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/sunrpc/svcauth_gss.h>
42#include <linux/sunrpc/clnt.h>
43#include "xdr4.h"
44#include "vfs.h"
45#include "current_stateid.h"
46
47#define NFSDDBG_FACILITY NFSDDBG_PROC
48
49/* Globals */
50time_t nfsd4_lease = 90; /* default lease time */
51time_t nfsd4_grace = 90;
52static time_t boot_time;
53
54#define all_ones {{~0,~0},~0}
55static const stateid_t one_stateid = {
56 .si_generation = ~0,
57 .si_opaque = all_ones,
58};
59static const stateid_t zero_stateid = {
60 /* all fields zero */
61};
62static const stateid_t currentstateid = {
63 .si_generation = 1,
64};
65
66static u64 current_sessionid = 1;
67
68#define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
69#define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
70#define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
71
72/* forward declarations */
73static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
74
75/* Locking: */
76
77/* Currently used for almost all code touching nfsv4 state: */
78static DEFINE_MUTEX(client_mutex);
79
80/*
81 * Currently used for the del_recall_lru and file hash table. In an
82 * effort to decrease the scope of the client_mutex, this spinlock may
83 * eventually cover more:
84 */
85static DEFINE_SPINLOCK(recall_lock);
86
87static struct kmem_cache *openowner_slab = NULL;
88static struct kmem_cache *lockowner_slab = NULL;
89static struct kmem_cache *file_slab = NULL;
90static struct kmem_cache *stateid_slab = NULL;
91static struct kmem_cache *deleg_slab = NULL;
92
93void
94nfs4_lock_state(void)
95{
96 mutex_lock(&client_mutex);
97}
98
99static void free_session(struct kref *);
100
101/* Must be called under the client_lock */
102static void nfsd4_put_session_locked(struct nfsd4_session *ses)
103{
104 kref_put(&ses->se_ref, free_session);
105}
106
107static void nfsd4_get_session(struct nfsd4_session *ses)
108{
109 kref_get(&ses->se_ref);
110}
111
112void
113nfs4_unlock_state(void)
114{
115 mutex_unlock(&client_mutex);
116}
117
118static inline u32
119opaque_hashval(const void *ptr, int nbytes)
120{
121 unsigned char *cptr = (unsigned char *) ptr;
122
123 u32 x = 0;
124 while (nbytes--) {
125 x *= 37;
126 x += *cptr++;
127 }
128 return x;
129}
130
131static struct list_head del_recall_lru;
132
133static void nfsd4_free_file(struct nfs4_file *f)
134{
135 kmem_cache_free(file_slab, f);
136}
137
138static inline void
139put_nfs4_file(struct nfs4_file *fi)
140{
141 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
142 list_del(&fi->fi_hash);
143 spin_unlock(&recall_lock);
144 iput(fi->fi_inode);
145 nfsd4_free_file(fi);
146 }
147}
148
149static inline void
150get_nfs4_file(struct nfs4_file *fi)
151{
152 atomic_inc(&fi->fi_ref);
153}
154
155static int num_delegations;
156unsigned int max_delegations;
157
158/*
159 * Open owner state (share locks)
160 */
161
162/* hash tables for lock and open owners */
163#define OWNER_HASH_BITS 8
164#define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
165#define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
166
167static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
168{
169 unsigned int ret;
170
171 ret = opaque_hashval(ownername->data, ownername->len);
172 ret += clientid;
173 return ret & OWNER_HASH_MASK;
174}
175
176static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
177
178/* hash table for nfs4_file */
179#define FILE_HASH_BITS 8
180#define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
181
182static unsigned int file_hashval(struct inode *ino)
183{
184 /* XXX: why are we hashing on inode pointer, anyway? */
185 return hash_ptr(ino, FILE_HASH_BITS);
186}
187
188static struct list_head file_hashtbl[FILE_HASH_SIZE];
189
190static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
191{
192 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
193 atomic_inc(&fp->fi_access[oflag]);
194}
195
196static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
197{
198 if (oflag == O_RDWR) {
199 __nfs4_file_get_access(fp, O_RDONLY);
200 __nfs4_file_get_access(fp, O_WRONLY);
201 } else
202 __nfs4_file_get_access(fp, oflag);
203}
204
205static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
206{
207 if (fp->fi_fds[oflag]) {
208 fput(fp->fi_fds[oflag]);
209 fp->fi_fds[oflag] = NULL;
210 }
211}
212
213static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
214{
215 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
216 nfs4_file_put_fd(fp, oflag);
217 /*
218 * It's also safe to get rid of the RDWR open *if*
219 * we no longer have need of the other kind of access
220 * or if we already have the other kind of open:
221 */
222 if (fp->fi_fds[1-oflag]
223 || atomic_read(&fp->fi_access[1 - oflag]) == 0)
224 nfs4_file_put_fd(fp, O_RDWR);
225 }
226}
227
228static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
229{
230 if (oflag == O_RDWR) {
231 __nfs4_file_put_access(fp, O_RDONLY);
232 __nfs4_file_put_access(fp, O_WRONLY);
233 } else
234 __nfs4_file_put_access(fp, oflag);
235}
236
237static inline int get_new_stid(struct nfs4_stid *stid)
238{
239 static int min_stateid = 0;
240 struct idr *stateids = &stid->sc_client->cl_stateids;
241 int new_stid;
242 int error;
243
244 error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
245 /*
246 * Note: the necessary preallocation was done in
247 * nfs4_alloc_stateid(). The idr code caps the number of
248 * preallocations that can exist at a time, but the state lock
249 * prevents anyone from using ours before we get here:
250 */
251 BUG_ON(error);
252 /*
253 * It shouldn't be a problem to reuse an opaque stateid value.
254 * I don't think it is for 4.1. But with 4.0 I worry that, for
255 * example, a stray write retransmission could be accepted by
256 * the server when it should have been rejected. Therefore,
257 * adopt a trick from the sctp code to attempt to maximize the
258 * amount of time until an id is reused, by ensuring they always
259 * "increase" (mod INT_MAX):
260 */
261
262 min_stateid = new_stid+1;
263 if (min_stateid == INT_MAX)
264 min_stateid = 0;
265 return new_stid;
266}
267
268static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
269{
270 stateid_t *s = &stid->sc_stateid;
271 int new_id;
272
273 stid->sc_type = type;
274 stid->sc_client = cl;
275 s->si_opaque.so_clid = cl->cl_clientid;
276 new_id = get_new_stid(stid);
277 s->si_opaque.so_id = (u32)new_id;
278 /* Will be incremented before return to client: */
279 s->si_generation = 0;
280}
281
282static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
283{
284 struct idr *stateids = &cl->cl_stateids;
285
286 if (!idr_pre_get(stateids, GFP_KERNEL))
287 return NULL;
288 /*
289 * Note: if we fail here (or any time between now and the time
290 * we actually get the new idr), we won't need to undo the idr
291 * preallocation, since the idr code caps the number of
292 * preallocated entries.
293 */
294 return kmem_cache_alloc(slab, GFP_KERNEL);
295}
296
297static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
298{
299 return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
300}
301
302static struct nfs4_delegation *
303alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
304{
305 struct nfs4_delegation *dp;
306 struct nfs4_file *fp = stp->st_file;
307
308 dprintk("NFSD alloc_init_deleg\n");
309 /*
310 * Major work on the lease subsystem (for example, to support
311 * calbacks on stat) will be required before we can support
312 * write delegations properly.
313 */
314 if (type != NFS4_OPEN_DELEGATE_READ)
315 return NULL;
316 if (fp->fi_had_conflict)
317 return NULL;
318 if (num_delegations > max_delegations)
319 return NULL;
320 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
321 if (dp == NULL)
322 return dp;
323 init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
324 /*
325 * delegation seqid's are never incremented. The 4.1 special
326 * meaning of seqid 0 isn't meaningful, really, but let's avoid
327 * 0 anyway just for consistency and use 1:
328 */
329 dp->dl_stid.sc_stateid.si_generation = 1;
330 num_delegations++;
331 INIT_LIST_HEAD(&dp->dl_perfile);
332 INIT_LIST_HEAD(&dp->dl_perclnt);
333 INIT_LIST_HEAD(&dp->dl_recall_lru);
334 get_nfs4_file(fp);
335 dp->dl_file = fp;
336 dp->dl_type = type;
337 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
338 dp->dl_time = 0;
339 atomic_set(&dp->dl_count, 1);
340 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
341 return dp;
342}
343
344void
345nfs4_put_delegation(struct nfs4_delegation *dp)
346{
347 if (atomic_dec_and_test(&dp->dl_count)) {
348 dprintk("NFSD: freeing dp %p\n",dp);
349 put_nfs4_file(dp->dl_file);
350 kmem_cache_free(deleg_slab, dp);
351 num_delegations--;
352 }
353}
354
355static void nfs4_put_deleg_lease(struct nfs4_file *fp)
356{
357 if (atomic_dec_and_test(&fp->fi_delegees)) {
358 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
359 fp->fi_lease = NULL;
360 fput(fp->fi_deleg_file);
361 fp->fi_deleg_file = NULL;
362 }
363}
364
365static void unhash_stid(struct nfs4_stid *s)
366{
367 struct idr *stateids = &s->sc_client->cl_stateids;
368
369 idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
370}
371
372/* Called under the state lock. */
373static void
374unhash_delegation(struct nfs4_delegation *dp)
375{
376 unhash_stid(&dp->dl_stid);
377 list_del_init(&dp->dl_perclnt);
378 spin_lock(&recall_lock);
379 list_del_init(&dp->dl_perfile);
380 list_del_init(&dp->dl_recall_lru);
381 spin_unlock(&recall_lock);
382 nfs4_put_deleg_lease(dp->dl_file);
383 nfs4_put_delegation(dp);
384}
385
386/*
387 * SETCLIENTID state
388 */
389
390/* client_lock protects the client lru list and session hash table */
391static DEFINE_SPINLOCK(client_lock);
392
393/* Hash tables for nfs4_clientid state */
394#define CLIENT_HASH_BITS 4
395#define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
396#define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
397
398static unsigned int clientid_hashval(u32 id)
399{
400 return id & CLIENT_HASH_MASK;
401}
402
403static unsigned int clientstr_hashval(const char *name)
404{
405 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
406}
407
408/*
409 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
410 * used in reboot/reset lease grace period processing
411 *
412 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
413 * setclientid_confirmed info.
414 *
415 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
416 * setclientid info.
417 *
418 * client_lru holds client queue ordered by nfs4_client.cl_time
419 * for lease renewal.
420 *
421 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
422 * for last close replay.
423 */
424static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
425static int reclaim_str_hashtbl_size = 0;
426static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
427static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
428static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
429static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
430static struct list_head client_lru;
431static struct list_head close_lru;
432
433/*
434 * We store the NONE, READ, WRITE, and BOTH bits separately in the
435 * st_{access,deny}_bmap field of the stateid, in order to track not
436 * only what share bits are currently in force, but also what
437 * combinations of share bits previous opens have used. This allows us
438 * to enforce the recommendation of rfc 3530 14.2.19 that the server
439 * return an error if the client attempt to downgrade to a combination
440 * of share bits not explicable by closing some of its previous opens.
441 *
442 * XXX: This enforcement is actually incomplete, since we don't keep
443 * track of access/deny bit combinations; so, e.g., we allow:
444 *
445 * OPEN allow read, deny write
446 * OPEN allow both, deny none
447 * DOWNGRADE allow read, deny none
448 *
449 * which we should reject.
450 */
451static unsigned int
452bmap_to_share_mode(unsigned long bmap) {
453 int i;
454 unsigned int access = 0;
455
456 for (i = 1; i < 4; i++) {
457 if (test_bit(i, &bmap))
458 access |= i;
459 }
460 return access;
461}
462
463static bool
464test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
465 unsigned int access, deny;
466
467 access = bmap_to_share_mode(stp->st_access_bmap);
468 deny = bmap_to_share_mode(stp->st_deny_bmap);
469 if ((access & open->op_share_deny) || (deny & open->op_share_access))
470 return false;
471 return true;
472}
473
474/* set share access for a given stateid */
475static inline void
476set_access(u32 access, struct nfs4_ol_stateid *stp)
477{
478 __set_bit(access, &stp->st_access_bmap);
479}
480
481/* clear share access for a given stateid */
482static inline void
483clear_access(u32 access, struct nfs4_ol_stateid *stp)
484{
485 __clear_bit(access, &stp->st_access_bmap);
486}
487
488/* test whether a given stateid has access */
489static inline bool
490test_access(u32 access, struct nfs4_ol_stateid *stp)
491{
492 return test_bit(access, &stp->st_access_bmap);
493}
494
495/* set share deny for a given stateid */
496static inline void
497set_deny(u32 access, struct nfs4_ol_stateid *stp)
498{
499 __set_bit(access, &stp->st_deny_bmap);
500}
501
502/* clear share deny for a given stateid */
503static inline void
504clear_deny(u32 access, struct nfs4_ol_stateid *stp)
505{
506 __clear_bit(access, &stp->st_deny_bmap);
507}
508
509/* test whether a given stateid is denying specific access */
510static inline bool
511test_deny(u32 access, struct nfs4_ol_stateid *stp)
512{
513 return test_bit(access, &stp->st_deny_bmap);
514}
515
516static int nfs4_access_to_omode(u32 access)
517{
518 switch (access & NFS4_SHARE_ACCESS_BOTH) {
519 case NFS4_SHARE_ACCESS_READ:
520 return O_RDONLY;
521 case NFS4_SHARE_ACCESS_WRITE:
522 return O_WRONLY;
523 case NFS4_SHARE_ACCESS_BOTH:
524 return O_RDWR;
525 }
526 BUG();
527}
528
529/* release all access and file references for a given stateid */
530static void
531release_all_access(struct nfs4_ol_stateid *stp)
532{
533 int i;
534
535 for (i = 1; i < 4; i++) {
536 if (test_access(i, stp))
537 nfs4_file_put_access(stp->st_file,
538 nfs4_access_to_omode(i));
539 clear_access(i, stp);
540 }
541}
542
543static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
544{
545 list_del(&stp->st_perfile);
546 list_del(&stp->st_perstateowner);
547}
548
549static void close_generic_stateid(struct nfs4_ol_stateid *stp)
550{
551 release_all_access(stp);
552 put_nfs4_file(stp->st_file);
553 stp->st_file = NULL;
554}
555
556static void free_generic_stateid(struct nfs4_ol_stateid *stp)
557{
558 kmem_cache_free(stateid_slab, stp);
559}
560
561static void release_lock_stateid(struct nfs4_ol_stateid *stp)
562{
563 struct file *file;
564
565 unhash_generic_stateid(stp);
566 unhash_stid(&stp->st_stid);
567 file = find_any_file(stp->st_file);
568 if (file)
569 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
570 close_generic_stateid(stp);
571 free_generic_stateid(stp);
572}
573
574static void unhash_lockowner(struct nfs4_lockowner *lo)
575{
576 struct nfs4_ol_stateid *stp;
577
578 list_del(&lo->lo_owner.so_strhash);
579 list_del(&lo->lo_perstateid);
580 list_del(&lo->lo_owner_ino_hash);
581 while (!list_empty(&lo->lo_owner.so_stateids)) {
582 stp = list_first_entry(&lo->lo_owner.so_stateids,
583 struct nfs4_ol_stateid, st_perstateowner);
584 release_lock_stateid(stp);
585 }
586}
587
588static void release_lockowner(struct nfs4_lockowner *lo)
589{
590 unhash_lockowner(lo);
591 nfs4_free_lockowner(lo);
592}
593
594static void
595release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
596{
597 struct nfs4_lockowner *lo;
598
599 while (!list_empty(&open_stp->st_lockowners)) {
600 lo = list_entry(open_stp->st_lockowners.next,
601 struct nfs4_lockowner, lo_perstateid);
602 release_lockowner(lo);
603 }
604}
605
606static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
607{
608 unhash_generic_stateid(stp);
609 release_stateid_lockowners(stp);
610 close_generic_stateid(stp);
611}
612
613static void release_open_stateid(struct nfs4_ol_stateid *stp)
614{
615 unhash_open_stateid(stp);
616 unhash_stid(&stp->st_stid);
617 free_generic_stateid(stp);
618}
619
620static void unhash_openowner(struct nfs4_openowner *oo)
621{
622 struct nfs4_ol_stateid *stp;
623
624 list_del(&oo->oo_owner.so_strhash);
625 list_del(&oo->oo_perclient);
626 while (!list_empty(&oo->oo_owner.so_stateids)) {
627 stp = list_first_entry(&oo->oo_owner.so_stateids,
628 struct nfs4_ol_stateid, st_perstateowner);
629 release_open_stateid(stp);
630 }
631}
632
633static void release_last_closed_stateid(struct nfs4_openowner *oo)
634{
635 struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
636
637 if (s) {
638 unhash_stid(&s->st_stid);
639 free_generic_stateid(s);
640 oo->oo_last_closed_stid = NULL;
641 }
642}
643
644static void release_openowner(struct nfs4_openowner *oo)
645{
646 unhash_openowner(oo);
647 list_del(&oo->oo_close_lru);
648 release_last_closed_stateid(oo);
649 nfs4_free_openowner(oo);
650}
651
652#define SESSION_HASH_SIZE 512
653static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
654
655static inline int
656hash_sessionid(struct nfs4_sessionid *sessionid)
657{
658 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
659
660 return sid->sequence % SESSION_HASH_SIZE;
661}
662
663#ifdef NFSD_DEBUG
664static inline void
665dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
666{
667 u32 *ptr = (u32 *)(&sessionid->data[0]);
668 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
669}
670#else
671static inline void
672dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
673{
674}
675#endif
676
677
678static void
679gen_sessionid(struct nfsd4_session *ses)
680{
681 struct nfs4_client *clp = ses->se_client;
682 struct nfsd4_sessionid *sid;
683
684 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
685 sid->clientid = clp->cl_clientid;
686 sid->sequence = current_sessionid++;
687 sid->reserved = 0;
688}
689
690/*
691 * The protocol defines ca_maxresponssize_cached to include the size of
692 * the rpc header, but all we need to cache is the data starting after
693 * the end of the initial SEQUENCE operation--the rest we regenerate
694 * each time. Therefore we can advertise a ca_maxresponssize_cached
695 * value that is the number of bytes in our cache plus a few additional
696 * bytes. In order to stay on the safe side, and not promise more than
697 * we can cache, those additional bytes must be the minimum possible: 24
698 * bytes of rpc header (xid through accept state, with AUTH_NULL
699 * verifier), 12 for the compound header (with zero-length tag), and 44
700 * for the SEQUENCE op response:
701 */
702#define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
703
704static void
705free_session_slots(struct nfsd4_session *ses)
706{
707 int i;
708
709 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
710 kfree(ses->se_slots[i]);
711}
712
713/*
714 * We don't actually need to cache the rpc and session headers, so we
715 * can allocate a little less for each slot:
716 */
717static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
718{
719 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
720}
721
722static int nfsd4_sanitize_slot_size(u32 size)
723{
724 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
725 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
726
727 return size;
728}
729
730/*
731 * XXX: If we run out of reserved DRC memory we could (up to a point)
732 * re-negotiate active sessions and reduce their slot usage to make
733 * room for new connections. For now we just fail the create session.
734 */
735static int nfsd4_get_drc_mem(int slotsize, u32 num)
736{
737 int avail;
738
739 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
740
741 spin_lock(&nfsd_drc_lock);
742 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
743 nfsd_drc_max_mem - nfsd_drc_mem_used);
744 num = min_t(int, num, avail / slotsize);
745 nfsd_drc_mem_used += num * slotsize;
746 spin_unlock(&nfsd_drc_lock);
747
748 return num;
749}
750
751static void nfsd4_put_drc_mem(int slotsize, int num)
752{
753 spin_lock(&nfsd_drc_lock);
754 nfsd_drc_mem_used -= slotsize * num;
755 spin_unlock(&nfsd_drc_lock);
756}
757
758static struct nfsd4_session *alloc_session(int slotsize, int numslots)
759{
760 struct nfsd4_session *new;
761 int mem, i;
762
763 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
764 + sizeof(struct nfsd4_session) > PAGE_SIZE);
765 mem = numslots * sizeof(struct nfsd4_slot *);
766
767 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
768 if (!new)
769 return NULL;
770 /* allocate each struct nfsd4_slot and data cache in one piece */
771 for (i = 0; i < numslots; i++) {
772 mem = sizeof(struct nfsd4_slot) + slotsize;
773 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
774 if (!new->se_slots[i])
775 goto out_free;
776 }
777 return new;
778out_free:
779 while (i--)
780 kfree(new->se_slots[i]);
781 kfree(new);
782 return NULL;
783}
784
785static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
786{
787 u32 maxrpc = nfsd_serv->sv_max_mesg;
788
789 new->maxreqs = numslots;
790 new->maxresp_cached = min_t(u32, req->maxresp_cached,
791 slotsize + NFSD_MIN_HDR_SEQ_SZ);
792 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
793 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
794 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
795}
796
797static void free_conn(struct nfsd4_conn *c)
798{
799 svc_xprt_put(c->cn_xprt);
800 kfree(c);
801}
802
803static void nfsd4_conn_lost(struct svc_xpt_user *u)
804{
805 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
806 struct nfs4_client *clp = c->cn_session->se_client;
807
808 spin_lock(&clp->cl_lock);
809 if (!list_empty(&c->cn_persession)) {
810 list_del(&c->cn_persession);
811 free_conn(c);
812 }
813 spin_unlock(&clp->cl_lock);
814 nfsd4_probe_callback(clp);
815}
816
817static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
818{
819 struct nfsd4_conn *conn;
820
821 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
822 if (!conn)
823 return NULL;
824 svc_xprt_get(rqstp->rq_xprt);
825 conn->cn_xprt = rqstp->rq_xprt;
826 conn->cn_flags = flags;
827 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
828 return conn;
829}
830
831static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
832{
833 conn->cn_session = ses;
834 list_add(&conn->cn_persession, &ses->se_conns);
835}
836
837static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
838{
839 struct nfs4_client *clp = ses->se_client;
840
841 spin_lock(&clp->cl_lock);
842 __nfsd4_hash_conn(conn, ses);
843 spin_unlock(&clp->cl_lock);
844}
845
846static int nfsd4_register_conn(struct nfsd4_conn *conn)
847{
848 conn->cn_xpt_user.callback = nfsd4_conn_lost;
849 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
850}
851
852static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
853{
854 struct nfsd4_conn *conn;
855 int ret;
856
857 conn = alloc_conn(rqstp, dir);
858 if (!conn)
859 return nfserr_jukebox;
860 nfsd4_hash_conn(conn, ses);
861 ret = nfsd4_register_conn(conn);
862 if (ret)
863 /* oops; xprt is already down: */
864 nfsd4_conn_lost(&conn->cn_xpt_user);
865 return nfs_ok;
866}
867
868static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
869{
870 u32 dir = NFS4_CDFC4_FORE;
871
872 if (ses->se_flags & SESSION4_BACK_CHAN)
873 dir |= NFS4_CDFC4_BACK;
874
875 return nfsd4_new_conn(rqstp, ses, dir);
876}
877
878/* must be called under client_lock */
879static void nfsd4_del_conns(struct nfsd4_session *s)
880{
881 struct nfs4_client *clp = s->se_client;
882 struct nfsd4_conn *c;
883
884 spin_lock(&clp->cl_lock);
885 while (!list_empty(&s->se_conns)) {
886 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
887 list_del_init(&c->cn_persession);
888 spin_unlock(&clp->cl_lock);
889
890 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
891 free_conn(c);
892
893 spin_lock(&clp->cl_lock);
894 }
895 spin_unlock(&clp->cl_lock);
896}
897
898static void free_session(struct kref *kref)
899{
900 struct nfsd4_session *ses;
901 int mem;
902
903 lockdep_assert_held(&client_lock);
904 ses = container_of(kref, struct nfsd4_session, se_ref);
905 nfsd4_del_conns(ses);
906 spin_lock(&nfsd_drc_lock);
907 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
908 nfsd_drc_mem_used -= mem;
909 spin_unlock(&nfsd_drc_lock);
910 free_session_slots(ses);
911 kfree(ses);
912}
913
914void nfsd4_put_session(struct nfsd4_session *ses)
915{
916 spin_lock(&client_lock);
917 nfsd4_put_session_locked(ses);
918 spin_unlock(&client_lock);
919}
920
921static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
922{
923 struct nfsd4_session *new;
924 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
925 int numslots, slotsize;
926 __be32 status;
927 int idx;
928
929 /*
930 * Note decreasing slot size below client's request may
931 * make it difficult for client to function correctly, whereas
932 * decreasing the number of slots will (just?) affect
933 * performance. When short on memory we therefore prefer to
934 * decrease number of slots instead of their size.
935 */
936 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
937 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
938 if (numslots < 1)
939 return NULL;
940
941 new = alloc_session(slotsize, numslots);
942 if (!new) {
943 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
944 return NULL;
945 }
946 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
947
948 new->se_client = clp;
949 gen_sessionid(new);
950
951 INIT_LIST_HEAD(&new->se_conns);
952
953 new->se_cb_seq_nr = 1;
954 new->se_flags = cses->flags;
955 new->se_cb_prog = cses->callback_prog;
956 kref_init(&new->se_ref);
957 idx = hash_sessionid(&new->se_sessionid);
958 spin_lock(&client_lock);
959 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
960 spin_lock(&clp->cl_lock);
961 list_add(&new->se_perclnt, &clp->cl_sessions);
962 spin_unlock(&clp->cl_lock);
963 spin_unlock(&client_lock);
964
965 status = nfsd4_new_conn_from_crses(rqstp, new);
966 /* whoops: benny points out, status is ignored! (err, or bogus) */
967 if (status) {
968 spin_lock(&client_lock);
969 free_session(&new->se_ref);
970 spin_unlock(&client_lock);
971 return NULL;
972 }
973 if (cses->flags & SESSION4_BACK_CHAN) {
974 struct sockaddr *sa = svc_addr(rqstp);
975 /*
976 * This is a little silly; with sessions there's no real
977 * use for the callback address. Use the peer address
978 * as a reasonable default for now, but consider fixing
979 * the rpc client not to require an address in the
980 * future:
981 */
982 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
983 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
984 }
985 nfsd4_probe_callback(clp);
986 return new;
987}
988
989/* caller must hold client_lock */
990static struct nfsd4_session *
991find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
992{
993 struct nfsd4_session *elem;
994 int idx;
995
996 dump_sessionid(__func__, sessionid);
997 idx = hash_sessionid(sessionid);
998 /* Search in the appropriate list */
999 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
1000 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1001 NFS4_MAX_SESSIONID_LEN)) {
1002 return elem;
1003 }
1004 }
1005
1006 dprintk("%s: session not found\n", __func__);
1007 return NULL;
1008}
1009
1010/* caller must hold client_lock */
1011static void
1012unhash_session(struct nfsd4_session *ses)
1013{
1014 list_del(&ses->se_hash);
1015 spin_lock(&ses->se_client->cl_lock);
1016 list_del(&ses->se_perclnt);
1017 spin_unlock(&ses->se_client->cl_lock);
1018}
1019
1020/* must be called under the client_lock */
1021static inline void
1022renew_client_locked(struct nfs4_client *clp)
1023{
1024 if (is_client_expired(clp)) {
1025 WARN_ON(1);
1026 printk("%s: client (clientid %08x/%08x) already expired\n",
1027 __func__,
1028 clp->cl_clientid.cl_boot,
1029 clp->cl_clientid.cl_id);
1030 return;
1031 }
1032
1033 dprintk("renewing client (clientid %08x/%08x)\n",
1034 clp->cl_clientid.cl_boot,
1035 clp->cl_clientid.cl_id);
1036 list_move_tail(&clp->cl_lru, &client_lru);
1037 clp->cl_time = get_seconds();
1038}
1039
1040static inline void
1041renew_client(struct nfs4_client *clp)
1042{
1043 spin_lock(&client_lock);
1044 renew_client_locked(clp);
1045 spin_unlock(&client_lock);
1046}
1047
1048/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1049static int
1050STALE_CLIENTID(clientid_t *clid)
1051{
1052 if (clid->cl_boot == boot_time)
1053 return 0;
1054 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1055 clid->cl_boot, clid->cl_id, boot_time);
1056 return 1;
1057}
1058
1059/*
1060 * XXX Should we use a slab cache ?
1061 * This type of memory management is somewhat inefficient, but we use it
1062 * anyway since SETCLIENTID is not a common operation.
1063 */
1064static struct nfs4_client *alloc_client(struct xdr_netobj name)
1065{
1066 struct nfs4_client *clp;
1067
1068 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1069 if (clp == NULL)
1070 return NULL;
1071 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1072 if (clp->cl_name.data == NULL) {
1073 kfree(clp);
1074 return NULL;
1075 }
1076 clp->cl_name.len = name.len;
1077 return clp;
1078}
1079
1080static inline void
1081free_client(struct nfs4_client *clp)
1082{
1083 lockdep_assert_held(&client_lock);
1084 while (!list_empty(&clp->cl_sessions)) {
1085 struct nfsd4_session *ses;
1086 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1087 se_perclnt);
1088 list_del(&ses->se_perclnt);
1089 nfsd4_put_session_locked(ses);
1090 }
1091 free_svc_cred(&clp->cl_cred);
1092 kfree(clp->cl_name.data);
1093 kfree(clp);
1094}
1095
1096void
1097release_session_client(struct nfsd4_session *session)
1098{
1099 struct nfs4_client *clp = session->se_client;
1100
1101 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1102 return;
1103 if (is_client_expired(clp)) {
1104 free_client(clp);
1105 session->se_client = NULL;
1106 } else
1107 renew_client_locked(clp);
1108 spin_unlock(&client_lock);
1109}
1110
1111/* must be called under the client_lock */
1112static inline void
1113unhash_client_locked(struct nfs4_client *clp)
1114{
1115 struct nfsd4_session *ses;
1116
1117 mark_client_expired(clp);
1118 list_del(&clp->cl_lru);
1119 spin_lock(&clp->cl_lock);
1120 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1121 list_del_init(&ses->se_hash);
1122 spin_unlock(&clp->cl_lock);
1123}
1124
1125static void
1126expire_client(struct nfs4_client *clp)
1127{
1128 struct nfs4_openowner *oo;
1129 struct nfs4_delegation *dp;
1130 struct list_head reaplist;
1131
1132 INIT_LIST_HEAD(&reaplist);
1133 spin_lock(&recall_lock);
1134 while (!list_empty(&clp->cl_delegations)) {
1135 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1136 list_del_init(&dp->dl_perclnt);
1137 list_move(&dp->dl_recall_lru, &reaplist);
1138 }
1139 spin_unlock(&recall_lock);
1140 while (!list_empty(&reaplist)) {
1141 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1142 unhash_delegation(dp);
1143 }
1144 while (!list_empty(&clp->cl_openowners)) {
1145 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1146 release_openowner(oo);
1147 }
1148 nfsd4_shutdown_callback(clp);
1149 if (clp->cl_cb_conn.cb_xprt)
1150 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1151 list_del(&clp->cl_idhash);
1152 list_del(&clp->cl_strhash);
1153 spin_lock(&client_lock);
1154 unhash_client_locked(clp);
1155 if (atomic_read(&clp->cl_refcount) == 0)
1156 free_client(clp);
1157 spin_unlock(&client_lock);
1158}
1159
1160static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1161{
1162 memcpy(target->cl_verifier.data, source->data,
1163 sizeof(target->cl_verifier.data));
1164}
1165
1166static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1167{
1168 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1169 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1170}
1171
1172static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1173{
1174 if (source->cr_principal) {
1175 target->cr_principal =
1176 kstrdup(source->cr_principal, GFP_KERNEL);
1177 if (target->cr_principal == NULL)
1178 return -ENOMEM;
1179 } else
1180 target->cr_principal = NULL;
1181 target->cr_flavor = source->cr_flavor;
1182 target->cr_uid = source->cr_uid;
1183 target->cr_gid = source->cr_gid;
1184 target->cr_group_info = source->cr_group_info;
1185 get_group_info(target->cr_group_info);
1186 return 0;
1187}
1188
1189static int same_name(const char *n1, const char *n2)
1190{
1191 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1192}
1193
1194static int
1195same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1196{
1197 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1198}
1199
1200static int
1201same_clid(clientid_t *cl1, clientid_t *cl2)
1202{
1203 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1204}
1205
1206static bool groups_equal(struct group_info *g1, struct group_info *g2)
1207{
1208 int i;
1209
1210 if (g1->ngroups != g2->ngroups)
1211 return false;
1212 for (i=0; i<g1->ngroups; i++)
1213 if (GROUP_AT(g1, i) != GROUP_AT(g2, i))
1214 return false;
1215 return true;
1216}
1217
1218static bool
1219same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1220{
1221 if ((cr1->cr_flavor != cr2->cr_flavor)
1222 || (cr1->cr_uid != cr2->cr_uid)
1223 || (cr1->cr_gid != cr2->cr_gid)
1224 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1225 return false;
1226 if (cr1->cr_principal == cr2->cr_principal)
1227 return true;
1228 if (!cr1->cr_principal || !cr2->cr_principal)
1229 return false;
1230 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1231}
1232
1233static void gen_clid(struct nfs4_client *clp)
1234{
1235 static u32 current_clientid = 1;
1236
1237 clp->cl_clientid.cl_boot = boot_time;
1238 clp->cl_clientid.cl_id = current_clientid++;
1239}
1240
1241static void gen_confirm(struct nfs4_client *clp)
1242{
1243 __be32 verf[2];
1244 static u32 i;
1245
1246 verf[0] = (__be32)get_seconds();
1247 verf[1] = (__be32)i++;
1248 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1249}
1250
1251static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1252{
1253 return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1254}
1255
1256static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1257{
1258 struct nfs4_stid *s;
1259
1260 s = find_stateid(cl, t);
1261 if (!s)
1262 return NULL;
1263 if (typemask & s->sc_type)
1264 return s;
1265 return NULL;
1266}
1267
1268static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1269 struct svc_rqst *rqstp, nfs4_verifier *verf)
1270{
1271 struct nfs4_client *clp;
1272 struct sockaddr *sa = svc_addr(rqstp);
1273 int ret;
1274
1275 clp = alloc_client(name);
1276 if (clp == NULL)
1277 return NULL;
1278
1279 INIT_LIST_HEAD(&clp->cl_sessions);
1280 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1281 if (ret) {
1282 spin_lock(&client_lock);
1283 free_client(clp);
1284 spin_unlock(&client_lock);
1285 return NULL;
1286 }
1287 idr_init(&clp->cl_stateids);
1288 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1289 atomic_set(&clp->cl_refcount, 0);
1290 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1291 INIT_LIST_HEAD(&clp->cl_idhash);
1292 INIT_LIST_HEAD(&clp->cl_strhash);
1293 INIT_LIST_HEAD(&clp->cl_openowners);
1294 INIT_LIST_HEAD(&clp->cl_delegations);
1295 INIT_LIST_HEAD(&clp->cl_lru);
1296 INIT_LIST_HEAD(&clp->cl_callbacks);
1297 spin_lock_init(&clp->cl_lock);
1298 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1299 clp->cl_time = get_seconds();
1300 clear_bit(0, &clp->cl_cb_slot_busy);
1301 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1302 copy_verf(clp, verf);
1303 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1304 gen_confirm(clp);
1305 clp->cl_cb_session = NULL;
1306 return clp;
1307}
1308
1309static void
1310add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1311{
1312 unsigned int idhashval;
1313
1314 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1315 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1316 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1317 renew_client(clp);
1318}
1319
1320static void
1321move_to_confirmed(struct nfs4_client *clp)
1322{
1323 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1324 unsigned int strhashval;
1325
1326 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1327 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1328 strhashval = clientstr_hashval(clp->cl_recdir);
1329 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1330 renew_client(clp);
1331}
1332
1333static struct nfs4_client *
1334find_confirmed_client(clientid_t *clid)
1335{
1336 struct nfs4_client *clp;
1337 unsigned int idhashval = clientid_hashval(clid->cl_id);
1338
1339 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1340 if (same_clid(&clp->cl_clientid, clid)) {
1341 renew_client(clp);
1342 return clp;
1343 }
1344 }
1345 return NULL;
1346}
1347
1348static struct nfs4_client *
1349find_unconfirmed_client(clientid_t *clid)
1350{
1351 struct nfs4_client *clp;
1352 unsigned int idhashval = clientid_hashval(clid->cl_id);
1353
1354 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1355 if (same_clid(&clp->cl_clientid, clid))
1356 return clp;
1357 }
1358 return NULL;
1359}
1360
1361static bool clp_used_exchangeid(struct nfs4_client *clp)
1362{
1363 return clp->cl_exchange_flags != 0;
1364}
1365
1366static struct nfs4_client *
1367find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1368{
1369 struct nfs4_client *clp;
1370
1371 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1372 if (same_name(clp->cl_recdir, dname))
1373 return clp;
1374 }
1375 return NULL;
1376}
1377
1378static struct nfs4_client *
1379find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1380{
1381 struct nfs4_client *clp;
1382
1383 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1384 if (same_name(clp->cl_recdir, dname))
1385 return clp;
1386 }
1387 return NULL;
1388}
1389
1390static void
1391gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1392{
1393 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1394 struct sockaddr *sa = svc_addr(rqstp);
1395 u32 scopeid = rpc_get_scope_id(sa);
1396 unsigned short expected_family;
1397
1398 /* Currently, we only support tcp and tcp6 for the callback channel */
1399 if (se->se_callback_netid_len == 3 &&
1400 !memcmp(se->se_callback_netid_val, "tcp", 3))
1401 expected_family = AF_INET;
1402 else if (se->se_callback_netid_len == 4 &&
1403 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1404 expected_family = AF_INET6;
1405 else
1406 goto out_err;
1407
1408 conn->cb_addrlen = rpc_uaddr2sockaddr(&init_net, se->se_callback_addr_val,
1409 se->se_callback_addr_len,
1410 (struct sockaddr *)&conn->cb_addr,
1411 sizeof(conn->cb_addr));
1412
1413 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1414 goto out_err;
1415
1416 if (conn->cb_addr.ss_family == AF_INET6)
1417 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1418
1419 conn->cb_prog = se->se_callback_prog;
1420 conn->cb_ident = se->se_callback_ident;
1421 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1422 return;
1423out_err:
1424 conn->cb_addr.ss_family = AF_UNSPEC;
1425 conn->cb_addrlen = 0;
1426 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1427 "will not receive delegations\n",
1428 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1429
1430 return;
1431}
1432
1433/*
1434 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1435 */
1436void
1437nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1438{
1439 struct nfsd4_slot *slot = resp->cstate.slot;
1440 unsigned int base;
1441
1442 dprintk("--> %s slot %p\n", __func__, slot);
1443
1444 slot->sl_opcnt = resp->opcnt;
1445 slot->sl_status = resp->cstate.status;
1446
1447 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1448 if (nfsd4_not_cached(resp)) {
1449 slot->sl_datalen = 0;
1450 return;
1451 }
1452 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1453 base = (char *)resp->cstate.datap -
1454 (char *)resp->xbuf->head[0].iov_base;
1455 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1456 slot->sl_datalen))
1457 WARN("%s: sessions DRC could not cache compound\n", __func__);
1458 return;
1459}
1460
1461/*
1462 * Encode the replay sequence operation from the slot values.
1463 * If cachethis is FALSE encode the uncached rep error on the next
1464 * operation which sets resp->p and increments resp->opcnt for
1465 * nfs4svc_encode_compoundres.
1466 *
1467 */
1468static __be32
1469nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1470 struct nfsd4_compoundres *resp)
1471{
1472 struct nfsd4_op *op;
1473 struct nfsd4_slot *slot = resp->cstate.slot;
1474
1475 /* Encode the replayed sequence operation */
1476 op = &args->ops[resp->opcnt - 1];
1477 nfsd4_encode_operation(resp, op);
1478
1479 /* Return nfserr_retry_uncached_rep in next operation. */
1480 if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1481 op = &args->ops[resp->opcnt++];
1482 op->status = nfserr_retry_uncached_rep;
1483 nfsd4_encode_operation(resp, op);
1484 }
1485 return op->status;
1486}
1487
1488/*
1489 * The sequence operation is not cached because we can use the slot and
1490 * session values.
1491 */
1492__be32
1493nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1494 struct nfsd4_sequence *seq)
1495{
1496 struct nfsd4_slot *slot = resp->cstate.slot;
1497 __be32 status;
1498
1499 dprintk("--> %s slot %p\n", __func__, slot);
1500
1501 /* Either returns 0 or nfserr_retry_uncached */
1502 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1503 if (status == nfserr_retry_uncached_rep)
1504 return status;
1505
1506 /* The sequence operation has been encoded, cstate->datap set. */
1507 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1508
1509 resp->opcnt = slot->sl_opcnt;
1510 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1511 status = slot->sl_status;
1512
1513 return status;
1514}
1515
1516/*
1517 * Set the exchange_id flags returned by the server.
1518 */
1519static void
1520nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1521{
1522 /* pNFS is not supported */
1523 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1524
1525 /* Referrals are supported, Migration is not. */
1526 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1527
1528 /* set the wire flags to return to client. */
1529 clid->flags = new->cl_exchange_flags;
1530}
1531
1532static bool client_has_state(struct nfs4_client *clp)
1533{
1534 /*
1535 * Note clp->cl_openowners check isn't quite right: there's no
1536 * need to count owners without stateid's.
1537 *
1538 * Also note we should probably be using this in 4.0 case too.
1539 */
1540 return !list_empty(&clp->cl_openowners)
1541 || !list_empty(&clp->cl_delegations)
1542 || !list_empty(&clp->cl_sessions);
1543}
1544
1545__be32
1546nfsd4_exchange_id(struct svc_rqst *rqstp,
1547 struct nfsd4_compound_state *cstate,
1548 struct nfsd4_exchange_id *exid)
1549{
1550 struct nfs4_client *unconf, *conf, *new;
1551 __be32 status;
1552 unsigned int strhashval;
1553 char dname[HEXDIR_LEN];
1554 char addr_str[INET6_ADDRSTRLEN];
1555 nfs4_verifier verf = exid->verifier;
1556 struct sockaddr *sa = svc_addr(rqstp);
1557 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1558
1559 rpc_ntop(sa, addr_str, sizeof(addr_str));
1560 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1561 "ip_addr=%s flags %x, spa_how %d\n",
1562 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1563 addr_str, exid->flags, exid->spa_how);
1564
1565 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1566 return nfserr_inval;
1567
1568 /* Currently only support SP4_NONE */
1569 switch (exid->spa_how) {
1570 case SP4_NONE:
1571 break;
1572 case SP4_SSV:
1573 return nfserr_serverfault;
1574 default:
1575 BUG(); /* checked by xdr code */
1576 case SP4_MACH_CRED:
1577 return nfserr_serverfault; /* no excuse :-/ */
1578 }
1579
1580 status = nfs4_make_rec_clidname(dname, &exid->clname);
1581
1582 if (status)
1583 return status;
1584
1585 strhashval = clientstr_hashval(dname);
1586
1587 /* Cases below refer to rfc 5661 section 18.35.4: */
1588 nfs4_lock_state();
1589 conf = find_confirmed_client_by_str(dname, strhashval);
1590 if (conf) {
1591 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1592 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1593
1594 if (update) {
1595 if (!clp_used_exchangeid(conf)) { /* buggy client */
1596 status = nfserr_inval;
1597 goto out;
1598 }
1599 if (!creds_match) { /* case 9 */
1600 status = nfserr_perm;
1601 goto out;
1602 }
1603 if (!verfs_match) { /* case 8 */
1604 status = nfserr_not_same;
1605 goto out;
1606 }
1607 /* case 6 */
1608 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1609 new = conf;
1610 goto out_copy;
1611 }
1612 if (!creds_match) { /* case 3 */
1613 if (client_has_state(conf)) {
1614 status = nfserr_clid_inuse;
1615 goto out;
1616 }
1617 expire_client(conf);
1618 goto out_new;
1619 }
1620 if (verfs_match) { /* case 2 */
1621 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1622 new = conf;
1623 goto out_copy;
1624 }
1625 /* case 5, client reboot */
1626 goto out_new;
1627 }
1628
1629 if (update) { /* case 7 */
1630 status = nfserr_noent;
1631 goto out;
1632 }
1633
1634 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1635 if (unconf) /* case 4, possible retry or client restart */
1636 expire_client(unconf);
1637
1638 /* case 1 (normal case) */
1639out_new:
1640 new = create_client(exid->clname, dname, rqstp, &verf);
1641 if (new == NULL) {
1642 status = nfserr_jukebox;
1643 goto out;
1644 }
1645
1646 gen_clid(new);
1647 add_to_unconfirmed(new, strhashval);
1648out_copy:
1649 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1650 exid->clientid.cl_id = new->cl_clientid.cl_id;
1651
1652 exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1653 nfsd4_set_ex_flags(new, exid);
1654
1655 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1656 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1657 status = nfs_ok;
1658
1659out:
1660 nfs4_unlock_state();
1661 return status;
1662}
1663
1664static __be32
1665check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1666{
1667 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1668 slot_seqid);
1669
1670 /* The slot is in use, and no response has been sent. */
1671 if (slot_inuse) {
1672 if (seqid == slot_seqid)
1673 return nfserr_jukebox;
1674 else
1675 return nfserr_seq_misordered;
1676 }
1677 /* Note unsigned 32-bit arithmetic handles wraparound: */
1678 if (likely(seqid == slot_seqid + 1))
1679 return nfs_ok;
1680 if (seqid == slot_seqid)
1681 return nfserr_replay_cache;
1682 return nfserr_seq_misordered;
1683}
1684
1685/*
1686 * Cache the create session result into the create session single DRC
1687 * slot cache by saving the xdr structure. sl_seqid has been set.
1688 * Do this for solo or embedded create session operations.
1689 */
1690static void
1691nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1692 struct nfsd4_clid_slot *slot, __be32 nfserr)
1693{
1694 slot->sl_status = nfserr;
1695 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1696}
1697
1698static __be32
1699nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1700 struct nfsd4_clid_slot *slot)
1701{
1702 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1703 return slot->sl_status;
1704}
1705
1706#define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1707 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1708 1 + /* MIN tag is length with zero, only length */ \
1709 3 + /* version, opcount, opcode */ \
1710 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1711 /* seqid, slotID, slotID, cache */ \
1712 4 ) * sizeof(__be32))
1713
1714#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1715 2 + /* verifier: AUTH_NULL, length 0 */\
1716 1 + /* status */ \
1717 1 + /* MIN tag is length with zero, only length */ \
1718 3 + /* opcount, opcode, opstatus*/ \
1719 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1720 /* seqid, slotID, slotID, slotID, status */ \
1721 5 ) * sizeof(__be32))
1722
1723static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1724{
1725 return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1726 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1727}
1728
1729__be32
1730nfsd4_create_session(struct svc_rqst *rqstp,
1731 struct nfsd4_compound_state *cstate,
1732 struct nfsd4_create_session *cr_ses)
1733{
1734 struct sockaddr *sa = svc_addr(rqstp);
1735 struct nfs4_client *conf, *unconf;
1736 struct nfsd4_session *new;
1737 struct nfsd4_clid_slot *cs_slot = NULL;
1738 bool confirm_me = false;
1739 __be32 status = 0;
1740
1741 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1742 return nfserr_inval;
1743
1744 nfs4_lock_state();
1745 unconf = find_unconfirmed_client(&cr_ses->clientid);
1746 conf = find_confirmed_client(&cr_ses->clientid);
1747
1748 if (conf) {
1749 cs_slot = &conf->cl_cs_slot;
1750 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1751 if (status == nfserr_replay_cache) {
1752 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1753 goto out;
1754 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1755 status = nfserr_seq_misordered;
1756 goto out;
1757 }
1758 } else if (unconf) {
1759 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1760 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1761 status = nfserr_clid_inuse;
1762 goto out;
1763 }
1764 cs_slot = &unconf->cl_cs_slot;
1765 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1766 if (status) {
1767 /* an unconfirmed replay returns misordered */
1768 status = nfserr_seq_misordered;
1769 goto out;
1770 }
1771 confirm_me = true;
1772 conf = unconf;
1773 } else {
1774 status = nfserr_stale_clientid;
1775 goto out;
1776 }
1777
1778 /*
1779 * XXX: we should probably set this at creation time, and check
1780 * for consistent minorversion use throughout:
1781 */
1782 conf->cl_minorversion = 1;
1783 /*
1784 * We do not support RDMA or persistent sessions
1785 */
1786 cr_ses->flags &= ~SESSION4_PERSIST;
1787 cr_ses->flags &= ~SESSION4_RDMA;
1788
1789 status = nfserr_toosmall;
1790 if (check_forechannel_attrs(cr_ses->fore_channel))
1791 goto out;
1792
1793 status = nfserr_jukebox;
1794 new = alloc_init_session(rqstp, conf, cr_ses);
1795 if (!new)
1796 goto out;
1797 status = nfs_ok;
1798 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1799 NFS4_MAX_SESSIONID_LEN);
1800 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1801 sizeof(struct nfsd4_channel_attrs));
1802 cs_slot->sl_seqid++;
1803 cr_ses->seqid = cs_slot->sl_seqid;
1804
1805 /* cache solo and embedded create sessions under the state lock */
1806 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1807 if (confirm_me) {
1808 unsigned int hash = clientstr_hashval(unconf->cl_recdir);
1809 struct nfs4_client *old =
1810 find_confirmed_client_by_str(conf->cl_recdir, hash);
1811 if (old)
1812 expire_client(old);
1813 move_to_confirmed(conf);
1814 }
1815out:
1816 nfs4_unlock_state();
1817 dprintk("%s returns %d\n", __func__, ntohl(status));
1818 return status;
1819}
1820
1821static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1822{
1823 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1824 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1825
1826 return argp->opcnt == resp->opcnt;
1827}
1828
1829static __be32 nfsd4_map_bcts_dir(u32 *dir)
1830{
1831 switch (*dir) {
1832 case NFS4_CDFC4_FORE:
1833 case NFS4_CDFC4_BACK:
1834 return nfs_ok;
1835 case NFS4_CDFC4_FORE_OR_BOTH:
1836 case NFS4_CDFC4_BACK_OR_BOTH:
1837 *dir = NFS4_CDFC4_BOTH;
1838 return nfs_ok;
1839 };
1840 return nfserr_inval;
1841}
1842
1843__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1844 struct nfsd4_compound_state *cstate,
1845 struct nfsd4_bind_conn_to_session *bcts)
1846{
1847 __be32 status;
1848
1849 if (!nfsd4_last_compound_op(rqstp))
1850 return nfserr_not_only_op;
1851 spin_lock(&client_lock);
1852 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1853 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1854 * client_lock iself: */
1855 if (cstate->session) {
1856 nfsd4_get_session(cstate->session);
1857 atomic_inc(&cstate->session->se_client->cl_refcount);
1858 }
1859 spin_unlock(&client_lock);
1860 if (!cstate->session)
1861 return nfserr_badsession;
1862
1863 status = nfsd4_map_bcts_dir(&bcts->dir);
1864 if (!status)
1865 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1866 return status;
1867}
1868
1869static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1870{
1871 if (!session)
1872 return 0;
1873 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1874}
1875
1876__be32
1877nfsd4_destroy_session(struct svc_rqst *r,
1878 struct nfsd4_compound_state *cstate,
1879 struct nfsd4_destroy_session *sessionid)
1880{
1881 struct nfsd4_session *ses;
1882 __be32 status = nfserr_badsession;
1883
1884 /* Notes:
1885 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1886 * - Should we return nfserr_back_chan_busy if waiting for
1887 * callbacks on to-be-destroyed session?
1888 * - Do we need to clear any callback info from previous session?
1889 */
1890
1891 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1892 if (!nfsd4_last_compound_op(r))
1893 return nfserr_not_only_op;
1894 }
1895 dump_sessionid(__func__, &sessionid->sessionid);
1896 spin_lock(&client_lock);
1897 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1898 if (!ses) {
1899 spin_unlock(&client_lock);
1900 goto out;
1901 }
1902
1903 unhash_session(ses);
1904 spin_unlock(&client_lock);
1905
1906 nfs4_lock_state();
1907 nfsd4_probe_callback_sync(ses->se_client);
1908 nfs4_unlock_state();
1909
1910 spin_lock(&client_lock);
1911 nfsd4_del_conns(ses);
1912 nfsd4_put_session_locked(ses);
1913 spin_unlock(&client_lock);
1914 status = nfs_ok;
1915out:
1916 dprintk("%s returns %d\n", __func__, ntohl(status));
1917 return status;
1918}
1919
1920static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1921{
1922 struct nfsd4_conn *c;
1923
1924 list_for_each_entry(c, &s->se_conns, cn_persession) {
1925 if (c->cn_xprt == xpt) {
1926 return c;
1927 }
1928 }
1929 return NULL;
1930}
1931
1932static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1933{
1934 struct nfs4_client *clp = ses->se_client;
1935 struct nfsd4_conn *c;
1936 int ret;
1937
1938 spin_lock(&clp->cl_lock);
1939 c = __nfsd4_find_conn(new->cn_xprt, ses);
1940 if (c) {
1941 spin_unlock(&clp->cl_lock);
1942 free_conn(new);
1943 return;
1944 }
1945 __nfsd4_hash_conn(new, ses);
1946 spin_unlock(&clp->cl_lock);
1947 ret = nfsd4_register_conn(new);
1948 if (ret)
1949 /* oops; xprt is already down: */
1950 nfsd4_conn_lost(&new->cn_xpt_user);
1951 return;
1952}
1953
1954static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1955{
1956 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1957
1958 return args->opcnt > session->se_fchannel.maxops;
1959}
1960
1961static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1962 struct nfsd4_session *session)
1963{
1964 struct xdr_buf *xb = &rqstp->rq_arg;
1965
1966 return xb->len > session->se_fchannel.maxreq_sz;
1967}
1968
1969__be32
1970nfsd4_sequence(struct svc_rqst *rqstp,
1971 struct nfsd4_compound_state *cstate,
1972 struct nfsd4_sequence *seq)
1973{
1974 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1975 struct nfsd4_session *session;
1976 struct nfsd4_slot *slot;
1977 struct nfsd4_conn *conn;
1978 __be32 status;
1979
1980 if (resp->opcnt != 1)
1981 return nfserr_sequence_pos;
1982
1983 /*
1984 * Will be either used or freed by nfsd4_sequence_check_conn
1985 * below.
1986 */
1987 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1988 if (!conn)
1989 return nfserr_jukebox;
1990
1991 spin_lock(&client_lock);
1992 status = nfserr_badsession;
1993 session = find_in_sessionid_hashtbl(&seq->sessionid);
1994 if (!session)
1995 goto out;
1996
1997 status = nfserr_too_many_ops;
1998 if (nfsd4_session_too_many_ops(rqstp, session))
1999 goto out;
2000
2001 status = nfserr_req_too_big;
2002 if (nfsd4_request_too_big(rqstp, session))
2003 goto out;
2004
2005 status = nfserr_badslot;
2006 if (seq->slotid >= session->se_fchannel.maxreqs)
2007 goto out;
2008
2009 slot = session->se_slots[seq->slotid];
2010 dprintk("%s: slotid %d\n", __func__, seq->slotid);
2011
2012 /* We do not negotiate the number of slots yet, so set the
2013 * maxslots to the session maxreqs which is used to encode
2014 * sr_highest_slotid and the sr_target_slot id to maxslots */
2015 seq->maxslots = session->se_fchannel.maxreqs;
2016
2017 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2018 slot->sl_flags & NFSD4_SLOT_INUSE);
2019 if (status == nfserr_replay_cache) {
2020 status = nfserr_seq_misordered;
2021 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2022 goto out;
2023 cstate->slot = slot;
2024 cstate->session = session;
2025 /* Return the cached reply status and set cstate->status
2026 * for nfsd4_proc_compound processing */
2027 status = nfsd4_replay_cache_entry(resp, seq);
2028 cstate->status = nfserr_replay_cache;
2029 goto out;
2030 }
2031 if (status)
2032 goto out;
2033
2034 nfsd4_sequence_check_conn(conn, session);
2035 conn = NULL;
2036
2037 /* Success! bump slot seqid */
2038 slot->sl_seqid = seq->seqid;
2039 slot->sl_flags |= NFSD4_SLOT_INUSE;
2040 if (seq->cachethis)
2041 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2042 else
2043 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2044
2045 cstate->slot = slot;
2046 cstate->session = session;
2047
2048out:
2049 /* Hold a session reference until done processing the compound. */
2050 if (cstate->session) {
2051 struct nfs4_client *clp = session->se_client;
2052
2053 nfsd4_get_session(cstate->session);
2054 atomic_inc(&clp->cl_refcount);
2055 switch (clp->cl_cb_state) {
2056 case NFSD4_CB_DOWN:
2057 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2058 break;
2059 case NFSD4_CB_FAULT:
2060 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2061 break;
2062 default:
2063 seq->status_flags = 0;
2064 }
2065 }
2066 kfree(conn);
2067 spin_unlock(&client_lock);
2068 dprintk("%s: return %d\n", __func__, ntohl(status));
2069 return status;
2070}
2071
2072__be32
2073nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2074{
2075 struct nfs4_client *conf, *unconf, *clp;
2076 __be32 status = 0;
2077
2078 nfs4_lock_state();
2079 unconf = find_unconfirmed_client(&dc->clientid);
2080 conf = find_confirmed_client(&dc->clientid);
2081
2082 if (conf) {
2083 clp = conf;
2084
2085 if (!is_client_expired(conf) && client_has_state(conf)) {
2086 status = nfserr_clientid_busy;
2087 goto out;
2088 }
2089
2090 /* rfc5661 18.50.3 */
2091 if (cstate->session && conf == cstate->session->se_client) {
2092 status = nfserr_clientid_busy;
2093 goto out;
2094 }
2095 } else if (unconf)
2096 clp = unconf;
2097 else {
2098 status = nfserr_stale_clientid;
2099 goto out;
2100 }
2101
2102 expire_client(clp);
2103out:
2104 nfs4_unlock_state();
2105 dprintk("%s return %d\n", __func__, ntohl(status));
2106 return status;
2107}
2108
2109__be32
2110nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2111{
2112 __be32 status = 0;
2113
2114 if (rc->rca_one_fs) {
2115 if (!cstate->current_fh.fh_dentry)
2116 return nfserr_nofilehandle;
2117 /*
2118 * We don't take advantage of the rca_one_fs case.
2119 * That's OK, it's optional, we can safely ignore it.
2120 */
2121 return nfs_ok;
2122 }
2123
2124 nfs4_lock_state();
2125 status = nfserr_complete_already;
2126 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2127 &cstate->session->se_client->cl_flags))
2128 goto out;
2129
2130 status = nfserr_stale_clientid;
2131 if (is_client_expired(cstate->session->se_client))
2132 /*
2133 * The following error isn't really legal.
2134 * But we only get here if the client just explicitly
2135 * destroyed the client. Surely it no longer cares what
2136 * error it gets back on an operation for the dead
2137 * client.
2138 */
2139 goto out;
2140
2141 status = nfs_ok;
2142 nfsd4_client_record_create(cstate->session->se_client);
2143out:
2144 nfs4_unlock_state();
2145 return status;
2146}
2147
2148__be32
2149nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2150 struct nfsd4_setclientid *setclid)
2151{
2152 struct xdr_netobj clname = setclid->se_name;
2153 nfs4_verifier clverifier = setclid->se_verf;
2154 unsigned int strhashval;
2155 struct nfs4_client *conf, *unconf, *new;
2156 __be32 status;
2157 char dname[HEXDIR_LEN];
2158
2159 status = nfs4_make_rec_clidname(dname, &clname);
2160 if (status)
2161 return status;
2162
2163 strhashval = clientstr_hashval(dname);
2164
2165 /* Cases below refer to rfc 3530 section 14.2.33: */
2166 nfs4_lock_state();
2167 conf = find_confirmed_client_by_str(dname, strhashval);
2168 if (conf) {
2169 /* case 0: */
2170 status = nfserr_clid_inuse;
2171 if (clp_used_exchangeid(conf))
2172 goto out;
2173 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2174 char addr_str[INET6_ADDRSTRLEN];
2175 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2176 sizeof(addr_str));
2177 dprintk("NFSD: setclientid: string in use by client "
2178 "at %s\n", addr_str);
2179 goto out;
2180 }
2181 }
2182 unconf = find_unconfirmed_client_by_str(dname, strhashval);
2183 if (unconf)
2184 expire_client(unconf);
2185 status = nfserr_jukebox;
2186 new = create_client(clname, dname, rqstp, &clverifier);
2187 if (new == NULL)
2188 goto out;
2189 if (conf && same_verf(&conf->cl_verifier, &clverifier))
2190 /* case 1: probable callback update */
2191 copy_clid(new, conf);
2192 else /* case 4 (new client) or cases 2, 3 (client reboot): */
2193 gen_clid(new);
2194 /*
2195 * XXX: we should probably set this at creation time, and check
2196 * for consistent minorversion use throughout:
2197 */
2198 new->cl_minorversion = 0;
2199 gen_callback(new, setclid, rqstp);
2200 add_to_unconfirmed(new, strhashval);
2201 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2202 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2203 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2204 status = nfs_ok;
2205out:
2206 nfs4_unlock_state();
2207 return status;
2208}
2209
2210
2211__be32
2212nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2213 struct nfsd4_compound_state *cstate,
2214 struct nfsd4_setclientid_confirm *setclientid_confirm)
2215{
2216 struct nfs4_client *conf, *unconf;
2217 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2218 clientid_t * clid = &setclientid_confirm->sc_clientid;
2219 __be32 status;
2220
2221 if (STALE_CLIENTID(clid))
2222 return nfserr_stale_clientid;
2223 nfs4_lock_state();
2224
2225 conf = find_confirmed_client(clid);
2226 unconf = find_unconfirmed_client(clid);
2227 /*
2228 * We try hard to give out unique clientid's, so if we get an
2229 * attempt to confirm the same clientid with a different cred,
2230 * there's a bug somewhere. Let's charitably assume it's our
2231 * bug.
2232 */
2233 status = nfserr_serverfault;
2234 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2235 goto out;
2236 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2237 goto out;
2238 /* cases below refer to rfc 3530 section 14.2.34: */
2239 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2240 if (conf && !unconf) /* case 2: probable retransmit */
2241 status = nfs_ok;
2242 else /* case 4: client hasn't noticed we rebooted yet? */
2243 status = nfserr_stale_clientid;
2244 goto out;
2245 }
2246 status = nfs_ok;
2247 if (conf) { /* case 1: callback update */
2248 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2249 nfsd4_probe_callback(conf);
2250 expire_client(unconf);
2251 } else { /* case 3: normal case; new or rebooted client */
2252 unsigned int hash = clientstr_hashval(unconf->cl_recdir);
2253
2254 conf = find_confirmed_client_by_str(unconf->cl_recdir, hash);
2255 if (conf) {
2256 nfsd4_client_record_remove(conf);
2257 expire_client(conf);
2258 }
2259 move_to_confirmed(unconf);
2260 nfsd4_probe_callback(unconf);
2261 }
2262out:
2263 nfs4_unlock_state();
2264 return status;
2265}
2266
2267static struct nfs4_file *nfsd4_alloc_file(void)
2268{
2269 return kmem_cache_alloc(file_slab, GFP_KERNEL);
2270}
2271
2272/* OPEN Share state helper functions */
2273static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2274{
2275 unsigned int hashval = file_hashval(ino);
2276
2277 atomic_set(&fp->fi_ref, 1);
2278 INIT_LIST_HEAD(&fp->fi_hash);
2279 INIT_LIST_HEAD(&fp->fi_stateids);
2280 INIT_LIST_HEAD(&fp->fi_delegations);
2281 fp->fi_inode = igrab(ino);
2282 fp->fi_had_conflict = false;
2283 fp->fi_lease = NULL;
2284 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2285 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2286 spin_lock(&recall_lock);
2287 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2288 spin_unlock(&recall_lock);
2289}
2290
2291static void
2292nfsd4_free_slab(struct kmem_cache **slab)
2293{
2294 if (*slab == NULL)
2295 return;
2296 kmem_cache_destroy(*slab);
2297 *slab = NULL;
2298}
2299
2300void
2301nfsd4_free_slabs(void)
2302{
2303 nfsd4_free_slab(&openowner_slab);
2304 nfsd4_free_slab(&lockowner_slab);
2305 nfsd4_free_slab(&file_slab);
2306 nfsd4_free_slab(&stateid_slab);
2307 nfsd4_free_slab(&deleg_slab);
2308}
2309
2310int
2311nfsd4_init_slabs(void)
2312{
2313 openowner_slab = kmem_cache_create("nfsd4_openowners",
2314 sizeof(struct nfs4_openowner), 0, 0, NULL);
2315 if (openowner_slab == NULL)
2316 goto out_nomem;
2317 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2318 sizeof(struct nfs4_openowner), 0, 0, NULL);
2319 if (lockowner_slab == NULL)
2320 goto out_nomem;
2321 file_slab = kmem_cache_create("nfsd4_files",
2322 sizeof(struct nfs4_file), 0, 0, NULL);
2323 if (file_slab == NULL)
2324 goto out_nomem;
2325 stateid_slab = kmem_cache_create("nfsd4_stateids",
2326 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2327 if (stateid_slab == NULL)
2328 goto out_nomem;
2329 deleg_slab = kmem_cache_create("nfsd4_delegations",
2330 sizeof(struct nfs4_delegation), 0, 0, NULL);
2331 if (deleg_slab == NULL)
2332 goto out_nomem;
2333 return 0;
2334out_nomem:
2335 nfsd4_free_slabs();
2336 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2337 return -ENOMEM;
2338}
2339
2340void nfs4_free_openowner(struct nfs4_openowner *oo)
2341{
2342 kfree(oo->oo_owner.so_owner.data);
2343 kmem_cache_free(openowner_slab, oo);
2344}
2345
2346void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2347{
2348 kfree(lo->lo_owner.so_owner.data);
2349 kmem_cache_free(lockowner_slab, lo);
2350}
2351
2352static void init_nfs4_replay(struct nfs4_replay *rp)
2353{
2354 rp->rp_status = nfserr_serverfault;
2355 rp->rp_buflen = 0;
2356 rp->rp_buf = rp->rp_ibuf;
2357}
2358
2359static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2360{
2361 struct nfs4_stateowner *sop;
2362
2363 sop = kmem_cache_alloc(slab, GFP_KERNEL);
2364 if (!sop)
2365 return NULL;
2366
2367 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2368 if (!sop->so_owner.data) {
2369 kmem_cache_free(slab, sop);
2370 return NULL;
2371 }
2372 sop->so_owner.len = owner->len;
2373
2374 INIT_LIST_HEAD(&sop->so_stateids);
2375 sop->so_client = clp;
2376 init_nfs4_replay(&sop->so_replay);
2377 return sop;
2378}
2379
2380static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2381{
2382 list_add(&oo->oo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
2383 list_add(&oo->oo_perclient, &clp->cl_openowners);
2384}
2385
2386static struct nfs4_openowner *
2387alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2388 struct nfs4_openowner *oo;
2389
2390 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2391 if (!oo)
2392 return NULL;
2393 oo->oo_owner.so_is_open_owner = 1;
2394 oo->oo_owner.so_seqid = open->op_seqid;
2395 oo->oo_flags = NFS4_OO_NEW;
2396 oo->oo_time = 0;
2397 oo->oo_last_closed_stid = NULL;
2398 INIT_LIST_HEAD(&oo->oo_close_lru);
2399 hash_openowner(oo, clp, strhashval);
2400 return oo;
2401}
2402
2403static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2404 struct nfs4_openowner *oo = open->op_openowner;
2405 struct nfs4_client *clp = oo->oo_owner.so_client;
2406
2407 init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2408 INIT_LIST_HEAD(&stp->st_lockowners);
2409 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2410 list_add(&stp->st_perfile, &fp->fi_stateids);
2411 stp->st_stateowner = &oo->oo_owner;
2412 get_nfs4_file(fp);
2413 stp->st_file = fp;
2414 stp->st_access_bmap = 0;
2415 stp->st_deny_bmap = 0;
2416 set_access(open->op_share_access, stp);
2417 set_deny(open->op_share_deny, stp);
2418 stp->st_openstp = NULL;
2419}
2420
2421static void
2422move_to_close_lru(struct nfs4_openowner *oo)
2423{
2424 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2425
2426 list_move_tail(&oo->oo_close_lru, &close_lru);
2427 oo->oo_time = get_seconds();
2428}
2429
2430static int
2431same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2432 clientid_t *clid)
2433{
2434 return (sop->so_owner.len == owner->len) &&
2435 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2436 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2437}
2438
2439static struct nfs4_openowner *
2440find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2441{
2442 struct nfs4_stateowner *so;
2443 struct nfs4_openowner *oo;
2444
2445 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2446 if (!so->so_is_open_owner)
2447 continue;
2448 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2449 oo = openowner(so);
2450 renew_client(oo->oo_owner.so_client);
2451 return oo;
2452 }
2453 }
2454 return NULL;
2455}
2456
2457/* search file_hashtbl[] for file */
2458static struct nfs4_file *
2459find_file(struct inode *ino)
2460{
2461 unsigned int hashval = file_hashval(ino);
2462 struct nfs4_file *fp;
2463
2464 spin_lock(&recall_lock);
2465 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2466 if (fp->fi_inode == ino) {
2467 get_nfs4_file(fp);
2468 spin_unlock(&recall_lock);
2469 return fp;
2470 }
2471 }
2472 spin_unlock(&recall_lock);
2473 return NULL;
2474}
2475
2476/*
2477 * Called to check deny when READ with all zero stateid or
2478 * WRITE with all zero or all one stateid
2479 */
2480static __be32
2481nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2482{
2483 struct inode *ino = current_fh->fh_dentry->d_inode;
2484 struct nfs4_file *fp;
2485 struct nfs4_ol_stateid *stp;
2486 __be32 ret;
2487
2488 dprintk("NFSD: nfs4_share_conflict\n");
2489
2490 fp = find_file(ino);
2491 if (!fp)
2492 return nfs_ok;
2493 ret = nfserr_locked;
2494 /* Search for conflicting share reservations */
2495 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2496 if (test_deny(deny_type, stp) ||
2497 test_deny(NFS4_SHARE_DENY_BOTH, stp))
2498 goto out;
2499 }
2500 ret = nfs_ok;
2501out:
2502 put_nfs4_file(fp);
2503 return ret;
2504}
2505
2506static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2507{
2508 /* We're assuming the state code never drops its reference
2509 * without first removing the lease. Since we're in this lease
2510 * callback (and since the lease code is serialized by the kernel
2511 * lock) we know the server hasn't removed the lease yet, we know
2512 * it's safe to take a reference: */
2513 atomic_inc(&dp->dl_count);
2514
2515 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2516
2517 /* only place dl_time is set. protected by lock_flocks*/
2518 dp->dl_time = get_seconds();
2519
2520 nfsd4_cb_recall(dp);
2521}
2522
2523/* Called from break_lease() with lock_flocks() held. */
2524static void nfsd_break_deleg_cb(struct file_lock *fl)
2525{
2526 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2527 struct nfs4_delegation *dp;
2528
2529 BUG_ON(!fp);
2530 /* We assume break_lease is only called once per lease: */
2531 BUG_ON(fp->fi_had_conflict);
2532 /*
2533 * We don't want the locks code to timeout the lease for us;
2534 * we'll remove it ourself if a delegation isn't returned
2535 * in time:
2536 */
2537 fl->fl_break_time = 0;
2538
2539 spin_lock(&recall_lock);
2540 fp->fi_had_conflict = true;
2541 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2542 nfsd_break_one_deleg(dp);
2543 spin_unlock(&recall_lock);
2544}
2545
2546static
2547int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2548{
2549 if (arg & F_UNLCK)
2550 return lease_modify(onlist, arg);
2551 else
2552 return -EAGAIN;
2553}
2554
2555static const struct lock_manager_operations nfsd_lease_mng_ops = {
2556 .lm_break = nfsd_break_deleg_cb,
2557 .lm_change = nfsd_change_deleg_cb,
2558};
2559
2560static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2561{
2562 if (nfsd4_has_session(cstate))
2563 return nfs_ok;
2564 if (seqid == so->so_seqid - 1)
2565 return nfserr_replay_me;
2566 if (seqid == so->so_seqid)
2567 return nfs_ok;
2568 return nfserr_bad_seqid;
2569}
2570
2571__be32
2572nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2573 struct nfsd4_open *open)
2574{
2575 clientid_t *clientid = &open->op_clientid;
2576 struct nfs4_client *clp = NULL;
2577 unsigned int strhashval;
2578 struct nfs4_openowner *oo = NULL;
2579 __be32 status;
2580
2581 if (STALE_CLIENTID(&open->op_clientid))
2582 return nfserr_stale_clientid;
2583 /*
2584 * In case we need it later, after we've already created the
2585 * file and don't want to risk a further failure:
2586 */
2587 open->op_file = nfsd4_alloc_file();
2588 if (open->op_file == NULL)
2589 return nfserr_jukebox;
2590
2591 strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2592 oo = find_openstateowner_str(strhashval, open);
2593 open->op_openowner = oo;
2594 if (!oo) {
2595 clp = find_confirmed_client(clientid);
2596 if (clp == NULL)
2597 return nfserr_expired;
2598 goto new_owner;
2599 }
2600 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2601 /* Replace unconfirmed owners without checking for replay. */
2602 clp = oo->oo_owner.so_client;
2603 release_openowner(oo);
2604 open->op_openowner = NULL;
2605 goto new_owner;
2606 }
2607 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2608 if (status)
2609 return status;
2610 clp = oo->oo_owner.so_client;
2611 goto alloc_stateid;
2612new_owner:
2613 oo = alloc_init_open_stateowner(strhashval, clp, open);
2614 if (oo == NULL)
2615 return nfserr_jukebox;
2616 open->op_openowner = oo;
2617alloc_stateid:
2618 open->op_stp = nfs4_alloc_stateid(clp);
2619 if (!open->op_stp)
2620 return nfserr_jukebox;
2621 return nfs_ok;
2622}
2623
2624static inline __be32
2625nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2626{
2627 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2628 return nfserr_openmode;
2629 else
2630 return nfs_ok;
2631}
2632
2633static int share_access_to_flags(u32 share_access)
2634{
2635 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2636}
2637
2638static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2639{
2640 struct nfs4_stid *ret;
2641
2642 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2643 if (!ret)
2644 return NULL;
2645 return delegstateid(ret);
2646}
2647
2648static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2649{
2650 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2651 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2652}
2653
2654static __be32
2655nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2656 struct nfs4_delegation **dp)
2657{
2658 int flags;
2659 __be32 status = nfserr_bad_stateid;
2660
2661 *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2662 if (*dp == NULL)
2663 goto out;
2664 flags = share_access_to_flags(open->op_share_access);
2665 status = nfs4_check_delegmode(*dp, flags);
2666 if (status)
2667 *dp = NULL;
2668out:
2669 if (!nfsd4_is_deleg_cur(open))
2670 return nfs_ok;
2671 if (status)
2672 return status;
2673 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2674 return nfs_ok;
2675}
2676
2677static __be32
2678nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2679{
2680 struct nfs4_ol_stateid *local;
2681 struct nfs4_openowner *oo = open->op_openowner;
2682
2683 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2684 /* ignore lock owners */
2685 if (local->st_stateowner->so_is_open_owner == 0)
2686 continue;
2687 /* remember if we have seen this open owner */
2688 if (local->st_stateowner == &oo->oo_owner)
2689 *stpp = local;
2690 /* check for conflicting share reservations */
2691 if (!test_share(local, open))
2692 return nfserr_share_denied;
2693 }
2694 return nfs_ok;
2695}
2696
2697static void nfs4_free_stateid(struct nfs4_ol_stateid *s)
2698{
2699 kmem_cache_free(stateid_slab, s);
2700}
2701
2702static inline int nfs4_access_to_access(u32 nfs4_access)
2703{
2704 int flags = 0;
2705
2706 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2707 flags |= NFSD_MAY_READ;
2708 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2709 flags |= NFSD_MAY_WRITE;
2710 return flags;
2711}
2712
2713static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2714 struct svc_fh *cur_fh, struct nfsd4_open *open)
2715{
2716 __be32 status;
2717 int oflag = nfs4_access_to_omode(open->op_share_access);
2718 int access = nfs4_access_to_access(open->op_share_access);
2719
2720 if (!fp->fi_fds[oflag]) {
2721 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2722 &fp->fi_fds[oflag]);
2723 if (status)
2724 return status;
2725 }
2726 nfs4_file_get_access(fp, oflag);
2727
2728 return nfs_ok;
2729}
2730
2731static inline __be32
2732nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2733 struct nfsd4_open *open)
2734{
2735 struct iattr iattr = {
2736 .ia_valid = ATTR_SIZE,
2737 .ia_size = 0,
2738 };
2739 if (!open->op_truncate)
2740 return 0;
2741 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2742 return nfserr_inval;
2743 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2744}
2745
2746static __be32
2747nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
2748{
2749 u32 op_share_access = open->op_share_access;
2750 bool new_access;
2751 __be32 status;
2752
2753 new_access = !test_access(op_share_access, stp);
2754 if (new_access) {
2755 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2756 if (status)
2757 return status;
2758 }
2759 status = nfsd4_truncate(rqstp, cur_fh, open);
2760 if (status) {
2761 if (new_access) {
2762 int oflag = nfs4_access_to_omode(op_share_access);
2763 nfs4_file_put_access(fp, oflag);
2764 }
2765 return status;
2766 }
2767 /* remember the open */
2768 set_access(op_share_access, stp);
2769 set_deny(open->op_share_deny, stp);
2770
2771 return nfs_ok;
2772}
2773
2774
2775static void
2776nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2777{
2778 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2779}
2780
2781/* Should we give out recallable state?: */
2782static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2783{
2784 if (clp->cl_cb_state == NFSD4_CB_UP)
2785 return true;
2786 /*
2787 * In the sessions case, since we don't have to establish a
2788 * separate connection for callbacks, we assume it's OK
2789 * until we hear otherwise:
2790 */
2791 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2792}
2793
2794static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2795{
2796 struct file_lock *fl;
2797
2798 fl = locks_alloc_lock();
2799 if (!fl)
2800 return NULL;
2801 locks_init_lock(fl);
2802 fl->fl_lmops = &nfsd_lease_mng_ops;
2803 fl->fl_flags = FL_LEASE;
2804 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2805 fl->fl_end = OFFSET_MAX;
2806 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2807 fl->fl_pid = current->tgid;
2808 return fl;
2809}
2810
2811static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2812{
2813 struct nfs4_file *fp = dp->dl_file;
2814 struct file_lock *fl;
2815 int status;
2816
2817 fl = nfs4_alloc_init_lease(dp, flag);
2818 if (!fl)
2819 return -ENOMEM;
2820 fl->fl_file = find_readable_file(fp);
2821 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2822 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2823 if (status) {
2824 list_del_init(&dp->dl_perclnt);
2825 locks_free_lock(fl);
2826 return -ENOMEM;
2827 }
2828 fp->fi_lease = fl;
2829 fp->fi_deleg_file = fl->fl_file;
2830 get_file(fp->fi_deleg_file);
2831 atomic_set(&fp->fi_delegees, 1);
2832 list_add(&dp->dl_perfile, &fp->fi_delegations);
2833 return 0;
2834}
2835
2836static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2837{
2838 struct nfs4_file *fp = dp->dl_file;
2839
2840 if (!fp->fi_lease)
2841 return nfs4_setlease(dp, flag);
2842 spin_lock(&recall_lock);
2843 if (fp->fi_had_conflict) {
2844 spin_unlock(&recall_lock);
2845 return -EAGAIN;
2846 }
2847 atomic_inc(&fp->fi_delegees);
2848 list_add(&dp->dl_perfile, &fp->fi_delegations);
2849 spin_unlock(&recall_lock);
2850 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2851 return 0;
2852}
2853
2854static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2855{
2856 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2857 if (status == -EAGAIN)
2858 open->op_why_no_deleg = WND4_CONTENTION;
2859 else {
2860 open->op_why_no_deleg = WND4_RESOURCE;
2861 switch (open->op_deleg_want) {
2862 case NFS4_SHARE_WANT_READ_DELEG:
2863 case NFS4_SHARE_WANT_WRITE_DELEG:
2864 case NFS4_SHARE_WANT_ANY_DELEG:
2865 break;
2866 case NFS4_SHARE_WANT_CANCEL:
2867 open->op_why_no_deleg = WND4_CANCELLED;
2868 break;
2869 case NFS4_SHARE_WANT_NO_DELEG:
2870 BUG(); /* not supposed to get here */
2871 }
2872 }
2873}
2874
2875/*
2876 * Attempt to hand out a delegation.
2877 */
2878static void
2879nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2880{
2881 struct nfs4_delegation *dp;
2882 struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2883 int cb_up;
2884 int status = 0, flag = 0;
2885
2886 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2887 flag = NFS4_OPEN_DELEGATE_NONE;
2888 open->op_recall = 0;
2889 switch (open->op_claim_type) {
2890 case NFS4_OPEN_CLAIM_PREVIOUS:
2891 if (!cb_up)
2892 open->op_recall = 1;
2893 flag = open->op_delegate_type;
2894 if (flag == NFS4_OPEN_DELEGATE_NONE)
2895 goto out;
2896 break;
2897 case NFS4_OPEN_CLAIM_NULL:
2898 /* Let's not give out any delegations till everyone's
2899 * had the chance to reclaim theirs.... */
2900 if (locks_in_grace())
2901 goto out;
2902 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2903 goto out;
2904 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2905 flag = NFS4_OPEN_DELEGATE_WRITE;
2906 else
2907 flag = NFS4_OPEN_DELEGATE_READ;
2908 break;
2909 default:
2910 goto out;
2911 }
2912
2913 dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2914 if (dp == NULL)
2915 goto out_no_deleg;
2916 status = nfs4_set_delegation(dp, flag);
2917 if (status)
2918 goto out_free;
2919
2920 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2921
2922 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2923 STATEID_VAL(&dp->dl_stid.sc_stateid));
2924out:
2925 open->op_delegate_type = flag;
2926 if (flag == NFS4_OPEN_DELEGATE_NONE) {
2927 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2928 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2929 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2930
2931 /* 4.1 client asking for a delegation? */
2932 if (open->op_deleg_want)
2933 nfsd4_open_deleg_none_ext(open, status);
2934 }
2935 return;
2936out_free:
2937 nfs4_put_delegation(dp);
2938out_no_deleg:
2939 flag = NFS4_OPEN_DELEGATE_NONE;
2940 goto out;
2941}
2942
2943static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
2944 struct nfs4_delegation *dp)
2945{
2946 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
2947 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2948 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2949 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
2950 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
2951 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2952 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2953 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
2954 }
2955 /* Otherwise the client must be confused wanting a delegation
2956 * it already has, therefore we don't return
2957 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
2958 */
2959}
2960
2961/*
2962 * called with nfs4_lock_state() held.
2963 */
2964__be32
2965nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2966{
2967 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2968 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
2969 struct nfs4_file *fp = NULL;
2970 struct inode *ino = current_fh->fh_dentry->d_inode;
2971 struct nfs4_ol_stateid *stp = NULL;
2972 struct nfs4_delegation *dp = NULL;
2973 __be32 status;
2974
2975 /*
2976 * Lookup file; if found, lookup stateid and check open request,
2977 * and check for delegations in the process of being recalled.
2978 * If not found, create the nfs4_file struct
2979 */
2980 fp = find_file(ino);
2981 if (fp) {
2982 if ((status = nfs4_check_open(fp, open, &stp)))
2983 goto out;
2984 status = nfs4_check_deleg(cl, fp, open, &dp);
2985 if (status)
2986 goto out;
2987 } else {
2988 status = nfserr_bad_stateid;
2989 if (nfsd4_is_deleg_cur(open))
2990 goto out;
2991 status = nfserr_jukebox;
2992 fp = open->op_file;
2993 open->op_file = NULL;
2994 nfsd4_init_file(fp, ino);
2995 }
2996
2997 /*
2998 * OPEN the file, or upgrade an existing OPEN.
2999 * If truncate fails, the OPEN fails.
3000 */
3001 if (stp) {
3002 /* Stateid was found, this is an OPEN upgrade */
3003 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3004 if (status)
3005 goto out;
3006 } else {
3007 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3008 if (status)
3009 goto out;
3010 stp = open->op_stp;
3011 open->op_stp = NULL;
3012 init_open_stateid(stp, fp, open);
3013 status = nfsd4_truncate(rqstp, current_fh, open);
3014 if (status) {
3015 release_open_stateid(stp);
3016 goto out;
3017 }
3018 }
3019 update_stateid(&stp->st_stid.sc_stateid);
3020 memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3021
3022 if (nfsd4_has_session(&resp->cstate)) {
3023 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3024
3025 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3026 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3027 open->op_why_no_deleg = WND4_NOT_WANTED;
3028 goto nodeleg;
3029 }
3030 }
3031
3032 /*
3033 * Attempt to hand out a delegation. No error return, because the
3034 * OPEN succeeds even if we fail.
3035 */
3036 nfs4_open_delegation(current_fh, open, stp);
3037nodeleg:
3038 status = nfs_ok;
3039
3040 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3041 STATEID_VAL(&stp->st_stid.sc_stateid));
3042out:
3043 /* 4.1 client trying to upgrade/downgrade delegation? */
3044 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3045 open->op_deleg_want)
3046 nfsd4_deleg_xgrade_none_ext(open, dp);
3047
3048 if (fp)
3049 put_nfs4_file(fp);
3050 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3051 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3052 /*
3053 * To finish the open response, we just need to set the rflags.
3054 */
3055 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3056 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3057 !nfsd4_has_session(&resp->cstate))
3058 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3059
3060 return status;
3061}
3062
3063void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3064{
3065 if (open->op_openowner) {
3066 struct nfs4_openowner *oo = open->op_openowner;
3067
3068 if (!list_empty(&oo->oo_owner.so_stateids))
3069 list_del_init(&oo->oo_close_lru);
3070 if (oo->oo_flags & NFS4_OO_NEW) {
3071 if (status) {
3072 release_openowner(oo);
3073 open->op_openowner = NULL;
3074 } else
3075 oo->oo_flags &= ~NFS4_OO_NEW;
3076 }
3077 }
3078 if (open->op_file)
3079 nfsd4_free_file(open->op_file);
3080 if (open->op_stp)
3081 nfs4_free_stateid(open->op_stp);
3082}
3083
3084__be32
3085nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3086 clientid_t *clid)
3087{
3088 struct nfs4_client *clp;
3089 __be32 status;
3090
3091 nfs4_lock_state();
3092 dprintk("process_renew(%08x/%08x): starting\n",
3093 clid->cl_boot, clid->cl_id);
3094 status = nfserr_stale_clientid;
3095 if (STALE_CLIENTID(clid))
3096 goto out;
3097 clp = find_confirmed_client(clid);
3098 status = nfserr_expired;
3099 if (clp == NULL) {
3100 /* We assume the client took too long to RENEW. */
3101 dprintk("nfsd4_renew: clientid not found!\n");
3102 goto out;
3103 }
3104 status = nfserr_cb_path_down;
3105 if (!list_empty(&clp->cl_delegations)
3106 && clp->cl_cb_state != NFSD4_CB_UP)
3107 goto out;
3108 status = nfs_ok;
3109out:
3110 nfs4_unlock_state();
3111 return status;
3112}
3113
3114static struct lock_manager nfsd4_manager = {
3115};
3116
3117static bool grace_ended;
3118
3119static void
3120nfsd4_end_grace(void)
3121{
3122 /* do nothing if grace period already ended */
3123 if (grace_ended)
3124 return;
3125
3126 dprintk("NFSD: end of grace period\n");
3127 grace_ended = true;
3128 nfsd4_record_grace_done(&init_net, boot_time);
3129 locks_end_grace(&nfsd4_manager);
3130 /*
3131 * Now that every NFSv4 client has had the chance to recover and
3132 * to see the (possibly new, possibly shorter) lease time, we
3133 * can safely set the next grace time to the current lease time:
3134 */
3135 nfsd4_grace = nfsd4_lease;
3136}
3137
3138static time_t
3139nfs4_laundromat(void)
3140{
3141 struct nfs4_client *clp;
3142 struct nfs4_openowner *oo;
3143 struct nfs4_delegation *dp;
3144 struct list_head *pos, *next, reaplist;
3145 time_t cutoff = get_seconds() - nfsd4_lease;
3146 time_t t, clientid_val = nfsd4_lease;
3147 time_t u, test_val = nfsd4_lease;
3148
3149 nfs4_lock_state();
3150
3151 dprintk("NFSD: laundromat service - starting\n");
3152 nfsd4_end_grace();
3153 INIT_LIST_HEAD(&reaplist);
3154 spin_lock(&client_lock);
3155 list_for_each_safe(pos, next, &client_lru) {
3156 clp = list_entry(pos, struct nfs4_client, cl_lru);
3157 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3158 t = clp->cl_time - cutoff;
3159 if (clientid_val > t)
3160 clientid_val = t;
3161 break;
3162 }
3163 if (atomic_read(&clp->cl_refcount)) {
3164 dprintk("NFSD: client in use (clientid %08x)\n",
3165 clp->cl_clientid.cl_id);
3166 continue;
3167 }
3168 unhash_client_locked(clp);
3169 list_add(&clp->cl_lru, &reaplist);
3170 }
3171 spin_unlock(&client_lock);
3172 list_for_each_safe(pos, next, &reaplist) {
3173 clp = list_entry(pos, struct nfs4_client, cl_lru);
3174 dprintk("NFSD: purging unused client (clientid %08x)\n",
3175 clp->cl_clientid.cl_id);
3176 nfsd4_client_record_remove(clp);
3177 expire_client(clp);
3178 }
3179 spin_lock(&recall_lock);
3180 list_for_each_safe(pos, next, &del_recall_lru) {
3181 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3182 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3183 u = dp->dl_time - cutoff;
3184 if (test_val > u)
3185 test_val = u;
3186 break;
3187 }
3188 list_move(&dp->dl_recall_lru, &reaplist);
3189 }
3190 spin_unlock(&recall_lock);
3191 list_for_each_safe(pos, next, &reaplist) {
3192 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3193 unhash_delegation(dp);
3194 }
3195 test_val = nfsd4_lease;
3196 list_for_each_safe(pos, next, &close_lru) {
3197 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3198 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3199 u = oo->oo_time - cutoff;
3200 if (test_val > u)
3201 test_val = u;
3202 break;
3203 }
3204 release_openowner(oo);
3205 }
3206 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3207 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3208 nfs4_unlock_state();
3209 return clientid_val;
3210}
3211
3212static struct workqueue_struct *laundry_wq;
3213static void laundromat_main(struct work_struct *);
3214static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3215
3216static void
3217laundromat_main(struct work_struct *not_used)
3218{
3219 time_t t;
3220
3221 t = nfs4_laundromat();
3222 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3223 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3224}
3225
3226static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3227{
3228 if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3229 return nfserr_bad_stateid;
3230 return nfs_ok;
3231}
3232
3233static int
3234STALE_STATEID(stateid_t *stateid)
3235{
3236 if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3237 return 0;
3238 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3239 STATEID_VAL(stateid));
3240 return 1;
3241}
3242
3243static inline int
3244access_permit_read(struct nfs4_ol_stateid *stp)
3245{
3246 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3247 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3248 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3249}
3250
3251static inline int
3252access_permit_write(struct nfs4_ol_stateid *stp)
3253{
3254 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3255 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3256}
3257
3258static
3259__be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3260{
3261 __be32 status = nfserr_openmode;
3262
3263 /* For lock stateid's, we test the parent open, not the lock: */
3264 if (stp->st_openstp)
3265 stp = stp->st_openstp;
3266 if ((flags & WR_STATE) && !access_permit_write(stp))
3267 goto out;
3268 if ((flags & RD_STATE) && !access_permit_read(stp))
3269 goto out;
3270 status = nfs_ok;
3271out:
3272 return status;
3273}
3274
3275static inline __be32
3276check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3277{
3278 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3279 return nfs_ok;
3280 else if (locks_in_grace()) {
3281 /* Answer in remaining cases depends on existence of
3282 * conflicting state; so we must wait out the grace period. */
3283 return nfserr_grace;
3284 } else if (flags & WR_STATE)
3285 return nfs4_share_conflict(current_fh,
3286 NFS4_SHARE_DENY_WRITE);
3287 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3288 return nfs4_share_conflict(current_fh,
3289 NFS4_SHARE_DENY_READ);
3290}
3291
3292/*
3293 * Allow READ/WRITE during grace period on recovered state only for files
3294 * that are not able to provide mandatory locking.
3295 */
3296static inline int
3297grace_disallows_io(struct inode *inode)
3298{
3299 return locks_in_grace() && mandatory_lock(inode);
3300}
3301
3302/* Returns true iff a is later than b: */
3303static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3304{
3305 return (s32)a->si_generation - (s32)b->si_generation > 0;
3306}
3307
3308static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3309{
3310 /*
3311 * When sessions are used the stateid generation number is ignored
3312 * when it is zero.
3313 */
3314 if (has_session && in->si_generation == 0)
3315 return nfs_ok;
3316
3317 if (in->si_generation == ref->si_generation)
3318 return nfs_ok;
3319
3320 /* If the client sends us a stateid from the future, it's buggy: */
3321 if (stateid_generation_after(in, ref))
3322 return nfserr_bad_stateid;
3323 /*
3324 * However, we could see a stateid from the past, even from a
3325 * non-buggy client. For example, if the client sends a lock
3326 * while some IO is outstanding, the lock may bump si_generation
3327 * while the IO is still in flight. The client could avoid that
3328 * situation by waiting for responses on all the IO requests,
3329 * but better performance may result in retrying IO that
3330 * receives an old_stateid error if requests are rarely
3331 * reordered in flight:
3332 */
3333 return nfserr_old_stateid;
3334}
3335
3336__be32 nfs4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3337{
3338 struct nfs4_stid *s;
3339 struct nfs4_ol_stateid *ols;
3340 __be32 status;
3341
3342 if (STALE_STATEID(stateid))
3343 return nfserr_stale_stateid;
3344
3345 s = find_stateid(cl, stateid);
3346 if (!s)
3347 return nfserr_stale_stateid;
3348 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3349 if (status)
3350 return status;
3351 if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3352 return nfs_ok;
3353 ols = openlockstateid(s);
3354 if (ols->st_stateowner->so_is_open_owner
3355 && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3356 return nfserr_bad_stateid;
3357 return nfs_ok;
3358}
3359
3360static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3361{
3362 struct nfs4_client *cl;
3363
3364 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3365 return nfserr_bad_stateid;
3366 if (STALE_STATEID(stateid))
3367 return nfserr_stale_stateid;
3368 cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3369 if (!cl)
3370 return nfserr_expired;
3371 *s = find_stateid_by_type(cl, stateid, typemask);
3372 if (!*s)
3373 return nfserr_bad_stateid;
3374 return nfs_ok;
3375
3376}
3377
3378/*
3379* Checks for stateid operations
3380*/
3381__be32
3382nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3383 stateid_t *stateid, int flags, struct file **filpp)
3384{
3385 struct nfs4_stid *s;
3386 struct nfs4_ol_stateid *stp = NULL;
3387 struct nfs4_delegation *dp = NULL;
3388 struct svc_fh *current_fh = &cstate->current_fh;
3389 struct inode *ino = current_fh->fh_dentry->d_inode;
3390 __be32 status;
3391
3392 if (filpp)
3393 *filpp = NULL;
3394
3395 if (grace_disallows_io(ino))
3396 return nfserr_grace;
3397
3398 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3399 return check_special_stateids(current_fh, stateid, flags);
3400
3401 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s);
3402 if (status)
3403 return status;
3404 status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3405 if (status)
3406 goto out;
3407 switch (s->sc_type) {
3408 case NFS4_DELEG_STID:
3409 dp = delegstateid(s);
3410 status = nfs4_check_delegmode(dp, flags);
3411 if (status)
3412 goto out;
3413 if (filpp) {
3414 *filpp = dp->dl_file->fi_deleg_file;
3415 BUG_ON(!*filpp);
3416 }
3417 break;
3418 case NFS4_OPEN_STID:
3419 case NFS4_LOCK_STID:
3420 stp = openlockstateid(s);
3421 status = nfs4_check_fh(current_fh, stp);
3422 if (status)
3423 goto out;
3424 if (stp->st_stateowner->so_is_open_owner
3425 && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3426 goto out;
3427 status = nfs4_check_openmode(stp, flags);
3428 if (status)
3429 goto out;
3430 if (filpp) {
3431 if (flags & RD_STATE)
3432 *filpp = find_readable_file(stp->st_file);
3433 else
3434 *filpp = find_writeable_file(stp->st_file);
3435 }
3436 break;
3437 default:
3438 return nfserr_bad_stateid;
3439 }
3440 status = nfs_ok;
3441out:
3442 return status;
3443}
3444
3445static __be32
3446nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3447{
3448 if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3449 return nfserr_locks_held;
3450 release_lock_stateid(stp);
3451 return nfs_ok;
3452}
3453
3454/*
3455 * Test if the stateid is valid
3456 */
3457__be32
3458nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3459 struct nfsd4_test_stateid *test_stateid)
3460{
3461 struct nfsd4_test_stateid_id *stateid;
3462 struct nfs4_client *cl = cstate->session->se_client;
3463
3464 nfs4_lock_state();
3465 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3466 stateid->ts_id_status = nfs4_validate_stateid(cl, &stateid->ts_id_stateid);
3467 nfs4_unlock_state();
3468
3469 return nfs_ok;
3470}
3471
3472__be32
3473nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3474 struct nfsd4_free_stateid *free_stateid)
3475{
3476 stateid_t *stateid = &free_stateid->fr_stateid;
3477 struct nfs4_stid *s;
3478 struct nfs4_client *cl = cstate->session->se_client;
3479 __be32 ret = nfserr_bad_stateid;
3480
3481 nfs4_lock_state();
3482 s = find_stateid(cl, stateid);
3483 if (!s)
3484 goto out;
3485 switch (s->sc_type) {
3486 case NFS4_DELEG_STID:
3487 ret = nfserr_locks_held;
3488 goto out;
3489 case NFS4_OPEN_STID:
3490 case NFS4_LOCK_STID:
3491 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3492 if (ret)
3493 goto out;
3494 if (s->sc_type == NFS4_LOCK_STID)
3495 ret = nfsd4_free_lock_stateid(openlockstateid(s));
3496 else
3497 ret = nfserr_locks_held;
3498 break;
3499 default:
3500 ret = nfserr_bad_stateid;
3501 }
3502out:
3503 nfs4_unlock_state();
3504 return ret;
3505}
3506
3507static inline int
3508setlkflg (int type)
3509{
3510 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3511 RD_STATE : WR_STATE;
3512}
3513
3514static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3515{
3516 struct svc_fh *current_fh = &cstate->current_fh;
3517 struct nfs4_stateowner *sop = stp->st_stateowner;
3518 __be32 status;
3519
3520 status = nfsd4_check_seqid(cstate, sop, seqid);
3521 if (status)
3522 return status;
3523 if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3524 /*
3525 * "Closed" stateid's exist *only* to return
3526 * nfserr_replay_me from the previous step.
3527 */
3528 return nfserr_bad_stateid;
3529 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3530 if (status)
3531 return status;
3532 return nfs4_check_fh(current_fh, stp);
3533}
3534
3535/*
3536 * Checks for sequence id mutating operations.
3537 */
3538static __be32
3539nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3540 stateid_t *stateid, char typemask,
3541 struct nfs4_ol_stateid **stpp)
3542{
3543 __be32 status;
3544 struct nfs4_stid *s;
3545
3546 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3547 seqid, STATEID_VAL(stateid));
3548
3549 *stpp = NULL;
3550 status = nfsd4_lookup_stateid(stateid, typemask, &s);
3551 if (status)
3552 return status;
3553 *stpp = openlockstateid(s);
3554 cstate->replay_owner = (*stpp)->st_stateowner;
3555
3556 return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3557}
3558
3559static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3560{
3561 __be32 status;
3562 struct nfs4_openowner *oo;
3563
3564 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3565 NFS4_OPEN_STID, stpp);
3566 if (status)
3567 return status;
3568 oo = openowner((*stpp)->st_stateowner);
3569 if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3570 return nfserr_bad_stateid;
3571 return nfs_ok;
3572}
3573
3574__be32
3575nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3576 struct nfsd4_open_confirm *oc)
3577{
3578 __be32 status;
3579 struct nfs4_openowner *oo;
3580 struct nfs4_ol_stateid *stp;
3581
3582 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3583 (int)cstate->current_fh.fh_dentry->d_name.len,
3584 cstate->current_fh.fh_dentry->d_name.name);
3585
3586 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3587 if (status)
3588 return status;
3589
3590 nfs4_lock_state();
3591
3592 status = nfs4_preprocess_seqid_op(cstate,
3593 oc->oc_seqid, &oc->oc_req_stateid,
3594 NFS4_OPEN_STID, &stp);
3595 if (status)
3596 goto out;
3597 oo = openowner(stp->st_stateowner);
3598 status = nfserr_bad_stateid;
3599 if (oo->oo_flags & NFS4_OO_CONFIRMED)
3600 goto out;
3601 oo->oo_flags |= NFS4_OO_CONFIRMED;
3602 update_stateid(&stp->st_stid.sc_stateid);
3603 memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3604 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3605 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3606
3607 nfsd4_client_record_create(oo->oo_owner.so_client);
3608 status = nfs_ok;
3609out:
3610 if (!cstate->replay_owner)
3611 nfs4_unlock_state();
3612 return status;
3613}
3614
3615static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3616{
3617 if (!test_access(access, stp))
3618 return;
3619 nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3620 clear_access(access, stp);
3621}
3622
3623static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3624{
3625 switch (to_access) {
3626 case NFS4_SHARE_ACCESS_READ:
3627 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3628 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3629 break;
3630 case NFS4_SHARE_ACCESS_WRITE:
3631 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3632 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3633 break;
3634 case NFS4_SHARE_ACCESS_BOTH:
3635 break;
3636 default:
3637 BUG();
3638 }
3639}
3640
3641static void
3642reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3643{
3644 int i;
3645 for (i = 0; i < 4; i++) {
3646 if ((i & deny) != i)
3647 clear_deny(i, stp);
3648 }
3649}
3650
3651__be32
3652nfsd4_open_downgrade(struct svc_rqst *rqstp,
3653 struct nfsd4_compound_state *cstate,
3654 struct nfsd4_open_downgrade *od)
3655{
3656 __be32 status;
3657 struct nfs4_ol_stateid *stp;
3658
3659 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3660 (int)cstate->current_fh.fh_dentry->d_name.len,
3661 cstate->current_fh.fh_dentry->d_name.name);
3662
3663 /* We don't yet support WANT bits: */
3664 if (od->od_deleg_want)
3665 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3666 od->od_deleg_want);
3667
3668 nfs4_lock_state();
3669 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3670 &od->od_stateid, &stp);
3671 if (status)
3672 goto out;
3673 status = nfserr_inval;
3674 if (!test_access(od->od_share_access, stp)) {
3675 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3676 stp->st_access_bmap, od->od_share_access);
3677 goto out;
3678 }
3679 if (!test_deny(od->od_share_deny, stp)) {
3680 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3681 stp->st_deny_bmap, od->od_share_deny);
3682 goto out;
3683 }
3684 nfs4_stateid_downgrade(stp, od->od_share_access);
3685
3686 reset_union_bmap_deny(od->od_share_deny, stp);
3687
3688 update_stateid(&stp->st_stid.sc_stateid);
3689 memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3690 status = nfs_ok;
3691out:
3692 if (!cstate->replay_owner)
3693 nfs4_unlock_state();
3694 return status;
3695}
3696
3697void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3698{
3699 struct nfs4_openowner *oo;
3700 struct nfs4_ol_stateid *s;
3701
3702 if (!so->so_is_open_owner)
3703 return;
3704 oo = openowner(so);
3705 s = oo->oo_last_closed_stid;
3706 if (!s)
3707 return;
3708 if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3709 /* Release the last_closed_stid on the next seqid bump: */
3710 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3711 return;
3712 }
3713 oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3714 release_last_closed_stateid(oo);
3715}
3716
3717static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3718{
3719 unhash_open_stateid(s);
3720 s->st_stid.sc_type = NFS4_CLOSED_STID;
3721}
3722
3723/*
3724 * nfs4_unlock_state() called after encode
3725 */
3726__be32
3727nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3728 struct nfsd4_close *close)
3729{
3730 __be32 status;
3731 struct nfs4_openowner *oo;
3732 struct nfs4_ol_stateid *stp;
3733
3734 dprintk("NFSD: nfsd4_close on file %.*s\n",
3735 (int)cstate->current_fh.fh_dentry->d_name.len,
3736 cstate->current_fh.fh_dentry->d_name.name);
3737
3738 nfs4_lock_state();
3739 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3740 &close->cl_stateid,
3741 NFS4_OPEN_STID|NFS4_CLOSED_STID,
3742 &stp);
3743 if (status)
3744 goto out;
3745 oo = openowner(stp->st_stateowner);
3746 status = nfs_ok;
3747 update_stateid(&stp->st_stid.sc_stateid);
3748 memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3749
3750 nfsd4_close_open_stateid(stp);
3751 oo->oo_last_closed_stid = stp;
3752
3753 /* place unused nfs4_stateowners on so_close_lru list to be
3754 * released by the laundromat service after the lease period
3755 * to enable us to handle CLOSE replay
3756 */
3757 if (list_empty(&oo->oo_owner.so_stateids))
3758 move_to_close_lru(oo);
3759out:
3760 if (!cstate->replay_owner)
3761 nfs4_unlock_state();
3762 return status;
3763}
3764
3765__be32
3766nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3767 struct nfsd4_delegreturn *dr)
3768{
3769 struct nfs4_delegation *dp;
3770 stateid_t *stateid = &dr->dr_stateid;
3771 struct nfs4_stid *s;
3772 struct inode *inode;
3773 __be32 status;
3774
3775 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3776 return status;
3777 inode = cstate->current_fh.fh_dentry->d_inode;
3778
3779 nfs4_lock_state();
3780 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s);
3781 if (status)
3782 goto out;
3783 dp = delegstateid(s);
3784 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3785 if (status)
3786 goto out;
3787
3788 unhash_delegation(dp);
3789out:
3790 nfs4_unlock_state();
3791
3792 return status;
3793}
3794
3795
3796#define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3797
3798#define LOCKOWNER_INO_HASH_BITS 8
3799#define LOCKOWNER_INO_HASH_SIZE (1 << LOCKOWNER_INO_HASH_BITS)
3800#define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3801
3802static inline u64
3803end_offset(u64 start, u64 len)
3804{
3805 u64 end;
3806
3807 end = start + len;
3808 return end >= start ? end: NFS4_MAX_UINT64;
3809}
3810
3811/* last octet in a range */
3812static inline u64
3813last_byte_offset(u64 start, u64 len)
3814{
3815 u64 end;
3816
3817 BUG_ON(!len);
3818 end = start + len;
3819 return end > start ? end - 1: NFS4_MAX_UINT64;
3820}
3821
3822static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3823{
3824 return (file_hashval(inode) + cl_id
3825 + opaque_hashval(ownername->data, ownername->len))
3826 & LOCKOWNER_INO_HASH_MASK;
3827}
3828
3829static struct list_head lockowner_ino_hashtbl[LOCKOWNER_INO_HASH_SIZE];
3830
3831/*
3832 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3833 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3834 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3835 * locking, this prevents us from being completely protocol-compliant. The
3836 * real solution to this problem is to start using unsigned file offsets in
3837 * the VFS, but this is a very deep change!
3838 */
3839static inline void
3840nfs4_transform_lock_offset(struct file_lock *lock)
3841{
3842 if (lock->fl_start < 0)
3843 lock->fl_start = OFFSET_MAX;
3844 if (lock->fl_end < 0)
3845 lock->fl_end = OFFSET_MAX;
3846}
3847
3848/* Hack!: For now, we're defining this just so we can use a pointer to it
3849 * as a unique cookie to identify our (NFSv4's) posix locks. */
3850static const struct lock_manager_operations nfsd_posix_mng_ops = {
3851};
3852
3853static inline void
3854nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3855{
3856 struct nfs4_lockowner *lo;
3857
3858 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3859 lo = (struct nfs4_lockowner *) fl->fl_owner;
3860 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3861 lo->lo_owner.so_owner.len, GFP_KERNEL);
3862 if (!deny->ld_owner.data)
3863 /* We just don't care that much */
3864 goto nevermind;
3865 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3866 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3867 } else {
3868nevermind:
3869 deny->ld_owner.len = 0;
3870 deny->ld_owner.data = NULL;
3871 deny->ld_clientid.cl_boot = 0;
3872 deny->ld_clientid.cl_id = 0;
3873 }
3874 deny->ld_start = fl->fl_start;
3875 deny->ld_length = NFS4_MAX_UINT64;
3876 if (fl->fl_end != NFS4_MAX_UINT64)
3877 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3878 deny->ld_type = NFS4_READ_LT;
3879 if (fl->fl_type != F_RDLCK)
3880 deny->ld_type = NFS4_WRITE_LT;
3881}
3882
3883static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3884{
3885 struct nfs4_ol_stateid *lst;
3886
3887 if (!same_owner_str(&lo->lo_owner, owner, clid))
3888 return false;
3889 lst = list_first_entry(&lo->lo_owner.so_stateids,
3890 struct nfs4_ol_stateid, st_perstateowner);
3891 return lst->st_file->fi_inode == inode;
3892}
3893
3894static struct nfs4_lockowner *
3895find_lockowner_str(struct inode *inode, clientid_t *clid,
3896 struct xdr_netobj *owner)
3897{
3898 unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3899 struct nfs4_lockowner *lo;
3900
3901 list_for_each_entry(lo, &lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3902 if (same_lockowner_ino(lo, inode, clid, owner))
3903 return lo;
3904 }
3905 return NULL;
3906}
3907
3908static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3909{
3910 struct inode *inode = open_stp->st_file->fi_inode;
3911 unsigned int inohash = lockowner_ino_hashval(inode,
3912 clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
3913
3914 list_add(&lo->lo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
3915 list_add(&lo->lo_owner_ino_hash, &lockowner_ino_hashtbl[inohash]);
3916 list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3917}
3918
3919/*
3920 * Alloc a lock owner structure.
3921 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3922 * occurred.
3923 *
3924 * strhashval = ownerstr_hashval
3925 */
3926
3927static struct nfs4_lockowner *
3928alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3929 struct nfs4_lockowner *lo;
3930
3931 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
3932 if (!lo)
3933 return NULL;
3934 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
3935 lo->lo_owner.so_is_open_owner = 0;
3936 /* It is the openowner seqid that will be incremented in encode in the
3937 * case of new lockowners; so increment the lock seqid manually: */
3938 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
3939 hash_lockowner(lo, strhashval, clp, open_stp);
3940 return lo;
3941}
3942
3943static struct nfs4_ol_stateid *
3944alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
3945{
3946 struct nfs4_ol_stateid *stp;
3947 struct nfs4_client *clp = lo->lo_owner.so_client;
3948
3949 stp = nfs4_alloc_stateid(clp);
3950 if (stp == NULL)
3951 return NULL;
3952 init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
3953 list_add(&stp->st_perfile, &fp->fi_stateids);
3954 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
3955 stp->st_stateowner = &lo->lo_owner;
3956 get_nfs4_file(fp);
3957 stp->st_file = fp;
3958 stp->st_access_bmap = 0;
3959 stp->st_deny_bmap = open_stp->st_deny_bmap;
3960 stp->st_openstp = open_stp;
3961 return stp;
3962}
3963
3964static int
3965check_lock_length(u64 offset, u64 length)
3966{
3967 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3968 LOFF_OVERFLOW(offset, length)));
3969}
3970
3971static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
3972{
3973 struct nfs4_file *fp = lock_stp->st_file;
3974 int oflag = nfs4_access_to_omode(access);
3975
3976 if (test_access(access, lock_stp))
3977 return;
3978 nfs4_file_get_access(fp, oflag);
3979 set_access(access, lock_stp);
3980}
3981
3982static __be32 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate, struct nfs4_ol_stateid *ost, struct nfsd4_lock *lock, struct nfs4_ol_stateid **lst, bool *new)
3983{
3984 struct nfs4_file *fi = ost->st_file;
3985 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
3986 struct nfs4_client *cl = oo->oo_owner.so_client;
3987 struct nfs4_lockowner *lo;
3988 unsigned int strhashval;
3989
3990 lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid, &lock->v.new.owner);
3991 if (lo) {
3992 if (!cstate->minorversion)
3993 return nfserr_bad_seqid;
3994 /* XXX: a lockowner always has exactly one stateid: */
3995 *lst = list_first_entry(&lo->lo_owner.so_stateids,
3996 struct nfs4_ol_stateid, st_perstateowner);
3997 return nfs_ok;
3998 }
3999 strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4000 &lock->v.new.owner);
4001 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4002 if (lo == NULL)
4003 return nfserr_jukebox;
4004 *lst = alloc_init_lock_stateid(lo, fi, ost);
4005 if (*lst == NULL) {
4006 release_lockowner(lo);
4007 return nfserr_jukebox;
4008 }
4009 *new = true;
4010 return nfs_ok;
4011}
4012
4013/*
4014 * LOCK operation
4015 */
4016__be32
4017nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4018 struct nfsd4_lock *lock)
4019{
4020 struct nfs4_openowner *open_sop = NULL;
4021 struct nfs4_lockowner *lock_sop = NULL;
4022 struct nfs4_ol_stateid *lock_stp;
4023 struct file *filp = NULL;
4024 struct file_lock file_lock;
4025 struct file_lock conflock;
4026 __be32 status = 0;
4027 bool new_state = false;
4028 int lkflg;
4029 int err;
4030
4031 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4032 (long long) lock->lk_offset,
4033 (long long) lock->lk_length);
4034
4035 if (check_lock_length(lock->lk_offset, lock->lk_length))
4036 return nfserr_inval;
4037
4038 if ((status = fh_verify(rqstp, &cstate->current_fh,
4039 S_IFREG, NFSD_MAY_LOCK))) {
4040 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4041 return status;
4042 }
4043
4044 nfs4_lock_state();
4045
4046 if (lock->lk_is_new) {
4047 /*
4048 * Client indicates that this is a new lockowner.
4049 * Use open owner and open stateid to create lock owner and
4050 * lock stateid.
4051 */
4052 struct nfs4_ol_stateid *open_stp = NULL;
4053
4054 if (nfsd4_has_session(cstate))
4055 /* See rfc 5661 18.10.3: given clientid is ignored: */
4056 memcpy(&lock->v.new.clientid,
4057 &cstate->session->se_client->cl_clientid,
4058 sizeof(clientid_t));
4059
4060 status = nfserr_stale_clientid;
4061 if (STALE_CLIENTID(&lock->lk_new_clientid))
4062 goto out;
4063
4064 /* validate and update open stateid and open seqid */
4065 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4066 lock->lk_new_open_seqid,
4067 &lock->lk_new_open_stateid,
4068 &open_stp);
4069 if (status)
4070 goto out;
4071 open_sop = openowner(open_stp->st_stateowner);
4072 status = nfserr_bad_stateid;
4073 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4074 &lock->v.new.clientid))
4075 goto out;
4076 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4077 &lock_stp, &new_state);
4078 if (status)
4079 goto out;
4080 } else {
4081 /* lock (lock owner + lock stateid) already exists */
4082 status = nfs4_preprocess_seqid_op(cstate,
4083 lock->lk_old_lock_seqid,
4084 &lock->lk_old_lock_stateid,
4085 NFS4_LOCK_STID, &lock_stp);
4086 if (status)
4087 goto out;
4088 }
4089 lock_sop = lockowner(lock_stp->st_stateowner);
4090
4091 lkflg = setlkflg(lock->lk_type);
4092 status = nfs4_check_openmode(lock_stp, lkflg);
4093 if (status)
4094 goto out;
4095
4096 status = nfserr_grace;
4097 if (locks_in_grace() && !lock->lk_reclaim)
4098 goto out;
4099 status = nfserr_no_grace;
4100 if (!locks_in_grace() && lock->lk_reclaim)
4101 goto out;
4102
4103 locks_init_lock(&file_lock);
4104 switch (lock->lk_type) {
4105 case NFS4_READ_LT:
4106 case NFS4_READW_LT:
4107 filp = find_readable_file(lock_stp->st_file);
4108 if (filp)
4109 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4110 file_lock.fl_type = F_RDLCK;
4111 break;
4112 case NFS4_WRITE_LT:
4113 case NFS4_WRITEW_LT:
4114 filp = find_writeable_file(lock_stp->st_file);
4115 if (filp)
4116 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4117 file_lock.fl_type = F_WRLCK;
4118 break;
4119 default:
4120 status = nfserr_inval;
4121 goto out;
4122 }
4123 if (!filp) {
4124 status = nfserr_openmode;
4125 goto out;
4126 }
4127 file_lock.fl_owner = (fl_owner_t)lock_sop;
4128 file_lock.fl_pid = current->tgid;
4129 file_lock.fl_file = filp;
4130 file_lock.fl_flags = FL_POSIX;
4131 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4132
4133 file_lock.fl_start = lock->lk_offset;
4134 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4135 nfs4_transform_lock_offset(&file_lock);
4136
4137 /*
4138 * Try to lock the file in the VFS.
4139 * Note: locks.c uses the BKL to protect the inode's lock list.
4140 */
4141
4142 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4143 switch (-err) {
4144 case 0: /* success! */
4145 update_stateid(&lock_stp->st_stid.sc_stateid);
4146 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4147 sizeof(stateid_t));
4148 status = 0;
4149 break;
4150 case (EAGAIN): /* conflock holds conflicting lock */
4151 status = nfserr_denied;
4152 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4153 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4154 break;
4155 case (EDEADLK):
4156 status = nfserr_deadlock;
4157 break;
4158 default:
4159 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4160 status = nfserrno(err);
4161 break;
4162 }
4163out:
4164 if (status && new_state)
4165 release_lockowner(lock_sop);
4166 if (!cstate->replay_owner)
4167 nfs4_unlock_state();
4168 return status;
4169}
4170
4171/*
4172 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4173 * so we do a temporary open here just to get an open file to pass to
4174 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
4175 * inode operation.)
4176 */
4177static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4178{
4179 struct file *file;
4180 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4181 if (!err) {
4182 err = nfserrno(vfs_test_lock(file, lock));
4183 nfsd_close(file);
4184 }
4185 return err;
4186}
4187
4188/*
4189 * LOCKT operation
4190 */
4191__be32
4192nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4193 struct nfsd4_lockt *lockt)
4194{
4195 struct inode *inode;
4196 struct file_lock file_lock;
4197 struct nfs4_lockowner *lo;
4198 __be32 status;
4199
4200 if (locks_in_grace())
4201 return nfserr_grace;
4202
4203 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4204 return nfserr_inval;
4205
4206 nfs4_lock_state();
4207
4208 status = nfserr_stale_clientid;
4209 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4210 goto out;
4211
4212 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4213 goto out;
4214
4215 inode = cstate->current_fh.fh_dentry->d_inode;
4216 locks_init_lock(&file_lock);
4217 switch (lockt->lt_type) {
4218 case NFS4_READ_LT:
4219 case NFS4_READW_LT:
4220 file_lock.fl_type = F_RDLCK;
4221 break;
4222 case NFS4_WRITE_LT:
4223 case NFS4_WRITEW_LT:
4224 file_lock.fl_type = F_WRLCK;
4225 break;
4226 default:
4227 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4228 status = nfserr_inval;
4229 goto out;
4230 }
4231
4232 lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4233 if (lo)
4234 file_lock.fl_owner = (fl_owner_t)lo;
4235 file_lock.fl_pid = current->tgid;
4236 file_lock.fl_flags = FL_POSIX;
4237
4238 file_lock.fl_start = lockt->lt_offset;
4239 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4240
4241 nfs4_transform_lock_offset(&file_lock);
4242
4243 status = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4244 if (status)
4245 goto out;
4246
4247 if (file_lock.fl_type != F_UNLCK) {
4248 status = nfserr_denied;
4249 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4250 }
4251out:
4252 nfs4_unlock_state();
4253 return status;
4254}
4255
4256__be32
4257nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4258 struct nfsd4_locku *locku)
4259{
4260 struct nfs4_ol_stateid *stp;
4261 struct file *filp = NULL;
4262 struct file_lock file_lock;
4263 __be32 status;
4264 int err;
4265
4266 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4267 (long long) locku->lu_offset,
4268 (long long) locku->lu_length);
4269
4270 if (check_lock_length(locku->lu_offset, locku->lu_length))
4271 return nfserr_inval;
4272
4273 nfs4_lock_state();
4274
4275 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4276 &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4277 if (status)
4278 goto out;
4279 filp = find_any_file(stp->st_file);
4280 if (!filp) {
4281 status = nfserr_lock_range;
4282 goto out;
4283 }
4284 BUG_ON(!filp);
4285 locks_init_lock(&file_lock);
4286 file_lock.fl_type = F_UNLCK;
4287 file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4288 file_lock.fl_pid = current->tgid;
4289 file_lock.fl_file = filp;
4290 file_lock.fl_flags = FL_POSIX;
4291 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4292 file_lock.fl_start = locku->lu_offset;
4293
4294 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4295 nfs4_transform_lock_offset(&file_lock);
4296
4297 /*
4298 * Try to unlock the file in the VFS.
4299 */
4300 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4301 if (err) {
4302 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4303 goto out_nfserr;
4304 }
4305 /*
4306 * OK, unlock succeeded; the only thing left to do is update the stateid.
4307 */
4308 update_stateid(&stp->st_stid.sc_stateid);
4309 memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4310
4311out:
4312 if (!cstate->replay_owner)
4313 nfs4_unlock_state();
4314 return status;
4315
4316out_nfserr:
4317 status = nfserrno(err);
4318 goto out;
4319}
4320
4321/*
4322 * returns
4323 * 1: locks held by lockowner
4324 * 0: no locks held by lockowner
4325 */
4326static int
4327check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4328{
4329 struct file_lock **flpp;
4330 struct inode *inode = filp->fi_inode;
4331 int status = 0;
4332
4333 lock_flocks();
4334 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4335 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4336 status = 1;
4337 goto out;
4338 }
4339 }
4340out:
4341 unlock_flocks();
4342 return status;
4343}
4344
4345__be32
4346nfsd4_release_lockowner(struct svc_rqst *rqstp,
4347 struct nfsd4_compound_state *cstate,
4348 struct nfsd4_release_lockowner *rlockowner)
4349{
4350 clientid_t *clid = &rlockowner->rl_clientid;
4351 struct nfs4_stateowner *sop;
4352 struct nfs4_lockowner *lo;
4353 struct nfs4_ol_stateid *stp;
4354 struct xdr_netobj *owner = &rlockowner->rl_owner;
4355 struct list_head matches;
4356 unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4357 __be32 status;
4358
4359 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4360 clid->cl_boot, clid->cl_id);
4361
4362 /* XXX check for lease expiration */
4363
4364 status = nfserr_stale_clientid;
4365 if (STALE_CLIENTID(clid))
4366 return status;
4367
4368 nfs4_lock_state();
4369
4370 status = nfserr_locks_held;
4371 INIT_LIST_HEAD(&matches);
4372
4373 list_for_each_entry(sop, &ownerstr_hashtbl[hashval], so_strhash) {
4374 if (sop->so_is_open_owner)
4375 continue;
4376 if (!same_owner_str(sop, owner, clid))
4377 continue;
4378 list_for_each_entry(stp, &sop->so_stateids,
4379 st_perstateowner) {
4380 lo = lockowner(sop);
4381 if (check_for_locks(stp->st_file, lo))
4382 goto out;
4383 list_add(&lo->lo_list, &matches);
4384 }
4385 }
4386 /* Clients probably won't expect us to return with some (but not all)
4387 * of the lockowner state released; so don't release any until all
4388 * have been checked. */
4389 status = nfs_ok;
4390 while (!list_empty(&matches)) {
4391 lo = list_entry(matches.next, struct nfs4_lockowner,
4392 lo_list);
4393 /* unhash_stateowner deletes so_perclient only
4394 * for openowners. */
4395 list_del(&lo->lo_list);
4396 release_lockowner(lo);
4397 }
4398out:
4399 nfs4_unlock_state();
4400 return status;
4401}
4402
4403static inline struct nfs4_client_reclaim *
4404alloc_reclaim(void)
4405{
4406 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4407}
4408
4409int
4410nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4411{
4412 unsigned int strhashval = clientstr_hashval(name);
4413 struct nfs4_client *clp;
4414
4415 clp = find_confirmed_client_by_str(name, strhashval);
4416 if (!clp)
4417 return 0;
4418 return test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
4419}
4420
4421/*
4422 * failure => all reset bets are off, nfserr_no_grace...
4423 */
4424int
4425nfs4_client_to_reclaim(const char *name)
4426{
4427 unsigned int strhashval;
4428 struct nfs4_client_reclaim *crp = NULL;
4429
4430 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4431 crp = alloc_reclaim();
4432 if (!crp)
4433 return 0;
4434 strhashval = clientstr_hashval(name);
4435 INIT_LIST_HEAD(&crp->cr_strhash);
4436 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4437 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4438 reclaim_str_hashtbl_size++;
4439 return 1;
4440}
4441
4442void
4443nfs4_release_reclaim(void)
4444{
4445 struct nfs4_client_reclaim *crp = NULL;
4446 int i;
4447
4448 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4449 while (!list_empty(&reclaim_str_hashtbl[i])) {
4450 crp = list_entry(reclaim_str_hashtbl[i].next,
4451 struct nfs4_client_reclaim, cr_strhash);
4452 list_del(&crp->cr_strhash);
4453 kfree(crp);
4454 reclaim_str_hashtbl_size--;
4455 }
4456 }
4457 BUG_ON(reclaim_str_hashtbl_size);
4458}
4459
4460/*
4461 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4462struct nfs4_client_reclaim *
4463nfsd4_find_reclaim_client(struct nfs4_client *clp)
4464{
4465 unsigned int strhashval;
4466 struct nfs4_client_reclaim *crp = NULL;
4467
4468 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4469 clp->cl_name.len, clp->cl_name.data,
4470 clp->cl_recdir);
4471
4472 /* find clp->cl_name in reclaim_str_hashtbl */
4473 strhashval = clientstr_hashval(clp->cl_recdir);
4474 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4475 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4476 return crp;
4477 }
4478 }
4479 return NULL;
4480}
4481
4482/*
4483* Called from OPEN. Look for clientid in reclaim list.
4484*/
4485__be32
4486nfs4_check_open_reclaim(clientid_t *clid)
4487{
4488 struct nfs4_client *clp;
4489
4490 /* find clientid in conf_id_hashtbl */
4491 clp = find_confirmed_client(clid);
4492 if (clp == NULL)
4493 return nfserr_reclaim_bad;
4494
4495 return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4496}
4497
4498#ifdef CONFIG_NFSD_FAULT_INJECTION
4499
4500void nfsd_forget_clients(u64 num)
4501{
4502 struct nfs4_client *clp, *next;
4503 int count = 0;
4504
4505 nfs4_lock_state();
4506 list_for_each_entry_safe(clp, next, &client_lru, cl_lru) {
4507 nfsd4_client_record_remove(clp);
4508 expire_client(clp);
4509 if (++count == num)
4510 break;
4511 }
4512 nfs4_unlock_state();
4513
4514 printk(KERN_INFO "NFSD: Forgot %d clients", count);
4515}
4516
4517static void release_lockowner_sop(struct nfs4_stateowner *sop)
4518{
4519 release_lockowner(lockowner(sop));
4520}
4521
4522static void release_openowner_sop(struct nfs4_stateowner *sop)
4523{
4524 release_openowner(openowner(sop));
4525}
4526
4527static int nfsd_release_n_owners(u64 num, bool is_open_owner,
4528 void (*release_sop)(struct nfs4_stateowner *))
4529{
4530 int i, count = 0;
4531 struct nfs4_stateowner *sop, *next;
4532
4533 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4534 list_for_each_entry_safe(sop, next, &ownerstr_hashtbl[i], so_strhash) {
4535 if (sop->so_is_open_owner != is_open_owner)
4536 continue;
4537 release_sop(sop);
4538 if (++count == num)
4539 return count;
4540 }
4541 }
4542 return count;
4543}
4544
4545void nfsd_forget_locks(u64 num)
4546{
4547 int count;
4548
4549 nfs4_lock_state();
4550 count = nfsd_release_n_owners(num, false, release_lockowner_sop);
4551 nfs4_unlock_state();
4552
4553 printk(KERN_INFO "NFSD: Forgot %d locks", count);
4554}
4555
4556void nfsd_forget_openowners(u64 num)
4557{
4558 int count;
4559
4560 nfs4_lock_state();
4561 count = nfsd_release_n_owners(num, true, release_openowner_sop);
4562 nfs4_unlock_state();
4563
4564 printk(KERN_INFO "NFSD: Forgot %d open owners", count);
4565}
4566
4567int nfsd_process_n_delegations(u64 num, void (*deleg_func)(struct nfs4_delegation *))
4568{
4569 int i, count = 0;
4570 struct nfs4_file *fp, *fnext;
4571 struct nfs4_delegation *dp, *dnext;
4572
4573 for (i = 0; i < FILE_HASH_SIZE; i++) {
4574 list_for_each_entry_safe(fp, fnext, &file_hashtbl[i], fi_hash) {
4575 list_for_each_entry_safe(dp, dnext, &fp->fi_delegations, dl_perfile) {
4576 deleg_func(dp);
4577 if (++count == num)
4578 return count;
4579 }
4580 }
4581 }
4582
4583 return count;
4584}
4585
4586void nfsd_forget_delegations(u64 num)
4587{
4588 unsigned int count;
4589
4590 nfs4_lock_state();
4591 count = nfsd_process_n_delegations(num, unhash_delegation);
4592 nfs4_unlock_state();
4593
4594 printk(KERN_INFO "NFSD: Forgot %d delegations", count);
4595}
4596
4597void nfsd_recall_delegations(u64 num)
4598{
4599 unsigned int count;
4600
4601 nfs4_lock_state();
4602 spin_lock(&recall_lock);
4603 count = nfsd_process_n_delegations(num, nfsd_break_one_deleg);
4604 spin_unlock(&recall_lock);
4605 nfs4_unlock_state();
4606
4607 printk(KERN_INFO "NFSD: Recalled %d delegations", count);
4608}
4609
4610#endif /* CONFIG_NFSD_FAULT_INJECTION */
4611
4612/* initialization to perform at module load time: */
4613
4614void
4615nfs4_state_init(void)
4616{
4617 int i;
4618
4619 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4620 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4621 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4622 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4623 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4624 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4625 }
4626 for (i = 0; i < SESSION_HASH_SIZE; i++)
4627 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4628 for (i = 0; i < FILE_HASH_SIZE; i++) {
4629 INIT_LIST_HEAD(&file_hashtbl[i]);
4630 }
4631 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4632 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4633 }
4634 for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4635 INIT_LIST_HEAD(&lockowner_ino_hashtbl[i]);
4636 INIT_LIST_HEAD(&close_lru);
4637 INIT_LIST_HEAD(&client_lru);
4638 INIT_LIST_HEAD(&del_recall_lru);
4639 reclaim_str_hashtbl_size = 0;
4640}
4641
4642/*
4643 * Since the lifetime of a delegation isn't limited to that of an open, a
4644 * client may quite reasonably hang on to a delegation as long as it has
4645 * the inode cached. This becomes an obvious problem the first time a
4646 * client's inode cache approaches the size of the server's total memory.
4647 *
4648 * For now we avoid this problem by imposing a hard limit on the number
4649 * of delegations, which varies according to the server's memory size.
4650 */
4651static void
4652set_max_delegations(void)
4653{
4654 /*
4655 * Allow at most 4 delegations per megabyte of RAM. Quick
4656 * estimates suggest that in the worst case (where every delegation
4657 * is for a different inode), a delegation could take about 1.5K,
4658 * giving a worst case usage of about 6% of memory.
4659 */
4660 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4661}
4662
4663/* initialization to perform when the nfsd service is started: */
4664
4665int
4666nfs4_state_start(void)
4667{
4668 int ret;
4669
4670 /*
4671 * FIXME: For now, we hang most of the pernet global stuff off of
4672 * init_net until nfsd is fully containerized. Eventually, we'll
4673 * need to pass a net pointer into this function, take a reference
4674 * to that instead and then do most of the rest of this on a per-net
4675 * basis.
4676 */
4677 get_net(&init_net);
4678 nfsd4_client_tracking_init(&init_net);
4679 boot_time = get_seconds();
4680 locks_start_grace(&nfsd4_manager);
4681 grace_ended = false;
4682 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4683 nfsd4_grace);
4684 ret = set_callback_cred();
4685 if (ret) {
4686 ret = -ENOMEM;
4687 goto out_recovery;
4688 }
4689 laundry_wq = create_singlethread_workqueue("nfsd4");
4690 if (laundry_wq == NULL) {
4691 ret = -ENOMEM;
4692 goto out_recovery;
4693 }
4694 ret = nfsd4_create_callback_queue();
4695 if (ret)
4696 goto out_free_laundry;
4697 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4698 set_max_delegations();
4699 return 0;
4700out_free_laundry:
4701 destroy_workqueue(laundry_wq);
4702out_recovery:
4703 nfsd4_client_tracking_exit(&init_net);
4704 put_net(&init_net);
4705 return ret;
4706}
4707
4708static void
4709__nfs4_state_shutdown(void)
4710{
4711 int i;
4712 struct nfs4_client *clp = NULL;
4713 struct nfs4_delegation *dp = NULL;
4714 struct list_head *pos, *next, reaplist;
4715
4716 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4717 while (!list_empty(&conf_id_hashtbl[i])) {
4718 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4719 expire_client(clp);
4720 }
4721 while (!list_empty(&unconf_str_hashtbl[i])) {
4722 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4723 expire_client(clp);
4724 }
4725 }
4726 INIT_LIST_HEAD(&reaplist);
4727 spin_lock(&recall_lock);
4728 list_for_each_safe(pos, next, &del_recall_lru) {
4729 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4730 list_move(&dp->dl_recall_lru, &reaplist);
4731 }
4732 spin_unlock(&recall_lock);
4733 list_for_each_safe(pos, next, &reaplist) {
4734 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4735 unhash_delegation(dp);
4736 }
4737
4738 nfsd4_client_tracking_exit(&init_net);
4739 put_net(&init_net);
4740}
4741
4742void
4743nfs4_state_shutdown(void)
4744{
4745 cancel_delayed_work_sync(&laundromat_work);
4746 destroy_workqueue(laundry_wq);
4747 locks_end_grace(&nfsd4_manager);
4748 nfs4_lock_state();
4749 __nfs4_state_shutdown();
4750 nfs4_unlock_state();
4751 nfsd4_destroy_callback_queue();
4752}
4753
4754static void
4755get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4756{
4757 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4758 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4759}
4760
4761static void
4762put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4763{
4764 if (cstate->minorversion) {
4765 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4766 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4767 }
4768}
4769
4770void
4771clear_current_stateid(struct nfsd4_compound_state *cstate)
4772{
4773 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4774}
4775
4776/*
4777 * functions to set current state id
4778 */
4779void
4780nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4781{
4782 put_stateid(cstate, &odp->od_stateid);
4783}
4784
4785void
4786nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
4787{
4788 put_stateid(cstate, &open->op_stateid);
4789}
4790
4791void
4792nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4793{
4794 put_stateid(cstate, &close->cl_stateid);
4795}
4796
4797void
4798nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
4799{
4800 put_stateid(cstate, &lock->lk_resp_stateid);
4801}
4802
4803/*
4804 * functions to consume current state id
4805 */
4806
4807void
4808nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4809{
4810 get_stateid(cstate, &odp->od_stateid);
4811}
4812
4813void
4814nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
4815{
4816 get_stateid(cstate, &drp->dr_stateid);
4817}
4818
4819void
4820nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
4821{
4822 get_stateid(cstate, &fsp->fr_stateid);
4823}
4824
4825void
4826nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
4827{
4828 get_stateid(cstate, &setattr->sa_stateid);
4829}
4830
4831void
4832nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4833{
4834 get_stateid(cstate, &close->cl_stateid);
4835}
4836
4837void
4838nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
4839{
4840 get_stateid(cstate, &locku->lu_stateid);
4841}
4842
4843void
4844nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
4845{
4846 get_stateid(cstate, &read->rd_stateid);
4847}
4848
4849void
4850nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
4851{
4852 get_stateid(cstate, &write->wr_stateid);
4853}
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 <linux/fsnotify.h>
47#include <linux/rhashtable.h>
48#include <linux/nfs_ssc.h>
49
50#include "xdr4.h"
51#include "xdr4cb.h"
52#include "vfs.h"
53#include "current_stateid.h"
54
55#include "netns.h"
56#include "pnfs.h"
57#include "filecache.h"
58#include "trace.h"
59
60#define NFSDDBG_FACILITY NFSDDBG_PROC
61
62#define all_ones {{ ~0, ~0}, ~0}
63static const stateid_t one_stateid = {
64 .si_generation = ~0,
65 .si_opaque = all_ones,
66};
67static const stateid_t zero_stateid = {
68 /* all fields zero */
69};
70static const stateid_t currentstateid = {
71 .si_generation = 1,
72};
73static const stateid_t close_stateid = {
74 .si_generation = 0xffffffffU,
75};
76
77static u64 current_sessionid = 1;
78
79#define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
80#define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
81#define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
82#define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
83
84/* forward declarations */
85static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
86static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
87void nfsd4_end_grace(struct nfsd_net *nn);
88static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
89static void nfsd4_file_hash_remove(struct nfs4_file *fi);
90
91/* Locking: */
92
93/*
94 * Currently used for the del_recall_lru and file hash table. In an
95 * effort to decrease the scope of the client_mutex, this spinlock may
96 * eventually cover more:
97 */
98static DEFINE_SPINLOCK(state_lock);
99
100enum nfsd4_st_mutex_lock_subclass {
101 OPEN_STATEID_MUTEX = 0,
102 LOCK_STATEID_MUTEX = 1,
103};
104
105/*
106 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
107 * the refcount on the open stateid to drop.
108 */
109static DECLARE_WAIT_QUEUE_HEAD(close_wq);
110
111/*
112 * A waitqueue where a writer to clients/#/ctl destroying a client can
113 * wait for cl_rpc_users to drop to 0 and then for the client to be
114 * unhashed.
115 */
116static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
117
118static struct kmem_cache *client_slab;
119static struct kmem_cache *openowner_slab;
120static struct kmem_cache *lockowner_slab;
121static struct kmem_cache *file_slab;
122static struct kmem_cache *stateid_slab;
123static struct kmem_cache *deleg_slab;
124static struct kmem_cache *odstate_slab;
125
126static void free_session(struct nfsd4_session *);
127
128static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
129static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
130
131static struct workqueue_struct *laundry_wq;
132
133int nfsd4_create_laundry_wq(void)
134{
135 int rc = 0;
136
137 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
138 if (laundry_wq == NULL)
139 rc = -ENOMEM;
140 return rc;
141}
142
143void nfsd4_destroy_laundry_wq(void)
144{
145 destroy_workqueue(laundry_wq);
146}
147
148static bool is_session_dead(struct nfsd4_session *ses)
149{
150 return ses->se_flags & NFS4_SESSION_DEAD;
151}
152
153static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
154{
155 if (atomic_read(&ses->se_ref) > ref_held_by_me)
156 return nfserr_jukebox;
157 ses->se_flags |= NFS4_SESSION_DEAD;
158 return nfs_ok;
159}
160
161static bool is_client_expired(struct nfs4_client *clp)
162{
163 return clp->cl_time == 0;
164}
165
166static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
167 struct nfs4_client *clp)
168{
169 if (clp->cl_state != NFSD4_ACTIVE)
170 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
171}
172
173static __be32 get_client_locked(struct nfs4_client *clp)
174{
175 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
176
177 lockdep_assert_held(&nn->client_lock);
178
179 if (is_client_expired(clp))
180 return nfserr_expired;
181 atomic_inc(&clp->cl_rpc_users);
182 nfsd4_dec_courtesy_client_count(nn, clp);
183 clp->cl_state = NFSD4_ACTIVE;
184 return nfs_ok;
185}
186
187/* must be called under the client_lock */
188static inline void
189renew_client_locked(struct nfs4_client *clp)
190{
191 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
192
193 if (is_client_expired(clp)) {
194 WARN_ON(1);
195 printk("%s: client (clientid %08x/%08x) already expired\n",
196 __func__,
197 clp->cl_clientid.cl_boot,
198 clp->cl_clientid.cl_id);
199 return;
200 }
201
202 list_move_tail(&clp->cl_lru, &nn->client_lru);
203 clp->cl_time = ktime_get_boottime_seconds();
204 nfsd4_dec_courtesy_client_count(nn, clp);
205 clp->cl_state = NFSD4_ACTIVE;
206}
207
208static void put_client_renew_locked(struct nfs4_client *clp)
209{
210 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
211
212 lockdep_assert_held(&nn->client_lock);
213
214 if (!atomic_dec_and_test(&clp->cl_rpc_users))
215 return;
216 if (!is_client_expired(clp))
217 renew_client_locked(clp);
218 else
219 wake_up_all(&expiry_wq);
220}
221
222static void put_client_renew(struct nfs4_client *clp)
223{
224 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
225
226 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
227 return;
228 if (!is_client_expired(clp))
229 renew_client_locked(clp);
230 else
231 wake_up_all(&expiry_wq);
232 spin_unlock(&nn->client_lock);
233}
234
235static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
236{
237 __be32 status;
238
239 if (is_session_dead(ses))
240 return nfserr_badsession;
241 status = get_client_locked(ses->se_client);
242 if (status)
243 return status;
244 atomic_inc(&ses->se_ref);
245 return nfs_ok;
246}
247
248static void nfsd4_put_session_locked(struct nfsd4_session *ses)
249{
250 struct nfs4_client *clp = ses->se_client;
251 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
252
253 lockdep_assert_held(&nn->client_lock);
254
255 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
256 free_session(ses);
257 put_client_renew_locked(clp);
258}
259
260static void nfsd4_put_session(struct nfsd4_session *ses)
261{
262 struct nfs4_client *clp = ses->se_client;
263 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
264
265 spin_lock(&nn->client_lock);
266 nfsd4_put_session_locked(ses);
267 spin_unlock(&nn->client_lock);
268}
269
270static struct nfsd4_blocked_lock *
271find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
272 struct nfsd_net *nn)
273{
274 struct nfsd4_blocked_lock *cur, *found = NULL;
275
276 spin_lock(&nn->blocked_locks_lock);
277 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
278 if (fh_match(fh, &cur->nbl_fh)) {
279 list_del_init(&cur->nbl_list);
280 WARN_ON(list_empty(&cur->nbl_lru));
281 list_del_init(&cur->nbl_lru);
282 found = cur;
283 break;
284 }
285 }
286 spin_unlock(&nn->blocked_locks_lock);
287 if (found)
288 locks_delete_block(&found->nbl_lock);
289 return found;
290}
291
292static struct nfsd4_blocked_lock *
293find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
294 struct nfsd_net *nn)
295{
296 struct nfsd4_blocked_lock *nbl;
297
298 nbl = find_blocked_lock(lo, fh, nn);
299 if (!nbl) {
300 nbl = kmalloc(sizeof(*nbl), GFP_KERNEL);
301 if (nbl) {
302 INIT_LIST_HEAD(&nbl->nbl_list);
303 INIT_LIST_HEAD(&nbl->nbl_lru);
304 fh_copy_shallow(&nbl->nbl_fh, fh);
305 locks_init_lock(&nbl->nbl_lock);
306 kref_init(&nbl->nbl_kref);
307 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
308 &nfsd4_cb_notify_lock_ops,
309 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
310 }
311 }
312 return nbl;
313}
314
315static void
316free_nbl(struct kref *kref)
317{
318 struct nfsd4_blocked_lock *nbl;
319
320 nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
321 kfree(nbl);
322}
323
324static void
325free_blocked_lock(struct nfsd4_blocked_lock *nbl)
326{
327 locks_delete_block(&nbl->nbl_lock);
328 locks_release_private(&nbl->nbl_lock);
329 kref_put(&nbl->nbl_kref, free_nbl);
330}
331
332static void
333remove_blocked_locks(struct nfs4_lockowner *lo)
334{
335 struct nfs4_client *clp = lo->lo_owner.so_client;
336 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
337 struct nfsd4_blocked_lock *nbl;
338 LIST_HEAD(reaplist);
339
340 /* Dequeue all blocked locks */
341 spin_lock(&nn->blocked_locks_lock);
342 while (!list_empty(&lo->lo_blocked)) {
343 nbl = list_first_entry(&lo->lo_blocked,
344 struct nfsd4_blocked_lock,
345 nbl_list);
346 list_del_init(&nbl->nbl_list);
347 WARN_ON(list_empty(&nbl->nbl_lru));
348 list_move(&nbl->nbl_lru, &reaplist);
349 }
350 spin_unlock(&nn->blocked_locks_lock);
351
352 /* Now free them */
353 while (!list_empty(&reaplist)) {
354 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
355 nbl_lru);
356 list_del_init(&nbl->nbl_lru);
357 free_blocked_lock(nbl);
358 }
359}
360
361static void
362nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
363{
364 struct nfsd4_blocked_lock *nbl = container_of(cb,
365 struct nfsd4_blocked_lock, nbl_cb);
366 locks_delete_block(&nbl->nbl_lock);
367}
368
369static int
370nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
371{
372 trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
373
374 /*
375 * Since this is just an optimization, we don't try very hard if it
376 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
377 * just quit trying on anything else.
378 */
379 switch (task->tk_status) {
380 case -NFS4ERR_DELAY:
381 rpc_delay(task, 1 * HZ);
382 return 0;
383 default:
384 return 1;
385 }
386}
387
388static void
389nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
390{
391 struct nfsd4_blocked_lock *nbl = container_of(cb,
392 struct nfsd4_blocked_lock, nbl_cb);
393
394 free_blocked_lock(nbl);
395}
396
397static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
398 .prepare = nfsd4_cb_notify_lock_prepare,
399 .done = nfsd4_cb_notify_lock_done,
400 .release = nfsd4_cb_notify_lock_release,
401};
402
403/*
404 * We store the NONE, READ, WRITE, and BOTH bits separately in the
405 * st_{access,deny}_bmap field of the stateid, in order to track not
406 * only what share bits are currently in force, but also what
407 * combinations of share bits previous opens have used. This allows us
408 * to enforce the recommendation in
409 * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
410 * the server return an error if the client attempt to downgrade to a
411 * combination of share bits not explicable by closing some of its
412 * previous opens.
413 *
414 * This enforcement is arguably incomplete, since we don't keep
415 * track of access/deny bit combinations; so, e.g., we allow:
416 *
417 * OPEN allow read, deny write
418 * OPEN allow both, deny none
419 * DOWNGRADE allow read, deny none
420 *
421 * which we should reject.
422 *
423 * But you could also argue that our current code is already overkill,
424 * since it only exists to return NFS4ERR_INVAL on incorrect client
425 * behavior.
426 */
427static unsigned int
428bmap_to_share_mode(unsigned long bmap)
429{
430 int i;
431 unsigned int access = 0;
432
433 for (i = 1; i < 4; i++) {
434 if (test_bit(i, &bmap))
435 access |= i;
436 }
437 return access;
438}
439
440/* set share access for a given stateid */
441static inline void
442set_access(u32 access, struct nfs4_ol_stateid *stp)
443{
444 unsigned char mask = 1 << access;
445
446 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
447 stp->st_access_bmap |= mask;
448}
449
450/* clear share access for a given stateid */
451static inline void
452clear_access(u32 access, struct nfs4_ol_stateid *stp)
453{
454 unsigned char mask = 1 << access;
455
456 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
457 stp->st_access_bmap &= ~mask;
458}
459
460/* test whether a given stateid has access */
461static inline bool
462test_access(u32 access, struct nfs4_ol_stateid *stp)
463{
464 unsigned char mask = 1 << access;
465
466 return (bool)(stp->st_access_bmap & mask);
467}
468
469/* set share deny for a given stateid */
470static inline void
471set_deny(u32 deny, struct nfs4_ol_stateid *stp)
472{
473 unsigned char mask = 1 << deny;
474
475 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
476 stp->st_deny_bmap |= mask;
477}
478
479/* clear share deny for a given stateid */
480static inline void
481clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
482{
483 unsigned char mask = 1 << deny;
484
485 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
486 stp->st_deny_bmap &= ~mask;
487}
488
489/* test whether a given stateid is denying specific access */
490static inline bool
491test_deny(u32 deny, struct nfs4_ol_stateid *stp)
492{
493 unsigned char mask = 1 << deny;
494
495 return (bool)(stp->st_deny_bmap & mask);
496}
497
498static int nfs4_access_to_omode(u32 access)
499{
500 switch (access & NFS4_SHARE_ACCESS_BOTH) {
501 case NFS4_SHARE_ACCESS_READ:
502 return O_RDONLY;
503 case NFS4_SHARE_ACCESS_WRITE:
504 return O_WRONLY;
505 case NFS4_SHARE_ACCESS_BOTH:
506 return O_RDWR;
507 }
508 WARN_ON_ONCE(1);
509 return O_RDONLY;
510}
511
512static inline int
513access_permit_read(struct nfs4_ol_stateid *stp)
514{
515 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
516 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
517 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
518}
519
520static inline int
521access_permit_write(struct nfs4_ol_stateid *stp)
522{
523 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
524 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
525}
526
527static inline struct nfs4_stateowner *
528nfs4_get_stateowner(struct nfs4_stateowner *sop)
529{
530 atomic_inc(&sop->so_count);
531 return sop;
532}
533
534static int
535same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
536{
537 return (sop->so_owner.len == owner->len) &&
538 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
539}
540
541static struct nfs4_openowner *
542find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
543 struct nfs4_client *clp)
544{
545 struct nfs4_stateowner *so;
546
547 lockdep_assert_held(&clp->cl_lock);
548
549 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
550 so_strhash) {
551 if (!so->so_is_open_owner)
552 continue;
553 if (same_owner_str(so, &open->op_owner))
554 return openowner(nfs4_get_stateowner(so));
555 }
556 return NULL;
557}
558
559static struct nfs4_openowner *
560find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
561 struct nfs4_client *clp)
562{
563 struct nfs4_openowner *oo;
564
565 spin_lock(&clp->cl_lock);
566 oo = find_openstateowner_str_locked(hashval, open, clp);
567 spin_unlock(&clp->cl_lock);
568 return oo;
569}
570
571static inline u32
572opaque_hashval(const void *ptr, int nbytes)
573{
574 unsigned char *cptr = (unsigned char *) ptr;
575
576 u32 x = 0;
577 while (nbytes--) {
578 x *= 37;
579 x += *cptr++;
580 }
581 return x;
582}
583
584static void nfsd4_free_file_rcu(struct rcu_head *rcu)
585{
586 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
587
588 kmem_cache_free(file_slab, fp);
589}
590
591void
592put_nfs4_file(struct nfs4_file *fi)
593{
594 if (refcount_dec_and_test(&fi->fi_ref)) {
595 nfsd4_file_hash_remove(fi);
596 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
597 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
598 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
599 }
600}
601
602static struct nfsd_file *
603find_writeable_file_locked(struct nfs4_file *f)
604{
605 struct nfsd_file *ret;
606
607 lockdep_assert_held(&f->fi_lock);
608
609 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
610 if (!ret)
611 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
612 return ret;
613}
614
615static struct nfsd_file *
616find_writeable_file(struct nfs4_file *f)
617{
618 struct nfsd_file *ret;
619
620 spin_lock(&f->fi_lock);
621 ret = find_writeable_file_locked(f);
622 spin_unlock(&f->fi_lock);
623
624 return ret;
625}
626
627static struct nfsd_file *
628find_readable_file_locked(struct nfs4_file *f)
629{
630 struct nfsd_file *ret;
631
632 lockdep_assert_held(&f->fi_lock);
633
634 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
635 if (!ret)
636 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
637 return ret;
638}
639
640static struct nfsd_file *
641find_readable_file(struct nfs4_file *f)
642{
643 struct nfsd_file *ret;
644
645 spin_lock(&f->fi_lock);
646 ret = find_readable_file_locked(f);
647 spin_unlock(&f->fi_lock);
648
649 return ret;
650}
651
652static struct nfsd_file *
653find_rw_file(struct nfs4_file *f)
654{
655 struct nfsd_file *ret;
656
657 spin_lock(&f->fi_lock);
658 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
659 spin_unlock(&f->fi_lock);
660
661 return ret;
662}
663
664struct nfsd_file *
665find_any_file(struct nfs4_file *f)
666{
667 struct nfsd_file *ret;
668
669 if (!f)
670 return NULL;
671 spin_lock(&f->fi_lock);
672 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
673 if (!ret) {
674 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
675 if (!ret)
676 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
677 }
678 spin_unlock(&f->fi_lock);
679 return ret;
680}
681
682static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
683{
684 lockdep_assert_held(&f->fi_lock);
685
686 if (f->fi_fds[O_RDWR])
687 return f->fi_fds[O_RDWR];
688 if (f->fi_fds[O_WRONLY])
689 return f->fi_fds[O_WRONLY];
690 if (f->fi_fds[O_RDONLY])
691 return f->fi_fds[O_RDONLY];
692 return NULL;
693}
694
695static atomic_long_t num_delegations;
696unsigned long max_delegations;
697
698/*
699 * Open owner state (share locks)
700 */
701
702/* hash tables for lock and open owners */
703#define OWNER_HASH_BITS 8
704#define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
705#define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
706
707static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
708{
709 unsigned int ret;
710
711 ret = opaque_hashval(ownername->data, ownername->len);
712 return ret & OWNER_HASH_MASK;
713}
714
715static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
716
717static const struct rhashtable_params nfs4_file_rhash_params = {
718 .key_len = sizeof_field(struct nfs4_file, fi_inode),
719 .key_offset = offsetof(struct nfs4_file, fi_inode),
720 .head_offset = offsetof(struct nfs4_file, fi_rlist),
721
722 /*
723 * Start with a single page hash table to reduce resizing churn
724 * on light workloads.
725 */
726 .min_size = 256,
727 .automatic_shrinking = true,
728};
729
730/*
731 * Check if courtesy clients have conflicting access and resolve it if possible
732 *
733 * access: is op_share_access if share_access is true.
734 * Check if access mode, op_share_access, would conflict with
735 * the current deny mode of the file 'fp'.
736 * access: is op_share_deny if share_access is false.
737 * Check if the deny mode, op_share_deny, would conflict with
738 * current access of the file 'fp'.
739 * stp: skip checking this entry.
740 * new_stp: normal open, not open upgrade.
741 *
742 * Function returns:
743 * false - access/deny mode conflict with normal client.
744 * true - no conflict or conflict with courtesy client(s) is resolved.
745 */
746static bool
747nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
748 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
749{
750 struct nfs4_ol_stateid *st;
751 bool resolvable = true;
752 unsigned char bmap;
753 struct nfsd_net *nn;
754 struct nfs4_client *clp;
755
756 lockdep_assert_held(&fp->fi_lock);
757 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
758 /* ignore lock stateid */
759 if (st->st_openstp)
760 continue;
761 if (st == stp && new_stp)
762 continue;
763 /* check file access against deny mode or vice versa */
764 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
765 if (!(access & bmap_to_share_mode(bmap)))
766 continue;
767 clp = st->st_stid.sc_client;
768 if (try_to_expire_client(clp))
769 continue;
770 resolvable = false;
771 break;
772 }
773 if (resolvable) {
774 clp = stp->st_stid.sc_client;
775 nn = net_generic(clp->net, nfsd_net_id);
776 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
777 }
778 return resolvable;
779}
780
781static void
782__nfs4_file_get_access(struct nfs4_file *fp, u32 access)
783{
784 lockdep_assert_held(&fp->fi_lock);
785
786 if (access & NFS4_SHARE_ACCESS_WRITE)
787 atomic_inc(&fp->fi_access[O_WRONLY]);
788 if (access & NFS4_SHARE_ACCESS_READ)
789 atomic_inc(&fp->fi_access[O_RDONLY]);
790}
791
792static __be32
793nfs4_file_get_access(struct nfs4_file *fp, u32 access)
794{
795 lockdep_assert_held(&fp->fi_lock);
796
797 /* Does this access mode make sense? */
798 if (access & ~NFS4_SHARE_ACCESS_BOTH)
799 return nfserr_inval;
800
801 /* Does it conflict with a deny mode already set? */
802 if ((access & fp->fi_share_deny) != 0)
803 return nfserr_share_denied;
804
805 __nfs4_file_get_access(fp, access);
806 return nfs_ok;
807}
808
809static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
810{
811 /* Common case is that there is no deny mode. */
812 if (deny) {
813 /* Does this deny mode make sense? */
814 if (deny & ~NFS4_SHARE_DENY_BOTH)
815 return nfserr_inval;
816
817 if ((deny & NFS4_SHARE_DENY_READ) &&
818 atomic_read(&fp->fi_access[O_RDONLY]))
819 return nfserr_share_denied;
820
821 if ((deny & NFS4_SHARE_DENY_WRITE) &&
822 atomic_read(&fp->fi_access[O_WRONLY]))
823 return nfserr_share_denied;
824 }
825 return nfs_ok;
826}
827
828static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
829{
830 might_lock(&fp->fi_lock);
831
832 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
833 struct nfsd_file *f1 = NULL;
834 struct nfsd_file *f2 = NULL;
835
836 swap(f1, fp->fi_fds[oflag]);
837 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
838 swap(f2, fp->fi_fds[O_RDWR]);
839 spin_unlock(&fp->fi_lock);
840 if (f1)
841 nfsd_file_put(f1);
842 if (f2)
843 nfsd_file_put(f2);
844 }
845}
846
847static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
848{
849 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
850
851 if (access & NFS4_SHARE_ACCESS_WRITE)
852 __nfs4_file_put_access(fp, O_WRONLY);
853 if (access & NFS4_SHARE_ACCESS_READ)
854 __nfs4_file_put_access(fp, O_RDONLY);
855}
856
857/*
858 * Allocate a new open/delegation state counter. This is needed for
859 * pNFS for proper return on close semantics.
860 *
861 * Note that we only allocate it for pNFS-enabled exports, otherwise
862 * all pointers to struct nfs4_clnt_odstate are always NULL.
863 */
864static struct nfs4_clnt_odstate *
865alloc_clnt_odstate(struct nfs4_client *clp)
866{
867 struct nfs4_clnt_odstate *co;
868
869 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
870 if (co) {
871 co->co_client = clp;
872 refcount_set(&co->co_odcount, 1);
873 }
874 return co;
875}
876
877static void
878hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
879{
880 struct nfs4_file *fp = co->co_file;
881
882 lockdep_assert_held(&fp->fi_lock);
883 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
884}
885
886static inline void
887get_clnt_odstate(struct nfs4_clnt_odstate *co)
888{
889 if (co)
890 refcount_inc(&co->co_odcount);
891}
892
893static void
894put_clnt_odstate(struct nfs4_clnt_odstate *co)
895{
896 struct nfs4_file *fp;
897
898 if (!co)
899 return;
900
901 fp = co->co_file;
902 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
903 list_del(&co->co_perfile);
904 spin_unlock(&fp->fi_lock);
905
906 nfsd4_return_all_file_layouts(co->co_client, fp);
907 kmem_cache_free(odstate_slab, co);
908 }
909}
910
911static struct nfs4_clnt_odstate *
912find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
913{
914 struct nfs4_clnt_odstate *co;
915 struct nfs4_client *cl;
916
917 if (!new)
918 return NULL;
919
920 cl = new->co_client;
921
922 spin_lock(&fp->fi_lock);
923 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
924 if (co->co_client == cl) {
925 get_clnt_odstate(co);
926 goto out;
927 }
928 }
929 co = new;
930 co->co_file = fp;
931 hash_clnt_odstate_locked(new);
932out:
933 spin_unlock(&fp->fi_lock);
934 return co;
935}
936
937struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
938 void (*sc_free)(struct nfs4_stid *))
939{
940 struct nfs4_stid *stid;
941 int new_id;
942
943 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
944 if (!stid)
945 return NULL;
946
947 idr_preload(GFP_KERNEL);
948 spin_lock(&cl->cl_lock);
949 /* Reserving 0 for start of file in nfsdfs "states" file: */
950 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
951 spin_unlock(&cl->cl_lock);
952 idr_preload_end();
953 if (new_id < 0)
954 goto out_free;
955
956 stid->sc_free = sc_free;
957 stid->sc_client = cl;
958 stid->sc_stateid.si_opaque.so_id = new_id;
959 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
960 /* Will be incremented before return to client: */
961 refcount_set(&stid->sc_count, 1);
962 spin_lock_init(&stid->sc_lock);
963 INIT_LIST_HEAD(&stid->sc_cp_list);
964
965 /*
966 * It shouldn't be a problem to reuse an opaque stateid value.
967 * I don't think it is for 4.1. But with 4.0 I worry that, for
968 * example, a stray write retransmission could be accepted by
969 * the server when it should have been rejected. Therefore,
970 * adopt a trick from the sctp code to attempt to maximize the
971 * amount of time until an id is reused, by ensuring they always
972 * "increase" (mod INT_MAX):
973 */
974 return stid;
975out_free:
976 kmem_cache_free(slab, stid);
977 return NULL;
978}
979
980/*
981 * Create a unique stateid_t to represent each COPY.
982 */
983static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
984 unsigned char cs_type)
985{
986 int new_id;
987
988 stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
989 stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
990
991 idr_preload(GFP_KERNEL);
992 spin_lock(&nn->s2s_cp_lock);
993 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
994 stid->cs_stid.si_opaque.so_id = new_id;
995 stid->cs_stid.si_generation = 1;
996 spin_unlock(&nn->s2s_cp_lock);
997 idr_preload_end();
998 if (new_id < 0)
999 return 0;
1000 stid->cs_type = cs_type;
1001 return 1;
1002}
1003
1004int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1005{
1006 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
1007}
1008
1009struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1010 struct nfs4_stid *p_stid)
1011{
1012 struct nfs4_cpntf_state *cps;
1013
1014 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1015 if (!cps)
1016 return NULL;
1017 cps->cpntf_time = ktime_get_boottime_seconds();
1018 refcount_set(&cps->cp_stateid.cs_count, 1);
1019 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1020 goto out_free;
1021 spin_lock(&nn->s2s_cp_lock);
1022 list_add(&cps->cp_list, &p_stid->sc_cp_list);
1023 spin_unlock(&nn->s2s_cp_lock);
1024 return cps;
1025out_free:
1026 kfree(cps);
1027 return NULL;
1028}
1029
1030void nfs4_free_copy_state(struct nfsd4_copy *copy)
1031{
1032 struct nfsd_net *nn;
1033
1034 if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1035 return;
1036 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1037 spin_lock(&nn->s2s_cp_lock);
1038 idr_remove(&nn->s2s_cp_stateids,
1039 copy->cp_stateid.cs_stid.si_opaque.so_id);
1040 spin_unlock(&nn->s2s_cp_lock);
1041}
1042
1043static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1044{
1045 struct nfs4_cpntf_state *cps;
1046 struct nfsd_net *nn;
1047
1048 nn = net_generic(net, nfsd_net_id);
1049 spin_lock(&nn->s2s_cp_lock);
1050 while (!list_empty(&stid->sc_cp_list)) {
1051 cps = list_first_entry(&stid->sc_cp_list,
1052 struct nfs4_cpntf_state, cp_list);
1053 _free_cpntf_state_locked(nn, cps);
1054 }
1055 spin_unlock(&nn->s2s_cp_lock);
1056}
1057
1058static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1059{
1060 struct nfs4_stid *stid;
1061
1062 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1063 if (!stid)
1064 return NULL;
1065
1066 return openlockstateid(stid);
1067}
1068
1069static void nfs4_free_deleg(struct nfs4_stid *stid)
1070{
1071 struct nfs4_delegation *dp = delegstateid(stid);
1072
1073 WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1074 WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1075 WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1076 WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1077 kmem_cache_free(deleg_slab, stid);
1078 atomic_long_dec(&num_delegations);
1079}
1080
1081/*
1082 * When we recall a delegation, we should be careful not to hand it
1083 * out again straight away.
1084 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1085 * in which the filehandles of recalled delegations are "stored".
1086 * If a filehandle appear in either filter, a delegation is blocked.
1087 * When a delegation is recalled, the filehandle is stored in the "new"
1088 * filter.
1089 * Every 30 seconds we swap the filters and clear the "new" one,
1090 * unless both are empty of course.
1091 *
1092 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
1093 * low 3 bytes as hash-table indices.
1094 *
1095 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1096 * is used to manage concurrent access. Testing does not need the lock
1097 * except when swapping the two filters.
1098 */
1099static DEFINE_SPINLOCK(blocked_delegations_lock);
1100static struct bloom_pair {
1101 int entries, old_entries;
1102 time64_t swap_time;
1103 int new; /* index into 'set' */
1104 DECLARE_BITMAP(set[2], 256);
1105} blocked_delegations;
1106
1107static int delegation_blocked(struct knfsd_fh *fh)
1108{
1109 u32 hash;
1110 struct bloom_pair *bd = &blocked_delegations;
1111
1112 if (bd->entries == 0)
1113 return 0;
1114 if (ktime_get_seconds() - bd->swap_time > 30) {
1115 spin_lock(&blocked_delegations_lock);
1116 if (ktime_get_seconds() - bd->swap_time > 30) {
1117 bd->entries -= bd->old_entries;
1118 bd->old_entries = bd->entries;
1119 memset(bd->set[bd->new], 0,
1120 sizeof(bd->set[0]));
1121 bd->new = 1-bd->new;
1122 bd->swap_time = ktime_get_seconds();
1123 }
1124 spin_unlock(&blocked_delegations_lock);
1125 }
1126 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1127 if (test_bit(hash&255, bd->set[0]) &&
1128 test_bit((hash>>8)&255, bd->set[0]) &&
1129 test_bit((hash>>16)&255, bd->set[0]))
1130 return 1;
1131
1132 if (test_bit(hash&255, bd->set[1]) &&
1133 test_bit((hash>>8)&255, bd->set[1]) &&
1134 test_bit((hash>>16)&255, bd->set[1]))
1135 return 1;
1136
1137 return 0;
1138}
1139
1140static void block_delegations(struct knfsd_fh *fh)
1141{
1142 u32 hash;
1143 struct bloom_pair *bd = &blocked_delegations;
1144
1145 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1146
1147 spin_lock(&blocked_delegations_lock);
1148 __set_bit(hash&255, bd->set[bd->new]);
1149 __set_bit((hash>>8)&255, bd->set[bd->new]);
1150 __set_bit((hash>>16)&255, bd->set[bd->new]);
1151 if (bd->entries == 0)
1152 bd->swap_time = ktime_get_seconds();
1153 bd->entries += 1;
1154 spin_unlock(&blocked_delegations_lock);
1155}
1156
1157static struct nfs4_delegation *
1158alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1159 struct nfs4_clnt_odstate *odstate, u32 dl_type)
1160{
1161 struct nfs4_delegation *dp;
1162 struct nfs4_stid *stid;
1163 long n;
1164
1165 dprintk("NFSD alloc_init_deleg\n");
1166 n = atomic_long_inc_return(&num_delegations);
1167 if (n < 0 || n > max_delegations)
1168 goto out_dec;
1169 if (delegation_blocked(&fp->fi_fhandle))
1170 goto out_dec;
1171 stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1172 if (stid == NULL)
1173 goto out_dec;
1174 dp = delegstateid(stid);
1175
1176 /*
1177 * delegation seqid's are never incremented. The 4.1 special
1178 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1179 * 0 anyway just for consistency and use 1:
1180 */
1181 dp->dl_stid.sc_stateid.si_generation = 1;
1182 INIT_LIST_HEAD(&dp->dl_perfile);
1183 INIT_LIST_HEAD(&dp->dl_perclnt);
1184 INIT_LIST_HEAD(&dp->dl_recall_lru);
1185 dp->dl_clnt_odstate = odstate;
1186 get_clnt_odstate(odstate);
1187 dp->dl_type = dl_type;
1188 dp->dl_retries = 1;
1189 dp->dl_recalled = false;
1190 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1191 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1192 get_nfs4_file(fp);
1193 dp->dl_stid.sc_file = fp;
1194 return dp;
1195out_dec:
1196 atomic_long_dec(&num_delegations);
1197 return NULL;
1198}
1199
1200void
1201nfs4_put_stid(struct nfs4_stid *s)
1202{
1203 struct nfs4_file *fp = s->sc_file;
1204 struct nfs4_client *clp = s->sc_client;
1205
1206 might_lock(&clp->cl_lock);
1207
1208 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1209 wake_up_all(&close_wq);
1210 return;
1211 }
1212 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1213 nfs4_free_cpntf_statelist(clp->net, s);
1214 spin_unlock(&clp->cl_lock);
1215 s->sc_free(s);
1216 if (fp)
1217 put_nfs4_file(fp);
1218}
1219
1220void
1221nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1222{
1223 stateid_t *src = &stid->sc_stateid;
1224
1225 spin_lock(&stid->sc_lock);
1226 if (unlikely(++src->si_generation == 0))
1227 src->si_generation = 1;
1228 memcpy(dst, src, sizeof(*dst));
1229 spin_unlock(&stid->sc_lock);
1230}
1231
1232static void put_deleg_file(struct nfs4_file *fp)
1233{
1234 struct nfsd_file *nf = NULL;
1235
1236 spin_lock(&fp->fi_lock);
1237 if (--fp->fi_delegees == 0)
1238 swap(nf, fp->fi_deleg_file);
1239 spin_unlock(&fp->fi_lock);
1240
1241 if (nf)
1242 nfsd_file_put(nf);
1243}
1244
1245static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1246{
1247 struct nfs4_file *fp = dp->dl_stid.sc_file;
1248 struct nfsd_file *nf = fp->fi_deleg_file;
1249
1250 WARN_ON_ONCE(!fp->fi_delegees);
1251
1252 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1253 put_deleg_file(fp);
1254}
1255
1256static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1257{
1258 put_clnt_odstate(dp->dl_clnt_odstate);
1259 nfs4_unlock_deleg_lease(dp);
1260 nfs4_put_stid(&dp->dl_stid);
1261}
1262
1263void nfs4_unhash_stid(struct nfs4_stid *s)
1264{
1265 s->sc_type = 0;
1266}
1267
1268/**
1269 * nfs4_delegation_exists - Discover if this delegation already exists
1270 * @clp: a pointer to the nfs4_client we're granting a delegation to
1271 * @fp: a pointer to the nfs4_file we're granting a delegation on
1272 *
1273 * Return:
1274 * On success: true iff an existing delegation is found
1275 */
1276
1277static bool
1278nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1279{
1280 struct nfs4_delegation *searchdp = NULL;
1281 struct nfs4_client *searchclp = NULL;
1282
1283 lockdep_assert_held(&state_lock);
1284 lockdep_assert_held(&fp->fi_lock);
1285
1286 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1287 searchclp = searchdp->dl_stid.sc_client;
1288 if (clp == searchclp) {
1289 return true;
1290 }
1291 }
1292 return false;
1293}
1294
1295/**
1296 * hash_delegation_locked - Add a delegation to the appropriate lists
1297 * @dp: a pointer to the nfs4_delegation we are adding.
1298 * @fp: a pointer to the nfs4_file we're granting a delegation on
1299 *
1300 * Return:
1301 * On success: NULL if the delegation was successfully hashed.
1302 *
1303 * On error: -EAGAIN if one was previously granted to this
1304 * nfs4_client for this nfs4_file. Delegation is not hashed.
1305 *
1306 */
1307
1308static int
1309hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1310{
1311 struct nfs4_client *clp = dp->dl_stid.sc_client;
1312
1313 lockdep_assert_held(&state_lock);
1314 lockdep_assert_held(&fp->fi_lock);
1315
1316 if (nfs4_delegation_exists(clp, fp))
1317 return -EAGAIN;
1318 refcount_inc(&dp->dl_stid.sc_count);
1319 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1320 list_add(&dp->dl_perfile, &fp->fi_delegations);
1321 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1322 return 0;
1323}
1324
1325static bool delegation_hashed(struct nfs4_delegation *dp)
1326{
1327 return !(list_empty(&dp->dl_perfile));
1328}
1329
1330static bool
1331unhash_delegation_locked(struct nfs4_delegation *dp)
1332{
1333 struct nfs4_file *fp = dp->dl_stid.sc_file;
1334
1335 lockdep_assert_held(&state_lock);
1336
1337 if (!delegation_hashed(dp))
1338 return false;
1339
1340 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1341 /* Ensure that deleg break won't try to requeue it */
1342 ++dp->dl_time;
1343 spin_lock(&fp->fi_lock);
1344 list_del_init(&dp->dl_perclnt);
1345 list_del_init(&dp->dl_recall_lru);
1346 list_del_init(&dp->dl_perfile);
1347 spin_unlock(&fp->fi_lock);
1348 return true;
1349}
1350
1351static void destroy_delegation(struct nfs4_delegation *dp)
1352{
1353 bool unhashed;
1354
1355 spin_lock(&state_lock);
1356 unhashed = unhash_delegation_locked(dp);
1357 spin_unlock(&state_lock);
1358 if (unhashed)
1359 destroy_unhashed_deleg(dp);
1360}
1361
1362static void revoke_delegation(struct nfs4_delegation *dp)
1363{
1364 struct nfs4_client *clp = dp->dl_stid.sc_client;
1365
1366 WARN_ON(!list_empty(&dp->dl_recall_lru));
1367
1368 trace_nfsd_stid_revoke(&dp->dl_stid);
1369
1370 if (clp->cl_minorversion) {
1371 spin_lock(&clp->cl_lock);
1372 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1373 refcount_inc(&dp->dl_stid.sc_count);
1374 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1375 spin_unlock(&clp->cl_lock);
1376 }
1377 destroy_unhashed_deleg(dp);
1378}
1379
1380/*
1381 * SETCLIENTID state
1382 */
1383
1384static unsigned int clientid_hashval(u32 id)
1385{
1386 return id & CLIENT_HASH_MASK;
1387}
1388
1389static unsigned int clientstr_hashval(struct xdr_netobj name)
1390{
1391 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1392}
1393
1394/*
1395 * A stateid that had a deny mode associated with it is being released
1396 * or downgraded. Recalculate the deny mode on the file.
1397 */
1398static void
1399recalculate_deny_mode(struct nfs4_file *fp)
1400{
1401 struct nfs4_ol_stateid *stp;
1402
1403 spin_lock(&fp->fi_lock);
1404 fp->fi_share_deny = 0;
1405 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1406 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1407 spin_unlock(&fp->fi_lock);
1408}
1409
1410static void
1411reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1412{
1413 int i;
1414 bool change = false;
1415
1416 for (i = 1; i < 4; i++) {
1417 if ((i & deny) != i) {
1418 change = true;
1419 clear_deny(i, stp);
1420 }
1421 }
1422
1423 /* Recalculate per-file deny mode if there was a change */
1424 if (change)
1425 recalculate_deny_mode(stp->st_stid.sc_file);
1426}
1427
1428/* release all access and file references for a given stateid */
1429static void
1430release_all_access(struct nfs4_ol_stateid *stp)
1431{
1432 int i;
1433 struct nfs4_file *fp = stp->st_stid.sc_file;
1434
1435 if (fp && stp->st_deny_bmap != 0)
1436 recalculate_deny_mode(fp);
1437
1438 for (i = 1; i < 4; i++) {
1439 if (test_access(i, stp))
1440 nfs4_file_put_access(stp->st_stid.sc_file, i);
1441 clear_access(i, stp);
1442 }
1443}
1444
1445static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1446{
1447 kfree(sop->so_owner.data);
1448 sop->so_ops->so_free(sop);
1449}
1450
1451static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1452{
1453 struct nfs4_client *clp = sop->so_client;
1454
1455 might_lock(&clp->cl_lock);
1456
1457 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1458 return;
1459 sop->so_ops->so_unhash(sop);
1460 spin_unlock(&clp->cl_lock);
1461 nfs4_free_stateowner(sop);
1462}
1463
1464static bool
1465nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1466{
1467 return list_empty(&stp->st_perfile);
1468}
1469
1470static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1471{
1472 struct nfs4_file *fp = stp->st_stid.sc_file;
1473
1474 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1475
1476 if (list_empty(&stp->st_perfile))
1477 return false;
1478
1479 spin_lock(&fp->fi_lock);
1480 list_del_init(&stp->st_perfile);
1481 spin_unlock(&fp->fi_lock);
1482 list_del(&stp->st_perstateowner);
1483 return true;
1484}
1485
1486static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1487{
1488 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1489
1490 put_clnt_odstate(stp->st_clnt_odstate);
1491 release_all_access(stp);
1492 if (stp->st_stateowner)
1493 nfs4_put_stateowner(stp->st_stateowner);
1494 WARN_ON(!list_empty(&stid->sc_cp_list));
1495 kmem_cache_free(stateid_slab, stid);
1496}
1497
1498static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1499{
1500 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1501 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1502 struct nfsd_file *nf;
1503
1504 nf = find_any_file(stp->st_stid.sc_file);
1505 if (nf) {
1506 get_file(nf->nf_file);
1507 filp_close(nf->nf_file, (fl_owner_t)lo);
1508 nfsd_file_put(nf);
1509 }
1510 nfs4_free_ol_stateid(stid);
1511}
1512
1513/*
1514 * Put the persistent reference to an already unhashed generic stateid, while
1515 * holding the cl_lock. If it's the last reference, then put it onto the
1516 * reaplist for later destruction.
1517 */
1518static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1519 struct list_head *reaplist)
1520{
1521 struct nfs4_stid *s = &stp->st_stid;
1522 struct nfs4_client *clp = s->sc_client;
1523
1524 lockdep_assert_held(&clp->cl_lock);
1525
1526 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1527
1528 if (!refcount_dec_and_test(&s->sc_count)) {
1529 wake_up_all(&close_wq);
1530 return;
1531 }
1532
1533 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1534 list_add(&stp->st_locks, reaplist);
1535}
1536
1537static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1538{
1539 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1540
1541 if (!unhash_ol_stateid(stp))
1542 return false;
1543 list_del_init(&stp->st_locks);
1544 nfs4_unhash_stid(&stp->st_stid);
1545 return true;
1546}
1547
1548static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1549{
1550 struct nfs4_client *clp = stp->st_stid.sc_client;
1551 bool unhashed;
1552
1553 spin_lock(&clp->cl_lock);
1554 unhashed = unhash_lock_stateid(stp);
1555 spin_unlock(&clp->cl_lock);
1556 if (unhashed)
1557 nfs4_put_stid(&stp->st_stid);
1558}
1559
1560static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1561{
1562 struct nfs4_client *clp = lo->lo_owner.so_client;
1563
1564 lockdep_assert_held(&clp->cl_lock);
1565
1566 list_del_init(&lo->lo_owner.so_strhash);
1567}
1568
1569/*
1570 * Free a list of generic stateids that were collected earlier after being
1571 * fully unhashed.
1572 */
1573static void
1574free_ol_stateid_reaplist(struct list_head *reaplist)
1575{
1576 struct nfs4_ol_stateid *stp;
1577 struct nfs4_file *fp;
1578
1579 might_sleep();
1580
1581 while (!list_empty(reaplist)) {
1582 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1583 st_locks);
1584 list_del(&stp->st_locks);
1585 fp = stp->st_stid.sc_file;
1586 stp->st_stid.sc_free(&stp->st_stid);
1587 if (fp)
1588 put_nfs4_file(fp);
1589 }
1590}
1591
1592static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1593 struct list_head *reaplist)
1594{
1595 struct nfs4_ol_stateid *stp;
1596
1597 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1598
1599 while (!list_empty(&open_stp->st_locks)) {
1600 stp = list_entry(open_stp->st_locks.next,
1601 struct nfs4_ol_stateid, st_locks);
1602 WARN_ON(!unhash_lock_stateid(stp));
1603 put_ol_stateid_locked(stp, reaplist);
1604 }
1605}
1606
1607static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1608 struct list_head *reaplist)
1609{
1610 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1611
1612 if (!unhash_ol_stateid(stp))
1613 return false;
1614 release_open_stateid_locks(stp, reaplist);
1615 return true;
1616}
1617
1618static void release_open_stateid(struct nfs4_ol_stateid *stp)
1619{
1620 LIST_HEAD(reaplist);
1621
1622 spin_lock(&stp->st_stid.sc_client->cl_lock);
1623 if (unhash_open_stateid(stp, &reaplist))
1624 put_ol_stateid_locked(stp, &reaplist);
1625 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1626 free_ol_stateid_reaplist(&reaplist);
1627}
1628
1629static void unhash_openowner_locked(struct nfs4_openowner *oo)
1630{
1631 struct nfs4_client *clp = oo->oo_owner.so_client;
1632
1633 lockdep_assert_held(&clp->cl_lock);
1634
1635 list_del_init(&oo->oo_owner.so_strhash);
1636 list_del_init(&oo->oo_perclient);
1637}
1638
1639static void release_last_closed_stateid(struct nfs4_openowner *oo)
1640{
1641 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1642 nfsd_net_id);
1643 struct nfs4_ol_stateid *s;
1644
1645 spin_lock(&nn->client_lock);
1646 s = oo->oo_last_closed_stid;
1647 if (s) {
1648 list_del_init(&oo->oo_close_lru);
1649 oo->oo_last_closed_stid = NULL;
1650 }
1651 spin_unlock(&nn->client_lock);
1652 if (s)
1653 nfs4_put_stid(&s->st_stid);
1654}
1655
1656static void release_openowner(struct nfs4_openowner *oo)
1657{
1658 struct nfs4_ol_stateid *stp;
1659 struct nfs4_client *clp = oo->oo_owner.so_client;
1660 struct list_head reaplist;
1661
1662 INIT_LIST_HEAD(&reaplist);
1663
1664 spin_lock(&clp->cl_lock);
1665 unhash_openowner_locked(oo);
1666 while (!list_empty(&oo->oo_owner.so_stateids)) {
1667 stp = list_first_entry(&oo->oo_owner.so_stateids,
1668 struct nfs4_ol_stateid, st_perstateowner);
1669 if (unhash_open_stateid(stp, &reaplist))
1670 put_ol_stateid_locked(stp, &reaplist);
1671 }
1672 spin_unlock(&clp->cl_lock);
1673 free_ol_stateid_reaplist(&reaplist);
1674 release_last_closed_stateid(oo);
1675 nfs4_put_stateowner(&oo->oo_owner);
1676}
1677
1678static inline int
1679hash_sessionid(struct nfs4_sessionid *sessionid)
1680{
1681 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1682
1683 return sid->sequence % SESSION_HASH_SIZE;
1684}
1685
1686#ifdef CONFIG_SUNRPC_DEBUG
1687static inline void
1688dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1689{
1690 u32 *ptr = (u32 *)(&sessionid->data[0]);
1691 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1692}
1693#else
1694static inline void
1695dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1696{
1697}
1698#endif
1699
1700/*
1701 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1702 * won't be used for replay.
1703 */
1704void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1705{
1706 struct nfs4_stateowner *so = cstate->replay_owner;
1707
1708 if (nfserr == nfserr_replay_me)
1709 return;
1710
1711 if (!seqid_mutating_err(ntohl(nfserr))) {
1712 nfsd4_cstate_clear_replay(cstate);
1713 return;
1714 }
1715 if (!so)
1716 return;
1717 if (so->so_is_open_owner)
1718 release_last_closed_stateid(openowner(so));
1719 so->so_seqid++;
1720 return;
1721}
1722
1723static void
1724gen_sessionid(struct nfsd4_session *ses)
1725{
1726 struct nfs4_client *clp = ses->se_client;
1727 struct nfsd4_sessionid *sid;
1728
1729 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1730 sid->clientid = clp->cl_clientid;
1731 sid->sequence = current_sessionid++;
1732 sid->reserved = 0;
1733}
1734
1735/*
1736 * The protocol defines ca_maxresponssize_cached to include the size of
1737 * the rpc header, but all we need to cache is the data starting after
1738 * the end of the initial SEQUENCE operation--the rest we regenerate
1739 * each time. Therefore we can advertise a ca_maxresponssize_cached
1740 * value that is the number of bytes in our cache plus a few additional
1741 * bytes. In order to stay on the safe side, and not promise more than
1742 * we can cache, those additional bytes must be the minimum possible: 24
1743 * bytes of rpc header (xid through accept state, with AUTH_NULL
1744 * verifier), 12 for the compound header (with zero-length tag), and 44
1745 * for the SEQUENCE op response:
1746 */
1747#define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1748
1749static void
1750free_session_slots(struct nfsd4_session *ses)
1751{
1752 int i;
1753
1754 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1755 free_svc_cred(&ses->se_slots[i]->sl_cred);
1756 kfree(ses->se_slots[i]);
1757 }
1758}
1759
1760/*
1761 * We don't actually need to cache the rpc and session headers, so we
1762 * can allocate a little less for each slot:
1763 */
1764static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1765{
1766 u32 size;
1767
1768 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1769 size = 0;
1770 else
1771 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1772 return size + sizeof(struct nfsd4_slot);
1773}
1774
1775/*
1776 * XXX: If we run out of reserved DRC memory we could (up to a point)
1777 * re-negotiate active sessions and reduce their slot usage to make
1778 * room for new connections. For now we just fail the create session.
1779 */
1780static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1781{
1782 u32 slotsize = slot_bytes(ca);
1783 u32 num = ca->maxreqs;
1784 unsigned long avail, total_avail;
1785 unsigned int scale_factor;
1786
1787 spin_lock(&nfsd_drc_lock);
1788 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1789 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1790 else
1791 /* We have handed out more space than we chose in
1792 * set_max_drc() to allow. That isn't really a
1793 * problem as long as that doesn't make us think we
1794 * have lots more due to integer overflow.
1795 */
1796 total_avail = 0;
1797 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1798 /*
1799 * Never use more than a fraction of the remaining memory,
1800 * unless it's the only way to give this client a slot.
1801 * The chosen fraction is either 1/8 or 1/number of threads,
1802 * whichever is smaller. This ensures there are adequate
1803 * slots to support multiple clients per thread.
1804 * Give the client one slot even if that would require
1805 * over-allocation--it is better than failure.
1806 */
1807 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1808
1809 avail = clamp_t(unsigned long, avail, slotsize,
1810 total_avail/scale_factor);
1811 num = min_t(int, num, avail / slotsize);
1812 num = max_t(int, num, 1);
1813 nfsd_drc_mem_used += num * slotsize;
1814 spin_unlock(&nfsd_drc_lock);
1815
1816 return num;
1817}
1818
1819static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1820{
1821 int slotsize = slot_bytes(ca);
1822
1823 spin_lock(&nfsd_drc_lock);
1824 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1825 spin_unlock(&nfsd_drc_lock);
1826}
1827
1828static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1829 struct nfsd4_channel_attrs *battrs)
1830{
1831 int numslots = fattrs->maxreqs;
1832 int slotsize = slot_bytes(fattrs);
1833 struct nfsd4_session *new;
1834 int i;
1835
1836 BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1837 > PAGE_SIZE);
1838
1839 new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1840 if (!new)
1841 return NULL;
1842 /* allocate each struct nfsd4_slot and data cache in one piece */
1843 for (i = 0; i < numslots; i++) {
1844 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1845 if (!new->se_slots[i])
1846 goto out_free;
1847 }
1848
1849 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1850 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1851
1852 return new;
1853out_free:
1854 while (i--)
1855 kfree(new->se_slots[i]);
1856 kfree(new);
1857 return NULL;
1858}
1859
1860static void free_conn(struct nfsd4_conn *c)
1861{
1862 svc_xprt_put(c->cn_xprt);
1863 kfree(c);
1864}
1865
1866static void nfsd4_conn_lost(struct svc_xpt_user *u)
1867{
1868 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1869 struct nfs4_client *clp = c->cn_session->se_client;
1870
1871 trace_nfsd_cb_lost(clp);
1872
1873 spin_lock(&clp->cl_lock);
1874 if (!list_empty(&c->cn_persession)) {
1875 list_del(&c->cn_persession);
1876 free_conn(c);
1877 }
1878 nfsd4_probe_callback(clp);
1879 spin_unlock(&clp->cl_lock);
1880}
1881
1882static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1883{
1884 struct nfsd4_conn *conn;
1885
1886 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1887 if (!conn)
1888 return NULL;
1889 svc_xprt_get(rqstp->rq_xprt);
1890 conn->cn_xprt = rqstp->rq_xprt;
1891 conn->cn_flags = flags;
1892 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1893 return conn;
1894}
1895
1896static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1897{
1898 conn->cn_session = ses;
1899 list_add(&conn->cn_persession, &ses->se_conns);
1900}
1901
1902static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1903{
1904 struct nfs4_client *clp = ses->se_client;
1905
1906 spin_lock(&clp->cl_lock);
1907 __nfsd4_hash_conn(conn, ses);
1908 spin_unlock(&clp->cl_lock);
1909}
1910
1911static int nfsd4_register_conn(struct nfsd4_conn *conn)
1912{
1913 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1914 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1915}
1916
1917static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1918{
1919 int ret;
1920
1921 nfsd4_hash_conn(conn, ses);
1922 ret = nfsd4_register_conn(conn);
1923 if (ret)
1924 /* oops; xprt is already down: */
1925 nfsd4_conn_lost(&conn->cn_xpt_user);
1926 /* We may have gained or lost a callback channel: */
1927 nfsd4_probe_callback_sync(ses->se_client);
1928}
1929
1930static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1931{
1932 u32 dir = NFS4_CDFC4_FORE;
1933
1934 if (cses->flags & SESSION4_BACK_CHAN)
1935 dir |= NFS4_CDFC4_BACK;
1936 return alloc_conn(rqstp, dir);
1937}
1938
1939/* must be called under client_lock */
1940static void nfsd4_del_conns(struct nfsd4_session *s)
1941{
1942 struct nfs4_client *clp = s->se_client;
1943 struct nfsd4_conn *c;
1944
1945 spin_lock(&clp->cl_lock);
1946 while (!list_empty(&s->se_conns)) {
1947 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1948 list_del_init(&c->cn_persession);
1949 spin_unlock(&clp->cl_lock);
1950
1951 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1952 free_conn(c);
1953
1954 spin_lock(&clp->cl_lock);
1955 }
1956 spin_unlock(&clp->cl_lock);
1957}
1958
1959static void __free_session(struct nfsd4_session *ses)
1960{
1961 free_session_slots(ses);
1962 kfree(ses);
1963}
1964
1965static void free_session(struct nfsd4_session *ses)
1966{
1967 nfsd4_del_conns(ses);
1968 nfsd4_put_drc_mem(&ses->se_fchannel);
1969 __free_session(ses);
1970}
1971
1972static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1973{
1974 int idx;
1975 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1976
1977 new->se_client = clp;
1978 gen_sessionid(new);
1979
1980 INIT_LIST_HEAD(&new->se_conns);
1981
1982 new->se_cb_seq_nr = 1;
1983 new->se_flags = cses->flags;
1984 new->se_cb_prog = cses->callback_prog;
1985 new->se_cb_sec = cses->cb_sec;
1986 atomic_set(&new->se_ref, 0);
1987 idx = hash_sessionid(&new->se_sessionid);
1988 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1989 spin_lock(&clp->cl_lock);
1990 list_add(&new->se_perclnt, &clp->cl_sessions);
1991 spin_unlock(&clp->cl_lock);
1992
1993 {
1994 struct sockaddr *sa = svc_addr(rqstp);
1995 /*
1996 * This is a little silly; with sessions there's no real
1997 * use for the callback address. Use the peer address
1998 * as a reasonable default for now, but consider fixing
1999 * the rpc client not to require an address in the
2000 * future:
2001 */
2002 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2003 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2004 }
2005}
2006
2007/* caller must hold client_lock */
2008static struct nfsd4_session *
2009__find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2010{
2011 struct nfsd4_session *elem;
2012 int idx;
2013 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2014
2015 lockdep_assert_held(&nn->client_lock);
2016
2017 dump_sessionid(__func__, sessionid);
2018 idx = hash_sessionid(sessionid);
2019 /* Search in the appropriate list */
2020 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2021 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2022 NFS4_MAX_SESSIONID_LEN)) {
2023 return elem;
2024 }
2025 }
2026
2027 dprintk("%s: session not found\n", __func__);
2028 return NULL;
2029}
2030
2031static struct nfsd4_session *
2032find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2033 __be32 *ret)
2034{
2035 struct nfsd4_session *session;
2036 __be32 status = nfserr_badsession;
2037
2038 session = __find_in_sessionid_hashtbl(sessionid, net);
2039 if (!session)
2040 goto out;
2041 status = nfsd4_get_session_locked(session);
2042 if (status)
2043 session = NULL;
2044out:
2045 *ret = status;
2046 return session;
2047}
2048
2049/* caller must hold client_lock */
2050static void
2051unhash_session(struct nfsd4_session *ses)
2052{
2053 struct nfs4_client *clp = ses->se_client;
2054 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2055
2056 lockdep_assert_held(&nn->client_lock);
2057
2058 list_del(&ses->se_hash);
2059 spin_lock(&ses->se_client->cl_lock);
2060 list_del(&ses->se_perclnt);
2061 spin_unlock(&ses->se_client->cl_lock);
2062}
2063
2064/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2065static int
2066STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2067{
2068 /*
2069 * We're assuming the clid was not given out from a boot
2070 * precisely 2^32 (about 136 years) before this one. That seems
2071 * a safe assumption:
2072 */
2073 if (clid->cl_boot == (u32)nn->boot_time)
2074 return 0;
2075 trace_nfsd_clid_stale(clid);
2076 return 1;
2077}
2078
2079/*
2080 * XXX Should we use a slab cache ?
2081 * This type of memory management is somewhat inefficient, but we use it
2082 * anyway since SETCLIENTID is not a common operation.
2083 */
2084static struct nfs4_client *alloc_client(struct xdr_netobj name,
2085 struct nfsd_net *nn)
2086{
2087 struct nfs4_client *clp;
2088 int i;
2089
2090 if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2091 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2092 return NULL;
2093 }
2094 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2095 if (clp == NULL)
2096 return NULL;
2097 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2098 if (clp->cl_name.data == NULL)
2099 goto err_no_name;
2100 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2101 sizeof(struct list_head),
2102 GFP_KERNEL);
2103 if (!clp->cl_ownerstr_hashtbl)
2104 goto err_no_hashtbl;
2105 for (i = 0; i < OWNER_HASH_SIZE; i++)
2106 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2107 INIT_LIST_HEAD(&clp->cl_sessions);
2108 idr_init(&clp->cl_stateids);
2109 atomic_set(&clp->cl_rpc_users, 0);
2110 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2111 clp->cl_state = NFSD4_ACTIVE;
2112 atomic_inc(&nn->nfs4_client_count);
2113 atomic_set(&clp->cl_delegs_in_recall, 0);
2114 INIT_LIST_HEAD(&clp->cl_idhash);
2115 INIT_LIST_HEAD(&clp->cl_openowners);
2116 INIT_LIST_HEAD(&clp->cl_delegations);
2117 INIT_LIST_HEAD(&clp->cl_lru);
2118 INIT_LIST_HEAD(&clp->cl_revoked);
2119#ifdef CONFIG_NFSD_PNFS
2120 INIT_LIST_HEAD(&clp->cl_lo_states);
2121#endif
2122 INIT_LIST_HEAD(&clp->async_copies);
2123 spin_lock_init(&clp->async_lock);
2124 spin_lock_init(&clp->cl_lock);
2125 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2126 return clp;
2127err_no_hashtbl:
2128 kfree(clp->cl_name.data);
2129err_no_name:
2130 kmem_cache_free(client_slab, clp);
2131 return NULL;
2132}
2133
2134static void __free_client(struct kref *k)
2135{
2136 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2137 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2138
2139 free_svc_cred(&clp->cl_cred);
2140 kfree(clp->cl_ownerstr_hashtbl);
2141 kfree(clp->cl_name.data);
2142 kfree(clp->cl_nii_domain.data);
2143 kfree(clp->cl_nii_name.data);
2144 idr_destroy(&clp->cl_stateids);
2145 kfree(clp->cl_ra);
2146 kmem_cache_free(client_slab, clp);
2147}
2148
2149static void drop_client(struct nfs4_client *clp)
2150{
2151 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2152}
2153
2154static void
2155free_client(struct nfs4_client *clp)
2156{
2157 while (!list_empty(&clp->cl_sessions)) {
2158 struct nfsd4_session *ses;
2159 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2160 se_perclnt);
2161 list_del(&ses->se_perclnt);
2162 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2163 free_session(ses);
2164 }
2165 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2166 if (clp->cl_nfsd_dentry) {
2167 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2168 clp->cl_nfsd_dentry = NULL;
2169 wake_up_all(&expiry_wq);
2170 }
2171 drop_client(clp);
2172}
2173
2174/* must be called under the client_lock */
2175static void
2176unhash_client_locked(struct nfs4_client *clp)
2177{
2178 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2179 struct nfsd4_session *ses;
2180
2181 lockdep_assert_held(&nn->client_lock);
2182
2183 /* Mark the client as expired! */
2184 clp->cl_time = 0;
2185 /* Make it invisible */
2186 if (!list_empty(&clp->cl_idhash)) {
2187 list_del_init(&clp->cl_idhash);
2188 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2189 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2190 else
2191 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2192 }
2193 list_del_init(&clp->cl_lru);
2194 spin_lock(&clp->cl_lock);
2195 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2196 list_del_init(&ses->se_hash);
2197 spin_unlock(&clp->cl_lock);
2198}
2199
2200static void
2201unhash_client(struct nfs4_client *clp)
2202{
2203 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2204
2205 spin_lock(&nn->client_lock);
2206 unhash_client_locked(clp);
2207 spin_unlock(&nn->client_lock);
2208}
2209
2210static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2211{
2212 if (atomic_read(&clp->cl_rpc_users))
2213 return nfserr_jukebox;
2214 unhash_client_locked(clp);
2215 return nfs_ok;
2216}
2217
2218static void
2219__destroy_client(struct nfs4_client *clp)
2220{
2221 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2222 int i;
2223 struct nfs4_openowner *oo;
2224 struct nfs4_delegation *dp;
2225 struct list_head reaplist;
2226
2227 INIT_LIST_HEAD(&reaplist);
2228 spin_lock(&state_lock);
2229 while (!list_empty(&clp->cl_delegations)) {
2230 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2231 WARN_ON(!unhash_delegation_locked(dp));
2232 list_add(&dp->dl_recall_lru, &reaplist);
2233 }
2234 spin_unlock(&state_lock);
2235 while (!list_empty(&reaplist)) {
2236 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2237 list_del_init(&dp->dl_recall_lru);
2238 destroy_unhashed_deleg(dp);
2239 }
2240 while (!list_empty(&clp->cl_revoked)) {
2241 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2242 list_del_init(&dp->dl_recall_lru);
2243 nfs4_put_stid(&dp->dl_stid);
2244 }
2245 while (!list_empty(&clp->cl_openowners)) {
2246 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2247 nfs4_get_stateowner(&oo->oo_owner);
2248 release_openowner(oo);
2249 }
2250 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2251 struct nfs4_stateowner *so, *tmp;
2252
2253 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2254 so_strhash) {
2255 /* Should be no openowners at this point */
2256 WARN_ON_ONCE(so->so_is_open_owner);
2257 remove_blocked_locks(lockowner(so));
2258 }
2259 }
2260 nfsd4_return_all_client_layouts(clp);
2261 nfsd4_shutdown_copy(clp);
2262 nfsd4_shutdown_callback(clp);
2263 if (clp->cl_cb_conn.cb_xprt)
2264 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2265 atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2266 nfsd4_dec_courtesy_client_count(nn, clp);
2267 free_client(clp);
2268 wake_up_all(&expiry_wq);
2269}
2270
2271static void
2272destroy_client(struct nfs4_client *clp)
2273{
2274 unhash_client(clp);
2275 __destroy_client(clp);
2276}
2277
2278static void inc_reclaim_complete(struct nfs4_client *clp)
2279{
2280 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2281
2282 if (!nn->track_reclaim_completes)
2283 return;
2284 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2285 return;
2286 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2287 nn->reclaim_str_hashtbl_size) {
2288 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2289 clp->net->ns.inum);
2290 nfsd4_end_grace(nn);
2291 }
2292}
2293
2294static void expire_client(struct nfs4_client *clp)
2295{
2296 unhash_client(clp);
2297 nfsd4_client_record_remove(clp);
2298 __destroy_client(clp);
2299}
2300
2301static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2302{
2303 memcpy(target->cl_verifier.data, source->data,
2304 sizeof(target->cl_verifier.data));
2305}
2306
2307static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2308{
2309 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2310 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2311}
2312
2313static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2314{
2315 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2316 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2317 GFP_KERNEL);
2318 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2319 if ((source->cr_principal && !target->cr_principal) ||
2320 (source->cr_raw_principal && !target->cr_raw_principal) ||
2321 (source->cr_targ_princ && !target->cr_targ_princ))
2322 return -ENOMEM;
2323
2324 target->cr_flavor = source->cr_flavor;
2325 target->cr_uid = source->cr_uid;
2326 target->cr_gid = source->cr_gid;
2327 target->cr_group_info = source->cr_group_info;
2328 get_group_info(target->cr_group_info);
2329 target->cr_gss_mech = source->cr_gss_mech;
2330 if (source->cr_gss_mech)
2331 gss_mech_get(source->cr_gss_mech);
2332 return 0;
2333}
2334
2335static int
2336compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2337{
2338 if (o1->len < o2->len)
2339 return -1;
2340 if (o1->len > o2->len)
2341 return 1;
2342 return memcmp(o1->data, o2->data, o1->len);
2343}
2344
2345static int
2346same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2347{
2348 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2349}
2350
2351static int
2352same_clid(clientid_t *cl1, clientid_t *cl2)
2353{
2354 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2355}
2356
2357static bool groups_equal(struct group_info *g1, struct group_info *g2)
2358{
2359 int i;
2360
2361 if (g1->ngroups != g2->ngroups)
2362 return false;
2363 for (i=0; i<g1->ngroups; i++)
2364 if (!gid_eq(g1->gid[i], g2->gid[i]))
2365 return false;
2366 return true;
2367}
2368
2369/*
2370 * RFC 3530 language requires clid_inuse be returned when the
2371 * "principal" associated with a requests differs from that previously
2372 * used. We use uid, gid's, and gss principal string as our best
2373 * approximation. We also don't want to allow non-gss use of a client
2374 * established using gss: in theory cr_principal should catch that
2375 * change, but in practice cr_principal can be null even in the gss case
2376 * since gssd doesn't always pass down a principal string.
2377 */
2378static bool is_gss_cred(struct svc_cred *cr)
2379{
2380 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2381 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2382}
2383
2384
2385static bool
2386same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2387{
2388 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2389 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2390 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2391 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2392 return false;
2393 /* XXX: check that cr_targ_princ fields match ? */
2394 if (cr1->cr_principal == cr2->cr_principal)
2395 return true;
2396 if (!cr1->cr_principal || !cr2->cr_principal)
2397 return false;
2398 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2399}
2400
2401static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2402{
2403 struct svc_cred *cr = &rqstp->rq_cred;
2404 u32 service;
2405
2406 if (!cr->cr_gss_mech)
2407 return false;
2408 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2409 return service == RPC_GSS_SVC_INTEGRITY ||
2410 service == RPC_GSS_SVC_PRIVACY;
2411}
2412
2413bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2414{
2415 struct svc_cred *cr = &rqstp->rq_cred;
2416
2417 if (!cl->cl_mach_cred)
2418 return true;
2419 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2420 return false;
2421 if (!svc_rqst_integrity_protected(rqstp))
2422 return false;
2423 if (cl->cl_cred.cr_raw_principal)
2424 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2425 cr->cr_raw_principal);
2426 if (!cr->cr_principal)
2427 return false;
2428 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2429}
2430
2431static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2432{
2433 __be32 verf[2];
2434
2435 /*
2436 * This is opaque to client, so no need to byte-swap. Use
2437 * __force to keep sparse happy
2438 */
2439 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2440 verf[1] = (__force __be32)nn->clverifier_counter++;
2441 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2442}
2443
2444static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2445{
2446 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2447 clp->cl_clientid.cl_id = nn->clientid_counter++;
2448 gen_confirm(clp, nn);
2449}
2450
2451static struct nfs4_stid *
2452find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2453{
2454 struct nfs4_stid *ret;
2455
2456 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2457 if (!ret || !ret->sc_type)
2458 return NULL;
2459 return ret;
2460}
2461
2462static struct nfs4_stid *
2463find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2464{
2465 struct nfs4_stid *s;
2466
2467 spin_lock(&cl->cl_lock);
2468 s = find_stateid_locked(cl, t);
2469 if (s != NULL) {
2470 if (typemask & s->sc_type)
2471 refcount_inc(&s->sc_count);
2472 else
2473 s = NULL;
2474 }
2475 spin_unlock(&cl->cl_lock);
2476 return s;
2477}
2478
2479static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2480{
2481 struct nfsdfs_client *nc;
2482 nc = get_nfsdfs_client(inode);
2483 if (!nc)
2484 return NULL;
2485 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2486}
2487
2488static void seq_quote_mem(struct seq_file *m, char *data, int len)
2489{
2490 seq_printf(m, "\"");
2491 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2492 seq_printf(m, "\"");
2493}
2494
2495static const char *cb_state2str(int state)
2496{
2497 switch (state) {
2498 case NFSD4_CB_UP:
2499 return "UP";
2500 case NFSD4_CB_UNKNOWN:
2501 return "UNKNOWN";
2502 case NFSD4_CB_DOWN:
2503 return "DOWN";
2504 case NFSD4_CB_FAULT:
2505 return "FAULT";
2506 }
2507 return "UNDEFINED";
2508}
2509
2510static int client_info_show(struct seq_file *m, void *v)
2511{
2512 struct inode *inode = file_inode(m->file);
2513 struct nfs4_client *clp;
2514 u64 clid;
2515
2516 clp = get_nfsdfs_clp(inode);
2517 if (!clp)
2518 return -ENXIO;
2519 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2520 seq_printf(m, "clientid: 0x%llx\n", clid);
2521 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2522
2523 if (clp->cl_state == NFSD4_COURTESY)
2524 seq_puts(m, "status: courtesy\n");
2525 else if (clp->cl_state == NFSD4_EXPIRABLE)
2526 seq_puts(m, "status: expirable\n");
2527 else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2528 seq_puts(m, "status: confirmed\n");
2529 else
2530 seq_puts(m, "status: unconfirmed\n");
2531 seq_printf(m, "seconds from last renew: %lld\n",
2532 ktime_get_boottime_seconds() - clp->cl_time);
2533 seq_printf(m, "name: ");
2534 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2535 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2536 if (clp->cl_nii_domain.data) {
2537 seq_printf(m, "Implementation domain: ");
2538 seq_quote_mem(m, clp->cl_nii_domain.data,
2539 clp->cl_nii_domain.len);
2540 seq_printf(m, "\nImplementation name: ");
2541 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2542 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2543 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2544 }
2545 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2546 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2547 drop_client(clp);
2548
2549 return 0;
2550}
2551
2552DEFINE_SHOW_ATTRIBUTE(client_info);
2553
2554static void *states_start(struct seq_file *s, loff_t *pos)
2555 __acquires(&clp->cl_lock)
2556{
2557 struct nfs4_client *clp = s->private;
2558 unsigned long id = *pos;
2559 void *ret;
2560
2561 spin_lock(&clp->cl_lock);
2562 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2563 *pos = id;
2564 return ret;
2565}
2566
2567static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2568{
2569 struct nfs4_client *clp = s->private;
2570 unsigned long id = *pos;
2571 void *ret;
2572
2573 id = *pos;
2574 id++;
2575 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2576 *pos = id;
2577 return ret;
2578}
2579
2580static void states_stop(struct seq_file *s, void *v)
2581 __releases(&clp->cl_lock)
2582{
2583 struct nfs4_client *clp = s->private;
2584
2585 spin_unlock(&clp->cl_lock);
2586}
2587
2588static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2589{
2590 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2591}
2592
2593static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2594{
2595 struct inode *inode = file_inode(f->nf_file);
2596
2597 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2598 MAJOR(inode->i_sb->s_dev),
2599 MINOR(inode->i_sb->s_dev),
2600 inode->i_ino);
2601}
2602
2603static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2604{
2605 seq_printf(s, "owner: ");
2606 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2607}
2608
2609static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2610{
2611 seq_printf(s, "0x%.8x", stid->si_generation);
2612 seq_printf(s, "%12phN", &stid->si_opaque);
2613}
2614
2615static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2616{
2617 struct nfs4_ol_stateid *ols;
2618 struct nfs4_file *nf;
2619 struct nfsd_file *file;
2620 struct nfs4_stateowner *oo;
2621 unsigned int access, deny;
2622
2623 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2624 return 0; /* XXX: or SEQ_SKIP? */
2625 ols = openlockstateid(st);
2626 oo = ols->st_stateowner;
2627 nf = st->sc_file;
2628
2629 spin_lock(&nf->fi_lock);
2630 file = find_any_file_locked(nf);
2631 if (!file)
2632 goto out;
2633
2634 seq_printf(s, "- ");
2635 nfs4_show_stateid(s, &st->sc_stateid);
2636 seq_printf(s, ": { type: open, ");
2637
2638 access = bmap_to_share_mode(ols->st_access_bmap);
2639 deny = bmap_to_share_mode(ols->st_deny_bmap);
2640
2641 seq_printf(s, "access: %s%s, ",
2642 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2643 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2644 seq_printf(s, "deny: %s%s, ",
2645 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2646 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2647
2648 nfs4_show_superblock(s, file);
2649 seq_printf(s, ", ");
2650 nfs4_show_fname(s, file);
2651 seq_printf(s, ", ");
2652 nfs4_show_owner(s, oo);
2653 seq_printf(s, " }\n");
2654out:
2655 spin_unlock(&nf->fi_lock);
2656 return 0;
2657}
2658
2659static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2660{
2661 struct nfs4_ol_stateid *ols;
2662 struct nfs4_file *nf;
2663 struct nfsd_file *file;
2664 struct nfs4_stateowner *oo;
2665
2666 ols = openlockstateid(st);
2667 oo = ols->st_stateowner;
2668 nf = st->sc_file;
2669 spin_lock(&nf->fi_lock);
2670 file = find_any_file_locked(nf);
2671 if (!file)
2672 goto out;
2673
2674 seq_printf(s, "- ");
2675 nfs4_show_stateid(s, &st->sc_stateid);
2676 seq_printf(s, ": { type: lock, ");
2677
2678 /*
2679 * Note: a lock stateid isn't really the same thing as a lock,
2680 * it's the locking state held by one owner on a file, and there
2681 * may be multiple (or no) lock ranges associated with it.
2682 * (Same for the matter is true of open stateids.)
2683 */
2684
2685 nfs4_show_superblock(s, file);
2686 /* XXX: open stateid? */
2687 seq_printf(s, ", ");
2688 nfs4_show_fname(s, file);
2689 seq_printf(s, ", ");
2690 nfs4_show_owner(s, oo);
2691 seq_printf(s, " }\n");
2692out:
2693 spin_unlock(&nf->fi_lock);
2694 return 0;
2695}
2696
2697static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2698{
2699 struct nfs4_delegation *ds;
2700 struct nfs4_file *nf;
2701 struct nfsd_file *file;
2702
2703 ds = delegstateid(st);
2704 nf = st->sc_file;
2705 spin_lock(&nf->fi_lock);
2706 file = nf->fi_deleg_file;
2707 if (!file)
2708 goto out;
2709
2710 seq_printf(s, "- ");
2711 nfs4_show_stateid(s, &st->sc_stateid);
2712 seq_printf(s, ": { type: deleg, ");
2713
2714 /* Kinda dead code as long as we only support read delegs: */
2715 seq_printf(s, "access: %s, ",
2716 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2717
2718 /* XXX: lease time, whether it's being recalled. */
2719
2720 nfs4_show_superblock(s, file);
2721 seq_printf(s, ", ");
2722 nfs4_show_fname(s, file);
2723 seq_printf(s, " }\n");
2724out:
2725 spin_unlock(&nf->fi_lock);
2726 return 0;
2727}
2728
2729static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2730{
2731 struct nfs4_layout_stateid *ls;
2732 struct nfsd_file *file;
2733
2734 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2735 file = ls->ls_file;
2736
2737 seq_printf(s, "- ");
2738 nfs4_show_stateid(s, &st->sc_stateid);
2739 seq_printf(s, ": { type: layout, ");
2740
2741 /* XXX: What else would be useful? */
2742
2743 nfs4_show_superblock(s, file);
2744 seq_printf(s, ", ");
2745 nfs4_show_fname(s, file);
2746 seq_printf(s, " }\n");
2747
2748 return 0;
2749}
2750
2751static int states_show(struct seq_file *s, void *v)
2752{
2753 struct nfs4_stid *st = v;
2754
2755 switch (st->sc_type) {
2756 case NFS4_OPEN_STID:
2757 return nfs4_show_open(s, st);
2758 case NFS4_LOCK_STID:
2759 return nfs4_show_lock(s, st);
2760 case NFS4_DELEG_STID:
2761 return nfs4_show_deleg(s, st);
2762 case NFS4_LAYOUT_STID:
2763 return nfs4_show_layout(s, st);
2764 default:
2765 return 0; /* XXX: or SEQ_SKIP? */
2766 }
2767 /* XXX: copy stateids? */
2768}
2769
2770static struct seq_operations states_seq_ops = {
2771 .start = states_start,
2772 .next = states_next,
2773 .stop = states_stop,
2774 .show = states_show
2775};
2776
2777static int client_states_open(struct inode *inode, struct file *file)
2778{
2779 struct seq_file *s;
2780 struct nfs4_client *clp;
2781 int ret;
2782
2783 clp = get_nfsdfs_clp(inode);
2784 if (!clp)
2785 return -ENXIO;
2786
2787 ret = seq_open(file, &states_seq_ops);
2788 if (ret)
2789 return ret;
2790 s = file->private_data;
2791 s->private = clp;
2792 return 0;
2793}
2794
2795static int client_opens_release(struct inode *inode, struct file *file)
2796{
2797 struct seq_file *m = file->private_data;
2798 struct nfs4_client *clp = m->private;
2799
2800 /* XXX: alternatively, we could get/drop in seq start/stop */
2801 drop_client(clp);
2802 return seq_release(inode, file);
2803}
2804
2805static const struct file_operations client_states_fops = {
2806 .open = client_states_open,
2807 .read = seq_read,
2808 .llseek = seq_lseek,
2809 .release = client_opens_release,
2810};
2811
2812/*
2813 * Normally we refuse to destroy clients that are in use, but here the
2814 * administrator is telling us to just do it. We also want to wait
2815 * so the caller has a guarantee that the client's locks are gone by
2816 * the time the write returns:
2817 */
2818static void force_expire_client(struct nfs4_client *clp)
2819{
2820 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2821 bool already_expired;
2822
2823 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2824
2825 spin_lock(&nn->client_lock);
2826 clp->cl_time = 0;
2827 spin_unlock(&nn->client_lock);
2828
2829 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2830 spin_lock(&nn->client_lock);
2831 already_expired = list_empty(&clp->cl_lru);
2832 if (!already_expired)
2833 unhash_client_locked(clp);
2834 spin_unlock(&nn->client_lock);
2835
2836 if (!already_expired)
2837 expire_client(clp);
2838 else
2839 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2840}
2841
2842static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2843 size_t size, loff_t *pos)
2844{
2845 char *data;
2846 struct nfs4_client *clp;
2847
2848 data = simple_transaction_get(file, buf, size);
2849 if (IS_ERR(data))
2850 return PTR_ERR(data);
2851 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2852 return -EINVAL;
2853 clp = get_nfsdfs_clp(file_inode(file));
2854 if (!clp)
2855 return -ENXIO;
2856 force_expire_client(clp);
2857 drop_client(clp);
2858 return 7;
2859}
2860
2861static const struct file_operations client_ctl_fops = {
2862 .write = client_ctl_write,
2863 .release = simple_transaction_release,
2864};
2865
2866static const struct tree_descr client_files[] = {
2867 [0] = {"info", &client_info_fops, S_IRUSR},
2868 [1] = {"states", &client_states_fops, S_IRUSR},
2869 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2870 [3] = {""},
2871};
2872
2873static int
2874nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
2875 struct rpc_task *task)
2876{
2877 trace_nfsd_cb_recall_any_done(cb, task);
2878 switch (task->tk_status) {
2879 case -NFS4ERR_DELAY:
2880 rpc_delay(task, 2 * HZ);
2881 return 0;
2882 default:
2883 return 1;
2884 }
2885}
2886
2887static void
2888nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
2889{
2890 struct nfs4_client *clp = cb->cb_clp;
2891 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2892
2893 spin_lock(&nn->client_lock);
2894 clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
2895 put_client_renew_locked(clp);
2896 spin_unlock(&nn->client_lock);
2897}
2898
2899static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
2900 .done = nfsd4_cb_recall_any_done,
2901 .release = nfsd4_cb_recall_any_release,
2902};
2903
2904static struct nfs4_client *create_client(struct xdr_netobj name,
2905 struct svc_rqst *rqstp, nfs4_verifier *verf)
2906{
2907 struct nfs4_client *clp;
2908 struct sockaddr *sa = svc_addr(rqstp);
2909 int ret;
2910 struct net *net = SVC_NET(rqstp);
2911 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2912 struct dentry *dentries[ARRAY_SIZE(client_files)];
2913
2914 clp = alloc_client(name, nn);
2915 if (clp == NULL)
2916 return NULL;
2917
2918 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2919 if (ret) {
2920 free_client(clp);
2921 return NULL;
2922 }
2923 gen_clid(clp, nn);
2924 kref_init(&clp->cl_nfsdfs.cl_ref);
2925 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2926 clp->cl_time = ktime_get_boottime_seconds();
2927 clear_bit(0, &clp->cl_cb_slot_busy);
2928 copy_verf(clp, verf);
2929 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2930 clp->cl_cb_session = NULL;
2931 clp->net = net;
2932 clp->cl_nfsd_dentry = nfsd_client_mkdir(
2933 nn, &clp->cl_nfsdfs,
2934 clp->cl_clientid.cl_id - nn->clientid_base,
2935 client_files, dentries);
2936 clp->cl_nfsd_info_dentry = dentries[0];
2937 if (!clp->cl_nfsd_dentry) {
2938 free_client(clp);
2939 return NULL;
2940 }
2941 clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
2942 if (!clp->cl_ra) {
2943 free_client(clp);
2944 return NULL;
2945 }
2946 clp->cl_ra_time = 0;
2947 nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
2948 NFSPROC4_CLNT_CB_RECALL_ANY);
2949 return clp;
2950}
2951
2952static void
2953add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2954{
2955 struct rb_node **new = &(root->rb_node), *parent = NULL;
2956 struct nfs4_client *clp;
2957
2958 while (*new) {
2959 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2960 parent = *new;
2961
2962 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2963 new = &((*new)->rb_left);
2964 else
2965 new = &((*new)->rb_right);
2966 }
2967
2968 rb_link_node(&new_clp->cl_namenode, parent, new);
2969 rb_insert_color(&new_clp->cl_namenode, root);
2970}
2971
2972static struct nfs4_client *
2973find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2974{
2975 int cmp;
2976 struct rb_node *node = root->rb_node;
2977 struct nfs4_client *clp;
2978
2979 while (node) {
2980 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2981 cmp = compare_blob(&clp->cl_name, name);
2982 if (cmp > 0)
2983 node = node->rb_left;
2984 else if (cmp < 0)
2985 node = node->rb_right;
2986 else
2987 return clp;
2988 }
2989 return NULL;
2990}
2991
2992static void
2993add_to_unconfirmed(struct nfs4_client *clp)
2994{
2995 unsigned int idhashval;
2996 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2997
2998 lockdep_assert_held(&nn->client_lock);
2999
3000 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3001 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3002 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3003 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3004 renew_client_locked(clp);
3005}
3006
3007static void
3008move_to_confirmed(struct nfs4_client *clp)
3009{
3010 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3011 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3012
3013 lockdep_assert_held(&nn->client_lock);
3014
3015 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3016 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3017 add_clp_to_name_tree(clp, &nn->conf_name_tree);
3018 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3019 trace_nfsd_clid_confirmed(&clp->cl_clientid);
3020 renew_client_locked(clp);
3021}
3022
3023static struct nfs4_client *
3024find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3025{
3026 struct nfs4_client *clp;
3027 unsigned int idhashval = clientid_hashval(clid->cl_id);
3028
3029 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3030 if (same_clid(&clp->cl_clientid, clid)) {
3031 if ((bool)clp->cl_minorversion != sessions)
3032 return NULL;
3033 renew_client_locked(clp);
3034 return clp;
3035 }
3036 }
3037 return NULL;
3038}
3039
3040static struct nfs4_client *
3041find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3042{
3043 struct list_head *tbl = nn->conf_id_hashtbl;
3044
3045 lockdep_assert_held(&nn->client_lock);
3046 return find_client_in_id_table(tbl, clid, sessions);
3047}
3048
3049static struct nfs4_client *
3050find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3051{
3052 struct list_head *tbl = nn->unconf_id_hashtbl;
3053
3054 lockdep_assert_held(&nn->client_lock);
3055 return find_client_in_id_table(tbl, clid, sessions);
3056}
3057
3058static bool clp_used_exchangeid(struct nfs4_client *clp)
3059{
3060 return clp->cl_exchange_flags != 0;
3061}
3062
3063static struct nfs4_client *
3064find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3065{
3066 lockdep_assert_held(&nn->client_lock);
3067 return find_clp_in_name_tree(name, &nn->conf_name_tree);
3068}
3069
3070static struct nfs4_client *
3071find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3072{
3073 lockdep_assert_held(&nn->client_lock);
3074 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3075}
3076
3077static void
3078gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3079{
3080 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3081 struct sockaddr *sa = svc_addr(rqstp);
3082 u32 scopeid = rpc_get_scope_id(sa);
3083 unsigned short expected_family;
3084
3085 /* Currently, we only support tcp and tcp6 for the callback channel */
3086 if (se->se_callback_netid_len == 3 &&
3087 !memcmp(se->se_callback_netid_val, "tcp", 3))
3088 expected_family = AF_INET;
3089 else if (se->se_callback_netid_len == 4 &&
3090 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3091 expected_family = AF_INET6;
3092 else
3093 goto out_err;
3094
3095 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3096 se->se_callback_addr_len,
3097 (struct sockaddr *)&conn->cb_addr,
3098 sizeof(conn->cb_addr));
3099
3100 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3101 goto out_err;
3102
3103 if (conn->cb_addr.ss_family == AF_INET6)
3104 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3105
3106 conn->cb_prog = se->se_callback_prog;
3107 conn->cb_ident = se->se_callback_ident;
3108 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3109 trace_nfsd_cb_args(clp, conn);
3110 return;
3111out_err:
3112 conn->cb_addr.ss_family = AF_UNSPEC;
3113 conn->cb_addrlen = 0;
3114 trace_nfsd_cb_nodelegs(clp);
3115 return;
3116}
3117
3118/*
3119 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3120 */
3121static void
3122nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3123{
3124 struct xdr_buf *buf = resp->xdr->buf;
3125 struct nfsd4_slot *slot = resp->cstate.slot;
3126 unsigned int base;
3127
3128 dprintk("--> %s slot %p\n", __func__, slot);
3129
3130 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3131 slot->sl_opcnt = resp->opcnt;
3132 slot->sl_status = resp->cstate.status;
3133 free_svc_cred(&slot->sl_cred);
3134 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3135
3136 if (!nfsd4_cache_this(resp)) {
3137 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3138 return;
3139 }
3140 slot->sl_flags |= NFSD4_SLOT_CACHED;
3141
3142 base = resp->cstate.data_offset;
3143 slot->sl_datalen = buf->len - base;
3144 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3145 WARN(1, "%s: sessions DRC could not cache compound\n",
3146 __func__);
3147 return;
3148}
3149
3150/*
3151 * Encode the replay sequence operation from the slot values.
3152 * If cachethis is FALSE encode the uncached rep error on the next
3153 * operation which sets resp->p and increments resp->opcnt for
3154 * nfs4svc_encode_compoundres.
3155 *
3156 */
3157static __be32
3158nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3159 struct nfsd4_compoundres *resp)
3160{
3161 struct nfsd4_op *op;
3162 struct nfsd4_slot *slot = resp->cstate.slot;
3163
3164 /* Encode the replayed sequence operation */
3165 op = &args->ops[resp->opcnt - 1];
3166 nfsd4_encode_operation(resp, op);
3167
3168 if (slot->sl_flags & NFSD4_SLOT_CACHED)
3169 return op->status;
3170 if (args->opcnt == 1) {
3171 /*
3172 * The original operation wasn't a solo sequence--we
3173 * always cache those--so this retry must not match the
3174 * original:
3175 */
3176 op->status = nfserr_seq_false_retry;
3177 } else {
3178 op = &args->ops[resp->opcnt++];
3179 op->status = nfserr_retry_uncached_rep;
3180 nfsd4_encode_operation(resp, op);
3181 }
3182 return op->status;
3183}
3184
3185/*
3186 * The sequence operation is not cached because we can use the slot and
3187 * session values.
3188 */
3189static __be32
3190nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3191 struct nfsd4_sequence *seq)
3192{
3193 struct nfsd4_slot *slot = resp->cstate.slot;
3194 struct xdr_stream *xdr = resp->xdr;
3195 __be32 *p;
3196 __be32 status;
3197
3198 dprintk("--> %s slot %p\n", __func__, slot);
3199
3200 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3201 if (status)
3202 return status;
3203
3204 p = xdr_reserve_space(xdr, slot->sl_datalen);
3205 if (!p) {
3206 WARN_ON_ONCE(1);
3207 return nfserr_serverfault;
3208 }
3209 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3210 xdr_commit_encode(xdr);
3211
3212 resp->opcnt = slot->sl_opcnt;
3213 return slot->sl_status;
3214}
3215
3216/*
3217 * Set the exchange_id flags returned by the server.
3218 */
3219static void
3220nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3221{
3222#ifdef CONFIG_NFSD_PNFS
3223 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3224#else
3225 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3226#endif
3227
3228 /* Referrals are supported, Migration is not. */
3229 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3230
3231 /* set the wire flags to return to client. */
3232 clid->flags = new->cl_exchange_flags;
3233}
3234
3235static bool client_has_openowners(struct nfs4_client *clp)
3236{
3237 struct nfs4_openowner *oo;
3238
3239 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3240 if (!list_empty(&oo->oo_owner.so_stateids))
3241 return true;
3242 }
3243 return false;
3244}
3245
3246static bool client_has_state(struct nfs4_client *clp)
3247{
3248 return client_has_openowners(clp)
3249#ifdef CONFIG_NFSD_PNFS
3250 || !list_empty(&clp->cl_lo_states)
3251#endif
3252 || !list_empty(&clp->cl_delegations)
3253 || !list_empty(&clp->cl_sessions)
3254 || !list_empty(&clp->async_copies);
3255}
3256
3257static __be32 copy_impl_id(struct nfs4_client *clp,
3258 struct nfsd4_exchange_id *exid)
3259{
3260 if (!exid->nii_domain.data)
3261 return 0;
3262 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3263 if (!clp->cl_nii_domain.data)
3264 return nfserr_jukebox;
3265 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3266 if (!clp->cl_nii_name.data)
3267 return nfserr_jukebox;
3268 clp->cl_nii_time = exid->nii_time;
3269 return 0;
3270}
3271
3272__be32
3273nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3274 union nfsd4_op_u *u)
3275{
3276 struct nfsd4_exchange_id *exid = &u->exchange_id;
3277 struct nfs4_client *conf, *new;
3278 struct nfs4_client *unconf = NULL;
3279 __be32 status;
3280 char addr_str[INET6_ADDRSTRLEN];
3281 nfs4_verifier verf = exid->verifier;
3282 struct sockaddr *sa = svc_addr(rqstp);
3283 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3284 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3285
3286 rpc_ntop(sa, addr_str, sizeof(addr_str));
3287 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3288 "ip_addr=%s flags %x, spa_how %u\n",
3289 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3290 addr_str, exid->flags, exid->spa_how);
3291
3292 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3293 return nfserr_inval;
3294
3295 new = create_client(exid->clname, rqstp, &verf);
3296 if (new == NULL)
3297 return nfserr_jukebox;
3298 status = copy_impl_id(new, exid);
3299 if (status)
3300 goto out_nolock;
3301
3302 switch (exid->spa_how) {
3303 case SP4_MACH_CRED:
3304 exid->spo_must_enforce[0] = 0;
3305 exid->spo_must_enforce[1] = (
3306 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3307 1 << (OP_EXCHANGE_ID - 32) |
3308 1 << (OP_CREATE_SESSION - 32) |
3309 1 << (OP_DESTROY_SESSION - 32) |
3310 1 << (OP_DESTROY_CLIENTID - 32));
3311
3312 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3313 1 << (OP_OPEN_DOWNGRADE) |
3314 1 << (OP_LOCKU) |
3315 1 << (OP_DELEGRETURN));
3316
3317 exid->spo_must_allow[1] &= (
3318 1 << (OP_TEST_STATEID - 32) |
3319 1 << (OP_FREE_STATEID - 32));
3320 if (!svc_rqst_integrity_protected(rqstp)) {
3321 status = nfserr_inval;
3322 goto out_nolock;
3323 }
3324 /*
3325 * Sometimes userspace doesn't give us a principal.
3326 * Which is a bug, really. Anyway, we can't enforce
3327 * MACH_CRED in that case, better to give up now:
3328 */
3329 if (!new->cl_cred.cr_principal &&
3330 !new->cl_cred.cr_raw_principal) {
3331 status = nfserr_serverfault;
3332 goto out_nolock;
3333 }
3334 new->cl_mach_cred = true;
3335 break;
3336 case SP4_NONE:
3337 break;
3338 default: /* checked by xdr code */
3339 WARN_ON_ONCE(1);
3340 fallthrough;
3341 case SP4_SSV:
3342 status = nfserr_encr_alg_unsupp;
3343 goto out_nolock;
3344 }
3345
3346 /* Cases below refer to rfc 5661 section 18.35.4: */
3347 spin_lock(&nn->client_lock);
3348 conf = find_confirmed_client_by_name(&exid->clname, nn);
3349 if (conf) {
3350 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3351 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3352
3353 if (update) {
3354 if (!clp_used_exchangeid(conf)) { /* buggy client */
3355 status = nfserr_inval;
3356 goto out;
3357 }
3358 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3359 status = nfserr_wrong_cred;
3360 goto out;
3361 }
3362 if (!creds_match) { /* case 9 */
3363 status = nfserr_perm;
3364 goto out;
3365 }
3366 if (!verfs_match) { /* case 8 */
3367 status = nfserr_not_same;
3368 goto out;
3369 }
3370 /* case 6 */
3371 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3372 trace_nfsd_clid_confirmed_r(conf);
3373 goto out_copy;
3374 }
3375 if (!creds_match) { /* case 3 */
3376 if (client_has_state(conf)) {
3377 status = nfserr_clid_inuse;
3378 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3379 goto out;
3380 }
3381 goto out_new;
3382 }
3383 if (verfs_match) { /* case 2 */
3384 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3385 trace_nfsd_clid_confirmed_r(conf);
3386 goto out_copy;
3387 }
3388 /* case 5, client reboot */
3389 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3390 conf = NULL;
3391 goto out_new;
3392 }
3393
3394 if (update) { /* case 7 */
3395 status = nfserr_noent;
3396 goto out;
3397 }
3398
3399 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3400 if (unconf) /* case 4, possible retry or client restart */
3401 unhash_client_locked(unconf);
3402
3403 /* case 1, new owner ID */
3404 trace_nfsd_clid_fresh(new);
3405
3406out_new:
3407 if (conf) {
3408 status = mark_client_expired_locked(conf);
3409 if (status)
3410 goto out;
3411 trace_nfsd_clid_replaced(&conf->cl_clientid);
3412 }
3413 new->cl_minorversion = cstate->minorversion;
3414 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3415 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3416
3417 add_to_unconfirmed(new);
3418 swap(new, conf);
3419out_copy:
3420 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3421 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3422
3423 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3424 nfsd4_set_ex_flags(conf, exid);
3425
3426 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3427 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3428 status = nfs_ok;
3429
3430out:
3431 spin_unlock(&nn->client_lock);
3432out_nolock:
3433 if (new)
3434 expire_client(new);
3435 if (unconf) {
3436 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3437 expire_client(unconf);
3438 }
3439 return status;
3440}
3441
3442static __be32
3443check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3444{
3445 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3446 slot_seqid);
3447
3448 /* The slot is in use, and no response has been sent. */
3449 if (slot_inuse) {
3450 if (seqid == slot_seqid)
3451 return nfserr_jukebox;
3452 else
3453 return nfserr_seq_misordered;
3454 }
3455 /* Note unsigned 32-bit arithmetic handles wraparound: */
3456 if (likely(seqid == slot_seqid + 1))
3457 return nfs_ok;
3458 if (seqid == slot_seqid)
3459 return nfserr_replay_cache;
3460 return nfserr_seq_misordered;
3461}
3462
3463/*
3464 * Cache the create session result into the create session single DRC
3465 * slot cache by saving the xdr structure. sl_seqid has been set.
3466 * Do this for solo or embedded create session operations.
3467 */
3468static void
3469nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3470 struct nfsd4_clid_slot *slot, __be32 nfserr)
3471{
3472 slot->sl_status = nfserr;
3473 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3474}
3475
3476static __be32
3477nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3478 struct nfsd4_clid_slot *slot)
3479{
3480 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3481 return slot->sl_status;
3482}
3483
3484#define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3485 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3486 1 + /* MIN tag is length with zero, only length */ \
3487 3 + /* version, opcount, opcode */ \
3488 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3489 /* seqid, slotID, slotID, cache */ \
3490 4 ) * sizeof(__be32))
3491
3492#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3493 2 + /* verifier: AUTH_NULL, length 0 */\
3494 1 + /* status */ \
3495 1 + /* MIN tag is length with zero, only length */ \
3496 3 + /* opcount, opcode, opstatus*/ \
3497 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3498 /* seqid, slotID, slotID, slotID, status */ \
3499 5 ) * sizeof(__be32))
3500
3501static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3502{
3503 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3504
3505 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3506 return nfserr_toosmall;
3507 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3508 return nfserr_toosmall;
3509 ca->headerpadsz = 0;
3510 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3511 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3512 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3513 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3514 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3515 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3516 /*
3517 * Note decreasing slot size below client's request may make it
3518 * difficult for client to function correctly, whereas
3519 * decreasing the number of slots will (just?) affect
3520 * performance. When short on memory we therefore prefer to
3521 * decrease number of slots instead of their size. Clients that
3522 * request larger slots than they need will get poor results:
3523 * Note that we always allow at least one slot, because our
3524 * accounting is soft and provides no guarantees either way.
3525 */
3526 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3527
3528 return nfs_ok;
3529}
3530
3531/*
3532 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3533 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3534 */
3535#define RPC_MAX_HEADER_WITH_AUTH_SYS \
3536 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3537
3538#define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3539 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3540
3541#define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3542 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3543#define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3544 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3545 sizeof(__be32))
3546
3547static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3548{
3549 ca->headerpadsz = 0;
3550
3551 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3552 return nfserr_toosmall;
3553 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3554 return nfserr_toosmall;
3555 ca->maxresp_cached = 0;
3556 if (ca->maxops < 2)
3557 return nfserr_toosmall;
3558
3559 return nfs_ok;
3560}
3561
3562static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3563{
3564 switch (cbs->flavor) {
3565 case RPC_AUTH_NULL:
3566 case RPC_AUTH_UNIX:
3567 return nfs_ok;
3568 default:
3569 /*
3570 * GSS case: the spec doesn't allow us to return this
3571 * error. But it also doesn't allow us not to support
3572 * GSS.
3573 * I'd rather this fail hard than return some error the
3574 * client might think it can already handle:
3575 */
3576 return nfserr_encr_alg_unsupp;
3577 }
3578}
3579
3580__be32
3581nfsd4_create_session(struct svc_rqst *rqstp,
3582 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3583{
3584 struct nfsd4_create_session *cr_ses = &u->create_session;
3585 struct sockaddr *sa = svc_addr(rqstp);
3586 struct nfs4_client *conf, *unconf;
3587 struct nfs4_client *old = NULL;
3588 struct nfsd4_session *new;
3589 struct nfsd4_conn *conn;
3590 struct nfsd4_clid_slot *cs_slot = NULL;
3591 __be32 status = 0;
3592 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3593
3594 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3595 return nfserr_inval;
3596 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3597 if (status)
3598 return status;
3599 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3600 if (status)
3601 return status;
3602 status = check_backchannel_attrs(&cr_ses->back_channel);
3603 if (status)
3604 goto out_release_drc_mem;
3605 status = nfserr_jukebox;
3606 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3607 if (!new)
3608 goto out_release_drc_mem;
3609 conn = alloc_conn_from_crses(rqstp, cr_ses);
3610 if (!conn)
3611 goto out_free_session;
3612
3613 spin_lock(&nn->client_lock);
3614 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3615 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3616 WARN_ON_ONCE(conf && unconf);
3617
3618 if (conf) {
3619 status = nfserr_wrong_cred;
3620 if (!nfsd4_mach_creds_match(conf, rqstp))
3621 goto out_free_conn;
3622 cs_slot = &conf->cl_cs_slot;
3623 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3624 if (status) {
3625 if (status == nfserr_replay_cache)
3626 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3627 goto out_free_conn;
3628 }
3629 } else if (unconf) {
3630 status = nfserr_clid_inuse;
3631 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3632 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3633 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3634 goto out_free_conn;
3635 }
3636 status = nfserr_wrong_cred;
3637 if (!nfsd4_mach_creds_match(unconf, rqstp))
3638 goto out_free_conn;
3639 cs_slot = &unconf->cl_cs_slot;
3640 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3641 if (status) {
3642 /* an unconfirmed replay returns misordered */
3643 status = nfserr_seq_misordered;
3644 goto out_free_conn;
3645 }
3646 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3647 if (old) {
3648 status = mark_client_expired_locked(old);
3649 if (status) {
3650 old = NULL;
3651 goto out_free_conn;
3652 }
3653 trace_nfsd_clid_replaced(&old->cl_clientid);
3654 }
3655 move_to_confirmed(unconf);
3656 conf = unconf;
3657 } else {
3658 status = nfserr_stale_clientid;
3659 goto out_free_conn;
3660 }
3661 status = nfs_ok;
3662 /* Persistent sessions are not supported */
3663 cr_ses->flags &= ~SESSION4_PERSIST;
3664 /* Upshifting from TCP to RDMA is not supported */
3665 cr_ses->flags &= ~SESSION4_RDMA;
3666
3667 init_session(rqstp, new, conf, cr_ses);
3668 nfsd4_get_session_locked(new);
3669
3670 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3671 NFS4_MAX_SESSIONID_LEN);
3672 cs_slot->sl_seqid++;
3673 cr_ses->seqid = cs_slot->sl_seqid;
3674
3675 /* cache solo and embedded create sessions under the client_lock */
3676 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3677 spin_unlock(&nn->client_lock);
3678 if (conf == unconf)
3679 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3680 /* init connection and backchannel */
3681 nfsd4_init_conn(rqstp, conn, new);
3682 nfsd4_put_session(new);
3683 if (old)
3684 expire_client(old);
3685 return status;
3686out_free_conn:
3687 spin_unlock(&nn->client_lock);
3688 free_conn(conn);
3689 if (old)
3690 expire_client(old);
3691out_free_session:
3692 __free_session(new);
3693out_release_drc_mem:
3694 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3695 return status;
3696}
3697
3698static __be32 nfsd4_map_bcts_dir(u32 *dir)
3699{
3700 switch (*dir) {
3701 case NFS4_CDFC4_FORE:
3702 case NFS4_CDFC4_BACK:
3703 return nfs_ok;
3704 case NFS4_CDFC4_FORE_OR_BOTH:
3705 case NFS4_CDFC4_BACK_OR_BOTH:
3706 *dir = NFS4_CDFC4_BOTH;
3707 return nfs_ok;
3708 }
3709 return nfserr_inval;
3710}
3711
3712__be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3713 struct nfsd4_compound_state *cstate,
3714 union nfsd4_op_u *u)
3715{
3716 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3717 struct nfsd4_session *session = cstate->session;
3718 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3719 __be32 status;
3720
3721 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3722 if (status)
3723 return status;
3724 spin_lock(&nn->client_lock);
3725 session->se_cb_prog = bc->bc_cb_program;
3726 session->se_cb_sec = bc->bc_cb_sec;
3727 spin_unlock(&nn->client_lock);
3728
3729 nfsd4_probe_callback(session->se_client);
3730
3731 return nfs_ok;
3732}
3733
3734static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3735{
3736 struct nfsd4_conn *c;
3737
3738 list_for_each_entry(c, &s->se_conns, cn_persession) {
3739 if (c->cn_xprt == xpt) {
3740 return c;
3741 }
3742 }
3743 return NULL;
3744}
3745
3746static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3747 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3748{
3749 struct nfs4_client *clp = session->se_client;
3750 struct svc_xprt *xpt = rqst->rq_xprt;
3751 struct nfsd4_conn *c;
3752 __be32 status;
3753
3754 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3755 spin_lock(&clp->cl_lock);
3756 c = __nfsd4_find_conn(xpt, session);
3757 if (!c)
3758 status = nfserr_noent;
3759 else if (req == c->cn_flags)
3760 status = nfs_ok;
3761 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3762 c->cn_flags != NFS4_CDFC4_BACK)
3763 status = nfs_ok;
3764 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3765 c->cn_flags != NFS4_CDFC4_FORE)
3766 status = nfs_ok;
3767 else
3768 status = nfserr_inval;
3769 spin_unlock(&clp->cl_lock);
3770 if (status == nfs_ok && conn)
3771 *conn = c;
3772 return status;
3773}
3774
3775__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3776 struct nfsd4_compound_state *cstate,
3777 union nfsd4_op_u *u)
3778{
3779 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3780 __be32 status;
3781 struct nfsd4_conn *conn;
3782 struct nfsd4_session *session;
3783 struct net *net = SVC_NET(rqstp);
3784 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3785
3786 if (!nfsd4_last_compound_op(rqstp))
3787 return nfserr_not_only_op;
3788 spin_lock(&nn->client_lock);
3789 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3790 spin_unlock(&nn->client_lock);
3791 if (!session)
3792 goto out_no_session;
3793 status = nfserr_wrong_cred;
3794 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3795 goto out;
3796 status = nfsd4_match_existing_connection(rqstp, session,
3797 bcts->dir, &conn);
3798 if (status == nfs_ok) {
3799 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
3800 bcts->dir == NFS4_CDFC4_BACK)
3801 conn->cn_flags |= NFS4_CDFC4_BACK;
3802 nfsd4_probe_callback(session->se_client);
3803 goto out;
3804 }
3805 if (status == nfserr_inval)
3806 goto out;
3807 status = nfsd4_map_bcts_dir(&bcts->dir);
3808 if (status)
3809 goto out;
3810 conn = alloc_conn(rqstp, bcts->dir);
3811 status = nfserr_jukebox;
3812 if (!conn)
3813 goto out;
3814 nfsd4_init_conn(rqstp, conn, session);
3815 status = nfs_ok;
3816out:
3817 nfsd4_put_session(session);
3818out_no_session:
3819 return status;
3820}
3821
3822static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3823{
3824 if (!cstate->session)
3825 return false;
3826 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3827}
3828
3829__be32
3830nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3831 union nfsd4_op_u *u)
3832{
3833 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3834 struct nfsd4_session *ses;
3835 __be32 status;
3836 int ref_held_by_me = 0;
3837 struct net *net = SVC_NET(r);
3838 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3839
3840 status = nfserr_not_only_op;
3841 if (nfsd4_compound_in_session(cstate, sessionid)) {
3842 if (!nfsd4_last_compound_op(r))
3843 goto out;
3844 ref_held_by_me++;
3845 }
3846 dump_sessionid(__func__, sessionid);
3847 spin_lock(&nn->client_lock);
3848 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3849 if (!ses)
3850 goto out_client_lock;
3851 status = nfserr_wrong_cred;
3852 if (!nfsd4_mach_creds_match(ses->se_client, r))
3853 goto out_put_session;
3854 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3855 if (status)
3856 goto out_put_session;
3857 unhash_session(ses);
3858 spin_unlock(&nn->client_lock);
3859
3860 nfsd4_probe_callback_sync(ses->se_client);
3861
3862 spin_lock(&nn->client_lock);
3863 status = nfs_ok;
3864out_put_session:
3865 nfsd4_put_session_locked(ses);
3866out_client_lock:
3867 spin_unlock(&nn->client_lock);
3868out:
3869 return status;
3870}
3871
3872static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3873{
3874 struct nfs4_client *clp = ses->se_client;
3875 struct nfsd4_conn *c;
3876 __be32 status = nfs_ok;
3877 int ret;
3878
3879 spin_lock(&clp->cl_lock);
3880 c = __nfsd4_find_conn(new->cn_xprt, ses);
3881 if (c)
3882 goto out_free;
3883 status = nfserr_conn_not_bound_to_session;
3884 if (clp->cl_mach_cred)
3885 goto out_free;
3886 __nfsd4_hash_conn(new, ses);
3887 spin_unlock(&clp->cl_lock);
3888 ret = nfsd4_register_conn(new);
3889 if (ret)
3890 /* oops; xprt is already down: */
3891 nfsd4_conn_lost(&new->cn_xpt_user);
3892 return nfs_ok;
3893out_free:
3894 spin_unlock(&clp->cl_lock);
3895 free_conn(new);
3896 return status;
3897}
3898
3899static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3900{
3901 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3902
3903 return args->opcnt > session->se_fchannel.maxops;
3904}
3905
3906static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3907 struct nfsd4_session *session)
3908{
3909 struct xdr_buf *xb = &rqstp->rq_arg;
3910
3911 return xb->len > session->se_fchannel.maxreq_sz;
3912}
3913
3914static bool replay_matches_cache(struct svc_rqst *rqstp,
3915 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3916{
3917 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3918
3919 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3920 (bool)seq->cachethis)
3921 return false;
3922 /*
3923 * If there's an error then the reply can have fewer ops than
3924 * the call.
3925 */
3926 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3927 return false;
3928 /*
3929 * But if we cached a reply with *more* ops than the call you're
3930 * sending us now, then this new call is clearly not really a
3931 * replay of the old one:
3932 */
3933 if (slot->sl_opcnt > argp->opcnt)
3934 return false;
3935 /* This is the only check explicitly called by spec: */
3936 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3937 return false;
3938 /*
3939 * There may be more comparisons we could actually do, but the
3940 * spec doesn't require us to catch every case where the calls
3941 * don't match (that would require caching the call as well as
3942 * the reply), so we don't bother.
3943 */
3944 return true;
3945}
3946
3947__be32
3948nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3949 union nfsd4_op_u *u)
3950{
3951 struct nfsd4_sequence *seq = &u->sequence;
3952 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3953 struct xdr_stream *xdr = resp->xdr;
3954 struct nfsd4_session *session;
3955 struct nfs4_client *clp;
3956 struct nfsd4_slot *slot;
3957 struct nfsd4_conn *conn;
3958 __be32 status;
3959 int buflen;
3960 struct net *net = SVC_NET(rqstp);
3961 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3962
3963 if (resp->opcnt != 1)
3964 return nfserr_sequence_pos;
3965
3966 /*
3967 * Will be either used or freed by nfsd4_sequence_check_conn
3968 * below.
3969 */
3970 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3971 if (!conn)
3972 return nfserr_jukebox;
3973
3974 spin_lock(&nn->client_lock);
3975 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3976 if (!session)
3977 goto out_no_session;
3978 clp = session->se_client;
3979
3980 status = nfserr_too_many_ops;
3981 if (nfsd4_session_too_many_ops(rqstp, session))
3982 goto out_put_session;
3983
3984 status = nfserr_req_too_big;
3985 if (nfsd4_request_too_big(rqstp, session))
3986 goto out_put_session;
3987
3988 status = nfserr_badslot;
3989 if (seq->slotid >= session->se_fchannel.maxreqs)
3990 goto out_put_session;
3991
3992 slot = session->se_slots[seq->slotid];
3993 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3994
3995 /* We do not negotiate the number of slots yet, so set the
3996 * maxslots to the session maxreqs which is used to encode
3997 * sr_highest_slotid and the sr_target_slot id to maxslots */
3998 seq->maxslots = session->se_fchannel.maxreqs;
3999
4000 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4001 slot->sl_flags & NFSD4_SLOT_INUSE);
4002 if (status == nfserr_replay_cache) {
4003 status = nfserr_seq_misordered;
4004 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4005 goto out_put_session;
4006 status = nfserr_seq_false_retry;
4007 if (!replay_matches_cache(rqstp, seq, slot))
4008 goto out_put_session;
4009 cstate->slot = slot;
4010 cstate->session = session;
4011 cstate->clp = clp;
4012 /* Return the cached reply status and set cstate->status
4013 * for nfsd4_proc_compound processing */
4014 status = nfsd4_replay_cache_entry(resp, seq);
4015 cstate->status = nfserr_replay_cache;
4016 goto out;
4017 }
4018 if (status)
4019 goto out_put_session;
4020
4021 status = nfsd4_sequence_check_conn(conn, session);
4022 conn = NULL;
4023 if (status)
4024 goto out_put_session;
4025
4026 buflen = (seq->cachethis) ?
4027 session->se_fchannel.maxresp_cached :
4028 session->se_fchannel.maxresp_sz;
4029 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4030 nfserr_rep_too_big;
4031 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4032 goto out_put_session;
4033 svc_reserve(rqstp, buflen);
4034
4035 status = nfs_ok;
4036 /* Success! bump slot seqid */
4037 slot->sl_seqid = seq->seqid;
4038 slot->sl_flags |= NFSD4_SLOT_INUSE;
4039 if (seq->cachethis)
4040 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4041 else
4042 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4043
4044 cstate->slot = slot;
4045 cstate->session = session;
4046 cstate->clp = clp;
4047
4048out:
4049 switch (clp->cl_cb_state) {
4050 case NFSD4_CB_DOWN:
4051 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4052 break;
4053 case NFSD4_CB_FAULT:
4054 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4055 break;
4056 default:
4057 seq->status_flags = 0;
4058 }
4059 if (!list_empty(&clp->cl_revoked))
4060 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4061out_no_session:
4062 if (conn)
4063 free_conn(conn);
4064 spin_unlock(&nn->client_lock);
4065 return status;
4066out_put_session:
4067 nfsd4_put_session_locked(session);
4068 goto out_no_session;
4069}
4070
4071void
4072nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4073{
4074 struct nfsd4_compound_state *cs = &resp->cstate;
4075
4076 if (nfsd4_has_session(cs)) {
4077 if (cs->status != nfserr_replay_cache) {
4078 nfsd4_store_cache_entry(resp);
4079 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4080 }
4081 /* Drop session reference that was taken in nfsd4_sequence() */
4082 nfsd4_put_session(cs->session);
4083 } else if (cs->clp)
4084 put_client_renew(cs->clp);
4085}
4086
4087__be32
4088nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4089 struct nfsd4_compound_state *cstate,
4090 union nfsd4_op_u *u)
4091{
4092 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4093 struct nfs4_client *conf, *unconf;
4094 struct nfs4_client *clp = NULL;
4095 __be32 status = 0;
4096 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4097
4098 spin_lock(&nn->client_lock);
4099 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4100 conf = find_confirmed_client(&dc->clientid, true, nn);
4101 WARN_ON_ONCE(conf && unconf);
4102
4103 if (conf) {
4104 if (client_has_state(conf)) {
4105 status = nfserr_clientid_busy;
4106 goto out;
4107 }
4108 status = mark_client_expired_locked(conf);
4109 if (status)
4110 goto out;
4111 clp = conf;
4112 } else if (unconf)
4113 clp = unconf;
4114 else {
4115 status = nfserr_stale_clientid;
4116 goto out;
4117 }
4118 if (!nfsd4_mach_creds_match(clp, rqstp)) {
4119 clp = NULL;
4120 status = nfserr_wrong_cred;
4121 goto out;
4122 }
4123 trace_nfsd_clid_destroyed(&clp->cl_clientid);
4124 unhash_client_locked(clp);
4125out:
4126 spin_unlock(&nn->client_lock);
4127 if (clp)
4128 expire_client(clp);
4129 return status;
4130}
4131
4132__be32
4133nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4134 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4135{
4136 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4137 struct nfs4_client *clp = cstate->clp;
4138 __be32 status = 0;
4139
4140 if (rc->rca_one_fs) {
4141 if (!cstate->current_fh.fh_dentry)
4142 return nfserr_nofilehandle;
4143 /*
4144 * We don't take advantage of the rca_one_fs case.
4145 * That's OK, it's optional, we can safely ignore it.
4146 */
4147 return nfs_ok;
4148 }
4149
4150 status = nfserr_complete_already;
4151 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4152 goto out;
4153
4154 status = nfserr_stale_clientid;
4155 if (is_client_expired(clp))
4156 /*
4157 * The following error isn't really legal.
4158 * But we only get here if the client just explicitly
4159 * destroyed the client. Surely it no longer cares what
4160 * error it gets back on an operation for the dead
4161 * client.
4162 */
4163 goto out;
4164
4165 status = nfs_ok;
4166 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4167 nfsd4_client_record_create(clp);
4168 inc_reclaim_complete(clp);
4169out:
4170 return status;
4171}
4172
4173__be32
4174nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4175 union nfsd4_op_u *u)
4176{
4177 struct nfsd4_setclientid *setclid = &u->setclientid;
4178 struct xdr_netobj clname = setclid->se_name;
4179 nfs4_verifier clverifier = setclid->se_verf;
4180 struct nfs4_client *conf, *new;
4181 struct nfs4_client *unconf = NULL;
4182 __be32 status;
4183 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4184
4185 new = create_client(clname, rqstp, &clverifier);
4186 if (new == NULL)
4187 return nfserr_jukebox;
4188 spin_lock(&nn->client_lock);
4189 conf = find_confirmed_client_by_name(&clname, nn);
4190 if (conf && client_has_state(conf)) {
4191 status = nfserr_clid_inuse;
4192 if (clp_used_exchangeid(conf))
4193 goto out;
4194 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4195 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4196 goto out;
4197 }
4198 }
4199 unconf = find_unconfirmed_client_by_name(&clname, nn);
4200 if (unconf)
4201 unhash_client_locked(unconf);
4202 if (conf) {
4203 if (same_verf(&conf->cl_verifier, &clverifier)) {
4204 copy_clid(new, conf);
4205 gen_confirm(new, nn);
4206 } else
4207 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4208 &clverifier);
4209 } else
4210 trace_nfsd_clid_fresh(new);
4211 new->cl_minorversion = 0;
4212 gen_callback(new, setclid, rqstp);
4213 add_to_unconfirmed(new);
4214 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4215 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4216 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4217 new = NULL;
4218 status = nfs_ok;
4219out:
4220 spin_unlock(&nn->client_lock);
4221 if (new)
4222 free_client(new);
4223 if (unconf) {
4224 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4225 expire_client(unconf);
4226 }
4227 return status;
4228}
4229
4230__be32
4231nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4232 struct nfsd4_compound_state *cstate,
4233 union nfsd4_op_u *u)
4234{
4235 struct nfsd4_setclientid_confirm *setclientid_confirm =
4236 &u->setclientid_confirm;
4237 struct nfs4_client *conf, *unconf;
4238 struct nfs4_client *old = NULL;
4239 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4240 clientid_t * clid = &setclientid_confirm->sc_clientid;
4241 __be32 status;
4242 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4243
4244 if (STALE_CLIENTID(clid, nn))
4245 return nfserr_stale_clientid;
4246
4247 spin_lock(&nn->client_lock);
4248 conf = find_confirmed_client(clid, false, nn);
4249 unconf = find_unconfirmed_client(clid, false, nn);
4250 /*
4251 * We try hard to give out unique clientid's, so if we get an
4252 * attempt to confirm the same clientid with a different cred,
4253 * the client may be buggy; this should never happen.
4254 *
4255 * Nevertheless, RFC 7530 recommends INUSE for this case:
4256 */
4257 status = nfserr_clid_inuse;
4258 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4259 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4260 goto out;
4261 }
4262 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4263 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4264 goto out;
4265 }
4266 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4267 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4268 status = nfs_ok;
4269 } else
4270 status = nfserr_stale_clientid;
4271 goto out;
4272 }
4273 status = nfs_ok;
4274 if (conf) {
4275 old = unconf;
4276 unhash_client_locked(old);
4277 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4278 } else {
4279 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4280 if (old) {
4281 status = nfserr_clid_inuse;
4282 if (client_has_state(old)
4283 && !same_creds(&unconf->cl_cred,
4284 &old->cl_cred)) {
4285 old = NULL;
4286 goto out;
4287 }
4288 status = mark_client_expired_locked(old);
4289 if (status) {
4290 old = NULL;
4291 goto out;
4292 }
4293 trace_nfsd_clid_replaced(&old->cl_clientid);
4294 }
4295 move_to_confirmed(unconf);
4296 conf = unconf;
4297 }
4298 get_client_locked(conf);
4299 spin_unlock(&nn->client_lock);
4300 if (conf == unconf)
4301 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4302 nfsd4_probe_callback(conf);
4303 spin_lock(&nn->client_lock);
4304 put_client_renew_locked(conf);
4305out:
4306 spin_unlock(&nn->client_lock);
4307 if (old)
4308 expire_client(old);
4309 return status;
4310}
4311
4312static struct nfs4_file *nfsd4_alloc_file(void)
4313{
4314 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4315}
4316
4317/* OPEN Share state helper functions */
4318
4319static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4320{
4321 refcount_set(&fp->fi_ref, 1);
4322 spin_lock_init(&fp->fi_lock);
4323 INIT_LIST_HEAD(&fp->fi_stateids);
4324 INIT_LIST_HEAD(&fp->fi_delegations);
4325 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4326 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4327 fp->fi_deleg_file = NULL;
4328 fp->fi_had_conflict = false;
4329 fp->fi_share_deny = 0;
4330 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4331 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4332 fp->fi_aliased = false;
4333 fp->fi_inode = d_inode(fh->fh_dentry);
4334#ifdef CONFIG_NFSD_PNFS
4335 INIT_LIST_HEAD(&fp->fi_lo_states);
4336 atomic_set(&fp->fi_lo_recalls, 0);
4337#endif
4338}
4339
4340void
4341nfsd4_free_slabs(void)
4342{
4343 kmem_cache_destroy(client_slab);
4344 kmem_cache_destroy(openowner_slab);
4345 kmem_cache_destroy(lockowner_slab);
4346 kmem_cache_destroy(file_slab);
4347 kmem_cache_destroy(stateid_slab);
4348 kmem_cache_destroy(deleg_slab);
4349 kmem_cache_destroy(odstate_slab);
4350}
4351
4352int
4353nfsd4_init_slabs(void)
4354{
4355 client_slab = kmem_cache_create("nfsd4_clients",
4356 sizeof(struct nfs4_client), 0, 0, NULL);
4357 if (client_slab == NULL)
4358 goto out;
4359 openowner_slab = kmem_cache_create("nfsd4_openowners",
4360 sizeof(struct nfs4_openowner), 0, 0, NULL);
4361 if (openowner_slab == NULL)
4362 goto out_free_client_slab;
4363 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4364 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4365 if (lockowner_slab == NULL)
4366 goto out_free_openowner_slab;
4367 file_slab = kmem_cache_create("nfsd4_files",
4368 sizeof(struct nfs4_file), 0, 0, NULL);
4369 if (file_slab == NULL)
4370 goto out_free_lockowner_slab;
4371 stateid_slab = kmem_cache_create("nfsd4_stateids",
4372 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4373 if (stateid_slab == NULL)
4374 goto out_free_file_slab;
4375 deleg_slab = kmem_cache_create("nfsd4_delegations",
4376 sizeof(struct nfs4_delegation), 0, 0, NULL);
4377 if (deleg_slab == NULL)
4378 goto out_free_stateid_slab;
4379 odstate_slab = kmem_cache_create("nfsd4_odstate",
4380 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4381 if (odstate_slab == NULL)
4382 goto out_free_deleg_slab;
4383 return 0;
4384
4385out_free_deleg_slab:
4386 kmem_cache_destroy(deleg_slab);
4387out_free_stateid_slab:
4388 kmem_cache_destroy(stateid_slab);
4389out_free_file_slab:
4390 kmem_cache_destroy(file_slab);
4391out_free_lockowner_slab:
4392 kmem_cache_destroy(lockowner_slab);
4393out_free_openowner_slab:
4394 kmem_cache_destroy(openowner_slab);
4395out_free_client_slab:
4396 kmem_cache_destroy(client_slab);
4397out:
4398 return -ENOMEM;
4399}
4400
4401static unsigned long
4402nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4403{
4404 int count;
4405 struct nfsd_net *nn = shrink->private_data;
4406
4407 count = atomic_read(&nn->nfsd_courtesy_clients);
4408 if (!count)
4409 count = atomic_long_read(&num_delegations);
4410 if (count)
4411 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4412 return (unsigned long)count;
4413}
4414
4415static unsigned long
4416nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4417{
4418 return SHRINK_STOP;
4419}
4420
4421void
4422nfsd4_init_leases_net(struct nfsd_net *nn)
4423{
4424 struct sysinfo si;
4425 u64 max_clients;
4426
4427 nn->nfsd4_lease = 90; /* default lease time */
4428 nn->nfsd4_grace = 90;
4429 nn->somebody_reclaimed = false;
4430 nn->track_reclaim_completes = false;
4431 nn->clverifier_counter = get_random_u32();
4432 nn->clientid_base = get_random_u32();
4433 nn->clientid_counter = nn->clientid_base + 1;
4434 nn->s2s_cp_cl_id = nn->clientid_counter++;
4435
4436 atomic_set(&nn->nfs4_client_count, 0);
4437 si_meminfo(&si);
4438 max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4439 max_clients *= NFS4_CLIENTS_PER_GB;
4440 nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4441
4442 atomic_set(&nn->nfsd_courtesy_clients, 0);
4443}
4444
4445static void init_nfs4_replay(struct nfs4_replay *rp)
4446{
4447 rp->rp_status = nfserr_serverfault;
4448 rp->rp_buflen = 0;
4449 rp->rp_buf = rp->rp_ibuf;
4450 mutex_init(&rp->rp_mutex);
4451}
4452
4453static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4454 struct nfs4_stateowner *so)
4455{
4456 if (!nfsd4_has_session(cstate)) {
4457 mutex_lock(&so->so_replay.rp_mutex);
4458 cstate->replay_owner = nfs4_get_stateowner(so);
4459 }
4460}
4461
4462void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4463{
4464 struct nfs4_stateowner *so = cstate->replay_owner;
4465
4466 if (so != NULL) {
4467 cstate->replay_owner = NULL;
4468 mutex_unlock(&so->so_replay.rp_mutex);
4469 nfs4_put_stateowner(so);
4470 }
4471}
4472
4473static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4474{
4475 struct nfs4_stateowner *sop;
4476
4477 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4478 if (!sop)
4479 return NULL;
4480
4481 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4482 if (!sop->so_owner.data) {
4483 kmem_cache_free(slab, sop);
4484 return NULL;
4485 }
4486
4487 INIT_LIST_HEAD(&sop->so_stateids);
4488 sop->so_client = clp;
4489 init_nfs4_replay(&sop->so_replay);
4490 atomic_set(&sop->so_count, 1);
4491 return sop;
4492}
4493
4494static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4495{
4496 lockdep_assert_held(&clp->cl_lock);
4497
4498 list_add(&oo->oo_owner.so_strhash,
4499 &clp->cl_ownerstr_hashtbl[strhashval]);
4500 list_add(&oo->oo_perclient, &clp->cl_openowners);
4501}
4502
4503static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4504{
4505 unhash_openowner_locked(openowner(so));
4506}
4507
4508static void nfs4_free_openowner(struct nfs4_stateowner *so)
4509{
4510 struct nfs4_openowner *oo = openowner(so);
4511
4512 kmem_cache_free(openowner_slab, oo);
4513}
4514
4515static const struct nfs4_stateowner_operations openowner_ops = {
4516 .so_unhash = nfs4_unhash_openowner,
4517 .so_free = nfs4_free_openowner,
4518};
4519
4520static struct nfs4_ol_stateid *
4521nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4522{
4523 struct nfs4_ol_stateid *local, *ret = NULL;
4524 struct nfs4_openowner *oo = open->op_openowner;
4525
4526 lockdep_assert_held(&fp->fi_lock);
4527
4528 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4529 /* ignore lock owners */
4530 if (local->st_stateowner->so_is_open_owner == 0)
4531 continue;
4532 if (local->st_stateowner != &oo->oo_owner)
4533 continue;
4534 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4535 ret = local;
4536 refcount_inc(&ret->st_stid.sc_count);
4537 break;
4538 }
4539 }
4540 return ret;
4541}
4542
4543static __be32
4544nfsd4_verify_open_stid(struct nfs4_stid *s)
4545{
4546 __be32 ret = nfs_ok;
4547
4548 switch (s->sc_type) {
4549 default:
4550 break;
4551 case 0:
4552 case NFS4_CLOSED_STID:
4553 case NFS4_CLOSED_DELEG_STID:
4554 ret = nfserr_bad_stateid;
4555 break;
4556 case NFS4_REVOKED_DELEG_STID:
4557 ret = nfserr_deleg_revoked;
4558 }
4559 return ret;
4560}
4561
4562/* Lock the stateid st_mutex, and deal with races with CLOSE */
4563static __be32
4564nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4565{
4566 __be32 ret;
4567
4568 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4569 ret = nfsd4_verify_open_stid(&stp->st_stid);
4570 if (ret != nfs_ok)
4571 mutex_unlock(&stp->st_mutex);
4572 return ret;
4573}
4574
4575static struct nfs4_ol_stateid *
4576nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4577{
4578 struct nfs4_ol_stateid *stp;
4579 for (;;) {
4580 spin_lock(&fp->fi_lock);
4581 stp = nfsd4_find_existing_open(fp, open);
4582 spin_unlock(&fp->fi_lock);
4583 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4584 break;
4585 nfs4_put_stid(&stp->st_stid);
4586 }
4587 return stp;
4588}
4589
4590static struct nfs4_openowner *
4591alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4592 struct nfsd4_compound_state *cstate)
4593{
4594 struct nfs4_client *clp = cstate->clp;
4595 struct nfs4_openowner *oo, *ret;
4596
4597 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4598 if (!oo)
4599 return NULL;
4600 oo->oo_owner.so_ops = &openowner_ops;
4601 oo->oo_owner.so_is_open_owner = 1;
4602 oo->oo_owner.so_seqid = open->op_seqid;
4603 oo->oo_flags = 0;
4604 if (nfsd4_has_session(cstate))
4605 oo->oo_flags |= NFS4_OO_CONFIRMED;
4606 oo->oo_time = 0;
4607 oo->oo_last_closed_stid = NULL;
4608 INIT_LIST_HEAD(&oo->oo_close_lru);
4609 spin_lock(&clp->cl_lock);
4610 ret = find_openstateowner_str_locked(strhashval, open, clp);
4611 if (ret == NULL) {
4612 hash_openowner(oo, clp, strhashval);
4613 ret = oo;
4614 } else
4615 nfs4_free_stateowner(&oo->oo_owner);
4616
4617 spin_unlock(&clp->cl_lock);
4618 return ret;
4619}
4620
4621static struct nfs4_ol_stateid *
4622init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4623{
4624
4625 struct nfs4_openowner *oo = open->op_openowner;
4626 struct nfs4_ol_stateid *retstp = NULL;
4627 struct nfs4_ol_stateid *stp;
4628
4629 stp = open->op_stp;
4630 /* We are moving these outside of the spinlocks to avoid the warnings */
4631 mutex_init(&stp->st_mutex);
4632 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4633
4634retry:
4635 spin_lock(&oo->oo_owner.so_client->cl_lock);
4636 spin_lock(&fp->fi_lock);
4637
4638 retstp = nfsd4_find_existing_open(fp, open);
4639 if (retstp)
4640 goto out_unlock;
4641
4642 open->op_stp = NULL;
4643 refcount_inc(&stp->st_stid.sc_count);
4644 stp->st_stid.sc_type = NFS4_OPEN_STID;
4645 INIT_LIST_HEAD(&stp->st_locks);
4646 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4647 get_nfs4_file(fp);
4648 stp->st_stid.sc_file = fp;
4649 stp->st_access_bmap = 0;
4650 stp->st_deny_bmap = 0;
4651 stp->st_openstp = NULL;
4652 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4653 list_add(&stp->st_perfile, &fp->fi_stateids);
4654
4655out_unlock:
4656 spin_unlock(&fp->fi_lock);
4657 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4658 if (retstp) {
4659 /* Handle races with CLOSE */
4660 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4661 nfs4_put_stid(&retstp->st_stid);
4662 goto retry;
4663 }
4664 /* To keep mutex tracking happy */
4665 mutex_unlock(&stp->st_mutex);
4666 stp = retstp;
4667 }
4668 return stp;
4669}
4670
4671/*
4672 * In the 4.0 case we need to keep the owners around a little while to handle
4673 * CLOSE replay. We still do need to release any file access that is held by
4674 * them before returning however.
4675 */
4676static void
4677move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4678{
4679 struct nfs4_ol_stateid *last;
4680 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4681 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4682 nfsd_net_id);
4683
4684 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4685
4686 /*
4687 * We know that we hold one reference via nfsd4_close, and another
4688 * "persistent" reference for the client. If the refcount is higher
4689 * than 2, then there are still calls in progress that are using this
4690 * stateid. We can't put the sc_file reference until they are finished.
4691 * Wait for the refcount to drop to 2. Since it has been unhashed,
4692 * there should be no danger of the refcount going back up again at
4693 * this point.
4694 */
4695 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4696
4697 release_all_access(s);
4698 if (s->st_stid.sc_file) {
4699 put_nfs4_file(s->st_stid.sc_file);
4700 s->st_stid.sc_file = NULL;
4701 }
4702
4703 spin_lock(&nn->client_lock);
4704 last = oo->oo_last_closed_stid;
4705 oo->oo_last_closed_stid = s;
4706 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4707 oo->oo_time = ktime_get_boottime_seconds();
4708 spin_unlock(&nn->client_lock);
4709 if (last)
4710 nfs4_put_stid(&last->st_stid);
4711}
4712
4713static noinline_for_stack struct nfs4_file *
4714nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4715{
4716 struct inode *inode = d_inode(fhp->fh_dentry);
4717 struct rhlist_head *tmp, *list;
4718 struct nfs4_file *fi;
4719
4720 rcu_read_lock();
4721 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4722 nfs4_file_rhash_params);
4723 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4724 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4725 if (refcount_inc_not_zero(&fi->fi_ref)) {
4726 rcu_read_unlock();
4727 return fi;
4728 }
4729 }
4730 }
4731 rcu_read_unlock();
4732 return NULL;
4733}
4734
4735/*
4736 * On hash insertion, identify entries with the same inode but
4737 * distinct filehandles. They will all be on the list returned
4738 * by rhltable_lookup().
4739 *
4740 * inode->i_lock prevents racing insertions from adding an entry
4741 * for the same inode/fhp pair twice.
4742 */
4743static noinline_for_stack struct nfs4_file *
4744nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
4745{
4746 struct inode *inode = d_inode(fhp->fh_dentry);
4747 struct rhlist_head *tmp, *list;
4748 struct nfs4_file *ret = NULL;
4749 bool alias_found = false;
4750 struct nfs4_file *fi;
4751 int err;
4752
4753 rcu_read_lock();
4754 spin_lock(&inode->i_lock);
4755
4756 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4757 nfs4_file_rhash_params);
4758 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4759 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4760 if (refcount_inc_not_zero(&fi->fi_ref))
4761 ret = fi;
4762 } else
4763 fi->fi_aliased = alias_found = true;
4764 }
4765 if (ret)
4766 goto out_unlock;
4767
4768 nfsd4_file_init(fhp, new);
4769 err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
4770 nfs4_file_rhash_params);
4771 if (err)
4772 goto out_unlock;
4773
4774 new->fi_aliased = alias_found;
4775 ret = new;
4776
4777out_unlock:
4778 spin_unlock(&inode->i_lock);
4779 rcu_read_unlock();
4780 return ret;
4781}
4782
4783static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
4784{
4785 rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
4786 nfs4_file_rhash_params);
4787}
4788
4789/*
4790 * Called to check deny when READ with all zero stateid or
4791 * WRITE with all zero or all one stateid
4792 */
4793static __be32
4794nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4795{
4796 struct nfs4_file *fp;
4797 __be32 ret = nfs_ok;
4798
4799 fp = nfsd4_file_hash_lookup(current_fh);
4800 if (!fp)
4801 return ret;
4802
4803 /* Check for conflicting share reservations */
4804 spin_lock(&fp->fi_lock);
4805 if (fp->fi_share_deny & deny_type)
4806 ret = nfserr_locked;
4807 spin_unlock(&fp->fi_lock);
4808 put_nfs4_file(fp);
4809 return ret;
4810}
4811
4812static bool nfsd4_deleg_present(const struct inode *inode)
4813{
4814 struct file_lock_context *ctx = locks_inode_context(inode);
4815
4816 return ctx && !list_empty_careful(&ctx->flc_lease);
4817}
4818
4819/**
4820 * nfsd_wait_for_delegreturn - wait for delegations to be returned
4821 * @rqstp: the RPC transaction being executed
4822 * @inode: in-core inode of the file being waited for
4823 *
4824 * The timeout prevents deadlock if all nfsd threads happen to be
4825 * tied up waiting for returning delegations.
4826 *
4827 * Return values:
4828 * %true: delegation was returned
4829 * %false: timed out waiting for delegreturn
4830 */
4831bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
4832{
4833 long __maybe_unused timeo;
4834
4835 timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
4836 NFSD_DELEGRETURN_TIMEOUT);
4837 trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
4838 return timeo > 0;
4839}
4840
4841static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4842{
4843 struct nfs4_delegation *dp = cb_to_delegation(cb);
4844 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4845 nfsd_net_id);
4846
4847 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4848
4849 /*
4850 * We can't do this in nfsd_break_deleg_cb because it is
4851 * already holding inode->i_lock.
4852 *
4853 * If the dl_time != 0, then we know that it has already been
4854 * queued for a lease break. Don't queue it again.
4855 */
4856 spin_lock(&state_lock);
4857 if (delegation_hashed(dp) && dp->dl_time == 0) {
4858 dp->dl_time = ktime_get_boottime_seconds();
4859 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4860 }
4861 spin_unlock(&state_lock);
4862}
4863
4864static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4865 struct rpc_task *task)
4866{
4867 struct nfs4_delegation *dp = cb_to_delegation(cb);
4868
4869 trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
4870
4871 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4872 dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4873 return 1;
4874
4875 switch (task->tk_status) {
4876 case 0:
4877 return 1;
4878 case -NFS4ERR_DELAY:
4879 rpc_delay(task, 2 * HZ);
4880 return 0;
4881 case -EBADHANDLE:
4882 case -NFS4ERR_BAD_STATEID:
4883 /*
4884 * Race: client probably got cb_recall before open reply
4885 * granting delegation.
4886 */
4887 if (dp->dl_retries--) {
4888 rpc_delay(task, 2 * HZ);
4889 return 0;
4890 }
4891 fallthrough;
4892 default:
4893 return 1;
4894 }
4895}
4896
4897static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4898{
4899 struct nfs4_delegation *dp = cb_to_delegation(cb);
4900
4901 nfs4_put_stid(&dp->dl_stid);
4902}
4903
4904static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4905 .prepare = nfsd4_cb_recall_prepare,
4906 .done = nfsd4_cb_recall_done,
4907 .release = nfsd4_cb_recall_release,
4908};
4909
4910static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4911{
4912 /*
4913 * We're assuming the state code never drops its reference
4914 * without first removing the lease. Since we're in this lease
4915 * callback (and since the lease code is serialized by the
4916 * flc_lock) we know the server hasn't removed the lease yet, and
4917 * we know it's safe to take a reference.
4918 */
4919 refcount_inc(&dp->dl_stid.sc_count);
4920 WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
4921}
4922
4923/* Called from break_lease() with flc_lock held. */
4924static bool
4925nfsd_break_deleg_cb(struct file_lock *fl)
4926{
4927 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4928 struct nfs4_file *fp = dp->dl_stid.sc_file;
4929 struct nfs4_client *clp = dp->dl_stid.sc_client;
4930 struct nfsd_net *nn;
4931
4932 trace_nfsd_cb_recall(&dp->dl_stid);
4933
4934 dp->dl_recalled = true;
4935 atomic_inc(&clp->cl_delegs_in_recall);
4936 if (try_to_expire_client(clp)) {
4937 nn = net_generic(clp->net, nfsd_net_id);
4938 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
4939 }
4940
4941 /*
4942 * We don't want the locks code to timeout the lease for us;
4943 * we'll remove it ourself if a delegation isn't returned
4944 * in time:
4945 */
4946 fl->fl_break_time = 0;
4947
4948 fp->fi_had_conflict = true;
4949 nfsd_break_one_deleg(dp);
4950 return false;
4951}
4952
4953/**
4954 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
4955 * @fl: Lock state to check
4956 *
4957 * Return values:
4958 * %true: Lease conflict was resolved
4959 * %false: Lease conflict was not resolved.
4960 */
4961static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4962{
4963 struct nfs4_delegation *dl = fl->fl_owner;
4964 struct svc_rqst *rqst;
4965 struct nfs4_client *clp;
4966
4967 if (!i_am_nfsd())
4968 return false;
4969 rqst = kthread_data(current);
4970 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4971 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4972 return false;
4973 clp = *(rqst->rq_lease_breaker);
4974 return dl->dl_stid.sc_client == clp;
4975}
4976
4977static int
4978nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4979 struct list_head *dispose)
4980{
4981 struct nfs4_delegation *dp = (struct nfs4_delegation *)onlist->fl_owner;
4982 struct nfs4_client *clp = dp->dl_stid.sc_client;
4983
4984 if (arg & F_UNLCK) {
4985 if (dp->dl_recalled)
4986 atomic_dec(&clp->cl_delegs_in_recall);
4987 return lease_modify(onlist, arg, dispose);
4988 } else
4989 return -EAGAIN;
4990}
4991
4992static const struct lock_manager_operations nfsd_lease_mng_ops = {
4993 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4994 .lm_break = nfsd_break_deleg_cb,
4995 .lm_change = nfsd_change_deleg_cb,
4996};
4997
4998static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4999{
5000 if (nfsd4_has_session(cstate))
5001 return nfs_ok;
5002 if (seqid == so->so_seqid - 1)
5003 return nfserr_replay_me;
5004 if (seqid == so->so_seqid)
5005 return nfs_ok;
5006 return nfserr_bad_seqid;
5007}
5008
5009static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5010 struct nfsd_net *nn)
5011{
5012 struct nfs4_client *found;
5013
5014 spin_lock(&nn->client_lock);
5015 found = find_confirmed_client(clid, sessions, nn);
5016 if (found)
5017 atomic_inc(&found->cl_rpc_users);
5018 spin_unlock(&nn->client_lock);
5019 return found;
5020}
5021
5022static __be32 set_client(clientid_t *clid,
5023 struct nfsd4_compound_state *cstate,
5024 struct nfsd_net *nn)
5025{
5026 if (cstate->clp) {
5027 if (!same_clid(&cstate->clp->cl_clientid, clid))
5028 return nfserr_stale_clientid;
5029 return nfs_ok;
5030 }
5031 if (STALE_CLIENTID(clid, nn))
5032 return nfserr_stale_clientid;
5033 /*
5034 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5035 * set cstate->clp), so session = false:
5036 */
5037 cstate->clp = lookup_clientid(clid, false, nn);
5038 if (!cstate->clp)
5039 return nfserr_expired;
5040 return nfs_ok;
5041}
5042
5043__be32
5044nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5045 struct nfsd4_open *open, struct nfsd_net *nn)
5046{
5047 clientid_t *clientid = &open->op_clientid;
5048 struct nfs4_client *clp = NULL;
5049 unsigned int strhashval;
5050 struct nfs4_openowner *oo = NULL;
5051 __be32 status;
5052
5053 /*
5054 * In case we need it later, after we've already created the
5055 * file and don't want to risk a further failure:
5056 */
5057 open->op_file = nfsd4_alloc_file();
5058 if (open->op_file == NULL)
5059 return nfserr_jukebox;
5060
5061 status = set_client(clientid, cstate, nn);
5062 if (status)
5063 return status;
5064 clp = cstate->clp;
5065
5066 strhashval = ownerstr_hashval(&open->op_owner);
5067 oo = find_openstateowner_str(strhashval, open, clp);
5068 open->op_openowner = oo;
5069 if (!oo) {
5070 goto new_owner;
5071 }
5072 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5073 /* Replace unconfirmed owners without checking for replay. */
5074 release_openowner(oo);
5075 open->op_openowner = NULL;
5076 goto new_owner;
5077 }
5078 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5079 if (status)
5080 return status;
5081 goto alloc_stateid;
5082new_owner:
5083 oo = alloc_init_open_stateowner(strhashval, open, cstate);
5084 if (oo == NULL)
5085 return nfserr_jukebox;
5086 open->op_openowner = oo;
5087alloc_stateid:
5088 open->op_stp = nfs4_alloc_open_stateid(clp);
5089 if (!open->op_stp)
5090 return nfserr_jukebox;
5091
5092 if (nfsd4_has_session(cstate) &&
5093 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5094 open->op_odstate = alloc_clnt_odstate(clp);
5095 if (!open->op_odstate)
5096 return nfserr_jukebox;
5097 }
5098
5099 return nfs_ok;
5100}
5101
5102static inline __be32
5103nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5104{
5105 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5106 return nfserr_openmode;
5107 else
5108 return nfs_ok;
5109}
5110
5111static int share_access_to_flags(u32 share_access)
5112{
5113 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5114}
5115
5116static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
5117{
5118 struct nfs4_stid *ret;
5119
5120 ret = find_stateid_by_type(cl, s,
5121 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
5122 if (!ret)
5123 return NULL;
5124 return delegstateid(ret);
5125}
5126
5127static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5128{
5129 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5130 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5131}
5132
5133static __be32
5134nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5135 struct nfs4_delegation **dp)
5136{
5137 int flags;
5138 __be32 status = nfserr_bad_stateid;
5139 struct nfs4_delegation *deleg;
5140
5141 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5142 if (deleg == NULL)
5143 goto out;
5144 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
5145 nfs4_put_stid(&deleg->dl_stid);
5146 if (cl->cl_minorversion)
5147 status = nfserr_deleg_revoked;
5148 goto out;
5149 }
5150 flags = share_access_to_flags(open->op_share_access);
5151 status = nfs4_check_delegmode(deleg, flags);
5152 if (status) {
5153 nfs4_put_stid(&deleg->dl_stid);
5154 goto out;
5155 }
5156 *dp = deleg;
5157out:
5158 if (!nfsd4_is_deleg_cur(open))
5159 return nfs_ok;
5160 if (status)
5161 return status;
5162 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5163 return nfs_ok;
5164}
5165
5166static inline int nfs4_access_to_access(u32 nfs4_access)
5167{
5168 int flags = 0;
5169
5170 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5171 flags |= NFSD_MAY_READ;
5172 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5173 flags |= NFSD_MAY_WRITE;
5174 return flags;
5175}
5176
5177static inline __be32
5178nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5179 struct nfsd4_open *open)
5180{
5181 struct iattr iattr = {
5182 .ia_valid = ATTR_SIZE,
5183 .ia_size = 0,
5184 };
5185 struct nfsd_attrs attrs = {
5186 .na_iattr = &iattr,
5187 };
5188 if (!open->op_truncate)
5189 return 0;
5190 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5191 return nfserr_inval;
5192 return nfsd_setattr(rqstp, fh, &attrs, 0, (time64_t)0);
5193}
5194
5195static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5196 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5197 struct nfsd4_open *open, bool new_stp)
5198{
5199 struct nfsd_file *nf = NULL;
5200 __be32 status;
5201 int oflag = nfs4_access_to_omode(open->op_share_access);
5202 int access = nfs4_access_to_access(open->op_share_access);
5203 unsigned char old_access_bmap, old_deny_bmap;
5204
5205 spin_lock(&fp->fi_lock);
5206
5207 /*
5208 * Are we trying to set a deny mode that would conflict with
5209 * current access?
5210 */
5211 status = nfs4_file_check_deny(fp, open->op_share_deny);
5212 if (status != nfs_ok) {
5213 if (status != nfserr_share_denied) {
5214 spin_unlock(&fp->fi_lock);
5215 goto out;
5216 }
5217 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5218 stp, open->op_share_deny, false))
5219 status = nfserr_jukebox;
5220 spin_unlock(&fp->fi_lock);
5221 goto out;
5222 }
5223
5224 /* set access to the file */
5225 status = nfs4_file_get_access(fp, open->op_share_access);
5226 if (status != nfs_ok) {
5227 if (status != nfserr_share_denied) {
5228 spin_unlock(&fp->fi_lock);
5229 goto out;
5230 }
5231 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5232 stp, open->op_share_access, true))
5233 status = nfserr_jukebox;
5234 spin_unlock(&fp->fi_lock);
5235 goto out;
5236 }
5237
5238 /* Set access bits in stateid */
5239 old_access_bmap = stp->st_access_bmap;
5240 set_access(open->op_share_access, stp);
5241
5242 /* Set new deny mask */
5243 old_deny_bmap = stp->st_deny_bmap;
5244 set_deny(open->op_share_deny, stp);
5245 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5246
5247 if (!fp->fi_fds[oflag]) {
5248 spin_unlock(&fp->fi_lock);
5249
5250 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5251 open->op_filp, &nf);
5252 if (status != nfs_ok)
5253 goto out_put_access;
5254
5255 spin_lock(&fp->fi_lock);
5256 if (!fp->fi_fds[oflag]) {
5257 fp->fi_fds[oflag] = nf;
5258 nf = NULL;
5259 }
5260 }
5261 spin_unlock(&fp->fi_lock);
5262 if (nf)
5263 nfsd_file_put(nf);
5264
5265 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5266 access));
5267 if (status)
5268 goto out_put_access;
5269
5270 status = nfsd4_truncate(rqstp, cur_fh, open);
5271 if (status)
5272 goto out_put_access;
5273out:
5274 return status;
5275out_put_access:
5276 stp->st_access_bmap = old_access_bmap;
5277 nfs4_file_put_access(fp, open->op_share_access);
5278 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5279 goto out;
5280}
5281
5282static __be32
5283nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5284 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5285 struct nfsd4_open *open)
5286{
5287 __be32 status;
5288 unsigned char old_deny_bmap = stp->st_deny_bmap;
5289
5290 if (!test_access(open->op_share_access, stp))
5291 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5292
5293 /* test and set deny mode */
5294 spin_lock(&fp->fi_lock);
5295 status = nfs4_file_check_deny(fp, open->op_share_deny);
5296 switch (status) {
5297 case nfs_ok:
5298 set_deny(open->op_share_deny, stp);
5299 fp->fi_share_deny |=
5300 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5301 break;
5302 case nfserr_share_denied:
5303 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5304 stp, open->op_share_deny, false))
5305 status = nfserr_jukebox;
5306 break;
5307 }
5308 spin_unlock(&fp->fi_lock);
5309
5310 if (status != nfs_ok)
5311 return status;
5312
5313 status = nfsd4_truncate(rqstp, cur_fh, open);
5314 if (status != nfs_ok)
5315 reset_union_bmap_deny(old_deny_bmap, stp);
5316 return status;
5317}
5318
5319/* Should we give out recallable state?: */
5320static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5321{
5322 if (clp->cl_cb_state == NFSD4_CB_UP)
5323 return true;
5324 /*
5325 * In the sessions case, since we don't have to establish a
5326 * separate connection for callbacks, we assume it's OK
5327 * until we hear otherwise:
5328 */
5329 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5330}
5331
5332static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5333 int flag)
5334{
5335 struct file_lock *fl;
5336
5337 fl = locks_alloc_lock();
5338 if (!fl)
5339 return NULL;
5340 fl->fl_lmops = &nfsd_lease_mng_ops;
5341 fl->fl_flags = FL_DELEG;
5342 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5343 fl->fl_end = OFFSET_MAX;
5344 fl->fl_owner = (fl_owner_t)dp;
5345 fl->fl_pid = current->tgid;
5346 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5347 return fl;
5348}
5349
5350static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5351 struct nfs4_file *fp)
5352{
5353 struct nfs4_ol_stateid *st;
5354 struct file *f = fp->fi_deleg_file->nf_file;
5355 struct inode *ino = file_inode(f);
5356 int writes;
5357
5358 writes = atomic_read(&ino->i_writecount);
5359 if (!writes)
5360 return 0;
5361 /*
5362 * There could be multiple filehandles (hence multiple
5363 * nfs4_files) referencing this file, but that's not too
5364 * common; let's just give up in that case rather than
5365 * trying to go look up all the clients using that other
5366 * nfs4_file as well:
5367 */
5368 if (fp->fi_aliased)
5369 return -EAGAIN;
5370 /*
5371 * If there's a close in progress, make sure that we see it
5372 * clear any fi_fds[] entries before we see it decrement
5373 * i_writecount:
5374 */
5375 smp_mb__after_atomic();
5376
5377 if (fp->fi_fds[O_WRONLY])
5378 writes--;
5379 if (fp->fi_fds[O_RDWR])
5380 writes--;
5381 if (writes > 0)
5382 return -EAGAIN; /* There may be non-NFSv4 writers */
5383 /*
5384 * It's possible there are non-NFSv4 write opens in progress,
5385 * but if they haven't incremented i_writecount yet then they
5386 * also haven't called break lease yet; so, they'll break this
5387 * lease soon enough. So, all that's left to check for is NFSv4
5388 * opens:
5389 */
5390 spin_lock(&fp->fi_lock);
5391 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5392 if (st->st_openstp == NULL /* it's an open */ &&
5393 access_permit_write(st) &&
5394 st->st_stid.sc_client != clp) {
5395 spin_unlock(&fp->fi_lock);
5396 return -EAGAIN;
5397 }
5398 }
5399 spin_unlock(&fp->fi_lock);
5400 /*
5401 * There's a small chance that we could be racing with another
5402 * NFSv4 open. However, any open that hasn't added itself to
5403 * the fi_stateids list also hasn't called break_lease yet; so,
5404 * they'll break this lease soon enough.
5405 */
5406 return 0;
5407}
5408
5409/*
5410 * It's possible that between opening the dentry and setting the delegation,
5411 * that it has been renamed or unlinked. Redo the lookup to verify that this
5412 * hasn't happened.
5413 */
5414static int
5415nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5416 struct svc_fh *parent)
5417{
5418 struct svc_export *exp;
5419 struct dentry *child;
5420 __be32 err;
5421
5422 err = nfsd_lookup_dentry(open->op_rqstp, parent,
5423 open->op_fname, open->op_fnamelen,
5424 &exp, &child);
5425
5426 if (err)
5427 return -EAGAIN;
5428
5429 exp_put(exp);
5430 dput(child);
5431 if (child != file_dentry(fp->fi_deleg_file->nf_file))
5432 return -EAGAIN;
5433
5434 return 0;
5435}
5436
5437/*
5438 * We avoid breaking delegations held by a client due to its own activity, but
5439 * clearing setuid/setgid bits on a write is an implicit activity and the client
5440 * may not notice and continue using the old mode. Avoid giving out a delegation
5441 * on setuid/setgid files when the client is requesting an open for write.
5442 */
5443static int
5444nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5445{
5446 struct inode *inode = file_inode(nf->nf_file);
5447
5448 if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5449 (inode->i_mode & (S_ISUID|S_ISGID)))
5450 return -EAGAIN;
5451 return 0;
5452}
5453
5454static struct nfs4_delegation *
5455nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5456 struct svc_fh *parent)
5457{
5458 int status = 0;
5459 struct nfs4_client *clp = stp->st_stid.sc_client;
5460 struct nfs4_file *fp = stp->st_stid.sc_file;
5461 struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5462 struct nfs4_delegation *dp;
5463 struct nfsd_file *nf = NULL;
5464 struct file_lock *fl;
5465 u32 dl_type;
5466
5467 /*
5468 * The fi_had_conflict and nfs_get_existing_delegation checks
5469 * here are just optimizations; we'll need to recheck them at
5470 * the end:
5471 */
5472 if (fp->fi_had_conflict)
5473 return ERR_PTR(-EAGAIN);
5474
5475 /*
5476 * Try for a write delegation first. RFC8881 section 10.4 says:
5477 *
5478 * "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5479 * on its own, all opens."
5480 *
5481 * Furthermore the client can use a write delegation for most READ
5482 * operations as well, so we require a O_RDWR file here.
5483 *
5484 * Offer a write delegation in the case of a BOTH open, and ensure
5485 * we get the O_RDWR descriptor.
5486 */
5487 if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5488 nf = find_rw_file(fp);
5489 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5490 }
5491
5492 /*
5493 * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5494 * file for some reason, then try for a read delegation instead.
5495 */
5496 if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5497 nf = find_readable_file(fp);
5498 dl_type = NFS4_OPEN_DELEGATE_READ;
5499 }
5500
5501 if (!nf)
5502 return ERR_PTR(-EAGAIN);
5503
5504 spin_lock(&state_lock);
5505 spin_lock(&fp->fi_lock);
5506 if (nfs4_delegation_exists(clp, fp))
5507 status = -EAGAIN;
5508 else if (nfsd4_verify_setuid_write(open, nf))
5509 status = -EAGAIN;
5510 else if (!fp->fi_deleg_file) {
5511 fp->fi_deleg_file = nf;
5512 /* increment early to prevent fi_deleg_file from being
5513 * cleared */
5514 fp->fi_delegees = 1;
5515 nf = NULL;
5516 } else
5517 fp->fi_delegees++;
5518 spin_unlock(&fp->fi_lock);
5519 spin_unlock(&state_lock);
5520 if (nf)
5521 nfsd_file_put(nf);
5522 if (status)
5523 return ERR_PTR(status);
5524
5525 status = -ENOMEM;
5526 dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5527 if (!dp)
5528 goto out_delegees;
5529
5530 fl = nfs4_alloc_init_lease(dp, dl_type);
5531 if (!fl)
5532 goto out_clnt_odstate;
5533
5534 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5535 if (fl)
5536 locks_free_lock(fl);
5537 if (status)
5538 goto out_clnt_odstate;
5539
5540 if (parent) {
5541 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5542 if (status)
5543 goto out_unlock;
5544 }
5545
5546 status = nfsd4_check_conflicting_opens(clp, fp);
5547 if (status)
5548 goto out_unlock;
5549
5550 /*
5551 * Now that the deleg is set, check again to ensure that nothing
5552 * raced in and changed the mode while we weren't lookng.
5553 */
5554 status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5555 if (status)
5556 goto out_unlock;
5557
5558 status = -EAGAIN;
5559 if (fp->fi_had_conflict)
5560 goto out_unlock;
5561
5562 spin_lock(&state_lock);
5563 spin_lock(&fp->fi_lock);
5564 status = hash_delegation_locked(dp, fp);
5565 spin_unlock(&fp->fi_lock);
5566 spin_unlock(&state_lock);
5567
5568 if (status)
5569 goto out_unlock;
5570
5571 return dp;
5572out_unlock:
5573 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5574out_clnt_odstate:
5575 put_clnt_odstate(dp->dl_clnt_odstate);
5576 nfs4_put_stid(&dp->dl_stid);
5577out_delegees:
5578 put_deleg_file(fp);
5579 return ERR_PTR(status);
5580}
5581
5582static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5583{
5584 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5585 if (status == -EAGAIN)
5586 open->op_why_no_deleg = WND4_CONTENTION;
5587 else {
5588 open->op_why_no_deleg = WND4_RESOURCE;
5589 switch (open->op_deleg_want) {
5590 case NFS4_SHARE_WANT_READ_DELEG:
5591 case NFS4_SHARE_WANT_WRITE_DELEG:
5592 case NFS4_SHARE_WANT_ANY_DELEG:
5593 break;
5594 case NFS4_SHARE_WANT_CANCEL:
5595 open->op_why_no_deleg = WND4_CANCELLED;
5596 break;
5597 case NFS4_SHARE_WANT_NO_DELEG:
5598 WARN_ON_ONCE(1);
5599 }
5600 }
5601}
5602
5603/*
5604 * The Linux NFS server does not offer write delegations to NFSv4.0
5605 * clients in order to avoid conflicts between write delegations and
5606 * GETATTRs requesting CHANGE or SIZE attributes.
5607 *
5608 * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5609 * begins each COMPOUND contains a client ID. Delegation recall can
5610 * be avoided when the server recognizes the client sending a
5611 * GETATTR also holds write delegation it conflicts with.
5612 *
5613 * However, the NFSv4.0 protocol does not enable a server to
5614 * determine that a GETATTR originated from the client holding the
5615 * conflicting delegation versus coming from some other client. Per
5616 * RFC 7530 Section 16.7.5, the server must recall or send a
5617 * CB_GETATTR even when the GETATTR originates from the client that
5618 * holds the conflicting delegation.
5619 *
5620 * An NFSv4.0 client can trigger a pathological situation if it
5621 * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5622 * the same COMPOUND. COMPOUND execution will always stop at the
5623 * GETATTR and the DELEGRETURN will never get executed. The server
5624 * eventually revokes the delegation, which can result in loss of
5625 * open or lock state.
5626 */
5627static void
5628nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5629 struct svc_fh *currentfh)
5630{
5631 struct nfs4_delegation *dp;
5632 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5633 struct nfs4_client *clp = stp->st_stid.sc_client;
5634 struct svc_fh *parent = NULL;
5635 int cb_up;
5636 int status = 0;
5637
5638 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5639 open->op_recall = false;
5640 switch (open->op_claim_type) {
5641 case NFS4_OPEN_CLAIM_PREVIOUS:
5642 if (!cb_up)
5643 open->op_recall = true;
5644 break;
5645 case NFS4_OPEN_CLAIM_NULL:
5646 parent = currentfh;
5647 fallthrough;
5648 case NFS4_OPEN_CLAIM_FH:
5649 /*
5650 * Let's not give out any delegations till everyone's
5651 * had the chance to reclaim theirs, *and* until
5652 * NLM locks have all been reclaimed:
5653 */
5654 if (locks_in_grace(clp->net))
5655 goto out_no_deleg;
5656 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5657 goto out_no_deleg;
5658 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5659 !clp->cl_minorversion)
5660 goto out_no_deleg;
5661 break;
5662 default:
5663 goto out_no_deleg;
5664 }
5665 dp = nfs4_set_delegation(open, stp, parent);
5666 if (IS_ERR(dp))
5667 goto out_no_deleg;
5668
5669 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5670
5671 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5672 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5673 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5674 } else {
5675 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5676 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5677 }
5678 nfs4_put_stid(&dp->dl_stid);
5679 return;
5680out_no_deleg:
5681 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5682 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5683 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5684 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5685 open->op_recall = true;
5686 }
5687
5688 /* 4.1 client asking for a delegation? */
5689 if (open->op_deleg_want)
5690 nfsd4_open_deleg_none_ext(open, status);
5691 return;
5692}
5693
5694static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5695 struct nfs4_delegation *dp)
5696{
5697 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5698 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5699 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5700 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5701 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5702 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5703 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5704 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5705 }
5706 /* Otherwise the client must be confused wanting a delegation
5707 * it already has, therefore we don't return
5708 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5709 */
5710}
5711
5712/**
5713 * nfsd4_process_open2 - finish open processing
5714 * @rqstp: the RPC transaction being executed
5715 * @current_fh: NFSv4 COMPOUND's current filehandle
5716 * @open: OPEN arguments
5717 *
5718 * If successful, (1) truncate the file if open->op_truncate was
5719 * set, (2) set open->op_stateid, (3) set open->op_delegation.
5720 *
5721 * Returns %nfs_ok on success; otherwise an nfs4stat value in
5722 * network byte order is returned.
5723 */
5724__be32
5725nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5726{
5727 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5728 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5729 struct nfs4_file *fp = NULL;
5730 struct nfs4_ol_stateid *stp = NULL;
5731 struct nfs4_delegation *dp = NULL;
5732 __be32 status;
5733 bool new_stp = false;
5734
5735 /*
5736 * Lookup file; if found, lookup stateid and check open request,
5737 * and check for delegations in the process of being recalled.
5738 * If not found, create the nfs4_file struct
5739 */
5740 fp = nfsd4_file_hash_insert(open->op_file, current_fh);
5741 if (unlikely(!fp))
5742 return nfserr_jukebox;
5743 if (fp != open->op_file) {
5744 status = nfs4_check_deleg(cl, open, &dp);
5745 if (status)
5746 goto out;
5747 stp = nfsd4_find_and_lock_existing_open(fp, open);
5748 } else {
5749 open->op_file = NULL;
5750 status = nfserr_bad_stateid;
5751 if (nfsd4_is_deleg_cur(open))
5752 goto out;
5753 }
5754
5755 if (!stp) {
5756 stp = init_open_stateid(fp, open);
5757 if (!open->op_stp)
5758 new_stp = true;
5759 }
5760
5761 /*
5762 * OPEN the file, or upgrade an existing OPEN.
5763 * If truncate fails, the OPEN fails.
5764 *
5765 * stp is already locked.
5766 */
5767 if (!new_stp) {
5768 /* Stateid was found, this is an OPEN upgrade */
5769 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5770 if (status) {
5771 mutex_unlock(&stp->st_mutex);
5772 goto out;
5773 }
5774 } else {
5775 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
5776 if (status) {
5777 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5778 release_open_stateid(stp);
5779 mutex_unlock(&stp->st_mutex);
5780 goto out;
5781 }
5782
5783 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5784 open->op_odstate);
5785 if (stp->st_clnt_odstate == open->op_odstate)
5786 open->op_odstate = NULL;
5787 }
5788
5789 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5790 mutex_unlock(&stp->st_mutex);
5791
5792 if (nfsd4_has_session(&resp->cstate)) {
5793 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5794 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5795 open->op_why_no_deleg = WND4_NOT_WANTED;
5796 goto nodeleg;
5797 }
5798 }
5799
5800 /*
5801 * Attempt to hand out a delegation. No error return, because the
5802 * OPEN succeeds even if we fail.
5803 */
5804 nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
5805nodeleg:
5806 status = nfs_ok;
5807 trace_nfsd_open(&stp->st_stid.sc_stateid);
5808out:
5809 /* 4.1 client trying to upgrade/downgrade delegation? */
5810 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5811 open->op_deleg_want)
5812 nfsd4_deleg_xgrade_none_ext(open, dp);
5813
5814 if (fp)
5815 put_nfs4_file(fp);
5816 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5817 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5818 /*
5819 * To finish the open response, we just need to set the rflags.
5820 */
5821 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5822 if (nfsd4_has_session(&resp->cstate))
5823 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5824 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5825 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5826
5827 if (dp)
5828 nfs4_put_stid(&dp->dl_stid);
5829 if (stp)
5830 nfs4_put_stid(&stp->st_stid);
5831
5832 return status;
5833}
5834
5835void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5836 struct nfsd4_open *open)
5837{
5838 if (open->op_openowner) {
5839 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5840
5841 nfsd4_cstate_assign_replay(cstate, so);
5842 nfs4_put_stateowner(so);
5843 }
5844 if (open->op_file)
5845 kmem_cache_free(file_slab, open->op_file);
5846 if (open->op_stp)
5847 nfs4_put_stid(&open->op_stp->st_stid);
5848 if (open->op_odstate)
5849 kmem_cache_free(odstate_slab, open->op_odstate);
5850}
5851
5852__be32
5853nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5854 union nfsd4_op_u *u)
5855{
5856 clientid_t *clid = &u->renew;
5857 struct nfs4_client *clp;
5858 __be32 status;
5859 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5860
5861 trace_nfsd_clid_renew(clid);
5862 status = set_client(clid, cstate, nn);
5863 if (status)
5864 return status;
5865 clp = cstate->clp;
5866 if (!list_empty(&clp->cl_delegations)
5867 && clp->cl_cb_state != NFSD4_CB_UP)
5868 return nfserr_cb_path_down;
5869 return nfs_ok;
5870}
5871
5872void
5873nfsd4_end_grace(struct nfsd_net *nn)
5874{
5875 /* do nothing if grace period already ended */
5876 if (nn->grace_ended)
5877 return;
5878
5879 trace_nfsd_grace_complete(nn);
5880 nn->grace_ended = true;
5881 /*
5882 * If the server goes down again right now, an NFSv4
5883 * client will still be allowed to reclaim after it comes back up,
5884 * even if it hasn't yet had a chance to reclaim state this time.
5885 *
5886 */
5887 nfsd4_record_grace_done(nn);
5888 /*
5889 * At this point, NFSv4 clients can still reclaim. But if the
5890 * server crashes, any that have not yet reclaimed will be out
5891 * of luck on the next boot.
5892 *
5893 * (NFSv4.1+ clients are considered to have reclaimed once they
5894 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5895 * have reclaimed after their first OPEN.)
5896 */
5897 locks_end_grace(&nn->nfsd4_manager);
5898 /*
5899 * At this point, and once lockd and/or any other containers
5900 * exit their grace period, further reclaims will fail and
5901 * regular locking can resume.
5902 */
5903}
5904
5905/*
5906 * If we've waited a lease period but there are still clients trying to
5907 * reclaim, wait a little longer to give them a chance to finish.
5908 */
5909static bool clients_still_reclaiming(struct nfsd_net *nn)
5910{
5911 time64_t double_grace_period_end = nn->boot_time +
5912 2 * nn->nfsd4_lease;
5913
5914 if (nn->track_reclaim_completes &&
5915 atomic_read(&nn->nr_reclaim_complete) ==
5916 nn->reclaim_str_hashtbl_size)
5917 return false;
5918 if (!nn->somebody_reclaimed)
5919 return false;
5920 nn->somebody_reclaimed = false;
5921 /*
5922 * If we've given them *two* lease times to reclaim, and they're
5923 * still not done, give up:
5924 */
5925 if (ktime_get_boottime_seconds() > double_grace_period_end)
5926 return false;
5927 return true;
5928}
5929
5930struct laundry_time {
5931 time64_t cutoff;
5932 time64_t new_timeo;
5933};
5934
5935static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5936{
5937 time64_t time_remaining;
5938
5939 if (last_refresh < lt->cutoff)
5940 return true;
5941 time_remaining = last_refresh - lt->cutoff;
5942 lt->new_timeo = min(lt->new_timeo, time_remaining);
5943 return false;
5944}
5945
5946#ifdef CONFIG_NFSD_V4_2_INTER_SSC
5947void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5948{
5949 spin_lock_init(&nn->nfsd_ssc_lock);
5950 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5951 init_waitqueue_head(&nn->nfsd_ssc_waitq);
5952}
5953EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5954
5955/*
5956 * This is called when nfsd is being shutdown, after all inter_ssc
5957 * cleanup were done, to destroy the ssc delayed unmount list.
5958 */
5959static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5960{
5961 struct nfsd4_ssc_umount_item *ni = NULL;
5962 struct nfsd4_ssc_umount_item *tmp;
5963
5964 spin_lock(&nn->nfsd_ssc_lock);
5965 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5966 list_del(&ni->nsui_list);
5967 spin_unlock(&nn->nfsd_ssc_lock);
5968 mntput(ni->nsui_vfsmount);
5969 kfree(ni);
5970 spin_lock(&nn->nfsd_ssc_lock);
5971 }
5972 spin_unlock(&nn->nfsd_ssc_lock);
5973}
5974
5975static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5976{
5977 bool do_wakeup = false;
5978 struct nfsd4_ssc_umount_item *ni = NULL;
5979 struct nfsd4_ssc_umount_item *tmp;
5980
5981 spin_lock(&nn->nfsd_ssc_lock);
5982 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5983 if (time_after(jiffies, ni->nsui_expire)) {
5984 if (refcount_read(&ni->nsui_refcnt) > 1)
5985 continue;
5986
5987 /* mark being unmount */
5988 ni->nsui_busy = true;
5989 spin_unlock(&nn->nfsd_ssc_lock);
5990 mntput(ni->nsui_vfsmount);
5991 spin_lock(&nn->nfsd_ssc_lock);
5992
5993 /* waiters need to start from begin of list */
5994 list_del(&ni->nsui_list);
5995 kfree(ni);
5996
5997 /* wakeup ssc_connect waiters */
5998 do_wakeup = true;
5999 continue;
6000 }
6001 break;
6002 }
6003 if (do_wakeup)
6004 wake_up_all(&nn->nfsd_ssc_waitq);
6005 spin_unlock(&nn->nfsd_ssc_lock);
6006}
6007#endif
6008
6009/* Check if any lock belonging to this lockowner has any blockers */
6010static bool
6011nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6012{
6013 struct file_lock_context *ctx;
6014 struct nfs4_ol_stateid *stp;
6015 struct nfs4_file *nf;
6016
6017 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6018 nf = stp->st_stid.sc_file;
6019 ctx = locks_inode_context(nf->fi_inode);
6020 if (!ctx)
6021 continue;
6022 if (locks_owner_has_blockers(ctx, lo))
6023 return true;
6024 }
6025 return false;
6026}
6027
6028static bool
6029nfs4_anylock_blockers(struct nfs4_client *clp)
6030{
6031 int i;
6032 struct nfs4_stateowner *so;
6033 struct nfs4_lockowner *lo;
6034
6035 if (atomic_read(&clp->cl_delegs_in_recall))
6036 return true;
6037 spin_lock(&clp->cl_lock);
6038 for (i = 0; i < OWNER_HASH_SIZE; i++) {
6039 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6040 so_strhash) {
6041 if (so->so_is_open_owner)
6042 continue;
6043 lo = lockowner(so);
6044 if (nfs4_lockowner_has_blockers(lo)) {
6045 spin_unlock(&clp->cl_lock);
6046 return true;
6047 }
6048 }
6049 }
6050 spin_unlock(&clp->cl_lock);
6051 return false;
6052}
6053
6054static void
6055nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6056 struct laundry_time *lt)
6057{
6058 unsigned int maxreap, reapcnt = 0;
6059 struct list_head *pos, *next;
6060 struct nfs4_client *clp;
6061
6062 maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6063 NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6064 INIT_LIST_HEAD(reaplist);
6065 spin_lock(&nn->client_lock);
6066 list_for_each_safe(pos, next, &nn->client_lru) {
6067 clp = list_entry(pos, struct nfs4_client, cl_lru);
6068 if (clp->cl_state == NFSD4_EXPIRABLE)
6069 goto exp_client;
6070 if (!state_expired(lt, clp->cl_time))
6071 break;
6072 if (!atomic_read(&clp->cl_rpc_users)) {
6073 if (clp->cl_state == NFSD4_ACTIVE)
6074 atomic_inc(&nn->nfsd_courtesy_clients);
6075 clp->cl_state = NFSD4_COURTESY;
6076 }
6077 if (!client_has_state(clp))
6078 goto exp_client;
6079 if (!nfs4_anylock_blockers(clp))
6080 if (reapcnt >= maxreap)
6081 continue;
6082exp_client:
6083 if (!mark_client_expired_locked(clp)) {
6084 list_add(&clp->cl_lru, reaplist);
6085 reapcnt++;
6086 }
6087 }
6088 spin_unlock(&nn->client_lock);
6089}
6090
6091static void
6092nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6093 struct list_head *reaplist)
6094{
6095 unsigned int maxreap = 0, reapcnt = 0;
6096 struct list_head *pos, *next;
6097 struct nfs4_client *clp;
6098
6099 maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6100 INIT_LIST_HEAD(reaplist);
6101
6102 spin_lock(&nn->client_lock);
6103 list_for_each_safe(pos, next, &nn->client_lru) {
6104 clp = list_entry(pos, struct nfs4_client, cl_lru);
6105 if (clp->cl_state == NFSD4_ACTIVE)
6106 break;
6107 if (reapcnt >= maxreap)
6108 break;
6109 if (!mark_client_expired_locked(clp)) {
6110 list_add(&clp->cl_lru, reaplist);
6111 reapcnt++;
6112 }
6113 }
6114 spin_unlock(&nn->client_lock);
6115}
6116
6117static void
6118nfs4_process_client_reaplist(struct list_head *reaplist)
6119{
6120 struct list_head *pos, *next;
6121 struct nfs4_client *clp;
6122
6123 list_for_each_safe(pos, next, reaplist) {
6124 clp = list_entry(pos, struct nfs4_client, cl_lru);
6125 trace_nfsd_clid_purged(&clp->cl_clientid);
6126 list_del_init(&clp->cl_lru);
6127 expire_client(clp);
6128 }
6129}
6130
6131static time64_t
6132nfs4_laundromat(struct nfsd_net *nn)
6133{
6134 struct nfs4_openowner *oo;
6135 struct nfs4_delegation *dp;
6136 struct nfs4_ol_stateid *stp;
6137 struct nfsd4_blocked_lock *nbl;
6138 struct list_head *pos, *next, reaplist;
6139 struct laundry_time lt = {
6140 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6141 .new_timeo = nn->nfsd4_lease
6142 };
6143 struct nfs4_cpntf_state *cps;
6144 copy_stateid_t *cps_t;
6145 int i;
6146
6147 if (clients_still_reclaiming(nn)) {
6148 lt.new_timeo = 0;
6149 goto out;
6150 }
6151 nfsd4_end_grace(nn);
6152
6153 spin_lock(&nn->s2s_cp_lock);
6154 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6155 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6156 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6157 state_expired(<, cps->cpntf_time))
6158 _free_cpntf_state_locked(nn, cps);
6159 }
6160 spin_unlock(&nn->s2s_cp_lock);
6161 nfs4_get_client_reaplist(nn, &reaplist, <);
6162 nfs4_process_client_reaplist(&reaplist);
6163
6164 spin_lock(&state_lock);
6165 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6166 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6167 if (!state_expired(<, dp->dl_time))
6168 break;
6169 WARN_ON(!unhash_delegation_locked(dp));
6170 list_add(&dp->dl_recall_lru, &reaplist);
6171 }
6172 spin_unlock(&state_lock);
6173 while (!list_empty(&reaplist)) {
6174 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6175 dl_recall_lru);
6176 list_del_init(&dp->dl_recall_lru);
6177 revoke_delegation(dp);
6178 }
6179
6180 spin_lock(&nn->client_lock);
6181 while (!list_empty(&nn->close_lru)) {
6182 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6183 oo_close_lru);
6184 if (!state_expired(<, oo->oo_time))
6185 break;
6186 list_del_init(&oo->oo_close_lru);
6187 stp = oo->oo_last_closed_stid;
6188 oo->oo_last_closed_stid = NULL;
6189 spin_unlock(&nn->client_lock);
6190 nfs4_put_stid(&stp->st_stid);
6191 spin_lock(&nn->client_lock);
6192 }
6193 spin_unlock(&nn->client_lock);
6194
6195 /*
6196 * It's possible for a client to try and acquire an already held lock
6197 * that is being held for a long time, and then lose interest in it.
6198 * So, we clean out any un-revisited request after a lease period
6199 * under the assumption that the client is no longer interested.
6200 *
6201 * RFC5661, sec. 9.6 states that the client must not rely on getting
6202 * notifications and must continue to poll for locks, even when the
6203 * server supports them. Thus this shouldn't lead to clients blocking
6204 * indefinitely once the lock does become free.
6205 */
6206 BUG_ON(!list_empty(&reaplist));
6207 spin_lock(&nn->blocked_locks_lock);
6208 while (!list_empty(&nn->blocked_locks_lru)) {
6209 nbl = list_first_entry(&nn->blocked_locks_lru,
6210 struct nfsd4_blocked_lock, nbl_lru);
6211 if (!state_expired(<, nbl->nbl_time))
6212 break;
6213 list_move(&nbl->nbl_lru, &reaplist);
6214 list_del_init(&nbl->nbl_list);
6215 }
6216 spin_unlock(&nn->blocked_locks_lock);
6217
6218 while (!list_empty(&reaplist)) {
6219 nbl = list_first_entry(&reaplist,
6220 struct nfsd4_blocked_lock, nbl_lru);
6221 list_del_init(&nbl->nbl_lru);
6222 free_blocked_lock(nbl);
6223 }
6224#ifdef CONFIG_NFSD_V4_2_INTER_SSC
6225 /* service the server-to-server copy delayed unmount list */
6226 nfsd4_ssc_expire_umount(nn);
6227#endif
6228out:
6229 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6230}
6231
6232static void laundromat_main(struct work_struct *);
6233
6234static void
6235laundromat_main(struct work_struct *laundry)
6236{
6237 time64_t t;
6238 struct delayed_work *dwork = to_delayed_work(laundry);
6239 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6240 laundromat_work);
6241
6242 t = nfs4_laundromat(nn);
6243 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6244}
6245
6246static void
6247courtesy_client_reaper(struct nfsd_net *nn)
6248{
6249 struct list_head reaplist;
6250
6251 nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6252 nfs4_process_client_reaplist(&reaplist);
6253}
6254
6255static void
6256deleg_reaper(struct nfsd_net *nn)
6257{
6258 struct list_head *pos, *next;
6259 struct nfs4_client *clp;
6260 struct list_head cblist;
6261
6262 INIT_LIST_HEAD(&cblist);
6263 spin_lock(&nn->client_lock);
6264 list_for_each_safe(pos, next, &nn->client_lru) {
6265 clp = list_entry(pos, struct nfs4_client, cl_lru);
6266 if (clp->cl_state != NFSD4_ACTIVE ||
6267 list_empty(&clp->cl_delegations) ||
6268 atomic_read(&clp->cl_delegs_in_recall) ||
6269 test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6270 (ktime_get_boottime_seconds() -
6271 clp->cl_ra_time < 5)) {
6272 continue;
6273 }
6274 list_add(&clp->cl_ra_cblist, &cblist);
6275
6276 /* release in nfsd4_cb_recall_any_release */
6277 atomic_inc(&clp->cl_rpc_users);
6278 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6279 clp->cl_ra_time = ktime_get_boottime_seconds();
6280 }
6281 spin_unlock(&nn->client_lock);
6282
6283 while (!list_empty(&cblist)) {
6284 clp = list_first_entry(&cblist, struct nfs4_client,
6285 cl_ra_cblist);
6286 list_del_init(&clp->cl_ra_cblist);
6287 clp->cl_ra->ra_keep = 0;
6288 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6289 trace_nfsd_cb_recall_any(clp->cl_ra);
6290 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6291 }
6292}
6293
6294static void
6295nfsd4_state_shrinker_worker(struct work_struct *work)
6296{
6297 struct nfsd_net *nn = container_of(work, struct nfsd_net,
6298 nfsd_shrinker_work);
6299
6300 courtesy_client_reaper(nn);
6301 deleg_reaper(nn);
6302}
6303
6304static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6305{
6306 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6307 return nfserr_bad_stateid;
6308 return nfs_ok;
6309}
6310
6311static
6312__be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6313{
6314 __be32 status = nfserr_openmode;
6315
6316 /* For lock stateid's, we test the parent open, not the lock: */
6317 if (stp->st_openstp)
6318 stp = stp->st_openstp;
6319 if ((flags & WR_STATE) && !access_permit_write(stp))
6320 goto out;
6321 if ((flags & RD_STATE) && !access_permit_read(stp))
6322 goto out;
6323 status = nfs_ok;
6324out:
6325 return status;
6326}
6327
6328static inline __be32
6329check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6330{
6331 if (ONE_STATEID(stateid) && (flags & RD_STATE))
6332 return nfs_ok;
6333 else if (opens_in_grace(net)) {
6334 /* Answer in remaining cases depends on existence of
6335 * conflicting state; so we must wait out the grace period. */
6336 return nfserr_grace;
6337 } else if (flags & WR_STATE)
6338 return nfs4_share_conflict(current_fh,
6339 NFS4_SHARE_DENY_WRITE);
6340 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6341 return nfs4_share_conflict(current_fh,
6342 NFS4_SHARE_DENY_READ);
6343}
6344
6345static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6346{
6347 /*
6348 * When sessions are used the stateid generation number is ignored
6349 * when it is zero.
6350 */
6351 if (has_session && in->si_generation == 0)
6352 return nfs_ok;
6353
6354 if (in->si_generation == ref->si_generation)
6355 return nfs_ok;
6356
6357 /* If the client sends us a stateid from the future, it's buggy: */
6358 if (nfsd4_stateid_generation_after(in, ref))
6359 return nfserr_bad_stateid;
6360 /*
6361 * However, we could see a stateid from the past, even from a
6362 * non-buggy client. For example, if the client sends a lock
6363 * while some IO is outstanding, the lock may bump si_generation
6364 * while the IO is still in flight. The client could avoid that
6365 * situation by waiting for responses on all the IO requests,
6366 * but better performance may result in retrying IO that
6367 * receives an old_stateid error if requests are rarely
6368 * reordered in flight:
6369 */
6370 return nfserr_old_stateid;
6371}
6372
6373static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6374{
6375 __be32 ret;
6376
6377 spin_lock(&s->sc_lock);
6378 ret = nfsd4_verify_open_stid(s);
6379 if (ret == nfs_ok)
6380 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6381 spin_unlock(&s->sc_lock);
6382 return ret;
6383}
6384
6385static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6386{
6387 if (ols->st_stateowner->so_is_open_owner &&
6388 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6389 return nfserr_bad_stateid;
6390 return nfs_ok;
6391}
6392
6393static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6394{
6395 struct nfs4_stid *s;
6396 __be32 status = nfserr_bad_stateid;
6397
6398 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6399 CLOSE_STATEID(stateid))
6400 return status;
6401 spin_lock(&cl->cl_lock);
6402 s = find_stateid_locked(cl, stateid);
6403 if (!s)
6404 goto out_unlock;
6405 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6406 if (status)
6407 goto out_unlock;
6408 switch (s->sc_type) {
6409 case NFS4_DELEG_STID:
6410 status = nfs_ok;
6411 break;
6412 case NFS4_REVOKED_DELEG_STID:
6413 status = nfserr_deleg_revoked;
6414 break;
6415 case NFS4_OPEN_STID:
6416 case NFS4_LOCK_STID:
6417 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6418 break;
6419 default:
6420 printk("unknown stateid type %x\n", s->sc_type);
6421 fallthrough;
6422 case NFS4_CLOSED_STID:
6423 case NFS4_CLOSED_DELEG_STID:
6424 status = nfserr_bad_stateid;
6425 }
6426out_unlock:
6427 spin_unlock(&cl->cl_lock);
6428 return status;
6429}
6430
6431__be32
6432nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6433 stateid_t *stateid, unsigned char typemask,
6434 struct nfs4_stid **s, struct nfsd_net *nn)
6435{
6436 __be32 status;
6437 struct nfs4_stid *stid;
6438 bool return_revoked = false;
6439
6440 /*
6441 * only return revoked delegations if explicitly asked.
6442 * otherwise we report revoked or bad_stateid status.
6443 */
6444 if (typemask & NFS4_REVOKED_DELEG_STID)
6445 return_revoked = true;
6446 else if (typemask & NFS4_DELEG_STID)
6447 typemask |= NFS4_REVOKED_DELEG_STID;
6448
6449 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6450 CLOSE_STATEID(stateid))
6451 return nfserr_bad_stateid;
6452 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6453 if (status == nfserr_stale_clientid) {
6454 if (cstate->session)
6455 return nfserr_bad_stateid;
6456 return nfserr_stale_stateid;
6457 }
6458 if (status)
6459 return status;
6460 stid = find_stateid_by_type(cstate->clp, stateid, typemask);
6461 if (!stid)
6462 return nfserr_bad_stateid;
6463 if ((stid->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
6464 nfs4_put_stid(stid);
6465 if (cstate->minorversion)
6466 return nfserr_deleg_revoked;
6467 return nfserr_bad_stateid;
6468 }
6469 *s = stid;
6470 return nfs_ok;
6471}
6472
6473static struct nfsd_file *
6474nfs4_find_file(struct nfs4_stid *s, int flags)
6475{
6476 struct nfsd_file *ret = NULL;
6477
6478 if (!s)
6479 return NULL;
6480
6481 switch (s->sc_type) {
6482 case NFS4_DELEG_STID:
6483 spin_lock(&s->sc_file->fi_lock);
6484 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6485 spin_unlock(&s->sc_file->fi_lock);
6486 break;
6487 case NFS4_OPEN_STID:
6488 case NFS4_LOCK_STID:
6489 if (flags & RD_STATE)
6490 ret = find_readable_file(s->sc_file);
6491 else
6492 ret = find_writeable_file(s->sc_file);
6493 }
6494
6495 return ret;
6496}
6497
6498static __be32
6499nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6500{
6501 __be32 status;
6502
6503 status = nfsd4_check_openowner_confirmed(ols);
6504 if (status)
6505 return status;
6506 return nfs4_check_openmode(ols, flags);
6507}
6508
6509static __be32
6510nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6511 struct nfsd_file **nfp, int flags)
6512{
6513 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6514 struct nfsd_file *nf;
6515 __be32 status;
6516
6517 nf = nfs4_find_file(s, flags);
6518 if (nf) {
6519 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6520 acc | NFSD_MAY_OWNER_OVERRIDE);
6521 if (status) {
6522 nfsd_file_put(nf);
6523 goto out;
6524 }
6525 } else {
6526 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6527 if (status)
6528 return status;
6529 }
6530 *nfp = nf;
6531out:
6532 return status;
6533}
6534static void
6535_free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6536{
6537 WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6538 if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6539 return;
6540 list_del(&cps->cp_list);
6541 idr_remove(&nn->s2s_cp_stateids,
6542 cps->cp_stateid.cs_stid.si_opaque.so_id);
6543 kfree(cps);
6544}
6545/*
6546 * A READ from an inter server to server COPY will have a
6547 * copy stateid. Look up the copy notify stateid from the
6548 * idr structure and take a reference on it.
6549 */
6550__be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6551 struct nfs4_client *clp,
6552 struct nfs4_cpntf_state **cps)
6553{
6554 copy_stateid_t *cps_t;
6555 struct nfs4_cpntf_state *state = NULL;
6556
6557 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6558 return nfserr_bad_stateid;
6559 spin_lock(&nn->s2s_cp_lock);
6560 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6561 if (cps_t) {
6562 state = container_of(cps_t, struct nfs4_cpntf_state,
6563 cp_stateid);
6564 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6565 state = NULL;
6566 goto unlock;
6567 }
6568 if (!clp)
6569 refcount_inc(&state->cp_stateid.cs_count);
6570 else
6571 _free_cpntf_state_locked(nn, state);
6572 }
6573unlock:
6574 spin_unlock(&nn->s2s_cp_lock);
6575 if (!state)
6576 return nfserr_bad_stateid;
6577 if (!clp)
6578 *cps = state;
6579 return 0;
6580}
6581
6582static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6583 struct nfs4_stid **stid)
6584{
6585 __be32 status;
6586 struct nfs4_cpntf_state *cps = NULL;
6587 struct nfs4_client *found;
6588
6589 status = manage_cpntf_state(nn, st, NULL, &cps);
6590 if (status)
6591 return status;
6592
6593 cps->cpntf_time = ktime_get_boottime_seconds();
6594
6595 status = nfserr_expired;
6596 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6597 if (!found)
6598 goto out;
6599
6600 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6601 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
6602 if (*stid)
6603 status = nfs_ok;
6604 else
6605 status = nfserr_bad_stateid;
6606
6607 put_client_renew(found);
6608out:
6609 nfs4_put_cpntf_state(nn, cps);
6610 return status;
6611}
6612
6613void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6614{
6615 spin_lock(&nn->s2s_cp_lock);
6616 _free_cpntf_state_locked(nn, cps);
6617 spin_unlock(&nn->s2s_cp_lock);
6618}
6619
6620/**
6621 * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6622 * @rqstp: incoming request from client
6623 * @cstate: current compound state
6624 * @fhp: filehandle associated with requested stateid
6625 * @stateid: stateid (provided by client)
6626 * @flags: flags describing type of operation to be done
6627 * @nfp: optional nfsd_file return pointer (may be NULL)
6628 * @cstid: optional returned nfs4_stid pointer (may be NULL)
6629 *
6630 * Given info from the client, look up a nfs4_stid for the operation. On
6631 * success, it returns a reference to the nfs4_stid and/or the nfsd_file
6632 * associated with it.
6633 */
6634__be32
6635nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6636 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6637 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6638 struct nfs4_stid **cstid)
6639{
6640 struct net *net = SVC_NET(rqstp);
6641 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6642 struct nfs4_stid *s = NULL;
6643 __be32 status;
6644
6645 if (nfp)
6646 *nfp = NULL;
6647
6648 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6649 if (cstid)
6650 status = nfserr_bad_stateid;
6651 else
6652 status = check_special_stateids(net, fhp, stateid,
6653 flags);
6654 goto done;
6655 }
6656
6657 status = nfsd4_lookup_stateid(cstate, stateid,
6658 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
6659 &s, nn);
6660 if (status == nfserr_bad_stateid)
6661 status = find_cpntf_state(nn, stateid, &s);
6662 if (status)
6663 return status;
6664 status = nfsd4_stid_check_stateid_generation(stateid, s,
6665 nfsd4_has_session(cstate));
6666 if (status)
6667 goto out;
6668
6669 switch (s->sc_type) {
6670 case NFS4_DELEG_STID:
6671 status = nfs4_check_delegmode(delegstateid(s), flags);
6672 break;
6673 case NFS4_OPEN_STID:
6674 case NFS4_LOCK_STID:
6675 status = nfs4_check_olstateid(openlockstateid(s), flags);
6676 break;
6677 default:
6678 status = nfserr_bad_stateid;
6679 break;
6680 }
6681 if (status)
6682 goto out;
6683 status = nfs4_check_fh(fhp, s);
6684
6685done:
6686 if (status == nfs_ok && nfp)
6687 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
6688out:
6689 if (s) {
6690 if (!status && cstid)
6691 *cstid = s;
6692 else
6693 nfs4_put_stid(s);
6694 }
6695 return status;
6696}
6697
6698/*
6699 * Test if the stateid is valid
6700 */
6701__be32
6702nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6703 union nfsd4_op_u *u)
6704{
6705 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
6706 struct nfsd4_test_stateid_id *stateid;
6707 struct nfs4_client *cl = cstate->clp;
6708
6709 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
6710 stateid->ts_id_status =
6711 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
6712
6713 return nfs_ok;
6714}
6715
6716static __be32
6717nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
6718{
6719 struct nfs4_ol_stateid *stp = openlockstateid(s);
6720 __be32 ret;
6721
6722 ret = nfsd4_lock_ol_stateid(stp);
6723 if (ret)
6724 goto out_put_stid;
6725
6726 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6727 if (ret)
6728 goto out;
6729
6730 ret = nfserr_locks_held;
6731 if (check_for_locks(stp->st_stid.sc_file,
6732 lockowner(stp->st_stateowner)))
6733 goto out;
6734
6735 release_lock_stateid(stp);
6736 ret = nfs_ok;
6737
6738out:
6739 mutex_unlock(&stp->st_mutex);
6740out_put_stid:
6741 nfs4_put_stid(s);
6742 return ret;
6743}
6744
6745__be32
6746nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6747 union nfsd4_op_u *u)
6748{
6749 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
6750 stateid_t *stateid = &free_stateid->fr_stateid;
6751 struct nfs4_stid *s;
6752 struct nfs4_delegation *dp;
6753 struct nfs4_client *cl = cstate->clp;
6754 __be32 ret = nfserr_bad_stateid;
6755
6756 spin_lock(&cl->cl_lock);
6757 s = find_stateid_locked(cl, stateid);
6758 if (!s)
6759 goto out_unlock;
6760 spin_lock(&s->sc_lock);
6761 switch (s->sc_type) {
6762 case NFS4_DELEG_STID:
6763 ret = nfserr_locks_held;
6764 break;
6765 case NFS4_OPEN_STID:
6766 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6767 if (ret)
6768 break;
6769 ret = nfserr_locks_held;
6770 break;
6771 case NFS4_LOCK_STID:
6772 spin_unlock(&s->sc_lock);
6773 refcount_inc(&s->sc_count);
6774 spin_unlock(&cl->cl_lock);
6775 ret = nfsd4_free_lock_stateid(stateid, s);
6776 goto out;
6777 case NFS4_REVOKED_DELEG_STID:
6778 spin_unlock(&s->sc_lock);
6779 dp = delegstateid(s);
6780 list_del_init(&dp->dl_recall_lru);
6781 spin_unlock(&cl->cl_lock);
6782 nfs4_put_stid(s);
6783 ret = nfs_ok;
6784 goto out;
6785 /* Default falls through and returns nfserr_bad_stateid */
6786 }
6787 spin_unlock(&s->sc_lock);
6788out_unlock:
6789 spin_unlock(&cl->cl_lock);
6790out:
6791 return ret;
6792}
6793
6794static inline int
6795setlkflg (int type)
6796{
6797 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6798 RD_STATE : WR_STATE;
6799}
6800
6801static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6802{
6803 struct svc_fh *current_fh = &cstate->current_fh;
6804 struct nfs4_stateowner *sop = stp->st_stateowner;
6805 __be32 status;
6806
6807 status = nfsd4_check_seqid(cstate, sop, seqid);
6808 if (status)
6809 return status;
6810 status = nfsd4_lock_ol_stateid(stp);
6811 if (status != nfs_ok)
6812 return status;
6813 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6814 if (status == nfs_ok)
6815 status = nfs4_check_fh(current_fh, &stp->st_stid);
6816 if (status != nfs_ok)
6817 mutex_unlock(&stp->st_mutex);
6818 return status;
6819}
6820
6821/**
6822 * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
6823 * @cstate: compund state
6824 * @seqid: seqid (provided by client)
6825 * @stateid: stateid (provided by client)
6826 * @typemask: mask of allowable types for this operation
6827 * @stpp: return pointer for the stateid found
6828 * @nn: net namespace for request
6829 *
6830 * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
6831 * return it in @stpp. On a nfs_ok return, the returned stateid will
6832 * have its st_mutex locked.
6833 */
6834static __be32
6835nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6836 stateid_t *stateid, char typemask,
6837 struct nfs4_ol_stateid **stpp,
6838 struct nfsd_net *nn)
6839{
6840 __be32 status;
6841 struct nfs4_stid *s;
6842 struct nfs4_ol_stateid *stp = NULL;
6843
6844 trace_nfsd_preprocess(seqid, stateid);
6845
6846 *stpp = NULL;
6847 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6848 if (status)
6849 return status;
6850 stp = openlockstateid(s);
6851 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6852
6853 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6854 if (!status)
6855 *stpp = stp;
6856 else
6857 nfs4_put_stid(&stp->st_stid);
6858 return status;
6859}
6860
6861static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6862 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6863{
6864 __be32 status;
6865 struct nfs4_openowner *oo;
6866 struct nfs4_ol_stateid *stp;
6867
6868 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6869 NFS4_OPEN_STID, &stp, nn);
6870 if (status)
6871 return status;
6872 oo = openowner(stp->st_stateowner);
6873 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6874 mutex_unlock(&stp->st_mutex);
6875 nfs4_put_stid(&stp->st_stid);
6876 return nfserr_bad_stateid;
6877 }
6878 *stpp = stp;
6879 return nfs_ok;
6880}
6881
6882__be32
6883nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6884 union nfsd4_op_u *u)
6885{
6886 struct nfsd4_open_confirm *oc = &u->open_confirm;
6887 __be32 status;
6888 struct nfs4_openowner *oo;
6889 struct nfs4_ol_stateid *stp;
6890 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6891
6892 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6893 cstate->current_fh.fh_dentry);
6894
6895 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6896 if (status)
6897 return status;
6898
6899 status = nfs4_preprocess_seqid_op(cstate,
6900 oc->oc_seqid, &oc->oc_req_stateid,
6901 NFS4_OPEN_STID, &stp, nn);
6902 if (status)
6903 goto out;
6904 oo = openowner(stp->st_stateowner);
6905 status = nfserr_bad_stateid;
6906 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6907 mutex_unlock(&stp->st_mutex);
6908 goto put_stateid;
6909 }
6910 oo->oo_flags |= NFS4_OO_CONFIRMED;
6911 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6912 mutex_unlock(&stp->st_mutex);
6913 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6914 nfsd4_client_record_create(oo->oo_owner.so_client);
6915 status = nfs_ok;
6916put_stateid:
6917 nfs4_put_stid(&stp->st_stid);
6918out:
6919 nfsd4_bump_seqid(cstate, status);
6920 return status;
6921}
6922
6923static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6924{
6925 if (!test_access(access, stp))
6926 return;
6927 nfs4_file_put_access(stp->st_stid.sc_file, access);
6928 clear_access(access, stp);
6929}
6930
6931static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6932{
6933 switch (to_access) {
6934 case NFS4_SHARE_ACCESS_READ:
6935 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6936 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6937 break;
6938 case NFS4_SHARE_ACCESS_WRITE:
6939 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6940 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6941 break;
6942 case NFS4_SHARE_ACCESS_BOTH:
6943 break;
6944 default:
6945 WARN_ON_ONCE(1);
6946 }
6947}
6948
6949__be32
6950nfsd4_open_downgrade(struct svc_rqst *rqstp,
6951 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6952{
6953 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6954 __be32 status;
6955 struct nfs4_ol_stateid *stp;
6956 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6957
6958 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6959 cstate->current_fh.fh_dentry);
6960
6961 /* We don't yet support WANT bits: */
6962 if (od->od_deleg_want)
6963 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6964 od->od_deleg_want);
6965
6966 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6967 &od->od_stateid, &stp, nn);
6968 if (status)
6969 goto out;
6970 status = nfserr_inval;
6971 if (!test_access(od->od_share_access, stp)) {
6972 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6973 stp->st_access_bmap, od->od_share_access);
6974 goto put_stateid;
6975 }
6976 if (!test_deny(od->od_share_deny, stp)) {
6977 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6978 stp->st_deny_bmap, od->od_share_deny);
6979 goto put_stateid;
6980 }
6981 nfs4_stateid_downgrade(stp, od->od_share_access);
6982 reset_union_bmap_deny(od->od_share_deny, stp);
6983 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6984 status = nfs_ok;
6985put_stateid:
6986 mutex_unlock(&stp->st_mutex);
6987 nfs4_put_stid(&stp->st_stid);
6988out:
6989 nfsd4_bump_seqid(cstate, status);
6990 return status;
6991}
6992
6993static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6994{
6995 struct nfs4_client *clp = s->st_stid.sc_client;
6996 bool unhashed;
6997 LIST_HEAD(reaplist);
6998 struct nfs4_ol_stateid *stp;
6999
7000 spin_lock(&clp->cl_lock);
7001 unhashed = unhash_open_stateid(s, &reaplist);
7002
7003 if (clp->cl_minorversion) {
7004 if (unhashed)
7005 put_ol_stateid_locked(s, &reaplist);
7006 spin_unlock(&clp->cl_lock);
7007 list_for_each_entry(stp, &reaplist, st_locks)
7008 nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7009 free_ol_stateid_reaplist(&reaplist);
7010 } else {
7011 spin_unlock(&clp->cl_lock);
7012 free_ol_stateid_reaplist(&reaplist);
7013 if (unhashed)
7014 move_to_close_lru(s, clp->net);
7015 }
7016}
7017
7018/*
7019 * nfs4_unlock_state() called after encode
7020 */
7021__be32
7022nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7023 union nfsd4_op_u *u)
7024{
7025 struct nfsd4_close *close = &u->close;
7026 __be32 status;
7027 struct nfs4_ol_stateid *stp;
7028 struct net *net = SVC_NET(rqstp);
7029 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7030
7031 dprintk("NFSD: nfsd4_close on file %pd\n",
7032 cstate->current_fh.fh_dentry);
7033
7034 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7035 &close->cl_stateid,
7036 NFS4_OPEN_STID|NFS4_CLOSED_STID,
7037 &stp, nn);
7038 nfsd4_bump_seqid(cstate, status);
7039 if (status)
7040 goto out;
7041
7042 stp->st_stid.sc_type = NFS4_CLOSED_STID;
7043
7044 /*
7045 * Technically we don't _really_ have to increment or copy it, since
7046 * it should just be gone after this operation and we clobber the
7047 * copied value below, but we continue to do so here just to ensure
7048 * that racing ops see that there was a state change.
7049 */
7050 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7051
7052 nfsd4_close_open_stateid(stp);
7053 mutex_unlock(&stp->st_mutex);
7054
7055 /* v4.1+ suggests that we send a special stateid in here, since the
7056 * clients should just ignore this anyway. Since this is not useful
7057 * for v4.0 clients either, we set it to the special close_stateid
7058 * universally.
7059 *
7060 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7061 */
7062 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7063
7064 /* put reference from nfs4_preprocess_seqid_op */
7065 nfs4_put_stid(&stp->st_stid);
7066out:
7067 return status;
7068}
7069
7070__be32
7071nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7072 union nfsd4_op_u *u)
7073{
7074 struct nfsd4_delegreturn *dr = &u->delegreturn;
7075 struct nfs4_delegation *dp;
7076 stateid_t *stateid = &dr->dr_stateid;
7077 struct nfs4_stid *s;
7078 __be32 status;
7079 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7080
7081 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7082 return status;
7083
7084 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
7085 if (status)
7086 goto out;
7087 dp = delegstateid(s);
7088 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7089 if (status)
7090 goto put_stateid;
7091
7092 trace_nfsd_deleg_return(stateid);
7093 wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7094 destroy_delegation(dp);
7095put_stateid:
7096 nfs4_put_stid(&dp->dl_stid);
7097out:
7098 return status;
7099}
7100
7101/* last octet in a range */
7102static inline u64
7103last_byte_offset(u64 start, u64 len)
7104{
7105 u64 end;
7106
7107 WARN_ON_ONCE(!len);
7108 end = start + len;
7109 return end > start ? end - 1: NFS4_MAX_UINT64;
7110}
7111
7112/*
7113 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7114 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7115 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
7116 * locking, this prevents us from being completely protocol-compliant. The
7117 * real solution to this problem is to start using unsigned file offsets in
7118 * the VFS, but this is a very deep change!
7119 */
7120static inline void
7121nfs4_transform_lock_offset(struct file_lock *lock)
7122{
7123 if (lock->fl_start < 0)
7124 lock->fl_start = OFFSET_MAX;
7125 if (lock->fl_end < 0)
7126 lock->fl_end = OFFSET_MAX;
7127}
7128
7129static fl_owner_t
7130nfsd4_lm_get_owner(fl_owner_t owner)
7131{
7132 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7133
7134 nfs4_get_stateowner(&lo->lo_owner);
7135 return owner;
7136}
7137
7138static void
7139nfsd4_lm_put_owner(fl_owner_t owner)
7140{
7141 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7142
7143 if (lo)
7144 nfs4_put_stateowner(&lo->lo_owner);
7145}
7146
7147/* return pointer to struct nfs4_client if client is expirable */
7148static bool
7149nfsd4_lm_lock_expirable(struct file_lock *cfl)
7150{
7151 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)cfl->fl_owner;
7152 struct nfs4_client *clp = lo->lo_owner.so_client;
7153 struct nfsd_net *nn;
7154
7155 if (try_to_expire_client(clp)) {
7156 nn = net_generic(clp->net, nfsd_net_id);
7157 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7158 return true;
7159 }
7160 return false;
7161}
7162
7163/* schedule laundromat to run immediately and wait for it to complete */
7164static void
7165nfsd4_lm_expire_lock(void)
7166{
7167 flush_workqueue(laundry_wq);
7168}
7169
7170static void
7171nfsd4_lm_notify(struct file_lock *fl)
7172{
7173 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
7174 struct net *net = lo->lo_owner.so_client->net;
7175 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7176 struct nfsd4_blocked_lock *nbl = container_of(fl,
7177 struct nfsd4_blocked_lock, nbl_lock);
7178 bool queue = false;
7179
7180 /* An empty list means that something else is going to be using it */
7181 spin_lock(&nn->blocked_locks_lock);
7182 if (!list_empty(&nbl->nbl_list)) {
7183 list_del_init(&nbl->nbl_list);
7184 list_del_init(&nbl->nbl_lru);
7185 queue = true;
7186 }
7187 spin_unlock(&nn->blocked_locks_lock);
7188
7189 if (queue) {
7190 trace_nfsd_cb_notify_lock(lo, nbl);
7191 nfsd4_run_cb(&nbl->nbl_cb);
7192 }
7193}
7194
7195static const struct lock_manager_operations nfsd_posix_mng_ops = {
7196 .lm_mod_owner = THIS_MODULE,
7197 .lm_notify = nfsd4_lm_notify,
7198 .lm_get_owner = nfsd4_lm_get_owner,
7199 .lm_put_owner = nfsd4_lm_put_owner,
7200 .lm_lock_expirable = nfsd4_lm_lock_expirable,
7201 .lm_expire_lock = nfsd4_lm_expire_lock,
7202};
7203
7204static inline void
7205nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7206{
7207 struct nfs4_lockowner *lo;
7208
7209 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7210 lo = (struct nfs4_lockowner *) fl->fl_owner;
7211 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7212 GFP_KERNEL);
7213 if (!deny->ld_owner.data)
7214 /* We just don't care that much */
7215 goto nevermind;
7216 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7217 } else {
7218nevermind:
7219 deny->ld_owner.len = 0;
7220 deny->ld_owner.data = NULL;
7221 deny->ld_clientid.cl_boot = 0;
7222 deny->ld_clientid.cl_id = 0;
7223 }
7224 deny->ld_start = fl->fl_start;
7225 deny->ld_length = NFS4_MAX_UINT64;
7226 if (fl->fl_end != NFS4_MAX_UINT64)
7227 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7228 deny->ld_type = NFS4_READ_LT;
7229 if (fl->fl_type != F_RDLCK)
7230 deny->ld_type = NFS4_WRITE_LT;
7231}
7232
7233static struct nfs4_lockowner *
7234find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7235{
7236 unsigned int strhashval = ownerstr_hashval(owner);
7237 struct nfs4_stateowner *so;
7238
7239 lockdep_assert_held(&clp->cl_lock);
7240
7241 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7242 so_strhash) {
7243 if (so->so_is_open_owner)
7244 continue;
7245 if (same_owner_str(so, owner))
7246 return lockowner(nfs4_get_stateowner(so));
7247 }
7248 return NULL;
7249}
7250
7251static struct nfs4_lockowner *
7252find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7253{
7254 struct nfs4_lockowner *lo;
7255
7256 spin_lock(&clp->cl_lock);
7257 lo = find_lockowner_str_locked(clp, owner);
7258 spin_unlock(&clp->cl_lock);
7259 return lo;
7260}
7261
7262static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7263{
7264 unhash_lockowner_locked(lockowner(sop));
7265}
7266
7267static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7268{
7269 struct nfs4_lockowner *lo = lockowner(sop);
7270
7271 kmem_cache_free(lockowner_slab, lo);
7272}
7273
7274static const struct nfs4_stateowner_operations lockowner_ops = {
7275 .so_unhash = nfs4_unhash_lockowner,
7276 .so_free = nfs4_free_lockowner,
7277};
7278
7279/*
7280 * Alloc a lock owner structure.
7281 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7282 * occurred.
7283 *
7284 * strhashval = ownerstr_hashval
7285 */
7286static struct nfs4_lockowner *
7287alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7288 struct nfs4_ol_stateid *open_stp,
7289 struct nfsd4_lock *lock)
7290{
7291 struct nfs4_lockowner *lo, *ret;
7292
7293 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7294 if (!lo)
7295 return NULL;
7296 INIT_LIST_HEAD(&lo->lo_blocked);
7297 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7298 lo->lo_owner.so_is_open_owner = 0;
7299 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7300 lo->lo_owner.so_ops = &lockowner_ops;
7301 spin_lock(&clp->cl_lock);
7302 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7303 if (ret == NULL) {
7304 list_add(&lo->lo_owner.so_strhash,
7305 &clp->cl_ownerstr_hashtbl[strhashval]);
7306 ret = lo;
7307 } else
7308 nfs4_free_stateowner(&lo->lo_owner);
7309
7310 spin_unlock(&clp->cl_lock);
7311 return ret;
7312}
7313
7314static struct nfs4_ol_stateid *
7315find_lock_stateid(const struct nfs4_lockowner *lo,
7316 const struct nfs4_ol_stateid *ost)
7317{
7318 struct nfs4_ol_stateid *lst;
7319
7320 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7321
7322 /* If ost is not hashed, ost->st_locks will not be valid */
7323 if (!nfs4_ol_stateid_unhashed(ost))
7324 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7325 if (lst->st_stateowner == &lo->lo_owner) {
7326 refcount_inc(&lst->st_stid.sc_count);
7327 return lst;
7328 }
7329 }
7330 return NULL;
7331}
7332
7333static struct nfs4_ol_stateid *
7334init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7335 struct nfs4_file *fp, struct inode *inode,
7336 struct nfs4_ol_stateid *open_stp)
7337{
7338 struct nfs4_client *clp = lo->lo_owner.so_client;
7339 struct nfs4_ol_stateid *retstp;
7340
7341 mutex_init(&stp->st_mutex);
7342 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7343retry:
7344 spin_lock(&clp->cl_lock);
7345 if (nfs4_ol_stateid_unhashed(open_stp))
7346 goto out_close;
7347 retstp = find_lock_stateid(lo, open_stp);
7348 if (retstp)
7349 goto out_found;
7350 refcount_inc(&stp->st_stid.sc_count);
7351 stp->st_stid.sc_type = NFS4_LOCK_STID;
7352 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7353 get_nfs4_file(fp);
7354 stp->st_stid.sc_file = fp;
7355 stp->st_access_bmap = 0;
7356 stp->st_deny_bmap = open_stp->st_deny_bmap;
7357 stp->st_openstp = open_stp;
7358 spin_lock(&fp->fi_lock);
7359 list_add(&stp->st_locks, &open_stp->st_locks);
7360 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7361 list_add(&stp->st_perfile, &fp->fi_stateids);
7362 spin_unlock(&fp->fi_lock);
7363 spin_unlock(&clp->cl_lock);
7364 return stp;
7365out_found:
7366 spin_unlock(&clp->cl_lock);
7367 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7368 nfs4_put_stid(&retstp->st_stid);
7369 goto retry;
7370 }
7371 /* To keep mutex tracking happy */
7372 mutex_unlock(&stp->st_mutex);
7373 return retstp;
7374out_close:
7375 spin_unlock(&clp->cl_lock);
7376 mutex_unlock(&stp->st_mutex);
7377 return NULL;
7378}
7379
7380static struct nfs4_ol_stateid *
7381find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7382 struct inode *inode, struct nfs4_ol_stateid *ost,
7383 bool *new)
7384{
7385 struct nfs4_stid *ns = NULL;
7386 struct nfs4_ol_stateid *lst;
7387 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7388 struct nfs4_client *clp = oo->oo_owner.so_client;
7389
7390 *new = false;
7391 spin_lock(&clp->cl_lock);
7392 lst = find_lock_stateid(lo, ost);
7393 spin_unlock(&clp->cl_lock);
7394 if (lst != NULL) {
7395 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7396 goto out;
7397 nfs4_put_stid(&lst->st_stid);
7398 }
7399 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7400 if (ns == NULL)
7401 return NULL;
7402
7403 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7404 if (lst == openlockstateid(ns))
7405 *new = true;
7406 else
7407 nfs4_put_stid(ns);
7408out:
7409 return lst;
7410}
7411
7412static int
7413check_lock_length(u64 offset, u64 length)
7414{
7415 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7416 (length > ~offset)));
7417}
7418
7419static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7420{
7421 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7422
7423 lockdep_assert_held(&fp->fi_lock);
7424
7425 if (test_access(access, lock_stp))
7426 return;
7427 __nfs4_file_get_access(fp, access);
7428 set_access(access, lock_stp);
7429}
7430
7431static __be32
7432lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7433 struct nfs4_ol_stateid *ost,
7434 struct nfsd4_lock *lock,
7435 struct nfs4_ol_stateid **plst, bool *new)
7436{
7437 __be32 status;
7438 struct nfs4_file *fi = ost->st_stid.sc_file;
7439 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7440 struct nfs4_client *cl = oo->oo_owner.so_client;
7441 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7442 struct nfs4_lockowner *lo;
7443 struct nfs4_ol_stateid *lst;
7444 unsigned int strhashval;
7445
7446 lo = find_lockowner_str(cl, &lock->lk_new_owner);
7447 if (!lo) {
7448 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7449 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7450 if (lo == NULL)
7451 return nfserr_jukebox;
7452 } else {
7453 /* with an existing lockowner, seqids must be the same */
7454 status = nfserr_bad_seqid;
7455 if (!cstate->minorversion &&
7456 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7457 goto out;
7458 }
7459
7460 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7461 if (lst == NULL) {
7462 status = nfserr_jukebox;
7463 goto out;
7464 }
7465
7466 status = nfs_ok;
7467 *plst = lst;
7468out:
7469 nfs4_put_stateowner(&lo->lo_owner);
7470 return status;
7471}
7472
7473/*
7474 * LOCK operation
7475 */
7476__be32
7477nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7478 union nfsd4_op_u *u)
7479{
7480 struct nfsd4_lock *lock = &u->lock;
7481 struct nfs4_openowner *open_sop = NULL;
7482 struct nfs4_lockowner *lock_sop = NULL;
7483 struct nfs4_ol_stateid *lock_stp = NULL;
7484 struct nfs4_ol_stateid *open_stp = NULL;
7485 struct nfs4_file *fp;
7486 struct nfsd_file *nf = NULL;
7487 struct nfsd4_blocked_lock *nbl = NULL;
7488 struct file_lock *file_lock = NULL;
7489 struct file_lock *conflock = NULL;
7490 struct super_block *sb;
7491 __be32 status = 0;
7492 int lkflg;
7493 int err;
7494 bool new = false;
7495 unsigned char fl_type;
7496 unsigned int fl_flags = FL_POSIX;
7497 struct net *net = SVC_NET(rqstp);
7498 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7499
7500 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7501 (long long) lock->lk_offset,
7502 (long long) lock->lk_length);
7503
7504 if (check_lock_length(lock->lk_offset, lock->lk_length))
7505 return nfserr_inval;
7506
7507 if ((status = fh_verify(rqstp, &cstate->current_fh,
7508 S_IFREG, NFSD_MAY_LOCK))) {
7509 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7510 return status;
7511 }
7512 sb = cstate->current_fh.fh_dentry->d_sb;
7513
7514 if (lock->lk_is_new) {
7515 if (nfsd4_has_session(cstate))
7516 /* See rfc 5661 18.10.3: given clientid is ignored: */
7517 memcpy(&lock->lk_new_clientid,
7518 &cstate->clp->cl_clientid,
7519 sizeof(clientid_t));
7520
7521 /* validate and update open stateid and open seqid */
7522 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7523 lock->lk_new_open_seqid,
7524 &lock->lk_new_open_stateid,
7525 &open_stp, nn);
7526 if (status)
7527 goto out;
7528 mutex_unlock(&open_stp->st_mutex);
7529 open_sop = openowner(open_stp->st_stateowner);
7530 status = nfserr_bad_stateid;
7531 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7532 &lock->lk_new_clientid))
7533 goto out;
7534 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7535 &lock_stp, &new);
7536 } else {
7537 status = nfs4_preprocess_seqid_op(cstate,
7538 lock->lk_old_lock_seqid,
7539 &lock->lk_old_lock_stateid,
7540 NFS4_LOCK_STID, &lock_stp, nn);
7541 }
7542 if (status)
7543 goto out;
7544 lock_sop = lockowner(lock_stp->st_stateowner);
7545
7546 lkflg = setlkflg(lock->lk_type);
7547 status = nfs4_check_openmode(lock_stp, lkflg);
7548 if (status)
7549 goto out;
7550
7551 status = nfserr_grace;
7552 if (locks_in_grace(net) && !lock->lk_reclaim)
7553 goto out;
7554 status = nfserr_no_grace;
7555 if (!locks_in_grace(net) && lock->lk_reclaim)
7556 goto out;
7557
7558 if (lock->lk_reclaim)
7559 fl_flags |= FL_RECLAIM;
7560
7561 fp = lock_stp->st_stid.sc_file;
7562 switch (lock->lk_type) {
7563 case NFS4_READW_LT:
7564 if (nfsd4_has_session(cstate) ||
7565 exportfs_lock_op_is_async(sb->s_export_op))
7566 fl_flags |= FL_SLEEP;
7567 fallthrough;
7568 case NFS4_READ_LT:
7569 spin_lock(&fp->fi_lock);
7570 nf = find_readable_file_locked(fp);
7571 if (nf)
7572 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7573 spin_unlock(&fp->fi_lock);
7574 fl_type = F_RDLCK;
7575 break;
7576 case NFS4_WRITEW_LT:
7577 if (nfsd4_has_session(cstate) ||
7578 exportfs_lock_op_is_async(sb->s_export_op))
7579 fl_flags |= FL_SLEEP;
7580 fallthrough;
7581 case NFS4_WRITE_LT:
7582 spin_lock(&fp->fi_lock);
7583 nf = find_writeable_file_locked(fp);
7584 if (nf)
7585 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7586 spin_unlock(&fp->fi_lock);
7587 fl_type = F_WRLCK;
7588 break;
7589 default:
7590 status = nfserr_inval;
7591 goto out;
7592 }
7593
7594 if (!nf) {
7595 status = nfserr_openmode;
7596 goto out;
7597 }
7598
7599 /*
7600 * Most filesystems with their own ->lock operations will block
7601 * the nfsd thread waiting to acquire the lock. That leads to
7602 * deadlocks (we don't want every nfsd thread tied up waiting
7603 * for file locks), so don't attempt blocking lock notifications
7604 * on those filesystems:
7605 */
7606 if (!exportfs_lock_op_is_async(sb->s_export_op))
7607 fl_flags &= ~FL_SLEEP;
7608
7609 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7610 if (!nbl) {
7611 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7612 status = nfserr_jukebox;
7613 goto out;
7614 }
7615
7616 file_lock = &nbl->nbl_lock;
7617 file_lock->fl_type = fl_type;
7618 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7619 file_lock->fl_pid = current->tgid;
7620 file_lock->fl_file = nf->nf_file;
7621 file_lock->fl_flags = fl_flags;
7622 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7623 file_lock->fl_start = lock->lk_offset;
7624 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7625 nfs4_transform_lock_offset(file_lock);
7626
7627 conflock = locks_alloc_lock();
7628 if (!conflock) {
7629 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7630 status = nfserr_jukebox;
7631 goto out;
7632 }
7633
7634 if (fl_flags & FL_SLEEP) {
7635 nbl->nbl_time = ktime_get_boottime_seconds();
7636 spin_lock(&nn->blocked_locks_lock);
7637 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7638 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7639 kref_get(&nbl->nbl_kref);
7640 spin_unlock(&nn->blocked_locks_lock);
7641 }
7642
7643 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7644 switch (err) {
7645 case 0: /* success! */
7646 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7647 status = 0;
7648 if (lock->lk_reclaim)
7649 nn->somebody_reclaimed = true;
7650 break;
7651 case FILE_LOCK_DEFERRED:
7652 kref_put(&nbl->nbl_kref, free_nbl);
7653 nbl = NULL;
7654 fallthrough;
7655 case -EAGAIN: /* conflock holds conflicting lock */
7656 status = nfserr_denied;
7657 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
7658 nfs4_set_lock_denied(conflock, &lock->lk_denied);
7659 break;
7660 case -EDEADLK:
7661 status = nfserr_deadlock;
7662 break;
7663 default:
7664 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
7665 status = nfserrno(err);
7666 break;
7667 }
7668out:
7669 if (nbl) {
7670 /* dequeue it if we queued it before */
7671 if (fl_flags & FL_SLEEP) {
7672 spin_lock(&nn->blocked_locks_lock);
7673 if (!list_empty(&nbl->nbl_list) &&
7674 !list_empty(&nbl->nbl_lru)) {
7675 list_del_init(&nbl->nbl_list);
7676 list_del_init(&nbl->nbl_lru);
7677 kref_put(&nbl->nbl_kref, free_nbl);
7678 }
7679 /* nbl can use one of lists to be linked to reaplist */
7680 spin_unlock(&nn->blocked_locks_lock);
7681 }
7682 free_blocked_lock(nbl);
7683 }
7684 if (nf)
7685 nfsd_file_put(nf);
7686 if (lock_stp) {
7687 /* Bump seqid manually if the 4.0 replay owner is openowner */
7688 if (cstate->replay_owner &&
7689 cstate->replay_owner != &lock_sop->lo_owner &&
7690 seqid_mutating_err(ntohl(status)))
7691 lock_sop->lo_owner.so_seqid++;
7692
7693 /*
7694 * If this is a new, never-before-used stateid, and we are
7695 * returning an error, then just go ahead and release it.
7696 */
7697 if (status && new)
7698 release_lock_stateid(lock_stp);
7699
7700 mutex_unlock(&lock_stp->st_mutex);
7701
7702 nfs4_put_stid(&lock_stp->st_stid);
7703 }
7704 if (open_stp)
7705 nfs4_put_stid(&open_stp->st_stid);
7706 nfsd4_bump_seqid(cstate, status);
7707 if (conflock)
7708 locks_free_lock(conflock);
7709 return status;
7710}
7711
7712void nfsd4_lock_release(union nfsd4_op_u *u)
7713{
7714 struct nfsd4_lock *lock = &u->lock;
7715 struct nfsd4_lock_denied *deny = &lock->lk_denied;
7716
7717 kfree(deny->ld_owner.data);
7718}
7719
7720/*
7721 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
7722 * so we do a temporary open here just to get an open file to pass to
7723 * vfs_test_lock.
7724 */
7725static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
7726{
7727 struct nfsd_file *nf;
7728 struct inode *inode;
7729 __be32 err;
7730
7731 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
7732 if (err)
7733 return err;
7734 inode = fhp->fh_dentry->d_inode;
7735 inode_lock(inode); /* to block new leases till after test_lock: */
7736 err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
7737 if (err)
7738 goto out;
7739 lock->fl_file = nf->nf_file;
7740 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
7741 lock->fl_file = NULL;
7742out:
7743 inode_unlock(inode);
7744 nfsd_file_put(nf);
7745 return err;
7746}
7747
7748/*
7749 * LOCKT operation
7750 */
7751__be32
7752nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7753 union nfsd4_op_u *u)
7754{
7755 struct nfsd4_lockt *lockt = &u->lockt;
7756 struct file_lock *file_lock = NULL;
7757 struct nfs4_lockowner *lo = NULL;
7758 __be32 status;
7759 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7760
7761 if (locks_in_grace(SVC_NET(rqstp)))
7762 return nfserr_grace;
7763
7764 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
7765 return nfserr_inval;
7766
7767 if (!nfsd4_has_session(cstate)) {
7768 status = set_client(&lockt->lt_clientid, cstate, nn);
7769 if (status)
7770 goto out;
7771 }
7772
7773 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7774 goto out;
7775
7776 file_lock = locks_alloc_lock();
7777 if (!file_lock) {
7778 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7779 status = nfserr_jukebox;
7780 goto out;
7781 }
7782
7783 switch (lockt->lt_type) {
7784 case NFS4_READ_LT:
7785 case NFS4_READW_LT:
7786 file_lock->fl_type = F_RDLCK;
7787 break;
7788 case NFS4_WRITE_LT:
7789 case NFS4_WRITEW_LT:
7790 file_lock->fl_type = F_WRLCK;
7791 break;
7792 default:
7793 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
7794 status = nfserr_inval;
7795 goto out;
7796 }
7797
7798 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
7799 if (lo)
7800 file_lock->fl_owner = (fl_owner_t)lo;
7801 file_lock->fl_pid = current->tgid;
7802 file_lock->fl_flags = FL_POSIX;
7803
7804 file_lock->fl_start = lockt->lt_offset;
7805 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
7806
7807 nfs4_transform_lock_offset(file_lock);
7808
7809 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
7810 if (status)
7811 goto out;
7812
7813 if (file_lock->fl_type != F_UNLCK) {
7814 status = nfserr_denied;
7815 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
7816 }
7817out:
7818 if (lo)
7819 nfs4_put_stateowner(&lo->lo_owner);
7820 if (file_lock)
7821 locks_free_lock(file_lock);
7822 return status;
7823}
7824
7825void nfsd4_lockt_release(union nfsd4_op_u *u)
7826{
7827 struct nfsd4_lockt *lockt = &u->lockt;
7828 struct nfsd4_lock_denied *deny = &lockt->lt_denied;
7829
7830 kfree(deny->ld_owner.data);
7831}
7832
7833__be32
7834nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7835 union nfsd4_op_u *u)
7836{
7837 struct nfsd4_locku *locku = &u->locku;
7838 struct nfs4_ol_stateid *stp;
7839 struct nfsd_file *nf = NULL;
7840 struct file_lock *file_lock = NULL;
7841 __be32 status;
7842 int err;
7843 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7844
7845 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
7846 (long long) locku->lu_offset,
7847 (long long) locku->lu_length);
7848
7849 if (check_lock_length(locku->lu_offset, locku->lu_length))
7850 return nfserr_inval;
7851
7852 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
7853 &locku->lu_stateid, NFS4_LOCK_STID,
7854 &stp, nn);
7855 if (status)
7856 goto out;
7857 nf = find_any_file(stp->st_stid.sc_file);
7858 if (!nf) {
7859 status = nfserr_lock_range;
7860 goto put_stateid;
7861 }
7862 file_lock = locks_alloc_lock();
7863 if (!file_lock) {
7864 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7865 status = nfserr_jukebox;
7866 goto put_file;
7867 }
7868
7869 file_lock->fl_type = F_UNLCK;
7870 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7871 file_lock->fl_pid = current->tgid;
7872 file_lock->fl_file = nf->nf_file;
7873 file_lock->fl_flags = FL_POSIX;
7874 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7875 file_lock->fl_start = locku->lu_offset;
7876
7877 file_lock->fl_end = last_byte_offset(locku->lu_offset,
7878 locku->lu_length);
7879 nfs4_transform_lock_offset(file_lock);
7880
7881 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7882 if (err) {
7883 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7884 goto out_nfserr;
7885 }
7886 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7887put_file:
7888 nfsd_file_put(nf);
7889put_stateid:
7890 mutex_unlock(&stp->st_mutex);
7891 nfs4_put_stid(&stp->st_stid);
7892out:
7893 nfsd4_bump_seqid(cstate, status);
7894 if (file_lock)
7895 locks_free_lock(file_lock);
7896 return status;
7897
7898out_nfserr:
7899 status = nfserrno(err);
7900 goto put_file;
7901}
7902
7903/*
7904 * returns
7905 * true: locks held by lockowner
7906 * false: no locks held by lockowner
7907 */
7908static bool
7909check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7910{
7911 struct file_lock *fl;
7912 int status = false;
7913 struct nfsd_file *nf;
7914 struct inode *inode;
7915 struct file_lock_context *flctx;
7916
7917 spin_lock(&fp->fi_lock);
7918 nf = find_any_file_locked(fp);
7919 if (!nf) {
7920 /* Any valid lock stateid should have some sort of access */
7921 WARN_ON_ONCE(1);
7922 goto out;
7923 }
7924
7925 inode = file_inode(nf->nf_file);
7926 flctx = locks_inode_context(inode);
7927
7928 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7929 spin_lock(&flctx->flc_lock);
7930 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7931 if (fl->fl_owner == (fl_owner_t)lowner) {
7932 status = true;
7933 break;
7934 }
7935 }
7936 spin_unlock(&flctx->flc_lock);
7937 }
7938out:
7939 spin_unlock(&fp->fi_lock);
7940 return status;
7941}
7942
7943/**
7944 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
7945 * @rqstp: RPC transaction
7946 * @cstate: NFSv4 COMPOUND state
7947 * @u: RELEASE_LOCKOWNER arguments
7948 *
7949 * Check if theree are any locks still held and if not - free the lockowner
7950 * and any lock state that is owned.
7951 *
7952 * Return values:
7953 * %nfs_ok: lockowner released or not found
7954 * %nfserr_locks_held: lockowner still in use
7955 * %nfserr_stale_clientid: clientid no longer active
7956 * %nfserr_expired: clientid not recognized
7957 */
7958__be32
7959nfsd4_release_lockowner(struct svc_rqst *rqstp,
7960 struct nfsd4_compound_state *cstate,
7961 union nfsd4_op_u *u)
7962{
7963 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7964 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7965 clientid_t *clid = &rlockowner->rl_clientid;
7966 struct nfs4_ol_stateid *stp;
7967 struct nfs4_lockowner *lo;
7968 struct nfs4_client *clp;
7969 LIST_HEAD(reaplist);
7970 __be32 status;
7971
7972 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7973 clid->cl_boot, clid->cl_id);
7974
7975 status = set_client(clid, cstate, nn);
7976 if (status)
7977 return status;
7978 clp = cstate->clp;
7979
7980 spin_lock(&clp->cl_lock);
7981 lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
7982 if (!lo) {
7983 spin_unlock(&clp->cl_lock);
7984 return nfs_ok;
7985 }
7986
7987 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
7988 if (check_for_locks(stp->st_stid.sc_file, lo)) {
7989 spin_unlock(&clp->cl_lock);
7990 nfs4_put_stateowner(&lo->lo_owner);
7991 return nfserr_locks_held;
7992 }
7993 }
7994 unhash_lockowner_locked(lo);
7995 while (!list_empty(&lo->lo_owner.so_stateids)) {
7996 stp = list_first_entry(&lo->lo_owner.so_stateids,
7997 struct nfs4_ol_stateid,
7998 st_perstateowner);
7999 WARN_ON(!unhash_lock_stateid(stp));
8000 put_ol_stateid_locked(stp, &reaplist);
8001 }
8002 spin_unlock(&clp->cl_lock);
8003
8004 free_ol_stateid_reaplist(&reaplist);
8005 remove_blocked_locks(lo);
8006 nfs4_put_stateowner(&lo->lo_owner);
8007 return nfs_ok;
8008}
8009
8010static inline struct nfs4_client_reclaim *
8011alloc_reclaim(void)
8012{
8013 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8014}
8015
8016bool
8017nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8018{
8019 struct nfs4_client_reclaim *crp;
8020
8021 crp = nfsd4_find_reclaim_client(name, nn);
8022 return (crp && crp->cr_clp);
8023}
8024
8025/*
8026 * failure => all reset bets are off, nfserr_no_grace...
8027 *
8028 * The caller is responsible for freeing name.data if NULL is returned (it
8029 * will be freed in nfs4_remove_reclaim_record in the normal case).
8030 */
8031struct nfs4_client_reclaim *
8032nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8033 struct nfsd_net *nn)
8034{
8035 unsigned int strhashval;
8036 struct nfs4_client_reclaim *crp;
8037
8038 crp = alloc_reclaim();
8039 if (crp) {
8040 strhashval = clientstr_hashval(name);
8041 INIT_LIST_HEAD(&crp->cr_strhash);
8042 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8043 crp->cr_name.data = name.data;
8044 crp->cr_name.len = name.len;
8045 crp->cr_princhash.data = princhash.data;
8046 crp->cr_princhash.len = princhash.len;
8047 crp->cr_clp = NULL;
8048 nn->reclaim_str_hashtbl_size++;
8049 }
8050 return crp;
8051}
8052
8053void
8054nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8055{
8056 list_del(&crp->cr_strhash);
8057 kfree(crp->cr_name.data);
8058 kfree(crp->cr_princhash.data);
8059 kfree(crp);
8060 nn->reclaim_str_hashtbl_size--;
8061}
8062
8063void
8064nfs4_release_reclaim(struct nfsd_net *nn)
8065{
8066 struct nfs4_client_reclaim *crp = NULL;
8067 int i;
8068
8069 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8070 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8071 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8072 struct nfs4_client_reclaim, cr_strhash);
8073 nfs4_remove_reclaim_record(crp, nn);
8074 }
8075 }
8076 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8077}
8078
8079/*
8080 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8081struct nfs4_client_reclaim *
8082nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8083{
8084 unsigned int strhashval;
8085 struct nfs4_client_reclaim *crp = NULL;
8086
8087 strhashval = clientstr_hashval(name);
8088 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8089 if (compare_blob(&crp->cr_name, &name) == 0) {
8090 return crp;
8091 }
8092 }
8093 return NULL;
8094}
8095
8096__be32
8097nfs4_check_open_reclaim(struct nfs4_client *clp)
8098{
8099 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8100 return nfserr_no_grace;
8101
8102 if (nfsd4_client_record_check(clp))
8103 return nfserr_reclaim_bad;
8104
8105 return nfs_ok;
8106}
8107
8108/*
8109 * Since the lifetime of a delegation isn't limited to that of an open, a
8110 * client may quite reasonably hang on to a delegation as long as it has
8111 * the inode cached. This becomes an obvious problem the first time a
8112 * client's inode cache approaches the size of the server's total memory.
8113 *
8114 * For now we avoid this problem by imposing a hard limit on the number
8115 * of delegations, which varies according to the server's memory size.
8116 */
8117static void
8118set_max_delegations(void)
8119{
8120 /*
8121 * Allow at most 4 delegations per megabyte of RAM. Quick
8122 * estimates suggest that in the worst case (where every delegation
8123 * is for a different inode), a delegation could take about 1.5K,
8124 * giving a worst case usage of about 6% of memory.
8125 */
8126 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8127}
8128
8129static int nfs4_state_create_net(struct net *net)
8130{
8131 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8132 int i;
8133
8134 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8135 sizeof(struct list_head),
8136 GFP_KERNEL);
8137 if (!nn->conf_id_hashtbl)
8138 goto err;
8139 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8140 sizeof(struct list_head),
8141 GFP_KERNEL);
8142 if (!nn->unconf_id_hashtbl)
8143 goto err_unconf_id;
8144 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8145 sizeof(struct list_head),
8146 GFP_KERNEL);
8147 if (!nn->sessionid_hashtbl)
8148 goto err_sessionid;
8149
8150 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8151 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8152 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8153 }
8154 for (i = 0; i < SESSION_HASH_SIZE; i++)
8155 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8156 nn->conf_name_tree = RB_ROOT;
8157 nn->unconf_name_tree = RB_ROOT;
8158 nn->boot_time = ktime_get_real_seconds();
8159 nn->grace_ended = false;
8160 nn->nfsd4_manager.block_opens = true;
8161 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8162 INIT_LIST_HEAD(&nn->client_lru);
8163 INIT_LIST_HEAD(&nn->close_lru);
8164 INIT_LIST_HEAD(&nn->del_recall_lru);
8165 spin_lock_init(&nn->client_lock);
8166 spin_lock_init(&nn->s2s_cp_lock);
8167 idr_init(&nn->s2s_cp_stateids);
8168
8169 spin_lock_init(&nn->blocked_locks_lock);
8170 INIT_LIST_HEAD(&nn->blocked_locks_lru);
8171
8172 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8173 INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8174 get_net(net);
8175
8176 nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8177 if (!nn->nfsd_client_shrinker)
8178 goto err_shrinker;
8179
8180 nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8181 nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8182 nn->nfsd_client_shrinker->private_data = nn;
8183
8184 shrinker_register(nn->nfsd_client_shrinker);
8185
8186 return 0;
8187
8188err_shrinker:
8189 put_net(net);
8190 kfree(nn->sessionid_hashtbl);
8191err_sessionid:
8192 kfree(nn->unconf_id_hashtbl);
8193err_unconf_id:
8194 kfree(nn->conf_id_hashtbl);
8195err:
8196 return -ENOMEM;
8197}
8198
8199static void
8200nfs4_state_destroy_net(struct net *net)
8201{
8202 int i;
8203 struct nfs4_client *clp = NULL;
8204 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8205
8206 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8207 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8208 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8209 destroy_client(clp);
8210 }
8211 }
8212
8213 WARN_ON(!list_empty(&nn->blocked_locks_lru));
8214
8215 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8216 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8217 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8218 destroy_client(clp);
8219 }
8220 }
8221
8222 kfree(nn->sessionid_hashtbl);
8223 kfree(nn->unconf_id_hashtbl);
8224 kfree(nn->conf_id_hashtbl);
8225 put_net(net);
8226}
8227
8228int
8229nfs4_state_start_net(struct net *net)
8230{
8231 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8232 int ret;
8233
8234 ret = nfs4_state_create_net(net);
8235 if (ret)
8236 return ret;
8237 locks_start_grace(net, &nn->nfsd4_manager);
8238 nfsd4_client_tracking_init(net);
8239 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8240 goto skip_grace;
8241 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8242 nn->nfsd4_grace, net->ns.inum);
8243 trace_nfsd_grace_start(nn);
8244 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8245 return 0;
8246
8247skip_grace:
8248 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8249 net->ns.inum);
8250 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8251 nfsd4_end_grace(nn);
8252 return 0;
8253}
8254
8255/* initialization to perform when the nfsd service is started: */
8256
8257int
8258nfs4_state_start(void)
8259{
8260 int ret;
8261
8262 ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8263 if (ret)
8264 return ret;
8265
8266 ret = nfsd4_create_callback_queue();
8267 if (ret) {
8268 rhltable_destroy(&nfs4_file_rhltable);
8269 return ret;
8270 }
8271
8272 set_max_delegations();
8273 return 0;
8274}
8275
8276void
8277nfs4_state_shutdown_net(struct net *net)
8278{
8279 struct nfs4_delegation *dp = NULL;
8280 struct list_head *pos, *next, reaplist;
8281 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8282
8283 shrinker_free(nn->nfsd_client_shrinker);
8284 cancel_work(&nn->nfsd_shrinker_work);
8285 cancel_delayed_work_sync(&nn->laundromat_work);
8286 locks_end_grace(&nn->nfsd4_manager);
8287
8288 INIT_LIST_HEAD(&reaplist);
8289 spin_lock(&state_lock);
8290 list_for_each_safe(pos, next, &nn->del_recall_lru) {
8291 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8292 WARN_ON(!unhash_delegation_locked(dp));
8293 list_add(&dp->dl_recall_lru, &reaplist);
8294 }
8295 spin_unlock(&state_lock);
8296 list_for_each_safe(pos, next, &reaplist) {
8297 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8298 list_del_init(&dp->dl_recall_lru);
8299 destroy_unhashed_deleg(dp);
8300 }
8301
8302 nfsd4_client_tracking_exit(net);
8303 nfs4_state_destroy_net(net);
8304#ifdef CONFIG_NFSD_V4_2_INTER_SSC
8305 nfsd4_ssc_shutdown_umount(nn);
8306#endif
8307}
8308
8309void
8310nfs4_state_shutdown(void)
8311{
8312 nfsd4_destroy_callback_queue();
8313 rhltable_destroy(&nfs4_file_rhltable);
8314}
8315
8316static void
8317get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8318{
8319 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8320 CURRENT_STATEID(stateid))
8321 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8322}
8323
8324static void
8325put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8326{
8327 if (cstate->minorversion) {
8328 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8329 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8330 }
8331}
8332
8333void
8334clear_current_stateid(struct nfsd4_compound_state *cstate)
8335{
8336 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8337}
8338
8339/*
8340 * functions to set current state id
8341 */
8342void
8343nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8344 union nfsd4_op_u *u)
8345{
8346 put_stateid(cstate, &u->open_downgrade.od_stateid);
8347}
8348
8349void
8350nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8351 union nfsd4_op_u *u)
8352{
8353 put_stateid(cstate, &u->open.op_stateid);
8354}
8355
8356void
8357nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8358 union nfsd4_op_u *u)
8359{
8360 put_stateid(cstate, &u->close.cl_stateid);
8361}
8362
8363void
8364nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8365 union nfsd4_op_u *u)
8366{
8367 put_stateid(cstate, &u->lock.lk_resp_stateid);
8368}
8369
8370/*
8371 * functions to consume current state id
8372 */
8373
8374void
8375nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8376 union nfsd4_op_u *u)
8377{
8378 get_stateid(cstate, &u->open_downgrade.od_stateid);
8379}
8380
8381void
8382nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8383 union nfsd4_op_u *u)
8384{
8385 get_stateid(cstate, &u->delegreturn.dr_stateid);
8386}
8387
8388void
8389nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8390 union nfsd4_op_u *u)
8391{
8392 get_stateid(cstate, &u->free_stateid.fr_stateid);
8393}
8394
8395void
8396nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8397 union nfsd4_op_u *u)
8398{
8399 get_stateid(cstate, &u->setattr.sa_stateid);
8400}
8401
8402void
8403nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8404 union nfsd4_op_u *u)
8405{
8406 get_stateid(cstate, &u->close.cl_stateid);
8407}
8408
8409void
8410nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8411 union nfsd4_op_u *u)
8412{
8413 get_stateid(cstate, &u->locku.lu_stateid);
8414}
8415
8416void
8417nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8418 union nfsd4_op_u *u)
8419{
8420 get_stateid(cstate, &u->read.rd_stateid);
8421}
8422
8423void
8424nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8425 union nfsd4_op_u *u)
8426{
8427 get_stateid(cstate, &u->write.wr_stateid);
8428}
8429
8430/**
8431 * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8432 * @rqstp: RPC transaction context
8433 * @inode: file to be checked for a conflict
8434 *
8435 * This function is called when there is a conflict between a write
8436 * delegation and a change/size GETATTR from another client. The server
8437 * must either use the CB_GETATTR to get the current values of the
8438 * attributes from the client that holds the delegation or recall the
8439 * delegation before replying to the GETATTR. See RFC 8881 section
8440 * 18.7.4.
8441 *
8442 * The current implementation does not support CB_GETATTR yet. However
8443 * this can avoid recalling the delegation could be added in follow up
8444 * work.
8445 *
8446 * Returns 0 if there is no conflict; otherwise an nfs_stat
8447 * code is returned.
8448 */
8449__be32
8450nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode)
8451{
8452 __be32 status;
8453 struct file_lock_context *ctx;
8454 struct file_lock *fl;
8455 struct nfs4_delegation *dp;
8456
8457 ctx = locks_inode_context(inode);
8458 if (!ctx)
8459 return 0;
8460 spin_lock(&ctx->flc_lock);
8461 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
8462 if (fl->fl_flags == FL_LAYOUT)
8463 continue;
8464 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8465 /*
8466 * non-nfs lease, if it's a lease with F_RDLCK then
8467 * we are done; there isn't any write delegation
8468 * on this inode
8469 */
8470 if (fl->fl_type == F_RDLCK)
8471 break;
8472 goto break_lease;
8473 }
8474 if (fl->fl_type == F_WRLCK) {
8475 dp = fl->fl_owner;
8476 if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8477 spin_unlock(&ctx->flc_lock);
8478 return 0;
8479 }
8480break_lease:
8481 spin_unlock(&ctx->flc_lock);
8482 nfsd_stats_wdeleg_getattr_inc();
8483 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8484 if (status != nfserr_jukebox ||
8485 !nfsd_wait_for_delegreturn(rqstp, inode))
8486 return status;
8487 return 0;
8488 }
8489 break;
8490 }
8491 spin_unlock(&ctx->flc_lock);
8492 return 0;
8493}