_NFSD_PNFS
2329 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2330#else
2331 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2332#endif
2333
2334 /* Referrals are supported, Migration is not. */
2335 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2336
2337 /* set the wire flags to return to client. */
2338 clid->flags = new->cl_exchange_flags;
2339}
2340
2341static bool client_has_openowners(struct nfs4_client *clp)
2342{
2343 struct nfs4_openowner *oo;
2344
2345 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2346 if (!list_empty(&oo->oo_owner.so_stateids))
2347 return true;
2348 }
2349 return false;
2350}
2351
2352static bool client_has_state(struct nfs4_client *clp)
2353{
2354 return client_has_openowners(clp)
2355#ifdef CONFIG_NFSD_PNFS
2356 || !list_empty(&clp->cl_lo_states)
2357#endif
2358 || !list_empty(&clp->cl_delegations)
2359 || !list_empty(&clp->cl_sessions);
2360}
2361
2362__be32
2363nfsd4_exchange_id(struct svc_rqst *rqstp,
2364 struct nfsd4_compound_state *cstate,
2365 struct nfsd4_exchange_id *exid)
2366{
2367 struct nfs4_client *conf, *new;
2368 struct nfs4_client *unconf = NULL;
2369 __be32 status;
2370 char addr_str[INET6_ADDRSTRLEN];
2371 nfs4_verifier verf = exid->verifier;
2372 struct sockaddr *sa = svc_addr(rqstp);
2373 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2374 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2375
2376 rpc_ntop(sa, addr_str, sizeof(addr_str));
2377 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2378 "ip_addr=%s flags %x, spa_how %d\n",
2379 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2380 addr_str, exid->flags, exid->spa_how);
2381
2382 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2383 return nfserr_inval;
2384
2385 new = create_client(exid->clname, rqstp, &verf);
2386 if (new == NULL)
2387 return nfserr_jukebox;
2388
2389 switch (exid->spa_how) {
2390 case SP4_MACH_CRED:
2391 if (!svc_rqst_integrity_protected(rqstp)) {
2392 status = nfserr_inval;
2393 goto out_nolock;
2394 }
2395 /*
2396 * Sometimes userspace doesn't give us a principal.
2397 * Which is a bug, really. Anyway, we can't enforce
2398 * MACH_CRED in that case, better to give up now:
2399 */
2400 if (!new->cl_cred.cr_principal &&
2401 !new->cl_cred.cr_raw_principal) {
2402 status = nfserr_serverfault;
2403 goto out_nolock;
2404 }
2405 new->cl_mach_cred = true;
2406 case SP4_NONE:
2407 break;
2408 default: /* checked by xdr code */
2409 WARN_ON_ONCE(1);
2410 case SP4_SSV:
2411 status = nfserr_encr_alg_unsupp;
2412 goto out_nolock;
2413 }
2414
2415 /* Cases below refer to rfc 5661 section 18.35.4: */
2416 spin_lock(&nn->client_lock);
2417 conf = find_confirmed_client_by_name(&exid->clname, nn);
2418 if (conf) {
2419 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2420 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2421
2422 if (update) {
2423 if (!clp_used_exchangeid(conf)) { /* buggy client */
2424 status = nfserr_inval;
2425 goto out;
2426 }
2427 if (!mach_creds_match(conf, rqstp)) {
2428 status = nfserr_wrong_cred;
2429 goto out;
2430 }
2431 if (!creds_match) { /* case 9 */
2432 status = nfserr_perm;
2433 goto out;
2434 }
2435 if (!verfs_match) { /* case 8 */
2436 status = nfserr_not_same;
2437 goto out;
2438 }
2439 /* case 6 */
2440 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2441 goto out_copy;
2442 }
2443 if (!creds_match) { /* case 3 */
2444 if (client_has_state(conf)) {
2445 status = nfserr_clid_inuse;
2446 goto out;
2447 }
2448 goto out_new;
2449 }
2450 if (verfs_match) { /* case 2 */
2451 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2452 goto out_copy;
2453 }
2454 /* case 5, client reboot */
2455 conf = NULL;
2456 goto out_new;
2457 }
2458
2459 if (update) { /* case 7 */
2460 status = nfserr_noent;
2461 goto out;
2462 }
2463
2464 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2465 if (unconf) /* case 4, possible retry or client restart */
2466 unhash_client_locked(unconf);
2467
2468 /* case 1 (normal case) */
2469out_new:
2470 if (conf) {
2471 status = mark_client_expired_locked(conf);
2472 if (status)
2473 goto out;
2474 }
2475 new->cl_minorversion = cstate->minorversion;
2476
2477 gen_clid(new, nn);
2478 add_to_unconfirmed(new);
2479 swap(new, conf);
2480out_copy:
2481 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2482 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2483
2484 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2485 nfsd4_set_ex_flags(conf, exid);
2486
2487 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2488 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2489 status = nfs_ok;
2490
2491out:
2492 spin_unlock(&nn->client_lock);
2493out_nolock:
2494 if (new)
2495 expire_client(new);
2496 if (unconf)
2497 expire_client(unconf);
2498 return status;
2499}
2500
2501static __be32
2502check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2503{
2504 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2505 slot_seqid);
2506
2507 /* The slot is in use, and no response has been sent. */
2508 if (slot_inuse) {
2509 if (seqid == slot_seqid)
2510 return nfserr_jukebox;
2511 else
2512 return nfserr_seq_misordered;
2513 }
2514 /* Note unsigned 32-bit arithmetic handles wraparound: */
2515 if (likely(seqid == slot_seqid + 1))
2516 return nfs_ok;
2517 if (seqid == slot_seqid)
2518 return nfserr_replay_cache;
2519 return nfserr_seq_misordered;
2520}
2521
2522/*
2523 * Cache the create session result into the create session single DRC
2524 * slot cache by saving the xdr structure. sl_seqid has been set.
2525 * Do this for solo or embedded create session operations.
2526 */
2527static void
2528nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2529 struct nfsd4_clid_slot *slot, __be32 nfserr)
2530{
2531 slot->sl_status = nfserr;
2532 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2533}
2534
2535static __be32
2536nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2537 struct nfsd4_clid_slot *slot)
2538{
2539 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2540 return slot->sl_status;
2541}
2542
2543#define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2544 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2545 1 + /* MIN tag is length with zero, only length */ \
2546 3 + /* version, opcount, opcode */ \
2547 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2548 /* seqid, slotID, slotID, cache */ \
2549 4 ) * sizeof(__be32))
2550
2551#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2552 2 + /* verifier: AUTH_NULL, length 0 */\
2553 1 + /* status */ \
2554 1 + /* MIN tag is length with zero, only length */ \
2555 3 + /* opcount, opcode, opstatus*/ \
2556 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2557 /* seqid, slotID, slotID, slotID, status */ \
2558 5 ) * sizeof(__be32))
2559
2560static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2561{
2562 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2563
2564 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2565 return nfserr_toosmall;
2566 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2567 return nfserr_toosmall;
2568 ca->headerpadsz = 0;
2569 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2570 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2571 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2572 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2573 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2574 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2575 /*
2576 * Note decreasing slot size below client's request may make it
2577 * difficult for client to function correctly, whereas
2578 * decreasing the number of slots will (just?) affect
2579 * performance. When short on memory we therefore prefer to
2580 * decrease number of slots instead of their size. Clients that
2581 * request larger slots than they need will get poor results:
2582 */
2583 ca->maxreqs = nfsd4_get_drc_mem(ca);
2584 if (!ca->maxreqs)
2585 return nfserr_jukebox;
2586
2587 return nfs_ok;
2588}
2589
2590/*
2591 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2592 * These are based on similar macros in linux/sunrpc/msg_prot.h .
2593 */
2594#define RPC_MAX_HEADER_WITH_AUTH_SYS \
2595 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2596
2597#define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2598 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2599
2600#define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2601 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2602#define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2603 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2604 sizeof(__be32))
2605
2606static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2607{
2608 ca->headerpadsz = 0;
2609
2610 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2611 return nfserr_toosmall;
2612 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2613 return nfserr_toosmall;
2614 ca->maxresp_cached = 0;
2615 if (ca->maxops < 2)
2616 return nfserr_toosmall;
2617
2618 return nfs_ok;
2619}
2620
2621static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2622{
2623 switch (cbs->flavor) {
2624 case RPC_AUTH_NULL:
2625 case RPC_AUTH_UNIX:
2626 return nfs_ok;
2627 default:
2628 /*
2629 * GSS case: the spec doesn't allow us to return this
2630 * error. But it also doesn't allow us not to support
2631 * GSS.
2632 * I'd rather this fail hard than return some error the
2633 * client might think it can already handle:
2634 */
2635 return nfserr_encr_alg_unsupp;
2636 }
2637}
2638
2639__be32
2640nfsd4_create_session(struct svc_rqst *rqstp,
2641 struct nfsd4_compound_state *cstate,
2642 struct nfsd4_create_session *cr_ses)
2643{
2644 struct sockaddr *sa = svc_addr(rqstp);
2645 struct nfs4_client *conf, *unconf;
2646 struct nfs4_client *old = NULL;
2647 struct nfsd4_session *new;
2648 struct nfsd4_conn *conn;
2649 struct nfsd4_clid_slot *cs_slot = NULL;
2650 __be32 status = 0;
2651 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2652
2653 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2654 return nfserr_inval;
2655 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2656 if (status)
2657 return status;
2658 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2659 if (status)
2660 return status;
2661 status = check_backchannel_attrs(&cr_ses->back_channel);
2662 if (status)
2663 goto out_release_drc_mem;
2664 status = nfserr_jukebox;
2665 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2666 if (!new)
2667 goto out_release_drc_mem;
2668 conn = alloc_conn_from_crses(rqstp, cr_ses);
2669 if (!conn)
2670 goto out_free_session;
2671
2672 spin_lock(&nn->client_lock);
2673 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2674 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2675 WARN_ON_ONCE(conf && unconf);
2676
2677 if (conf) {
2678 status = nfserr_wrong_cred;
2679 if (!mach_creds_match(conf, rqstp))
2680 goto out_free_conn;
2681 cs_slot = &conf->cl_cs_slot;
2682 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2683 if (status) {
2684 if (status == nfserr_replay_cache)
2685 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2686 goto out_free_conn;
2687 }
2688 } else if (unconf) {
2689 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2690 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2691 status = nfserr_clid_inuse;
2692 goto out_free_conn;
2693 }
2694 status = nfserr_wrong_cred;
2695 if (!mach_creds_match(unconf, rqstp))
2696 goto out_free_conn;
2697 cs_slot = &unconf->cl_cs_slot;
2698 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2699 if (status) {
2700 /* an unconfirmed replay returns misordered */
2701 status = nfserr_seq_misordered;
2702 goto out_free_conn;
2703 }
2704 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2705 if (old) {
2706 status = mark_client_expired_locked(old);
2707 if (status) {
2708 old = NULL;
2709 goto out_free_conn;
2710 }
2711 }
2712 move_to_confirmed(unconf);
2713 conf = unconf;
2714 } else {
2715 status = nfserr_stale_clientid;
2716 goto out_free_conn;
2717 }
2718 status = nfs_ok;
2719 /* Persistent sessions are not supported */
2720 cr_ses->flags &= ~SESSION4_PERSIST;
2721 /* Upshifting from TCP to RDMA is not supported */
2722 cr_ses->flags &= ~SESSION4_RDMA;
2723
2724 init_session(rqstp, new, conf, cr_ses);
2725 nfsd4_get_session_locked(new);
2726
2727 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2728 NFS4_MAX_SESSIONID_LEN);
2729 cs_slot->sl_seqid++;
2730 cr_ses->seqid = cs_slot->sl_seqid;
2731
2732 /* cache solo and embedded create sessions under the client_lock */
2733 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2734 spin_unlock(&nn->client_lock);
2735 /* init connection and backchannel */
2736 nfsd4_init_conn(rqstp, conn, new);
2737 nfsd4_put_session(new);
2738 if (old)
2739 expire_client(old);
2740 return status;
2741out_free_conn:
2742 spin_unlock(&nn->client_lock);
2743 free_conn(conn);
2744 if (old)
2745 expire_client(old);
2746out_free_session:
2747 __free_session(new);
2748out_release_drc_mem:
2749 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2750 return status;
2751}
2752
2753static __be32 nfsd4_map_bcts_dir(u32 *dir)
2754{
2755 switch (*dir) {
2756 case NFS4_CDFC4_FORE:
2757 case NFS4_CDFC4_BACK:
2758 return nfs_ok;
2759 case NFS4_CDFC4_FORE_OR_BOTH:
2760 case NFS4_CDFC4_BACK_OR_BOTH:
2761 *dir = NFS4_CDFC4_BOTH;
2762 return nfs_ok;
2763 };
2764 return nfserr_inval;
2765}
2766
2767__be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2768{
2769 struct nfsd4_session *session = cstate->session;
2770 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2771 __be32 status;
2772
2773 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2774 if (status)
2775 return status;
2776 spin_lock(&nn->client_lock);
2777 session->se_cb_prog = bc->bc_cb_program;
2778 session->se_cb_sec = bc->bc_cb_sec;
2779 spin_unlock(&nn->client_lock);
2780
2781 nfsd4_probe_callback(session->se_client);
2782
2783 return nfs_ok;
2784}
2785
2786__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2787 struct nfsd4_compound_state *cstate,
2788 struct nfsd4_bind_conn_to_session *bcts)
2789{
2790 __be32 status;
2791 struct nfsd4_conn *conn;
2792 struct nfsd4_session *session;
2793 struct net *net = SVC_NET(rqstp);
2794 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2795
2796 if (!nfsd4_last_compound_op(rqstp))
2797 return nfserr_not_only_op;
2798 spin_lock(&nn->client_lock);
2799 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2800 spin_unlock(&nn->client_lock);
2801 if (!session)
2802 goto out_no_session;
2803 status = nfserr_wrong_cred;
2804 if (!mach_creds_match(session->se_client, rqstp))
2805 goto out;
2806 status = nfsd4_map_bcts_dir(&bcts->dir);
2807 if (status)
2808 goto out;
2809 conn = alloc_conn(rqstp, bcts->dir);
2810 status = nfserr_jukebox;
2811 if (!conn)
2812 goto out;
2813 nfsd4_init_conn(rqstp, conn, session);
2814 status = nfs_ok;
2815out:
2816 nfsd4_put_session(session);
2817out_no_session:
2818 return status;
2819}
2820
2821static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2822{
2823 if (!session)
2824 return 0;
2825 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2826}
2827
2828__be32
2829nfsd4_destroy_session(struct svc_rqst *r,
2830 struct nfsd4_compound_state *cstate,
2831 struct nfsd4_destroy_session *sessionid)
2832{
2833 struct nfsd4_session *ses;
2834 __be32 status;
2835 int ref_held_by_me = 0;
2836 struct net *net = SVC_NET(r);
2837 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2838
2839 status = nfserr_not_only_op;
2840 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2841 if (!nfsd4_last_compound_op(r))
2842 goto out;
2843 ref_held_by_me++;
2844 }
2845 dump_sessionid(__func__, &sessionid->sessionid);
2846 spin_lock(&nn->client_lock);
2847 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2848 if (!ses)
2849 goto out_client_lock;
2850 status = nfserr_wrong_cred;
2851 if (!mach_creds_match(ses->se_client, r))
2852 goto out_put_session;
2853 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2854 if (status)
2855 goto out_put_session;
2856 unhash_session(ses);
2857 spin_unlock(&nn->client_lock);
2858
2859 nfsd4_probe_callback_sync(ses->se_client);
2860
2861 spin_lock(&nn->client_lock);
2862 status = nfs_ok;
2863out_put_session:
2864 nfsd4_put_session_locked(ses);
2865out_client_lock:
2866 spin_unlock(&nn->client_lock);
2867out:
2868 return status;
2869}
2870
2871static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2872{
2873 struct nfsd4_conn *c;
2874
2875 list_for_each_entry(c, &s->se_conns, cn_persession) {
2876 if (c->cn_xprt == xpt) {
2877 return c;
2878 }
2879 }
2880 return NULL;
2881}
2882
2883static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2884{
2885 struct nfs4_client *clp = ses->se_client;
2886 struct nfsd4_conn *c;
2887 __be32 status = nfs_ok;
2888 int ret;
2889
2890 spin_lock(&clp->cl_lock);
2891 c = __nfsd4_find_conn(new->cn_xprt, ses);
2892 if (c)
2893 goto out_free;
2894 status = nfserr_conn_not_bound_to_session;
2895 if (clp->cl_mach_cred)
2896 goto out_free;
2897 __nfsd4_hash_conn(new, ses);
2898 spin_unlock(&clp->cl_lock);
2899 ret = nfsd4_register_conn(new);
2900 if (ret)
2901 /* oops; xprt is already down: */
2902 nfsd4_conn_lost(&new->cn_xpt_user);
2903 return nfs_ok;
2904out_free:
2905 spin_unlock(&clp->cl_lock);
2906 free_conn(new);
2907 return status;
2908}
2909
2910static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2911{
2912 struct nfsd4_compoundargs *args = rqstp->rq_argp;
2913
2914 return args->opcnt > session->se_fchannel.maxops;
2915}
2916
2917static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2918 struct nfsd4_session *session)
2919{
2920 struct xdr_buf *xb = &rqstp->rq_arg;
2921
2922 return xb->len > session->se_fchannel.maxreq_sz;
2923}
2924
2925__be32
2926nfsd4_sequence(struct svc_rqst *rqstp,
2927 struct nfsd4_compound_state *cstate,
2928 struct nfsd4_sequence *seq)
2929{
2930 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2931 struct xdr_stream *xdr = &resp->xdr;
2932 struct nfsd4_session *session;
2933 struct nfs4_client *clp;
2934 struct nfsd4_slot *slot;
2935 struct nfsd4_conn *conn;
2936 __be32 status;
2937 int buflen;
2938 struct net *net = SVC_NET(rqstp);
2939 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2940
2941 if (resp->opcnt != 1)
2942 return nfserr_sequence_pos;
2943
2944 /*
2945 * Will be either used or freed by nfsd4_sequence_check_conn
2946 * below.
2947 */
2948 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2949 if (!conn)
2950 return nfserr_jukebox;
2951
2952 spin_lock(&nn->client_lock);
2953 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2954 if (!session)
2955 goto out_no_session;
2956 clp = session->se_client;
2957
2958 status = nfserr_too_many_ops;
2959 if (nfsd4_session_too_many_ops(rqstp, session))
2960 goto out_put_session;
2961
2962 status = nfserr_req_too_big;
2963 if (nfsd4_request_too_big(rqstp, session))
2964 goto out_put_session;
2965
2966 status = nfserr_badslot;
2967 if (seq->slotid >= session->se_fchannel.maxreqs)
2968 goto out_put_session;
2969
2970 slot = session->se_slots[seq->slotid];
2971 dprintk("%s: slotid %d\n", __func__, seq->slotid);
2972
2973 /* We do not negotiate the number of slots yet, so set the
2974 * maxslots to the session maxreqs which is used to encode
2975 * sr_highest_slotid and the sr_target_slot id to maxslots */
2976 seq->maxslots = session->se_fchannel.maxreqs;
2977
2978 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2979 slot->sl_flags & NFSD4_SLOT_INUSE);
2980 if (status == nfserr_replay_cache) {
2981 status = nfserr_seq_misordered;
2982 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2983 goto out_put_session;
2984 cstate->slot = slot;
2985 cstate->session = session;
2986 cstate->clp = clp;
2987 /* Return the cached reply status and set cstate->status
2988 * for nfsd4_proc_compound processing */
2989 status = nfsd4_replay_cache_entry(resp, seq);
2990 cstate->status = nfserr_replay_cache;
2991 goto out;
2992 }
2993 if (status)
2994 goto out_put_session;
2995
2996 status = nfsd4_sequence_check_conn(conn, session);
2997 conn = NULL;
2998 if (status)
2999 goto out_put_session;
3000
3001 buflen = (seq->cachethis) ?
3002 session->se_fchannel.maxresp_cached :
3003 session->se_fchannel.maxresp_sz;
3004 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3005 nfserr_rep_too_big;
3006 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3007 goto out_put_session;
3008 svc_reserve(rqstp, buflen);
3009
3010 status = nfs_ok;
3011 /* Success! bump slot seqid */
3012 slot->sl_seqid = seq->seqid;
3013 slot->sl_flags |= NFSD4_SLOT_INUSE;
3014 if (seq->cachethis)
3015 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3016 else
3017 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3018
3019 cstate->slot = slot;
3020 cstate->session = session;
3021 cstate->clp = clp;
3022
3023out:
3024 switch (clp->cl_cb_state) {
3025 case NFSD4_CB_DOWN:
3026 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3027 break;
3028 case NFSD4_CB_FAULT:
3029 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3030 break;
3031 default:
3032 seq->status_flags = 0;
3033 }
3034 if (!list_empty(&clp->cl_revoked))
3035 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3036out_no_session:
3037 if (conn)
3038 free_conn(conn);
3039 spin_unlock(&nn->client_lock);
3040 return status;
3041out_put_session:
3042 nfsd4_put_session_locked(session);
3043 goto out_no_session;
3044}
3045
3046void
3047nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3048{
3049 struct nfsd4_compound_state *cs = &resp->cstate;
3050
3051 if (nfsd4_has_session(cs)) {
3052 if (cs->status != nfserr_replay_cache) {
3053 nfsd4_store_cache_entry(resp);
3054 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3055 }
3056 /* Drop session reference that was taken in nfsd4_sequence() */
3057 nfsd4_put_session(cs->session);
3058 } else if (cs->clp)
3059 put_client_renew(cs->clp);
3060}
3061
3062__be32
3063nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3064{
3065 struct nfs4_client *conf, *unconf;
3066 struct nfs4_client *clp = NULL;
3067 __be32 status = 0;
3068 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3069
3070 spin_lock(&nn->client_lock);
3071 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3072 conf = find_confirmed_client(&dc->clientid, true, nn);
3073 WARN_ON_ONCE(conf && unconf);
3074
3075 if (conf) {
3076 if (client_has_state(conf)) {
3077 status = nfserr_clientid_busy;
3078 goto out;
3079 }
3080 status = mark_client_expired_locked(conf);
3081 if (status)
3082 goto out;
3083 clp = conf;
3084 } else if (unconf)
3085 clp = unconf;
3086 else {
3087 status = nfserr_stale_clientid;
3088 goto out;
3089 }
3090 if (!mach_creds_match(clp, rqstp)) {
3091 clp = NULL;
3092 status = nfserr_wrong_cred;
3093 goto out;
3094 }
3095 unhash_client_locked(clp);
3096out:
3097 spin_unlock(&nn->client_lock);
3098 if (clp)
3099 expire_client(clp);
3100 return status;
3101}
3102
3103__be32
3104nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3105{
3106 __be32 status = 0;
3107
3108 if (rc->rca_one_fs) {
3109 if (!cstate->current_fh.fh_dentry)
3110 return nfserr_nofilehandle;
3111 /*
3112 * We don't take advantage of the rca_one_fs case.
3113 * That's OK, it's optional, we can safely ignore it.
3114 */
3115 return nfs_ok;
3116 }
3117
3118 status = nfserr_complete_already;
3119 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3120 &cstate->session->se_client->cl_flags))
3121 goto out;
3122
3123 status = nfserr_stale_clientid;
3124 if (is_client_expired(cstate->session->se_client))
3125 /*
3126 * The following error isn't really legal.
3127 * But we only get here if the client just explicitly
3128 * destroyed the client. Surely it no longer cares what
3129 * error it gets back on an operation for the dead
3130 * client.
3131 */
3132 goto out;
3133
3134 status = nfs_ok;
3135 nfsd4_client_record_create(cstate->session->se_client);
3136out:
3137 return status;
3138}
3139
3140__be32
3141nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3142 struct nfsd4_setclientid *setclid)
3143{
3144 struct xdr_netobj clname = setclid->se_name;
3145 nfs4_verifier clverifier = setclid->se_verf;
3146 struct nfs4_client *conf, *new;
3147 struct nfs4_client *unconf = NULL;
3148 __be32 status;
3149 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3150
3151 new = create_client(clname, rqstp, &clverifier);
3152 if (new == NULL)
3153 return nfserr_jukebox;
3154 /* Cases below refer to rfc 3530 section 14.2.33: */
3155 spin_lock(&nn->client_lock);
3156 conf = find_confirmed_client_by_name(&clname, nn);
3157 if (conf && client_has_state(conf)) {
3158 /* case 0: */
3159 status = nfserr_clid_inuse;
3160 if (clp_used_exchangeid(conf))
3161 goto out;
3162 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3163 char addr_str[INET6_ADDRSTRLEN];
3164 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3165 sizeof(addr_str));
3166 dprintk("NFSD: setclientid: string in use by client "
3167 "at %s\n", addr_str);
3168 goto out;
3169 }
3170 }
3171 unconf = find_unconfirmed_client_by_name(&clname, nn);
3172 if (unconf)
3173 unhash_client_locked(unconf);
3174 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3175 /* case 1: probable callback update */
3176 copy_clid(new, conf);
3177 gen_confirm(new, nn);
3178 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3179 gen_clid(new, nn);
3180 new->cl_minorversion = 0;
3181 gen_callback(new, setclid, rqstp);
3182 add_to_unconfirmed(new);
3183 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3184 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3185 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3186 new = NULL;
3187 status = nfs_ok;
3188out:
3189 spin_unlock(&nn->client_lock);
3190 if (new)
3191 free_client(new);
3192 if (unconf)
3193 expire_client(unconf);
3194 return status;
3195}
3196
3197
3198__be32
3199nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3200 struct nfsd4_compound_state *cstate,
3201 struct nfsd4_setclientid_confirm *setclientid_confirm)
3202{
3203 struct nfs4_client *conf, *unconf;
3204 struct nfs4_client *old = NULL;
3205 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3206 clientid_t * clid = &setclientid_confirm->sc_clientid;
3207 __be32 status;
3208 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3209
3210 if (STALE_CLIENTID(clid, nn))
3211 return nfserr_stale_clientid;
3212
3213 spin_lock(&nn->client_lock);
3214 conf = find_confirmed_client(clid, false, nn);
3215 unconf = find_unconfirmed_client(clid, false, nn);
3216 /*
3217 * We try hard to give out unique clientid's, so if we get an
3218 * attempt to confirm the same clientid with a different cred,
3219 * the client may be buggy; this should never happen.
3220 *
3221 * Nevertheless, RFC 7530 recommends INUSE for this case:
3222 */
3223 status = nfserr_clid_inuse;
3224 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3225 goto out;
3226 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3227 goto out;
3228 /* cases below refer to rfc 3530 section 14.2.34: */
3229 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3230 if (conf && !unconf) /* case 2: probable retransmit */
3231 status = nfs_ok;
3232 else /* case 4: client hasn't noticed we rebooted yet? */
3233 status = nfserr_stale_clientid;
3234 goto out;
3235 }
3236 status = nfs_ok;
3237 if (conf) { /* case 1: callback update */
3238 old = unconf;
3239 unhash_client_locked(old);
3240 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3241 } else { /* case 3: normal case; new or rebooted client */
3242 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3243 if (old) {
3244 status = nfserr_clid_inuse;
3245 if (client_has_state(old)
3246 && !same_creds(&unconf->cl_cred,
3247 &old->cl_cred))
3248 goto out;
3249 status = mark_client_expired_locked(old);
3250 if (status) {
3251 old = NULL;
3252 goto out;
3253 }
3254 }
3255 move_to_confirmed(unconf);
3256 conf = unconf;
3257 }
3258 get_client_locked(conf);
3259 spin_unlock(&nn->client_lock);
3260 nfsd4_probe_callback(conf);
3261 spin_lock(&nn->client_lock);
3262 put_client_renew_locked(conf);
3263out:
3264 spin_unlock(&nn->client_lock);
3265 if (old)
3266 expire_client(old);
3267 return status;
3268}
3269
3270static struct nfs4_file *nfsd4_alloc_file(void)
3271{
3272 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3273}
3274
3275/* OPEN Share state helper functions */
3276static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3277 struct nfs4_file *fp)
3278{
3279 lockdep_assert_held(&state_lock);
3280
3281 atomic_set(&fp->fi_ref, 1);
3282 spin_lock_init(&fp->fi_lock);
3283 INIT_LIST_HEAD(&fp->fi_stateids);
3284 INIT_LIST_HEAD(&fp->fi_delegations);
3285 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3286 fh_copy_shallow(&fp->fi_fhandle, fh);
3287 fp->fi_deleg_file = NULL;
3288 fp->fi_had_conflict = false;
3289 fp->fi_share_deny = 0;
3290 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3291 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3292#ifdef CONFIG_NFSD_PNFS
3293 INIT_LIST_HEAD(&fp->fi_lo_states);
3294 atomic_set(&fp->fi_lo_recalls, 0);
3295#endif
3296 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3297}
3298
3299void
3300nfsd4_free_slabs(void)
3301{
3302 kmem_cache_destroy(odstate_slab);
3303 kmem_cache_destroy(openowner_slab);
3304 kmem_cache_destroy(lockowner_slab);
3305 kmem_cache_destroy(file_slab);
3306 kmem_cache_destroy(stateid_slab);
3307 kmem_cache_destroy(deleg_slab);
3308}
3309
3310int
3311nfsd4_init_slabs(void)
3312{
3313 openowner_slab = kmem_cache_create("nfsd4_openowners",
3314 sizeof(struct nfs4_openowner), 0, 0, NULL);
3315 if (openowner_slab == NULL)
3316 goto out;
3317 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3318 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3319 if (lockowner_slab == NULL)
3320 goto out_free_openowner_slab;
3321 file_slab = kmem_cache_create("nfsd4_files",
3322 sizeof(struct nfs4_file), 0, 0, NULL);
3323 if (file_slab == NULL)
3324 goto out_free_lockowner_slab;
3325 stateid_slab = kmem_cache_create("nfsd4_stateids",
3326 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3327 if (stateid_slab == NULL)
3328 goto out_free_file_slab;
3329 deleg_slab = kmem_cache_create("nfsd4_delegations",
3330 sizeof(struct nfs4_delegation), 0, 0, NULL);
3331 if (deleg_slab == NULL)
3332 goto out_free_stateid_slab;
3333 odstate_slab = kmem_cache_create("nfsd4_odstate",
3334 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3335 if (odstate_slab == NULL)
3336 goto out_free_deleg_slab;
3337 return 0;
3338
3339out_free_deleg_slab:
3340 kmem_cache_destroy(deleg_slab);
3341out_free_stateid_slab:
3342 kmem_cache_destroy(stateid_slab);
3343out_free_file_slab:
3344 kmem_cache_destroy(file_slab);
3345out_free_lockowner_slab:
3346 kmem_cache_destroy(lockowner_slab);
3347out_free_openowner_slab:
3348 kmem_cache_destroy(openowner_slab);
3349out:
3350 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3351 return -ENOMEM;
3352}
3353
3354static void init_nfs4_replay(struct nfs4_replay *rp)
3355{
3356 rp->rp_status = nfserr_serverfault;
3357 rp->rp_buflen = 0;
3358 rp->rp_buf = rp->rp_ibuf;
3359 mutex_init(&rp->rp_mutex);
3360}
3361
3362static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3363 struct nfs4_stateowner *so)
3364{
3365 if (!nfsd4_has_session(cstate)) {
3366 mutex_lock(&so->so_replay.rp_mutex);
3367 cstate->replay_owner = nfs4_get_stateowner(so);
3368 }
3369}
3370
3371void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3372{
3373 struct nfs4_stateowner *so = cstate->replay_owner;
3374
3375 if (so != NULL) {
3376 cstate->replay_owner = NULL;
3377 mutex_unlock(&so->so_replay.rp_mutex);
3378 nfs4_put_stateowner(so);
3379 }
3380}
3381
3382static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3383{
3384 struct nfs4_stateowner *sop;
3385
3386 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3387 if (!sop)
3388 return NULL;
3389
3390 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3391 if (!sop->so_owner.data) {
3392 kmem_cache_free(slab, sop);
3393 return NULL;
3394 }
3395 sop->so_owner.len = owner->len;
3396
3397 INIT_LIST_HEAD(&sop->so_stateids);
3398 sop->so_client = clp;
3399 init_nfs4_replay(&sop->so_replay);
3400 atomic_set(&sop->so_count, 1);
3401 return sop;
3402}
3403
3404static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3405{
3406 lockdep_assert_held(&clp->cl_lock);
3407
3408 list_add(&oo->oo_owner.so_strhash,
3409 &clp->cl_ownerstr_hashtbl[strhashval]);
3410 list_add(&oo->oo_perclient, &clp->cl_openowners);
3411}
3412
3413static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3414{
3415 unhash_openowner_locked(openowner(so));
3416}
3417
3418static void nfs4_free_openowner(struct nfs4_stateowner *so)
3419{
3420 struct nfs4_openowner *oo = openowner(so);
3421
3422 kmem_cache_free(openowner_slab, oo);
3423}
3424
3425static const struct nfs4_stateowner_operations openowner_ops = {
3426 .so_unhash = nfs4_unhash_openowner,
3427 .so_free = nfs4_free_openowner,
3428};
3429
3430static struct nfs4_ol_stateid *
3431nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3432{
3433 struct nfs4_ol_stateid *local, *ret = NULL;
3434 struct nfs4_openowner *oo = open->op_openowner;
3435
3436 lockdep_assert_held(&fp->fi_lock);
3437
3438 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3439 /* ignore lock owners */
3440 if (local->st_stateowner->so_is_open_owner == 0)
3441 continue;
3442 if (local->st_stateowner == &oo->oo_owner) {
3443 ret = local;
3444 atomic_inc(&ret->st_stid.sc_count);
3445 break;
3446 }
3447 }
3448 return ret;
3449}
3450
3451static struct nfs4_openowner *
3452alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3453 struct nfsd4_compound_state *cstate)
3454{
3455 struct nfs4_client *clp = cstate->clp;
3456 struct nfs4_openowner *oo, *ret;
3457
3458 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3459 if (!oo)
3460 return NULL;
3461 oo->oo_owner.so_ops = &openowner_ops;
3462 oo->oo_owner.so_is_open_owner = 1;
3463 oo->oo_owner.so_seqid = open->op_seqid;
3464 oo->oo_flags = 0;
3465 if (nfsd4_has_session(cstate))
3466 oo->oo_flags |= NFS4_OO_CONFIRMED;
3467 oo->oo_time = 0;
3468 oo->oo_last_closed_stid = NULL;
3469 INIT_LIST_HEAD(&oo->oo_close_lru);
3470 spin_lock(&clp->cl_lock);
3471 ret = find_openstateowner_str_locked(strhashval, open, clp);
3472 if (ret == NULL) {
3473 hash_openowner(oo, clp, strhashval);
3474 ret = oo;
3475 } else
3476 nfs4_free_stateowner(&oo->oo_owner);
3477
3478 spin_unlock(&clp->cl_lock);
3479 return ret;
3480}
3481
3482static struct nfs4_ol_stateid *
3483init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp,
3484 struct nfsd4_open *open)
3485{
3486
3487 struct nfs4_openowner *oo = open->op_openowner;
3488 struct nfs4_ol_stateid *retstp = NULL;
3489
3490 spin_lock(&oo->oo_owner.so_client->cl_lock);
3491 spin_lock(&fp->fi_lock);
3492
3493 retstp = nfsd4_find_existing_open(fp, open);
3494 if (retstp)
3495 goto out_unlock;
3496 atomic_inc(&stp->st_stid.sc_count);
3497 stp->st_stid.sc_type = NFS4_OPEN_STID;
3498 INIT_LIST_HEAD(&stp->st_locks);
3499 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3500 get_nfs4_file(fp);
3501 stp->st_stid.sc_file = fp;
3502 stp->st_access_bmap = 0;
3503 stp->st_deny_bmap = 0;
3504 stp->st_openstp = NULL;
3505 init_rwsem(&stp->st_rwsem);
3506 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3507 list_add(&stp->st_perfile, &fp->fi_stateids);
3508
3509out_unlock:
3510 spin_unlock(&fp->fi_lock);
3511 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3512 return retstp;
3513}
3514
3515/*
3516 * In the 4.0 case we need to keep the owners around a little while to handle
3517 * CLOSE replay. We still do need to release any file access that is held by
3518 * them before returning however.
3519 */
3520static void
3521move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3522{
3523 struct nfs4_ol_stateid *last;
3524 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3525 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3526 nfsd_net_id);
3527
3528 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3529
3530 /*
3531 * We know that we hold one reference via nfsd4_close, and another
3532 * "persistent" reference for the client. If the refcount is higher
3533 * than 2, then there are still calls in progress that are using this
3534 * stateid. We can't put the sc_file reference until they are finished.
3535 * Wait for the refcount to drop to 2. Since it has been unhashed,
3536 * there should be no danger of the refcount going back up again at
3537 * this point.
3538 */
3539 wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3540
3541 release_all_access(s);
3542 if (s->st_stid.sc_file) {
3543 put_nfs4_file(s->st_stid.sc_file);
3544 s->st_stid.sc_file = NULL;
3545 }
3546
3547 spin_lock(&nn->client_lock);
3548 last = oo->oo_last_closed_stid;
3549 oo->oo_last_closed_stid = s;
3550 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3551 oo->oo_time = get_seconds();
3552 spin_unlock(&nn->client_lock);
3553 if (last)
3554 nfs4_put_stid(&last->st_stid);
3555}
3556
3557/* search file_hashtbl[] for file */
3558static struct nfs4_file *
3559find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3560{
3561 struct nfs4_file *fp;
3562
3563 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3564 if (fh_match(&fp->fi_fhandle, fh)) {
3565 if (atomic_inc_not_zero(&fp->fi_ref))
3566 return fp;
3567 }
3568 }
3569 return NULL;
3570}
3571
3572struct nfs4_file *
3573find_file(struct knfsd_fh *fh)
3574{
3575 struct nfs4_file *fp;
3576 unsigned int hashval = file_hashval(fh);
3577
3578 rcu_read_lock();
3579 fp = find_file_locked(fh, hashval);
3580 rcu_read_unlock();
3581 return fp;
3582}
3583
3584static struct nfs4_file *
3585find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3586{
3587 struct nfs4_file *fp;
3588 unsigned int hashval = file_hashval(fh);
3589
3590 rcu_read_lock();
3591 fp = find_file_locked(fh, hashval);
3592 rcu_read_unlock();
3593 if (fp)
3594 return fp;
3595
3596 spin_lock(&state_lock);
3597 fp = find_file_locked(fh, hashval);
3598 if (likely(fp == NULL)) {
3599 nfsd4_init_file(fh, hashval, new);
3600 fp = new;
3601 }
3602 spin_unlock(&state_lock);
3603
3604 return fp;
3605}
3606
3607/*
3608 * Called to check deny when READ with all zero stateid or
3609 * WRITE with all zero or all one stateid
3610 */
3611static __be32
3612nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3613{
3614 struct nfs4_file *fp;
3615 __be32 ret = nfs_ok;
3616
3617 fp = find_file(¤t_fh->fh_handle);
3618 if (!fp)
3619 return ret;
3620 /* Check for conflicting share reservations */
3621 spin_lock(&fp->fi_lock);
3622 if (fp->fi_share_deny & deny_type)
3623 ret = nfserr_locked;
3624 spin_unlock(&fp->fi_lock);
3625 put_nfs4_file(fp);
3626 return ret;
3627}
3628
3629static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3630{
3631 struct nfs4_delegation *dp = cb_to_delegation(cb);
3632 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3633 nfsd_net_id);
3634
3635 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3636
3637 /*
3638 * We can't do this in nfsd_break_deleg_cb because it is
3639 * already holding inode->i_lock.
3640 *
3641 * If the dl_time != 0, then we know that it has already been
3642 * queued for a lease break. Don't queue it again.
3643 */
3644 spin_lock(&state_lock);
3645 if (dp->dl_time == 0) {
3646 dp->dl_time = get_seconds();
3647 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3648 }
3649 spin_unlock(&state_lock);
3650}
3651
3652static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3653 struct rpc_task *task)
3654{
3655 struct nfs4_delegation *dp = cb_to_delegation(cb);
3656
3657 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3658 return 1;
3659
3660 switch (task->tk_status) {
3661 case 0:
3662 return 1;
3663 case -EBADHANDLE:
3664 case -NFS4ERR_BAD_STATEID:
3665 /*
3666 * Race: client probably got cb_recall before open reply
3667 * granting delegation.
3668 */
3669 if (dp->dl_retries--) {
3670 rpc_delay(task, 2 * HZ);
3671 return 0;
3672 }
3673 /*FALLTHRU*/
3674 default:
3675 return -1;
3676 }
3677}
3678
3679static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3680{
3681 struct nfs4_delegation *dp = cb_to_delegation(cb);
3682
3683 nfs4_put_stid(&dp->dl_stid);
3684}
3685
3686static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3687 .prepare = nfsd4_cb_recall_prepare,
3688 .done = nfsd4_cb_recall_done,
3689 .release = nfsd4_cb_recall_release,
3690};
3691
3692static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3693{
3694 /*
3695 * We're assuming the state code never drops its reference
3696 * without first removing the lease. Since we're in this lease
3697 * callback (and since the lease code is serialized by the kernel
3698 * lock) we know the server hasn't removed the lease yet, we know
3699 * it's safe to take a reference.
3700 */
3701 atomic_inc(&dp->dl_stid.sc_count);
3702 nfsd4_run_cb(&dp->dl_recall);
3703}
3704
3705/* Called from break_lease() with i_lock held. */
3706static bool
3707nfsd_break_deleg_cb(struct file_lock *fl)
3708{
3709 bool ret = false;
3710 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3711 struct nfs4_delegation *dp;
3712
3713 if (!fp) {
3714 WARN(1, "(%p)->fl_owner NULL\n", fl);
3715 return ret;
3716 }
3717 if (fp->fi_had_conflict) {
3718 WARN(1, "duplicate break on %p\n", fp);
3719 return ret;
3720 }
3721 /*
3722 * We don't want the locks code to timeout the lease for us;
3723 * we'll remove it ourself if a delegation isn't returned
3724 * in time:
3725 */
3726 fl->fl_break_time = 0;
3727
3728 spin_lock(&fp->fi_lock);
3729 fp->fi_had_conflict = true;
3730 /*
3731 * If there are no delegations on the list, then return true
3732 * so that the lease code will go ahead and delete it.
3733 */
3734 if (list_empty(&fp->fi_delegations))
3735 ret = true;
3736 else
3737 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3738 nfsd_break_one_deleg(dp);
3739 spin_unlock(&fp->fi_lock);
3740 return ret;
3741}
3742
3743static int
3744nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3745 struct list_head *dispose)
3746{
3747 if (arg & F_UNLCK)
3748 return lease_modify(onlist, arg, dispose);
3749 else
3750 return -EAGAIN;
3751}
3752
3753static const struct lock_manager_operations nfsd_lease_mng_ops = {
3754 .lm_break = nfsd_break_deleg_cb,
3755 .lm_change = nfsd_change_deleg_cb,
3756};
3757
3758static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3759{
3760 if (nfsd4_has_session(cstate))
3761 return nfs_ok;
3762 if (seqid == so->so_seqid - 1)
3763 return nfserr_replay_me;
3764 if (seqid == so->so_seqid)
3765 return nfs_ok;
3766 return nfserr_bad_seqid;
3767}
3768
3769static __be32 lookup_clientid(clientid_t *clid,
3770 struct nfsd4_compound_state *cstate,
3771 struct nfsd_net *nn)
3772{
3773 struct nfs4_client *found;
3774
3775 if (cstate->clp) {
3776 found = cstate->clp;
3777 if (!same_clid(&found->cl_clientid, clid))
3778 return nfserr_stale_clientid;
3779 return nfs_ok;
3780 }
3781
3782 if (STALE_CLIENTID(clid, nn))
3783 return nfserr_stale_clientid;
3784
3785 /*
3786 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3787 * cached already then we know this is for is for v4.0 and "sessions"
3788 * will be false.
3789 */
3790 WARN_ON_ONCE(cstate->session);
3791 spin_lock(&nn->client_lock);
3792 found = find_confirmed_client(clid, false, nn);
3793 if (!found) {
3794 spin_unlock(&nn->client_lock);
3795 return nfserr_expired;
3796 }
3797 atomic_inc(&found->cl_refcount);
3798 spin_unlock(&nn->client_lock);
3799
3800 /* Cache the nfs4_client in cstate! */
3801 cstate->clp = found;
3802 return nfs_ok;
3803}
3804
3805__be32
3806nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3807 struct nfsd4_open *open, struct nfsd_net *nn)
3808{
3809 clientid_t *clientid = &open->op_clientid;
3810 struct nfs4_client *clp = NULL;
3811 unsigned int strhashval;
3812 struct nfs4_openowner *oo = NULL;
3813 __be32 status;
3814
3815 if (STALE_CLIENTID(&open->op_clientid, nn))
3816 return nfserr_stale_clientid;
3817 /*
3818 * In case we need it later, after we've already created the
3819 * file and don't want to risk a further failure:
3820 */
3821 open->op_file = nfsd4_alloc_file();
3822 if (open->op_file == NULL)
3823 return nfserr_jukebox;
3824
3825 status = lookup_clientid(clientid, cstate, nn);
3826 if (status)
3827 return status;
3828 clp = cstate->clp;
3829
3830 strhashval = ownerstr_hashval(&open->op_owner);
3831 oo = find_openstateowner_str(strhashval, open, clp);
3832 open->op_openowner = oo;
3833 if (!oo) {
3834 goto new_owner;
3835 }
3836 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3837 /* Replace unconfirmed owners without checking for replay. */
3838 release_openowner(oo);
3839 open->op_openowner = NULL;
3840 goto new_owner;
3841 }
3842 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3843 if (status)
3844 return status;
3845 goto alloc_stateid;
3846new_owner:
3847 oo = alloc_init_open_stateowner(strhashval, open, cstate);
3848 if (oo == NULL)
3849 return nfserr_jukebox;
3850 open->op_openowner = oo;
3851alloc_stateid:
3852 open->op_stp = nfs4_alloc_open_stateid(clp);
3853 if (!open->op_stp)
3854 return nfserr_jukebox;
3855
3856 if (nfsd4_has_session(cstate) &&
3857 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3858 open->op_odstate = alloc_clnt_odstate(clp);
3859 if (!open->op_odstate)
3860 return nfserr_jukebox;
3861 }
3862
3863 return nfs_ok;
3864}
3865
3866static inline __be32
3867nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3868{
3869 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3870 return nfserr_openmode;
3871 else
3872 return nfs_ok;
3873}
3874
3875static int share_access_to_flags(u32 share_access)
3876{
3877 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3878}
3879
3880static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3881{
3882 struct nfs4_stid *ret;
3883
3884 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3885 if (!ret)
3886 return NULL;
3887 return delegstateid(ret);
3888}
3889
3890static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3891{
3892 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3893 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3894}
3895
3896static __be32
3897nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3898 struct nfs4_delegation **dp)
3899{
3900 int flags;
3901 __be32 status = nfserr_bad_stateid;
3902 struct nfs4_delegation *deleg;
3903
3904 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3905 if (deleg == NULL)
3906 goto out;
3907 flags = share_access_to_flags(open->op_share_access);
3908 status = nfs4_check_delegmode(deleg, flags);
3909 if (status) {
3910 nfs4_put_stid(&deleg->dl_stid);
3911 goto out;
3912 }
3913 *dp = deleg;
3914out:
3915 if (!nfsd4_is_deleg_cur(open))
3916 return nfs_ok;
3917 if (status)
3918 return status;
3919 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3920 return nfs_ok;
3921}
3922
3923static inline int nfs4_access_to_access(u32 nfs4_access)
3924{
3925 int flags = 0;
3926
3927 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3928 flags |= NFSD_MAY_READ;
3929 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3930 flags |= NFSD_MAY_WRITE;
3931 return flags;
3932}
3933
3934static inline __be32
3935nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3936 struct nfsd4_open *open)
3937{
3938 struct iattr iattr = {
3939 .ia_valid = ATTR_SIZE,
3940 .ia_size = 0,
3941 };
3942 if (!open->op_truncate)
3943 return 0;
3944 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3945 return nfserr_inval;
3946 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3947}
3948
3949static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3950 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3951 struct nfsd4_open *open)
3952{
3953 struct file *filp = NULL;
3954 __be32 status;
3955 int oflag = nfs4_access_to_omode(open->op_share_access);
3956 int access = nfs4_access_to_access(open->op_share_access);
3957 unsigned char old_access_bmap, old_deny_bmap;
3958
3959 spin_lock(&fp->fi_lock);
3960
3961 /*
3962 * Are we trying to set a deny mode that would conflict with
3963 * current access?
3964 */
3965 status = nfs4_file_check_deny(fp, open->op_share_deny);
3966 if (status != nfs_ok) {
3967 spin_unlock(&fp->fi_lock);
3968 goto out;
3969 }
3970
3971 /* set access to the file */
3972 status = nfs4_file_get_access(fp, open->op_share_access);
3973 if (status != nfs_ok) {
3974 spin_unlock(&fp->fi_lock);
3975 goto out;
3976 }
3977
3978 /* Set access bits in stateid */
3979 old_access_bmap = stp->st_access_bmap;
3980 set_access(open->op_share_access, stp);
3981
3982 /* Set new deny mask */
3983 old_deny_bmap = stp->st_deny_bmap;
3984 set_deny(open->op_share_deny, stp);
3985 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3986
3987 if (!fp->fi_fds[oflag]) {
3988 spin_unlock(&fp->fi_lock);
3989 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3990 if (status)
3991 goto out_put_access;
3992 spin_lock(&fp->fi_lock);
3993 if (!fp->fi_fds[oflag]) {
3994 fp->fi_fds[oflag] = filp;
3995 filp = NULL;
3996 }
3997 }
3998 spin_unlock(&fp->fi_lock);
3999 if (filp)
4000 fput(filp);
4001
4002 status = nfsd4_truncate(rqstp, cur_fh, open);
4003 if (status)
4004 goto out_put_access;
4005out:
4006 return status;
4007out_put_access:
4008 stp->st_access_bmap = old_access_bmap;
4009 nfs4_file_put_access(fp, open->op_share_access);
4010 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4011 goto out;
4012}
4013
4014static __be32
4015nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4016{
4017 __be32 status;
4018 unsigned char old_deny_bmap = stp->st_deny_bmap;
4019
4020 if (!test_access(open->op_share_access, stp))
4021 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4022
4023 /* test and set deny mode */
4024 spin_lock(&fp->fi_lock);
4025 status = nfs4_file_check_deny(fp, open->op_share_deny);
4026 if (status == nfs_ok) {
4027 set_deny(open->op_share_deny, stp);
4028 fp->fi_share_deny |=
4029 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4030 }
4031 spin_unlock(&fp->fi_lock);
4032
4033 if (status != nfs_ok)
4034 return status;
4035
4036 status = nfsd4_truncate(rqstp, cur_fh, open);
4037 if (status != nfs_ok)
4038 reset_union_bmap_deny(old_deny_bmap, stp);
4039 return status;
4040}
4041
4042/* Should we give out recallable state?: */
4043static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4044{
4045 if (clp->cl_cb_state == NFSD4_CB_UP)
4046 return true;
4047 /*
4048 * In the sessions case, since we don't have to establish a
4049 * separate connection for callbacks, we assume it's OK
4050 * until we hear otherwise:
4051 */
4052 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4053}
4054
4055static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4056{
4057 struct file_lock *fl;
4058
4059 fl = locks_alloc_lock();
4060 if (!fl)
4061 return NULL;
4062 fl->fl_lmops = &nfsd_lease_mng_ops;
4063 fl->fl_flags = FL_DELEG;
4064 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4065 fl->fl_end = OFFSET_MAX;
4066 fl->fl_owner = (fl_owner_t)fp;
4067 fl->fl_pid = current->tgid;
4068 return fl;
4069}
4070
4071/**
4072 * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4073 * @dp: a pointer to the nfs4_delegation we're adding.
4074 *
4075 * Return:
4076 * On success: Return code will be 0 on success.
4077 *
4078 * On error: -EAGAIN if there was an existing delegation.
4079 * nonzero if there is an error in other cases.
4080 *
4081 */
4082
4083static int nfs4_setlease(struct nfs4_delegation *dp)
4084{
4085 struct nfs4_file *fp = dp->dl_stid.sc_file;
4086 struct file_lock *fl;
4087 struct file *filp;
4088 int status = 0;
4089
4090 fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4091 if (!fl)
4092 return -ENOMEM;
4093 filp = find_readable_file(fp);
4094 if (!filp) {
4095 /* We should always have a readable file here */
4096 WARN_ON_ONCE(1);
4097 locks_free_lock(fl);
4098 return -EBADF;
4099 }
4100 fl->fl_file = filp;
4101 status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4102 if (fl)
4103 locks_free_lock(fl);
4104 if (status)
4105 goto out_fput;
4106 spin_lock(&state_lock);
4107 spin_lock(&fp->fi_lock);
4108 /* Did the lease get broken before we took the lock? */
4109 status = -EAGAIN;
4110 if (fp->fi_had_conflict)
4111 goto out_unlock;
4112 /* Race breaker */
4113 if (fp->fi_deleg_file) {
4114 status = hash_delegation_locked(dp, fp);
4115 goto out_unlock;
4116 }
4117 fp->fi_deleg_file = filp;
4118 fp->fi_delegees = 0;
4119 status = hash_delegation_locked(dp, fp);
4120 spin_unlock(&fp->fi_lock);
4121 spin_unlock(&state_lock);
4122 if (status) {
4123 /* Should never happen, this is a new fi_deleg_file */
4124 WARN_ON_ONCE(1);
4125 goto out_fput;
4126 }
4127 return 0;
4128out_unlock:
4129 spin_unlock(&fp->fi_lock);
4130 spin_unlock(&state_lock);
4131out_fput:
4132 fput(filp);
4133 return status;
4134}
4135
4136static struct nfs4_delegation *
4137nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4138 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4139{
4140 int status;
4141 struct nfs4_delegation *dp;
4142
4143 if (fp->fi_had_conflict)
4144 return ERR_PTR(-EAGAIN);
4145
4146 spin_lock(&state_lock);
4147 spin_lock(&fp->fi_lock);
4148 status = nfs4_get_existing_delegation(clp, fp);
4149 spin_unlock(&fp->fi_lock);
4150 spin_unlock(&state_lock);
4151
4152 if (status)
4153 return ERR_PTR(status);
4154
4155 dp = alloc_init_deleg(clp, fh, odstate);
4156 if (!dp)
4157 return ERR_PTR(-ENOMEM);
4158
4159 get_nfs4_file(fp);
4160 spin_lock(&state_lock);
4161 spin_lock(&fp->fi_lock);
4162 dp->dl_stid.sc_file = fp;
4163 if (!fp->fi_deleg_file) {
4164 spin_unlock(&fp->fi_lock);
4165 spin_unlock(&state_lock);
4166 status = nfs4_setlease(dp);
4167 goto out;
4168 }
4169 if (fp->fi_had_conflict) {
4170 status = -EAGAIN;
4171 goto out_unlock;
4172 }
4173 status = hash_delegation_locked(dp, fp);
4174out_unlock:
4175 spin_unlock(&fp->fi_lock);
4176 spin_unlock(&state_lock);
4177out:
4178 if (status) {
4179 put_clnt_odstate(dp->dl_clnt_odstate);
4180 nfs4_put_stid(&dp->dl_stid);
4181 return ERR_PTR(status);
4182 }
4183 return dp;
4184}
4185
4186static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4187{
4188 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4189 if (status == -EAGAIN)
4190 open->op_why_no_deleg = WND4_CONTENTION;
4191 else {
4192 open->op_why_no_deleg = WND4_RESOURCE;
4193 switch (open->op_deleg_want) {
4194 case NFS4_SHARE_WANT_READ_DELEG:
4195 case NFS4_SHARE_WANT_WRITE_DELEG:
4196 case NFS4_SHARE_WANT_ANY_DELEG:
4197 break;
4198 case NFS4_SHARE_WANT_CANCEL:
4199 open->op_why_no_deleg = WND4_CANCELLED;
4200 break;
4201 case NFS4_SHARE_WANT_NO_DELEG:
4202 WARN_ON_ONCE(1);
4203 }
4204 }
4205}
4206
4207/*
4208 * Attempt to hand out a delegation.
4209 *
4210 * Note we don't support write delegations, and won't until the vfs has
4211 * proper support for them.
4212 */
4213static void
4214nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4215 struct nfs4_ol_stateid *stp)
4216{
4217 struct nfs4_delegation *dp;
4218 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4219 struct nfs4_client *clp = stp->st_stid.sc_client;
4220 int cb_up;
4221 int status = 0;
4222
4223 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4224 open->op_recall = 0;
4225 switch (open->op_claim_type) {
4226 case NFS4_OPEN_CLAIM_PREVIOUS:
4227 if (!cb_up)
4228 open->op_recall = 1;
4229 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4230 goto out_no_deleg;
4231 break;
4232 case NFS4_OPEN_CLAIM_NULL:
4233 case NFS4_OPEN_CLAIM_FH:
4234 /*
4235 * Let's not give out any delegations till everyone's
4236 * had the chance to reclaim theirs, *and* until
4237 * NLM locks have all been reclaimed:
4238 */
4239 if (locks_in_grace(clp->net))
4240 goto out_no_deleg;
4241 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4242 goto out_no_deleg;
4243 /*
4244 * Also, if the file was opened for write or
4245 * create, there's a good chance the client's
4246 * about to write to it, resulting in an
4247 * immediate recall (since we don't support
4248 * write delegations):
4249 */
4250 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4251 goto out_no_deleg;
4252 if (open->op_create == NFS4_OPEN_CREATE)
4253 goto out_no_deleg;
4254 break;
4255 default:
4256 goto out_no_deleg;
4257 }
4258 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4259 if (IS_ERR(dp))
4260 goto out_no_deleg;
4261
4262 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4263
4264 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4265 STATEID_VAL(&dp->dl_stid.sc_stateid));
4266 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4267 nfs4_put_stid(&dp->dl_stid);
4268 return;
4269out_no_deleg:
4270 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4271 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4272 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4273 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4274 open->op_recall = 1;
4275 }
4276
4277 /* 4.1 client asking for a delegation? */
4278 if (open->op_deleg_want)
4279 nfsd4_open_deleg_none_ext(open, status);
4280 return;
4281}
4282
4283static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4284 struct nfs4_delegation *dp)
4285{
4286 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4287 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4288 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4289 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4290 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4291 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4292 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4293 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4294 }
4295 /* Otherwise the client must be confused wanting a delegation
4296 * it already has, therefore we don't return
4297 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4298 */
4299}
4300
4301__be32
4302nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4303{
4304 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4305 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4306 struct nfs4_file *fp = NULL;
4307 struct nfs4_ol_stateid *stp = NULL;
4308 struct nfs4_ol_stateid *swapstp = NULL;
4309 struct nfs4_delegation *dp = NULL;
4310 __be32 status;
4311
4312 /*
4313 * Lookup file; if found, lookup stateid and check open request,
4314 * and check for delegations in the process of being recalled.
4315 * If not found, create the nfs4_file struct
4316 */
4317 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
4318 if (fp != open->op_file) {
4319 status = nfs4_check_deleg(cl, open, &dp);
4320 if (status)
4321 goto out;
4322 spin_lock(&fp->fi_lock);
4323 stp = nfsd4_find_existing_open(fp, open);
4324 spin_unlock(&fp->fi_lock);
4325 } else {
4326 open->op_file = NULL;
4327 status = nfserr_bad_stateid;
4328 if (nfsd4_is_deleg_cur(open))
4329 goto out;
4330 }
4331
4332 /*
4333 * OPEN the file, or upgrade an existing OPEN.
4334 * If truncate fails, the OPEN fails.
4335 */
4336 if (stp) {
4337 /* Stateid was found, this is an OPEN upgrade */
4338 down_read(&stp->st_rwsem);
4339 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4340 if (status) {
4341 up_read(&stp->st_rwsem);
4342 goto out;
4343 }
4344 } else {
4345 stp = open->op_stp;
4346 open->op_stp = NULL;
4347 swapstp = init_open_stateid(stp, fp, open);
4348 if (swapstp) {
4349 nfs4_put_stid(&stp->st_stid);
4350 stp = swapstp;
4351 down_read(&stp->st_rwsem);
4352 status = nfs4_upgrade_open(rqstp, fp, current_fh,
4353 stp, open);
4354 if (status) {
4355 up_read(&stp->st_rwsem);
4356 goto out;
4357 }
4358 goto upgrade_out;
4359 }
4360 down_read(&stp->st_rwsem);
4361 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4362 if (status) {
4363 up_read(&stp->st_rwsem);
4364 release_open_stateid(stp);
4365 goto out;
4366 }
4367
4368 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4369 open->op_odstate);
4370 if (stp->st_clnt_odstate == open->op_odstate)
4371 open->op_odstate = NULL;
4372 }
4373upgrade_out:
4374 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4375 up_read(&stp->st_rwsem);
4376
4377 if (nfsd4_has_session(&resp->cstate)) {
4378 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4379 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4380 open->op_why_no_deleg = WND4_NOT_WANTED;
4381 goto nodeleg;
4382 }
4383 }
4384
4385 /*
4386 * Attempt to hand out a delegation. No error return, because the
4387 * OPEN succeeds even if we fail.
4388 */
4389 nfs4_open_delegation(current_fh, open, stp);
4390nodeleg:
4391 status = nfs_ok;
4392
4393 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4394 STATEID_VAL(&stp->st_stid.sc_stateid));
4395out:
4396 /* 4.1 client trying to upgrade/downgrade delegation? */
4397 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4398 open->op_deleg_want)
4399 nfsd4_deleg_xgrade_none_ext(open, dp);
4400
4401 if (fp)
4402 put_nfs4_file(fp);
4403 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4404 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4405 /*
4406 * To finish the open response, we just need to set the rflags.
4407 */
4408 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4409 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4410 !nfsd4_has_session(&resp->cstate))
4411 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4412 if (dp)
4413 nfs4_put_stid(&dp->dl_stid);
4414 if (stp)
4415 nfs4_put_stid(&stp->st_stid);
4416
4417 return status;
4418}
4419
4420void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4421 struct nfsd4_open *open)
4422{
4423 if (open->op_openowner) {
4424 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4425
4426 nfsd4_cstate_assign_replay(cstate, so);
4427 nfs4_put_stateowner(so);
4428 }
4429 if (open->op_file)
4430 kmem_cache_free(file_slab, open->op_file);
4431 if (open->op_stp)
4432 nfs4_put_stid(&open->op_stp->st_stid);
4433 if (open->op_odstate)
4434 kmem_cache_free(odstate_slab, open->op_odstate);
4435}
4436
4437__be32
4438nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4439 clientid_t *clid)
4440{
4441 struct nfs4_client *clp;
4442 __be32 status;
4443 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4444
4445 dprintk("process_renew(%08x/%08x): starting\n",
4446 clid->cl_boot, clid->cl_id);
4447 status = lookup_clientid(clid, cstate, nn);
4448 if (status)
4449 goto out;
4450 clp = cstate->clp;
4451 status = nfserr_cb_path_down;
4452 if (!list_empty(&clp->cl_delegations)
4453 && clp->cl_cb_state != NFSD4_CB_UP)
4454 goto out;
4455 status = nfs_ok;
4456out:
4457 return status;
4458}
4459
4460void
4461nfsd4_end_grace(struct nfsd_net *nn)
4462{
4463 /* do nothing if grace period already ended */
4464 if (nn->grace_ended)
4465 return;
4466
4467 dprintk("NFSD: end of grace period\n");
4468 nn->grace_ended = true;
4469 /*
4470 * If the server goes down again right now, an NFSv4
4471 * client will still be allowed to reclaim after it comes back up,
4472 * even if it hasn't yet had a chance to reclaim state this time.
4473 *
4474 */
4475 nfsd4_record_grace_done(nn);
4476 /*
4477 * At this point, NFSv4 clients can still reclaim. But if the
4478 * server crashes, any that have not yet reclaimed will be out
4479 * of luck on the next boot.
4480 *
4481 * (NFSv4.1+ clients are considered to have reclaimed once they
4482 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4483 * have reclaimed after their first OPEN.)
4484 */
4485 locks_end_grace(&nn->nfsd4_manager);
4486 /*
4487 * At this point, and once lockd and/or any other containers
4488 * exit their grace period, further reclaims will fail and
4489 * regular locking can resume.
4490 */
4491}
4492
4493static time_t
4494nfs4_laundromat(struct nfsd_net *nn)
4495{
4496 struct nfs4_client *clp;
4497 struct nfs4_openowner *oo;
4498 struct nfs4_delegation *dp;
4499 struct nfs4_ol_stateid *stp;
4500 struct list_head *pos, *next, reaplist;
4501 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4502 time_t t, new_timeo = nn->nfsd4_lease;
4503
4504 dprintk("NFSD: laundromat service - starting\n");
4505 nfsd4_end_grace(nn);
4506 INIT_LIST_HEAD(&reaplist);
4507 spin_lock(&nn->client_lock);
4508 list_for_each_safe(pos, next, &nn->client_lru) {
4509 clp = list_entry(pos, struct nfs4_client, cl_lru);
4510 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4511 t = clp->cl_time - cutoff;
4512 new_timeo = min(new_timeo, t);
4513 break;
4514 }
4515 if (mark_client_expired_locked(clp)) {
4516 dprintk("NFSD: client in use (clientid %08x)\n",
4517 clp->cl_clientid.cl_id);
4518 continue;
4519 }
4520 list_add(&clp->cl_lru, &reaplist);
4521 }
4522 spin_unlock(&nn->client_lock);
4523 list_for_each_safe(pos, next, &reaplist) {
4524 clp = list_entry(pos, struct nfs4_client, cl_lru);
4525 dprintk("NFSD: purging unused client (clientid %08x)\n",
4526 clp->cl_clientid.cl_id);
4527 list_del_init(&clp->cl_lru);
4528 expire_client(clp);
4529 }
4530 spin_lock(&state_lock);
4531 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4532 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4533 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4534 t = dp->dl_time - cutoff;
4535 new_timeo = min(new_timeo, t);
4536 break;
4537 }
4538 WARN_ON(!unhash_delegation_locked(dp));
4539 list_add(&dp->dl_recall_lru, &reaplist);
4540 }
4541 spin_unlock(&state_lock);
4542 while (!list_empty(&reaplist)) {
4543 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4544 dl_recall_lru);
4545 list_del_init(&dp->dl_recall_lru);
4546 revoke_delegation(dp);
4547 }
4548
4549 spin_lock(&nn->client_lock);
4550 while (!list_empty(&nn->close_lru)) {
4551 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4552 oo_close_lru);
4553 if (time_after((unsigned long)oo->oo_time,
4554 (unsigned long)cutoff)) {
4555 t = oo->oo_time - cutoff;
4556 new_timeo = min(new_timeo, t);
4557 break;
4558 }
4559 list_del_init(&oo->oo_close_lru);
4560 stp = oo->oo_last_closed_stid;
4561 oo->oo_last_closed_stid = NULL;
4562 spin_unlock(&nn->client_lock);
4563 nfs4_put_stid(&stp->st_stid);
4564 spin_lock(&nn->client_lock);
4565 }
4566 spin_unlock(&nn->client_lock);
4567
4568 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4569 return new_timeo;
4570}
4571
4572static struct workqueue_struct *laundry_wq;
4573static void laundromat_main(struct work_struct *);
4574
4575static void
4576laundromat_main(struct work_struct *laundry)
4577{
4578 time_t t;
4579 struct delayed_work *dwork = to_delayed_work(laundry);
4580 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4581 laundromat_work);
4582
4583 t = nfs4_laundromat(nn);
4584 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4585 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4586}
4587
4588static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4589{
4590 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4591 return nfserr_bad_stateid;
4592 return nfs_ok;
4593}
4594
4595static inline int
4596access_permit_read(struct nfs4_ol_stateid *stp)
4597{
4598 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4599 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4600 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4601}
4602
4603static inline int
4604access_permit_write(struct nfs4_ol_stateid *stp)
4605{
4606 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4607 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4608}
4609
4610static
4611__be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4612{
4613 __be32 status = nfserr_openmode;
4614
4615 /* For lock stateid's, we test the parent open, not the lock: */
4616 if (stp->st_openstp)
4617 stp = stp->st_openstp;
4618 if ((flags & WR_STATE) && !access_permit_write(stp))
4619 goto out;
4620 if ((flags & RD_STATE) && !access_permit_read(stp))
4621 goto out;
4622 status = nfs_ok;
4623out:
4624 return status;
4625}
4626
4627static inline __be32
4628check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4629{
4630 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4631 return nfs_ok;
4632 else if (opens_in_grace(net)) {
4633 /* Answer in remaining cases depends on existence of
4634 * conflicting state; so we must wait out the grace period. */
4635 return nfserr_grace;
4636 } else if (flags & WR_STATE)
4637 return nfs4_share_conflict(current_fh,
4638 NFS4_SHARE_DENY_WRITE);
4639 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4640 return nfs4_share_conflict(current_fh,
4641 NFS4_SHARE_DENY_READ);
4642}
4643
4644/*
4645 * Allow READ/WRITE during grace period on recovered state only for files
4646 * that are not able to provide mandatory locking.
4647 */
4648static inline int
4649grace_disallows_io(struct net *net, struct inode *inode)
4650{
4651 return opens_in_grace(net) && mandatory_lock(inode);
4652}
4653
4654/* Returns true iff a is later than b: */
4655static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4656{
4657 return (s32)(a->si_generation - b->si_generation) > 0;
4658}
4659
4660static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4661{
4662 /*
4663 * When sessions are used the stateid generation number is ignored
4664 * when it is zero.
4665 */
4666 if (has_session && in->si_generation == 0)
4667 return nfs_ok;
4668
4669 if (in->si_generation == ref->si_generation)
4670 return nfs_ok;
4671
4672 /* If the client sends us a stateid from the future, it's buggy: */
4673 if (stateid_generation_after(in, ref))
4674 return nfserr_bad_stateid;
4675 /*
4676 * However, we could see a stateid from the past, even from a
4677 * non-buggy client. For example, if the client sends a lock
4678 * while some IO is outstanding, the lock may bump si_generation
4679 * while the IO is still in flight. The client could avoid that
4680 * situation by waiting for responses on all the IO requests,
4681 * but better performance may result in retrying IO that
4682 * receives an old_stateid error if requests are rarely
4683 * reordered in flight:
4684 */
4685 return nfserr_old_stateid;
4686}
4687
4688static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4689{
4690 if (ols->st_stateowner->so_is_open_owner &&
4691 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4692 return nfserr_bad_stateid;
4693 return nfs_ok;
4694}
4695
4696static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4697{
4698 struct nfs4_stid *s;
4699 __be32 status = nfserr_bad_stateid;
4700
4701 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4702 return status;
4703 /* Client debugging aid. */
4704 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4705 char addr_str[INET6_ADDRSTRLEN];
4706 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4707 sizeof(addr_str));
4708 pr_warn_ratelimited("NFSD: client %s testing state ID "
4709 "with incorrect client ID\n", addr_str);
4710 return status;
4711 }
4712 spin_lock(&cl->cl_lock);
4713 s = find_stateid_locked(cl, stateid);
4714 if (!s)
4715 goto out_unlock;
4716 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4717 if (status)
4718 goto out_unlock;
4719 switch (s->sc_type) {
4720 case NFS4_DELEG_STID:
4721 status = nfs_ok;
4722 break;
4723 case NFS4_REVOKED_DELEG_STID:
4724 status = nfserr_deleg_revoked;
4725 break;
4726 case NFS4_OPEN_STID:
4727 case NFS4_LOCK_STID:
4728 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4729 break;
4730 default:
4731 printk("unknown stateid type %x\n", s->sc_type);
4732 /* Fallthrough */
4733 case NFS4_CLOSED_STID:
4734 case NFS4_CLOSED_DELEG_STID:
4735 status = nfserr_bad_stateid;
4736 }
4737out_unlock:
4738 spin_unlock(&cl->cl_lock);
4739 return status;
4740}
4741
4742__be32
4743nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4744 stateid_t *stateid, unsigned char typemask,
4745 struct nfs4_stid **s, struct nfsd_net *nn)
4746{
4747 __be32 status;
4748
4749 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4750 return nfserr_bad_stateid;
4751 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4752 if (status == nfserr_stale_clientid) {
4753 if (cstate->session)
4754 return nfserr_bad_stateid;
4755 return nfserr_stale_stateid;
4756 }
4757 if (status)
4758 return status;
4759 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4760 if (!*s)
4761 return nfserr_bad_stateid;
4762 return nfs_ok;
4763}
4764
4765static struct file *
4766nfs4_find_file(struct nfs4_stid *s, int flags)
4767{
4768 if (!s)
4769 return NULL;
4770
4771 switch (s->sc_type) {
4772 case NFS4_DELEG_STID:
4773 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4774 return NULL;
4775 return get_file(s->sc_file->fi_deleg_file);
4776 case NFS4_OPEN_STID:
4777 case NFS4_LOCK_STID:
4778 if (flags & RD_STATE)
4779 return find_readable_file(s->sc_file);
4780 else
4781 return find_writeable_file(s->sc_file);
4782 break;
4783 }
4784
4785 return NULL;
4786}
4787
4788static __be32
4789nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4790{
4791 __be32 status;
4792
4793 status = nfsd4_check_openowner_confirmed(ols);
4794 if (status)
4795 return status;
4796 return nfs4_check_openmode(ols, flags);
4797}
4798
4799static __be32
4800nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4801 struct file **filpp, bool *tmp_file, int flags)
4802{
4803 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4804 struct file *file;
4805 __be32 status;
4806
4807 file = nfs4_find_file(s, flags);
4808 if (file) {
4809 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4810 acc | NFSD_MAY_OWNER_OVERRIDE);
4811 if (status) {
4812 fput(file);
4813 return status;
4814 }
4815
4816 *filpp = file;
4817 } else {
4818 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4819 if (status)
4820 return status;
4821
4822 if (tmp_file)
4823 *tmp_file = true;
4824 }
4825
4826 return 0;
4827}
4828
4829/*
4830 * Checks for stateid operations
4831 */
4832__be32
4833nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4834 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4835 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4836{
4837 struct inode *ino = d_inode(fhp->fh_dentry);
4838 struct net *net = SVC_NET(rqstp);
4839 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4840 struct nfs4_stid *s = NULL;
4841 __be32 status;
4842
4843 if (filpp)
4844 *filpp = NULL;
4845 if (tmp_file)
4846 *tmp_file = false;
4847
4848 if (grace_disallows_io(net, ino))
4849 return nfserr_grace;
4850
4851 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4852 status = check_special_stateids(net, fhp, stateid, flags);
4853 goto done;
4854 }
4855
4856 status = nfsd4_lookup_stateid(cstate, stateid,
4857 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4858 &s, nn);
4859 if (status)
4860 return status;
4861 status = check_stateid_generation(stateid, &s->sc_stateid,
4862 nfsd4_has_session(cstate));
4863 if (status)
4864 goto out;
4865
4866 switch (s->sc_type) {
4867 case NFS4_DELEG_STID:
4868 status = nfs4_check_delegmode(delegstateid(s), flags);
4869 break;
4870 case NFS4_OPEN_STID:
4871 case NFS4_LOCK_STID:
4872 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4873 break;
4874 default:
4875 status = nfserr_bad_stateid;
4876 break;
4877 }
4878 if (status)
4879 goto out;
4880 status = nfs4_check_fh(fhp, s);
4881
4882done:
4883 if (!status && filpp)
4884 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
4885out:
4886 if (s)
4887 nfs4_put_stid(s);
4888 return status;
4889}
4890
4891/*
4892 * Test if the stateid is valid
4893 */
4894__be32
4895nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4896 struct nfsd4_test_stateid *test_stateid)
4897{
4898 struct nfsd4_test_stateid_id *stateid;
4899 struct nfs4_client *cl = cstate->session->se_client;
4900
4901 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4902 stateid->ts_id_status =
4903 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4904
4905 return nfs_ok;
4906}
4907
4908__be32
4909nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4910 struct nfsd4_free_stateid *free_stateid)
4911{
4912 stateid_t *stateid = &free_stateid->fr_stateid;
4913 struct nfs4_stid *s;
4914 struct nfs4_delegation *dp;
4915 struct nfs4_ol_stateid *stp;
4916 struct nfs4_client *cl = cstate->session->se_client;
4917 __be32 ret = nfserr_bad_stateid;
4918
4919 spin_lock(&cl->cl_lock);
4920 s = find_stateid_locked(cl, stateid);
4921 if (!s)
4922 goto out_unlock;
4923 switch (s->sc_type) {
4924 case NFS4_DELEG_STID:
4925 ret = nfserr_locks_held;
4926 break;
4927 case NFS4_OPEN_STID:
4928 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4929 if (ret)
4930 break;
4931 ret = nfserr_locks_held;
4932 break;
4933 case NFS4_LOCK_STID:
4934 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4935 if (ret)
4936 break;
4937 stp = openlockstateid(s);
4938 ret = nfserr_locks_held;
4939 if (check_for_locks(stp->st_stid.sc_file,
4940 lockowner(stp->st_stateowner)))
4941 break;
4942 WARN_ON(!unhash_lock_stateid(stp));
4943 spin_unlock(&cl->cl_lock);
4944 nfs4_put_stid(s);
4945 ret = nfs_ok;
4946 goto out;
4947 case NFS4_REVOKED_DELEG_STID:
4948 dp = delegstateid(s);
4949 list_del_init(&dp->dl_recall_lru);
4950 spin_unlock(&cl->cl_lock);
4951 nfs4_put_stid(s);
4952 ret = nfs_ok;
4953 goto out;
4954 /* Default falls through and returns nfserr_bad_stateid */
4955 }
4956out_unlock:
4957 spin_unlock(&cl->cl_lock);
4958out:
4959 return ret;
4960}
4961
4962static inline int
4963setlkflg (int type)
4964{
4965 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4966 RD_STATE : WR_STATE;
4967}
4968
4969static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4970{
4971 struct svc_fh *current_fh = &cstate->current_fh;
4972 struct nfs4_stateowner *sop = stp->st_stateowner;
4973 __be32 status;
4974
4975 status = nfsd4_check_seqid(cstate, sop, seqid);
4976 if (status)
4977 return status;
4978 if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4979 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4980 /*
4981 * "Closed" stateid's exist *only* to return
4982 * nfserr_replay_me from the previous step, and
4983 * revoked delegations are kept only for free_stateid.
4984 */
4985 return nfserr_bad_stateid;
4986 down_write(&stp->st_rwsem);
4987 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4988 if (status == nfs_ok)
4989 status = nfs4_check_fh(current_fh, &stp->st_stid);
4990 if (status != nfs_ok)
4991 up_write(&stp->st_rwsem);
4992 return status;
4993}
4994
4995/*
4996 * Checks for sequence id mutating operations.
4997 */
4998static __be32
4999nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5000 stateid_t *stateid, char typemask,
5001 struct nfs4_ol_stateid **stpp,
5002 struct nfsd_net *nn)
5003{
5004 __be32 status;
5005 struct nfs4_stid *s;
5006 struct nfs4_ol_stateid *stp = NULL;
5007
5008 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5009 seqid, STATEID_VAL(stateid));
5010
5011 *stpp = NULL;
5012 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5013 if (status)
5014 return status;
5015 stp = openlockstateid(s);
5016 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5017
5018 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5019 if (!status)
5020 *stpp = stp;
5021 else
5022 nfs4_put_stid(&stp->st_stid);
5023 return status;
5024}
5025
5026static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5027 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5028{
5029 __be32 status;
5030 struct nfs4_openowner *oo;
5031 struct nfs4_ol_stateid *stp;
5032
5033 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5034 NFS4_OPEN_STID, &stp, nn);
5035 if (status)
5036 return status;
5037 oo = openowner(stp->st_stateowner);
5038 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5039 up_write(&stp->st_rwsem);
5040 nfs4_put_stid(&stp->st_stid);
5041 return nfserr_bad_stateid;
5042 }
5043 *stpp = stp;
5044 return nfs_ok;
5045}
5046
5047__be32
5048nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5049 struct nfsd4_open_confirm *oc)
5050{
5051 __be32 status;
5052 struct nfs4_openowner *oo;
5053 struct nfs4_ol_stateid *stp;
5054 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5055
5056 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5057 cstate->current_fh.fh_dentry);
5058
5059 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5060 if (status)
5061 return status;
5062
5063 status = nfs4_preprocess_seqid_op(cstate,
5064 oc->oc_seqid, &oc->oc_req_stateid,
5065 NFS4_OPEN_STID, &stp, nn);
5066 if (status)
5067 goto out;
5068 oo = openowner(stp->st_stateowner);
5069 status = nfserr_bad_stateid;
5070 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5071 up_write(&stp->st_rwsem);
5072 goto put_stateid;
5073 }
5074 oo->oo_flags |= NFS4_OO_CONFIRMED;
5075 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5076 up_write(&stp->st_rwsem);
5077 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5078 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5079
5080 nfsd4_client_record_create(oo->oo_owner.so_client);
5081 status = nfs_ok;
5082put_stateid:
5083 nfs4_put_stid(&stp->st_stid);
5084out:
5085 nfsd4_bump_seqid(cstate, status);
5086 return status;
5087}
5088
5089static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5090{
5091 if (!test_access(access, stp))
5092 return;
5093 nfs4_file_put_access(stp->st_stid.sc_file, access);
5094 clear_access(access, stp);
5095}
5096
5097static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5098{
5099 switch (to_access) {
5100 case NFS4_SHARE_ACCESS_READ:
5101 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5102 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5103 break;
5104 case NFS4_SHARE_ACCESS_WRITE:
5105 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5106 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5107 break;
5108 case NFS4_SHARE_ACCESS_BOTH:
5109 break;
5110 default:
5111 WARN_ON_ONCE(1);
5112 }
5113}
5114
5115__be32
5116nfsd4_open_downgrade(struct svc_rqst *rqstp,
5117 struct nfsd4_compound_state *cstate,
5118 struct nfsd4_open_downgrade *od)
5119{
5120 __be32 status;
5121 struct nfs4_ol_stateid *stp;
5122 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5123
5124 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5125 cstate->current_fh.fh_dentry);
5126
5127 /* We don't yet support WANT bits: */
5128 if (od->od_deleg_want)
5129 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5130 od->od_deleg_want);
5131
5132 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5133 &od->od_stateid, &stp, nn);
5134 if (status)
5135 goto out;
5136 status = nfserr_inval;
5137 if (!test_access(od->od_share_access, stp)) {
5138 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5139 stp->st_access_bmap, od->od_share_access);
5140 goto put_stateid;
5141 }
5142 if (!test_deny(od->od_share_deny, stp)) {
5143 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5144 stp->st_deny_bmap, od->od_share_deny);
5145 goto put_stateid;
5146 }
5147 nfs4_stateid_downgrade(stp, od->od_share_access);
5148 reset_union_bmap_deny(od->od_share_deny, stp);
5149 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5150 status = nfs_ok;
5151put_stateid:
5152 up_write(&stp->st_rwsem);
5153 nfs4_put_stid(&stp->st_stid);
5154out:
5155 nfsd4_bump_seqid(cstate, status);
5156 return status;
5157}
5158
5159static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5160{
5161 struct nfs4_client *clp = s->st_stid.sc_client;
5162 bool unhashed;
5163 LIST_HEAD(reaplist);
5164
5165 s->st_stid.sc_type = NFS4_CLOSED_STID;
5166 spin_lock(&clp->cl_lock);
5167 unhashed = unhash_open_stateid(s, &reaplist);
5168
5169 if (clp->cl_minorversion) {
5170 if (unhashed)
5171 put_ol_stateid_locked(s, &reaplist);
5172 spin_unlock(&clp->cl_lock);
5173 free_ol_stateid_reaplist(&reaplist);
5174 } else {
5175 spin_unlock(&clp->cl_lock);
5176 free_ol_stateid_reaplist(&reaplist);
5177 if (unhashed)
5178 move_to_close_lru(s, clp->net);
5179 }
5180}
5181
5182/*
5183 * nfs4_unlock_state() called after encode
5184 */
5185__be32
5186nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5187 struct nfsd4_close *close)
5188{
5189 __be32 status;
5190 struct nfs4_ol_stateid *stp;
5191 struct net *net = SVC_NET(rqstp);
5192 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5193
5194 dprintk("NFSD: nfsd4_close on file %pd\n",
5195 cstate->current_fh.fh_dentry);
5196
5197 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5198 &close->cl_stateid,
5199 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5200 &stp, nn);
5201 nfsd4_bump_seqid(cstate, status);
5202 if (status)
5203 goto out;
5204 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5205 up_write(&stp->st_rwsem);
5206
5207 nfsd4_close_open_stateid(stp);
5208
5209 /* put reference from nfs4_preprocess_seqid_op */
5210 nfs4_put_stid(&stp->st_stid);
5211out:
5212 return status;
5213}
5214
5215__be32
5216nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5217 struct nfsd4_delegreturn *dr)
5218{
5219 struct nfs4_delegation *dp;
5220 stateid_t *stateid = &dr->dr_stateid;
5221 struct nfs4_stid *s;
5222 __be32 status;
5223 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5224
5225 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5226 return status;
5227
5228 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5229 if (status)
5230 goto out;
5231 dp = delegstateid(s);
5232 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5233 if (status)
5234 goto put_stateid;
5235
5236 destroy_delegation(dp);
5237put_stateid:
5238 nfs4_put_stid(&dp->dl_stid);
5239out:
5240 return status;
5241}
5242
5243static inline u64
5244end_offset(u64 start, u64 len)
5245{
5246 u64 end;
5247
5248 end = start + len;
5249 return end >= start ? end: NFS4_MAX_UINT64;
5250}
5251
5252/* last octet in a range */
5253static inline u64
5254last_byte_offset(u64 start, u64 len)
5255{
5256 u64 end;
5257
5258 WARN_ON_ONCE(!len);
5259 end = start + len;
5260 return end > start ? end - 1: NFS4_MAX_UINT64;
5261}
5262
5263/*
5264 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5265 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5266 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
5267 * locking, this prevents us from being completely protocol-compliant. The
5268 * real solution to this problem is to start using unsigned file offsets in
5269 * the VFS, but this is a very deep change!
5270 */
5271static inline void
5272nfs4_transform_lock_offset(struct file_lock *lock)
5273{
5274 if (lock->fl_start < 0)
5275 lock->fl_start = OFFSET_MAX;
5276 if (lock->fl_end < 0)
5277 lock->fl_end = OFFSET_MAX;
5278}
5279
5280static fl_owner_t
5281nfsd4_fl_get_owner(fl_owner_t owner)
5282{
5283 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5284
5285 nfs4_get_stateowner(&lo->lo_owner);
5286 return owner;
5287}
5288
5289static void
5290nfsd4_fl_put_owner(fl_owner_t owner)
5291{
5292 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5293
5294 if (lo)
5295 nfs4_put_stateowner(&lo->lo_owner);
5296}
5297
5298static const struct lock_manager_operations nfsd_posix_mng_ops = {
5299 .lm_get_owner = nfsd4_fl_get_owner,
5300 .lm_put_owner = nfsd4_fl_put_owner,
5301};
5302
5303static inline void
5304nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5305{
5306 struct nfs4_lockowner *lo;
5307
5308 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5309 lo = (struct nfs4_lockowner *) fl->fl_owner;
5310 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5311 lo->lo_owner.so_owner.len, GFP_KERNEL);
5312 if (!deny->ld_owner.data)
5313 /* We just don't care that much */
5314 goto nevermind;
5315 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5316 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5317 } else {
5318nevermind:
5319 deny->ld_owner.len = 0;
5320 deny->ld_owner.data = NULL;
5321 deny->ld_clientid.cl_boot = 0;
5322 deny->ld_clientid.cl_id = 0;
5323 }
5324 deny->ld_start = fl->fl_start;
5325 deny->ld_length = NFS4_MAX_UINT64;
5326 if (fl->fl_end != NFS4_MAX_UINT64)
5327 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5328 deny->ld_type = NFS4_READ_LT;
5329 if (fl->fl_type != F_RDLCK)
5330 deny->ld_type = NFS4_WRITE_LT;
5331}
5332
5333static struct nfs4_lockowner *
5334find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5335{
5336 unsigned int strhashval = ownerstr_hashval(owner);
5337 struct nfs4_stateowner *so;
5338
5339 lockdep_assert_held(&clp->cl_lock);
5340
5341 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5342 so_strhash) {
5343 if (so->so_is_open_owner)
5344 continue;
5345 if (same_owner_str(so, owner))
5346 return lockowner(nfs4_get_stateowner(so));
5347 }
5348 return NULL;
5349}
5350
5351static struct nfs4_lockowner *
5352find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5353{
5354 struct nfs4_lockowner *lo;
5355
5356 spin_lock(&clp->cl_lock);
5357 lo = find_lockowner_str_locked(clp, owner);
5358 spin_unlock(&clp->cl_lock);
5359 return lo;
5360}
5361
5362static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5363{
5364 unhash_lockowner_locked(lockowner(sop));
5365}
5366
5367static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5368{
5369 struct nfs4_lockowner *lo = lockowner(sop);
5370
5371 kmem_cache_free(lockowner_slab, lo);
5372}
5373
5374static const struct nfs4_stateowner_operations lockowner_ops = {
5375 .so_unhash = nfs4_unhash_lockowner,
5376 .so_free = nfs4_free_lockowner,
5377};
5378
5379/*
5380 * Alloc a lock owner structure.
5381 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5382 * occurred.
5383 *
5384 * strhashval = ownerstr_hashval
5385 */
5386static struct nfs4_lockowner *
5387alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5388 struct nfs4_ol_stateid *open_stp,
5389 struct nfsd4_lock *lock)
5390{
5391 struct nfs4_lockowner *lo, *ret;
5392
5393 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5394 if (!lo)
5395 return NULL;
5396 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5397 lo->lo_owner.so_is_open_owner = 0;
5398 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5399 lo->lo_owner.so_ops = &lockowner_ops;
5400 spin_lock(&clp->cl_lock);
5401 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5402 if (ret == NULL) {
5403 list_add(&lo->lo_owner.so_strhash,
5404 &clp->cl_ownerstr_hashtbl[strhashval]);
5405 ret = lo;
5406 } else
5407 nfs4_free_stateowner(&lo->lo_owner);
5408
5409 spin_unlock(&clp->cl_lock);
5410 return ret;
5411}
5412
5413static void
5414init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5415 struct nfs4_file *fp, struct inode *inode,
5416 struct nfs4_ol_stateid *open_stp)
5417{
5418 struct nfs4_client *clp = lo->lo_owner.so_client;
5419
5420 lockdep_assert_held(&clp->cl_lock);
5421
5422 atomic_inc(&stp->st_stid.sc_count);
5423 stp->st_stid.sc_type = NFS4_LOCK_STID;
5424 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5425 get_nfs4_file(fp);
5426 stp->st_stid.sc_file = fp;
5427 stp->st_stid.sc_free = nfs4_free_lock_stateid;
5428 stp->st_access_bmap = 0;
5429 stp->st_deny_bmap = open_stp->st_deny_bmap;
5430 stp->st_openstp = open_stp;
5431 init_rwsem(&stp->st_rwsem);
5432 list_add(&stp->st_locks, &open_stp->st_locks);
5433 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5434 spin_lock(&fp->fi_lock);
5435 list_add(&stp->st_perfile, &fp->fi_stateids);
5436 spin_unlock(&fp->fi_lock);
5437}
5438
5439static struct nfs4_ol_stateid *
5440find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5441{
5442 struct nfs4_ol_stateid *lst;
5443 struct nfs4_client *clp = lo->lo_owner.so_client;
5444
5445 lockdep_assert_held(&clp->cl_lock);
5446
5447 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5448 if (lst->st_stid.sc_file == fp) {
5449 atomic_inc(&lst->st_stid.sc_count);
5450 return lst;
5451 }
5452 }
5453 return NULL;
5454}
5455
5456static struct nfs4_ol_stateid *
5457find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5458 struct inode *inode, struct nfs4_ol_stateid *ost,
5459 bool *new)
5460{
5461 struct nfs4_stid *ns = NULL;
5462 struct nfs4_ol_stateid *lst;
5463 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5464 struct nfs4_client *clp = oo->oo_owner.so_client;
5465
5466 spin_lock(&clp->cl_lock);
5467 lst = find_lock_stateid(lo, fi);
5468 if (lst == NULL) {
5469 spin_unlock(&clp->cl_lock);
5470 ns = nfs4_alloc_stid(clp, stateid_slab);
5471 if (ns == NULL)
5472 return NULL;
5473
5474 spin_lock(&clp->cl_lock);
5475 lst = find_lock_stateid(lo, fi);
5476 if (likely(!lst)) {
5477 lst = openlockstateid(ns);
5478 init_lock_stateid(lst, lo, fi, inode, ost);
5479 ns = NULL;
5480 *new = true;
5481 }
5482 }
5483 spin_unlock(&clp->cl_lock);
5484 if (ns)
5485 nfs4_put_stid(ns);
5486 return lst;
5487}
5488
5489static int
5490check_lock_length(u64 offset, u64 length)
5491{
5492 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5493 (length > ~offset)));
5494}
5495
5496static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5497{
5498 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5499
5500 lockdep_assert_held(&fp->fi_lock);
5501
5502 if (test_access(access, lock_stp))
5503 return;
5504 __nfs4_file_get_access(fp, access);
5505 set_access(access, lock_stp);
5506}
5507
5508static __be32
5509lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5510 struct nfs4_ol_stateid *ost,
5511 struct nfsd4_lock *lock,
5512 struct nfs4_ol_stateid **lst, bool *new)
5513{
5514 __be32 status;
5515 struct nfs4_file *fi = ost->st_stid.sc_file;
5516 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5517 struct nfs4_client *cl = oo->oo_owner.so_client;
5518 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5519 struct nfs4_lockowner *lo;
5520 unsigned int strhashval;
5521
5522 lo = find_lockowner_str(cl, &lock->lk_new_owner);
5523 if (!lo) {
5524 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5525 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5526 if (lo == NULL)
5527 return nfserr_jukebox;
5528 } else {
5529 /* with an existing lockowner, seqids must be the same */
5530 status = nfserr_bad_seqid;
5531 if (!cstate->minorversion &&
5532 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5533 goto out;
5534 }
5535
5536 *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5537 if (*lst == NULL) {
5538 status = nfserr_jukebox;
5539 goto out;
5540 }
5541 status = nfs_ok;
5542out:
5543 nfs4_put_stateowner(&lo->lo_owner);
5544 return status;
5545}
5546
5547/*
5548 * LOCK operation
5549 */
5550__be32
5551nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5552 struct nfsd4_lock *lock)
5553{
5554 struct nfs4_openowner *open_sop = NULL;
5555 struct nfs4_lockowner *lock_sop = NULL;
5556 struct nfs4_ol_stateid *lock_stp = NULL;
5557 struct nfs4_ol_stateid *open_stp = NULL;
5558 struct nfs4_file *fp;
5559 struct file *filp = NULL;
5560 struct file_lock *file_lock = NULL;
5561 struct file_lock *conflock = NULL;
5562 __be32 status = 0;
5563 int lkflg;
5564 int err;
5565 bool new = false;
5566 struct net *net = SVC_NET(rqstp);
5567 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5568
5569 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5570 (long long) lock->lk_offset,
5571 (long long) lock->lk_length);
5572
5573 if (check_lock_length(lock->lk_offset, lock->lk_length))
5574 return nfserr_inval;
5575
5576 if ((status = fh_verify(rqstp, &cstate->current_fh,
5577 S_IFREG, NFSD_MAY_LOCK))) {
5578 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5579 return status;
5580 }
5581
5582 if (lock->lk_is_new) {
5583 if (nfsd4_has_session(cstate))
5584 /* See rfc 5661 18.10.3: given clientid is ignored: */
5585 memcpy(&lock->lk_new_clientid,
5586 &cstate->session->se_client->cl_clientid,
5587 sizeof(clientid_t));
5588
5589 status = nfserr_stale_clientid;
5590 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5591 goto out;
5592
5593 /* validate and update open stateid and open seqid */
5594 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5595 lock->lk_new_open_seqid,
5596 &lock->lk_new_open_stateid,
5597 &open_stp, nn);
5598 if (status)
5599 goto out;
5600 up_write(&open_stp->st_rwsem);
5601 open_sop = openowner(open_stp->st_stateowner);
5602 status = nfserr_bad_stateid;
5603 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5604 &lock->lk_new_clientid))
5605 goto out;
5606 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5607 &lock_stp, &new);
5608 if (status == nfs_ok)
5609 down_write(&lock_stp->st_rwsem);
5610 } else {
5611 status = nfs4_preprocess_seqid_op(cstate,
5612 lock->lk_old_lock_seqid,
5613 &lock->lk_old_lock_stateid,
5614 NFS4_LOCK_STID, &lock_stp, nn);
5615 }
5616 if (status)
5617 goto out;
5618 lock_sop = lockowner(lock_stp->st_stateowner);
5619
5620 lkflg = setlkflg(lock->lk_type);
5621 status = nfs4_check_openmode(lock_stp, lkflg);
5622 if (status)
5623 goto out;
5624
5625 status = nfserr_grace;
5626 if (locks_in_grace(net) && !lock->lk_reclaim)
5627 goto out;
5628 status = nfserr_no_grace;
5629 if (!locks_in_grace(net) && lock->lk_reclaim)
5630 goto out;
5631
5632 file_lock = locks_alloc_lock();
5633 if (!file_lock) {
5634 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5635 status = nfserr_jukebox;
5636 goto out;
5637 }
5638
5639 fp = lock_stp->st_stid.sc_file;
5640 switch (lock->lk_type) {
5641 case NFS4_READ_LT:
5642 case NFS4_READW_LT:
5643 spin_lock(&fp->fi_lock);
5644 filp = find_readable_file_locked(fp);
5645 if (filp)
5646 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5647 spin_unlock(&fp->fi_lock);
5648 file_lock->fl_type = F_RDLCK;
5649 break;
5650 case NFS4_WRITE_LT:
5651 case NFS4_WRITEW_LT:
5652 spin_lock(&fp->fi_lock);
5653 filp = find_writeable_file_locked(fp);
5654 if (filp)
5655 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5656 spin_unlock(&fp->fi_lock);
5657 file_lock->fl_type = F_WRLCK;
5658 break;
5659 default:
5660 status = nfserr_inval;
5661 goto out;
5662 }
5663 if (!filp) {
5664 status = nfserr_openmode;
5665 goto out;
5666 }
5667
5668 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5669 file_lock->fl_pid = current->tgid;
5670 file_lock->fl_file = filp;
5671 file_lock->fl_flags = FL_POSIX;
5672 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5673 file_lock->fl_start = lock->lk_offset;
5674 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5675 nfs4_transform_lock_offset(file_lock);
5676
5677 conflock = locks_alloc_lock();
5678 if (!conflock) {
5679 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5680 status = nfserr_jukebox;
5681 goto out;
5682 }
5683
5684 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5685 switch (-err) {
5686 case 0: /* success! */
5687 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5688 status = 0;
5689 break;
5690 case (EAGAIN): /* conflock holds conflicting lock */
5691 status = nfserr_denied;
5692 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5693 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5694 break;
5695 case (EDEADLK):
5696 status = nfserr_deadlock;
5697 break;
5698 default:
5699 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5700 status = nfserrno(err);
5701 break;
5702 }
5703out:
5704 if (filp)
5705 fput(filp);
5706 if (lock_stp) {
5707 /* Bump seqid manually if the 4.0 replay owner is openowner */
5708 if (cstate->replay_owner &&
5709 cstate->replay_owner != &lock_sop->lo_owner &&
5710 seqid_mutating_err(ntohl(status)))
5711 lock_sop->lo_owner.so_seqid++;
5712
5713 up_write(&lock_stp->st_rwsem);
5714
5715 /*
5716 * If this is a new, never-before-used stateid, and we are
5717 * returning an error, then just go ahead and release it.
5718 */
5719 if (status && new)
5720 release_lock_stateid(lock_stp);
5721
5722 nfs4_put_stid(&lock_stp->st_stid);
5723 }
5724 if (open_stp)
5725 nfs4_put_stid(&open_stp->st_stid);
5726 nfsd4_bump_seqid(cstate, status);
5727 if (file_lock)
5728 locks_free_lock(file_lock);
5729 if (conflock)
5730 locks_free_lock(conflock);
5731 return status;
5732}
5733
5734/*
5735 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5736 * so we do a temporary open here just to get an open file to pass to
5737 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
5738 * inode operation.)
5739 */
5740static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5741{
5742 struct file *file;
5743 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5744 if (!err) {
5745 err = nfserrno(vfs_test_lock(file, lock));
5746 fput(file);
5747 }
5748 return err;
5749}
5750
5751/*
5752 * LOCKT operation
5753 */
5754__be32
5755nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5756 struct nfsd4_lockt *lockt)
5757{
5758 struct file_lock *file_lock = NULL;
5759 struct nfs4_lockowner *lo = NULL;
5760 __be32 status;
5761 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5762
5763 if (locks_in_grace(SVC_NET(rqstp)))
5764 return nfserr_grace;
5765
5766 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5767 return nfserr_inval;
5768
5769 if (!nfsd4_has_session(cstate)) {
5770 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5771 if (status)
5772 goto out;
5773 }
5774
5775 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5776 goto out;
5777
5778 file_lock = locks_alloc_lock();
5779 if (!file_lock) {
5780 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5781 status = nfserr_jukebox;
5782 goto out;
5783 }
5784
5785 switch (lockt->lt_type) {
5786 case NFS4_READ_LT:
5787 case NFS4_READW_LT:
5788 file_lock->fl_type = F_RDLCK;
5789 break;
5790 case NFS4_WRITE_LT:
5791 case NFS4_WRITEW_LT:
5792 file_lock->fl_type = F_WRLCK;
5793 break;
5794 default:
5795 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5796 status = nfserr_inval;
5797 goto out;
5798 }
5799
5800 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
5801 if (lo)
5802 file_lock->fl_owner = (fl_owner_t)lo;
5803 file_lock->fl_pid = current->tgid;
5804 file_lock->fl_flags = FL_POSIX;
5805
5806 file_lock->fl_start = lockt->lt_offset;
5807 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5808
5809 nfs4_transform_lock_offset(file_lock);
5810
5811 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5812 if (status)
5813 goto out;
5814
5815 if (file_lock->fl_type != F_UNLCK) {
5816 status = nfserr_denied;
5817 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5818 }
5819out:
5820 if (lo)
5821 nfs4_put_stateowner(&lo->lo_owner);
5822 if (file_lock)
5823 locks_free_lock(file_lock);
5824 return status;
5825}
5826
5827__be32
5828nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5829 struct nfsd4_locku *locku)
5830{
5831 struct nfs4_ol_stateid *stp;
5832 struct file *filp = NULL;
5833 struct file_lock *file_lock = NULL;
5834 __be32 status;
5835 int err;
5836 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5837
5838 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5839 (long long) locku->lu_offset,
5840 (long long) locku->lu_length);
5841
5842 if (check_lock_length(locku->lu_offset, locku->lu_length))
5843 return nfserr_inval;
5844
5845 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5846 &locku->lu_stateid, NFS4_LOCK_STID,
5847 &stp, nn);
5848 if (status)
5849 goto out;
5850 filp = find_any_file(stp->st_stid.sc_file);
5851 if (!filp) {
5852 status = nfserr_lock_range;
5853 goto put_stateid;
5854 }
5855 file_lock = locks_alloc_lock();
5856 if (!file_lock) {
5857 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5858 status = nfserr_jukebox;
5859 goto fput;
5860 }
5861
5862 file_lock->fl_type = F_UNLCK;
5863 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5864 file_lock->fl_pid = current->tgid;
5865 file_lock->fl_file = filp;
5866 file_lock->fl_flags = FL_POSIX;
5867 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5868 file_lock->fl_start = locku->lu_offset;
5869
5870 file_lock->fl_end = last_byte_offset(locku->lu_offset,
5871 locku->lu_length);
5872 nfs4_transform_lock_offset(file_lock);
5873
5874 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5875 if (err) {
5876 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5877 goto out_nfserr;
5878 }
5879 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
5880fput:
5881 fput(filp);
5882put_stateid:
5883 up_write(&stp->st_rwsem);
5884 nfs4_put_stid(&stp->st_stid);
5885out:
5886 nfsd4_bump_seqid(cstate, status);
5887 if (file_lock)
5888 locks_free_lock(file_lock);
5889 return status;
5890
5891out_nfserr:
5892 status = nfserrno(err);
5893 goto fput;
5894}
5895
5896/*
5897 * returns
5898 * true: locks held by lockowner
5899 * false: no locks held by lockowner
5900 */
5901static bool
5902check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5903{
5904 struct file_lock *fl;
5905 int status = false;
5906 struct file *filp = find_any_file(fp);
5907 struct inode *inode;
5908 struct file_lock_context *flctx;
5909
5910 if (!filp) {
5911 /* Any valid lock stateid should have some sort of access */
5912 WARN_ON_ONCE(1);
5913 return status;
5914 }
5915
5916 inode = file_inode(filp);
5917 flctx = inode->i_flctx;
5918
5919 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
5920 spin_lock(&flctx->flc_lock);
5921 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
5922 if (fl->fl_owner == (fl_owner_t)lowner) {
5923 status = true;
5924 break;
5925 }
5926 }
5927 spin_unlock(&flctx->flc_lock);
5928 }
5929 fput(filp);
5930 return status;
5931}
5932
5933__be32
5934nfsd4_release_lockowner(struct svc_rqst *rqstp,
5935 struct nfsd4_compound_state *cstate,
5936 struct nfsd4_release_lockowner *rlockowner)
5937{
5938 clientid_t *clid = &rlockowner->rl_clientid;
5939 struct nfs4_stateowner *sop;
5940 struct nfs4_lockowner *lo = NULL;
5941 struct nfs4_ol_stateid *stp;
5942 struct xdr_netobj *owner = &rlockowner->rl_owner;
5943 unsigned int hashval = ownerstr_hashval(owner);
5944 __be32 status;
5945 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5946 struct nfs4_client *clp;
5947
5948 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5949 clid->cl_boot, clid->cl_id);
5950
5951 status = lookup_clientid(clid, cstate, nn);
5952 if (status)
5953 return status;
5954
5955 clp = cstate->clp;
5956 /* Find the matching lock stateowner */
5957 spin_lock(&clp->cl_lock);
5958 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5959 so_strhash) {
5960
5961 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5962 continue;
5963
5964 /* see if there are still any locks associated with it */
5965 lo = lockowner(sop);
5966 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5967 if (check_for_locks(stp->st_stid.sc_file, lo)) {
5968 status = nfserr_locks_held;
5969 spin_unlock(&clp->cl_lock);
5970 return status;
5971 }
5972 }
5973
5974 nfs4_get_stateowner(sop);
5975 break;
5976 }
5977 spin_unlock(&clp->cl_lock);
5978 if (lo)
5979 release_lockowner(lo);
5980 return status;
5981}
5982
5983static inline struct nfs4_client_reclaim *
5984alloc_reclaim(void)
5985{
5986 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5987}
5988
5989bool
5990nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5991{
5992 struct nfs4_client_reclaim *crp;
5993
5994 crp = nfsd4_find_reclaim_client(name, nn);
5995 return (crp && crp->cr_clp);
5996}
5997
5998/*
5999 * failure => all reset bets are off, nfserr_no_grace...
6000 */
6001struct nfs4_client_reclaim *
6002nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6003{
6004 unsigned int strhashval;
6005 struct nfs4_client_reclaim *crp;
6006
6007 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6008 crp = alloc_reclaim();
6009 if (crp) {
6010 strhashval = clientstr_hashval(name);
6011 INIT_LIST_HEAD(&crp->cr_strhash);
6012 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6013 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6014 crp->cr_clp = NULL;
6015 nn->reclaim_str_hashtbl_size++;
6016 }
6017 return crp;
6018}
6019
6020void
6021nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6022{
6023 list_del(&crp->cr_strhash);
6024 kfree(crp);
6025 nn->reclaim_str_hashtbl_size--;
6026}
6027
6028void
6029nfs4_release_reclaim(struct nfsd_net *nn)
6030{
6031 struct nfs4_client_reclaim *crp = NULL;
6032 int i;
6033
6034 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6035 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6036 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6037 struct nfs4_client_reclaim, cr_strhash);
6038 nfs4_remove_reclaim_record(crp, nn);
6039 }
6040 }
6041 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6042}
6043
6044/*
6045 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6046struct nfs4_client_reclaim *
6047nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6048{
6049 unsigned int strhashval;
6050 struct nfs4_client_reclaim *crp = NULL;
6051
6052 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6053
6054 strhashval = clientstr_hashval(recdir);
6055 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6056 if (same_name(crp->cr_recdir, recdir)) {
6057 return crp;
6058 }
6059 }
6060 return NULL;
6061}
6062
6063/*
6064* Called from OPEN. Look for clientid in reclaim list.
6065*/
6066__be32
6067nfs4_check_open_reclaim(clientid_t *clid,
6068 struct nfsd4_compound_state *cstate,
6069 struct nfsd_net *nn)
6070{
6071 __be32 status;
6072
6073 /* find clientid in conf_id_hashtbl */
6074 status = lookup_clientid(clid, cstate, nn);
6075 if (status)
6076 return nfserr_reclaim_bad;
6077
6078 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6079 return nfserr_no_grace;
6080
6081 if (nfsd4_client_record_check(cstate->clp))
6082 return nfserr_reclaim_bad;
6083
6084 return nfs_ok;
6085}
6086
6087#ifdef CONFIG_NFSD_FAULT_INJECTION
6088static inline void
6089put_client(struct nfs4_client *clp)
6090{
6091 atomic_dec(&clp->cl_refcount);
6092}
6093
6094static struct nfs4_client *
6095nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6096{
6097 struct nfs4_client *clp;
6098 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6099 nfsd_net_id);
6100
6101 if (!nfsd_netns_ready(nn))
6102 return NULL;
6103
6104 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6105 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6106 return clp;
6107 }
6108 return NULL;
6109}
6110
6111u64
6112nfsd_inject_print_clients(void)
6113{
6114 struct nfs4_client *clp;
6115 u64 count = 0;
6116 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6117 nfsd_net_id);
6118 char buf[INET6_ADDRSTRLEN];
6119
6120 if (!nfsd_netns_ready(nn))
6121 return 0;
6122
6123 spin_lock(&nn->client_lock);
6124 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6125 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6126 pr_info("NFS Client: %s\n", buf);
6127 ++count;
6128 }
6129 spin_unlock(&nn->client_lock);
6130
6131 return count;
6132}
6133
6134u64
6135nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6136{
6137 u64 count = 0;
6138 struct nfs4_client *clp;
6139 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6140 nfsd_net_id);
6141
6142 if (!nfsd_netns_ready(nn))
6143 return count;
6144
6145 spin_lock(&nn->client_lock);
6146 clp = nfsd_find_client(addr, addr_size);
6147 if (clp) {
6148 if (mark_client_expired_locked(clp) == nfs_ok)
6149 ++count;
6150 else
6151 clp = NULL;
6152 }
6153 spin_unlock(&nn->client_lock);
6154
6155 if (clp)
6156 expire_client(clp);
6157
6158 return count;
6159}
6160
6161u64
6162nfsd_inject_forget_clients(u64 max)
6163{
6164 u64 count = 0;
6165 struct nfs4_client *clp, *next;
6166 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6167 nfsd_net_id);
6168 LIST_HEAD(reaplist);
6169
6170 if (!nfsd_netns_ready(nn))
6171 return count;
6172
6173 spin_lock(&nn->client_lock);
6174 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6175 if (mark_client_expired_locked(clp) == nfs_ok) {
6176 list_add(&clp->cl_lru, &reaplist);
6177 if (max != 0 && ++count >= max)
6178 break;
6179 }
6180 }
6181 spin_unlock(&nn->client_lock);
6182
6183 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6184 expire_client(clp);
6185
6186 return count;
6187}
6188
6189static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6190 const char *type)
6191{
6192 char buf[INET6_ADDRSTRLEN];
6193 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6194 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6195}
6196
6197static void
6198nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6199 struct list_head *collect)
6200{
6201 struct nfs4_client *clp = lst->st_stid.sc_client;
6202 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6203 nfsd_net_id);
6204
6205 if (!collect)
6206 return;
6207
6208 lockdep_assert_held(&nn->client_lock);
6209 atomic_inc(&clp->cl_refcount);
6210 list_add(&lst->st_locks, collect);
6211}
6212
6213static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6214 struct list_head *collect,
6215 bool (*func)(struct nfs4_ol_stateid *))
6216{
6217 struct nfs4_openowner *oop;
6218 struct nfs4_ol_stateid *stp, *st_next;
6219 struct nfs4_ol_stateid *lst, *lst_next;
6220 u64 count = 0;
6221
6222 spin_lock(&clp->cl_lock);
6223 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6224 list_for_each_entry_safe(stp, st_next,
6225 &oop->oo_owner.so_stateids, st_perstateowner) {
6226 list_for_each_entry_safe(lst, lst_next,
6227 &stp->st_locks, st_locks) {
6228 if (func) {
6229 if (func(lst))
6230 nfsd_inject_add_lock_to_list(lst,
6231 collect);
6232 }
6233 ++count;
6234 /*
6235 * Despite the fact that these functions deal
6236 * with 64-bit integers for "count", we must
6237 * ensure that it doesn't blow up the
6238 * clp->cl_refcount. Throw a warning if we
6239 * start to approach INT_MAX here.
6240 */
6241 WARN_ON_ONCE(count == (INT_MAX / 2));
6242 if (count == max)
6243 goto out;
6244 }
6245 }
6246 }
6247out:
6248 spin_unlock(&clp->cl_lock);
6249
6250 return count;
6251}
6252
6253static u64
6254nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6255 u64 max)
6256{
6257 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6258}
6259
6260static u64
6261nfsd_print_client_locks(struct nfs4_client *clp)
6262{
6263 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6264 nfsd_print_count(clp, count, "locked files");
6265 return count;
6266}
6267
6268u64
6269nfsd_inject_print_locks(void)
6270{
6271 struct nfs4_client *clp;
6272 u64 count = 0;
6273 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6274 nfsd_net_id);
6275
6276 if (!nfsd_netns_ready(nn))
6277 return 0;
6278
6279 spin_lock(&nn->client_lock);
6280 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6281 count += nfsd_print_client_locks(clp);
6282 spin_unlock(&nn->client_lock);
6283
6284 return count;
6285}
6286
6287static void
6288nfsd_reap_locks(struct list_head *reaplist)
6289{
6290 struct nfs4_client *clp;
6291 struct nfs4_ol_stateid *stp, *next;
6292
6293 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6294 list_del_init(&stp->st_locks);
6295 clp = stp->st_stid.sc_client;
6296 nfs4_put_stid(&stp->st_stid);
6297 put_client(clp);
6298 }
6299}
6300
6301u64
6302nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6303{
6304 unsigned int count = 0;
6305 struct nfs4_client *clp;
6306 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6307 nfsd_net_id);
6308 LIST_HEAD(reaplist);
6309
6310 if (!nfsd_netns_ready(nn))
6311 return count;
6312
6313 spin_lock(&nn->client_lock);
6314 clp = nfsd_find_client(addr, addr_size);
6315 if (clp)
6316 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6317 spin_unlock(&nn->client_lock);
6318 nfsd_reap_locks(&reaplist);
6319 return count;
6320}
6321
6322u64
6323nfsd_inject_forget_locks(u64 max)
6324{
6325 u64 count = 0;
6326 struct nfs4_client *clp;
6327 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6328 nfsd_net_id);
6329 LIST_HEAD(reaplist);
6330
6331 if (!nfsd_netns_ready(nn))
6332 return count;
6333
6334 spin_lock(&nn->client_lock);
6335 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6336 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6337 if (max != 0 && count >= max)
6338 break;
6339 }
6340 spin_unlock(&nn->client_lock);
6341 nfsd_reap_locks(&reaplist);
6342 return count;
6343}
6344
6345static u64
6346nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6347 struct list_head *collect,
6348 void (*func)(struct nfs4_openowner *))
6349{
6350 struct nfs4_openowner *oop, *next;
6351 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6352 nfsd_net_id);
6353 u64 count = 0;
6354
6355 lockdep_assert_held(&nn->client_lock);
6356
6357 spin_lock(&clp->cl_lock);
6358 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6359 if (func) {
6360 func(oop);
6361 if (collect) {
6362 atomic_inc(&clp->cl_refcount);
6363 list_add(&oop->oo_perclient, collect);
6364 }
6365 }
6366 ++count;
6367 /*
6368 * Despite the fact that these functions deal with
6369 * 64-bit integers for "count", we must ensure that
6370 * it doesn't blow up the clp->cl_refcount. Throw a
6371 * warning if we start to approach INT_MAX here.
6372 */
6373 WARN_ON_ONCE(count == (INT_MAX / 2));
6374 if (count == max)
6375 break;
6376 }
6377 spin_unlock(&clp->cl_lock);
6378
6379 return count;
6380}
6381
6382static u64
6383nfsd_print_client_openowners(struct nfs4_client *clp)
6384{
6385 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6386
6387 nfsd_print_count(clp, count, "openowners");
6388 return count;
6389}
6390
6391static u64
6392nfsd_collect_client_openowners(struct nfs4_client *clp,
6393 struct list_head *collect, u64 max)
6394{
6395 return nfsd_foreach_client_openowner(clp, max, collect,
6396 unhash_openowner_locked);
6397}
6398
6399u64
6400nfsd_inject_print_openowners(void)
6401{
6402 struct nfs4_client *clp;
6403 u64 count = 0;
6404 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6405 nfsd_net_id);
6406
6407 if (!nfsd_netns_ready(nn))
6408 return 0;
6409
6410 spin_lock(&nn->client_lock);
6411 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6412 count += nfsd_print_client_openowners(clp);
6413 spin_unlock(&nn->client_lock);
6414
6415 return count;
6416}
6417
6418static void
6419nfsd_reap_openowners(struct list_head *reaplist)
6420{
6421 struct nfs4_client *clp;
6422 struct nfs4_openowner *oop, *next;
6423
6424 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6425 list_del_init(&oop->oo_perclient);
6426 clp = oop->oo_owner.so_client;
6427 release_openowner(oop);
6428 put_client(clp);
6429 }
6430}
6431
6432u64
6433nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6434 size_t addr_size)
6435{
6436 unsigned int count = 0;
6437 struct nfs4_client *clp;
6438 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6439 nfsd_net_id);
6440 LIST_HEAD(reaplist);
6441
6442 if (!nfsd_netns_ready(nn))
6443 return count;
6444
6445 spin_lock(&nn->client_lock);
6446 clp = nfsd_find_client(addr, addr_size);
6447 if (clp)
6448 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6449 spin_unlock(&nn->client_lock);
6450 nfsd_reap_openowners(&reaplist);
6451 return count;
6452}
6453
6454u64
6455nfsd_inject_forget_openowners(u64 max)
6456{
6457 u64 count = 0;
6458 struct nfs4_client *clp;
6459 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6460 nfsd_net_id);
6461 LIST_HEAD(reaplist);
6462
6463 if (!nfsd_netns_ready(nn))
6464 return count;
6465
6466 spin_lock(&nn->client_lock);
6467 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6468 count += nfsd_collect_client_openowners(clp, &reaplist,
6469 max - count);
6470 if (max != 0 && count >= max)
6471 break;
6472 }
6473 spin_unlock(&nn->client_lock);
6474 nfsd_reap_openowners(&reaplist);
6475 return count;
6476}
6477
6478static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6479 struct list_head *victims)
6480{
6481 struct nfs4_delegation *dp, *next;
6482 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6483 nfsd_net_id);
6484 u64 count = 0;
6485
6486 lockdep_assert_held(&nn->client_lock);
6487
6488 spin_lock(&state_lock);
6489 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6490 if (victims) {
6491 /*
6492 * It's not safe to mess with delegations that have a
6493 * non-zero dl_time. They might have already been broken
6494 * and could be processed by the laundromat outside of
6495 * the state_lock. Just leave them be.
6496 */
6497 if (dp->dl_time != 0)
6498 continue;
6499
6500 atomic_inc(&clp->cl_refcount);
6501 WARN_ON(!unhash_delegation_locked(dp));
6502 list_add(&dp->dl_recall_lru, victims);
6503 }
6504 ++count;
6505 /*
6506 * Despite the fact that these functions deal with
6507 * 64-bit integers for "count", we must ensure that
6508 * it doesn't blow up the clp->cl_refcount. Throw a
6509 * warning if we start to approach INT_MAX here.
6510 */
6511 WARN_ON_ONCE(count == (INT_MAX / 2));
6512 if (count == max)
6513 break;
6514 }
6515 spin_unlock(&state_lock);
6516 return count;
6517}
6518
6519static u64
6520nfsd_print_client_delegations(struct nfs4_client *clp)
6521{
6522 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6523
6524 nfsd_print_count(clp, count, "delegations");
6525 return count;
6526}
6527
6528u64
6529nfsd_inject_print_delegations(void)
6530{
6531 struct nfs4_client *clp;
6532 u64 count = 0;
6533 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6534 nfsd_net_id);
6535
6536 if (!nfsd_netns_ready(nn))
6537 return 0;
6538
6539 spin_lock(&nn->client_lock);
6540 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6541 count += nfsd_print_client_delegations(clp);
6542 spin_unlock(&nn->client_lock);
6543
6544 return count;
6545}
6546
6547static void
6548nfsd_forget_delegations(struct list_head *reaplist)
6549{
6550 struct nfs4_client *clp;
6551 struct nfs4_delegation *dp, *next;
6552
6553 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6554 list_del_init(&dp->dl_recall_lru);
6555 clp = dp->dl_stid.sc_client;
6556 revoke_delegation(dp);
6557 put_client(clp);
6558 }
6559}
6560
6561u64
6562nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6563 size_t addr_size)
6564{
6565 u64 count = 0;
6566 struct nfs4_client *clp;
6567 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6568 nfsd_net_id);
6569 LIST_HEAD(reaplist);
6570
6571 if (!nfsd_netns_ready(nn))
6572 return count;
6573
6574 spin_lock(&nn->client_lock);
6575 clp = nfsd_find_client(addr, addr_size);
6576 if (clp)
6577 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6578 spin_unlock(&nn->client_lock);
6579
6580 nfsd_forget_delegations(&reaplist);
6581 return count;
6582}
6583
6584u64
6585nfsd_inject_forget_delegations(u64 max)
6586{
6587 u64 count = 0;
6588 struct nfs4_client *clp;
6589 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6590 nfsd_net_id);
6591 LIST_HEAD(reaplist);
6592
6593 if (!nfsd_netns_ready(nn))
6594 return count;
6595
6596 spin_lock(&nn->client_lock);
6597 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6598 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6599 if (max != 0 && count >= max)
6600 break;
6601 }
6602 spin_unlock(&nn->client_lock);
6603 nfsd_forget_delegations(&reaplist);
6604 return count;
6605}
6606
6607static void
6608nfsd_recall_delegations(struct list_head *reaplist)
6609{
6610 struct nfs4_client *clp;
6611 struct nfs4_delegation *dp, *next;
6612
6613 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6614 list_del_init(&dp->dl_recall_lru);
6615 clp = dp->dl_stid.sc_client;
6616 /*
6617 * We skipped all entries that had a zero dl_time before,
6618 * so we can now reset the dl_time back to 0. If a delegation
6619 * break comes in now, then it won't make any difference since
6620 * we're recalling it either way.
6621 */
6622 spin_lock(&state_lock);
6623 dp->dl_time = 0;
6624 spin_unlock(&state_lock);
6625 nfsd_break_one_deleg(dp);
6626 put_client(clp);
6627 }
6628}
6629
6630u64
6631nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6632 size_t addr_size)
6633{
6634 u64 count = 0;
6635 struct nfs4_client *clp;
6636 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6637 nfsd_net_id);
6638 LIST_HEAD(reaplist);
6639
6640 if (!nfsd_netns_ready(nn))
6641 return count;
6642
6643 spin_lock(&nn->client_lock);
6644 clp = nfsd_find_client(addr, addr_size);
6645 if (clp)
6646 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6647 spin_unlock(&nn->client_lock);
6648
6649 nfsd_recall_delegations(&reaplist);
6650 return count;
6651}
6652
6653u64
6654nfsd_inject_recall_delegations(u64 max)
6655{
6656 u64 count = 0;
6657 struct nfs4_client *clp, *next;
6658 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6659 nfsd_net_id);
6660 LIST_HEAD(reaplist);
6661
6662 if (!nfsd_netns_ready(nn))
6663 return count;
6664
6665 spin_lock(&nn->client_lock);
6666 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6667 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6668 if (max != 0 && ++count >= max)
6669 break;
6670 }
6671 spin_unlock(&nn->client_lock);
6672 nfsd_recall_delegations(&reaplist);
6673 return count;
6674}
6675#endif /* CONFIG_NFSD_FAULT_INJECTION */
6676
6677/*
6678 * Since the lifetime of a delegation isn't limited to that of an open, a
6679 * client may quite reasonably hang on to a delegation as long as it has
6680 * the inode cached. This becomes an obvious problem the first time a
6681 * client's inode cache approaches the size of the server's total memory.
6682 *
6683 * For now we avoid this problem by imposing a hard limit on the number
6684 * of delegations, which varies according to the server's memory size.
6685 */
6686static void
6687set_max_delegations(void)
6688{
6689 /*
6690 * Allow at most 4 delegations per megabyte of RAM. Quick
6691 * estimates suggest that in the worst case (where every delegation
6692 * is for a different inode), a delegation could take about 1.5K,
6693 * giving a worst case usage of about 6% of memory.
6694 */
6695 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6696}
6697
6698static int nfs4_state_create_net(struct net *net)
6699{
6700 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6701 int i;
6702
6703 nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6704 CLIENT_HASH_SIZE, GFP_KERNEL);
6705 if (!nn->conf_id_hashtbl)
6706 goto err;
6707 nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6708 CLIENT_HASH_SIZE, GFP_KERNEL);
6709 if (!nn->unconf_id_hashtbl)
6710 goto err_unconf_id;
6711 nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6712 SESSION_HASH_SIZE, GFP_KERNEL);
6713 if (!nn->sessionid_hashtbl)
6714 goto err_sessionid;
6715
6716 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6717 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6718 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6719 }
6720 for (i = 0; i < SESSION_HASH_SIZE; i++)
6721 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6722 nn->conf_name_tree = RB_ROOT;
6723 nn->unconf_name_tree = RB_ROOT;
6724 INIT_LIST_HEAD(&nn->client_lru);
6725 INIT_LIST_HEAD(&nn->close_lru);
6726 INIT_LIST_HEAD(&nn->del_recall_lru);
6727 spin_lock_init(&nn->client_lock);
6728
6729 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6730 get_net(net);
6731
6732 return 0;
6733
6734err_sessionid:
6735 kfree(nn->unconf_id_hashtbl);
6736err_unconf_id:
6737 kfree(nn->conf_id_hashtbl);
6738err:
6739 return -ENOMEM;
6740}
6741
6742static void
6743nfs4_state_destroy_net(struct net *net)
6744{
6745 int i;
6746 struct nfs4_client *clp = NULL;
6747 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6748
6749 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6750 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6751 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6752 destroy_client(clp);
6753 }
6754 }
6755
6756 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6757 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6758 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6759 destroy_client(clp);
6760 }
6761 }
6762
6763 kfree(nn->sessionid_hashtbl);
6764 kfree(nn->unconf_id_hashtbl);
6765 kfree(nn->conf_id_hashtbl);
6766 put_net(net);
6767}
6768
6769int
6770nfs4_state_start_net(struct net *net)
6771{
6772 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6773 int ret;
6774
6775 ret = nfs4_state_create_net(net);
6776 if (ret)
6777 return ret;
6778 nn->boot_time = get_seconds();
6779 nn->grace_ended = false;
6780 nn->nfsd4_manager.block_opens = true;
6781 locks_start_grace(net, &nn->nfsd4_manager);
6782 nfsd4_client_tracking_init(net);
6783 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6784 nn->nfsd4_grace, net);
6785 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6786 return 0;
6787}
6788
6789/* initialization to perform when the nfsd service is started: */
6790
6791int
6792nfs4_state_start(void)
6793{
6794 int ret;
6795
6796 ret = set_callback_cred();
6797 if (ret)
6798 return -ENOMEM;
6799 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
6800 if (laundry_wq == NULL) {
6801 ret = -ENOMEM;
6802 goto out_recovery;
6803 }
6804 ret = nfsd4_create_callback_queue();
6805 if (ret)
6806 goto out_free_laundry;
6807
6808 set_max_delegations();
6809
6810 return 0;
6811
6812out_free_laundry:
6813 destroy_workqueue(laundry_wq);
6814out_recovery:
6815 return ret;
6816}
6817
6818void
6819nfs4_state_shutdown_net(struct net *net)
6820{
6821 struct nfs4_delegation *dp = NULL;
6822 struct list_head *pos, *next, reaplist;
6823 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6824
6825 cancel_delayed_work_sync(&nn->laundromat_work);
6826 locks_end_grace(&nn->nfsd4_manager);
6827
6828 INIT_LIST_HEAD(&reaplist);
6829 spin_lock(&state_lock);
6830 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6831 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6832 WARN_ON(!unhash_delegation_locked(dp));
6833 list_add(&dp->dl_recall_lru, &reaplist);
6834 }
6835 spin_unlock(&state_lock);
6836 list_for_each_safe(pos, next, &reaplist) {
6837 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6838 list_del_init(&dp->dl_recall_lru);
6839 put_clnt_odstate(dp->dl_clnt_odstate);
6840 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6841 nfs4_put_stid(&dp->dl_stid);
6842 }
6843
6844 nfsd4_client_tracking_exit(net);
6845 nfs4_state_destroy_net(net);
6846}
6847
6848void
6849nfs4_state_shutdown(void)
6850{
6851 destroy_workqueue(laundry_wq);
6852 nfsd4_destroy_callback_queue();
6853}
6854
6855static void
6856get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6857{
6858 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6859 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6860}
6861
6862static void
6863put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6864{
6865 if (cstate->minorversion) {
6866 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6867 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6868 }
6869}
6870
6871void
6872clear_current_stateid(struct nfsd4_compound_state *cstate)
6873{
6874 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6875}
6876
6877/*
6878 * functions to set current state id
6879 */
6880void
6881nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6882{
6883 put_stateid(cstate, &odp->od_stateid);
6884}
6885
6886void
6887nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6888{
6889 put_stateid(cstate, &open->op_stateid);
6890}
6891
6892void
6893nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6894{
6895 put_stateid(cstate, &close->cl_stateid);
6896}
6897
6898void
6899nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6900{
6901 put_stateid(cstate, &lock->lk_resp_stateid);
6902}
6903
6904/*
6905 * functions to consume current state id
6906 */
6907
6908void
6909nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6910{
6911 get_stateid(cstate, &odp->od_stateid);
6912}
6913
6914void
6915nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6916{
6917 get_stateid(cstate, &drp->dr_stateid);
6918}
6919
6920void
6921nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6922{
6923 get_stateid(cstate, &fsp->fr_stateid);
6924}
6925
6926void
6927nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6928{
6929 get_stateid(cstate, &setattr->sa_stateid);
6930}
6931
6932void
6933nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6934{
6935 get_stateid(cstate, &close->cl_stateid);
6936}
6937
6938void
6939nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6940{
6941 get_stateid(cstate, &locku->lu_stateid);
6942}
6943
6944void
6945nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6946{
6947 get_stateid(cstate, &read->rd_stateid);
6948}
6949
6950void
6951nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6952{
6953 get_stateid(cstate, &write->wr_stateid);
6954}
1/*
2* Copyright (c) 2001 The Regents of the University of Michigan.
3* All rights reserved.
4*
5* Kendrick Smith <kmsmith@umich.edu>
6* Andy Adamson <kandros@umich.edu>
7*
8* Redistribution and use in source and binary forms, with or without
9* modification, are permitted provided that the following conditions
10* are met:
11*
12* 1. Redistributions of source code must retain the above copyright
13* notice, this list of conditions and the following disclaimer.
14* 2. Redistributions in binary form must reproduce the above copyright
15* notice, this list of conditions and the following disclaimer in the
16* documentation and/or other materials provided with the distribution.
17* 3. Neither the name of the University nor the names of its
18* contributors may be used to endorse or promote products derived
19* from this software without specific prior written permission.
20*
21* THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24* DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30* NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32*
33*/
34
35#include <linux/file.h>
36#include <linux/fs.h>
37#include <linux/slab.h>
38#include <linux/namei.h>
39#include <linux/swap.h>
40#include <linux/pagemap.h>
41#include <linux/ratelimit.h>
42#include <linux/sunrpc/svcauth_gss.h>
43#include <linux/sunrpc/addr.h>
44#include <linux/jhash.h>
45#include <linux/string_helpers.h>
46#include "xdr4.h"
47#include "xdr4cb.h"
48#include "vfs.h"
49#include "current_stateid.h"
50
51#include "netns.h"
52#include "pnfs.h"
53#include "filecache.h"
54
55#define NFSDDBG_FACILITY NFSDDBG_PROC
56
57#define all_ones {{~0,~0},~0}
58static const stateid_t one_stateid = {
59 .si_generation = ~0,
60 .si_opaque = all_ones,
61};
62static const stateid_t zero_stateid = {
63 /* all fields zero */
64};
65static const stateid_t currentstateid = {
66 .si_generation = 1,
67};
68static const stateid_t close_stateid = {
69 .si_generation = 0xffffffffU,
70};
71
72static u64 current_sessionid = 1;
73
74#define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
75#define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
76#define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
77#define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
78
79/* forward declarations */
80static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
81static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
82void nfsd4_end_grace(struct nfsd_net *nn);
83
84/* Locking: */
85
86/*
87 * Currently used for the del_recall_lru and file hash table. In an
88 * effort to decrease the scope of the client_mutex, this spinlock may
89 * eventually cover more:
90 */
91static DEFINE_SPINLOCK(state_lock);
92
93enum nfsd4_st_mutex_lock_subclass {
94 OPEN_STATEID_MUTEX = 0,
95 LOCK_STATEID_MUTEX = 1,
96};
97
98/*
99 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
100 * the refcount on the open stateid to drop.
101 */
102static DECLARE_WAIT_QUEUE_HEAD(close_wq);
103
104/*
105 * A waitqueue where a writer to clients/#/ctl destroying a client can
106 * wait for cl_rpc_users to drop to 0 and then for the client to be
107 * unhashed.
108 */
109static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
110
111static struct kmem_cache *client_slab;
112static struct kmem_cache *openowner_slab;
113static struct kmem_cache *lockowner_slab;
114static struct kmem_cache *file_slab;
115static struct kmem_cache *stateid_slab;
116static struct kmem_cache *deleg_slab;
117static struct kmem_cache *odstate_slab;
118
119static void free_session(struct nfsd4_session *);
120
121static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
122static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
123
124static bool is_session_dead(struct nfsd4_session *ses)
125{
126 return ses->se_flags & NFS4_SESSION_DEAD;
127}
128
129static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
130{
131 if (atomic_read(&ses->se_ref) > ref_held_by_me)
132 return nfserr_jukebox;
133 ses->se_flags |= NFS4_SESSION_DEAD;
134 return nfs_ok;
135}
136
137static bool is_client_expired(struct nfs4_client *clp)
138{
139 return clp->cl_time == 0;
140}
141
142static __be32 get_client_locked(struct nfs4_client *clp)
143{
144 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
145
146 lockdep_assert_held(&nn->client_lock);
147
148 if (is_client_expired(clp))
149 return nfserr_expired;
150 atomic_inc(&clp->cl_rpc_users);
151 return nfs_ok;
152}
153
154/* must be called under the client_lock */
155static inline void
156renew_client_locked(struct nfs4_client *clp)
157{
158 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
159
160 if (is_client_expired(clp)) {
161 WARN_ON(1);
162 printk("%s: client (clientid %08x/%08x) already expired\n",
163 __func__,
164 clp->cl_clientid.cl_boot,
165 clp->cl_clientid.cl_id);
166 return;
167 }
168
169 dprintk("renewing client (clientid %08x/%08x)\n",
170 clp->cl_clientid.cl_boot,
171 clp->cl_clientid.cl_id);
172 list_move_tail(&clp->cl_lru, &nn->client_lru);
173 clp->cl_time = get_seconds();
174}
175
176static void put_client_renew_locked(struct nfs4_client *clp)
177{
178 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
179
180 lockdep_assert_held(&nn->client_lock);
181
182 if (!atomic_dec_and_test(&clp->cl_rpc_users))
183 return;
184 if (!is_client_expired(clp))
185 renew_client_locked(clp);
186 else
187 wake_up_all(&expiry_wq);
188}
189
190static void put_client_renew(struct nfs4_client *clp)
191{
192 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
193
194 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
195 return;
196 if (!is_client_expired(clp))
197 renew_client_locked(clp);
198 else
199 wake_up_all(&expiry_wq);
200 spin_unlock(&nn->client_lock);
201}
202
203static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
204{
205 __be32 status;
206
207 if (is_session_dead(ses))
208 return nfserr_badsession;
209 status = get_client_locked(ses->se_client);
210 if (status)
211 return status;
212 atomic_inc(&ses->se_ref);
213 return nfs_ok;
214}
215
216static void nfsd4_put_session_locked(struct nfsd4_session *ses)
217{
218 struct nfs4_client *clp = ses->se_client;
219 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
220
221 lockdep_assert_held(&nn->client_lock);
222
223 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
224 free_session(ses);
225 put_client_renew_locked(clp);
226}
227
228static void nfsd4_put_session(struct nfsd4_session *ses)
229{
230 struct nfs4_client *clp = ses->se_client;
231 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
232
233 spin_lock(&nn->client_lock);
234 nfsd4_put_session_locked(ses);
235 spin_unlock(&nn->client_lock);
236}
237
238static struct nfsd4_blocked_lock *
239find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
240 struct nfsd_net *nn)
241{
242 struct nfsd4_blocked_lock *cur, *found = NULL;
243
244 spin_lock(&nn->blocked_locks_lock);
245 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
246 if (fh_match(fh, &cur->nbl_fh)) {
247 list_del_init(&cur->nbl_list);
248 list_del_init(&cur->nbl_lru);
249 found = cur;
250 break;
251 }
252 }
253 spin_unlock(&nn->blocked_locks_lock);
254 if (found)
255 locks_delete_block(&found->nbl_lock);
256 return found;
257}
258
259static struct nfsd4_blocked_lock *
260find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
261 struct nfsd_net *nn)
262{
263 struct nfsd4_blocked_lock *nbl;
264
265 nbl = find_blocked_lock(lo, fh, nn);
266 if (!nbl) {
267 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
268 if (nbl) {
269 fh_copy_shallow(&nbl->nbl_fh, fh);
270 locks_init_lock(&nbl->nbl_lock);
271 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
272 &nfsd4_cb_notify_lock_ops,
273 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
274 }
275 }
276 return nbl;
277}
278
279static void
280free_blocked_lock(struct nfsd4_blocked_lock *nbl)
281{
282 locks_delete_block(&nbl->nbl_lock);
283 locks_release_private(&nbl->nbl_lock);
284 kfree(nbl);
285}
286
287static void
288remove_blocked_locks(struct nfs4_lockowner *lo)
289{
290 struct nfs4_client *clp = lo->lo_owner.so_client;
291 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
292 struct nfsd4_blocked_lock *nbl;
293 LIST_HEAD(reaplist);
294
295 /* Dequeue all blocked locks */
296 spin_lock(&nn->blocked_locks_lock);
297 while (!list_empty(&lo->lo_blocked)) {
298 nbl = list_first_entry(&lo->lo_blocked,
299 struct nfsd4_blocked_lock,
300 nbl_list);
301 list_del_init(&nbl->nbl_list);
302 list_move(&nbl->nbl_lru, &reaplist);
303 }
304 spin_unlock(&nn->blocked_locks_lock);
305
306 /* Now free them */
307 while (!list_empty(&reaplist)) {
308 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
309 nbl_lru);
310 list_del_init(&nbl->nbl_lru);
311 free_blocked_lock(nbl);
312 }
313}
314
315static void
316nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
317{
318 struct nfsd4_blocked_lock *nbl = container_of(cb,
319 struct nfsd4_blocked_lock, nbl_cb);
320 locks_delete_block(&nbl->nbl_lock);
321}
322
323static int
324nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
325{
326 /*
327 * Since this is just an optimization, we don't try very hard if it
328 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
329 * just quit trying on anything else.
330 */
331 switch (task->tk_status) {
332 case -NFS4ERR_DELAY:
333 rpc_delay(task, 1 * HZ);
334 return 0;
335 default:
336 return 1;
337 }
338}
339
340static void
341nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
342{
343 struct nfsd4_blocked_lock *nbl = container_of(cb,
344 struct nfsd4_blocked_lock, nbl_cb);
345
346 free_blocked_lock(nbl);
347}
348
349static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
350 .prepare = nfsd4_cb_notify_lock_prepare,
351 .done = nfsd4_cb_notify_lock_done,
352 .release = nfsd4_cb_notify_lock_release,
353};
354
355static inline struct nfs4_stateowner *
356nfs4_get_stateowner(struct nfs4_stateowner *sop)
357{
358 atomic_inc(&sop->so_count);
359 return sop;
360}
361
362static int
363same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
364{
365 return (sop->so_owner.len == owner->len) &&
366 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
367}
368
369static struct nfs4_openowner *
370find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
371 struct nfs4_client *clp)
372{
373 struct nfs4_stateowner *so;
374
375 lockdep_assert_held(&clp->cl_lock);
376
377 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
378 so_strhash) {
379 if (!so->so_is_open_owner)
380 continue;
381 if (same_owner_str(so, &open->op_owner))
382 return openowner(nfs4_get_stateowner(so));
383 }
384 return NULL;
385}
386
387static struct nfs4_openowner *
388find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
389 struct nfs4_client *clp)
390{
391 struct nfs4_openowner *oo;
392
393 spin_lock(&clp->cl_lock);
394 oo = find_openstateowner_str_locked(hashval, open, clp);
395 spin_unlock(&clp->cl_lock);
396 return oo;
397}
398
399static inline u32
400opaque_hashval(const void *ptr, int nbytes)
401{
402 unsigned char *cptr = (unsigned char *) ptr;
403
404 u32 x = 0;
405 while (nbytes--) {
406 x *= 37;
407 x += *cptr++;
408 }
409 return x;
410}
411
412static void nfsd4_free_file_rcu(struct rcu_head *rcu)
413{
414 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
415
416 kmem_cache_free(file_slab, fp);
417}
418
419void
420put_nfs4_file(struct nfs4_file *fi)
421{
422 might_lock(&state_lock);
423
424 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
425 hlist_del_rcu(&fi->fi_hash);
426 spin_unlock(&state_lock);
427 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
428 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
429 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
430 }
431}
432
433static struct nfsd_file *
434__nfs4_get_fd(struct nfs4_file *f, int oflag)
435{
436 if (f->fi_fds[oflag])
437 return nfsd_file_get(f->fi_fds[oflag]);
438 return NULL;
439}
440
441static struct nfsd_file *
442find_writeable_file_locked(struct nfs4_file *f)
443{
444 struct nfsd_file *ret;
445
446 lockdep_assert_held(&f->fi_lock);
447
448 ret = __nfs4_get_fd(f, O_WRONLY);
449 if (!ret)
450 ret = __nfs4_get_fd(f, O_RDWR);
451 return ret;
452}
453
454static struct nfsd_file *
455find_writeable_file(struct nfs4_file *f)
456{
457 struct nfsd_file *ret;
458
459 spin_lock(&f->fi_lock);
460 ret = find_writeable_file_locked(f);
461 spin_unlock(&f->fi_lock);
462
463 return ret;
464}
465
466static struct nfsd_file *
467find_readable_file_locked(struct nfs4_file *f)
468{
469 struct nfsd_file *ret;
470
471 lockdep_assert_held(&f->fi_lock);
472
473 ret = __nfs4_get_fd(f, O_RDONLY);
474 if (!ret)
475 ret = __nfs4_get_fd(f, O_RDWR);
476 return ret;
477}
478
479static struct nfsd_file *
480find_readable_file(struct nfs4_file *f)
481{
482 struct nfsd_file *ret;
483
484 spin_lock(&f->fi_lock);
485 ret = find_readable_file_locked(f);
486 spin_unlock(&f->fi_lock);
487
488 return ret;
489}
490
491struct nfsd_file *
492find_any_file(struct nfs4_file *f)
493{
494 struct nfsd_file *ret;
495
496 spin_lock(&f->fi_lock);
497 ret = __nfs4_get_fd(f, O_RDWR);
498 if (!ret) {
499 ret = __nfs4_get_fd(f, O_WRONLY);
500 if (!ret)
501 ret = __nfs4_get_fd(f, O_RDONLY);
502 }
503 spin_unlock(&f->fi_lock);
504 return ret;
505}
506
507static atomic_long_t num_delegations;
508unsigned long max_delegations;
509
510/*
511 * Open owner state (share locks)
512 */
513
514/* hash tables for lock and open owners */
515#define OWNER_HASH_BITS 8
516#define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
517#define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
518
519static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
520{
521 unsigned int ret;
522
523 ret = opaque_hashval(ownername->data, ownername->len);
524 return ret & OWNER_HASH_MASK;
525}
526
527/* hash table for nfs4_file */
528#define FILE_HASH_BITS 8
529#define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
530
531static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
532{
533 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
534}
535
536static unsigned int file_hashval(struct knfsd_fh *fh)
537{
538 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
539}
540
541static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
542
543static void
544__nfs4_file_get_access(struct nfs4_file *fp, u32 access)
545{
546 lockdep_assert_held(&fp->fi_lock);
547
548 if (access & NFS4_SHARE_ACCESS_WRITE)
549 atomic_inc(&fp->fi_access[O_WRONLY]);
550 if (access & NFS4_SHARE_ACCESS_READ)
551 atomic_inc(&fp->fi_access[O_RDONLY]);
552}
553
554static __be32
555nfs4_file_get_access(struct nfs4_file *fp, u32 access)
556{
557 lockdep_assert_held(&fp->fi_lock);
558
559 /* Does this access mode make sense? */
560 if (access & ~NFS4_SHARE_ACCESS_BOTH)
561 return nfserr_inval;
562
563 /* Does it conflict with a deny mode already set? */
564 if ((access & fp->fi_share_deny) != 0)
565 return nfserr_share_denied;
566
567 __nfs4_file_get_access(fp, access);
568 return nfs_ok;
569}
570
571static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
572{
573 /* Common case is that there is no deny mode. */
574 if (deny) {
575 /* Does this deny mode make sense? */
576 if (deny & ~NFS4_SHARE_DENY_BOTH)
577 return nfserr_inval;
578
579 if ((deny & NFS4_SHARE_DENY_READ) &&
580 atomic_read(&fp->fi_access[O_RDONLY]))
581 return nfserr_share_denied;
582
583 if ((deny & NFS4_SHARE_DENY_WRITE) &&
584 atomic_read(&fp->fi_access[O_WRONLY]))
585 return nfserr_share_denied;
586 }
587 return nfs_ok;
588}
589
590static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
591{
592 might_lock(&fp->fi_lock);
593
594 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
595 struct nfsd_file *f1 = NULL;
596 struct nfsd_file *f2 = NULL;
597
598 swap(f1, fp->fi_fds[oflag]);
599 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
600 swap(f2, fp->fi_fds[O_RDWR]);
601 spin_unlock(&fp->fi_lock);
602 if (f1)
603 nfsd_file_put(f1);
604 if (f2)
605 nfsd_file_put(f2);
606 }
607}
608
609static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
610{
611 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
612
613 if (access & NFS4_SHARE_ACCESS_WRITE)
614 __nfs4_file_put_access(fp, O_WRONLY);
615 if (access & NFS4_SHARE_ACCESS_READ)
616 __nfs4_file_put_access(fp, O_RDONLY);
617}
618
619/*
620 * Allocate a new open/delegation state counter. This is needed for
621 * pNFS for proper return on close semantics.
622 *
623 * Note that we only allocate it for pNFS-enabled exports, otherwise
624 * all pointers to struct nfs4_clnt_odstate are always NULL.
625 */
626static struct nfs4_clnt_odstate *
627alloc_clnt_odstate(struct nfs4_client *clp)
628{
629 struct nfs4_clnt_odstate *co;
630
631 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
632 if (co) {
633 co->co_client = clp;
634 refcount_set(&co->co_odcount, 1);
635 }
636 return co;
637}
638
639static void
640hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
641{
642 struct nfs4_file *fp = co->co_file;
643
644 lockdep_assert_held(&fp->fi_lock);
645 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
646}
647
648static inline void
649get_clnt_odstate(struct nfs4_clnt_odstate *co)
650{
651 if (co)
652 refcount_inc(&co->co_odcount);
653}
654
655static void
656put_clnt_odstate(struct nfs4_clnt_odstate *co)
657{
658 struct nfs4_file *fp;
659
660 if (!co)
661 return;
662
663 fp = co->co_file;
664 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
665 list_del(&co->co_perfile);
666 spin_unlock(&fp->fi_lock);
667
668 nfsd4_return_all_file_layouts(co->co_client, fp);
669 kmem_cache_free(odstate_slab, co);
670 }
671}
672
673static struct nfs4_clnt_odstate *
674find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
675{
676 struct nfs4_clnt_odstate *co;
677 struct nfs4_client *cl;
678
679 if (!new)
680 return NULL;
681
682 cl = new->co_client;
683
684 spin_lock(&fp->fi_lock);
685 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
686 if (co->co_client == cl) {
687 get_clnt_odstate(co);
688 goto out;
689 }
690 }
691 co = new;
692 co->co_file = fp;
693 hash_clnt_odstate_locked(new);
694out:
695 spin_unlock(&fp->fi_lock);
696 return co;
697}
698
699struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
700 void (*sc_free)(struct nfs4_stid *))
701{
702 struct nfs4_stid *stid;
703 int new_id;
704
705 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
706 if (!stid)
707 return NULL;
708
709 idr_preload(GFP_KERNEL);
710 spin_lock(&cl->cl_lock);
711 /* Reserving 0 for start of file in nfsdfs "states" file: */
712 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
713 spin_unlock(&cl->cl_lock);
714 idr_preload_end();
715 if (new_id < 0)
716 goto out_free;
717
718 stid->sc_free = sc_free;
719 stid->sc_client = cl;
720 stid->sc_stateid.si_opaque.so_id = new_id;
721 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
722 /* Will be incremented before return to client: */
723 refcount_set(&stid->sc_count, 1);
724 spin_lock_init(&stid->sc_lock);
725
726 /*
727 * It shouldn't be a problem to reuse an opaque stateid value.
728 * I don't think it is for 4.1. But with 4.0 I worry that, for
729 * example, a stray write retransmission could be accepted by
730 * the server when it should have been rejected. Therefore,
731 * adopt a trick from the sctp code to attempt to maximize the
732 * amount of time until an id is reused, by ensuring they always
733 * "increase" (mod INT_MAX):
734 */
735 return stid;
736out_free:
737 kmem_cache_free(slab, stid);
738 return NULL;
739}
740
741/*
742 * Create a unique stateid_t to represent each COPY.
743 */
744int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
745{
746 int new_id;
747
748 idr_preload(GFP_KERNEL);
749 spin_lock(&nn->s2s_cp_lock);
750 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, copy, 0, 0, GFP_NOWAIT);
751 spin_unlock(&nn->s2s_cp_lock);
752 idr_preload_end();
753 if (new_id < 0)
754 return 0;
755 copy->cp_stateid.si_opaque.so_id = new_id;
756 copy->cp_stateid.si_opaque.so_clid.cl_boot = nn->boot_time;
757 copy->cp_stateid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
758 return 1;
759}
760
761void nfs4_free_cp_state(struct nfsd4_copy *copy)
762{
763 struct nfsd_net *nn;
764
765 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
766 spin_lock(&nn->s2s_cp_lock);
767 idr_remove(&nn->s2s_cp_stateids, copy->cp_stateid.si_opaque.so_id);
768 spin_unlock(&nn->s2s_cp_lock);
769}
770
771static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
772{
773 struct nfs4_stid *stid;
774
775 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
776 if (!stid)
777 return NULL;
778
779 return openlockstateid(stid);
780}
781
782static void nfs4_free_deleg(struct nfs4_stid *stid)
783{
784 kmem_cache_free(deleg_slab, stid);
785 atomic_long_dec(&num_delegations);
786}
787
788/*
789 * When we recall a delegation, we should be careful not to hand it
790 * out again straight away.
791 * To ensure this we keep a pair of bloom filters ('new' and 'old')
792 * in which the filehandles of recalled delegations are "stored".
793 * If a filehandle appear in either filter, a delegation is blocked.
794 * When a delegation is recalled, the filehandle is stored in the "new"
795 * filter.
796 * Every 30 seconds we swap the filters and clear the "new" one,
797 * unless both are empty of course.
798 *
799 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
800 * low 3 bytes as hash-table indices.
801 *
802 * 'blocked_delegations_lock', which is always taken in block_delegations(),
803 * is used to manage concurrent access. Testing does not need the lock
804 * except when swapping the two filters.
805 */
806static DEFINE_SPINLOCK(blocked_delegations_lock);
807static struct bloom_pair {
808 int entries, old_entries;
809 time_t swap_time;
810 int new; /* index into 'set' */
811 DECLARE_BITMAP(set[2], 256);
812} blocked_delegations;
813
814static int delegation_blocked(struct knfsd_fh *fh)
815{
816 u32 hash;
817 struct bloom_pair *bd = &blocked_delegations;
818
819 if (bd->entries == 0)
820 return 0;
821 if (seconds_since_boot() - bd->swap_time > 30) {
822 spin_lock(&blocked_delegations_lock);
823 if (seconds_since_boot() - bd->swap_time > 30) {
824 bd->entries -= bd->old_entries;
825 bd->old_entries = bd->entries;
826 memset(bd->set[bd->new], 0,
827 sizeof(bd->set[0]));
828 bd->new = 1-bd->new;
829 bd->swap_time = seconds_since_boot();
830 }
831 spin_unlock(&blocked_delegations_lock);
832 }
833 hash = jhash(&fh->fh_base, fh->fh_size, 0);
834 if (test_bit(hash&255, bd->set[0]) &&
835 test_bit((hash>>8)&255, bd->set[0]) &&
836 test_bit((hash>>16)&255, bd->set[0]))
837 return 1;
838
839 if (test_bit(hash&255, bd->set[1]) &&
840 test_bit((hash>>8)&255, bd->set[1]) &&
841 test_bit((hash>>16)&255, bd->set[1]))
842 return 1;
843
844 return 0;
845}
846
847static void block_delegations(struct knfsd_fh *fh)
848{
849 u32 hash;
850 struct bloom_pair *bd = &blocked_delegations;
851
852 hash = jhash(&fh->fh_base, fh->fh_size, 0);
853
854 spin_lock(&blocked_delegations_lock);
855 __set_bit(hash&255, bd->set[bd->new]);
856 __set_bit((hash>>8)&255, bd->set[bd->new]);
857 __set_bit((hash>>16)&255, bd->set[bd->new]);
858 if (bd->entries == 0)
859 bd->swap_time = seconds_since_boot();
860 bd->entries += 1;
861 spin_unlock(&blocked_delegations_lock);
862}
863
864static struct nfs4_delegation *
865alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
866 struct svc_fh *current_fh,
867 struct nfs4_clnt_odstate *odstate)
868{
869 struct nfs4_delegation *dp;
870 long n;
871
872 dprintk("NFSD alloc_init_deleg\n");
873 n = atomic_long_inc_return(&num_delegations);
874 if (n < 0 || n > max_delegations)
875 goto out_dec;
876 if (delegation_blocked(¤t_fh->fh_handle))
877 goto out_dec;
878 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
879 if (dp == NULL)
880 goto out_dec;
881
882 /*
883 * delegation seqid's are never incremented. The 4.1 special
884 * meaning of seqid 0 isn't meaningful, really, but let's avoid
885 * 0 anyway just for consistency and use 1:
886 */
887 dp->dl_stid.sc_stateid.si_generation = 1;
888 INIT_LIST_HEAD(&dp->dl_perfile);
889 INIT_LIST_HEAD(&dp->dl_perclnt);
890 INIT_LIST_HEAD(&dp->dl_recall_lru);
891 dp->dl_clnt_odstate = odstate;
892 get_clnt_odstate(odstate);
893 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
894 dp->dl_retries = 1;
895 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
896 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
897 get_nfs4_file(fp);
898 dp->dl_stid.sc_file = fp;
899 return dp;
900out_dec:
901 atomic_long_dec(&num_delegations);
902 return NULL;
903}
904
905void
906nfs4_put_stid(struct nfs4_stid *s)
907{
908 struct nfs4_file *fp = s->sc_file;
909 struct nfs4_client *clp = s->sc_client;
910
911 might_lock(&clp->cl_lock);
912
913 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
914 wake_up_all(&close_wq);
915 return;
916 }
917 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
918 spin_unlock(&clp->cl_lock);
919 s->sc_free(s);
920 if (fp)
921 put_nfs4_file(fp);
922}
923
924void
925nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
926{
927 stateid_t *src = &stid->sc_stateid;
928
929 spin_lock(&stid->sc_lock);
930 if (unlikely(++src->si_generation == 0))
931 src->si_generation = 1;
932 memcpy(dst, src, sizeof(*dst));
933 spin_unlock(&stid->sc_lock);
934}
935
936static void put_deleg_file(struct nfs4_file *fp)
937{
938 struct nfsd_file *nf = NULL;
939
940 spin_lock(&fp->fi_lock);
941 if (--fp->fi_delegees == 0)
942 swap(nf, fp->fi_deleg_file);
943 spin_unlock(&fp->fi_lock);
944
945 if (nf)
946 nfsd_file_put(nf);
947}
948
949static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
950{
951 struct nfs4_file *fp = dp->dl_stid.sc_file;
952 struct nfsd_file *nf = fp->fi_deleg_file;
953
954 WARN_ON_ONCE(!fp->fi_delegees);
955
956 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
957 put_deleg_file(fp);
958}
959
960static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
961{
962 put_clnt_odstate(dp->dl_clnt_odstate);
963 nfs4_unlock_deleg_lease(dp);
964 nfs4_put_stid(&dp->dl_stid);
965}
966
967void nfs4_unhash_stid(struct nfs4_stid *s)
968{
969 s->sc_type = 0;
970}
971
972/**
973 * nfs4_delegation_exists - Discover if this delegation already exists
974 * @clp: a pointer to the nfs4_client we're granting a delegation to
975 * @fp: a pointer to the nfs4_file we're granting a delegation on
976 *
977 * Return:
978 * On success: true iff an existing delegation is found
979 */
980
981static bool
982nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
983{
984 struct nfs4_delegation *searchdp = NULL;
985 struct nfs4_client *searchclp = NULL;
986
987 lockdep_assert_held(&state_lock);
988 lockdep_assert_held(&fp->fi_lock);
989
990 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
991 searchclp = searchdp->dl_stid.sc_client;
992 if (clp == searchclp) {
993 return true;
994 }
995 }
996 return false;
997}
998
999/**
1000 * hash_delegation_locked - Add a delegation to the appropriate lists
1001 * @dp: a pointer to the nfs4_delegation we are adding.
1002 * @fp: a pointer to the nfs4_file we're granting a delegation on
1003 *
1004 * Return:
1005 * On success: NULL if the delegation was successfully hashed.
1006 *
1007 * On error: -EAGAIN if one was previously granted to this
1008 * nfs4_client for this nfs4_file. Delegation is not hashed.
1009 *
1010 */
1011
1012static int
1013hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1014{
1015 struct nfs4_client *clp = dp->dl_stid.sc_client;
1016
1017 lockdep_assert_held(&state_lock);
1018 lockdep_assert_held(&fp->fi_lock);
1019
1020 if (nfs4_delegation_exists(clp, fp))
1021 return -EAGAIN;
1022 refcount_inc(&dp->dl_stid.sc_count);
1023 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1024 list_add(&dp->dl_perfile, &fp->fi_delegations);
1025 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1026 return 0;
1027}
1028
1029static bool
1030unhash_delegation_locked(struct nfs4_delegation *dp)
1031{
1032 struct nfs4_file *fp = dp->dl_stid.sc_file;
1033
1034 lockdep_assert_held(&state_lock);
1035
1036 if (list_empty(&dp->dl_perfile))
1037 return false;
1038
1039 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1040 /* Ensure that deleg break won't try to requeue it */
1041 ++dp->dl_time;
1042 spin_lock(&fp->fi_lock);
1043 list_del_init(&dp->dl_perclnt);
1044 list_del_init(&dp->dl_recall_lru);
1045 list_del_init(&dp->dl_perfile);
1046 spin_unlock(&fp->fi_lock);
1047 return true;
1048}
1049
1050static void destroy_delegation(struct nfs4_delegation *dp)
1051{
1052 bool unhashed;
1053
1054 spin_lock(&state_lock);
1055 unhashed = unhash_delegation_locked(dp);
1056 spin_unlock(&state_lock);
1057 if (unhashed)
1058 destroy_unhashed_deleg(dp);
1059}
1060
1061static void revoke_delegation(struct nfs4_delegation *dp)
1062{
1063 struct nfs4_client *clp = dp->dl_stid.sc_client;
1064
1065 WARN_ON(!list_empty(&dp->dl_recall_lru));
1066
1067 if (clp->cl_minorversion) {
1068 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1069 refcount_inc(&dp->dl_stid.sc_count);
1070 spin_lock(&clp->cl_lock);
1071 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1072 spin_unlock(&clp->cl_lock);
1073 }
1074 destroy_unhashed_deleg(dp);
1075}
1076
1077/*
1078 * SETCLIENTID state
1079 */
1080
1081static unsigned int clientid_hashval(u32 id)
1082{
1083 return id & CLIENT_HASH_MASK;
1084}
1085
1086static unsigned int clientstr_hashval(struct xdr_netobj name)
1087{
1088 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1089}
1090
1091/*
1092 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1093 * st_{access,deny}_bmap field of the stateid, in order to track not
1094 * only what share bits are currently in force, but also what
1095 * combinations of share bits previous opens have used. This allows us
1096 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1097 * return an error if the client attempt to downgrade to a combination
1098 * of share bits not explicable by closing some of its previous opens.
1099 *
1100 * XXX: This enforcement is actually incomplete, since we don't keep
1101 * track of access/deny bit combinations; so, e.g., we allow:
1102 *
1103 * OPEN allow read, deny write
1104 * OPEN allow both, deny none
1105 * DOWNGRADE allow read, deny none
1106 *
1107 * which we should reject.
1108 */
1109static unsigned int
1110bmap_to_share_mode(unsigned long bmap) {
1111 int i;
1112 unsigned int access = 0;
1113
1114 for (i = 1; i < 4; i++) {
1115 if (test_bit(i, &bmap))
1116 access |= i;
1117 }
1118 return access;
1119}
1120
1121/* set share access for a given stateid */
1122static inline void
1123set_access(u32 access, struct nfs4_ol_stateid *stp)
1124{
1125 unsigned char mask = 1 << access;
1126
1127 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1128 stp->st_access_bmap |= mask;
1129}
1130
1131/* clear share access for a given stateid */
1132static inline void
1133clear_access(u32 access, struct nfs4_ol_stateid *stp)
1134{
1135 unsigned char mask = 1 << access;
1136
1137 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1138 stp->st_access_bmap &= ~mask;
1139}
1140
1141/* test whether a given stateid has access */
1142static inline bool
1143test_access(u32 access, struct nfs4_ol_stateid *stp)
1144{
1145 unsigned char mask = 1 << access;
1146
1147 return (bool)(stp->st_access_bmap & mask);
1148}
1149
1150/* set share deny for a given stateid */
1151static inline void
1152set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1153{
1154 unsigned char mask = 1 << deny;
1155
1156 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1157 stp->st_deny_bmap |= mask;
1158}
1159
1160/* clear share deny for a given stateid */
1161static inline void
1162clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1163{
1164 unsigned char mask = 1 << deny;
1165
1166 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1167 stp->st_deny_bmap &= ~mask;
1168}
1169
1170/* test whether a given stateid is denying specific access */
1171static inline bool
1172test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1173{
1174 unsigned char mask = 1 << deny;
1175
1176 return (bool)(stp->st_deny_bmap & mask);
1177}
1178
1179static int nfs4_access_to_omode(u32 access)
1180{
1181 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1182 case NFS4_SHARE_ACCESS_READ:
1183 return O_RDONLY;
1184 case NFS4_SHARE_ACCESS_WRITE:
1185 return O_WRONLY;
1186 case NFS4_SHARE_ACCESS_BOTH:
1187 return O_RDWR;
1188 }
1189 WARN_ON_ONCE(1);
1190 return O_RDONLY;
1191}
1192
1193/*
1194 * A stateid that had a deny mode associated with it is being released
1195 * or downgraded. Recalculate the deny mode on the file.
1196 */
1197static void
1198recalculate_deny_mode(struct nfs4_file *fp)
1199{
1200 struct nfs4_ol_stateid *stp;
1201
1202 spin_lock(&fp->fi_lock);
1203 fp->fi_share_deny = 0;
1204 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1205 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1206 spin_unlock(&fp->fi_lock);
1207}
1208
1209static void
1210reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1211{
1212 int i;
1213 bool change = false;
1214
1215 for (i = 1; i < 4; i++) {
1216 if ((i & deny) != i) {
1217 change = true;
1218 clear_deny(i, stp);
1219 }
1220 }
1221
1222 /* Recalculate per-file deny mode if there was a change */
1223 if (change)
1224 recalculate_deny_mode(stp->st_stid.sc_file);
1225}
1226
1227/* release all access and file references for a given stateid */
1228static void
1229release_all_access(struct nfs4_ol_stateid *stp)
1230{
1231 int i;
1232 struct nfs4_file *fp = stp->st_stid.sc_file;
1233
1234 if (fp && stp->st_deny_bmap != 0)
1235 recalculate_deny_mode(fp);
1236
1237 for (i = 1; i < 4; i++) {
1238 if (test_access(i, stp))
1239 nfs4_file_put_access(stp->st_stid.sc_file, i);
1240 clear_access(i, stp);
1241 }
1242}
1243
1244static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1245{
1246 kfree(sop->so_owner.data);
1247 sop->so_ops->so_free(sop);
1248}
1249
1250static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1251{
1252 struct nfs4_client *clp = sop->so_client;
1253
1254 might_lock(&clp->cl_lock);
1255
1256 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1257 return;
1258 sop->so_ops->so_unhash(sop);
1259 spin_unlock(&clp->cl_lock);
1260 nfs4_free_stateowner(sop);
1261}
1262
1263static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1264{
1265 struct nfs4_file *fp = stp->st_stid.sc_file;
1266
1267 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1268
1269 if (list_empty(&stp->st_perfile))
1270 return false;
1271
1272 spin_lock(&fp->fi_lock);
1273 list_del_init(&stp->st_perfile);
1274 spin_unlock(&fp->fi_lock);
1275 list_del(&stp->st_perstateowner);
1276 return true;
1277}
1278
1279static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1280{
1281 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1282
1283 put_clnt_odstate(stp->st_clnt_odstate);
1284 release_all_access(stp);
1285 if (stp->st_stateowner)
1286 nfs4_put_stateowner(stp->st_stateowner);
1287 kmem_cache_free(stateid_slab, stid);
1288}
1289
1290static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1291{
1292 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1293 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1294 struct nfsd_file *nf;
1295
1296 nf = find_any_file(stp->st_stid.sc_file);
1297 if (nf) {
1298 get_file(nf->nf_file);
1299 filp_close(nf->nf_file, (fl_owner_t)lo);
1300 nfsd_file_put(nf);
1301 }
1302 nfs4_free_ol_stateid(stid);
1303}
1304
1305/*
1306 * Put the persistent reference to an already unhashed generic stateid, while
1307 * holding the cl_lock. If it's the last reference, then put it onto the
1308 * reaplist for later destruction.
1309 */
1310static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1311 struct list_head *reaplist)
1312{
1313 struct nfs4_stid *s = &stp->st_stid;
1314 struct nfs4_client *clp = s->sc_client;
1315
1316 lockdep_assert_held(&clp->cl_lock);
1317
1318 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1319
1320 if (!refcount_dec_and_test(&s->sc_count)) {
1321 wake_up_all(&close_wq);
1322 return;
1323 }
1324
1325 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1326 list_add(&stp->st_locks, reaplist);
1327}
1328
1329static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1330{
1331 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1332
1333 list_del_init(&stp->st_locks);
1334 nfs4_unhash_stid(&stp->st_stid);
1335 return unhash_ol_stateid(stp);
1336}
1337
1338static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1339{
1340 struct nfs4_client *clp = stp->st_stid.sc_client;
1341 bool unhashed;
1342
1343 spin_lock(&clp->cl_lock);
1344 unhashed = unhash_lock_stateid(stp);
1345 spin_unlock(&clp->cl_lock);
1346 if (unhashed)
1347 nfs4_put_stid(&stp->st_stid);
1348}
1349
1350static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1351{
1352 struct nfs4_client *clp = lo->lo_owner.so_client;
1353
1354 lockdep_assert_held(&clp->cl_lock);
1355
1356 list_del_init(&lo->lo_owner.so_strhash);
1357}
1358
1359/*
1360 * Free a list of generic stateids that were collected earlier after being
1361 * fully unhashed.
1362 */
1363static void
1364free_ol_stateid_reaplist(struct list_head *reaplist)
1365{
1366 struct nfs4_ol_stateid *stp;
1367 struct nfs4_file *fp;
1368
1369 might_sleep();
1370
1371 while (!list_empty(reaplist)) {
1372 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1373 st_locks);
1374 list_del(&stp->st_locks);
1375 fp = stp->st_stid.sc_file;
1376 stp->st_stid.sc_free(&stp->st_stid);
1377 if (fp)
1378 put_nfs4_file(fp);
1379 }
1380}
1381
1382static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1383 struct list_head *reaplist)
1384{
1385 struct nfs4_ol_stateid *stp;
1386
1387 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1388
1389 while (!list_empty(&open_stp->st_locks)) {
1390 stp = list_entry(open_stp->st_locks.next,
1391 struct nfs4_ol_stateid, st_locks);
1392 WARN_ON(!unhash_lock_stateid(stp));
1393 put_ol_stateid_locked(stp, reaplist);
1394 }
1395}
1396
1397static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1398 struct list_head *reaplist)
1399{
1400 bool unhashed;
1401
1402 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1403
1404 unhashed = unhash_ol_stateid(stp);
1405 release_open_stateid_locks(stp, reaplist);
1406 return unhashed;
1407}
1408
1409static void release_open_stateid(struct nfs4_ol_stateid *stp)
1410{
1411 LIST_HEAD(reaplist);
1412
1413 spin_lock(&stp->st_stid.sc_client->cl_lock);
1414 if (unhash_open_stateid(stp, &reaplist))
1415 put_ol_stateid_locked(stp, &reaplist);
1416 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1417 free_ol_stateid_reaplist(&reaplist);
1418}
1419
1420static void unhash_openowner_locked(struct nfs4_openowner *oo)
1421{
1422 struct nfs4_client *clp = oo->oo_owner.so_client;
1423
1424 lockdep_assert_held(&clp->cl_lock);
1425
1426 list_del_init(&oo->oo_owner.so_strhash);
1427 list_del_init(&oo->oo_perclient);
1428}
1429
1430static void release_last_closed_stateid(struct nfs4_openowner *oo)
1431{
1432 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1433 nfsd_net_id);
1434 struct nfs4_ol_stateid *s;
1435
1436 spin_lock(&nn->client_lock);
1437 s = oo->oo_last_closed_stid;
1438 if (s) {
1439 list_del_init(&oo->oo_close_lru);
1440 oo->oo_last_closed_stid = NULL;
1441 }
1442 spin_unlock(&nn->client_lock);
1443 if (s)
1444 nfs4_put_stid(&s->st_stid);
1445}
1446
1447static void release_openowner(struct nfs4_openowner *oo)
1448{
1449 struct nfs4_ol_stateid *stp;
1450 struct nfs4_client *clp = oo->oo_owner.so_client;
1451 struct list_head reaplist;
1452
1453 INIT_LIST_HEAD(&reaplist);
1454
1455 spin_lock(&clp->cl_lock);
1456 unhash_openowner_locked(oo);
1457 while (!list_empty(&oo->oo_owner.so_stateids)) {
1458 stp = list_first_entry(&oo->oo_owner.so_stateids,
1459 struct nfs4_ol_stateid, st_perstateowner);
1460 if (unhash_open_stateid(stp, &reaplist))
1461 put_ol_stateid_locked(stp, &reaplist);
1462 }
1463 spin_unlock(&clp->cl_lock);
1464 free_ol_stateid_reaplist(&reaplist);
1465 release_last_closed_stateid(oo);
1466 nfs4_put_stateowner(&oo->oo_owner);
1467}
1468
1469static inline int
1470hash_sessionid(struct nfs4_sessionid *sessionid)
1471{
1472 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1473
1474 return sid->sequence % SESSION_HASH_SIZE;
1475}
1476
1477#ifdef CONFIG_SUNRPC_DEBUG
1478static inline void
1479dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1480{
1481 u32 *ptr = (u32 *)(&sessionid->data[0]);
1482 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1483}
1484#else
1485static inline void
1486dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1487{
1488}
1489#endif
1490
1491/*
1492 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1493 * won't be used for replay.
1494 */
1495void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1496{
1497 struct nfs4_stateowner *so = cstate->replay_owner;
1498
1499 if (nfserr == nfserr_replay_me)
1500 return;
1501
1502 if (!seqid_mutating_err(ntohl(nfserr))) {
1503 nfsd4_cstate_clear_replay(cstate);
1504 return;
1505 }
1506 if (!so)
1507 return;
1508 if (so->so_is_open_owner)
1509 release_last_closed_stateid(openowner(so));
1510 so->so_seqid++;
1511 return;
1512}
1513
1514static void
1515gen_sessionid(struct nfsd4_session *ses)
1516{
1517 struct nfs4_client *clp = ses->se_client;
1518 struct nfsd4_sessionid *sid;
1519
1520 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1521 sid->clientid = clp->cl_clientid;
1522 sid->sequence = current_sessionid++;
1523 sid->reserved = 0;
1524}
1525
1526/*
1527 * The protocol defines ca_maxresponssize_cached to include the size of
1528 * the rpc header, but all we need to cache is the data starting after
1529 * the end of the initial SEQUENCE operation--the rest we regenerate
1530 * each time. Therefore we can advertise a ca_maxresponssize_cached
1531 * value that is the number of bytes in our cache plus a few additional
1532 * bytes. In order to stay on the safe side, and not promise more than
1533 * we can cache, those additional bytes must be the minimum possible: 24
1534 * bytes of rpc header (xid through accept state, with AUTH_NULL
1535 * verifier), 12 for the compound header (with zero-length tag), and 44
1536 * for the SEQUENCE op response:
1537 */
1538#define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1539
1540static void
1541free_session_slots(struct nfsd4_session *ses)
1542{
1543 int i;
1544
1545 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1546 free_svc_cred(&ses->se_slots[i]->sl_cred);
1547 kfree(ses->se_slots[i]);
1548 }
1549}
1550
1551/*
1552 * We don't actually need to cache the rpc and session headers, so we
1553 * can allocate a little less for each slot:
1554 */
1555static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1556{
1557 u32 size;
1558
1559 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1560 size = 0;
1561 else
1562 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1563 return size + sizeof(struct nfsd4_slot);
1564}
1565
1566/*
1567 * XXX: If we run out of reserved DRC memory we could (up to a point)
1568 * re-negotiate active sessions and reduce their slot usage to make
1569 * room for new connections. For now we just fail the create session.
1570 */
1571static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1572{
1573 u32 slotsize = slot_bytes(ca);
1574 u32 num = ca->maxreqs;
1575 unsigned long avail, total_avail;
1576 unsigned int scale_factor;
1577
1578 spin_lock(&nfsd_drc_lock);
1579 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1580 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1581 else
1582 /* We have handed out more space than we chose in
1583 * set_max_drc() to allow. That isn't really a
1584 * problem as long as that doesn't make us think we
1585 * have lots more due to integer overflow.
1586 */
1587 total_avail = 0;
1588 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1589 /*
1590 * Never use more than a fraction of the remaining memory,
1591 * unless it's the only way to give this client a slot.
1592 * The chosen fraction is either 1/8 or 1/number of threads,
1593 * whichever is smaller. This ensures there are adequate
1594 * slots to support multiple clients per thread.
1595 * Give the client one slot even if that would require
1596 * over-allocation--it is better than failure.
1597 */
1598 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1599
1600 avail = clamp_t(unsigned long, avail, slotsize,
1601 total_avail/scale_factor);
1602 num = min_t(int, num, avail / slotsize);
1603 num = max_t(int, num, 1);
1604 nfsd_drc_mem_used += num * slotsize;
1605 spin_unlock(&nfsd_drc_lock);
1606
1607 return num;
1608}
1609
1610static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1611{
1612 int slotsize = slot_bytes(ca);
1613
1614 spin_lock(&nfsd_drc_lock);
1615 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1616 spin_unlock(&nfsd_drc_lock);
1617}
1618
1619static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1620 struct nfsd4_channel_attrs *battrs)
1621{
1622 int numslots = fattrs->maxreqs;
1623 int slotsize = slot_bytes(fattrs);
1624 struct nfsd4_session *new;
1625 int mem, i;
1626
1627 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1628 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1629 mem = numslots * sizeof(struct nfsd4_slot *);
1630
1631 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1632 if (!new)
1633 return NULL;
1634 /* allocate each struct nfsd4_slot and data cache in one piece */
1635 for (i = 0; i < numslots; i++) {
1636 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1637 if (!new->se_slots[i])
1638 goto out_free;
1639 }
1640
1641 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1642 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1643
1644 return new;
1645out_free:
1646 while (i--)
1647 kfree(new->se_slots[i]);
1648 kfree(new);
1649 return NULL;
1650}
1651
1652static void free_conn(struct nfsd4_conn *c)
1653{
1654 svc_xprt_put(c->cn_xprt);
1655 kfree(c);
1656}
1657
1658static void nfsd4_conn_lost(struct svc_xpt_user *u)
1659{
1660 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1661 struct nfs4_client *clp = c->cn_session->se_client;
1662
1663 spin_lock(&clp->cl_lock);
1664 if (!list_empty(&c->cn_persession)) {
1665 list_del(&c->cn_persession);
1666 free_conn(c);
1667 }
1668 nfsd4_probe_callback(clp);
1669 spin_unlock(&clp->cl_lock);
1670}
1671
1672static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1673{
1674 struct nfsd4_conn *conn;
1675
1676 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1677 if (!conn)
1678 return NULL;
1679 svc_xprt_get(rqstp->rq_xprt);
1680 conn->cn_xprt = rqstp->rq_xprt;
1681 conn->cn_flags = flags;
1682 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1683 return conn;
1684}
1685
1686static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1687{
1688 conn->cn_session = ses;
1689 list_add(&conn->cn_persession, &ses->se_conns);
1690}
1691
1692static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1693{
1694 struct nfs4_client *clp = ses->se_client;
1695
1696 spin_lock(&clp->cl_lock);
1697 __nfsd4_hash_conn(conn, ses);
1698 spin_unlock(&clp->cl_lock);
1699}
1700
1701static int nfsd4_register_conn(struct nfsd4_conn *conn)
1702{
1703 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1704 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1705}
1706
1707static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1708{
1709 int ret;
1710
1711 nfsd4_hash_conn(conn, ses);
1712 ret = nfsd4_register_conn(conn);
1713 if (ret)
1714 /* oops; xprt is already down: */
1715 nfsd4_conn_lost(&conn->cn_xpt_user);
1716 /* We may have gained or lost a callback channel: */
1717 nfsd4_probe_callback_sync(ses->se_client);
1718}
1719
1720static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1721{
1722 u32 dir = NFS4_CDFC4_FORE;
1723
1724 if (cses->flags & SESSION4_BACK_CHAN)
1725 dir |= NFS4_CDFC4_BACK;
1726 return alloc_conn(rqstp, dir);
1727}
1728
1729/* must be called under client_lock */
1730static void nfsd4_del_conns(struct nfsd4_session *s)
1731{
1732 struct nfs4_client *clp = s->se_client;
1733 struct nfsd4_conn *c;
1734
1735 spin_lock(&clp->cl_lock);
1736 while (!list_empty(&s->se_conns)) {
1737 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1738 list_del_init(&c->cn_persession);
1739 spin_unlock(&clp->cl_lock);
1740
1741 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1742 free_conn(c);
1743
1744 spin_lock(&clp->cl_lock);
1745 }
1746 spin_unlock(&clp->cl_lock);
1747}
1748
1749static void __free_session(struct nfsd4_session *ses)
1750{
1751 free_session_slots(ses);
1752 kfree(ses);
1753}
1754
1755static void free_session(struct nfsd4_session *ses)
1756{
1757 nfsd4_del_conns(ses);
1758 nfsd4_put_drc_mem(&ses->se_fchannel);
1759 __free_session(ses);
1760}
1761
1762static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1763{
1764 int idx;
1765 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1766
1767 new->se_client = clp;
1768 gen_sessionid(new);
1769
1770 INIT_LIST_HEAD(&new->se_conns);
1771
1772 new->se_cb_seq_nr = 1;
1773 new->se_flags = cses->flags;
1774 new->se_cb_prog = cses->callback_prog;
1775 new->se_cb_sec = cses->cb_sec;
1776 atomic_set(&new->se_ref, 0);
1777 idx = hash_sessionid(&new->se_sessionid);
1778 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1779 spin_lock(&clp->cl_lock);
1780 list_add(&new->se_perclnt, &clp->cl_sessions);
1781 spin_unlock(&clp->cl_lock);
1782
1783 {
1784 struct sockaddr *sa = svc_addr(rqstp);
1785 /*
1786 * This is a little silly; with sessions there's no real
1787 * use for the callback address. Use the peer address
1788 * as a reasonable default for now, but consider fixing
1789 * the rpc client not to require an address in the
1790 * future:
1791 */
1792 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1793 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1794 }
1795}
1796
1797/* caller must hold client_lock */
1798static struct nfsd4_session *
1799__find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1800{
1801 struct nfsd4_session *elem;
1802 int idx;
1803 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1804
1805 lockdep_assert_held(&nn->client_lock);
1806
1807 dump_sessionid(__func__, sessionid);
1808 idx = hash_sessionid(sessionid);
1809 /* Search in the appropriate list */
1810 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1811 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1812 NFS4_MAX_SESSIONID_LEN)) {
1813 return elem;
1814 }
1815 }
1816
1817 dprintk("%s: session not found\n", __func__);
1818 return NULL;
1819}
1820
1821static struct nfsd4_session *
1822find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1823 __be32 *ret)
1824{
1825 struct nfsd4_session *session;
1826 __be32 status = nfserr_badsession;
1827
1828 session = __find_in_sessionid_hashtbl(sessionid, net);
1829 if (!session)
1830 goto out;
1831 status = nfsd4_get_session_locked(session);
1832 if (status)
1833 session = NULL;
1834out:
1835 *ret = status;
1836 return session;
1837}
1838
1839/* caller must hold client_lock */
1840static void
1841unhash_session(struct nfsd4_session *ses)
1842{
1843 struct nfs4_client *clp = ses->se_client;
1844 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1845
1846 lockdep_assert_held(&nn->client_lock);
1847
1848 list_del(&ses->se_hash);
1849 spin_lock(&ses->se_client->cl_lock);
1850 list_del(&ses->se_perclnt);
1851 spin_unlock(&ses->se_client->cl_lock);
1852}
1853
1854/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1855static int
1856STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1857{
1858 /*
1859 * We're assuming the clid was not given out from a boot
1860 * precisely 2^32 (about 136 years) before this one. That seems
1861 * a safe assumption:
1862 */
1863 if (clid->cl_boot == (u32)nn->boot_time)
1864 return 0;
1865 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1866 clid->cl_boot, clid->cl_id, nn->boot_time);
1867 return 1;
1868}
1869
1870/*
1871 * XXX Should we use a slab cache ?
1872 * This type of memory management is somewhat inefficient, but we use it
1873 * anyway since SETCLIENTID is not a common operation.
1874 */
1875static struct nfs4_client *alloc_client(struct xdr_netobj name)
1876{
1877 struct nfs4_client *clp;
1878 int i;
1879
1880 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1881 if (clp == NULL)
1882 return NULL;
1883 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1884 if (clp->cl_name.data == NULL)
1885 goto err_no_name;
1886 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1887 sizeof(struct list_head),
1888 GFP_KERNEL);
1889 if (!clp->cl_ownerstr_hashtbl)
1890 goto err_no_hashtbl;
1891 for (i = 0; i < OWNER_HASH_SIZE; i++)
1892 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1893 INIT_LIST_HEAD(&clp->cl_sessions);
1894 idr_init(&clp->cl_stateids);
1895 atomic_set(&clp->cl_rpc_users, 0);
1896 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1897 INIT_LIST_HEAD(&clp->cl_idhash);
1898 INIT_LIST_HEAD(&clp->cl_openowners);
1899 INIT_LIST_HEAD(&clp->cl_delegations);
1900 INIT_LIST_HEAD(&clp->cl_lru);
1901 INIT_LIST_HEAD(&clp->cl_revoked);
1902#ifdef CONFIG_NFSD_PNFS
1903 INIT_LIST_HEAD(&clp->cl_lo_states);
1904#endif
1905 INIT_LIST_HEAD(&clp->async_copies);
1906 spin_lock_init(&clp->async_lock);
1907 spin_lock_init(&clp->cl_lock);
1908 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1909 return clp;
1910err_no_hashtbl:
1911 kfree(clp->cl_name.data);
1912err_no_name:
1913 kmem_cache_free(client_slab, clp);
1914 return NULL;
1915}
1916
1917static void __free_client(struct kref *k)
1918{
1919 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1920 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1921
1922 free_svc_cred(&clp->cl_cred);
1923 kfree(clp->cl_ownerstr_hashtbl);
1924 kfree(clp->cl_name.data);
1925 kfree(clp->cl_nii_domain.data);
1926 kfree(clp->cl_nii_name.data);
1927 idr_destroy(&clp->cl_stateids);
1928 kmem_cache_free(client_slab, clp);
1929}
1930
1931static void drop_client(struct nfs4_client *clp)
1932{
1933 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1934}
1935
1936static void
1937free_client(struct nfs4_client *clp)
1938{
1939 while (!list_empty(&clp->cl_sessions)) {
1940 struct nfsd4_session *ses;
1941 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1942 se_perclnt);
1943 list_del(&ses->se_perclnt);
1944 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1945 free_session(ses);
1946 }
1947 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1948 if (clp->cl_nfsd_dentry) {
1949 nfsd_client_rmdir(clp->cl_nfsd_dentry);
1950 clp->cl_nfsd_dentry = NULL;
1951 wake_up_all(&expiry_wq);
1952 }
1953 drop_client(clp);
1954}
1955
1956/* must be called under the client_lock */
1957static void
1958unhash_client_locked(struct nfs4_client *clp)
1959{
1960 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1961 struct nfsd4_session *ses;
1962
1963 lockdep_assert_held(&nn->client_lock);
1964
1965 /* Mark the client as expired! */
1966 clp->cl_time = 0;
1967 /* Make it invisible */
1968 if (!list_empty(&clp->cl_idhash)) {
1969 list_del_init(&clp->cl_idhash);
1970 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1971 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1972 else
1973 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1974 }
1975 list_del_init(&clp->cl_lru);
1976 spin_lock(&clp->cl_lock);
1977 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1978 list_del_init(&ses->se_hash);
1979 spin_unlock(&clp->cl_lock);
1980}
1981
1982static void
1983unhash_client(struct nfs4_client *clp)
1984{
1985 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1986
1987 spin_lock(&nn->client_lock);
1988 unhash_client_locked(clp);
1989 spin_unlock(&nn->client_lock);
1990}
1991
1992static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1993{
1994 if (atomic_read(&clp->cl_rpc_users))
1995 return nfserr_jukebox;
1996 unhash_client_locked(clp);
1997 return nfs_ok;
1998}
1999
2000static void
2001__destroy_client(struct nfs4_client *clp)
2002{
2003 int i;
2004 struct nfs4_openowner *oo;
2005 struct nfs4_delegation *dp;
2006 struct list_head reaplist;
2007
2008 INIT_LIST_HEAD(&reaplist);
2009 spin_lock(&state_lock);
2010 while (!list_empty(&clp->cl_delegations)) {
2011 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2012 WARN_ON(!unhash_delegation_locked(dp));
2013 list_add(&dp->dl_recall_lru, &reaplist);
2014 }
2015 spin_unlock(&state_lock);
2016 while (!list_empty(&reaplist)) {
2017 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2018 list_del_init(&dp->dl_recall_lru);
2019 destroy_unhashed_deleg(dp);
2020 }
2021 while (!list_empty(&clp->cl_revoked)) {
2022 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2023 list_del_init(&dp->dl_recall_lru);
2024 nfs4_put_stid(&dp->dl_stid);
2025 }
2026 while (!list_empty(&clp->cl_openowners)) {
2027 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2028 nfs4_get_stateowner(&oo->oo_owner);
2029 release_openowner(oo);
2030 }
2031 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2032 struct nfs4_stateowner *so, *tmp;
2033
2034 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2035 so_strhash) {
2036 /* Should be no openowners at this point */
2037 WARN_ON_ONCE(so->so_is_open_owner);
2038 remove_blocked_locks(lockowner(so));
2039 }
2040 }
2041 nfsd4_return_all_client_layouts(clp);
2042 nfsd4_shutdown_copy(clp);
2043 nfsd4_shutdown_callback(clp);
2044 if (clp->cl_cb_conn.cb_xprt)
2045 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2046 free_client(clp);
2047 wake_up_all(&expiry_wq);
2048}
2049
2050static void
2051destroy_client(struct nfs4_client *clp)
2052{
2053 unhash_client(clp);
2054 __destroy_client(clp);
2055}
2056
2057static void inc_reclaim_complete(struct nfs4_client *clp)
2058{
2059 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2060
2061 if (!nn->track_reclaim_completes)
2062 return;
2063 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2064 return;
2065 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2066 nn->reclaim_str_hashtbl_size) {
2067 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2068 clp->net->ns.inum);
2069 nfsd4_end_grace(nn);
2070 }
2071}
2072
2073static void expire_client(struct nfs4_client *clp)
2074{
2075 unhash_client(clp);
2076 nfsd4_client_record_remove(clp);
2077 __destroy_client(clp);
2078}
2079
2080static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2081{
2082 memcpy(target->cl_verifier.data, source->data,
2083 sizeof(target->cl_verifier.data));
2084}
2085
2086static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2087{
2088 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2089 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2090}
2091
2092static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2093{
2094 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2095 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2096 GFP_KERNEL);
2097 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2098 if ((source->cr_principal && !target->cr_principal) ||
2099 (source->cr_raw_principal && !target->cr_raw_principal) ||
2100 (source->cr_targ_princ && !target->cr_targ_princ))
2101 return -ENOMEM;
2102
2103 target->cr_flavor = source->cr_flavor;
2104 target->cr_uid = source->cr_uid;
2105 target->cr_gid = source->cr_gid;
2106 target->cr_group_info = source->cr_group_info;
2107 get_group_info(target->cr_group_info);
2108 target->cr_gss_mech = source->cr_gss_mech;
2109 if (source->cr_gss_mech)
2110 gss_mech_get(source->cr_gss_mech);
2111 return 0;
2112}
2113
2114static int
2115compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2116{
2117 if (o1->len < o2->len)
2118 return -1;
2119 if (o1->len > o2->len)
2120 return 1;
2121 return memcmp(o1->data, o2->data, o1->len);
2122}
2123
2124static int
2125same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2126{
2127 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2128}
2129
2130static int
2131same_clid(clientid_t *cl1, clientid_t *cl2)
2132{
2133 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2134}
2135
2136static bool groups_equal(struct group_info *g1, struct group_info *g2)
2137{
2138 int i;
2139
2140 if (g1->ngroups != g2->ngroups)
2141 return false;
2142 for (i=0; i<g1->ngroups; i++)
2143 if (!gid_eq(g1->gid[i], g2->gid[i]))
2144 return false;
2145 return true;
2146}
2147
2148/*
2149 * RFC 3530 language requires clid_inuse be returned when the
2150 * "principal" associated with a requests differs from that previously
2151 * used. We use uid, gid's, and gss principal string as our best
2152 * approximation. We also don't want to allow non-gss use of a client
2153 * established using gss: in theory cr_principal should catch that
2154 * change, but in practice cr_principal can be null even in the gss case
2155 * since gssd doesn't always pass down a principal string.
2156 */
2157static bool is_gss_cred(struct svc_cred *cr)
2158{
2159 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2160 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2161}
2162
2163
2164static bool
2165same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2166{
2167 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2168 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2169 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2170 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2171 return false;
2172 /* XXX: check that cr_targ_princ fields match ? */
2173 if (cr1->cr_principal == cr2->cr_principal)
2174 return true;
2175 if (!cr1->cr_principal || !cr2->cr_principal)
2176 return false;
2177 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2178}
2179
2180static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2181{
2182 struct svc_cred *cr = &rqstp->rq_cred;
2183 u32 service;
2184
2185 if (!cr->cr_gss_mech)
2186 return false;
2187 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2188 return service == RPC_GSS_SVC_INTEGRITY ||
2189 service == RPC_GSS_SVC_PRIVACY;
2190}
2191
2192bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2193{
2194 struct svc_cred *cr = &rqstp->rq_cred;
2195
2196 if (!cl->cl_mach_cred)
2197 return true;
2198 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2199 return false;
2200 if (!svc_rqst_integrity_protected(rqstp))
2201 return false;
2202 if (cl->cl_cred.cr_raw_principal)
2203 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2204 cr->cr_raw_principal);
2205 if (!cr->cr_principal)
2206 return false;
2207 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2208}
2209
2210static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2211{
2212 __be32 verf[2];
2213
2214 /*
2215 * This is opaque to client, so no need to byte-swap. Use
2216 * __force to keep sparse happy
2217 */
2218 verf[0] = (__force __be32)get_seconds();
2219 verf[1] = (__force __be32)nn->clverifier_counter++;
2220 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2221}
2222
2223static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2224{
2225 clp->cl_clientid.cl_boot = nn->boot_time;
2226 clp->cl_clientid.cl_id = nn->clientid_counter++;
2227 gen_confirm(clp, nn);
2228}
2229
2230static struct nfs4_stid *
2231find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2232{
2233 struct nfs4_stid *ret;
2234
2235 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2236 if (!ret || !ret->sc_type)
2237 return NULL;
2238 return ret;
2239}
2240
2241static struct nfs4_stid *
2242find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2243{
2244 struct nfs4_stid *s;
2245
2246 spin_lock(&cl->cl_lock);
2247 s = find_stateid_locked(cl, t);
2248 if (s != NULL) {
2249 if (typemask & s->sc_type)
2250 refcount_inc(&s->sc_count);
2251 else
2252 s = NULL;
2253 }
2254 spin_unlock(&cl->cl_lock);
2255 return s;
2256}
2257
2258static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2259{
2260 struct nfsdfs_client *nc;
2261 nc = get_nfsdfs_client(inode);
2262 if (!nc)
2263 return NULL;
2264 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2265}
2266
2267static void seq_quote_mem(struct seq_file *m, char *data, int len)
2268{
2269 seq_printf(m, "\"");
2270 seq_escape_mem_ascii(m, data, len);
2271 seq_printf(m, "\"");
2272}
2273
2274static int client_info_show(struct seq_file *m, void *v)
2275{
2276 struct inode *inode = m->private;
2277 struct nfs4_client *clp;
2278 u64 clid;
2279
2280 clp = get_nfsdfs_clp(inode);
2281 if (!clp)
2282 return -ENXIO;
2283 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2284 seq_printf(m, "clientid: 0x%llx\n", clid);
2285 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2286 seq_printf(m, "name: ");
2287 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2288 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2289 if (clp->cl_nii_domain.data) {
2290 seq_printf(m, "Implementation domain: ");
2291 seq_quote_mem(m, clp->cl_nii_domain.data,
2292 clp->cl_nii_domain.len);
2293 seq_printf(m, "\nImplementation name: ");
2294 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2295 seq_printf(m, "\nImplementation time: [%ld, %ld]\n",
2296 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2297 }
2298 drop_client(clp);
2299
2300 return 0;
2301}
2302
2303static int client_info_open(struct inode *inode, struct file *file)
2304{
2305 return single_open(file, client_info_show, inode);
2306}
2307
2308static const struct file_operations client_info_fops = {
2309 .open = client_info_open,
2310 .read = seq_read,
2311 .llseek = seq_lseek,
2312 .release = single_release,
2313};
2314
2315static void *states_start(struct seq_file *s, loff_t *pos)
2316 __acquires(&clp->cl_lock)
2317{
2318 struct nfs4_client *clp = s->private;
2319 unsigned long id = *pos;
2320 void *ret;
2321
2322 spin_lock(&clp->cl_lock);
2323 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2324 *pos = id;
2325 return ret;
2326}
2327
2328static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2329{
2330 struct nfs4_client *clp = s->private;
2331 unsigned long id = *pos;
2332 void *ret;
2333
2334 id = *pos;
2335 id++;
2336 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2337 *pos = id;
2338 return ret;
2339}
2340
2341static void states_stop(struct seq_file *s, void *v)
2342 __releases(&clp->cl_lock)
2343{
2344 struct nfs4_client *clp = s->private;
2345
2346 spin_unlock(&clp->cl_lock);
2347}
2348
2349static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2350{
2351 struct inode *inode = f->nf_inode;
2352
2353 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2354 MAJOR(inode->i_sb->s_dev),
2355 MINOR(inode->i_sb->s_dev),
2356 inode->i_ino);
2357}
2358
2359static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2360{
2361 seq_printf(s, "owner: ");
2362 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2363}
2364
2365static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2366{
2367 struct nfs4_ol_stateid *ols;
2368 struct nfs4_file *nf;
2369 struct nfsd_file *file;
2370 struct nfs4_stateowner *oo;
2371 unsigned int access, deny;
2372
2373 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2374 return 0; /* XXX: or SEQ_SKIP? */
2375 ols = openlockstateid(st);
2376 oo = ols->st_stateowner;
2377 nf = st->sc_file;
2378 file = find_any_file(nf);
2379
2380 seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2381
2382 access = bmap_to_share_mode(ols->st_access_bmap);
2383 deny = bmap_to_share_mode(ols->st_deny_bmap);
2384
2385 seq_printf(s, "access: \%s\%s, ",
2386 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2387 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2388 seq_printf(s, "deny: \%s\%s, ",
2389 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2390 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2391
2392 nfs4_show_superblock(s, file);
2393 seq_printf(s, ", ");
2394 nfs4_show_owner(s, oo);
2395 seq_printf(s, " }\n");
2396 nfsd_file_put(file);
2397
2398 return 0;
2399}
2400
2401static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2402{
2403 struct nfs4_ol_stateid *ols;
2404 struct nfs4_file *nf;
2405 struct nfsd_file *file;
2406 struct nfs4_stateowner *oo;
2407
2408 ols = openlockstateid(st);
2409 oo = ols->st_stateowner;
2410 nf = st->sc_file;
2411 file = find_any_file(nf);
2412
2413 seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2414
2415 /*
2416 * Note: a lock stateid isn't really the same thing as a lock,
2417 * it's the locking state held by one owner on a file, and there
2418 * may be multiple (or no) lock ranges associated with it.
2419 * (Same for the matter is true of open stateids.)
2420 */
2421
2422 nfs4_show_superblock(s, file);
2423 /* XXX: open stateid? */
2424 seq_printf(s, ", ");
2425 nfs4_show_owner(s, oo);
2426 seq_printf(s, " }\n");
2427 nfsd_file_put(file);
2428
2429 return 0;
2430}
2431
2432static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2433{
2434 struct nfs4_delegation *ds;
2435 struct nfs4_file *nf;
2436 struct nfsd_file *file;
2437
2438 ds = delegstateid(st);
2439 nf = st->sc_file;
2440 file = nf->fi_deleg_file;
2441
2442 seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2443
2444 /* Kinda dead code as long as we only support read delegs: */
2445 seq_printf(s, "access: %s, ",
2446 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2447
2448 /* XXX: lease time, whether it's being recalled. */
2449
2450 nfs4_show_superblock(s, file);
2451 seq_printf(s, " }\n");
2452
2453 return 0;
2454}
2455
2456static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2457{
2458 struct nfs4_layout_stateid *ls;
2459 struct nfsd_file *file;
2460
2461 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2462 file = ls->ls_file;
2463
2464 seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2465
2466 /* XXX: What else would be useful? */
2467
2468 nfs4_show_superblock(s, file);
2469 seq_printf(s, " }\n");
2470
2471 return 0;
2472}
2473
2474static int states_show(struct seq_file *s, void *v)
2475{
2476 struct nfs4_stid *st = v;
2477
2478 switch (st->sc_type) {
2479 case NFS4_OPEN_STID:
2480 return nfs4_show_open(s, st);
2481 case NFS4_LOCK_STID:
2482 return nfs4_show_lock(s, st);
2483 case NFS4_DELEG_STID:
2484 return nfs4_show_deleg(s, st);
2485 case NFS4_LAYOUT_STID:
2486 return nfs4_show_layout(s, st);
2487 default:
2488 return 0; /* XXX: or SEQ_SKIP? */
2489 }
2490 /* XXX: copy stateids? */
2491}
2492
2493static struct seq_operations states_seq_ops = {
2494 .start = states_start,
2495 .next = states_next,
2496 .stop = states_stop,
2497 .show = states_show
2498};
2499
2500static int client_states_open(struct inode *inode, struct file *file)
2501{
2502 struct seq_file *s;
2503 struct nfs4_client *clp;
2504 int ret;
2505
2506 clp = get_nfsdfs_clp(inode);
2507 if (!clp)
2508 return -ENXIO;
2509
2510 ret = seq_open(file, &states_seq_ops);
2511 if (ret)
2512 return ret;
2513 s = file->private_data;
2514 s->private = clp;
2515 return 0;
2516}
2517
2518static int client_opens_release(struct inode *inode, struct file *file)
2519{
2520 struct seq_file *m = file->private_data;
2521 struct nfs4_client *clp = m->private;
2522
2523 /* XXX: alternatively, we could get/drop in seq start/stop */
2524 drop_client(clp);
2525 return 0;
2526}
2527
2528static const struct file_operations client_states_fops = {
2529 .open = client_states_open,
2530 .read = seq_read,
2531 .llseek = seq_lseek,
2532 .release = client_opens_release,
2533};
2534
2535/*
2536 * Normally we refuse to destroy clients that are in use, but here the
2537 * administrator is telling us to just do it. We also want to wait
2538 * so the caller has a guarantee that the client's locks are gone by
2539 * the time the write returns:
2540 */
2541static void force_expire_client(struct nfs4_client *clp)
2542{
2543 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2544 bool already_expired;
2545
2546 spin_lock(&clp->cl_lock);
2547 clp->cl_time = 0;
2548 spin_unlock(&clp->cl_lock);
2549
2550 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2551 spin_lock(&nn->client_lock);
2552 already_expired = list_empty(&clp->cl_lru);
2553 if (!already_expired)
2554 unhash_client_locked(clp);
2555 spin_unlock(&nn->client_lock);
2556
2557 if (!already_expired)
2558 expire_client(clp);
2559 else
2560 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2561}
2562
2563static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2564 size_t size, loff_t *pos)
2565{
2566 char *data;
2567 struct nfs4_client *clp;
2568
2569 data = simple_transaction_get(file, buf, size);
2570 if (IS_ERR(data))
2571 return PTR_ERR(data);
2572 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2573 return -EINVAL;
2574 clp = get_nfsdfs_clp(file_inode(file));
2575 if (!clp)
2576 return -ENXIO;
2577 force_expire_client(clp);
2578 drop_client(clp);
2579 return 7;
2580}
2581
2582static const struct file_operations client_ctl_fops = {
2583 .write = client_ctl_write,
2584 .release = simple_transaction_release,
2585};
2586
2587static const struct tree_descr client_files[] = {
2588 [0] = {"info", &client_info_fops, S_IRUSR},
2589 [1] = {"states", &client_states_fops, S_IRUSR},
2590 [2] = {"ctl", &client_ctl_fops, S_IRUSR|S_IWUSR},
2591 [3] = {""},
2592};
2593
2594static struct nfs4_client *create_client(struct xdr_netobj name,
2595 struct svc_rqst *rqstp, nfs4_verifier *verf)
2596{
2597 struct nfs4_client *clp;
2598 struct sockaddr *sa = svc_addr(rqstp);
2599 int ret;
2600 struct net *net = SVC_NET(rqstp);
2601 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2602
2603 clp = alloc_client(name);
2604 if (clp == NULL)
2605 return NULL;
2606
2607 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2608 if (ret) {
2609 free_client(clp);
2610 return NULL;
2611 }
2612 gen_clid(clp, nn);
2613 kref_init(&clp->cl_nfsdfs.cl_ref);
2614 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2615 clp->cl_time = get_seconds();
2616 clear_bit(0, &clp->cl_cb_slot_busy);
2617 copy_verf(clp, verf);
2618 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2619 clp->cl_cb_session = NULL;
2620 clp->net = net;
2621 clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2622 clp->cl_clientid.cl_id - nn->clientid_base,
2623 client_files);
2624 if (!clp->cl_nfsd_dentry) {
2625 free_client(clp);
2626 return NULL;
2627 }
2628 return clp;
2629}
2630
2631static void
2632add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2633{
2634 struct rb_node **new = &(root->rb_node), *parent = NULL;
2635 struct nfs4_client *clp;
2636
2637 while (*new) {
2638 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2639 parent = *new;
2640
2641 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2642 new = &((*new)->rb_left);
2643 else
2644 new = &((*new)->rb_right);
2645 }
2646
2647 rb_link_node(&new_clp->cl_namenode, parent, new);
2648 rb_insert_color(&new_clp->cl_namenode, root);
2649}
2650
2651static struct nfs4_client *
2652find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2653{
2654 int cmp;
2655 struct rb_node *node = root->rb_node;
2656 struct nfs4_client *clp;
2657
2658 while (node) {
2659 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2660 cmp = compare_blob(&clp->cl_name, name);
2661 if (cmp > 0)
2662 node = node->rb_left;
2663 else if (cmp < 0)
2664 node = node->rb_right;
2665 else
2666 return clp;
2667 }
2668 return NULL;
2669}
2670
2671static void
2672add_to_unconfirmed(struct nfs4_client *clp)
2673{
2674 unsigned int idhashval;
2675 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2676
2677 lockdep_assert_held(&nn->client_lock);
2678
2679 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2680 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2681 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2682 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2683 renew_client_locked(clp);
2684}
2685
2686static void
2687move_to_confirmed(struct nfs4_client *clp)
2688{
2689 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2690 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2691
2692 lockdep_assert_held(&nn->client_lock);
2693
2694 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2695 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2696 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2697 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2698 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2699 renew_client_locked(clp);
2700}
2701
2702static struct nfs4_client *
2703find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2704{
2705 struct nfs4_client *clp;
2706 unsigned int idhashval = clientid_hashval(clid->cl_id);
2707
2708 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2709 if (same_clid(&clp->cl_clientid, clid)) {
2710 if ((bool)clp->cl_minorversion != sessions)
2711 return NULL;
2712 renew_client_locked(clp);
2713 return clp;
2714 }
2715 }
2716 return NULL;
2717}
2718
2719static struct nfs4_client *
2720find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2721{
2722 struct list_head *tbl = nn->conf_id_hashtbl;
2723
2724 lockdep_assert_held(&nn->client_lock);
2725 return find_client_in_id_table(tbl, clid, sessions);
2726}
2727
2728static struct nfs4_client *
2729find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2730{
2731 struct list_head *tbl = nn->unconf_id_hashtbl;
2732
2733 lockdep_assert_held(&nn->client_lock);
2734 return find_client_in_id_table(tbl, clid, sessions);
2735}
2736
2737static bool clp_used_exchangeid(struct nfs4_client *clp)
2738{
2739 return clp->cl_exchange_flags != 0;
2740}
2741
2742static struct nfs4_client *
2743find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2744{
2745 lockdep_assert_held(&nn->client_lock);
2746 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2747}
2748
2749static struct nfs4_client *
2750find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2751{
2752 lockdep_assert_held(&nn->client_lock);
2753 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2754}
2755
2756static void
2757gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2758{
2759 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2760 struct sockaddr *sa = svc_addr(rqstp);
2761 u32 scopeid = rpc_get_scope_id(sa);
2762 unsigned short expected_family;
2763
2764 /* Currently, we only support tcp and tcp6 for the callback channel */
2765 if (se->se_callback_netid_len == 3 &&
2766 !memcmp(se->se_callback_netid_val, "tcp", 3))
2767 expected_family = AF_INET;
2768 else if (se->se_callback_netid_len == 4 &&
2769 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2770 expected_family = AF_INET6;
2771 else
2772 goto out_err;
2773
2774 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2775 se->se_callback_addr_len,
2776 (struct sockaddr *)&conn->cb_addr,
2777 sizeof(conn->cb_addr));
2778
2779 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2780 goto out_err;
2781
2782 if (conn->cb_addr.ss_family == AF_INET6)
2783 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2784
2785 conn->cb_prog = se->se_callback_prog;
2786 conn->cb_ident = se->se_callback_ident;
2787 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2788 return;
2789out_err:
2790 conn->cb_addr.ss_family = AF_UNSPEC;
2791 conn->cb_addrlen = 0;
2792 dprintk("NFSD: this client (clientid %08x/%08x) "
2793 "will not receive delegations\n",
2794 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2795
2796 return;
2797}
2798
2799/*
2800 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2801 */
2802static void
2803nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2804{
2805 struct xdr_buf *buf = resp->xdr.buf;
2806 struct nfsd4_slot *slot = resp->cstate.slot;
2807 unsigned int base;
2808
2809 dprintk("--> %s slot %p\n", __func__, slot);
2810
2811 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2812 slot->sl_opcnt = resp->opcnt;
2813 slot->sl_status = resp->cstate.status;
2814 free_svc_cred(&slot->sl_cred);
2815 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2816
2817 if (!nfsd4_cache_this(resp)) {
2818 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2819 return;
2820 }
2821 slot->sl_flags |= NFSD4_SLOT_CACHED;
2822
2823 base = resp->cstate.data_offset;
2824 slot->sl_datalen = buf->len - base;
2825 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2826 WARN(1, "%s: sessions DRC could not cache compound\n",
2827 __func__);
2828 return;
2829}
2830
2831/*
2832 * Encode the replay sequence operation from the slot values.
2833 * If cachethis is FALSE encode the uncached rep error on the next
2834 * operation which sets resp->p and increments resp->opcnt for
2835 * nfs4svc_encode_compoundres.
2836 *
2837 */
2838static __be32
2839nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2840 struct nfsd4_compoundres *resp)
2841{
2842 struct nfsd4_op *op;
2843 struct nfsd4_slot *slot = resp->cstate.slot;
2844
2845 /* Encode the replayed sequence operation */
2846 op = &args->ops[resp->opcnt - 1];
2847 nfsd4_encode_operation(resp, op);
2848
2849 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2850 return op->status;
2851 if (args->opcnt == 1) {
2852 /*
2853 * The original operation wasn't a solo sequence--we
2854 * always cache those--so this retry must not match the
2855 * original:
2856 */
2857 op->status = nfserr_seq_false_retry;
2858 } else {
2859 op = &args->ops[resp->opcnt++];
2860 op->status = nfserr_retry_uncached_rep;
2861 nfsd4_encode_operation(resp, op);
2862 }
2863 return op->status;
2864}
2865
2866/*
2867 * The sequence operation is not cached because we can use the slot and
2868 * session values.
2869 */
2870static __be32
2871nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2872 struct nfsd4_sequence *seq)
2873{
2874 struct nfsd4_slot *slot = resp->cstate.slot;
2875 struct xdr_stream *xdr = &resp->xdr;
2876 __be32 *p;
2877 __be32 status;
2878
2879 dprintk("--> %s slot %p\n", __func__, slot);
2880
2881 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2882 if (status)
2883 return status;
2884
2885 p = xdr_reserve_space(xdr, slot->sl_datalen);
2886 if (!p) {
2887 WARN_ON_ONCE(1);
2888 return nfserr_serverfault;
2889 }
2890 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2891 xdr_commit_encode(xdr);
2892
2893 resp->opcnt = slot->sl_opcnt;
2894 return slot->sl_status;
2895}
2896
2897/*
2898 * Set the exchange_id flags returned by the server.
2899 */
2900static void
2901nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2902{
2903#ifdef CONFIG_NFSD_PNFS
2904 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2905#else
2906 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2907#endif
2908
2909 /* Referrals are supported, Migration is not. */
2910 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2911
2912 /* set the wire flags to return to client. */
2913 clid->flags = new->cl_exchange_flags;
2914}
2915
2916static bool client_has_openowners(struct nfs4_client *clp)
2917{
2918 struct nfs4_openowner *oo;
2919
2920 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2921 if (!list_empty(&oo->oo_owner.so_stateids))
2922 return true;
2923 }
2924 return false;
2925}
2926
2927static bool client_has_state(struct nfs4_client *clp)
2928{
2929 return client_has_openowners(clp)
2930#ifdef CONFIG_NFSD_PNFS
2931 || !list_empty(&clp->cl_lo_states)
2932#endif
2933 || !list_empty(&clp->cl_delegations)
2934 || !list_empty(&clp->cl_sessions)
2935 || !list_empty(&clp->async_copies);
2936}
2937
2938static __be32 copy_impl_id(struct nfs4_client *clp,
2939 struct nfsd4_exchange_id *exid)
2940{
2941 if (!exid->nii_domain.data)
2942 return 0;
2943 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
2944 if (!clp->cl_nii_domain.data)
2945 return nfserr_jukebox;
2946 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
2947 if (!clp->cl_nii_name.data)
2948 return nfserr_jukebox;
2949 clp->cl_nii_time.tv_sec = exid->nii_time.tv_sec;
2950 clp->cl_nii_time.tv_nsec = exid->nii_time.tv_nsec;
2951 return 0;
2952}
2953
2954__be32
2955nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2956 union nfsd4_op_u *u)
2957{
2958 struct nfsd4_exchange_id *exid = &u->exchange_id;
2959 struct nfs4_client *conf, *new;
2960 struct nfs4_client *unconf = NULL;
2961 __be32 status;
2962 char addr_str[INET6_ADDRSTRLEN];
2963 nfs4_verifier verf = exid->verifier;
2964 struct sockaddr *sa = svc_addr(rqstp);
2965 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2966 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2967
2968 rpc_ntop(sa, addr_str, sizeof(addr_str));
2969 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2970 "ip_addr=%s flags %x, spa_how %d\n",
2971 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2972 addr_str, exid->flags, exid->spa_how);
2973
2974 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2975 return nfserr_inval;
2976
2977 new = create_client(exid->clname, rqstp, &verf);
2978 if (new == NULL)
2979 return nfserr_jukebox;
2980 status = copy_impl_id(new, exid);
2981 if (status)
2982 goto out_nolock;
2983
2984 switch (exid->spa_how) {
2985 case SP4_MACH_CRED:
2986 exid->spo_must_enforce[0] = 0;
2987 exid->spo_must_enforce[1] = (
2988 1 << (OP_BIND_CONN_TO_SESSION - 32) |
2989 1 << (OP_EXCHANGE_ID - 32) |
2990 1 << (OP_CREATE_SESSION - 32) |
2991 1 << (OP_DESTROY_SESSION - 32) |
2992 1 << (OP_DESTROY_CLIENTID - 32));
2993
2994 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2995 1 << (OP_OPEN_DOWNGRADE) |
2996 1 << (OP_LOCKU) |
2997 1 << (OP_DELEGRETURN));
2998
2999 exid->spo_must_allow[1] &= (
3000 1 << (OP_TEST_STATEID - 32) |
3001 1 << (OP_FREE_STATEID - 32));
3002 if (!svc_rqst_integrity_protected(rqstp)) {
3003 status = nfserr_inval;
3004 goto out_nolock;
3005 }
3006 /*
3007 * Sometimes userspace doesn't give us a principal.
3008 * Which is a bug, really. Anyway, we can't enforce
3009 * MACH_CRED in that case, better to give up now:
3010 */
3011 if (!new->cl_cred.cr_principal &&
3012 !new->cl_cred.cr_raw_principal) {
3013 status = nfserr_serverfault;
3014 goto out_nolock;
3015 }
3016 new->cl_mach_cred = true;
3017 case SP4_NONE:
3018 break;
3019 default: /* checked by xdr code */
3020 WARN_ON_ONCE(1);
3021 /* fall through */
3022 case SP4_SSV:
3023 status = nfserr_encr_alg_unsupp;
3024 goto out_nolock;
3025 }
3026
3027 /* Cases below refer to rfc 5661 section 18.35.4: */
3028 spin_lock(&nn->client_lock);
3029 conf = find_confirmed_client_by_name(&exid->clname, nn);
3030 if (conf) {
3031 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3032 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3033
3034 if (update) {
3035 if (!clp_used_exchangeid(conf)) { /* buggy client */
3036 status = nfserr_inval;
3037 goto out;
3038 }
3039 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3040 status = nfserr_wrong_cred;
3041 goto out;
3042 }
3043 if (!creds_match) { /* case 9 */
3044 status = nfserr_perm;
3045 goto out;
3046 }
3047 if (!verfs_match) { /* case 8 */
3048 status = nfserr_not_same;
3049 goto out;
3050 }
3051 /* case 6 */
3052 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3053 goto out_copy;
3054 }
3055 if (!creds_match) { /* case 3 */
3056 if (client_has_state(conf)) {
3057 status = nfserr_clid_inuse;
3058 goto out;
3059 }
3060 goto out_new;
3061 }
3062 if (verfs_match) { /* case 2 */
3063 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3064 goto out_copy;
3065 }
3066 /* case 5, client reboot */
3067 conf = NULL;
3068 goto out_new;
3069 }
3070
3071 if (update) { /* case 7 */
3072 status = nfserr_noent;
3073 goto out;
3074 }
3075
3076 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3077 if (unconf) /* case 4, possible retry or client restart */
3078 unhash_client_locked(unconf);
3079
3080 /* case 1 (normal case) */
3081out_new:
3082 if (conf) {
3083 status = mark_client_expired_locked(conf);
3084 if (status)
3085 goto out;
3086 }
3087 new->cl_minorversion = cstate->minorversion;
3088 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3089 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3090
3091 add_to_unconfirmed(new);
3092 swap(new, conf);
3093out_copy:
3094 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3095 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3096
3097 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3098 nfsd4_set_ex_flags(conf, exid);
3099
3100 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3101 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3102 status = nfs_ok;
3103
3104out:
3105 spin_unlock(&nn->client_lock);
3106out_nolock:
3107 if (new)
3108 expire_client(new);
3109 if (unconf)
3110 expire_client(unconf);
3111 return status;
3112}
3113
3114static __be32
3115check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3116{
3117 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3118 slot_seqid);
3119
3120 /* The slot is in use, and no response has been sent. */
3121 if (slot_inuse) {
3122 if (seqid == slot_seqid)
3123 return nfserr_jukebox;
3124 else
3125 return nfserr_seq_misordered;
3126 }
3127 /* Note unsigned 32-bit arithmetic handles wraparound: */
3128 if (likely(seqid == slot_seqid + 1))
3129 return nfs_ok;
3130 if (seqid == slot_seqid)
3131 return nfserr_replay_cache;
3132 return nfserr_seq_misordered;
3133}
3134
3135/*
3136 * Cache the create session result into the create session single DRC
3137 * slot cache by saving the xdr structure. sl_seqid has been set.
3138 * Do this for solo or embedded create session operations.
3139 */
3140static void
3141nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3142 struct nfsd4_clid_slot *slot, __be32 nfserr)
3143{
3144 slot->sl_status = nfserr;
3145 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3146}
3147
3148static __be32
3149nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3150 struct nfsd4_clid_slot *slot)
3151{
3152 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3153 return slot->sl_status;
3154}
3155
3156#define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3157 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3158 1 + /* MIN tag is length with zero, only length */ \
3159 3 + /* version, opcount, opcode */ \
3160 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3161 /* seqid, slotID, slotID, cache */ \
3162 4 ) * sizeof(__be32))
3163
3164#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3165 2 + /* verifier: AUTH_NULL, length 0 */\
3166 1 + /* status */ \
3167 1 + /* MIN tag is length with zero, only length */ \
3168 3 + /* opcount, opcode, opstatus*/ \
3169 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3170 /* seqid, slotID, slotID, slotID, status */ \
3171 5 ) * sizeof(__be32))
3172
3173static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3174{
3175 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3176
3177 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3178 return nfserr_toosmall;
3179 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3180 return nfserr_toosmall;
3181 ca->headerpadsz = 0;
3182 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3183 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3184 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3185 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3186 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3187 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3188 /*
3189 * Note decreasing slot size below client's request may make it
3190 * difficult for client to function correctly, whereas
3191 * decreasing the number of slots will (just?) affect
3192 * performance. When short on memory we therefore prefer to
3193 * decrease number of slots instead of their size. Clients that
3194 * request larger slots than they need will get poor results:
3195 * Note that we always allow at least one slot, because our
3196 * accounting is soft and provides no guarantees either way.
3197 */
3198 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3199
3200 return nfs_ok;
3201}
3202
3203/*
3204 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3205 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3206 */
3207#define RPC_MAX_HEADER_WITH_AUTH_SYS \
3208 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3209
3210#define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3211 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3212
3213#define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3214 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3215#define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3216 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3217 sizeof(__be32))
3218
3219static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3220{
3221 ca->headerpadsz = 0;
3222
3223 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3224 return nfserr_toosmall;
3225 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3226 return nfserr_toosmall;
3227 ca->maxresp_cached = 0;
3228 if (ca->maxops < 2)
3229 return nfserr_toosmall;
3230
3231 return nfs_ok;
3232}
3233
3234static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3235{
3236 switch (cbs->flavor) {
3237 case RPC_AUTH_NULL:
3238 case RPC_AUTH_UNIX:
3239 return nfs_ok;
3240 default:
3241 /*
3242 * GSS case: the spec doesn't allow us to return this
3243 * error. But it also doesn't allow us not to support
3244 * GSS.
3245 * I'd rather this fail hard than return some error the
3246 * client might think it can already handle:
3247 */
3248 return nfserr_encr_alg_unsupp;
3249 }
3250}
3251
3252__be32
3253nfsd4_create_session(struct svc_rqst *rqstp,
3254 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3255{
3256 struct nfsd4_create_session *cr_ses = &u->create_session;
3257 struct sockaddr *sa = svc_addr(rqstp);
3258 struct nfs4_client *conf, *unconf;
3259 struct nfs4_client *old = NULL;
3260 struct nfsd4_session *new;
3261 struct nfsd4_conn *conn;
3262 struct nfsd4_clid_slot *cs_slot = NULL;
3263 __be32 status = 0;
3264 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3265
3266 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3267 return nfserr_inval;
3268 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3269 if (status)
3270 return status;
3271 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3272 if (status)
3273 return status;
3274 status = check_backchannel_attrs(&cr_ses->back_channel);
3275 if (status)
3276 goto out_release_drc_mem;
3277 status = nfserr_jukebox;
3278 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3279 if (!new)
3280 goto out_release_drc_mem;
3281 conn = alloc_conn_from_crses(rqstp, cr_ses);
3282 if (!conn)
3283 goto out_free_session;
3284
3285 spin_lock(&nn->client_lock);
3286 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3287 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3288 WARN_ON_ONCE(conf && unconf);
3289
3290 if (conf) {
3291 status = nfserr_wrong_cred;
3292 if (!nfsd4_mach_creds_match(conf, rqstp))
3293 goto out_free_conn;
3294 cs_slot = &conf->cl_cs_slot;
3295 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3296 if (status) {
3297 if (status == nfserr_replay_cache)
3298 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3299 goto out_free_conn;
3300 }
3301 } else if (unconf) {
3302 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3303 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3304 status = nfserr_clid_inuse;
3305 goto out_free_conn;
3306 }
3307 status = nfserr_wrong_cred;
3308 if (!nfsd4_mach_creds_match(unconf, rqstp))
3309 goto out_free_conn;
3310 cs_slot = &unconf->cl_cs_slot;
3311 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3312 if (status) {
3313 /* an unconfirmed replay returns misordered */
3314 status = nfserr_seq_misordered;
3315 goto out_free_conn;
3316 }
3317 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3318 if (old) {
3319 status = mark_client_expired_locked(old);
3320 if (status) {
3321 old = NULL;
3322 goto out_free_conn;
3323 }
3324 }
3325 move_to_confirmed(unconf);
3326 conf = unconf;
3327 } else {
3328 status = nfserr_stale_clientid;
3329 goto out_free_conn;
3330 }
3331 status = nfs_ok;
3332 /* Persistent sessions are not supported */
3333 cr_ses->flags &= ~SESSION4_PERSIST;
3334 /* Upshifting from TCP to RDMA is not supported */
3335 cr_ses->flags &= ~SESSION4_RDMA;
3336
3337 init_session(rqstp, new, conf, cr_ses);
3338 nfsd4_get_session_locked(new);
3339
3340 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3341 NFS4_MAX_SESSIONID_LEN);
3342 cs_slot->sl_seqid++;
3343 cr_ses->seqid = cs_slot->sl_seqid;
3344
3345 /* cache solo and embedded create sessions under the client_lock */
3346 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3347 spin_unlock(&nn->client_lock);
3348 /* init connection and backchannel */
3349 nfsd4_init_conn(rqstp, conn, new);
3350 nfsd4_put_session(new);
3351 if (old)
3352 expire_client(old);
3353 return status;
3354out_free_conn:
3355 spin_unlock(&nn->client_lock);
3356 free_conn(conn);
3357 if (old)
3358 expire_client(old);
3359out_free_session:
3360 __free_session(new);
3361out_release_drc_mem:
3362 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3363 return status;
3364}
3365
3366static __be32 nfsd4_map_bcts_dir(u32 *dir)
3367{
3368 switch (*dir) {
3369 case NFS4_CDFC4_FORE:
3370 case NFS4_CDFC4_BACK:
3371 return nfs_ok;
3372 case NFS4_CDFC4_FORE_OR_BOTH:
3373 case NFS4_CDFC4_BACK_OR_BOTH:
3374 *dir = NFS4_CDFC4_BOTH;
3375 return nfs_ok;
3376 };
3377 return nfserr_inval;
3378}
3379
3380__be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3381 struct nfsd4_compound_state *cstate,
3382 union nfsd4_op_u *u)
3383{
3384 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3385 struct nfsd4_session *session = cstate->session;
3386 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3387 __be32 status;
3388
3389 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3390 if (status)
3391 return status;
3392 spin_lock(&nn->client_lock);
3393 session->se_cb_prog = bc->bc_cb_program;
3394 session->se_cb_sec = bc->bc_cb_sec;
3395 spin_unlock(&nn->client_lock);
3396
3397 nfsd4_probe_callback(session->se_client);
3398
3399 return nfs_ok;
3400}
3401
3402__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3403 struct nfsd4_compound_state *cstate,
3404 union nfsd4_op_u *u)
3405{
3406 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3407 __be32 status;
3408 struct nfsd4_conn *conn;
3409 struct nfsd4_session *session;
3410 struct net *net = SVC_NET(rqstp);
3411 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3412
3413 if (!nfsd4_last_compound_op(rqstp))
3414 return nfserr_not_only_op;
3415 spin_lock(&nn->client_lock);
3416 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3417 spin_unlock(&nn->client_lock);
3418 if (!session)
3419 goto out_no_session;
3420 status = nfserr_wrong_cred;
3421 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3422 goto out;
3423 status = nfsd4_map_bcts_dir(&bcts->dir);
3424 if (status)
3425 goto out;
3426 conn = alloc_conn(rqstp, bcts->dir);
3427 status = nfserr_jukebox;
3428 if (!conn)
3429 goto out;
3430 nfsd4_init_conn(rqstp, conn, session);
3431 status = nfs_ok;
3432out:
3433 nfsd4_put_session(session);
3434out_no_session:
3435 return status;
3436}
3437
3438static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3439{
3440 if (!cstate->session)
3441 return false;
3442 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3443}
3444
3445__be32
3446nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3447 union nfsd4_op_u *u)
3448{
3449 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3450 struct nfsd4_session *ses;
3451 __be32 status;
3452 int ref_held_by_me = 0;
3453 struct net *net = SVC_NET(r);
3454 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3455
3456 status = nfserr_not_only_op;
3457 if (nfsd4_compound_in_session(cstate, sessionid)) {
3458 if (!nfsd4_last_compound_op(r))
3459 goto out;
3460 ref_held_by_me++;
3461 }
3462 dump_sessionid(__func__, sessionid);
3463 spin_lock(&nn->client_lock);
3464 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3465 if (!ses)
3466 goto out_client_lock;
3467 status = nfserr_wrong_cred;
3468 if (!nfsd4_mach_creds_match(ses->se_client, r))
3469 goto out_put_session;
3470 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3471 if (status)
3472 goto out_put_session;
3473 unhash_session(ses);
3474 spin_unlock(&nn->client_lock);
3475
3476 nfsd4_probe_callback_sync(ses->se_client);
3477
3478 spin_lock(&nn->client_lock);
3479 status = nfs_ok;
3480out_put_session:
3481 nfsd4_put_session_locked(ses);
3482out_client_lock:
3483 spin_unlock(&nn->client_lock);
3484out:
3485 return status;
3486}
3487
3488static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3489{
3490 struct nfsd4_conn *c;
3491
3492 list_for_each_entry(c, &s->se_conns, cn_persession) {
3493 if (c->cn_xprt == xpt) {
3494 return c;
3495 }
3496 }
3497 return NULL;
3498}
3499
3500static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3501{
3502 struct nfs4_client *clp = ses->se_client;
3503 struct nfsd4_conn *c;
3504 __be32 status = nfs_ok;
3505 int ret;
3506
3507 spin_lock(&clp->cl_lock);
3508 c = __nfsd4_find_conn(new->cn_xprt, ses);
3509 if (c)
3510 goto out_free;
3511 status = nfserr_conn_not_bound_to_session;
3512 if (clp->cl_mach_cred)
3513 goto out_free;
3514 __nfsd4_hash_conn(new, ses);
3515 spin_unlock(&clp->cl_lock);
3516 ret = nfsd4_register_conn(new);
3517 if (ret)
3518 /* oops; xprt is already down: */
3519 nfsd4_conn_lost(&new->cn_xpt_user);
3520 return nfs_ok;
3521out_free:
3522 spin_unlock(&clp->cl_lock);
3523 free_conn(new);
3524 return status;
3525}
3526
3527static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3528{
3529 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3530
3531 return args->opcnt > session->se_fchannel.maxops;
3532}
3533
3534static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3535 struct nfsd4_session *session)
3536{
3537 struct xdr_buf *xb = &rqstp->rq_arg;
3538
3539 return xb->len > session->se_fchannel.maxreq_sz;
3540}
3541
3542static bool replay_matches_cache(struct svc_rqst *rqstp,
3543 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3544{
3545 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3546
3547 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3548 (bool)seq->cachethis)
3549 return false;
3550 /*
3551 * If there's an error than the reply can have fewer ops than
3552 * the call. But if we cached a reply with *more* ops than the
3553 * call you're sending us now, then this new call is clearly not
3554 * really a replay of the old one:
3555 */
3556 if (slot->sl_opcnt < argp->opcnt)
3557 return false;
3558 /* This is the only check explicitly called by spec: */
3559 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3560 return false;
3561 /*
3562 * There may be more comparisons we could actually do, but the
3563 * spec doesn't require us to catch every case where the calls
3564 * don't match (that would require caching the call as well as
3565 * the reply), so we don't bother.
3566 */
3567 return true;
3568}
3569
3570__be32
3571nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3572 union nfsd4_op_u *u)
3573{
3574 struct nfsd4_sequence *seq = &u->sequence;
3575 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3576 struct xdr_stream *xdr = &resp->xdr;
3577 struct nfsd4_session *session;
3578 struct nfs4_client *clp;
3579 struct nfsd4_slot *slot;
3580 struct nfsd4_conn *conn;
3581 __be32 status;
3582 int buflen;
3583 struct net *net = SVC_NET(rqstp);
3584 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3585
3586 if (resp->opcnt != 1)
3587 return nfserr_sequence_pos;
3588
3589 /*
3590 * Will be either used or freed by nfsd4_sequence_check_conn
3591 * below.
3592 */
3593 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3594 if (!conn)
3595 return nfserr_jukebox;
3596
3597 spin_lock(&nn->client_lock);
3598 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3599 if (!session)
3600 goto out_no_session;
3601 clp = session->se_client;
3602
3603 status = nfserr_too_many_ops;
3604 if (nfsd4_session_too_many_ops(rqstp, session))
3605 goto out_put_session;
3606
3607 status = nfserr_req_too_big;
3608 if (nfsd4_request_too_big(rqstp, session))
3609 goto out_put_session;
3610
3611 status = nfserr_badslot;
3612 if (seq->slotid >= session->se_fchannel.maxreqs)
3613 goto out_put_session;
3614
3615 slot = session->se_slots[seq->slotid];
3616 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3617
3618 /* We do not negotiate the number of slots yet, so set the
3619 * maxslots to the session maxreqs which is used to encode
3620 * sr_highest_slotid and the sr_target_slot id to maxslots */
3621 seq->maxslots = session->se_fchannel.maxreqs;
3622
3623 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3624 slot->sl_flags & NFSD4_SLOT_INUSE);
3625 if (status == nfserr_replay_cache) {
3626 status = nfserr_seq_misordered;
3627 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3628 goto out_put_session;
3629 status = nfserr_seq_false_retry;
3630 if (!replay_matches_cache(rqstp, seq, slot))
3631 goto out_put_session;
3632 cstate->slot = slot;
3633 cstate->session = session;
3634 cstate->clp = clp;
3635 /* Return the cached reply status and set cstate->status
3636 * for nfsd4_proc_compound processing */
3637 status = nfsd4_replay_cache_entry(resp, seq);
3638 cstate->status = nfserr_replay_cache;
3639 goto out;
3640 }
3641 if (status)
3642 goto out_put_session;
3643
3644 status = nfsd4_sequence_check_conn(conn, session);
3645 conn = NULL;
3646 if (status)
3647 goto out_put_session;
3648
3649 buflen = (seq->cachethis) ?
3650 session->se_fchannel.maxresp_cached :
3651 session->se_fchannel.maxresp_sz;
3652 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3653 nfserr_rep_too_big;
3654 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3655 goto out_put_session;
3656 svc_reserve(rqstp, buflen);
3657
3658 status = nfs_ok;
3659 /* Success! bump slot seqid */
3660 slot->sl_seqid = seq->seqid;
3661 slot->sl_flags |= NFSD4_SLOT_INUSE;
3662 if (seq->cachethis)
3663 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3664 else
3665 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3666
3667 cstate->slot = slot;
3668 cstate->session = session;
3669 cstate->clp = clp;
3670
3671out:
3672 switch (clp->cl_cb_state) {
3673 case NFSD4_CB_DOWN:
3674 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3675 break;
3676 case NFSD4_CB_FAULT:
3677 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3678 break;
3679 default:
3680 seq->status_flags = 0;
3681 }
3682 if (!list_empty(&clp->cl_revoked))
3683 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3684out_no_session:
3685 if (conn)
3686 free_conn(conn);
3687 spin_unlock(&nn->client_lock);
3688 return status;
3689out_put_session:
3690 nfsd4_put_session_locked(session);
3691 goto out_no_session;
3692}
3693
3694void
3695nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3696{
3697 struct nfsd4_compound_state *cs = &resp->cstate;
3698
3699 if (nfsd4_has_session(cs)) {
3700 if (cs->status != nfserr_replay_cache) {
3701 nfsd4_store_cache_entry(resp);
3702 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3703 }
3704 /* Drop session reference that was taken in nfsd4_sequence() */
3705 nfsd4_put_session(cs->session);
3706 } else if (cs->clp)
3707 put_client_renew(cs->clp);
3708}
3709
3710__be32
3711nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3712 struct nfsd4_compound_state *cstate,
3713 union nfsd4_op_u *u)
3714{
3715 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3716 struct nfs4_client *conf, *unconf;
3717 struct nfs4_client *clp = NULL;
3718 __be32 status = 0;
3719 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3720
3721 spin_lock(&nn->client_lock);
3722 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3723 conf = find_confirmed_client(&dc->clientid, true, nn);
3724 WARN_ON_ONCE(conf && unconf);
3725
3726 if (conf) {
3727 if (client_has_state(conf)) {
3728 status = nfserr_clientid_busy;
3729 goto out;
3730 }
3731 status = mark_client_expired_locked(conf);
3732 if (status)
3733 goto out;
3734 clp = conf;
3735 } else if (unconf)
3736 clp = unconf;
3737 else {
3738 status = nfserr_stale_clientid;
3739 goto out;
3740 }
3741 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3742 clp = NULL;
3743 status = nfserr_wrong_cred;
3744 goto out;
3745 }
3746 unhash_client_locked(clp);
3747out:
3748 spin_unlock(&nn->client_lock);
3749 if (clp)
3750 expire_client(clp);
3751 return status;
3752}
3753
3754__be32
3755nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3756 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3757{
3758 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3759 __be32 status = 0;
3760
3761 if (rc->rca_one_fs) {
3762 if (!cstate->current_fh.fh_dentry)
3763 return nfserr_nofilehandle;
3764 /*
3765 * We don't take advantage of the rca_one_fs case.
3766 * That's OK, it's optional, we can safely ignore it.
3767 */
3768 return nfs_ok;
3769 }
3770
3771 status = nfserr_complete_already;
3772 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3773 &cstate->session->se_client->cl_flags))
3774 goto out;
3775
3776 status = nfserr_stale_clientid;
3777 if (is_client_expired(cstate->session->se_client))
3778 /*
3779 * The following error isn't really legal.
3780 * But we only get here if the client just explicitly
3781 * destroyed the client. Surely it no longer cares what
3782 * error it gets back on an operation for the dead
3783 * client.
3784 */
3785 goto out;
3786
3787 status = nfs_ok;
3788 nfsd4_client_record_create(cstate->session->se_client);
3789 inc_reclaim_complete(cstate->session->se_client);
3790out:
3791 return status;
3792}
3793
3794__be32
3795nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3796 union nfsd4_op_u *u)
3797{
3798 struct nfsd4_setclientid *setclid = &u->setclientid;
3799 struct xdr_netobj clname = setclid->se_name;
3800 nfs4_verifier clverifier = setclid->se_verf;
3801 struct nfs4_client *conf, *new;
3802 struct nfs4_client *unconf = NULL;
3803 __be32 status;
3804 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3805
3806 new = create_client(clname, rqstp, &clverifier);
3807 if (new == NULL)
3808 return nfserr_jukebox;
3809 /* Cases below refer to rfc 3530 section 14.2.33: */
3810 spin_lock(&nn->client_lock);
3811 conf = find_confirmed_client_by_name(&clname, nn);
3812 if (conf && client_has_state(conf)) {
3813 /* case 0: */
3814 status = nfserr_clid_inuse;
3815 if (clp_used_exchangeid(conf))
3816 goto out;
3817 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3818 char addr_str[INET6_ADDRSTRLEN];
3819 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3820 sizeof(addr_str));
3821 dprintk("NFSD: setclientid: string in use by client "
3822 "at %s\n", addr_str);
3823 goto out;
3824 }
3825 }
3826 unconf = find_unconfirmed_client_by_name(&clname, nn);
3827 if (unconf)
3828 unhash_client_locked(unconf);
3829 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3830 /* case 1: probable callback update */
3831 copy_clid(new, conf);
3832 gen_confirm(new, nn);
3833 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3834 ;
3835 new->cl_minorversion = 0;
3836 gen_callback(new, setclid, rqstp);
3837 add_to_unconfirmed(new);
3838 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3839 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3840 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3841 new = NULL;
3842 status = nfs_ok;
3843out:
3844 spin_unlock(&nn->client_lock);
3845 if (new)
3846 free_client(new);
3847 if (unconf)
3848 expire_client(unconf);
3849 return status;
3850}
3851
3852
3853__be32
3854nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3855 struct nfsd4_compound_state *cstate,
3856 union nfsd4_op_u *u)
3857{
3858 struct nfsd4_setclientid_confirm *setclientid_confirm =
3859 &u->setclientid_confirm;
3860 struct nfs4_client *conf, *unconf;
3861 struct nfs4_client *old = NULL;
3862 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3863 clientid_t * clid = &setclientid_confirm->sc_clientid;
3864 __be32 status;
3865 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3866
3867 if (STALE_CLIENTID(clid, nn))
3868 return nfserr_stale_clientid;
3869
3870 spin_lock(&nn->client_lock);
3871 conf = find_confirmed_client(clid, false, nn);
3872 unconf = find_unconfirmed_client(clid, false, nn);
3873 /*
3874 * We try hard to give out unique clientid's, so if we get an
3875 * attempt to confirm the same clientid with a different cred,
3876 * the client may be buggy; this should never happen.
3877 *
3878 * Nevertheless, RFC 7530 recommends INUSE for this case:
3879 */
3880 status = nfserr_clid_inuse;
3881 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3882 goto out;
3883 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3884 goto out;
3885 /* cases below refer to rfc 3530 section 14.2.34: */
3886 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3887 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3888 /* case 2: probable retransmit */
3889 status = nfs_ok;
3890 } else /* case 4: client hasn't noticed we rebooted yet? */
3891 status = nfserr_stale_clientid;
3892 goto out;
3893 }
3894 status = nfs_ok;
3895 if (conf) { /* case 1: callback update */
3896 old = unconf;
3897 unhash_client_locked(old);
3898 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3899 } else { /* case 3: normal case; new or rebooted client */
3900 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3901 if (old) {
3902 status = nfserr_clid_inuse;
3903 if (client_has_state(old)
3904 && !same_creds(&unconf->cl_cred,
3905 &old->cl_cred))
3906 goto out;
3907 status = mark_client_expired_locked(old);
3908 if (status) {
3909 old = NULL;
3910 goto out;
3911 }
3912 }
3913 move_to_confirmed(unconf);
3914 conf = unconf;
3915 }
3916 get_client_locked(conf);
3917 spin_unlock(&nn->client_lock);
3918 nfsd4_probe_callback(conf);
3919 spin_lock(&nn->client_lock);
3920 put_client_renew_locked(conf);
3921out:
3922 spin_unlock(&nn->client_lock);
3923 if (old)
3924 expire_client(old);
3925 return status;
3926}
3927
3928static struct nfs4_file *nfsd4_alloc_file(void)
3929{
3930 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3931}
3932
3933/* OPEN Share state helper functions */
3934static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3935 struct nfs4_file *fp)
3936{
3937 lockdep_assert_held(&state_lock);
3938
3939 refcount_set(&fp->fi_ref, 1);
3940 spin_lock_init(&fp->fi_lock);
3941 INIT_LIST_HEAD(&fp->fi_stateids);
3942 INIT_LIST_HEAD(&fp->fi_delegations);
3943 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3944 fh_copy_shallow(&fp->fi_fhandle, fh);
3945 fp->fi_deleg_file = NULL;
3946 fp->fi_had_conflict = false;
3947 fp->fi_share_deny = 0;
3948 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3949 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3950#ifdef CONFIG_NFSD_PNFS
3951 INIT_LIST_HEAD(&fp->fi_lo_states);
3952 atomic_set(&fp->fi_lo_recalls, 0);
3953#endif
3954 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3955}
3956
3957void
3958nfsd4_free_slabs(void)
3959{
3960 kmem_cache_destroy(client_slab);
3961 kmem_cache_destroy(openowner_slab);
3962 kmem_cache_destroy(lockowner_slab);
3963 kmem_cache_destroy(file_slab);
3964 kmem_cache_destroy(stateid_slab);
3965 kmem_cache_destroy(deleg_slab);
3966 kmem_cache_destroy(odstate_slab);
3967}
3968
3969int
3970nfsd4_init_slabs(void)
3971{
3972 client_slab = kmem_cache_create("nfsd4_clients",
3973 sizeof(struct nfs4_client), 0, 0, NULL);
3974 if (client_slab == NULL)
3975 goto out;
3976 openowner_slab = kmem_cache_create("nfsd4_openowners",
3977 sizeof(struct nfs4_openowner), 0, 0, NULL);
3978 if (openowner_slab == NULL)
3979 goto out_free_client_slab;
3980 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3981 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3982 if (lockowner_slab == NULL)
3983 goto out_free_openowner_slab;
3984 file_slab = kmem_cache_create("nfsd4_files",
3985 sizeof(struct nfs4_file), 0, 0, NULL);
3986 if (file_slab == NULL)
3987 goto out_free_lockowner_slab;
3988 stateid_slab = kmem_cache_create("nfsd4_stateids",
3989 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3990 if (stateid_slab == NULL)
3991 goto out_free_file_slab;
3992 deleg_slab = kmem_cache_create("nfsd4_delegations",
3993 sizeof(struct nfs4_delegation), 0, 0, NULL);
3994 if (deleg_slab == NULL)
3995 goto out_free_stateid_slab;
3996 odstate_slab = kmem_cache_create("nfsd4_odstate",
3997 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3998 if (odstate_slab == NULL)
3999 goto out_free_deleg_slab;
4000 return 0;
4001
4002out_free_deleg_slab:
4003 kmem_cache_destroy(deleg_slab);
4004out_free_stateid_slab:
4005 kmem_cache_destroy(stateid_slab);
4006out_free_file_slab:
4007 kmem_cache_destroy(file_slab);
4008out_free_lockowner_slab:
4009 kmem_cache_destroy(lockowner_slab);
4010out_free_openowner_slab:
4011 kmem_cache_destroy(openowner_slab);
4012out_free_client_slab:
4013 kmem_cache_destroy(client_slab);
4014out:
4015 dprintk("nfsd4: out of memory while initializing nfsv4\n");
4016 return -ENOMEM;
4017}
4018
4019static void init_nfs4_replay(struct nfs4_replay *rp)
4020{
4021 rp->rp_status = nfserr_serverfault;
4022 rp->rp_buflen = 0;
4023 rp->rp_buf = rp->rp_ibuf;
4024 mutex_init(&rp->rp_mutex);
4025}
4026
4027static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4028 struct nfs4_stateowner *so)
4029{
4030 if (!nfsd4_has_session(cstate)) {
4031 mutex_lock(&so->so_replay.rp_mutex);
4032 cstate->replay_owner = nfs4_get_stateowner(so);
4033 }
4034}
4035
4036void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4037{
4038 struct nfs4_stateowner *so = cstate->replay_owner;
4039
4040 if (so != NULL) {
4041 cstate->replay_owner = NULL;
4042 mutex_unlock(&so->so_replay.rp_mutex);
4043 nfs4_put_stateowner(so);
4044 }
4045}
4046
4047static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4048{
4049 struct nfs4_stateowner *sop;
4050
4051 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4052 if (!sop)
4053 return NULL;
4054
4055 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4056 if (!sop->so_owner.data) {
4057 kmem_cache_free(slab, sop);
4058 return NULL;
4059 }
4060
4061 INIT_LIST_HEAD(&sop->so_stateids);
4062 sop->so_client = clp;
4063 init_nfs4_replay(&sop->so_replay);
4064 atomic_set(&sop->so_count, 1);
4065 return sop;
4066}
4067
4068static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4069{
4070 lockdep_assert_held(&clp->cl_lock);
4071
4072 list_add(&oo->oo_owner.so_strhash,
4073 &clp->cl_ownerstr_hashtbl[strhashval]);
4074 list_add(&oo->oo_perclient, &clp->cl_openowners);
4075}
4076
4077static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4078{
4079 unhash_openowner_locked(openowner(so));
4080}
4081
4082static void nfs4_free_openowner(struct nfs4_stateowner *so)
4083{
4084 struct nfs4_openowner *oo = openowner(so);
4085
4086 kmem_cache_free(openowner_slab, oo);
4087}
4088
4089static const struct nfs4_stateowner_operations openowner_ops = {
4090 .so_unhash = nfs4_unhash_openowner,
4091 .so_free = nfs4_free_openowner,
4092};
4093
4094static struct nfs4_ol_stateid *
4095nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4096{
4097 struct nfs4_ol_stateid *local, *ret = NULL;
4098 struct nfs4_openowner *oo = open->op_openowner;
4099
4100 lockdep_assert_held(&fp->fi_lock);
4101
4102 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4103 /* ignore lock owners */
4104 if (local->st_stateowner->so_is_open_owner == 0)
4105 continue;
4106 if (local->st_stateowner != &oo->oo_owner)
4107 continue;
4108 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4109 ret = local;
4110 refcount_inc(&ret->st_stid.sc_count);
4111 break;
4112 }
4113 }
4114 return ret;
4115}
4116
4117static __be32
4118nfsd4_verify_open_stid(struct nfs4_stid *s)
4119{
4120 __be32 ret = nfs_ok;
4121
4122 switch (s->sc_type) {
4123 default:
4124 break;
4125 case 0:
4126 case NFS4_CLOSED_STID:
4127 case NFS4_CLOSED_DELEG_STID:
4128 ret = nfserr_bad_stateid;
4129 break;
4130 case NFS4_REVOKED_DELEG_STID:
4131 ret = nfserr_deleg_revoked;
4132 }
4133 return ret;
4134}
4135
4136/* Lock the stateid st_mutex, and deal with races with CLOSE */
4137static __be32
4138nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4139{
4140 __be32 ret;
4141
4142 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4143 ret = nfsd4_verify_open_stid(&stp->st_stid);
4144 if (ret != nfs_ok)
4145 mutex_unlock(&stp->st_mutex);
4146 return ret;
4147}
4148
4149static struct nfs4_ol_stateid *
4150nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4151{
4152 struct nfs4_ol_stateid *stp;
4153 for (;;) {
4154 spin_lock(&fp->fi_lock);
4155 stp = nfsd4_find_existing_open(fp, open);
4156 spin_unlock(&fp->fi_lock);
4157 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4158 break;
4159 nfs4_put_stid(&stp->st_stid);
4160 }
4161 return stp;
4162}
4163
4164static struct nfs4_openowner *
4165alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4166 struct nfsd4_compound_state *cstate)
4167{
4168 struct nfs4_client *clp = cstate->clp;
4169 struct nfs4_openowner *oo, *ret;
4170
4171 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4172 if (!oo)
4173 return NULL;
4174 oo->oo_owner.so_ops = &openowner_ops;
4175 oo->oo_owner.so_is_open_owner = 1;
4176 oo->oo_owner.so_seqid = open->op_seqid;
4177 oo->oo_flags = 0;
4178 if (nfsd4_has_session(cstate))
4179 oo->oo_flags |= NFS4_OO_CONFIRMED;
4180 oo->oo_time = 0;
4181 oo->oo_last_closed_stid = NULL;
4182 INIT_LIST_HEAD(&oo->oo_close_lru);
4183 spin_lock(&clp->cl_lock);
4184 ret = find_openstateowner_str_locked(strhashval, open, clp);
4185 if (ret == NULL) {
4186 hash_openowner(oo, clp, strhashval);
4187 ret = oo;
4188 } else
4189 nfs4_free_stateowner(&oo->oo_owner);
4190
4191 spin_unlock(&clp->cl_lock);
4192 return ret;
4193}
4194
4195static struct nfs4_ol_stateid *
4196init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4197{
4198
4199 struct nfs4_openowner *oo = open->op_openowner;
4200 struct nfs4_ol_stateid *retstp = NULL;
4201 struct nfs4_ol_stateid *stp;
4202
4203 stp = open->op_stp;
4204 /* We are moving these outside of the spinlocks to avoid the warnings */
4205 mutex_init(&stp->st_mutex);
4206 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4207
4208retry:
4209 spin_lock(&oo->oo_owner.so_client->cl_lock);
4210 spin_lock(&fp->fi_lock);
4211
4212 retstp = nfsd4_find_existing_open(fp, open);
4213 if (retstp)
4214 goto out_unlock;
4215
4216 open->op_stp = NULL;
4217 refcount_inc(&stp->st_stid.sc_count);
4218 stp->st_stid.sc_type = NFS4_OPEN_STID;
4219 INIT_LIST_HEAD(&stp->st_locks);
4220 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4221 get_nfs4_file(fp);
4222 stp->st_stid.sc_file = fp;
4223 stp->st_access_bmap = 0;
4224 stp->st_deny_bmap = 0;
4225 stp->st_openstp = NULL;
4226 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4227 list_add(&stp->st_perfile, &fp->fi_stateids);
4228
4229out_unlock:
4230 spin_unlock(&fp->fi_lock);
4231 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4232 if (retstp) {
4233 /* Handle races with CLOSE */
4234 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4235 nfs4_put_stid(&retstp->st_stid);
4236 goto retry;
4237 }
4238 /* To keep mutex tracking happy */
4239 mutex_unlock(&stp->st_mutex);
4240 stp = retstp;
4241 }
4242 return stp;
4243}
4244
4245/*
4246 * In the 4.0 case we need to keep the owners around a little while to handle
4247 * CLOSE replay. We still do need to release any file access that is held by
4248 * them before returning however.
4249 */
4250static void
4251move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4252{
4253 struct nfs4_ol_stateid *last;
4254 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4255 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4256 nfsd_net_id);
4257
4258 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4259
4260 /*
4261 * We know that we hold one reference via nfsd4_close, and another
4262 * "persistent" reference for the client. If the refcount is higher
4263 * than 2, then there are still calls in progress that are using this
4264 * stateid. We can't put the sc_file reference until they are finished.
4265 * Wait for the refcount to drop to 2. Since it has been unhashed,
4266 * there should be no danger of the refcount going back up again at
4267 * this point.
4268 */
4269 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4270
4271 release_all_access(s);
4272 if (s->st_stid.sc_file) {
4273 put_nfs4_file(s->st_stid.sc_file);
4274 s->st_stid.sc_file = NULL;
4275 }
4276
4277 spin_lock(&nn->client_lock);
4278 last = oo->oo_last_closed_stid;
4279 oo->oo_last_closed_stid = s;
4280 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4281 oo->oo_time = get_seconds();
4282 spin_unlock(&nn->client_lock);
4283 if (last)
4284 nfs4_put_stid(&last->st_stid);
4285}
4286
4287/* search file_hashtbl[] for file */
4288static struct nfs4_file *
4289find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4290{
4291 struct nfs4_file *fp;
4292
4293 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
4294 if (fh_match(&fp->fi_fhandle, fh)) {
4295 if (refcount_inc_not_zero(&fp->fi_ref))
4296 return fp;
4297 }
4298 }
4299 return NULL;
4300}
4301
4302struct nfs4_file *
4303find_file(struct knfsd_fh *fh)
4304{
4305 struct nfs4_file *fp;
4306 unsigned int hashval = file_hashval(fh);
4307
4308 rcu_read_lock();
4309 fp = find_file_locked(fh, hashval);
4310 rcu_read_unlock();
4311 return fp;
4312}
4313
4314static struct nfs4_file *
4315find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4316{
4317 struct nfs4_file *fp;
4318 unsigned int hashval = file_hashval(fh);
4319
4320 rcu_read_lock();
4321 fp = find_file_locked(fh, hashval);
4322 rcu_read_unlock();
4323 if (fp)
4324 return fp;
4325
4326 spin_lock(&state_lock);
4327 fp = find_file_locked(fh, hashval);
4328 if (likely(fp == NULL)) {
4329 nfsd4_init_file(fh, hashval, new);
4330 fp = new;
4331 }
4332 spin_unlock(&state_lock);
4333
4334 return fp;
4335}
4336
4337/*
4338 * Called to check deny when READ with all zero stateid or
4339 * WRITE with all zero or all one stateid
4340 */
4341static __be32
4342nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4343{
4344 struct nfs4_file *fp;
4345 __be32 ret = nfs_ok;
4346
4347 fp = find_file(¤t_fh->fh_handle);
4348 if (!fp)
4349 return ret;
4350 /* Check for conflicting share reservations */
4351 spin_lock(&fp->fi_lock);
4352 if (fp->fi_share_deny & deny_type)
4353 ret = nfserr_locked;
4354 spin_unlock(&fp->fi_lock);
4355 put_nfs4_file(fp);
4356 return ret;
4357}
4358
4359static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4360{
4361 struct nfs4_delegation *dp = cb_to_delegation(cb);
4362 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4363 nfsd_net_id);
4364
4365 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4366
4367 /*
4368 * We can't do this in nfsd_break_deleg_cb because it is
4369 * already holding inode->i_lock.
4370 *
4371 * If the dl_time != 0, then we know that it has already been
4372 * queued for a lease break. Don't queue it again.
4373 */
4374 spin_lock(&state_lock);
4375 if (dp->dl_time == 0) {
4376 dp->dl_time = get_seconds();
4377 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4378 }
4379 spin_unlock(&state_lock);
4380}
4381
4382static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4383 struct rpc_task *task)
4384{
4385 struct nfs4_delegation *dp = cb_to_delegation(cb);
4386
4387 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4388 return 1;
4389
4390 switch (task->tk_status) {
4391 case 0:
4392 return 1;
4393 case -NFS4ERR_DELAY:
4394 rpc_delay(task, 2 * HZ);
4395 return 0;
4396 case -EBADHANDLE:
4397 case -NFS4ERR_BAD_STATEID:
4398 /*
4399 * Race: client probably got cb_recall before open reply
4400 * granting delegation.
4401 */
4402 if (dp->dl_retries--) {
4403 rpc_delay(task, 2 * HZ);
4404 return 0;
4405 }
4406 /*FALLTHRU*/
4407 default:
4408 return 1;
4409 }
4410}
4411
4412static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4413{
4414 struct nfs4_delegation *dp = cb_to_delegation(cb);
4415
4416 nfs4_put_stid(&dp->dl_stid);
4417}
4418
4419static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4420 .prepare = nfsd4_cb_recall_prepare,
4421 .done = nfsd4_cb_recall_done,
4422 .release = nfsd4_cb_recall_release,
4423};
4424
4425static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4426{
4427 /*
4428 * We're assuming the state code never drops its reference
4429 * without first removing the lease. Since we're in this lease
4430 * callback (and since the lease code is serialized by the
4431 * i_lock) we know the server hasn't removed the lease yet, and
4432 * we know it's safe to take a reference.
4433 */
4434 refcount_inc(&dp->dl_stid.sc_count);
4435 nfsd4_run_cb(&dp->dl_recall);
4436}
4437
4438/* Called from break_lease() with i_lock held. */
4439static bool
4440nfsd_break_deleg_cb(struct file_lock *fl)
4441{
4442 bool ret = false;
4443 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4444 struct nfs4_file *fp = dp->dl_stid.sc_file;
4445
4446 /*
4447 * We don't want the locks code to timeout the lease for us;
4448 * we'll remove it ourself if a delegation isn't returned
4449 * in time:
4450 */
4451 fl->fl_break_time = 0;
4452
4453 spin_lock(&fp->fi_lock);
4454 fp->fi_had_conflict = true;
4455 nfsd_break_one_deleg(dp);
4456 spin_unlock(&fp->fi_lock);
4457 return ret;
4458}
4459
4460static int
4461nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4462 struct list_head *dispose)
4463{
4464 if (arg & F_UNLCK)
4465 return lease_modify(onlist, arg, dispose);
4466 else
4467 return -EAGAIN;
4468}
4469
4470static const struct lock_manager_operations nfsd_lease_mng_ops = {
4471 .lm_break = nfsd_break_deleg_cb,
4472 .lm_change = nfsd_change_deleg_cb,
4473};
4474
4475static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4476{
4477 if (nfsd4_has_session(cstate))
4478 return nfs_ok;
4479 if (seqid == so->so_seqid - 1)
4480 return nfserr_replay_me;
4481 if (seqid == so->so_seqid)
4482 return nfs_ok;
4483 return nfserr_bad_seqid;
4484}
4485
4486static __be32 lookup_clientid(clientid_t *clid,
4487 struct nfsd4_compound_state *cstate,
4488 struct nfsd_net *nn)
4489{
4490 struct nfs4_client *found;
4491
4492 if (cstate->clp) {
4493 found = cstate->clp;
4494 if (!same_clid(&found->cl_clientid, clid))
4495 return nfserr_stale_clientid;
4496 return nfs_ok;
4497 }
4498
4499 if (STALE_CLIENTID(clid, nn))
4500 return nfserr_stale_clientid;
4501
4502 /*
4503 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4504 * cached already then we know this is for is for v4.0 and "sessions"
4505 * will be false.
4506 */
4507 WARN_ON_ONCE(cstate->session);
4508 spin_lock(&nn->client_lock);
4509 found = find_confirmed_client(clid, false, nn);
4510 if (!found) {
4511 spin_unlock(&nn->client_lock);
4512 return nfserr_expired;
4513 }
4514 atomic_inc(&found->cl_rpc_users);
4515 spin_unlock(&nn->client_lock);
4516
4517 /* Cache the nfs4_client in cstate! */
4518 cstate->clp = found;
4519 return nfs_ok;
4520}
4521
4522__be32
4523nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4524 struct nfsd4_open *open, struct nfsd_net *nn)
4525{
4526 clientid_t *clientid = &open->op_clientid;
4527 struct nfs4_client *clp = NULL;
4528 unsigned int strhashval;
4529 struct nfs4_openowner *oo = NULL;
4530 __be32 status;
4531
4532 if (STALE_CLIENTID(&open->op_clientid, nn))
4533 return nfserr_stale_clientid;
4534 /*
4535 * In case we need it later, after we've already created the
4536 * file and don't want to risk a further failure:
4537 */
4538 open->op_file = nfsd4_alloc_file();
4539 if (open->op_file == NULL)
4540 return nfserr_jukebox;
4541
4542 status = lookup_clientid(clientid, cstate, nn);
4543 if (status)
4544 return status;
4545 clp = cstate->clp;
4546
4547 strhashval = ownerstr_hashval(&open->op_owner);
4548 oo = find_openstateowner_str(strhashval, open, clp);
4549 open->op_openowner = oo;
4550 if (!oo) {
4551 goto new_owner;
4552 }
4553 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4554 /* Replace unconfirmed owners without checking for replay. */
4555 release_openowner(oo);
4556 open->op_openowner = NULL;
4557 goto new_owner;
4558 }
4559 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4560 if (status)
4561 return status;
4562 goto alloc_stateid;
4563new_owner:
4564 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4565 if (oo == NULL)
4566 return nfserr_jukebox;
4567 open->op_openowner = oo;
4568alloc_stateid:
4569 open->op_stp = nfs4_alloc_open_stateid(clp);
4570 if (!open->op_stp)
4571 return nfserr_jukebox;
4572
4573 if (nfsd4_has_session(cstate) &&
4574 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4575 open->op_odstate = alloc_clnt_odstate(clp);
4576 if (!open->op_odstate)
4577 return nfserr_jukebox;
4578 }
4579
4580 return nfs_ok;
4581}
4582
4583static inline __be32
4584nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4585{
4586 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4587 return nfserr_openmode;
4588 else
4589 return nfs_ok;
4590}
4591
4592static int share_access_to_flags(u32 share_access)
4593{
4594 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4595}
4596
4597static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4598{
4599 struct nfs4_stid *ret;
4600
4601 ret = find_stateid_by_type(cl, s,
4602 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4603 if (!ret)
4604 return NULL;
4605 return delegstateid(ret);
4606}
4607
4608static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4609{
4610 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4611 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4612}
4613
4614static __be32
4615nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4616 struct nfs4_delegation **dp)
4617{
4618 int flags;
4619 __be32 status = nfserr_bad_stateid;
4620 struct nfs4_delegation *deleg;
4621
4622 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4623 if (deleg == NULL)
4624 goto out;
4625 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4626 nfs4_put_stid(&deleg->dl_stid);
4627 if (cl->cl_minorversion)
4628 status = nfserr_deleg_revoked;
4629 goto out;
4630 }
4631 flags = share_access_to_flags(open->op_share_access);
4632 status = nfs4_check_delegmode(deleg, flags);
4633 if (status) {
4634 nfs4_put_stid(&deleg->dl_stid);
4635 goto out;
4636 }
4637 *dp = deleg;
4638out:
4639 if (!nfsd4_is_deleg_cur(open))
4640 return nfs_ok;
4641 if (status)
4642 return status;
4643 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4644 return nfs_ok;
4645}
4646
4647static inline int nfs4_access_to_access(u32 nfs4_access)
4648{
4649 int flags = 0;
4650
4651 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4652 flags |= NFSD_MAY_READ;
4653 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4654 flags |= NFSD_MAY_WRITE;
4655 return flags;
4656}
4657
4658static inline __be32
4659nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4660 struct nfsd4_open *open)
4661{
4662 struct iattr iattr = {
4663 .ia_valid = ATTR_SIZE,
4664 .ia_size = 0,
4665 };
4666 if (!open->op_truncate)
4667 return 0;
4668 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4669 return nfserr_inval;
4670 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4671}
4672
4673static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4674 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4675 struct nfsd4_open *open)
4676{
4677 struct nfsd_file *nf = NULL;
4678 __be32 status;
4679 int oflag = nfs4_access_to_omode(open->op_share_access);
4680 int access = nfs4_access_to_access(open->op_share_access);
4681 unsigned char old_access_bmap, old_deny_bmap;
4682
4683 spin_lock(&fp->fi_lock);
4684
4685 /*
4686 * Are we trying to set a deny mode that would conflict with
4687 * current access?
4688 */
4689 status = nfs4_file_check_deny(fp, open->op_share_deny);
4690 if (status != nfs_ok) {
4691 spin_unlock(&fp->fi_lock);
4692 goto out;
4693 }
4694
4695 /* set access to the file */
4696 status = nfs4_file_get_access(fp, open->op_share_access);
4697 if (status != nfs_ok) {
4698 spin_unlock(&fp->fi_lock);
4699 goto out;
4700 }
4701
4702 /* Set access bits in stateid */
4703 old_access_bmap = stp->st_access_bmap;
4704 set_access(open->op_share_access, stp);
4705
4706 /* Set new deny mask */
4707 old_deny_bmap = stp->st_deny_bmap;
4708 set_deny(open->op_share_deny, stp);
4709 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4710
4711 if (!fp->fi_fds[oflag]) {
4712 spin_unlock(&fp->fi_lock);
4713 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4714 if (status)
4715 goto out_put_access;
4716 spin_lock(&fp->fi_lock);
4717 if (!fp->fi_fds[oflag]) {
4718 fp->fi_fds[oflag] = nf;
4719 nf = NULL;
4720 }
4721 }
4722 spin_unlock(&fp->fi_lock);
4723 if (nf)
4724 nfsd_file_put(nf);
4725
4726 status = nfsd4_truncate(rqstp, cur_fh, open);
4727 if (status)
4728 goto out_put_access;
4729out:
4730 return status;
4731out_put_access:
4732 stp->st_access_bmap = old_access_bmap;
4733 nfs4_file_put_access(fp, open->op_share_access);
4734 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4735 goto out;
4736}
4737
4738static __be32
4739nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4740{
4741 __be32 status;
4742 unsigned char old_deny_bmap = stp->st_deny_bmap;
4743
4744 if (!test_access(open->op_share_access, stp))
4745 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4746
4747 /* test and set deny mode */
4748 spin_lock(&fp->fi_lock);
4749 status = nfs4_file_check_deny(fp, open->op_share_deny);
4750 if (status == nfs_ok) {
4751 set_deny(open->op_share_deny, stp);
4752 fp->fi_share_deny |=
4753 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4754 }
4755 spin_unlock(&fp->fi_lock);
4756
4757 if (status != nfs_ok)
4758 return status;
4759
4760 status = nfsd4_truncate(rqstp, cur_fh, open);
4761 if (status != nfs_ok)
4762 reset_union_bmap_deny(old_deny_bmap, stp);
4763 return status;
4764}
4765
4766/* Should we give out recallable state?: */
4767static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4768{
4769 if (clp->cl_cb_state == NFSD4_CB_UP)
4770 return true;
4771 /*
4772 * In the sessions case, since we don't have to establish a
4773 * separate connection for callbacks, we assume it's OK
4774 * until we hear otherwise:
4775 */
4776 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4777}
4778
4779static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4780 int flag)
4781{
4782 struct file_lock *fl;
4783
4784 fl = locks_alloc_lock();
4785 if (!fl)
4786 return NULL;
4787 fl->fl_lmops = &nfsd_lease_mng_ops;
4788 fl->fl_flags = FL_DELEG;
4789 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4790 fl->fl_end = OFFSET_MAX;
4791 fl->fl_owner = (fl_owner_t)dp;
4792 fl->fl_pid = current->tgid;
4793 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4794 return fl;
4795}
4796
4797static struct nfs4_delegation *
4798nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4799 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4800{
4801 int status = 0;
4802 struct nfs4_delegation *dp;
4803 struct nfsd_file *nf;
4804 struct file_lock *fl;
4805
4806 /*
4807 * The fi_had_conflict and nfs_get_existing_delegation checks
4808 * here are just optimizations; we'll need to recheck them at
4809 * the end:
4810 */
4811 if (fp->fi_had_conflict)
4812 return ERR_PTR(-EAGAIN);
4813
4814 nf = find_readable_file(fp);
4815 if (!nf) {
4816 /* We should always have a readable file here */
4817 WARN_ON_ONCE(1);
4818 return ERR_PTR(-EBADF);
4819 }
4820 spin_lock(&state_lock);
4821 spin_lock(&fp->fi_lock);
4822 if (nfs4_delegation_exists(clp, fp))
4823 status = -EAGAIN;
4824 else if (!fp->fi_deleg_file) {
4825 fp->fi_deleg_file = nf;
4826 /* increment early to prevent fi_deleg_file from being
4827 * cleared */
4828 fp->fi_delegees = 1;
4829 nf = NULL;
4830 } else
4831 fp->fi_delegees++;
4832 spin_unlock(&fp->fi_lock);
4833 spin_unlock(&state_lock);
4834 if (nf)
4835 nfsd_file_put(nf);
4836 if (status)
4837 return ERR_PTR(status);
4838
4839 status = -ENOMEM;
4840 dp = alloc_init_deleg(clp, fp, fh, odstate);
4841 if (!dp)
4842 goto out_delegees;
4843
4844 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4845 if (!fl)
4846 goto out_clnt_odstate;
4847
4848 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
4849 if (fl)
4850 locks_free_lock(fl);
4851 if (status)
4852 goto out_clnt_odstate;
4853
4854 spin_lock(&state_lock);
4855 spin_lock(&fp->fi_lock);
4856 if (fp->fi_had_conflict)
4857 status = -EAGAIN;
4858 else
4859 status = hash_delegation_locked(dp, fp);
4860 spin_unlock(&fp->fi_lock);
4861 spin_unlock(&state_lock);
4862
4863 if (status)
4864 goto out_unlock;
4865
4866 return dp;
4867out_unlock:
4868 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
4869out_clnt_odstate:
4870 put_clnt_odstate(dp->dl_clnt_odstate);
4871 nfs4_put_stid(&dp->dl_stid);
4872out_delegees:
4873 put_deleg_file(fp);
4874 return ERR_PTR(status);
4875}
4876
4877static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4878{
4879 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4880 if (status == -EAGAIN)
4881 open->op_why_no_deleg = WND4_CONTENTION;
4882 else {
4883 open->op_why_no_deleg = WND4_RESOURCE;
4884 switch (open->op_deleg_want) {
4885 case NFS4_SHARE_WANT_READ_DELEG:
4886 case NFS4_SHARE_WANT_WRITE_DELEG:
4887 case NFS4_SHARE_WANT_ANY_DELEG:
4888 break;
4889 case NFS4_SHARE_WANT_CANCEL:
4890 open->op_why_no_deleg = WND4_CANCELLED;
4891 break;
4892 case NFS4_SHARE_WANT_NO_DELEG:
4893 WARN_ON_ONCE(1);
4894 }
4895 }
4896}
4897
4898/*
4899 * Attempt to hand out a delegation.
4900 *
4901 * Note we don't support write delegations, and won't until the vfs has
4902 * proper support for them.
4903 */
4904static void
4905nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4906 struct nfs4_ol_stateid *stp)
4907{
4908 struct nfs4_delegation *dp;
4909 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4910 struct nfs4_client *clp = stp->st_stid.sc_client;
4911 int cb_up;
4912 int status = 0;
4913
4914 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4915 open->op_recall = 0;
4916 switch (open->op_claim_type) {
4917 case NFS4_OPEN_CLAIM_PREVIOUS:
4918 if (!cb_up)
4919 open->op_recall = 1;
4920 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4921 goto out_no_deleg;
4922 break;
4923 case NFS4_OPEN_CLAIM_NULL:
4924 case NFS4_OPEN_CLAIM_FH:
4925 /*
4926 * Let's not give out any delegations till everyone's
4927 * had the chance to reclaim theirs, *and* until
4928 * NLM locks have all been reclaimed:
4929 */
4930 if (locks_in_grace(clp->net))
4931 goto out_no_deleg;
4932 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4933 goto out_no_deleg;
4934 /*
4935 * Also, if the file was opened for write or
4936 * create, there's a good chance the client's
4937 * about to write to it, resulting in an
4938 * immediate recall (since we don't support
4939 * write delegations):
4940 */
4941 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4942 goto out_no_deleg;
4943 if (open->op_create == NFS4_OPEN_CREATE)
4944 goto out_no_deleg;
4945 break;
4946 default:
4947 goto out_no_deleg;
4948 }
4949 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4950 if (IS_ERR(dp))
4951 goto out_no_deleg;
4952
4953 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4954
4955 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4956 STATEID_VAL(&dp->dl_stid.sc_stateid));
4957 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4958 nfs4_put_stid(&dp->dl_stid);
4959 return;
4960out_no_deleg:
4961 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4962 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4963 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4964 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4965 open->op_recall = 1;
4966 }
4967
4968 /* 4.1 client asking for a delegation? */
4969 if (open->op_deleg_want)
4970 nfsd4_open_deleg_none_ext(open, status);
4971 return;
4972}
4973
4974static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4975 struct nfs4_delegation *dp)
4976{
4977 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4978 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4979 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4980 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4981 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4982 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4983 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4984 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4985 }
4986 /* Otherwise the client must be confused wanting a delegation
4987 * it already has, therefore we don't return
4988 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4989 */
4990}
4991
4992__be32
4993nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4994{
4995 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4996 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4997 struct nfs4_file *fp = NULL;
4998 struct nfs4_ol_stateid *stp = NULL;
4999 struct nfs4_delegation *dp = NULL;
5000 __be32 status;
5001 bool new_stp = false;
5002
5003 /*
5004 * Lookup file; if found, lookup stateid and check open request,
5005 * and check for delegations in the process of being recalled.
5006 * If not found, create the nfs4_file struct
5007 */
5008 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
5009 if (fp != open->op_file) {
5010 status = nfs4_check_deleg(cl, open, &dp);
5011 if (status)
5012 goto out;
5013 stp = nfsd4_find_and_lock_existing_open(fp, open);
5014 } else {
5015 open->op_file = NULL;
5016 status = nfserr_bad_stateid;
5017 if (nfsd4_is_deleg_cur(open))
5018 goto out;
5019 }
5020
5021 if (!stp) {
5022 stp = init_open_stateid(fp, open);
5023 if (!open->op_stp)
5024 new_stp = true;
5025 }
5026
5027 /*
5028 * OPEN the file, or upgrade an existing OPEN.
5029 * If truncate fails, the OPEN fails.
5030 *
5031 * stp is already locked.
5032 */
5033 if (!new_stp) {
5034 /* Stateid was found, this is an OPEN upgrade */
5035 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5036 if (status) {
5037 mutex_unlock(&stp->st_mutex);
5038 goto out;
5039 }
5040 } else {
5041 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5042 if (status) {
5043 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5044 release_open_stateid(stp);
5045 mutex_unlock(&stp->st_mutex);
5046 goto out;
5047 }
5048
5049 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5050 open->op_odstate);
5051 if (stp->st_clnt_odstate == open->op_odstate)
5052 open->op_odstate = NULL;
5053 }
5054
5055 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5056 mutex_unlock(&stp->st_mutex);
5057
5058 if (nfsd4_has_session(&resp->cstate)) {
5059 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5060 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5061 open->op_why_no_deleg = WND4_NOT_WANTED;
5062 goto nodeleg;
5063 }
5064 }
5065
5066 /*
5067 * Attempt to hand out a delegation. No error return, because the
5068 * OPEN succeeds even if we fail.
5069 */
5070 nfs4_open_delegation(current_fh, open, stp);
5071nodeleg:
5072 status = nfs_ok;
5073
5074 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5075 STATEID_VAL(&stp->st_stid.sc_stateid));
5076out:
5077 /* 4.1 client trying to upgrade/downgrade delegation? */
5078 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5079 open->op_deleg_want)
5080 nfsd4_deleg_xgrade_none_ext(open, dp);
5081
5082 if (fp)
5083 put_nfs4_file(fp);
5084 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5085 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5086 /*
5087 * To finish the open response, we just need to set the rflags.
5088 */
5089 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5090 if (nfsd4_has_session(&resp->cstate))
5091 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5092 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5093 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5094
5095 if (dp)
5096 nfs4_put_stid(&dp->dl_stid);
5097 if (stp)
5098 nfs4_put_stid(&stp->st_stid);
5099
5100 return status;
5101}
5102
5103void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5104 struct nfsd4_open *open)
5105{
5106 if (open->op_openowner) {
5107 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5108
5109 nfsd4_cstate_assign_replay(cstate, so);
5110 nfs4_put_stateowner(so);
5111 }
5112 if (open->op_file)
5113 kmem_cache_free(file_slab, open->op_file);
5114 if (open->op_stp)
5115 nfs4_put_stid(&open->op_stp->st_stid);
5116 if (open->op_odstate)
5117 kmem_cache_free(odstate_slab, open->op_odstate);
5118}
5119
5120__be32
5121nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5122 union nfsd4_op_u *u)
5123{
5124 clientid_t *clid = &u->renew;
5125 struct nfs4_client *clp;
5126 __be32 status;
5127 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5128
5129 dprintk("process_renew(%08x/%08x): starting\n",
5130 clid->cl_boot, clid->cl_id);
5131 status = lookup_clientid(clid, cstate, nn);
5132 if (status)
5133 goto out;
5134 clp = cstate->clp;
5135 status = nfserr_cb_path_down;
5136 if (!list_empty(&clp->cl_delegations)
5137 && clp->cl_cb_state != NFSD4_CB_UP)
5138 goto out;
5139 status = nfs_ok;
5140out:
5141 return status;
5142}
5143
5144void
5145nfsd4_end_grace(struct nfsd_net *nn)
5146{
5147 /* do nothing if grace period already ended */
5148 if (nn->grace_ended)
5149 return;
5150
5151 nn->grace_ended = true;
5152 /*
5153 * If the server goes down again right now, an NFSv4
5154 * client will still be allowed to reclaim after it comes back up,
5155 * even if it hasn't yet had a chance to reclaim state this time.
5156 *
5157 */
5158 nfsd4_record_grace_done(nn);
5159 /*
5160 * At this point, NFSv4 clients can still reclaim. But if the
5161 * server crashes, any that have not yet reclaimed will be out
5162 * of luck on the next boot.
5163 *
5164 * (NFSv4.1+ clients are considered to have reclaimed once they
5165 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5166 * have reclaimed after their first OPEN.)
5167 */
5168 locks_end_grace(&nn->nfsd4_manager);
5169 /*
5170 * At this point, and once lockd and/or any other containers
5171 * exit their grace period, further reclaims will fail and
5172 * regular locking can resume.
5173 */
5174}
5175
5176/*
5177 * If we've waited a lease period but there are still clients trying to
5178 * reclaim, wait a little longer to give them a chance to finish.
5179 */
5180static bool clients_still_reclaiming(struct nfsd_net *nn)
5181{
5182 unsigned long now = get_seconds();
5183 unsigned long double_grace_period_end = nn->boot_time +
5184 2 * nn->nfsd4_lease;
5185
5186 if (nn->track_reclaim_completes &&
5187 atomic_read(&nn->nr_reclaim_complete) ==
5188 nn->reclaim_str_hashtbl_size)
5189 return false;
5190 if (!nn->somebody_reclaimed)
5191 return false;
5192 nn->somebody_reclaimed = false;
5193 /*
5194 * If we've given them *two* lease times to reclaim, and they're
5195 * still not done, give up:
5196 */
5197 if (time_after(now, double_grace_period_end))
5198 return false;
5199 return true;
5200}
5201
5202static time_t
5203nfs4_laundromat(struct nfsd_net *nn)
5204{
5205 struct nfs4_client *clp;
5206 struct nfs4_openowner *oo;
5207 struct nfs4_delegation *dp;
5208 struct nfs4_ol_stateid *stp;
5209 struct nfsd4_blocked_lock *nbl;
5210 struct list_head *pos, *next, reaplist;
5211 time_t cutoff = get_seconds() - nn->nfsd4_lease;
5212 time_t t, new_timeo = nn->nfsd4_lease;
5213
5214 dprintk("NFSD: laundromat service - starting\n");
5215
5216 if (clients_still_reclaiming(nn)) {
5217 new_timeo = 0;
5218 goto out;
5219 }
5220 dprintk("NFSD: end of grace period\n");
5221 nfsd4_end_grace(nn);
5222 INIT_LIST_HEAD(&reaplist);
5223 spin_lock(&nn->client_lock);
5224 list_for_each_safe(pos, next, &nn->client_lru) {
5225 clp = list_entry(pos, struct nfs4_client, cl_lru);
5226 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
5227 t = clp->cl_time - cutoff;
5228 new_timeo = min(new_timeo, t);
5229 break;
5230 }
5231 if (mark_client_expired_locked(clp)) {
5232 dprintk("NFSD: client in use (clientid %08x)\n",
5233 clp->cl_clientid.cl_id);
5234 continue;
5235 }
5236 list_add(&clp->cl_lru, &reaplist);
5237 }
5238 spin_unlock(&nn->client_lock);
5239 list_for_each_safe(pos, next, &reaplist) {
5240 clp = list_entry(pos, struct nfs4_client, cl_lru);
5241 dprintk("NFSD: purging unused client (clientid %08x)\n",
5242 clp->cl_clientid.cl_id);
5243 list_del_init(&clp->cl_lru);
5244 expire_client(clp);
5245 }
5246 spin_lock(&state_lock);
5247 list_for_each_safe(pos, next, &nn->del_recall_lru) {
5248 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5249 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
5250 t = dp->dl_time - cutoff;
5251 new_timeo = min(new_timeo, t);
5252 break;
5253 }
5254 WARN_ON(!unhash_delegation_locked(dp));
5255 list_add(&dp->dl_recall_lru, &reaplist);
5256 }
5257 spin_unlock(&state_lock);
5258 while (!list_empty(&reaplist)) {
5259 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5260 dl_recall_lru);
5261 list_del_init(&dp->dl_recall_lru);
5262 revoke_delegation(dp);
5263 }
5264
5265 spin_lock(&nn->client_lock);
5266 while (!list_empty(&nn->close_lru)) {
5267 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5268 oo_close_lru);
5269 if (time_after((unsigned long)oo->oo_time,
5270 (unsigned long)cutoff)) {
5271 t = oo->oo_time - cutoff;
5272 new_timeo = min(new_timeo, t);
5273 break;
5274 }
5275 list_del_init(&oo->oo_close_lru);
5276 stp = oo->oo_last_closed_stid;
5277 oo->oo_last_closed_stid = NULL;
5278 spin_unlock(&nn->client_lock);
5279 nfs4_put_stid(&stp->st_stid);
5280 spin_lock(&nn->client_lock);
5281 }
5282 spin_unlock(&nn->client_lock);
5283
5284 /*
5285 * It's possible for a client to try and acquire an already held lock
5286 * that is being held for a long time, and then lose interest in it.
5287 * So, we clean out any un-revisited request after a lease period
5288 * under the assumption that the client is no longer interested.
5289 *
5290 * RFC5661, sec. 9.6 states that the client must not rely on getting
5291 * notifications and must continue to poll for locks, even when the
5292 * server supports them. Thus this shouldn't lead to clients blocking
5293 * indefinitely once the lock does become free.
5294 */
5295 BUG_ON(!list_empty(&reaplist));
5296 spin_lock(&nn->blocked_locks_lock);
5297 while (!list_empty(&nn->blocked_locks_lru)) {
5298 nbl = list_first_entry(&nn->blocked_locks_lru,
5299 struct nfsd4_blocked_lock, nbl_lru);
5300 if (time_after((unsigned long)nbl->nbl_time,
5301 (unsigned long)cutoff)) {
5302 t = nbl->nbl_time - cutoff;
5303 new_timeo = min(new_timeo, t);
5304 break;
5305 }
5306 list_move(&nbl->nbl_lru, &reaplist);
5307 list_del_init(&nbl->nbl_list);
5308 }
5309 spin_unlock(&nn->blocked_locks_lock);
5310
5311 while (!list_empty(&reaplist)) {
5312 nbl = list_first_entry(&reaplist,
5313 struct nfsd4_blocked_lock, nbl_lru);
5314 list_del_init(&nbl->nbl_lru);
5315 free_blocked_lock(nbl);
5316 }
5317out:
5318 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5319 return new_timeo;
5320}
5321
5322static struct workqueue_struct *laundry_wq;
5323static void laundromat_main(struct work_struct *);
5324
5325static void
5326laundromat_main(struct work_struct *laundry)
5327{
5328 time_t t;
5329 struct delayed_work *dwork = to_delayed_work(laundry);
5330 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5331 laundromat_work);
5332
5333 t = nfs4_laundromat(nn);
5334 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
5335 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5336}
5337
5338static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5339{
5340 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5341 return nfserr_bad_stateid;
5342 return nfs_ok;
5343}
5344
5345static inline int
5346access_permit_read(struct nfs4_ol_stateid *stp)
5347{
5348 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5349 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5350 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5351}
5352
5353static inline int
5354access_permit_write(struct nfs4_ol_stateid *stp)
5355{
5356 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5357 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5358}
5359
5360static
5361__be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5362{
5363 __be32 status = nfserr_openmode;
5364
5365 /* For lock stateid's, we test the parent open, not the lock: */
5366 if (stp->st_openstp)
5367 stp = stp->st_openstp;
5368 if ((flags & WR_STATE) && !access_permit_write(stp))
5369 goto out;
5370 if ((flags & RD_STATE) && !access_permit_read(stp))
5371 goto out;
5372 status = nfs_ok;
5373out:
5374 return status;
5375}
5376
5377static inline __be32
5378check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5379{
5380 if (ONE_STATEID(stateid) && (flags & RD_STATE))
5381 return nfs_ok;
5382 else if (opens_in_grace(net)) {
5383 /* Answer in remaining cases depends on existence of
5384 * conflicting state; so we must wait out the grace period. */
5385 return nfserr_grace;
5386 } else if (flags & WR_STATE)
5387 return nfs4_share_conflict(current_fh,
5388 NFS4_SHARE_DENY_WRITE);
5389 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5390 return nfs4_share_conflict(current_fh,
5391 NFS4_SHARE_DENY_READ);
5392}
5393
5394/*
5395 * Allow READ/WRITE during grace period on recovered state only for files
5396 * that are not able to provide mandatory locking.
5397 */
5398static inline int
5399grace_disallows_io(struct net *net, struct inode *inode)
5400{
5401 return opens_in_grace(net) && mandatory_lock(inode);
5402}
5403
5404static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5405{
5406 /*
5407 * When sessions are used the stateid generation number is ignored
5408 * when it is zero.
5409 */
5410 if (has_session && in->si_generation == 0)
5411 return nfs_ok;
5412
5413 if (in->si_generation == ref->si_generation)
5414 return nfs_ok;
5415
5416 /* If the client sends us a stateid from the future, it's buggy: */
5417 if (nfsd4_stateid_generation_after(in, ref))
5418 return nfserr_bad_stateid;
5419 /*
5420 * However, we could see a stateid from the past, even from a
5421 * non-buggy client. For example, if the client sends a lock
5422 * while some IO is outstanding, the lock may bump si_generation
5423 * while the IO is still in flight. The client could avoid that
5424 * situation by waiting for responses on all the IO requests,
5425 * but better performance may result in retrying IO that
5426 * receives an old_stateid error if requests are rarely
5427 * reordered in flight:
5428 */
5429 return nfserr_old_stateid;
5430}
5431
5432static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5433{
5434 __be32 ret;
5435
5436 spin_lock(&s->sc_lock);
5437 ret = nfsd4_verify_open_stid(s);
5438 if (ret == nfs_ok)
5439 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5440 spin_unlock(&s->sc_lock);
5441 return ret;
5442}
5443
5444static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5445{
5446 if (ols->st_stateowner->so_is_open_owner &&
5447 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5448 return nfserr_bad_stateid;
5449 return nfs_ok;
5450}
5451
5452static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5453{
5454 struct nfs4_stid *s;
5455 __be32 status = nfserr_bad_stateid;
5456
5457 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5458 CLOSE_STATEID(stateid))
5459 return status;
5460 /* Client debugging aid. */
5461 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5462 char addr_str[INET6_ADDRSTRLEN];
5463 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5464 sizeof(addr_str));
5465 pr_warn_ratelimited("NFSD: client %s testing state ID "
5466 "with incorrect client ID\n", addr_str);
5467 return status;
5468 }
5469 spin_lock(&cl->cl_lock);
5470 s = find_stateid_locked(cl, stateid);
5471 if (!s)
5472 goto out_unlock;
5473 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5474 if (status)
5475 goto out_unlock;
5476 switch (s->sc_type) {
5477 case NFS4_DELEG_STID:
5478 status = nfs_ok;
5479 break;
5480 case NFS4_REVOKED_DELEG_STID:
5481 status = nfserr_deleg_revoked;
5482 break;
5483 case NFS4_OPEN_STID:
5484 case NFS4_LOCK_STID:
5485 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5486 break;
5487 default:
5488 printk("unknown stateid type %x\n", s->sc_type);
5489 /* Fallthrough */
5490 case NFS4_CLOSED_STID:
5491 case NFS4_CLOSED_DELEG_STID:
5492 status = nfserr_bad_stateid;
5493 }
5494out_unlock:
5495 spin_unlock(&cl->cl_lock);
5496 return status;
5497}
5498
5499__be32
5500nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5501 stateid_t *stateid, unsigned char typemask,
5502 struct nfs4_stid **s, struct nfsd_net *nn)
5503{
5504 __be32 status;
5505 bool return_revoked = false;
5506
5507 /*
5508 * only return revoked delegations if explicitly asked.
5509 * otherwise we report revoked or bad_stateid status.
5510 */
5511 if (typemask & NFS4_REVOKED_DELEG_STID)
5512 return_revoked = true;
5513 else if (typemask & NFS4_DELEG_STID)
5514 typemask |= NFS4_REVOKED_DELEG_STID;
5515
5516 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5517 CLOSE_STATEID(stateid))
5518 return nfserr_bad_stateid;
5519 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5520 if (status == nfserr_stale_clientid) {
5521 if (cstate->session)
5522 return nfserr_bad_stateid;
5523 return nfserr_stale_stateid;
5524 }
5525 if (status)
5526 return status;
5527 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5528 if (!*s)
5529 return nfserr_bad_stateid;
5530 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5531 nfs4_put_stid(*s);
5532 if (cstate->minorversion)
5533 return nfserr_deleg_revoked;
5534 return nfserr_bad_stateid;
5535 }
5536 return nfs_ok;
5537}
5538
5539static struct nfsd_file *
5540nfs4_find_file(struct nfs4_stid *s, int flags)
5541{
5542 if (!s)
5543 return NULL;
5544
5545 switch (s->sc_type) {
5546 case NFS4_DELEG_STID:
5547 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5548 return NULL;
5549 return nfsd_file_get(s->sc_file->fi_deleg_file);
5550 case NFS4_OPEN_STID:
5551 case NFS4_LOCK_STID:
5552 if (flags & RD_STATE)
5553 return find_readable_file(s->sc_file);
5554 else
5555 return find_writeable_file(s->sc_file);
5556 break;
5557 }
5558
5559 return NULL;
5560}
5561
5562static __be32
5563nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5564{
5565 __be32 status;
5566
5567 status = nfsd4_check_openowner_confirmed(ols);
5568 if (status)
5569 return status;
5570 return nfs4_check_openmode(ols, flags);
5571}
5572
5573static __be32
5574nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5575 struct nfsd_file **nfp, int flags)
5576{
5577 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5578 struct nfsd_file *nf;
5579 __be32 status;
5580
5581 nf = nfs4_find_file(s, flags);
5582 if (nf) {
5583 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5584 acc | NFSD_MAY_OWNER_OVERRIDE);
5585 if (status) {
5586 nfsd_file_put(nf);
5587 goto out;
5588 }
5589 } else {
5590 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5591 if (status)
5592 return status;
5593 }
5594 *nfp = nf;
5595out:
5596 return status;
5597}
5598
5599/*
5600 * Checks for stateid operations
5601 */
5602__be32
5603nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5604 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5605 stateid_t *stateid, int flags, struct nfsd_file **nfp)
5606{
5607 struct inode *ino = d_inode(fhp->fh_dentry);
5608 struct net *net = SVC_NET(rqstp);
5609 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5610 struct nfs4_stid *s = NULL;
5611 __be32 status;
5612
5613 if (nfp)
5614 *nfp = NULL;
5615
5616 if (grace_disallows_io(net, ino))
5617 return nfserr_grace;
5618
5619 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5620 status = check_special_stateids(net, fhp, stateid, flags);
5621 goto done;
5622 }
5623
5624 status = nfsd4_lookup_stateid(cstate, stateid,
5625 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5626 &s, nn);
5627 if (status)
5628 return status;
5629 status = nfsd4_stid_check_stateid_generation(stateid, s,
5630 nfsd4_has_session(cstate));
5631 if (status)
5632 goto out;
5633
5634 switch (s->sc_type) {
5635 case NFS4_DELEG_STID:
5636 status = nfs4_check_delegmode(delegstateid(s), flags);
5637 break;
5638 case NFS4_OPEN_STID:
5639 case NFS4_LOCK_STID:
5640 status = nfs4_check_olstateid(openlockstateid(s), flags);
5641 break;
5642 default:
5643 status = nfserr_bad_stateid;
5644 break;
5645 }
5646 if (status)
5647 goto out;
5648 status = nfs4_check_fh(fhp, s);
5649
5650done:
5651 if (status == nfs_ok && nfp)
5652 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5653out:
5654 if (s)
5655 nfs4_put_stid(s);
5656 return status;
5657}
5658
5659/*
5660 * Test if the stateid is valid
5661 */
5662__be32
5663nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5664 union nfsd4_op_u *u)
5665{
5666 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5667 struct nfsd4_test_stateid_id *stateid;
5668 struct nfs4_client *cl = cstate->session->se_client;
5669
5670 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5671 stateid->ts_id_status =
5672 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5673
5674 return nfs_ok;
5675}
5676
5677static __be32
5678nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5679{
5680 struct nfs4_ol_stateid *stp = openlockstateid(s);
5681 __be32 ret;
5682
5683 ret = nfsd4_lock_ol_stateid(stp);
5684 if (ret)
5685 goto out_put_stid;
5686
5687 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5688 if (ret)
5689 goto out;
5690
5691 ret = nfserr_locks_held;
5692 if (check_for_locks(stp->st_stid.sc_file,
5693 lockowner(stp->st_stateowner)))
5694 goto out;
5695
5696 release_lock_stateid(stp);
5697 ret = nfs_ok;
5698
5699out:
5700 mutex_unlock(&stp->st_mutex);
5701out_put_stid:
5702 nfs4_put_stid(s);
5703 return ret;
5704}
5705
5706__be32
5707nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5708 union nfsd4_op_u *u)
5709{
5710 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5711 stateid_t *stateid = &free_stateid->fr_stateid;
5712 struct nfs4_stid *s;
5713 struct nfs4_delegation *dp;
5714 struct nfs4_client *cl = cstate->session->se_client;
5715 __be32 ret = nfserr_bad_stateid;
5716
5717 spin_lock(&cl->cl_lock);
5718 s = find_stateid_locked(cl, stateid);
5719 if (!s)
5720 goto out_unlock;
5721 spin_lock(&s->sc_lock);
5722 switch (s->sc_type) {
5723 case NFS4_DELEG_STID:
5724 ret = nfserr_locks_held;
5725 break;
5726 case NFS4_OPEN_STID:
5727 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5728 if (ret)
5729 break;
5730 ret = nfserr_locks_held;
5731 break;
5732 case NFS4_LOCK_STID:
5733 spin_unlock(&s->sc_lock);
5734 refcount_inc(&s->sc_count);
5735 spin_unlock(&cl->cl_lock);
5736 ret = nfsd4_free_lock_stateid(stateid, s);
5737 goto out;
5738 case NFS4_REVOKED_DELEG_STID:
5739 spin_unlock(&s->sc_lock);
5740 dp = delegstateid(s);
5741 list_del_init(&dp->dl_recall_lru);
5742 spin_unlock(&cl->cl_lock);
5743 nfs4_put_stid(s);
5744 ret = nfs_ok;
5745 goto out;
5746 /* Default falls through and returns nfserr_bad_stateid */
5747 }
5748 spin_unlock(&s->sc_lock);
5749out_unlock:
5750 spin_unlock(&cl->cl_lock);
5751out:
5752 return ret;
5753}
5754
5755static inline int
5756setlkflg (int type)
5757{
5758 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5759 RD_STATE : WR_STATE;
5760}
5761
5762static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5763{
5764 struct svc_fh *current_fh = &cstate->current_fh;
5765 struct nfs4_stateowner *sop = stp->st_stateowner;
5766 __be32 status;
5767
5768 status = nfsd4_check_seqid(cstate, sop, seqid);
5769 if (status)
5770 return status;
5771 status = nfsd4_lock_ol_stateid(stp);
5772 if (status != nfs_ok)
5773 return status;
5774 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5775 if (status == nfs_ok)
5776 status = nfs4_check_fh(current_fh, &stp->st_stid);
5777 if (status != nfs_ok)
5778 mutex_unlock(&stp->st_mutex);
5779 return status;
5780}
5781
5782/*
5783 * Checks for sequence id mutating operations.
5784 */
5785static __be32
5786nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5787 stateid_t *stateid, char typemask,
5788 struct nfs4_ol_stateid **stpp,
5789 struct nfsd_net *nn)
5790{
5791 __be32 status;
5792 struct nfs4_stid *s;
5793 struct nfs4_ol_stateid *stp = NULL;
5794
5795 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5796 seqid, STATEID_VAL(stateid));
5797
5798 *stpp = NULL;
5799 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5800 if (status)
5801 return status;
5802 stp = openlockstateid(s);
5803 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5804
5805 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5806 if (!status)
5807 *stpp = stp;
5808 else
5809 nfs4_put_stid(&stp->st_stid);
5810 return status;
5811}
5812
5813static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5814 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5815{
5816 __be32 status;
5817 struct nfs4_openowner *oo;
5818 struct nfs4_ol_stateid *stp;
5819
5820 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5821 NFS4_OPEN_STID, &stp, nn);
5822 if (status)
5823 return status;
5824 oo = openowner(stp->st_stateowner);
5825 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5826 mutex_unlock(&stp->st_mutex);
5827 nfs4_put_stid(&stp->st_stid);
5828 return nfserr_bad_stateid;
5829 }
5830 *stpp = stp;
5831 return nfs_ok;
5832}
5833
5834__be32
5835nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5836 union nfsd4_op_u *u)
5837{
5838 struct nfsd4_open_confirm *oc = &u->open_confirm;
5839 __be32 status;
5840 struct nfs4_openowner *oo;
5841 struct nfs4_ol_stateid *stp;
5842 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5843
5844 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5845 cstate->current_fh.fh_dentry);
5846
5847 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5848 if (status)
5849 return status;
5850
5851 status = nfs4_preprocess_seqid_op(cstate,
5852 oc->oc_seqid, &oc->oc_req_stateid,
5853 NFS4_OPEN_STID, &stp, nn);
5854 if (status)
5855 goto out;
5856 oo = openowner(stp->st_stateowner);
5857 status = nfserr_bad_stateid;
5858 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5859 mutex_unlock(&stp->st_mutex);
5860 goto put_stateid;
5861 }
5862 oo->oo_flags |= NFS4_OO_CONFIRMED;
5863 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5864 mutex_unlock(&stp->st_mutex);
5865 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5866 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5867
5868 nfsd4_client_record_create(oo->oo_owner.so_client);
5869 status = nfs_ok;
5870put_stateid:
5871 nfs4_put_stid(&stp->st_stid);
5872out:
5873 nfsd4_bump_seqid(cstate, status);
5874 return status;
5875}
5876
5877static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5878{
5879 if (!test_access(access, stp))
5880 return;
5881 nfs4_file_put_access(stp->st_stid.sc_file, access);
5882 clear_access(access, stp);
5883}
5884
5885static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5886{
5887 switch (to_access) {
5888 case NFS4_SHARE_ACCESS_READ:
5889 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5890 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5891 break;
5892 case NFS4_SHARE_ACCESS_WRITE:
5893 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5894 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5895 break;
5896 case NFS4_SHARE_ACCESS_BOTH:
5897 break;
5898 default:
5899 WARN_ON_ONCE(1);
5900 }
5901}
5902
5903__be32
5904nfsd4_open_downgrade(struct svc_rqst *rqstp,
5905 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5906{
5907 struct nfsd4_open_downgrade *od = &u->open_downgrade;
5908 __be32 status;
5909 struct nfs4_ol_stateid *stp;
5910 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5911
5912 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5913 cstate->current_fh.fh_dentry);
5914
5915 /* We don't yet support WANT bits: */
5916 if (od->od_deleg_want)
5917 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5918 od->od_deleg_want);
5919
5920 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5921 &od->od_stateid, &stp, nn);
5922 if (status)
5923 goto out;
5924 status = nfserr_inval;
5925 if (!test_access(od->od_share_access, stp)) {
5926 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5927 stp->st_access_bmap, od->od_share_access);
5928 goto put_stateid;
5929 }
5930 if (!test_deny(od->od_share_deny, stp)) {
5931 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5932 stp->st_deny_bmap, od->od_share_deny);
5933 goto put_stateid;
5934 }
5935 nfs4_stateid_downgrade(stp, od->od_share_access);
5936 reset_union_bmap_deny(od->od_share_deny, stp);
5937 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5938 status = nfs_ok;
5939put_stateid:
5940 mutex_unlock(&stp->st_mutex);
5941 nfs4_put_stid(&stp->st_stid);
5942out:
5943 nfsd4_bump_seqid(cstate, status);
5944 return status;
5945}
5946
5947static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5948{
5949 struct nfs4_client *clp = s->st_stid.sc_client;
5950 bool unhashed;
5951 LIST_HEAD(reaplist);
5952
5953 spin_lock(&clp->cl_lock);
5954 unhashed = unhash_open_stateid(s, &reaplist);
5955
5956 if (clp->cl_minorversion) {
5957 if (unhashed)
5958 put_ol_stateid_locked(s, &reaplist);
5959 spin_unlock(&clp->cl_lock);
5960 free_ol_stateid_reaplist(&reaplist);
5961 } else {
5962 spin_unlock(&clp->cl_lock);
5963 free_ol_stateid_reaplist(&reaplist);
5964 if (unhashed)
5965 move_to_close_lru(s, clp->net);
5966 }
5967}
5968
5969/*
5970 * nfs4_unlock_state() called after encode
5971 */
5972__be32
5973nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5974 union nfsd4_op_u *u)
5975{
5976 struct nfsd4_close *close = &u->close;
5977 __be32 status;
5978 struct nfs4_ol_stateid *stp;
5979 struct net *net = SVC_NET(rqstp);
5980 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5981
5982 dprintk("NFSD: nfsd4_close on file %pd\n",
5983 cstate->current_fh.fh_dentry);
5984
5985 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5986 &close->cl_stateid,
5987 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5988 &stp, nn);
5989 nfsd4_bump_seqid(cstate, status);
5990 if (status)
5991 goto out;
5992
5993 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5994
5995 /*
5996 * Technically we don't _really_ have to increment or copy it, since
5997 * it should just be gone after this operation and we clobber the
5998 * copied value below, but we continue to do so here just to ensure
5999 * that racing ops see that there was a state change.
6000 */
6001 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6002
6003 nfsd4_close_open_stateid(stp);
6004 mutex_unlock(&stp->st_mutex);
6005
6006 /* v4.1+ suggests that we send a special stateid in here, since the
6007 * clients should just ignore this anyway. Since this is not useful
6008 * for v4.0 clients either, we set it to the special close_stateid
6009 * universally.
6010 *
6011 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6012 */
6013 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6014
6015 /* put reference from nfs4_preprocess_seqid_op */
6016 nfs4_put_stid(&stp->st_stid);
6017out:
6018 return status;
6019}
6020
6021__be32
6022nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6023 union nfsd4_op_u *u)
6024{
6025 struct nfsd4_delegreturn *dr = &u->delegreturn;
6026 struct nfs4_delegation *dp;
6027 stateid_t *stateid = &dr->dr_stateid;
6028 struct nfs4_stid *s;
6029 __be32 status;
6030 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6031
6032 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6033 return status;
6034
6035 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6036 if (status)
6037 goto out;
6038 dp = delegstateid(s);
6039 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6040 if (status)
6041 goto put_stateid;
6042
6043 destroy_delegation(dp);
6044put_stateid:
6045 nfs4_put_stid(&dp->dl_stid);
6046out:
6047 return status;
6048}
6049
6050static inline u64
6051end_offset(u64 start, u64 len)
6052{
6053 u64 end;
6054
6055 end = start + len;
6056 return end >= start ? end: NFS4_MAX_UINT64;
6057}
6058
6059/* last octet in a range */
6060static inline u64
6061last_byte_offset(u64 start, u64 len)
6062{
6063 u64 end;
6064
6065 WARN_ON_ONCE(!len);
6066 end = start + len;
6067 return end > start ? end - 1: NFS4_MAX_UINT64;
6068}
6069
6070/*
6071 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6072 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6073 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
6074 * locking, this prevents us from being completely protocol-compliant. The
6075 * real solution to this problem is to start using unsigned file offsets in
6076 * the VFS, but this is a very deep change!
6077 */
6078static inline void
6079nfs4_transform_lock_offset(struct file_lock *lock)
6080{
6081 if (lock->fl_start < 0)
6082 lock->fl_start = OFFSET_MAX;
6083 if (lock->fl_end < 0)
6084 lock->fl_end = OFFSET_MAX;
6085}
6086
6087static fl_owner_t
6088nfsd4_fl_get_owner(fl_owner_t owner)
6089{
6090 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6091
6092 nfs4_get_stateowner(&lo->lo_owner);
6093 return owner;
6094}
6095
6096static void
6097nfsd4_fl_put_owner(fl_owner_t owner)
6098{
6099 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6100
6101 if (lo)
6102 nfs4_put_stateowner(&lo->lo_owner);
6103}
6104
6105static void
6106nfsd4_lm_notify(struct file_lock *fl)
6107{
6108 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
6109 struct net *net = lo->lo_owner.so_client->net;
6110 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6111 struct nfsd4_blocked_lock *nbl = container_of(fl,
6112 struct nfsd4_blocked_lock, nbl_lock);
6113 bool queue = false;
6114
6115 /* An empty list means that something else is going to be using it */
6116 spin_lock(&nn->blocked_locks_lock);
6117 if (!list_empty(&nbl->nbl_list)) {
6118 list_del_init(&nbl->nbl_list);
6119 list_del_init(&nbl->nbl_lru);
6120 queue = true;
6121 }
6122 spin_unlock(&nn->blocked_locks_lock);
6123
6124 if (queue)
6125 nfsd4_run_cb(&nbl->nbl_cb);
6126}
6127
6128static const struct lock_manager_operations nfsd_posix_mng_ops = {
6129 .lm_notify = nfsd4_lm_notify,
6130 .lm_get_owner = nfsd4_fl_get_owner,
6131 .lm_put_owner = nfsd4_fl_put_owner,
6132};
6133
6134static inline void
6135nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6136{
6137 struct nfs4_lockowner *lo;
6138
6139 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6140 lo = (struct nfs4_lockowner *) fl->fl_owner;
6141 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6142 GFP_KERNEL);
6143 if (!deny->ld_owner.data)
6144 /* We just don't care that much */
6145 goto nevermind;
6146 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6147 } else {
6148nevermind:
6149 deny->ld_owner.len = 0;
6150 deny->ld_owner.data = NULL;
6151 deny->ld_clientid.cl_boot = 0;
6152 deny->ld_clientid.cl_id = 0;
6153 }
6154 deny->ld_start = fl->fl_start;
6155 deny->ld_length = NFS4_MAX_UINT64;
6156 if (fl->fl_end != NFS4_MAX_UINT64)
6157 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6158 deny->ld_type = NFS4_READ_LT;
6159 if (fl->fl_type != F_RDLCK)
6160 deny->ld_type = NFS4_WRITE_LT;
6161}
6162
6163static struct nfs4_lockowner *
6164find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6165{
6166 unsigned int strhashval = ownerstr_hashval(owner);
6167 struct nfs4_stateowner *so;
6168
6169 lockdep_assert_held(&clp->cl_lock);
6170
6171 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6172 so_strhash) {
6173 if (so->so_is_open_owner)
6174 continue;
6175 if (same_owner_str(so, owner))
6176 return lockowner(nfs4_get_stateowner(so));
6177 }
6178 return NULL;
6179}
6180
6181static struct nfs4_lockowner *
6182find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6183{
6184 struct nfs4_lockowner *lo;
6185
6186 spin_lock(&clp->cl_lock);
6187 lo = find_lockowner_str_locked(clp, owner);
6188 spin_unlock(&clp->cl_lock);
6189 return lo;
6190}
6191
6192static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6193{
6194 unhash_lockowner_locked(lockowner(sop));
6195}
6196
6197static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6198{
6199 struct nfs4_lockowner *lo = lockowner(sop);
6200
6201 kmem_cache_free(lockowner_slab, lo);
6202}
6203
6204static const struct nfs4_stateowner_operations lockowner_ops = {
6205 .so_unhash = nfs4_unhash_lockowner,
6206 .so_free = nfs4_free_lockowner,
6207};
6208
6209/*
6210 * Alloc a lock owner structure.
6211 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6212 * occurred.
6213 *
6214 * strhashval = ownerstr_hashval
6215 */
6216static struct nfs4_lockowner *
6217alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6218 struct nfs4_ol_stateid *open_stp,
6219 struct nfsd4_lock *lock)
6220{
6221 struct nfs4_lockowner *lo, *ret;
6222
6223 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6224 if (!lo)
6225 return NULL;
6226 INIT_LIST_HEAD(&lo->lo_blocked);
6227 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6228 lo->lo_owner.so_is_open_owner = 0;
6229 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6230 lo->lo_owner.so_ops = &lockowner_ops;
6231 spin_lock(&clp->cl_lock);
6232 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6233 if (ret == NULL) {
6234 list_add(&lo->lo_owner.so_strhash,
6235 &clp->cl_ownerstr_hashtbl[strhashval]);
6236 ret = lo;
6237 } else
6238 nfs4_free_stateowner(&lo->lo_owner);
6239
6240 spin_unlock(&clp->cl_lock);
6241 return ret;
6242}
6243
6244static struct nfs4_ol_stateid *
6245find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
6246{
6247 struct nfs4_ol_stateid *lst;
6248 struct nfs4_client *clp = lo->lo_owner.so_client;
6249
6250 lockdep_assert_held(&clp->cl_lock);
6251
6252 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
6253 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
6254 continue;
6255 if (lst->st_stid.sc_file == fp) {
6256 refcount_inc(&lst->st_stid.sc_count);
6257 return lst;
6258 }
6259 }
6260 return NULL;
6261}
6262
6263static struct nfs4_ol_stateid *
6264init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6265 struct nfs4_file *fp, struct inode *inode,
6266 struct nfs4_ol_stateid *open_stp)
6267{
6268 struct nfs4_client *clp = lo->lo_owner.so_client;
6269 struct nfs4_ol_stateid *retstp;
6270
6271 mutex_init(&stp->st_mutex);
6272 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6273retry:
6274 spin_lock(&clp->cl_lock);
6275 spin_lock(&fp->fi_lock);
6276 retstp = find_lock_stateid(lo, fp);
6277 if (retstp)
6278 goto out_unlock;
6279
6280 refcount_inc(&stp->st_stid.sc_count);
6281 stp->st_stid.sc_type = NFS4_LOCK_STID;
6282 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6283 get_nfs4_file(fp);
6284 stp->st_stid.sc_file = fp;
6285 stp->st_access_bmap = 0;
6286 stp->st_deny_bmap = open_stp->st_deny_bmap;
6287 stp->st_openstp = open_stp;
6288 list_add(&stp->st_locks, &open_stp->st_locks);
6289 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6290 list_add(&stp->st_perfile, &fp->fi_stateids);
6291out_unlock:
6292 spin_unlock(&fp->fi_lock);
6293 spin_unlock(&clp->cl_lock);
6294 if (retstp) {
6295 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6296 nfs4_put_stid(&retstp->st_stid);
6297 goto retry;
6298 }
6299 /* To keep mutex tracking happy */
6300 mutex_unlock(&stp->st_mutex);
6301 stp = retstp;
6302 }
6303 return stp;
6304}
6305
6306static struct nfs4_ol_stateid *
6307find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6308 struct inode *inode, struct nfs4_ol_stateid *ost,
6309 bool *new)
6310{
6311 struct nfs4_stid *ns = NULL;
6312 struct nfs4_ol_stateid *lst;
6313 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6314 struct nfs4_client *clp = oo->oo_owner.so_client;
6315
6316 *new = false;
6317 spin_lock(&clp->cl_lock);
6318 lst = find_lock_stateid(lo, fi);
6319 spin_unlock(&clp->cl_lock);
6320 if (lst != NULL) {
6321 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6322 goto out;
6323 nfs4_put_stid(&lst->st_stid);
6324 }
6325 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6326 if (ns == NULL)
6327 return NULL;
6328
6329 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6330 if (lst == openlockstateid(ns))
6331 *new = true;
6332 else
6333 nfs4_put_stid(ns);
6334out:
6335 return lst;
6336}
6337
6338static int
6339check_lock_length(u64 offset, u64 length)
6340{
6341 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6342 (length > ~offset)));
6343}
6344
6345static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6346{
6347 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6348
6349 lockdep_assert_held(&fp->fi_lock);
6350
6351 if (test_access(access, lock_stp))
6352 return;
6353 __nfs4_file_get_access(fp, access);
6354 set_access(access, lock_stp);
6355}
6356
6357static __be32
6358lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6359 struct nfs4_ol_stateid *ost,
6360 struct nfsd4_lock *lock,
6361 struct nfs4_ol_stateid **plst, bool *new)
6362{
6363 __be32 status;
6364 struct nfs4_file *fi = ost->st_stid.sc_file;
6365 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6366 struct nfs4_client *cl = oo->oo_owner.so_client;
6367 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6368 struct nfs4_lockowner *lo;
6369 struct nfs4_ol_stateid *lst;
6370 unsigned int strhashval;
6371
6372 lo = find_lockowner_str(cl, &lock->lk_new_owner);
6373 if (!lo) {
6374 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6375 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6376 if (lo == NULL)
6377 return nfserr_jukebox;
6378 } else {
6379 /* with an existing lockowner, seqids must be the same */
6380 status = nfserr_bad_seqid;
6381 if (!cstate->minorversion &&
6382 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6383 goto out;
6384 }
6385
6386 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6387 if (lst == NULL) {
6388 status = nfserr_jukebox;
6389 goto out;
6390 }
6391
6392 status = nfs_ok;
6393 *plst = lst;
6394out:
6395 nfs4_put_stateowner(&lo->lo_owner);
6396 return status;
6397}
6398
6399/*
6400 * LOCK operation
6401 */
6402__be32
6403nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6404 union nfsd4_op_u *u)
6405{
6406 struct nfsd4_lock *lock = &u->lock;
6407 struct nfs4_openowner *open_sop = NULL;
6408 struct nfs4_lockowner *lock_sop = NULL;
6409 struct nfs4_ol_stateid *lock_stp = NULL;
6410 struct nfs4_ol_stateid *open_stp = NULL;
6411 struct nfs4_file *fp;
6412 struct nfsd_file *nf = NULL;
6413 struct nfsd4_blocked_lock *nbl = NULL;
6414 struct file_lock *file_lock = NULL;
6415 struct file_lock *conflock = NULL;
6416 __be32 status = 0;
6417 int lkflg;
6418 int err;
6419 bool new = false;
6420 unsigned char fl_type;
6421 unsigned int fl_flags = FL_POSIX;
6422 struct net *net = SVC_NET(rqstp);
6423 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6424
6425 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6426 (long long) lock->lk_offset,
6427 (long long) lock->lk_length);
6428
6429 if (check_lock_length(lock->lk_offset, lock->lk_length))
6430 return nfserr_inval;
6431
6432 if ((status = fh_verify(rqstp, &cstate->current_fh,
6433 S_IFREG, NFSD_MAY_LOCK))) {
6434 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6435 return status;
6436 }
6437
6438 if (lock->lk_is_new) {
6439 if (nfsd4_has_session(cstate))
6440 /* See rfc 5661 18.10.3: given clientid is ignored: */
6441 memcpy(&lock->lk_new_clientid,
6442 &cstate->session->se_client->cl_clientid,
6443 sizeof(clientid_t));
6444
6445 status = nfserr_stale_clientid;
6446 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6447 goto out;
6448
6449 /* validate and update open stateid and open seqid */
6450 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6451 lock->lk_new_open_seqid,
6452 &lock->lk_new_open_stateid,
6453 &open_stp, nn);
6454 if (status)
6455 goto out;
6456 mutex_unlock(&open_stp->st_mutex);
6457 open_sop = openowner(open_stp->st_stateowner);
6458 status = nfserr_bad_stateid;
6459 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6460 &lock->lk_new_clientid))
6461 goto out;
6462 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6463 &lock_stp, &new);
6464 } else {
6465 status = nfs4_preprocess_seqid_op(cstate,
6466 lock->lk_old_lock_seqid,
6467 &lock->lk_old_lock_stateid,
6468 NFS4_LOCK_STID, &lock_stp, nn);
6469 }
6470 if (status)
6471 goto out;
6472 lock_sop = lockowner(lock_stp->st_stateowner);
6473
6474 lkflg = setlkflg(lock->lk_type);
6475 status = nfs4_check_openmode(lock_stp, lkflg);
6476 if (status)
6477 goto out;
6478
6479 status = nfserr_grace;
6480 if (locks_in_grace(net) && !lock->lk_reclaim)
6481 goto out;
6482 status = nfserr_no_grace;
6483 if (!locks_in_grace(net) && lock->lk_reclaim)
6484 goto out;
6485
6486 fp = lock_stp->st_stid.sc_file;
6487 switch (lock->lk_type) {
6488 case NFS4_READW_LT:
6489 if (nfsd4_has_session(cstate))
6490 fl_flags |= FL_SLEEP;
6491 /* Fallthrough */
6492 case NFS4_READ_LT:
6493 spin_lock(&fp->fi_lock);
6494 nf = find_readable_file_locked(fp);
6495 if (nf)
6496 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6497 spin_unlock(&fp->fi_lock);
6498 fl_type = F_RDLCK;
6499 break;
6500 case NFS4_WRITEW_LT:
6501 if (nfsd4_has_session(cstate))
6502 fl_flags |= FL_SLEEP;
6503 /* Fallthrough */
6504 case NFS4_WRITE_LT:
6505 spin_lock(&fp->fi_lock);
6506 nf = find_writeable_file_locked(fp);
6507 if (nf)
6508 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6509 spin_unlock(&fp->fi_lock);
6510 fl_type = F_WRLCK;
6511 break;
6512 default:
6513 status = nfserr_inval;
6514 goto out;
6515 }
6516
6517 if (!nf) {
6518 status = nfserr_openmode;
6519 goto out;
6520 }
6521
6522 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6523 if (!nbl) {
6524 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6525 status = nfserr_jukebox;
6526 goto out;
6527 }
6528
6529 file_lock = &nbl->nbl_lock;
6530 file_lock->fl_type = fl_type;
6531 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6532 file_lock->fl_pid = current->tgid;
6533 file_lock->fl_file = nf->nf_file;
6534 file_lock->fl_flags = fl_flags;
6535 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6536 file_lock->fl_start = lock->lk_offset;
6537 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6538 nfs4_transform_lock_offset(file_lock);
6539
6540 conflock = locks_alloc_lock();
6541 if (!conflock) {
6542 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6543 status = nfserr_jukebox;
6544 goto out;
6545 }
6546
6547 if (fl_flags & FL_SLEEP) {
6548 nbl->nbl_time = jiffies;
6549 spin_lock(&nn->blocked_locks_lock);
6550 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6551 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6552 spin_unlock(&nn->blocked_locks_lock);
6553 }
6554
6555 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6556 switch (err) {
6557 case 0: /* success! */
6558 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6559 status = 0;
6560 if (lock->lk_reclaim)
6561 nn->somebody_reclaimed = true;
6562 break;
6563 case FILE_LOCK_DEFERRED:
6564 nbl = NULL;
6565 /* Fallthrough */
6566 case -EAGAIN: /* conflock holds conflicting lock */
6567 status = nfserr_denied;
6568 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6569 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6570 break;
6571 case -EDEADLK:
6572 status = nfserr_deadlock;
6573 break;
6574 default:
6575 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6576 status = nfserrno(err);
6577 break;
6578 }
6579out:
6580 if (nbl) {
6581 /* dequeue it if we queued it before */
6582 if (fl_flags & FL_SLEEP) {
6583 spin_lock(&nn->blocked_locks_lock);
6584 list_del_init(&nbl->nbl_list);
6585 list_del_init(&nbl->nbl_lru);
6586 spin_unlock(&nn->blocked_locks_lock);
6587 }
6588 free_blocked_lock(nbl);
6589 }
6590 if (nf)
6591 nfsd_file_put(nf);
6592 if (lock_stp) {
6593 /* Bump seqid manually if the 4.0 replay owner is openowner */
6594 if (cstate->replay_owner &&
6595 cstate->replay_owner != &lock_sop->lo_owner &&
6596 seqid_mutating_err(ntohl(status)))
6597 lock_sop->lo_owner.so_seqid++;
6598
6599 /*
6600 * If this is a new, never-before-used stateid, and we are
6601 * returning an error, then just go ahead and release it.
6602 */
6603 if (status && new)
6604 release_lock_stateid(lock_stp);
6605
6606 mutex_unlock(&lock_stp->st_mutex);
6607
6608 nfs4_put_stid(&lock_stp->st_stid);
6609 }
6610 if (open_stp)
6611 nfs4_put_stid(&open_stp->st_stid);
6612 nfsd4_bump_seqid(cstate, status);
6613 if (conflock)
6614 locks_free_lock(conflock);
6615 return status;
6616}
6617
6618/*
6619 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6620 * so we do a temporary open here just to get an open file to pass to
6621 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6622 * inode operation.)
6623 */
6624static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6625{
6626 struct nfsd_file *nf;
6627 __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6628 if (!err) {
6629 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6630 nfsd_file_put(nf);
6631 }
6632 return err;
6633}
6634
6635/*
6636 * LOCKT operation
6637 */
6638__be32
6639nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6640 union nfsd4_op_u *u)
6641{
6642 struct nfsd4_lockt *lockt = &u->lockt;
6643 struct file_lock *file_lock = NULL;
6644 struct nfs4_lockowner *lo = NULL;
6645 __be32 status;
6646 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6647
6648 if (locks_in_grace(SVC_NET(rqstp)))
6649 return nfserr_grace;
6650
6651 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6652 return nfserr_inval;
6653
6654 if (!nfsd4_has_session(cstate)) {
6655 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6656 if (status)
6657 goto out;
6658 }
6659
6660 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6661 goto out;
6662
6663 file_lock = locks_alloc_lock();
6664 if (!file_lock) {
6665 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6666 status = nfserr_jukebox;
6667 goto out;
6668 }
6669
6670 switch (lockt->lt_type) {
6671 case NFS4_READ_LT:
6672 case NFS4_READW_LT:
6673 file_lock->fl_type = F_RDLCK;
6674 break;
6675 case NFS4_WRITE_LT:
6676 case NFS4_WRITEW_LT:
6677 file_lock->fl_type = F_WRLCK;
6678 break;
6679 default:
6680 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6681 status = nfserr_inval;
6682 goto out;
6683 }
6684
6685 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6686 if (lo)
6687 file_lock->fl_owner = (fl_owner_t)lo;
6688 file_lock->fl_pid = current->tgid;
6689 file_lock->fl_flags = FL_POSIX;
6690
6691 file_lock->fl_start = lockt->lt_offset;
6692 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6693
6694 nfs4_transform_lock_offset(file_lock);
6695
6696 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6697 if (status)
6698 goto out;
6699
6700 if (file_lock->fl_type != F_UNLCK) {
6701 status = nfserr_denied;
6702 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6703 }
6704out:
6705 if (lo)
6706 nfs4_put_stateowner(&lo->lo_owner);
6707 if (file_lock)
6708 locks_free_lock(file_lock);
6709 return status;
6710}
6711
6712__be32
6713nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6714 union nfsd4_op_u *u)
6715{
6716 struct nfsd4_locku *locku = &u->locku;
6717 struct nfs4_ol_stateid *stp;
6718 struct nfsd_file *nf = NULL;
6719 struct file_lock *file_lock = NULL;
6720 __be32 status;
6721 int err;
6722 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6723
6724 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6725 (long long) locku->lu_offset,
6726 (long long) locku->lu_length);
6727
6728 if (check_lock_length(locku->lu_offset, locku->lu_length))
6729 return nfserr_inval;
6730
6731 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6732 &locku->lu_stateid, NFS4_LOCK_STID,
6733 &stp, nn);
6734 if (status)
6735 goto out;
6736 nf = find_any_file(stp->st_stid.sc_file);
6737 if (!nf) {
6738 status = nfserr_lock_range;
6739 goto put_stateid;
6740 }
6741 file_lock = locks_alloc_lock();
6742 if (!file_lock) {
6743 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6744 status = nfserr_jukebox;
6745 goto put_file;
6746 }
6747
6748 file_lock->fl_type = F_UNLCK;
6749 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6750 file_lock->fl_pid = current->tgid;
6751 file_lock->fl_file = nf->nf_file;
6752 file_lock->fl_flags = FL_POSIX;
6753 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6754 file_lock->fl_start = locku->lu_offset;
6755
6756 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6757 locku->lu_length);
6758 nfs4_transform_lock_offset(file_lock);
6759
6760 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
6761 if (err) {
6762 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6763 goto out_nfserr;
6764 }
6765 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6766put_file:
6767 nfsd_file_put(nf);
6768put_stateid:
6769 mutex_unlock(&stp->st_mutex);
6770 nfs4_put_stid(&stp->st_stid);
6771out:
6772 nfsd4_bump_seqid(cstate, status);
6773 if (file_lock)
6774 locks_free_lock(file_lock);
6775 return status;
6776
6777out_nfserr:
6778 status = nfserrno(err);
6779 goto put_file;
6780}
6781
6782/*
6783 * returns
6784 * true: locks held by lockowner
6785 * false: no locks held by lockowner
6786 */
6787static bool
6788check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6789{
6790 struct file_lock *fl;
6791 int status = false;
6792 struct nfsd_file *nf = find_any_file(fp);
6793 struct inode *inode;
6794 struct file_lock_context *flctx;
6795
6796 if (!nf) {
6797 /* Any valid lock stateid should have some sort of access */
6798 WARN_ON_ONCE(1);
6799 return status;
6800 }
6801
6802 inode = locks_inode(nf->nf_file);
6803 flctx = inode->i_flctx;
6804
6805 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6806 spin_lock(&flctx->flc_lock);
6807 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6808 if (fl->fl_owner == (fl_owner_t)lowner) {
6809 status = true;
6810 break;
6811 }
6812 }
6813 spin_unlock(&flctx->flc_lock);
6814 }
6815 nfsd_file_put(nf);
6816 return status;
6817}
6818
6819__be32
6820nfsd4_release_lockowner(struct svc_rqst *rqstp,
6821 struct nfsd4_compound_state *cstate,
6822 union nfsd4_op_u *u)
6823{
6824 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6825 clientid_t *clid = &rlockowner->rl_clientid;
6826 struct nfs4_stateowner *sop;
6827 struct nfs4_lockowner *lo = NULL;
6828 struct nfs4_ol_stateid *stp;
6829 struct xdr_netobj *owner = &rlockowner->rl_owner;
6830 unsigned int hashval = ownerstr_hashval(owner);
6831 __be32 status;
6832 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6833 struct nfs4_client *clp;
6834 LIST_HEAD (reaplist);
6835
6836 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6837 clid->cl_boot, clid->cl_id);
6838
6839 status = lookup_clientid(clid, cstate, nn);
6840 if (status)
6841 return status;
6842
6843 clp = cstate->clp;
6844 /* Find the matching lock stateowner */
6845 spin_lock(&clp->cl_lock);
6846 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6847 so_strhash) {
6848
6849 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6850 continue;
6851
6852 /* see if there are still any locks associated with it */
6853 lo = lockowner(sop);
6854 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6855 if (check_for_locks(stp->st_stid.sc_file, lo)) {
6856 status = nfserr_locks_held;
6857 spin_unlock(&clp->cl_lock);
6858 return status;
6859 }
6860 }
6861
6862 nfs4_get_stateowner(sop);
6863 break;
6864 }
6865 if (!lo) {
6866 spin_unlock(&clp->cl_lock);
6867 return status;
6868 }
6869
6870 unhash_lockowner_locked(lo);
6871 while (!list_empty(&lo->lo_owner.so_stateids)) {
6872 stp = list_first_entry(&lo->lo_owner.so_stateids,
6873 struct nfs4_ol_stateid,
6874 st_perstateowner);
6875 WARN_ON(!unhash_lock_stateid(stp));
6876 put_ol_stateid_locked(stp, &reaplist);
6877 }
6878 spin_unlock(&clp->cl_lock);
6879 free_ol_stateid_reaplist(&reaplist);
6880 remove_blocked_locks(lo);
6881 nfs4_put_stateowner(&lo->lo_owner);
6882
6883 return status;
6884}
6885
6886static inline struct nfs4_client_reclaim *
6887alloc_reclaim(void)
6888{
6889 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6890}
6891
6892bool
6893nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
6894{
6895 struct nfs4_client_reclaim *crp;
6896
6897 crp = nfsd4_find_reclaim_client(name, nn);
6898 return (crp && crp->cr_clp);
6899}
6900
6901/*
6902 * failure => all reset bets are off, nfserr_no_grace...
6903 *
6904 * The caller is responsible for freeing name.data if NULL is returned (it
6905 * will be freed in nfs4_remove_reclaim_record in the normal case).
6906 */
6907struct nfs4_client_reclaim *
6908nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
6909 struct nfsd_net *nn)
6910{
6911 unsigned int strhashval;
6912 struct nfs4_client_reclaim *crp;
6913
6914 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
6915 crp = alloc_reclaim();
6916 if (crp) {
6917 strhashval = clientstr_hashval(name);
6918 INIT_LIST_HEAD(&crp->cr_strhash);
6919 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6920 crp->cr_name.data = name.data;
6921 crp->cr_name.len = name.len;
6922 crp->cr_princhash.data = princhash.data;
6923 crp->cr_princhash.len = princhash.len;
6924 crp->cr_clp = NULL;
6925 nn->reclaim_str_hashtbl_size++;
6926 }
6927 return crp;
6928}
6929
6930void
6931nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6932{
6933 list_del(&crp->cr_strhash);
6934 kfree(crp->cr_name.data);
6935 kfree(crp->cr_princhash.data);
6936 kfree(crp);
6937 nn->reclaim_str_hashtbl_size--;
6938}
6939
6940void
6941nfs4_release_reclaim(struct nfsd_net *nn)
6942{
6943 struct nfs4_client_reclaim *crp = NULL;
6944 int i;
6945
6946 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6947 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6948 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6949 struct nfs4_client_reclaim, cr_strhash);
6950 nfs4_remove_reclaim_record(crp, nn);
6951 }
6952 }
6953 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6954}
6955
6956/*
6957 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6958struct nfs4_client_reclaim *
6959nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
6960{
6961 unsigned int strhashval;
6962 struct nfs4_client_reclaim *crp = NULL;
6963
6964 dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
6965
6966 strhashval = clientstr_hashval(name);
6967 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6968 if (compare_blob(&crp->cr_name, &name) == 0) {
6969 return crp;
6970 }
6971 }
6972 return NULL;
6973}
6974
6975/*
6976* Called from OPEN. Look for clientid in reclaim list.
6977*/
6978__be32
6979nfs4_check_open_reclaim(clientid_t *clid,
6980 struct nfsd4_compound_state *cstate,
6981 struct nfsd_net *nn)
6982{
6983 __be32 status;
6984
6985 /* find clientid in conf_id_hashtbl */
6986 status = lookup_clientid(clid, cstate, nn);
6987 if (status)
6988 return nfserr_reclaim_bad;
6989
6990 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6991 return nfserr_no_grace;
6992
6993 if (nfsd4_client_record_check(cstate->clp))
6994 return nfserr_reclaim_bad;
6995
6996 return nfs_ok;
6997}
6998
6999#ifdef CONFIG_NFSD_FAULT_INJECTION
7000static inline void
7001put_client(struct nfs4_client *clp)
7002{
7003 atomic_dec(&clp->cl_rpc_users);
7004}
7005
7006static struct nfs4_client *
7007nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
7008{
7009 struct nfs4_client *clp;
7010 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7011 nfsd_net_id);
7012
7013 if (!nfsd_netns_ready(nn))
7014 return NULL;
7015
7016 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7017 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
7018 return clp;
7019 }
7020 return NULL;
7021}
7022
7023u64
7024nfsd_inject_print_clients(void)
7025{
7026 struct nfs4_client *clp;
7027 u64 count = 0;
7028 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7029 nfsd_net_id);
7030 char buf[INET6_ADDRSTRLEN];
7031
7032 if (!nfsd_netns_ready(nn))
7033 return 0;
7034
7035 spin_lock(&nn->client_lock);
7036 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7037 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7038 pr_info("NFS Client: %s\n", buf);
7039 ++count;
7040 }
7041 spin_unlock(&nn->client_lock);
7042
7043 return count;
7044}
7045
7046u64
7047nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7048{
7049 u64 count = 0;
7050 struct nfs4_client *clp;
7051 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7052 nfsd_net_id);
7053
7054 if (!nfsd_netns_ready(nn))
7055 return count;
7056
7057 spin_lock(&nn->client_lock);
7058 clp = nfsd_find_client(addr, addr_size);
7059 if (clp) {
7060 if (mark_client_expired_locked(clp) == nfs_ok)
7061 ++count;
7062 else
7063 clp = NULL;
7064 }
7065 spin_unlock(&nn->client_lock);
7066
7067 if (clp)
7068 expire_client(clp);
7069
7070 return count;
7071}
7072
7073u64
7074nfsd_inject_forget_clients(u64 max)
7075{
7076 u64 count = 0;
7077 struct nfs4_client *clp, *next;
7078 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7079 nfsd_net_id);
7080 LIST_HEAD(reaplist);
7081
7082 if (!nfsd_netns_ready(nn))
7083 return count;
7084
7085 spin_lock(&nn->client_lock);
7086 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7087 if (mark_client_expired_locked(clp) == nfs_ok) {
7088 list_add(&clp->cl_lru, &reaplist);
7089 if (max != 0 && ++count >= max)
7090 break;
7091 }
7092 }
7093 spin_unlock(&nn->client_lock);
7094
7095 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7096 expire_client(clp);
7097
7098 return count;
7099}
7100
7101static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7102 const char *type)
7103{
7104 char buf[INET6_ADDRSTRLEN];
7105 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7106 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7107}
7108
7109static void
7110nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7111 struct list_head *collect)
7112{
7113 struct nfs4_client *clp = lst->st_stid.sc_client;
7114 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7115 nfsd_net_id);
7116
7117 if (!collect)
7118 return;
7119
7120 lockdep_assert_held(&nn->client_lock);
7121 atomic_inc(&clp->cl_rpc_users);
7122 list_add(&lst->st_locks, collect);
7123}
7124
7125static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7126 struct list_head *collect,
7127 bool (*func)(struct nfs4_ol_stateid *))
7128{
7129 struct nfs4_openowner *oop;
7130 struct nfs4_ol_stateid *stp, *st_next;
7131 struct nfs4_ol_stateid *lst, *lst_next;
7132 u64 count = 0;
7133
7134 spin_lock(&clp->cl_lock);
7135 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7136 list_for_each_entry_safe(stp, st_next,
7137 &oop->oo_owner.so_stateids, st_perstateowner) {
7138 list_for_each_entry_safe(lst, lst_next,
7139 &stp->st_locks, st_locks) {
7140 if (func) {
7141 if (func(lst))
7142 nfsd_inject_add_lock_to_list(lst,
7143 collect);
7144 }
7145 ++count;
7146 /*
7147 * Despite the fact that these functions deal
7148 * with 64-bit integers for "count", we must
7149 * ensure that it doesn't blow up the
7150 * clp->cl_rpc_users. Throw a warning if we
7151 * start to approach INT_MAX here.
7152 */
7153 WARN_ON_ONCE(count == (INT_MAX / 2));
7154 if (count == max)
7155 goto out;
7156 }
7157 }
7158 }
7159out:
7160 spin_unlock(&clp->cl_lock);
7161
7162 return count;
7163}
7164
7165static u64
7166nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7167 u64 max)
7168{
7169 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7170}
7171
7172static u64
7173nfsd_print_client_locks(struct nfs4_client *clp)
7174{
7175 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7176 nfsd_print_count(clp, count, "locked files");
7177 return count;
7178}
7179
7180u64
7181nfsd_inject_print_locks(void)
7182{
7183 struct nfs4_client *clp;
7184 u64 count = 0;
7185 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7186 nfsd_net_id);
7187
7188 if (!nfsd_netns_ready(nn))
7189 return 0;
7190
7191 spin_lock(&nn->client_lock);
7192 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7193 count += nfsd_print_client_locks(clp);
7194 spin_unlock(&nn->client_lock);
7195
7196 return count;
7197}
7198
7199static void
7200nfsd_reap_locks(struct list_head *reaplist)
7201{
7202 struct nfs4_client *clp;
7203 struct nfs4_ol_stateid *stp, *next;
7204
7205 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7206 list_del_init(&stp->st_locks);
7207 clp = stp->st_stid.sc_client;
7208 nfs4_put_stid(&stp->st_stid);
7209 put_client(clp);
7210 }
7211}
7212
7213u64
7214nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7215{
7216 unsigned int count = 0;
7217 struct nfs4_client *clp;
7218 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7219 nfsd_net_id);
7220 LIST_HEAD(reaplist);
7221
7222 if (!nfsd_netns_ready(nn))
7223 return count;
7224
7225 spin_lock(&nn->client_lock);
7226 clp = nfsd_find_client(addr, addr_size);
7227 if (clp)
7228 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7229 spin_unlock(&nn->client_lock);
7230 nfsd_reap_locks(&reaplist);
7231 return count;
7232}
7233
7234u64
7235nfsd_inject_forget_locks(u64 max)
7236{
7237 u64 count = 0;
7238 struct nfs4_client *clp;
7239 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7240 nfsd_net_id);
7241 LIST_HEAD(reaplist);
7242
7243 if (!nfsd_netns_ready(nn))
7244 return count;
7245
7246 spin_lock(&nn->client_lock);
7247 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7248 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7249 if (max != 0 && count >= max)
7250 break;
7251 }
7252 spin_unlock(&nn->client_lock);
7253 nfsd_reap_locks(&reaplist);
7254 return count;
7255}
7256
7257static u64
7258nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7259 struct list_head *collect,
7260 void (*func)(struct nfs4_openowner *))
7261{
7262 struct nfs4_openowner *oop, *next;
7263 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7264 nfsd_net_id);
7265 u64 count = 0;
7266
7267 lockdep_assert_held(&nn->client_lock);
7268
7269 spin_lock(&clp->cl_lock);
7270 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7271 if (func) {
7272 func(oop);
7273 if (collect) {
7274 atomic_inc(&clp->cl_rpc_users);
7275 list_add(&oop->oo_perclient, collect);
7276 }
7277 }
7278 ++count;
7279 /*
7280 * Despite the fact that these functions deal with
7281 * 64-bit integers for "count", we must ensure that
7282 * it doesn't blow up the clp->cl_rpc_users. Throw a
7283 * warning if we start to approach INT_MAX here.
7284 */
7285 WARN_ON_ONCE(count == (INT_MAX / 2));
7286 if (count == max)
7287 break;
7288 }
7289 spin_unlock(&clp->cl_lock);
7290
7291 return count;
7292}
7293
7294static u64
7295nfsd_print_client_openowners(struct nfs4_client *clp)
7296{
7297 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7298
7299 nfsd_print_count(clp, count, "openowners");
7300 return count;
7301}
7302
7303static u64
7304nfsd_collect_client_openowners(struct nfs4_client *clp,
7305 struct list_head *collect, u64 max)
7306{
7307 return nfsd_foreach_client_openowner(clp, max, collect,
7308 unhash_openowner_locked);
7309}
7310
7311u64
7312nfsd_inject_print_openowners(void)
7313{
7314 struct nfs4_client *clp;
7315 u64 count = 0;
7316 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7317 nfsd_net_id);
7318
7319 if (!nfsd_netns_ready(nn))
7320 return 0;
7321
7322 spin_lock(&nn->client_lock);
7323 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7324 count += nfsd_print_client_openowners(clp);
7325 spin_unlock(&nn->client_lock);
7326
7327 return count;
7328}
7329
7330static void
7331nfsd_reap_openowners(struct list_head *reaplist)
7332{
7333 struct nfs4_client *clp;
7334 struct nfs4_openowner *oop, *next;
7335
7336 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7337 list_del_init(&oop->oo_perclient);
7338 clp = oop->oo_owner.so_client;
7339 release_openowner(oop);
7340 put_client(clp);
7341 }
7342}
7343
7344u64
7345nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7346 size_t addr_size)
7347{
7348 unsigned int count = 0;
7349 struct nfs4_client *clp;
7350 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7351 nfsd_net_id);
7352 LIST_HEAD(reaplist);
7353
7354 if (!nfsd_netns_ready(nn))
7355 return count;
7356
7357 spin_lock(&nn->client_lock);
7358 clp = nfsd_find_client(addr, addr_size);
7359 if (clp)
7360 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7361 spin_unlock(&nn->client_lock);
7362 nfsd_reap_openowners(&reaplist);
7363 return count;
7364}
7365
7366u64
7367nfsd_inject_forget_openowners(u64 max)
7368{
7369 u64 count = 0;
7370 struct nfs4_client *clp;
7371 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7372 nfsd_net_id);
7373 LIST_HEAD(reaplist);
7374
7375 if (!nfsd_netns_ready(nn))
7376 return count;
7377
7378 spin_lock(&nn->client_lock);
7379 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7380 count += nfsd_collect_client_openowners(clp, &reaplist,
7381 max - count);
7382 if (max != 0 && count >= max)
7383 break;
7384 }
7385 spin_unlock(&nn->client_lock);
7386 nfsd_reap_openowners(&reaplist);
7387 return count;
7388}
7389
7390static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7391 struct list_head *victims)
7392{
7393 struct nfs4_delegation *dp, *next;
7394 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7395 nfsd_net_id);
7396 u64 count = 0;
7397
7398 lockdep_assert_held(&nn->client_lock);
7399
7400 spin_lock(&state_lock);
7401 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7402 if (victims) {
7403 /*
7404 * It's not safe to mess with delegations that have a
7405 * non-zero dl_time. They might have already been broken
7406 * and could be processed by the laundromat outside of
7407 * the state_lock. Just leave them be.
7408 */
7409 if (dp->dl_time != 0)
7410 continue;
7411
7412 atomic_inc(&clp->cl_rpc_users);
7413 WARN_ON(!unhash_delegation_locked(dp));
7414 list_add(&dp->dl_recall_lru, victims);
7415 }
7416 ++count;
7417 /*
7418 * Despite the fact that these functions deal with
7419 * 64-bit integers for "count", we must ensure that
7420 * it doesn't blow up the clp->cl_rpc_users. Throw a
7421 * warning if we start to approach INT_MAX here.
7422 */
7423 WARN_ON_ONCE(count == (INT_MAX / 2));
7424 if (count == max)
7425 break;
7426 }
7427 spin_unlock(&state_lock);
7428 return count;
7429}
7430
7431static u64
7432nfsd_print_client_delegations(struct nfs4_client *clp)
7433{
7434 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7435
7436 nfsd_print_count(clp, count, "delegations");
7437 return count;
7438}
7439
7440u64
7441nfsd_inject_print_delegations(void)
7442{
7443 struct nfs4_client *clp;
7444 u64 count = 0;
7445 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7446 nfsd_net_id);
7447
7448 if (!nfsd_netns_ready(nn))
7449 return 0;
7450
7451 spin_lock(&nn->client_lock);
7452 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7453 count += nfsd_print_client_delegations(clp);
7454 spin_unlock(&nn->client_lock);
7455
7456 return count;
7457}
7458
7459static void
7460nfsd_forget_delegations(struct list_head *reaplist)
7461{
7462 struct nfs4_client *clp;
7463 struct nfs4_delegation *dp, *next;
7464
7465 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7466 list_del_init(&dp->dl_recall_lru);
7467 clp = dp->dl_stid.sc_client;
7468 revoke_delegation(dp);
7469 put_client(clp);
7470 }
7471}
7472
7473u64
7474nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7475 size_t addr_size)
7476{
7477 u64 count = 0;
7478 struct nfs4_client *clp;
7479 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7480 nfsd_net_id);
7481 LIST_HEAD(reaplist);
7482
7483 if (!nfsd_netns_ready(nn))
7484 return count;
7485
7486 spin_lock(&nn->client_lock);
7487 clp = nfsd_find_client(addr, addr_size);
7488 if (clp)
7489 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7490 spin_unlock(&nn->client_lock);
7491
7492 nfsd_forget_delegations(&reaplist);
7493 return count;
7494}
7495
7496u64
7497nfsd_inject_forget_delegations(u64 max)
7498{
7499 u64 count = 0;
7500 struct nfs4_client *clp;
7501 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7502 nfsd_net_id);
7503 LIST_HEAD(reaplist);
7504
7505 if (!nfsd_netns_ready(nn))
7506 return count;
7507
7508 spin_lock(&nn->client_lock);
7509 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7510 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7511 if (max != 0 && count >= max)
7512 break;
7513 }
7514 spin_unlock(&nn->client_lock);
7515 nfsd_forget_delegations(&reaplist);
7516 return count;
7517}
7518
7519static void
7520nfsd_recall_delegations(struct list_head *reaplist)
7521{
7522 struct nfs4_client *clp;
7523 struct nfs4_delegation *dp, *next;
7524
7525 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7526 list_del_init(&dp->dl_recall_lru);
7527 clp = dp->dl_stid.sc_client;
7528 /*
7529 * We skipped all entries that had a zero dl_time before,
7530 * so we can now reset the dl_time back to 0. If a delegation
7531 * break comes in now, then it won't make any difference since
7532 * we're recalling it either way.
7533 */
7534 spin_lock(&state_lock);
7535 dp->dl_time = 0;
7536 spin_unlock(&state_lock);
7537 nfsd_break_one_deleg(dp);
7538 put_client(clp);
7539 }
7540}
7541
7542u64
7543nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7544 size_t addr_size)
7545{
7546 u64 count = 0;
7547 struct nfs4_client *clp;
7548 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7549 nfsd_net_id);
7550 LIST_HEAD(reaplist);
7551
7552 if (!nfsd_netns_ready(nn))
7553 return count;
7554
7555 spin_lock(&nn->client_lock);
7556 clp = nfsd_find_client(addr, addr_size);
7557 if (clp)
7558 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7559 spin_unlock(&nn->client_lock);
7560
7561 nfsd_recall_delegations(&reaplist);
7562 return count;
7563}
7564
7565u64
7566nfsd_inject_recall_delegations(u64 max)
7567{
7568 u64 count = 0;
7569 struct nfs4_client *clp, *next;
7570 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7571 nfsd_net_id);
7572 LIST_HEAD(reaplist);
7573
7574 if (!nfsd_netns_ready(nn))
7575 return count;
7576
7577 spin_lock(&nn->client_lock);
7578 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7579 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7580 if (max != 0 && ++count >= max)
7581 break;
7582 }
7583 spin_unlock(&nn->client_lock);
7584 nfsd_recall_delegations(&reaplist);
7585 return count;
7586}
7587#endif /* CONFIG_NFSD_FAULT_INJECTION */
7588
7589/*
7590 * Since the lifetime of a delegation isn't limited to that of an open, a
7591 * client may quite reasonably hang on to a delegation as long as it has
7592 * the inode cached. This becomes an obvious problem the first time a
7593 * client's inode cache approaches the size of the server's total memory.
7594 *
7595 * For now we avoid this problem by imposing a hard limit on the number
7596 * of delegations, which varies according to the server's memory size.
7597 */
7598static void
7599set_max_delegations(void)
7600{
7601 /*
7602 * Allow at most 4 delegations per megabyte of RAM. Quick
7603 * estimates suggest that in the worst case (where every delegation
7604 * is for a different inode), a delegation could take about 1.5K,
7605 * giving a worst case usage of about 6% of memory.
7606 */
7607 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7608}
7609
7610static int nfs4_state_create_net(struct net *net)
7611{
7612 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7613 int i;
7614
7615 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7616 sizeof(struct list_head),
7617 GFP_KERNEL);
7618 if (!nn->conf_id_hashtbl)
7619 goto err;
7620 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7621 sizeof(struct list_head),
7622 GFP_KERNEL);
7623 if (!nn->unconf_id_hashtbl)
7624 goto err_unconf_id;
7625 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7626 sizeof(struct list_head),
7627 GFP_KERNEL);
7628 if (!nn->sessionid_hashtbl)
7629 goto err_sessionid;
7630
7631 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7632 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7633 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7634 }
7635 for (i = 0; i < SESSION_HASH_SIZE; i++)
7636 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7637 nn->conf_name_tree = RB_ROOT;
7638 nn->unconf_name_tree = RB_ROOT;
7639 nn->boot_time = get_seconds();
7640 nn->grace_ended = false;
7641 nn->nfsd4_manager.block_opens = true;
7642 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7643 INIT_LIST_HEAD(&nn->client_lru);
7644 INIT_LIST_HEAD(&nn->close_lru);
7645 INIT_LIST_HEAD(&nn->del_recall_lru);
7646 spin_lock_init(&nn->client_lock);
7647 spin_lock_init(&nn->s2s_cp_lock);
7648 idr_init(&nn->s2s_cp_stateids);
7649
7650 spin_lock_init(&nn->blocked_locks_lock);
7651 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7652
7653 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7654 get_net(net);
7655
7656 return 0;
7657
7658err_sessionid:
7659 kfree(nn->unconf_id_hashtbl);
7660err_unconf_id:
7661 kfree(nn->conf_id_hashtbl);
7662err:
7663 return -ENOMEM;
7664}
7665
7666static void
7667nfs4_state_destroy_net(struct net *net)
7668{
7669 int i;
7670 struct nfs4_client *clp = NULL;
7671 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7672
7673 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7674 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7675 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7676 destroy_client(clp);
7677 }
7678 }
7679
7680 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7681
7682 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7683 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7684 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7685 destroy_client(clp);
7686 }
7687 }
7688
7689 kfree(nn->sessionid_hashtbl);
7690 kfree(nn->unconf_id_hashtbl);
7691 kfree(nn->conf_id_hashtbl);
7692 put_net(net);
7693}
7694
7695int
7696nfs4_state_start_net(struct net *net)
7697{
7698 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7699 int ret;
7700
7701 ret = nfs4_state_create_net(net);
7702 if (ret)
7703 return ret;
7704 locks_start_grace(net, &nn->nfsd4_manager);
7705 nfsd4_client_tracking_init(net);
7706 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7707 goto skip_grace;
7708 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7709 nn->nfsd4_grace, net->ns.inum);
7710 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7711 return 0;
7712
7713skip_grace:
7714 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7715 net->ns.inum);
7716 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7717 nfsd4_end_grace(nn);
7718 return 0;
7719}
7720
7721/* initialization to perform when the nfsd service is started: */
7722
7723int
7724nfs4_state_start(void)
7725{
7726 int ret;
7727
7728 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7729 if (laundry_wq == NULL) {
7730 ret = -ENOMEM;
7731 goto out;
7732 }
7733 ret = nfsd4_create_callback_queue();
7734 if (ret)
7735 goto out_free_laundry;
7736
7737 set_max_delegations();
7738 return 0;
7739
7740out_free_laundry:
7741 destroy_workqueue(laundry_wq);
7742out:
7743 return ret;
7744}
7745
7746void
7747nfs4_state_shutdown_net(struct net *net)
7748{
7749 struct nfs4_delegation *dp = NULL;
7750 struct list_head *pos, *next, reaplist;
7751 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7752
7753 cancel_delayed_work_sync(&nn->laundromat_work);
7754 locks_end_grace(&nn->nfsd4_manager);
7755
7756 INIT_LIST_HEAD(&reaplist);
7757 spin_lock(&state_lock);
7758 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7759 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7760 WARN_ON(!unhash_delegation_locked(dp));
7761 list_add(&dp->dl_recall_lru, &reaplist);
7762 }
7763 spin_unlock(&state_lock);
7764 list_for_each_safe(pos, next, &reaplist) {
7765 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7766 list_del_init(&dp->dl_recall_lru);
7767 destroy_unhashed_deleg(dp);
7768 }
7769
7770 nfsd4_client_tracking_exit(net);
7771 nfs4_state_destroy_net(net);
7772}
7773
7774void
7775nfs4_state_shutdown(void)
7776{
7777 destroy_workqueue(laundry_wq);
7778 nfsd4_destroy_callback_queue();
7779}
7780
7781static void
7782get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7783{
7784 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7785 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7786}
7787
7788static void
7789put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7790{
7791 if (cstate->minorversion) {
7792 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7793 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7794 }
7795}
7796
7797void
7798clear_current_stateid(struct nfsd4_compound_state *cstate)
7799{
7800 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7801}
7802
7803/*
7804 * functions to set current state id
7805 */
7806void
7807nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7808 union nfsd4_op_u *u)
7809{
7810 put_stateid(cstate, &u->open_downgrade.od_stateid);
7811}
7812
7813void
7814nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7815 union nfsd4_op_u *u)
7816{
7817 put_stateid(cstate, &u->open.op_stateid);
7818}
7819
7820void
7821nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7822 union nfsd4_op_u *u)
7823{
7824 put_stateid(cstate, &u->close.cl_stateid);
7825}
7826
7827void
7828nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7829 union nfsd4_op_u *u)
7830{
7831 put_stateid(cstate, &u->lock.lk_resp_stateid);
7832}
7833
7834/*
7835 * functions to consume current state id
7836 */
7837
7838void
7839nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7840 union nfsd4_op_u *u)
7841{
7842 get_stateid(cstate, &u->open_downgrade.od_stateid);
7843}
7844
7845void
7846nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7847 union nfsd4_op_u *u)
7848{
7849 get_stateid(cstate, &u->delegreturn.dr_stateid);
7850}
7851
7852void
7853nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7854 union nfsd4_op_u *u)
7855{
7856 get_stateid(cstate, &u->free_stateid.fr_stateid);
7857}
7858
7859void
7860nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7861 union nfsd4_op_u *u)
7862{
7863 get_stateid(cstate, &u->setattr.sa_stateid);
7864}
7865
7866void
7867nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7868 union nfsd4_op_u *u)
7869{
7870 get_stateid(cstate, &u->close.cl_stateid);
7871}
7872
7873void
7874nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7875 union nfsd4_op_u *u)
7876{
7877 get_stateid(cstate, &u->locku.lu_stateid);
7878}
7879
7880void
7881nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7882 union nfsd4_op_u *u)
7883{
7884 get_stateid(cstate, &u->read.rd_stateid);
7885}
7886
7887void
7888nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7889 union nfsd4_op_u *u)
7890{
7891 get_stateid(cstate, &u->write.wr_stateid);
7892}