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
46#define NFSDDBG_FACILITY NFSDDBG_PROC
47
48/* Globals */
49time_t nfsd4_lease = 90; /* default lease time */
50time_t nfsd4_grace = 90;
51static time_t boot_time;
52static u32 current_ownerid = 1;
53static u32 current_fileid = 1;
54static u32 current_delegid = 1;
55static stateid_t zerostateid; /* bits all 0 */
56static stateid_t onestateid; /* bits all 1 */
57static u64 current_sessionid = 1;
58
59#define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
60#define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61
62/* forward declarations */
63static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
64static struct nfs4_stateid * search_for_stateid(stateid_t *stid);
65static struct nfs4_delegation * search_for_delegation(stateid_t *stid);
66static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
67static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
68static void nfs4_set_recdir(char *recdir);
69static int check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner);
70
71/* Locking: */
72
73/* Currently used for almost all code touching nfsv4 state: */
74static DEFINE_MUTEX(client_mutex);
75
76/*
77 * Currently used for the del_recall_lru and file hash table. In an
78 * effort to decrease the scope of the client_mutex, this spinlock may
79 * eventually cover more:
80 */
81static DEFINE_SPINLOCK(recall_lock);
82
83static struct kmem_cache *stateowner_slab = NULL;
84static struct kmem_cache *file_slab = NULL;
85static struct kmem_cache *stateid_slab = NULL;
86static struct kmem_cache *deleg_slab = NULL;
87
88void
89nfs4_lock_state(void)
90{
91 mutex_lock(&client_mutex);
92}
93
94void
95nfs4_unlock_state(void)
96{
97 mutex_unlock(&client_mutex);
98}
99
100static inline u32
101opaque_hashval(const void *ptr, int nbytes)
102{
103 unsigned char *cptr = (unsigned char *) ptr;
104
105 u32 x = 0;
106 while (nbytes--) {
107 x *= 37;
108 x += *cptr++;
109 }
110 return x;
111}
112
113static struct list_head del_recall_lru;
114
115static inline void
116put_nfs4_file(struct nfs4_file *fi)
117{
118 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
119 list_del(&fi->fi_hash);
120 spin_unlock(&recall_lock);
121 iput(fi->fi_inode);
122 kmem_cache_free(file_slab, fi);
123 }
124}
125
126static inline void
127get_nfs4_file(struct nfs4_file *fi)
128{
129 atomic_inc(&fi->fi_ref);
130}
131
132static int num_delegations;
133unsigned int max_delegations;
134
135/*
136 * Open owner state (share locks)
137 */
138
139/* hash tables for nfs4_stateowner */
140#define OWNER_HASH_BITS 8
141#define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
142#define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
143
144#define ownerid_hashval(id) \
145 ((id) & OWNER_HASH_MASK)
146#define ownerstr_hashval(clientid, ownername) \
147 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
148
149static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
150static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
151
152/* hash table for nfs4_file */
153#define FILE_HASH_BITS 8
154#define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
155
156/* hash table for (open)nfs4_stateid */
157#define STATEID_HASH_BITS 10
158#define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
159#define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
160
161#define file_hashval(x) \
162 hash_ptr(x, FILE_HASH_BITS)
163#define stateid_hashval(owner_id, file_id) \
164 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
165
166static struct list_head file_hashtbl[FILE_HASH_SIZE];
167static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
168
169static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
170{
171 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
172 atomic_inc(&fp->fi_access[oflag]);
173}
174
175static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
176{
177 if (oflag == O_RDWR) {
178 __nfs4_file_get_access(fp, O_RDONLY);
179 __nfs4_file_get_access(fp, O_WRONLY);
180 } else
181 __nfs4_file_get_access(fp, oflag);
182}
183
184static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
185{
186 if (fp->fi_fds[oflag]) {
187 fput(fp->fi_fds[oflag]);
188 fp->fi_fds[oflag] = NULL;
189 }
190}
191
192static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
193{
194 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
195 nfs4_file_put_fd(fp, O_RDWR);
196 nfs4_file_put_fd(fp, oflag);
197 }
198}
199
200static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
201{
202 if (oflag == O_RDWR) {
203 __nfs4_file_put_access(fp, O_RDONLY);
204 __nfs4_file_put_access(fp, O_WRONLY);
205 } else
206 __nfs4_file_put_access(fp, oflag);
207}
208
209static struct nfs4_delegation *
210alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
211{
212 struct nfs4_delegation *dp;
213 struct nfs4_file *fp = stp->st_file;
214
215 dprintk("NFSD alloc_init_deleg\n");
216 /*
217 * Major work on the lease subsystem (for example, to support
218 * calbacks on stat) will be required before we can support
219 * write delegations properly.
220 */
221 if (type != NFS4_OPEN_DELEGATE_READ)
222 return NULL;
223 if (fp->fi_had_conflict)
224 return NULL;
225 if (num_delegations > max_delegations)
226 return NULL;
227 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
228 if (dp == NULL)
229 return dp;
230 num_delegations++;
231 INIT_LIST_HEAD(&dp->dl_perfile);
232 INIT_LIST_HEAD(&dp->dl_perclnt);
233 INIT_LIST_HEAD(&dp->dl_recall_lru);
234 dp->dl_client = clp;
235 get_nfs4_file(fp);
236 dp->dl_file = fp;
237 dp->dl_type = type;
238 dp->dl_stateid.si_boot = boot_time;
239 dp->dl_stateid.si_stateownerid = current_delegid++;
240 dp->dl_stateid.si_fileid = 0;
241 dp->dl_stateid.si_generation = 0;
242 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
243 dp->dl_time = 0;
244 atomic_set(&dp->dl_count, 1);
245 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
246 return dp;
247}
248
249void
250nfs4_put_delegation(struct nfs4_delegation *dp)
251{
252 if (atomic_dec_and_test(&dp->dl_count)) {
253 dprintk("NFSD: freeing dp %p\n",dp);
254 put_nfs4_file(dp->dl_file);
255 kmem_cache_free(deleg_slab, dp);
256 num_delegations--;
257 }
258}
259
260static void nfs4_put_deleg_lease(struct nfs4_file *fp)
261{
262 if (atomic_dec_and_test(&fp->fi_delegees)) {
263 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
264 fp->fi_lease = NULL;
265 fput(fp->fi_deleg_file);
266 fp->fi_deleg_file = NULL;
267 }
268}
269
270/* Called under the state lock. */
271static void
272unhash_delegation(struct nfs4_delegation *dp)
273{
274 list_del_init(&dp->dl_perclnt);
275 spin_lock(&recall_lock);
276 list_del_init(&dp->dl_perfile);
277 list_del_init(&dp->dl_recall_lru);
278 spin_unlock(&recall_lock);
279 nfs4_put_deleg_lease(dp->dl_file);
280 nfs4_put_delegation(dp);
281}
282
283/*
284 * SETCLIENTID state
285 */
286
287/* client_lock protects the client lru list and session hash table */
288static DEFINE_SPINLOCK(client_lock);
289
290/* Hash tables for nfs4_clientid state */
291#define CLIENT_HASH_BITS 4
292#define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
293#define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
294
295#define clientid_hashval(id) \
296 ((id) & CLIENT_HASH_MASK)
297#define clientstr_hashval(name) \
298 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
299/*
300 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
301 * used in reboot/reset lease grace period processing
302 *
303 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
304 * setclientid_confirmed info.
305 *
306 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
307 * setclientid info.
308 *
309 * client_lru holds client queue ordered by nfs4_client.cl_time
310 * for lease renewal.
311 *
312 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
313 * for last close replay.
314 */
315static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
316static int reclaim_str_hashtbl_size = 0;
317static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
318static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
319static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
320static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
321static struct list_head client_lru;
322static struct list_head close_lru;
323
324/*
325 * We store the NONE, READ, WRITE, and BOTH bits separately in the
326 * st_{access,deny}_bmap field of the stateid, in order to track not
327 * only what share bits are currently in force, but also what
328 * combinations of share bits previous opens have used. This allows us
329 * to enforce the recommendation of rfc 3530 14.2.19 that the server
330 * return an error if the client attempt to downgrade to a combination
331 * of share bits not explicable by closing some of its previous opens.
332 *
333 * XXX: This enforcement is actually incomplete, since we don't keep
334 * track of access/deny bit combinations; so, e.g., we allow:
335 *
336 * OPEN allow read, deny write
337 * OPEN allow both, deny none
338 * DOWNGRADE allow read, deny none
339 *
340 * which we should reject.
341 */
342static void
343set_access(unsigned int *access, unsigned long bmap) {
344 int i;
345
346 *access = 0;
347 for (i = 1; i < 4; i++) {
348 if (test_bit(i, &bmap))
349 *access |= i;
350 }
351}
352
353static void
354set_deny(unsigned int *deny, unsigned long bmap) {
355 int i;
356
357 *deny = 0;
358 for (i = 0; i < 4; i++) {
359 if (test_bit(i, &bmap))
360 *deny |= i ;
361 }
362}
363
364static int
365test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
366 unsigned int access, deny;
367
368 set_access(&access, stp->st_access_bmap);
369 set_deny(&deny, stp->st_deny_bmap);
370 if ((access & open->op_share_deny) || (deny & open->op_share_access))
371 return 0;
372 return 1;
373}
374
375static int nfs4_access_to_omode(u32 access)
376{
377 switch (access & NFS4_SHARE_ACCESS_BOTH) {
378 case NFS4_SHARE_ACCESS_READ:
379 return O_RDONLY;
380 case NFS4_SHARE_ACCESS_WRITE:
381 return O_WRONLY;
382 case NFS4_SHARE_ACCESS_BOTH:
383 return O_RDWR;
384 }
385 BUG();
386}
387
388static void unhash_generic_stateid(struct nfs4_stateid *stp)
389{
390 list_del(&stp->st_hash);
391 list_del(&stp->st_perfile);
392 list_del(&stp->st_perstateowner);
393}
394
395static void free_generic_stateid(struct nfs4_stateid *stp)
396{
397 int i;
398
399 if (stp->st_access_bmap) {
400 for (i = 1; i < 4; i++) {
401 if (test_bit(i, &stp->st_access_bmap))
402 nfs4_file_put_access(stp->st_file,
403 nfs4_access_to_omode(i));
404 }
405 }
406 put_nfs4_file(stp->st_file);
407 kmem_cache_free(stateid_slab, stp);
408}
409
410static void release_lock_stateid(struct nfs4_stateid *stp)
411{
412 struct file *file;
413
414 unhash_generic_stateid(stp);
415 file = find_any_file(stp->st_file);
416 if (file)
417 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
418 free_generic_stateid(stp);
419}
420
421static void unhash_lockowner(struct nfs4_stateowner *sop)
422{
423 struct nfs4_stateid *stp;
424
425 list_del(&sop->so_idhash);
426 list_del(&sop->so_strhash);
427 list_del(&sop->so_perstateid);
428 while (!list_empty(&sop->so_stateids)) {
429 stp = list_first_entry(&sop->so_stateids,
430 struct nfs4_stateid, st_perstateowner);
431 release_lock_stateid(stp);
432 }
433}
434
435static void release_lockowner(struct nfs4_stateowner *sop)
436{
437 unhash_lockowner(sop);
438 nfs4_put_stateowner(sop);
439}
440
441static void
442release_stateid_lockowners(struct nfs4_stateid *open_stp)
443{
444 struct nfs4_stateowner *lock_sop;
445
446 while (!list_empty(&open_stp->st_lockowners)) {
447 lock_sop = list_entry(open_stp->st_lockowners.next,
448 struct nfs4_stateowner, so_perstateid);
449 /* list_del(&open_stp->st_lockowners); */
450 BUG_ON(lock_sop->so_is_open_owner);
451 release_lockowner(lock_sop);
452 }
453}
454
455static void release_open_stateid(struct nfs4_stateid *stp)
456{
457 unhash_generic_stateid(stp);
458 release_stateid_lockowners(stp);
459 free_generic_stateid(stp);
460}
461
462static void unhash_openowner(struct nfs4_stateowner *sop)
463{
464 struct nfs4_stateid *stp;
465
466 list_del(&sop->so_idhash);
467 list_del(&sop->so_strhash);
468 list_del(&sop->so_perclient);
469 list_del(&sop->so_perstateid); /* XXX: necessary? */
470 while (!list_empty(&sop->so_stateids)) {
471 stp = list_first_entry(&sop->so_stateids,
472 struct nfs4_stateid, st_perstateowner);
473 release_open_stateid(stp);
474 }
475}
476
477static void release_openowner(struct nfs4_stateowner *sop)
478{
479 unhash_openowner(sop);
480 list_del(&sop->so_close_lru);
481 nfs4_put_stateowner(sop);
482}
483
484#define SESSION_HASH_SIZE 512
485static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
486
487static inline int
488hash_sessionid(struct nfs4_sessionid *sessionid)
489{
490 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
491
492 return sid->sequence % SESSION_HASH_SIZE;
493}
494
495static inline void
496dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
497{
498 u32 *ptr = (u32 *)(&sessionid->data[0]);
499 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
500}
501
502static void
503gen_sessionid(struct nfsd4_session *ses)
504{
505 struct nfs4_client *clp = ses->se_client;
506 struct nfsd4_sessionid *sid;
507
508 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
509 sid->clientid = clp->cl_clientid;
510 sid->sequence = current_sessionid++;
511 sid->reserved = 0;
512}
513
514/*
515 * The protocol defines ca_maxresponssize_cached to include the size of
516 * the rpc header, but all we need to cache is the data starting after
517 * the end of the initial SEQUENCE operation--the rest we regenerate
518 * each time. Therefore we can advertise a ca_maxresponssize_cached
519 * value that is the number of bytes in our cache plus a few additional
520 * bytes. In order to stay on the safe side, and not promise more than
521 * we can cache, those additional bytes must be the minimum possible: 24
522 * bytes of rpc header (xid through accept state, with AUTH_NULL
523 * verifier), 12 for the compound header (with zero-length tag), and 44
524 * for the SEQUENCE op response:
525 */
526#define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
527
528static void
529free_session_slots(struct nfsd4_session *ses)
530{
531 int i;
532
533 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
534 kfree(ses->se_slots[i]);
535}
536
537/*
538 * We don't actually need to cache the rpc and session headers, so we
539 * can allocate a little less for each slot:
540 */
541static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
542{
543 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
544}
545
546static int nfsd4_sanitize_slot_size(u32 size)
547{
548 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
549 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
550
551 return size;
552}
553
554/*
555 * XXX: If we run out of reserved DRC memory we could (up to a point)
556 * re-negotiate active sessions and reduce their slot usage to make
557 * rooom for new connections. For now we just fail the create session.
558 */
559static int nfsd4_get_drc_mem(int slotsize, u32 num)
560{
561 int avail;
562
563 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
564
565 spin_lock(&nfsd_drc_lock);
566 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
567 nfsd_drc_max_mem - nfsd_drc_mem_used);
568 num = min_t(int, num, avail / slotsize);
569 nfsd_drc_mem_used += num * slotsize;
570 spin_unlock(&nfsd_drc_lock);
571
572 return num;
573}
574
575static void nfsd4_put_drc_mem(int slotsize, int num)
576{
577 spin_lock(&nfsd_drc_lock);
578 nfsd_drc_mem_used -= slotsize * num;
579 spin_unlock(&nfsd_drc_lock);
580}
581
582static struct nfsd4_session *alloc_session(int slotsize, int numslots)
583{
584 struct nfsd4_session *new;
585 int mem, i;
586
587 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
588 + sizeof(struct nfsd4_session) > PAGE_SIZE);
589 mem = numslots * sizeof(struct nfsd4_slot *);
590
591 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
592 if (!new)
593 return NULL;
594 /* allocate each struct nfsd4_slot and data cache in one piece */
595 for (i = 0; i < numslots; i++) {
596 mem = sizeof(struct nfsd4_slot) + slotsize;
597 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
598 if (!new->se_slots[i])
599 goto out_free;
600 }
601 return new;
602out_free:
603 while (i--)
604 kfree(new->se_slots[i]);
605 kfree(new);
606 return NULL;
607}
608
609static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
610{
611 u32 maxrpc = nfsd_serv->sv_max_mesg;
612
613 new->maxreqs = numslots;
614 new->maxresp_cached = min_t(u32, req->maxresp_cached,
615 slotsize + NFSD_MIN_HDR_SEQ_SZ);
616 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
617 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
618 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
619}
620
621static void free_conn(struct nfsd4_conn *c)
622{
623 svc_xprt_put(c->cn_xprt);
624 kfree(c);
625}
626
627static void nfsd4_conn_lost(struct svc_xpt_user *u)
628{
629 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
630 struct nfs4_client *clp = c->cn_session->se_client;
631
632 spin_lock(&clp->cl_lock);
633 if (!list_empty(&c->cn_persession)) {
634 list_del(&c->cn_persession);
635 free_conn(c);
636 }
637 spin_unlock(&clp->cl_lock);
638 nfsd4_probe_callback(clp);
639}
640
641static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
642{
643 struct nfsd4_conn *conn;
644
645 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
646 if (!conn)
647 return NULL;
648 svc_xprt_get(rqstp->rq_xprt);
649 conn->cn_xprt = rqstp->rq_xprt;
650 conn->cn_flags = flags;
651 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
652 return conn;
653}
654
655static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
656{
657 conn->cn_session = ses;
658 list_add(&conn->cn_persession, &ses->se_conns);
659}
660
661static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
662{
663 struct nfs4_client *clp = ses->se_client;
664
665 spin_lock(&clp->cl_lock);
666 __nfsd4_hash_conn(conn, ses);
667 spin_unlock(&clp->cl_lock);
668}
669
670static int nfsd4_register_conn(struct nfsd4_conn *conn)
671{
672 conn->cn_xpt_user.callback = nfsd4_conn_lost;
673 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
674}
675
676static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
677{
678 struct nfsd4_conn *conn;
679 int ret;
680
681 conn = alloc_conn(rqstp, dir);
682 if (!conn)
683 return nfserr_jukebox;
684 nfsd4_hash_conn(conn, ses);
685 ret = nfsd4_register_conn(conn);
686 if (ret)
687 /* oops; xprt is already down: */
688 nfsd4_conn_lost(&conn->cn_xpt_user);
689 return nfs_ok;
690}
691
692static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
693{
694 u32 dir = NFS4_CDFC4_FORE;
695
696 if (ses->se_flags & SESSION4_BACK_CHAN)
697 dir |= NFS4_CDFC4_BACK;
698
699 return nfsd4_new_conn(rqstp, ses, dir);
700}
701
702/* must be called under client_lock */
703static void nfsd4_del_conns(struct nfsd4_session *s)
704{
705 struct nfs4_client *clp = s->se_client;
706 struct nfsd4_conn *c;
707
708 spin_lock(&clp->cl_lock);
709 while (!list_empty(&s->se_conns)) {
710 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
711 list_del_init(&c->cn_persession);
712 spin_unlock(&clp->cl_lock);
713
714 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
715 free_conn(c);
716
717 spin_lock(&clp->cl_lock);
718 }
719 spin_unlock(&clp->cl_lock);
720}
721
722void free_session(struct kref *kref)
723{
724 struct nfsd4_session *ses;
725 int mem;
726
727 ses = container_of(kref, struct nfsd4_session, se_ref);
728 nfsd4_del_conns(ses);
729 spin_lock(&nfsd_drc_lock);
730 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
731 nfsd_drc_mem_used -= mem;
732 spin_unlock(&nfsd_drc_lock);
733 free_session_slots(ses);
734 kfree(ses);
735}
736
737static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
738{
739 struct nfsd4_session *new;
740 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
741 int numslots, slotsize;
742 int status;
743 int idx;
744
745 /*
746 * Note decreasing slot size below client's request may
747 * make it difficult for client to function correctly, whereas
748 * decreasing the number of slots will (just?) affect
749 * performance. When short on memory we therefore prefer to
750 * decrease number of slots instead of their size.
751 */
752 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
753 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
754 if (numslots < 1)
755 return NULL;
756
757 new = alloc_session(slotsize, numslots);
758 if (!new) {
759 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
760 return NULL;
761 }
762 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
763
764 new->se_client = clp;
765 gen_sessionid(new);
766
767 INIT_LIST_HEAD(&new->se_conns);
768
769 new->se_cb_seq_nr = 1;
770 new->se_flags = cses->flags;
771 new->se_cb_prog = cses->callback_prog;
772 kref_init(&new->se_ref);
773 idx = hash_sessionid(&new->se_sessionid);
774 spin_lock(&client_lock);
775 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
776 spin_lock(&clp->cl_lock);
777 list_add(&new->se_perclnt, &clp->cl_sessions);
778 spin_unlock(&clp->cl_lock);
779 spin_unlock(&client_lock);
780
781 status = nfsd4_new_conn_from_crses(rqstp, new);
782 /* whoops: benny points out, status is ignored! (err, or bogus) */
783 if (status) {
784 free_session(&new->se_ref);
785 return NULL;
786 }
787 if (cses->flags & SESSION4_BACK_CHAN) {
788 struct sockaddr *sa = svc_addr(rqstp);
789 /*
790 * This is a little silly; with sessions there's no real
791 * use for the callback address. Use the peer address
792 * as a reasonable default for now, but consider fixing
793 * the rpc client not to require an address in the
794 * future:
795 */
796 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
797 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
798 }
799 nfsd4_probe_callback(clp);
800 return new;
801}
802
803/* caller must hold client_lock */
804static struct nfsd4_session *
805find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
806{
807 struct nfsd4_session *elem;
808 int idx;
809
810 dump_sessionid(__func__, sessionid);
811 idx = hash_sessionid(sessionid);
812 /* Search in the appropriate list */
813 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
814 if (!memcmp(elem->se_sessionid.data, sessionid->data,
815 NFS4_MAX_SESSIONID_LEN)) {
816 return elem;
817 }
818 }
819
820 dprintk("%s: session not found\n", __func__);
821 return NULL;
822}
823
824/* caller must hold client_lock */
825static void
826unhash_session(struct nfsd4_session *ses)
827{
828 list_del(&ses->se_hash);
829 spin_lock(&ses->se_client->cl_lock);
830 list_del(&ses->se_perclnt);
831 spin_unlock(&ses->se_client->cl_lock);
832}
833
834/* must be called under the client_lock */
835static inline void
836renew_client_locked(struct nfs4_client *clp)
837{
838 if (is_client_expired(clp)) {
839 dprintk("%s: client (clientid %08x/%08x) already expired\n",
840 __func__,
841 clp->cl_clientid.cl_boot,
842 clp->cl_clientid.cl_id);
843 return;
844 }
845
846 /*
847 * Move client to the end to the LRU list.
848 */
849 dprintk("renewing client (clientid %08x/%08x)\n",
850 clp->cl_clientid.cl_boot,
851 clp->cl_clientid.cl_id);
852 list_move_tail(&clp->cl_lru, &client_lru);
853 clp->cl_time = get_seconds();
854}
855
856static inline void
857renew_client(struct nfs4_client *clp)
858{
859 spin_lock(&client_lock);
860 renew_client_locked(clp);
861 spin_unlock(&client_lock);
862}
863
864/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
865static int
866STALE_CLIENTID(clientid_t *clid)
867{
868 if (clid->cl_boot == boot_time)
869 return 0;
870 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
871 clid->cl_boot, clid->cl_id, boot_time);
872 return 1;
873}
874
875/*
876 * XXX Should we use a slab cache ?
877 * This type of memory management is somewhat inefficient, but we use it
878 * anyway since SETCLIENTID is not a common operation.
879 */
880static struct nfs4_client *alloc_client(struct xdr_netobj name)
881{
882 struct nfs4_client *clp;
883
884 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
885 if (clp == NULL)
886 return NULL;
887 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
888 if (clp->cl_name.data == NULL) {
889 kfree(clp);
890 return NULL;
891 }
892 memcpy(clp->cl_name.data, name.data, name.len);
893 clp->cl_name.len = name.len;
894 return clp;
895}
896
897static inline void
898free_client(struct nfs4_client *clp)
899{
900 while (!list_empty(&clp->cl_sessions)) {
901 struct nfsd4_session *ses;
902 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
903 se_perclnt);
904 list_del(&ses->se_perclnt);
905 nfsd4_put_session(ses);
906 }
907 if (clp->cl_cred.cr_group_info)
908 put_group_info(clp->cl_cred.cr_group_info);
909 kfree(clp->cl_principal);
910 kfree(clp->cl_name.data);
911 kfree(clp);
912}
913
914void
915release_session_client(struct nfsd4_session *session)
916{
917 struct nfs4_client *clp = session->se_client;
918
919 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
920 return;
921 if (is_client_expired(clp)) {
922 free_client(clp);
923 session->se_client = NULL;
924 } else
925 renew_client_locked(clp);
926 spin_unlock(&client_lock);
927}
928
929/* must be called under the client_lock */
930static inline void
931unhash_client_locked(struct nfs4_client *clp)
932{
933 struct nfsd4_session *ses;
934
935 mark_client_expired(clp);
936 list_del(&clp->cl_lru);
937 spin_lock(&clp->cl_lock);
938 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
939 list_del_init(&ses->se_hash);
940 spin_unlock(&clp->cl_lock);
941}
942
943static void
944expire_client(struct nfs4_client *clp)
945{
946 struct nfs4_stateowner *sop;
947 struct nfs4_delegation *dp;
948 struct list_head reaplist;
949
950 INIT_LIST_HEAD(&reaplist);
951 spin_lock(&recall_lock);
952 while (!list_empty(&clp->cl_delegations)) {
953 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
954 list_del_init(&dp->dl_perclnt);
955 list_move(&dp->dl_recall_lru, &reaplist);
956 }
957 spin_unlock(&recall_lock);
958 while (!list_empty(&reaplist)) {
959 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
960 list_del_init(&dp->dl_recall_lru);
961 unhash_delegation(dp);
962 }
963 while (!list_empty(&clp->cl_openowners)) {
964 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
965 release_openowner(sop);
966 }
967 nfsd4_shutdown_callback(clp);
968 if (clp->cl_cb_conn.cb_xprt)
969 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
970 list_del(&clp->cl_idhash);
971 list_del(&clp->cl_strhash);
972 spin_lock(&client_lock);
973 unhash_client_locked(clp);
974 if (atomic_read(&clp->cl_refcount) == 0)
975 free_client(clp);
976 spin_unlock(&client_lock);
977}
978
979static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
980{
981 memcpy(target->cl_verifier.data, source->data,
982 sizeof(target->cl_verifier.data));
983}
984
985static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
986{
987 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
988 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
989}
990
991static void copy_cred(struct svc_cred *target, struct svc_cred *source)
992{
993 target->cr_uid = source->cr_uid;
994 target->cr_gid = source->cr_gid;
995 target->cr_group_info = source->cr_group_info;
996 get_group_info(target->cr_group_info);
997}
998
999static int same_name(const char *n1, const char *n2)
1000{
1001 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1002}
1003
1004static int
1005same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1006{
1007 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1008}
1009
1010static int
1011same_clid(clientid_t *cl1, clientid_t *cl2)
1012{
1013 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1014}
1015
1016/* XXX what about NGROUP */
1017static int
1018same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1019{
1020 return cr1->cr_uid == cr2->cr_uid;
1021}
1022
1023static void gen_clid(struct nfs4_client *clp)
1024{
1025 static u32 current_clientid = 1;
1026
1027 clp->cl_clientid.cl_boot = boot_time;
1028 clp->cl_clientid.cl_id = current_clientid++;
1029}
1030
1031static void gen_confirm(struct nfs4_client *clp)
1032{
1033 static u32 i;
1034 u32 *p;
1035
1036 p = (u32 *)clp->cl_confirm.data;
1037 *p++ = get_seconds();
1038 *p++ = i++;
1039}
1040
1041static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1042 struct svc_rqst *rqstp, nfs4_verifier *verf)
1043{
1044 struct nfs4_client *clp;
1045 struct sockaddr *sa = svc_addr(rqstp);
1046 char *princ;
1047
1048 clp = alloc_client(name);
1049 if (clp == NULL)
1050 return NULL;
1051
1052 INIT_LIST_HEAD(&clp->cl_sessions);
1053
1054 princ = svc_gss_principal(rqstp);
1055 if (princ) {
1056 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1057 if (clp->cl_principal == NULL) {
1058 free_client(clp);
1059 return NULL;
1060 }
1061 }
1062
1063 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1064 atomic_set(&clp->cl_refcount, 0);
1065 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1066 INIT_LIST_HEAD(&clp->cl_idhash);
1067 INIT_LIST_HEAD(&clp->cl_strhash);
1068 INIT_LIST_HEAD(&clp->cl_openowners);
1069 INIT_LIST_HEAD(&clp->cl_delegations);
1070 INIT_LIST_HEAD(&clp->cl_lru);
1071 INIT_LIST_HEAD(&clp->cl_callbacks);
1072 spin_lock_init(&clp->cl_lock);
1073 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1074 clp->cl_time = get_seconds();
1075 clear_bit(0, &clp->cl_cb_slot_busy);
1076 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1077 copy_verf(clp, verf);
1078 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1079 clp->cl_flavor = rqstp->rq_flavor;
1080 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1081 gen_confirm(clp);
1082 clp->cl_cb_session = NULL;
1083 return clp;
1084}
1085
1086static int check_name(struct xdr_netobj name)
1087{
1088 if (name.len == 0)
1089 return 0;
1090 if (name.len > NFS4_OPAQUE_LIMIT) {
1091 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1092 return 0;
1093 }
1094 return 1;
1095}
1096
1097static void
1098add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1099{
1100 unsigned int idhashval;
1101
1102 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1103 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1104 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1105 renew_client(clp);
1106}
1107
1108static void
1109move_to_confirmed(struct nfs4_client *clp)
1110{
1111 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1112 unsigned int strhashval;
1113
1114 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1115 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1116 strhashval = clientstr_hashval(clp->cl_recdir);
1117 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1118 renew_client(clp);
1119}
1120
1121static struct nfs4_client *
1122find_confirmed_client(clientid_t *clid)
1123{
1124 struct nfs4_client *clp;
1125 unsigned int idhashval = clientid_hashval(clid->cl_id);
1126
1127 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1128 if (same_clid(&clp->cl_clientid, clid))
1129 return clp;
1130 }
1131 return NULL;
1132}
1133
1134static struct nfs4_client *
1135find_unconfirmed_client(clientid_t *clid)
1136{
1137 struct nfs4_client *clp;
1138 unsigned int idhashval = clientid_hashval(clid->cl_id);
1139
1140 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1141 if (same_clid(&clp->cl_clientid, clid))
1142 return clp;
1143 }
1144 return NULL;
1145}
1146
1147static bool clp_used_exchangeid(struct nfs4_client *clp)
1148{
1149 return clp->cl_exchange_flags != 0;
1150}
1151
1152static struct nfs4_client *
1153find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1154{
1155 struct nfs4_client *clp;
1156
1157 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1158 if (same_name(clp->cl_recdir, dname))
1159 return clp;
1160 }
1161 return NULL;
1162}
1163
1164static struct nfs4_client *
1165find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1166{
1167 struct nfs4_client *clp;
1168
1169 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1170 if (same_name(clp->cl_recdir, dname))
1171 return clp;
1172 }
1173 return NULL;
1174}
1175
1176static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1177{
1178 switch (family) {
1179 case AF_INET:
1180 ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1181 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1182 return;
1183 case AF_INET6:
1184 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1185 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1186 return;
1187 }
1188}
1189
1190static void
1191gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1192{
1193 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1194 struct sockaddr *sa = svc_addr(rqstp);
1195 u32 scopeid = rpc_get_scope_id(sa);
1196 unsigned short expected_family;
1197
1198 /* Currently, we only support tcp and tcp6 for the callback channel */
1199 if (se->se_callback_netid_len == 3 &&
1200 !memcmp(se->se_callback_netid_val, "tcp", 3))
1201 expected_family = AF_INET;
1202 else if (se->se_callback_netid_len == 4 &&
1203 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1204 expected_family = AF_INET6;
1205 else
1206 goto out_err;
1207
1208 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1209 se->se_callback_addr_len,
1210 (struct sockaddr *)&conn->cb_addr,
1211 sizeof(conn->cb_addr));
1212
1213 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1214 goto out_err;
1215
1216 if (conn->cb_addr.ss_family == AF_INET6)
1217 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1218
1219 conn->cb_prog = se->se_callback_prog;
1220 conn->cb_ident = se->se_callback_ident;
1221 rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1222 return;
1223out_err:
1224 conn->cb_addr.ss_family = AF_UNSPEC;
1225 conn->cb_addrlen = 0;
1226 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1227 "will not receive delegations\n",
1228 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1229
1230 return;
1231}
1232
1233/*
1234 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1235 */
1236void
1237nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1238{
1239 struct nfsd4_slot *slot = resp->cstate.slot;
1240 unsigned int base;
1241
1242 dprintk("--> %s slot %p\n", __func__, slot);
1243
1244 slot->sl_opcnt = resp->opcnt;
1245 slot->sl_status = resp->cstate.status;
1246
1247 if (nfsd4_not_cached(resp)) {
1248 slot->sl_datalen = 0;
1249 return;
1250 }
1251 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1252 base = (char *)resp->cstate.datap -
1253 (char *)resp->xbuf->head[0].iov_base;
1254 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1255 slot->sl_datalen))
1256 WARN("%s: sessions DRC could not cache compound\n", __func__);
1257 return;
1258}
1259
1260/*
1261 * Encode the replay sequence operation from the slot values.
1262 * If cachethis is FALSE encode the uncached rep error on the next
1263 * operation which sets resp->p and increments resp->opcnt for
1264 * nfs4svc_encode_compoundres.
1265 *
1266 */
1267static __be32
1268nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1269 struct nfsd4_compoundres *resp)
1270{
1271 struct nfsd4_op *op;
1272 struct nfsd4_slot *slot = resp->cstate.slot;
1273
1274 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1275 resp->opcnt, resp->cstate.slot->sl_cachethis);
1276
1277 /* Encode the replayed sequence operation */
1278 op = &args->ops[resp->opcnt - 1];
1279 nfsd4_encode_operation(resp, op);
1280
1281 /* Return nfserr_retry_uncached_rep in next operation. */
1282 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1283 op = &args->ops[resp->opcnt++];
1284 op->status = nfserr_retry_uncached_rep;
1285 nfsd4_encode_operation(resp, op);
1286 }
1287 return op->status;
1288}
1289
1290/*
1291 * The sequence operation is not cached because we can use the slot and
1292 * session values.
1293 */
1294__be32
1295nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1296 struct nfsd4_sequence *seq)
1297{
1298 struct nfsd4_slot *slot = resp->cstate.slot;
1299 __be32 status;
1300
1301 dprintk("--> %s slot %p\n", __func__, slot);
1302
1303 /* Either returns 0 or nfserr_retry_uncached */
1304 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1305 if (status == nfserr_retry_uncached_rep)
1306 return status;
1307
1308 /* The sequence operation has been encoded, cstate->datap set. */
1309 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1310
1311 resp->opcnt = slot->sl_opcnt;
1312 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1313 status = slot->sl_status;
1314
1315 return status;
1316}
1317
1318/*
1319 * Set the exchange_id flags returned by the server.
1320 */
1321static void
1322nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1323{
1324 /* pNFS is not supported */
1325 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1326
1327 /* Referrals are supported, Migration is not. */
1328 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1329
1330 /* set the wire flags to return to client. */
1331 clid->flags = new->cl_exchange_flags;
1332}
1333
1334__be32
1335nfsd4_exchange_id(struct svc_rqst *rqstp,
1336 struct nfsd4_compound_state *cstate,
1337 struct nfsd4_exchange_id *exid)
1338{
1339 struct nfs4_client *unconf, *conf, *new;
1340 int status;
1341 unsigned int strhashval;
1342 char dname[HEXDIR_LEN];
1343 char addr_str[INET6_ADDRSTRLEN];
1344 nfs4_verifier verf = exid->verifier;
1345 struct sockaddr *sa = svc_addr(rqstp);
1346
1347 rpc_ntop(sa, addr_str, sizeof(addr_str));
1348 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1349 "ip_addr=%s flags %x, spa_how %d\n",
1350 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1351 addr_str, exid->flags, exid->spa_how);
1352
1353 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1354 return nfserr_inval;
1355
1356 /* Currently only support SP4_NONE */
1357 switch (exid->spa_how) {
1358 case SP4_NONE:
1359 break;
1360 case SP4_SSV:
1361 return nfserr_serverfault;
1362 default:
1363 BUG(); /* checked by xdr code */
1364 case SP4_MACH_CRED:
1365 return nfserr_serverfault; /* no excuse :-/ */
1366 }
1367
1368 status = nfs4_make_rec_clidname(dname, &exid->clname);
1369
1370 if (status)
1371 goto error;
1372
1373 strhashval = clientstr_hashval(dname);
1374
1375 nfs4_lock_state();
1376 status = nfs_ok;
1377
1378 conf = find_confirmed_client_by_str(dname, strhashval);
1379 if (conf) {
1380 if (!clp_used_exchangeid(conf)) {
1381 status = nfserr_clid_inuse; /* XXX: ? */
1382 goto out;
1383 }
1384 if (!same_verf(&verf, &conf->cl_verifier)) {
1385 /* 18.35.4 case 8 */
1386 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1387 status = nfserr_not_same;
1388 goto out;
1389 }
1390 /* Client reboot: destroy old state */
1391 expire_client(conf);
1392 goto out_new;
1393 }
1394 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1395 /* 18.35.4 case 9 */
1396 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1397 status = nfserr_perm;
1398 goto out;
1399 }
1400 expire_client(conf);
1401 goto out_new;
1402 }
1403 /*
1404 * Set bit when the owner id and verifier map to an already
1405 * confirmed client id (18.35.3).
1406 */
1407 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1408
1409 /*
1410 * Falling into 18.35.4 case 2, possible router replay.
1411 * Leave confirmed record intact and return same result.
1412 */
1413 copy_verf(conf, &verf);
1414 new = conf;
1415 goto out_copy;
1416 }
1417
1418 /* 18.35.4 case 7 */
1419 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1420 status = nfserr_noent;
1421 goto out;
1422 }
1423
1424 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1425 if (unconf) {
1426 /*
1427 * Possible retry or client restart. Per 18.35.4 case 4,
1428 * a new unconfirmed record should be generated regardless
1429 * of whether any properties have changed.
1430 */
1431 expire_client(unconf);
1432 }
1433
1434out_new:
1435 /* Normal case */
1436 new = create_client(exid->clname, dname, rqstp, &verf);
1437 if (new == NULL) {
1438 status = nfserr_jukebox;
1439 goto out;
1440 }
1441
1442 gen_clid(new);
1443 add_to_unconfirmed(new, strhashval);
1444out_copy:
1445 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1446 exid->clientid.cl_id = new->cl_clientid.cl_id;
1447
1448 exid->seqid = 1;
1449 nfsd4_set_ex_flags(new, exid);
1450
1451 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1452 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1453 status = nfs_ok;
1454
1455out:
1456 nfs4_unlock_state();
1457error:
1458 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1459 return status;
1460}
1461
1462static int
1463check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1464{
1465 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1466 slot_seqid);
1467
1468 /* The slot is in use, and no response has been sent. */
1469 if (slot_inuse) {
1470 if (seqid == slot_seqid)
1471 return nfserr_jukebox;
1472 else
1473 return nfserr_seq_misordered;
1474 }
1475 /* Normal */
1476 if (likely(seqid == slot_seqid + 1))
1477 return nfs_ok;
1478 /* Replay */
1479 if (seqid == slot_seqid)
1480 return nfserr_replay_cache;
1481 /* Wraparound */
1482 if (seqid == 1 && (slot_seqid + 1) == 0)
1483 return nfs_ok;
1484 /* Misordered replay or misordered new request */
1485 return nfserr_seq_misordered;
1486}
1487
1488/*
1489 * Cache the create session result into the create session single DRC
1490 * slot cache by saving the xdr structure. sl_seqid has been set.
1491 * Do this for solo or embedded create session operations.
1492 */
1493static void
1494nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1495 struct nfsd4_clid_slot *slot, int nfserr)
1496{
1497 slot->sl_status = nfserr;
1498 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1499}
1500
1501static __be32
1502nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1503 struct nfsd4_clid_slot *slot)
1504{
1505 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1506 return slot->sl_status;
1507}
1508
1509#define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1510 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1511 1 + /* MIN tag is length with zero, only length */ \
1512 3 + /* version, opcount, opcode */ \
1513 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1514 /* seqid, slotID, slotID, cache */ \
1515 4 ) * sizeof(__be32))
1516
1517#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1518 2 + /* verifier: AUTH_NULL, length 0 */\
1519 1 + /* status */ \
1520 1 + /* MIN tag is length with zero, only length */ \
1521 3 + /* opcount, opcode, opstatus*/ \
1522 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1523 /* seqid, slotID, slotID, slotID, status */ \
1524 5 ) * sizeof(__be32))
1525
1526static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1527{
1528 return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1529 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1530}
1531
1532__be32
1533nfsd4_create_session(struct svc_rqst *rqstp,
1534 struct nfsd4_compound_state *cstate,
1535 struct nfsd4_create_session *cr_ses)
1536{
1537 struct sockaddr *sa = svc_addr(rqstp);
1538 struct nfs4_client *conf, *unconf;
1539 struct nfsd4_session *new;
1540 struct nfsd4_clid_slot *cs_slot = NULL;
1541 bool confirm_me = false;
1542 int status = 0;
1543
1544 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1545 return nfserr_inval;
1546
1547 nfs4_lock_state();
1548 unconf = find_unconfirmed_client(&cr_ses->clientid);
1549 conf = find_confirmed_client(&cr_ses->clientid);
1550
1551 if (conf) {
1552 cs_slot = &conf->cl_cs_slot;
1553 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1554 if (status == nfserr_replay_cache) {
1555 dprintk("Got a create_session replay! seqid= %d\n",
1556 cs_slot->sl_seqid);
1557 /* Return the cached reply status */
1558 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1559 goto out;
1560 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1561 status = nfserr_seq_misordered;
1562 dprintk("Sequence misordered!\n");
1563 dprintk("Expected seqid= %d but got seqid= %d\n",
1564 cs_slot->sl_seqid, cr_ses->seqid);
1565 goto out;
1566 }
1567 } else if (unconf) {
1568 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1569 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1570 status = nfserr_clid_inuse;
1571 goto out;
1572 }
1573
1574 cs_slot = &unconf->cl_cs_slot;
1575 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1576 if (status) {
1577 /* an unconfirmed replay returns misordered */
1578 status = nfserr_seq_misordered;
1579 goto out;
1580 }
1581
1582 confirm_me = true;
1583 conf = unconf;
1584 } else {
1585 status = nfserr_stale_clientid;
1586 goto out;
1587 }
1588
1589 /*
1590 * XXX: we should probably set this at creation time, and check
1591 * for consistent minorversion use throughout:
1592 */
1593 conf->cl_minorversion = 1;
1594 /*
1595 * We do not support RDMA or persistent sessions
1596 */
1597 cr_ses->flags &= ~SESSION4_PERSIST;
1598 cr_ses->flags &= ~SESSION4_RDMA;
1599
1600 status = nfserr_toosmall;
1601 if (check_forechannel_attrs(cr_ses->fore_channel))
1602 goto out;
1603
1604 status = nfserr_jukebox;
1605 new = alloc_init_session(rqstp, conf, cr_ses);
1606 if (!new)
1607 goto out;
1608 status = nfs_ok;
1609 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1610 NFS4_MAX_SESSIONID_LEN);
1611 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1612 sizeof(struct nfsd4_channel_attrs));
1613 cs_slot->sl_seqid++;
1614 cr_ses->seqid = cs_slot->sl_seqid;
1615
1616 /* cache solo and embedded create sessions under the state lock */
1617 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1618 if (confirm_me)
1619 move_to_confirmed(conf);
1620out:
1621 nfs4_unlock_state();
1622 dprintk("%s returns %d\n", __func__, ntohl(status));
1623 return status;
1624}
1625
1626static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1627{
1628 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1629 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1630
1631 return argp->opcnt == resp->opcnt;
1632}
1633
1634static __be32 nfsd4_map_bcts_dir(u32 *dir)
1635{
1636 switch (*dir) {
1637 case NFS4_CDFC4_FORE:
1638 case NFS4_CDFC4_BACK:
1639 return nfs_ok;
1640 case NFS4_CDFC4_FORE_OR_BOTH:
1641 case NFS4_CDFC4_BACK_OR_BOTH:
1642 *dir = NFS4_CDFC4_BOTH;
1643 return nfs_ok;
1644 };
1645 return nfserr_inval;
1646}
1647
1648__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1649 struct nfsd4_compound_state *cstate,
1650 struct nfsd4_bind_conn_to_session *bcts)
1651{
1652 __be32 status;
1653
1654 if (!nfsd4_last_compound_op(rqstp))
1655 return nfserr_not_only_op;
1656 spin_lock(&client_lock);
1657 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1658 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1659 * client_lock iself: */
1660 if (cstate->session) {
1661 nfsd4_get_session(cstate->session);
1662 atomic_inc(&cstate->session->se_client->cl_refcount);
1663 }
1664 spin_unlock(&client_lock);
1665 if (!cstate->session)
1666 return nfserr_badsession;
1667
1668 status = nfsd4_map_bcts_dir(&bcts->dir);
1669 if (!status)
1670 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1671 return status;
1672}
1673
1674static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1675{
1676 if (!session)
1677 return 0;
1678 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1679}
1680
1681__be32
1682nfsd4_destroy_session(struct svc_rqst *r,
1683 struct nfsd4_compound_state *cstate,
1684 struct nfsd4_destroy_session *sessionid)
1685{
1686 struct nfsd4_session *ses;
1687 u32 status = nfserr_badsession;
1688
1689 /* Notes:
1690 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1691 * - Should we return nfserr_back_chan_busy if waiting for
1692 * callbacks on to-be-destroyed session?
1693 * - Do we need to clear any callback info from previous session?
1694 */
1695
1696 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1697 if (!nfsd4_last_compound_op(r))
1698 return nfserr_not_only_op;
1699 }
1700 dump_sessionid(__func__, &sessionid->sessionid);
1701 spin_lock(&client_lock);
1702 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1703 if (!ses) {
1704 spin_unlock(&client_lock);
1705 goto out;
1706 }
1707
1708 unhash_session(ses);
1709 spin_unlock(&client_lock);
1710
1711 nfs4_lock_state();
1712 nfsd4_probe_callback_sync(ses->se_client);
1713 nfs4_unlock_state();
1714
1715 nfsd4_del_conns(ses);
1716
1717 nfsd4_put_session(ses);
1718 status = nfs_ok;
1719out:
1720 dprintk("%s returns %d\n", __func__, ntohl(status));
1721 return status;
1722}
1723
1724static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1725{
1726 struct nfsd4_conn *c;
1727
1728 list_for_each_entry(c, &s->se_conns, cn_persession) {
1729 if (c->cn_xprt == xpt) {
1730 return c;
1731 }
1732 }
1733 return NULL;
1734}
1735
1736static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1737{
1738 struct nfs4_client *clp = ses->se_client;
1739 struct nfsd4_conn *c;
1740 int ret;
1741
1742 spin_lock(&clp->cl_lock);
1743 c = __nfsd4_find_conn(new->cn_xprt, ses);
1744 if (c) {
1745 spin_unlock(&clp->cl_lock);
1746 free_conn(new);
1747 return;
1748 }
1749 __nfsd4_hash_conn(new, ses);
1750 spin_unlock(&clp->cl_lock);
1751 ret = nfsd4_register_conn(new);
1752 if (ret)
1753 /* oops; xprt is already down: */
1754 nfsd4_conn_lost(&new->cn_xpt_user);
1755 return;
1756}
1757
1758static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1759{
1760 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1761
1762 return args->opcnt > session->se_fchannel.maxops;
1763}
1764
1765static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1766 struct nfsd4_session *session)
1767{
1768 struct xdr_buf *xb = &rqstp->rq_arg;
1769
1770 return xb->len > session->se_fchannel.maxreq_sz;
1771}
1772
1773__be32
1774nfsd4_sequence(struct svc_rqst *rqstp,
1775 struct nfsd4_compound_state *cstate,
1776 struct nfsd4_sequence *seq)
1777{
1778 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1779 struct nfsd4_session *session;
1780 struct nfsd4_slot *slot;
1781 struct nfsd4_conn *conn;
1782 int status;
1783
1784 if (resp->opcnt != 1)
1785 return nfserr_sequence_pos;
1786
1787 /*
1788 * Will be either used or freed by nfsd4_sequence_check_conn
1789 * below.
1790 */
1791 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1792 if (!conn)
1793 return nfserr_jukebox;
1794
1795 spin_lock(&client_lock);
1796 status = nfserr_badsession;
1797 session = find_in_sessionid_hashtbl(&seq->sessionid);
1798 if (!session)
1799 goto out;
1800
1801 status = nfserr_too_many_ops;
1802 if (nfsd4_session_too_many_ops(rqstp, session))
1803 goto out;
1804
1805 status = nfserr_req_too_big;
1806 if (nfsd4_request_too_big(rqstp, session))
1807 goto out;
1808
1809 status = nfserr_badslot;
1810 if (seq->slotid >= session->se_fchannel.maxreqs)
1811 goto out;
1812
1813 slot = session->se_slots[seq->slotid];
1814 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1815
1816 /* We do not negotiate the number of slots yet, so set the
1817 * maxslots to the session maxreqs which is used to encode
1818 * sr_highest_slotid and the sr_target_slot id to maxslots */
1819 seq->maxslots = session->se_fchannel.maxreqs;
1820
1821 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1822 if (status == nfserr_replay_cache) {
1823 cstate->slot = slot;
1824 cstate->session = session;
1825 /* Return the cached reply status and set cstate->status
1826 * for nfsd4_proc_compound processing */
1827 status = nfsd4_replay_cache_entry(resp, seq);
1828 cstate->status = nfserr_replay_cache;
1829 goto out;
1830 }
1831 if (status)
1832 goto out;
1833
1834 nfsd4_sequence_check_conn(conn, session);
1835 conn = NULL;
1836
1837 /* Success! bump slot seqid */
1838 slot->sl_inuse = true;
1839 slot->sl_seqid = seq->seqid;
1840 slot->sl_cachethis = seq->cachethis;
1841
1842 cstate->slot = slot;
1843 cstate->session = session;
1844
1845out:
1846 /* Hold a session reference until done processing the compound. */
1847 if (cstate->session) {
1848 struct nfs4_client *clp = session->se_client;
1849
1850 nfsd4_get_session(cstate->session);
1851 atomic_inc(&clp->cl_refcount);
1852 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1853 seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1854 }
1855 kfree(conn);
1856 spin_unlock(&client_lock);
1857 dprintk("%s: return %d\n", __func__, ntohl(status));
1858 return status;
1859}
1860
1861__be32
1862nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1863{
1864 int status = 0;
1865
1866 if (rc->rca_one_fs) {
1867 if (!cstate->current_fh.fh_dentry)
1868 return nfserr_nofilehandle;
1869 /*
1870 * We don't take advantage of the rca_one_fs case.
1871 * That's OK, it's optional, we can safely ignore it.
1872 */
1873 return nfs_ok;
1874 }
1875
1876 nfs4_lock_state();
1877 status = nfserr_complete_already;
1878 if (cstate->session->se_client->cl_firststate)
1879 goto out;
1880
1881 status = nfserr_stale_clientid;
1882 if (is_client_expired(cstate->session->se_client))
1883 /*
1884 * The following error isn't really legal.
1885 * But we only get here if the client just explicitly
1886 * destroyed the client. Surely it no longer cares what
1887 * error it gets back on an operation for the dead
1888 * client.
1889 */
1890 goto out;
1891
1892 status = nfs_ok;
1893 nfsd4_create_clid_dir(cstate->session->se_client);
1894out:
1895 nfs4_unlock_state();
1896 return status;
1897}
1898
1899__be32
1900nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1901 struct nfsd4_setclientid *setclid)
1902{
1903 struct xdr_netobj clname = {
1904 .len = setclid->se_namelen,
1905 .data = setclid->se_name,
1906 };
1907 nfs4_verifier clverifier = setclid->se_verf;
1908 unsigned int strhashval;
1909 struct nfs4_client *conf, *unconf, *new;
1910 __be32 status;
1911 char dname[HEXDIR_LEN];
1912
1913 if (!check_name(clname))
1914 return nfserr_inval;
1915
1916 status = nfs4_make_rec_clidname(dname, &clname);
1917 if (status)
1918 return status;
1919
1920 /*
1921 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1922 * We get here on a DRC miss.
1923 */
1924
1925 strhashval = clientstr_hashval(dname);
1926
1927 nfs4_lock_state();
1928 conf = find_confirmed_client_by_str(dname, strhashval);
1929 if (conf) {
1930 /* RFC 3530 14.2.33 CASE 0: */
1931 status = nfserr_clid_inuse;
1932 if (clp_used_exchangeid(conf))
1933 goto out;
1934 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1935 char addr_str[INET6_ADDRSTRLEN];
1936 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1937 sizeof(addr_str));
1938 dprintk("NFSD: setclientid: string in use by client "
1939 "at %s\n", addr_str);
1940 goto out;
1941 }
1942 }
1943 /*
1944 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1945 * has a description of SETCLIENTID request processing consisting
1946 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1947 */
1948 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1949 status = nfserr_resource;
1950 if (!conf) {
1951 /*
1952 * RFC 3530 14.2.33 CASE 4:
1953 * placed first, because it is the normal case
1954 */
1955 if (unconf)
1956 expire_client(unconf);
1957 new = create_client(clname, dname, rqstp, &clverifier);
1958 if (new == NULL)
1959 goto out;
1960 gen_clid(new);
1961 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1962 /*
1963 * RFC 3530 14.2.33 CASE 1:
1964 * probable callback update
1965 */
1966 if (unconf) {
1967 /* Note this is removing unconfirmed {*x***},
1968 * which is stronger than RFC recommended {vxc**}.
1969 * This has the advantage that there is at most
1970 * one {*x***} in either list at any time.
1971 */
1972 expire_client(unconf);
1973 }
1974 new = create_client(clname, dname, rqstp, &clverifier);
1975 if (new == NULL)
1976 goto out;
1977 copy_clid(new, conf);
1978 } else if (!unconf) {
1979 /*
1980 * RFC 3530 14.2.33 CASE 2:
1981 * probable client reboot; state will be removed if
1982 * confirmed.
1983 */
1984 new = create_client(clname, dname, rqstp, &clverifier);
1985 if (new == NULL)
1986 goto out;
1987 gen_clid(new);
1988 } else {
1989 /*
1990 * RFC 3530 14.2.33 CASE 3:
1991 * probable client reboot; state will be removed if
1992 * confirmed.
1993 */
1994 expire_client(unconf);
1995 new = create_client(clname, dname, rqstp, &clverifier);
1996 if (new == NULL)
1997 goto out;
1998 gen_clid(new);
1999 }
2000 /*
2001 * XXX: we should probably set this at creation time, and check
2002 * for consistent minorversion use throughout:
2003 */
2004 new->cl_minorversion = 0;
2005 gen_callback(new, setclid, rqstp);
2006 add_to_unconfirmed(new, strhashval);
2007 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2008 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2009 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2010 status = nfs_ok;
2011out:
2012 nfs4_unlock_state();
2013 return status;
2014}
2015
2016
2017/*
2018 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2019 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2020 * bullets, labeled as CASE1 - CASE4 below.
2021 */
2022__be32
2023nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2024 struct nfsd4_compound_state *cstate,
2025 struct nfsd4_setclientid_confirm *setclientid_confirm)
2026{
2027 struct sockaddr *sa = svc_addr(rqstp);
2028 struct nfs4_client *conf, *unconf;
2029 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2030 clientid_t * clid = &setclientid_confirm->sc_clientid;
2031 __be32 status;
2032
2033 if (STALE_CLIENTID(clid))
2034 return nfserr_stale_clientid;
2035 /*
2036 * XXX The Duplicate Request Cache (DRC) has been checked (??)
2037 * We get here on a DRC miss.
2038 */
2039
2040 nfs4_lock_state();
2041
2042 conf = find_confirmed_client(clid);
2043 unconf = find_unconfirmed_client(clid);
2044
2045 status = nfserr_clid_inuse;
2046 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2047 goto out;
2048 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2049 goto out;
2050
2051 /*
2052 * section 14.2.34 of RFC 3530 has a description of
2053 * SETCLIENTID_CONFIRM request processing consisting
2054 * of 4 bullet points, labeled as CASE1 - CASE4 below.
2055 */
2056 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2057 /*
2058 * RFC 3530 14.2.34 CASE 1:
2059 * callback update
2060 */
2061 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2062 status = nfserr_clid_inuse;
2063 else {
2064 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2065 nfsd4_probe_callback(conf);
2066 expire_client(unconf);
2067 status = nfs_ok;
2068
2069 }
2070 } else if (conf && !unconf) {
2071 /*
2072 * RFC 3530 14.2.34 CASE 2:
2073 * probable retransmitted request; play it safe and
2074 * do nothing.
2075 */
2076 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2077 status = nfserr_clid_inuse;
2078 else
2079 status = nfs_ok;
2080 } else if (!conf && unconf
2081 && same_verf(&unconf->cl_confirm, &confirm)) {
2082 /*
2083 * RFC 3530 14.2.34 CASE 3:
2084 * Normal case; new or rebooted client:
2085 */
2086 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2087 status = nfserr_clid_inuse;
2088 } else {
2089 unsigned int hash =
2090 clientstr_hashval(unconf->cl_recdir);
2091 conf = find_confirmed_client_by_str(unconf->cl_recdir,
2092 hash);
2093 if (conf) {
2094 nfsd4_remove_clid_dir(conf);
2095 expire_client(conf);
2096 }
2097 move_to_confirmed(unconf);
2098 conf = unconf;
2099 nfsd4_probe_callback(conf);
2100 status = nfs_ok;
2101 }
2102 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2103 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2104 &confirm)))) {
2105 /*
2106 * RFC 3530 14.2.34 CASE 4:
2107 * Client probably hasn't noticed that we rebooted yet.
2108 */
2109 status = nfserr_stale_clientid;
2110 } else {
2111 /* check that we have hit one of the cases...*/
2112 status = nfserr_clid_inuse;
2113 }
2114out:
2115 nfs4_unlock_state();
2116 return status;
2117}
2118
2119/* OPEN Share state helper functions */
2120static inline struct nfs4_file *
2121alloc_init_file(struct inode *ino)
2122{
2123 struct nfs4_file *fp;
2124 unsigned int hashval = file_hashval(ino);
2125
2126 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2127 if (fp) {
2128 atomic_set(&fp->fi_ref, 1);
2129 INIT_LIST_HEAD(&fp->fi_hash);
2130 INIT_LIST_HEAD(&fp->fi_stateids);
2131 INIT_LIST_HEAD(&fp->fi_delegations);
2132 fp->fi_inode = igrab(ino);
2133 fp->fi_id = current_fileid++;
2134 fp->fi_had_conflict = false;
2135 fp->fi_lease = NULL;
2136 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2137 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2138 spin_lock(&recall_lock);
2139 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2140 spin_unlock(&recall_lock);
2141 return fp;
2142 }
2143 return NULL;
2144}
2145
2146static void
2147nfsd4_free_slab(struct kmem_cache **slab)
2148{
2149 if (*slab == NULL)
2150 return;
2151 kmem_cache_destroy(*slab);
2152 *slab = NULL;
2153}
2154
2155void
2156nfsd4_free_slabs(void)
2157{
2158 nfsd4_free_slab(&stateowner_slab);
2159 nfsd4_free_slab(&file_slab);
2160 nfsd4_free_slab(&stateid_slab);
2161 nfsd4_free_slab(&deleg_slab);
2162}
2163
2164static int
2165nfsd4_init_slabs(void)
2166{
2167 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2168 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2169 if (stateowner_slab == NULL)
2170 goto out_nomem;
2171 file_slab = kmem_cache_create("nfsd4_files",
2172 sizeof(struct nfs4_file), 0, 0, NULL);
2173 if (file_slab == NULL)
2174 goto out_nomem;
2175 stateid_slab = kmem_cache_create("nfsd4_stateids",
2176 sizeof(struct nfs4_stateid), 0, 0, NULL);
2177 if (stateid_slab == NULL)
2178 goto out_nomem;
2179 deleg_slab = kmem_cache_create("nfsd4_delegations",
2180 sizeof(struct nfs4_delegation), 0, 0, NULL);
2181 if (deleg_slab == NULL)
2182 goto out_nomem;
2183 return 0;
2184out_nomem:
2185 nfsd4_free_slabs();
2186 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2187 return -ENOMEM;
2188}
2189
2190void
2191nfs4_free_stateowner(struct kref *kref)
2192{
2193 struct nfs4_stateowner *sop =
2194 container_of(kref, struct nfs4_stateowner, so_ref);
2195 kfree(sop->so_owner.data);
2196 kmem_cache_free(stateowner_slab, sop);
2197}
2198
2199static inline struct nfs4_stateowner *
2200alloc_stateowner(struct xdr_netobj *owner)
2201{
2202 struct nfs4_stateowner *sop;
2203
2204 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2205 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2206 memcpy(sop->so_owner.data, owner->data, owner->len);
2207 sop->so_owner.len = owner->len;
2208 kref_init(&sop->so_ref);
2209 return sop;
2210 }
2211 kmem_cache_free(stateowner_slab, sop);
2212 }
2213 return NULL;
2214}
2215
2216static struct nfs4_stateowner *
2217alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2218 struct nfs4_stateowner *sop;
2219 struct nfs4_replay *rp;
2220 unsigned int idhashval;
2221
2222 if (!(sop = alloc_stateowner(&open->op_owner)))
2223 return NULL;
2224 idhashval = ownerid_hashval(current_ownerid);
2225 INIT_LIST_HEAD(&sop->so_idhash);
2226 INIT_LIST_HEAD(&sop->so_strhash);
2227 INIT_LIST_HEAD(&sop->so_perclient);
2228 INIT_LIST_HEAD(&sop->so_stateids);
2229 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2230 INIT_LIST_HEAD(&sop->so_close_lru);
2231 sop->so_time = 0;
2232 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2233 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2234 list_add(&sop->so_perclient, &clp->cl_openowners);
2235 sop->so_is_open_owner = 1;
2236 sop->so_id = current_ownerid++;
2237 sop->so_client = clp;
2238 sop->so_seqid = open->op_seqid;
2239 sop->so_confirmed = 0;
2240 rp = &sop->so_replay;
2241 rp->rp_status = nfserr_serverfault;
2242 rp->rp_buflen = 0;
2243 rp->rp_buf = rp->rp_ibuf;
2244 return sop;
2245}
2246
2247static inline void
2248init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2249 struct nfs4_stateowner *sop = open->op_stateowner;
2250 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2251
2252 INIT_LIST_HEAD(&stp->st_hash);
2253 INIT_LIST_HEAD(&stp->st_perstateowner);
2254 INIT_LIST_HEAD(&stp->st_lockowners);
2255 INIT_LIST_HEAD(&stp->st_perfile);
2256 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2257 list_add(&stp->st_perstateowner, &sop->so_stateids);
2258 list_add(&stp->st_perfile, &fp->fi_stateids);
2259 stp->st_stateowner = sop;
2260 get_nfs4_file(fp);
2261 stp->st_file = fp;
2262 stp->st_stateid.si_boot = boot_time;
2263 stp->st_stateid.si_stateownerid = sop->so_id;
2264 stp->st_stateid.si_fileid = fp->fi_id;
2265 stp->st_stateid.si_generation = 0;
2266 stp->st_access_bmap = 0;
2267 stp->st_deny_bmap = 0;
2268 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2269 &stp->st_access_bmap);
2270 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2271 stp->st_openstp = NULL;
2272}
2273
2274static void
2275move_to_close_lru(struct nfs4_stateowner *sop)
2276{
2277 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2278
2279 list_move_tail(&sop->so_close_lru, &close_lru);
2280 sop->so_time = get_seconds();
2281}
2282
2283static int
2284same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2285 clientid_t *clid)
2286{
2287 return (sop->so_owner.len == owner->len) &&
2288 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2289 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2290}
2291
2292static struct nfs4_stateowner *
2293find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2294{
2295 struct nfs4_stateowner *so = NULL;
2296
2297 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2298 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2299 return so;
2300 }
2301 return NULL;
2302}
2303
2304/* search file_hashtbl[] for file */
2305static struct nfs4_file *
2306find_file(struct inode *ino)
2307{
2308 unsigned int hashval = file_hashval(ino);
2309 struct nfs4_file *fp;
2310
2311 spin_lock(&recall_lock);
2312 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2313 if (fp->fi_inode == ino) {
2314 get_nfs4_file(fp);
2315 spin_unlock(&recall_lock);
2316 return fp;
2317 }
2318 }
2319 spin_unlock(&recall_lock);
2320 return NULL;
2321}
2322
2323static inline int access_valid(u32 x, u32 minorversion)
2324{
2325 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2326 return 0;
2327 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2328 return 0;
2329 x &= ~NFS4_SHARE_ACCESS_MASK;
2330 if (minorversion && x) {
2331 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2332 return 0;
2333 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2334 return 0;
2335 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2336 }
2337 if (x)
2338 return 0;
2339 return 1;
2340}
2341
2342static inline int deny_valid(u32 x)
2343{
2344 /* Note: unlike access bits, deny bits may be zero. */
2345 return x <= NFS4_SHARE_DENY_BOTH;
2346}
2347
2348/*
2349 * Called to check deny when READ with all zero stateid or
2350 * WRITE with all zero or all one stateid
2351 */
2352static __be32
2353nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2354{
2355 struct inode *ino = current_fh->fh_dentry->d_inode;
2356 struct nfs4_file *fp;
2357 struct nfs4_stateid *stp;
2358 __be32 ret;
2359
2360 dprintk("NFSD: nfs4_share_conflict\n");
2361
2362 fp = find_file(ino);
2363 if (!fp)
2364 return nfs_ok;
2365 ret = nfserr_locked;
2366 /* Search for conflicting share reservations */
2367 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2368 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2369 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2370 goto out;
2371 }
2372 ret = nfs_ok;
2373out:
2374 put_nfs4_file(fp);
2375 return ret;
2376}
2377
2378static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2379{
2380 /* We're assuming the state code never drops its reference
2381 * without first removing the lease. Since we're in this lease
2382 * callback (and since the lease code is serialized by the kernel
2383 * lock) we know the server hasn't removed the lease yet, we know
2384 * it's safe to take a reference: */
2385 atomic_inc(&dp->dl_count);
2386
2387 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2388
2389 /* only place dl_time is set. protected by lock_flocks*/
2390 dp->dl_time = get_seconds();
2391
2392 nfsd4_cb_recall(dp);
2393}
2394
2395/* Called from break_lease() with lock_flocks() held. */
2396static void nfsd_break_deleg_cb(struct file_lock *fl)
2397{
2398 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2399 struct nfs4_delegation *dp;
2400
2401 BUG_ON(!fp);
2402 /* We assume break_lease is only called once per lease: */
2403 BUG_ON(fp->fi_had_conflict);
2404 /*
2405 * We don't want the locks code to timeout the lease for us;
2406 * we'll remove it ourself if a delegation isn't returned
2407 * in time:
2408 */
2409 fl->fl_break_time = 0;
2410
2411 spin_lock(&recall_lock);
2412 fp->fi_had_conflict = true;
2413 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2414 nfsd_break_one_deleg(dp);
2415 spin_unlock(&recall_lock);
2416}
2417
2418static
2419int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2420{
2421 if (arg & F_UNLCK)
2422 return lease_modify(onlist, arg);
2423 else
2424 return -EAGAIN;
2425}
2426
2427static const struct lock_manager_operations nfsd_lease_mng_ops = {
2428 .lm_break = nfsd_break_deleg_cb,
2429 .lm_change = nfsd_change_deleg_cb,
2430};
2431
2432
2433__be32
2434nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2435 struct nfsd4_open *open)
2436{
2437 clientid_t *clientid = &open->op_clientid;
2438 struct nfs4_client *clp = NULL;
2439 unsigned int strhashval;
2440 struct nfs4_stateowner *sop = NULL;
2441
2442 if (!check_name(open->op_owner))
2443 return nfserr_inval;
2444
2445 if (STALE_CLIENTID(&open->op_clientid))
2446 return nfserr_stale_clientid;
2447
2448 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2449 sop = find_openstateowner_str(strhashval, open);
2450 open->op_stateowner = sop;
2451 if (!sop) {
2452 /* Make sure the client's lease hasn't expired. */
2453 clp = find_confirmed_client(clientid);
2454 if (clp == NULL)
2455 return nfserr_expired;
2456 goto renew;
2457 }
2458 /* When sessions are used, skip open sequenceid processing */
2459 if (nfsd4_has_session(cstate))
2460 goto renew;
2461 if (!sop->so_confirmed) {
2462 /* Replace unconfirmed owners without checking for replay. */
2463 clp = sop->so_client;
2464 release_openowner(sop);
2465 open->op_stateowner = NULL;
2466 goto renew;
2467 }
2468 if (open->op_seqid == sop->so_seqid - 1) {
2469 if (sop->so_replay.rp_buflen)
2470 return nfserr_replay_me;
2471 /* The original OPEN failed so spectacularly
2472 * that we don't even have replay data saved!
2473 * Therefore, we have no choice but to continue
2474 * processing this OPEN; presumably, we'll
2475 * fail again for the same reason.
2476 */
2477 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2478 goto renew;
2479 }
2480 if (open->op_seqid != sop->so_seqid)
2481 return nfserr_bad_seqid;
2482renew:
2483 if (open->op_stateowner == NULL) {
2484 sop = alloc_init_open_stateowner(strhashval, clp, open);
2485 if (sop == NULL)
2486 return nfserr_resource;
2487 open->op_stateowner = sop;
2488 }
2489 list_del_init(&sop->so_close_lru);
2490 renew_client(sop->so_client);
2491 return nfs_ok;
2492}
2493
2494static inline __be32
2495nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2496{
2497 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2498 return nfserr_openmode;
2499 else
2500 return nfs_ok;
2501}
2502
2503static struct nfs4_delegation *
2504find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2505{
2506 struct nfs4_delegation *dp;
2507
2508 spin_lock(&recall_lock);
2509 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2510 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid) {
2511 spin_unlock(&recall_lock);
2512 return dp;
2513 }
2514 spin_unlock(&recall_lock);
2515 return NULL;
2516}
2517
2518static int share_access_to_flags(u32 share_access)
2519{
2520 share_access &= ~NFS4_SHARE_WANT_MASK;
2521
2522 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2523}
2524
2525static __be32
2526nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2527 struct nfs4_delegation **dp)
2528{
2529 int flags;
2530 __be32 status = nfserr_bad_stateid;
2531
2532 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2533 if (*dp == NULL)
2534 goto out;
2535 flags = share_access_to_flags(open->op_share_access);
2536 status = nfs4_check_delegmode(*dp, flags);
2537 if (status)
2538 *dp = NULL;
2539out:
2540 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2541 return nfs_ok;
2542 if (status)
2543 return status;
2544 open->op_stateowner->so_confirmed = 1;
2545 return nfs_ok;
2546}
2547
2548static __be32
2549nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2550{
2551 struct nfs4_stateid *local;
2552 __be32 status = nfserr_share_denied;
2553 struct nfs4_stateowner *sop = open->op_stateowner;
2554
2555 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2556 /* ignore lock owners */
2557 if (local->st_stateowner->so_is_open_owner == 0)
2558 continue;
2559 /* remember if we have seen this open owner */
2560 if (local->st_stateowner == sop)
2561 *stpp = local;
2562 /* check for conflicting share reservations */
2563 if (!test_share(local, open))
2564 goto out;
2565 }
2566 status = 0;
2567out:
2568 return status;
2569}
2570
2571static inline struct nfs4_stateid *
2572nfs4_alloc_stateid(void)
2573{
2574 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2575}
2576
2577static inline int nfs4_access_to_access(u32 nfs4_access)
2578{
2579 int flags = 0;
2580
2581 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2582 flags |= NFSD_MAY_READ;
2583 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2584 flags |= NFSD_MAY_WRITE;
2585 return flags;
2586}
2587
2588static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2589 struct svc_fh *cur_fh, struct nfsd4_open *open)
2590{
2591 __be32 status;
2592 int oflag = nfs4_access_to_omode(open->op_share_access);
2593 int access = nfs4_access_to_access(open->op_share_access);
2594
2595 /* CLAIM_DELEGATE_CUR is used in response to a broken lease;
2596 * allowing it to break the lease and return EAGAIN leaves the
2597 * client unable to make progress in returning the delegation */
2598 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2599 access |= NFSD_MAY_NOT_BREAK_LEASE;
2600
2601 if (!fp->fi_fds[oflag]) {
2602 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2603 &fp->fi_fds[oflag]);
2604 if (status)
2605 return status;
2606 }
2607 nfs4_file_get_access(fp, oflag);
2608
2609 return nfs_ok;
2610}
2611
2612static __be32
2613nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2614 struct nfs4_file *fp, struct svc_fh *cur_fh,
2615 struct nfsd4_open *open)
2616{
2617 struct nfs4_stateid *stp;
2618 __be32 status;
2619
2620 stp = nfs4_alloc_stateid();
2621 if (stp == NULL)
2622 return nfserr_resource;
2623
2624 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2625 if (status) {
2626 kmem_cache_free(stateid_slab, stp);
2627 return status;
2628 }
2629 *stpp = stp;
2630 return 0;
2631}
2632
2633static inline __be32
2634nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2635 struct nfsd4_open *open)
2636{
2637 struct iattr iattr = {
2638 .ia_valid = ATTR_SIZE,
2639 .ia_size = 0,
2640 };
2641 if (!open->op_truncate)
2642 return 0;
2643 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2644 return nfserr_inval;
2645 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2646}
2647
2648static __be32
2649nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2650{
2651 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2652 bool new_access;
2653 __be32 status;
2654
2655 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2656 if (new_access) {
2657 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2658 if (status)
2659 return status;
2660 }
2661 status = nfsd4_truncate(rqstp, cur_fh, open);
2662 if (status) {
2663 if (new_access) {
2664 int oflag = nfs4_access_to_omode(op_share_access);
2665 nfs4_file_put_access(fp, oflag);
2666 }
2667 return status;
2668 }
2669 /* remember the open */
2670 __set_bit(op_share_access, &stp->st_access_bmap);
2671 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2672
2673 return nfs_ok;
2674}
2675
2676
2677static void
2678nfs4_set_claim_prev(struct nfsd4_open *open)
2679{
2680 open->op_stateowner->so_confirmed = 1;
2681 open->op_stateowner->so_client->cl_firststate = 1;
2682}
2683
2684/* Should we give out recallable state?: */
2685static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2686{
2687 if (clp->cl_cb_state == NFSD4_CB_UP)
2688 return true;
2689 /*
2690 * In the sessions case, since we don't have to establish a
2691 * separate connection for callbacks, we assume it's OK
2692 * until we hear otherwise:
2693 */
2694 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2695}
2696
2697static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2698{
2699 struct file_lock *fl;
2700
2701 fl = locks_alloc_lock();
2702 if (!fl)
2703 return NULL;
2704 locks_init_lock(fl);
2705 fl->fl_lmops = &nfsd_lease_mng_ops;
2706 fl->fl_flags = FL_LEASE;
2707 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2708 fl->fl_end = OFFSET_MAX;
2709 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2710 fl->fl_pid = current->tgid;
2711 return fl;
2712}
2713
2714static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2715{
2716 struct nfs4_file *fp = dp->dl_file;
2717 struct file_lock *fl;
2718 int status;
2719
2720 fl = nfs4_alloc_init_lease(dp, flag);
2721 if (!fl)
2722 return -ENOMEM;
2723 fl->fl_file = find_readable_file(fp);
2724 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2725 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2726 if (status) {
2727 list_del_init(&dp->dl_perclnt);
2728 locks_free_lock(fl);
2729 return -ENOMEM;
2730 }
2731 fp->fi_lease = fl;
2732 fp->fi_deleg_file = fl->fl_file;
2733 get_file(fp->fi_deleg_file);
2734 atomic_set(&fp->fi_delegees, 1);
2735 list_add(&dp->dl_perfile, &fp->fi_delegations);
2736 return 0;
2737}
2738
2739static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2740{
2741 struct nfs4_file *fp = dp->dl_file;
2742
2743 if (!fp->fi_lease)
2744 return nfs4_setlease(dp, flag);
2745 spin_lock(&recall_lock);
2746 if (fp->fi_had_conflict) {
2747 spin_unlock(&recall_lock);
2748 return -EAGAIN;
2749 }
2750 atomic_inc(&fp->fi_delegees);
2751 list_add(&dp->dl_perfile, &fp->fi_delegations);
2752 spin_unlock(&recall_lock);
2753 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2754 return 0;
2755}
2756
2757/*
2758 * Attempt to hand out a delegation.
2759 */
2760static void
2761nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2762{
2763 struct nfs4_delegation *dp;
2764 struct nfs4_stateowner *sop = stp->st_stateowner;
2765 int cb_up;
2766 int status, flag = 0;
2767
2768 cb_up = nfsd4_cb_channel_good(sop->so_client);
2769 flag = NFS4_OPEN_DELEGATE_NONE;
2770 open->op_recall = 0;
2771 switch (open->op_claim_type) {
2772 case NFS4_OPEN_CLAIM_PREVIOUS:
2773 if (!cb_up)
2774 open->op_recall = 1;
2775 flag = open->op_delegate_type;
2776 if (flag == NFS4_OPEN_DELEGATE_NONE)
2777 goto out;
2778 break;
2779 case NFS4_OPEN_CLAIM_NULL:
2780 /* Let's not give out any delegations till everyone's
2781 * had the chance to reclaim theirs.... */
2782 if (locks_in_grace())
2783 goto out;
2784 if (!cb_up || !sop->so_confirmed)
2785 goto out;
2786 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2787 flag = NFS4_OPEN_DELEGATE_WRITE;
2788 else
2789 flag = NFS4_OPEN_DELEGATE_READ;
2790 break;
2791 default:
2792 goto out;
2793 }
2794
2795 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2796 if (dp == NULL)
2797 goto out_no_deleg;
2798 status = nfs4_set_delegation(dp, flag);
2799 if (status)
2800 goto out_free;
2801
2802 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2803
2804 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2805 STATEID_VAL(&dp->dl_stateid));
2806out:
2807 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2808 && flag == NFS4_OPEN_DELEGATE_NONE
2809 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2810 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2811 open->op_delegate_type = flag;
2812 return;
2813out_free:
2814 nfs4_put_delegation(dp);
2815out_no_deleg:
2816 flag = NFS4_OPEN_DELEGATE_NONE;
2817 goto out;
2818}
2819
2820/*
2821 * called with nfs4_lock_state() held.
2822 */
2823__be32
2824nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2825{
2826 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2827 struct nfs4_file *fp = NULL;
2828 struct inode *ino = current_fh->fh_dentry->d_inode;
2829 struct nfs4_stateid *stp = NULL;
2830 struct nfs4_delegation *dp = NULL;
2831 __be32 status;
2832
2833 status = nfserr_inval;
2834 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2835 || !deny_valid(open->op_share_deny))
2836 goto out;
2837 /*
2838 * Lookup file; if found, lookup stateid and check open request,
2839 * and check for delegations in the process of being recalled.
2840 * If not found, create the nfs4_file struct
2841 */
2842 fp = find_file(ino);
2843 if (fp) {
2844 if ((status = nfs4_check_open(fp, open, &stp)))
2845 goto out;
2846 status = nfs4_check_deleg(fp, open, &dp);
2847 if (status)
2848 goto out;
2849 } else {
2850 status = nfserr_bad_stateid;
2851 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2852 goto out;
2853 status = nfserr_resource;
2854 fp = alloc_init_file(ino);
2855 if (fp == NULL)
2856 goto out;
2857 }
2858
2859 /*
2860 * OPEN the file, or upgrade an existing OPEN.
2861 * If truncate fails, the OPEN fails.
2862 */
2863 if (stp) {
2864 /* Stateid was found, this is an OPEN upgrade */
2865 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2866 if (status)
2867 goto out;
2868 update_stateid(&stp->st_stateid);
2869 } else {
2870 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2871 if (status)
2872 goto out;
2873 init_stateid(stp, fp, open);
2874 status = nfsd4_truncate(rqstp, current_fh, open);
2875 if (status) {
2876 release_open_stateid(stp);
2877 goto out;
2878 }
2879 if (nfsd4_has_session(&resp->cstate))
2880 update_stateid(&stp->st_stateid);
2881 }
2882 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2883
2884 if (nfsd4_has_session(&resp->cstate))
2885 open->op_stateowner->so_confirmed = 1;
2886
2887 /*
2888 * Attempt to hand out a delegation. No error return, because the
2889 * OPEN succeeds even if we fail.
2890 */
2891 nfs4_open_delegation(current_fh, open, stp);
2892
2893 status = nfs_ok;
2894
2895 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2896 STATEID_VAL(&stp->st_stateid));
2897out:
2898 if (fp)
2899 put_nfs4_file(fp);
2900 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2901 nfs4_set_claim_prev(open);
2902 /*
2903 * To finish the open response, we just need to set the rflags.
2904 */
2905 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2906 if (!open->op_stateowner->so_confirmed &&
2907 !nfsd4_has_session(&resp->cstate))
2908 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2909
2910 return status;
2911}
2912
2913__be32
2914nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2915 clientid_t *clid)
2916{
2917 struct nfs4_client *clp;
2918 __be32 status;
2919
2920 nfs4_lock_state();
2921 dprintk("process_renew(%08x/%08x): starting\n",
2922 clid->cl_boot, clid->cl_id);
2923 status = nfserr_stale_clientid;
2924 if (STALE_CLIENTID(clid))
2925 goto out;
2926 clp = find_confirmed_client(clid);
2927 status = nfserr_expired;
2928 if (clp == NULL) {
2929 /* We assume the client took too long to RENEW. */
2930 dprintk("nfsd4_renew: clientid not found!\n");
2931 goto out;
2932 }
2933 renew_client(clp);
2934 status = nfserr_cb_path_down;
2935 if (!list_empty(&clp->cl_delegations)
2936 && clp->cl_cb_state != NFSD4_CB_UP)
2937 goto out;
2938 status = nfs_ok;
2939out:
2940 nfs4_unlock_state();
2941 return status;
2942}
2943
2944static struct lock_manager nfsd4_manager = {
2945};
2946
2947static void
2948nfsd4_end_grace(void)
2949{
2950 dprintk("NFSD: end of grace period\n");
2951 nfsd4_recdir_purge_old();
2952 locks_end_grace(&nfsd4_manager);
2953 /*
2954 * Now that every NFSv4 client has had the chance to recover and
2955 * to see the (possibly new, possibly shorter) lease time, we
2956 * can safely set the next grace time to the current lease time:
2957 */
2958 nfsd4_grace = nfsd4_lease;
2959}
2960
2961static time_t
2962nfs4_laundromat(void)
2963{
2964 struct nfs4_client *clp;
2965 struct nfs4_stateowner *sop;
2966 struct nfs4_delegation *dp;
2967 struct list_head *pos, *next, reaplist;
2968 time_t cutoff = get_seconds() - nfsd4_lease;
2969 time_t t, clientid_val = nfsd4_lease;
2970 time_t u, test_val = nfsd4_lease;
2971
2972 nfs4_lock_state();
2973
2974 dprintk("NFSD: laundromat service - starting\n");
2975 if (locks_in_grace())
2976 nfsd4_end_grace();
2977 INIT_LIST_HEAD(&reaplist);
2978 spin_lock(&client_lock);
2979 list_for_each_safe(pos, next, &client_lru) {
2980 clp = list_entry(pos, struct nfs4_client, cl_lru);
2981 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2982 t = clp->cl_time - cutoff;
2983 if (clientid_val > t)
2984 clientid_val = t;
2985 break;
2986 }
2987 if (atomic_read(&clp->cl_refcount)) {
2988 dprintk("NFSD: client in use (clientid %08x)\n",
2989 clp->cl_clientid.cl_id);
2990 continue;
2991 }
2992 unhash_client_locked(clp);
2993 list_add(&clp->cl_lru, &reaplist);
2994 }
2995 spin_unlock(&client_lock);
2996 list_for_each_safe(pos, next, &reaplist) {
2997 clp = list_entry(pos, struct nfs4_client, cl_lru);
2998 dprintk("NFSD: purging unused client (clientid %08x)\n",
2999 clp->cl_clientid.cl_id);
3000 nfsd4_remove_clid_dir(clp);
3001 expire_client(clp);
3002 }
3003 spin_lock(&recall_lock);
3004 list_for_each_safe(pos, next, &del_recall_lru) {
3005 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3006 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3007 u = dp->dl_time - cutoff;
3008 if (test_val > u)
3009 test_val = u;
3010 break;
3011 }
3012 list_move(&dp->dl_recall_lru, &reaplist);
3013 }
3014 spin_unlock(&recall_lock);
3015 list_for_each_safe(pos, next, &reaplist) {
3016 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3017 list_del_init(&dp->dl_recall_lru);
3018 unhash_delegation(dp);
3019 }
3020 test_val = nfsd4_lease;
3021 list_for_each_safe(pos, next, &close_lru) {
3022 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
3023 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
3024 u = sop->so_time - cutoff;
3025 if (test_val > u)
3026 test_val = u;
3027 break;
3028 }
3029 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
3030 sop->so_id);
3031 release_openowner(sop);
3032 }
3033 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3034 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3035 nfs4_unlock_state();
3036 return clientid_val;
3037}
3038
3039static struct workqueue_struct *laundry_wq;
3040static void laundromat_main(struct work_struct *);
3041static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3042
3043static void
3044laundromat_main(struct work_struct *not_used)
3045{
3046 time_t t;
3047
3048 t = nfs4_laundromat();
3049 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3050 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3051}
3052
3053static struct nfs4_stateowner *
3054search_close_lru(u32 st_id, int flags)
3055{
3056 struct nfs4_stateowner *local = NULL;
3057
3058 if (flags & CLOSE_STATE) {
3059 list_for_each_entry(local, &close_lru, so_close_lru) {
3060 if (local->so_id == st_id)
3061 return local;
3062 }
3063 }
3064 return NULL;
3065}
3066
3067static inline int
3068nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
3069{
3070 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
3071}
3072
3073static int
3074STALE_STATEID(stateid_t *stateid)
3075{
3076 if (stateid->si_boot == boot_time)
3077 return 0;
3078 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3079 STATEID_VAL(stateid));
3080 return 1;
3081}
3082
3083static inline int
3084access_permit_read(unsigned long access_bmap)
3085{
3086 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3087 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3088 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3089}
3090
3091static inline int
3092access_permit_write(unsigned long access_bmap)
3093{
3094 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3095 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3096}
3097
3098static
3099__be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
3100{
3101 __be32 status = nfserr_openmode;
3102
3103 /* For lock stateid's, we test the parent open, not the lock: */
3104 if (stp->st_openstp)
3105 stp = stp->st_openstp;
3106 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3107 goto out;
3108 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3109 goto out;
3110 status = nfs_ok;
3111out:
3112 return status;
3113}
3114
3115static inline __be32
3116check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3117{
3118 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3119 return nfs_ok;
3120 else if (locks_in_grace()) {
3121 /* Answer in remaining cases depends on existence of
3122 * conflicting state; so we must wait out the grace period. */
3123 return nfserr_grace;
3124 } else if (flags & WR_STATE)
3125 return nfs4_share_conflict(current_fh,
3126 NFS4_SHARE_DENY_WRITE);
3127 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3128 return nfs4_share_conflict(current_fh,
3129 NFS4_SHARE_DENY_READ);
3130}
3131
3132/*
3133 * Allow READ/WRITE during grace period on recovered state only for files
3134 * that are not able to provide mandatory locking.
3135 */
3136static inline int
3137grace_disallows_io(struct inode *inode)
3138{
3139 return locks_in_grace() && mandatory_lock(inode);
3140}
3141
3142static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3143{
3144 /*
3145 * When sessions are used the stateid generation number is ignored
3146 * when it is zero.
3147 */
3148 if ((flags & HAS_SESSION) && in->si_generation == 0)
3149 goto out;
3150
3151 /* If the client sends us a stateid from the future, it's buggy: */
3152 if (in->si_generation > ref->si_generation)
3153 return nfserr_bad_stateid;
3154 /*
3155 * The following, however, can happen. For example, if the
3156 * client sends an open and some IO at the same time, the open
3157 * may bump si_generation while the IO is still in flight.
3158 * Thanks to hard links and renames, the client never knows what
3159 * file an open will affect. So it could avoid that situation
3160 * only by serializing all opens and IO from the same open
3161 * owner. To recover from the old_stateid error, the client
3162 * will just have to retry the IO:
3163 */
3164 if (in->si_generation < ref->si_generation)
3165 return nfserr_old_stateid;
3166out:
3167 return nfs_ok;
3168}
3169
3170static int is_delegation_stateid(stateid_t *stateid)
3171{
3172 return stateid->si_fileid == 0;
3173}
3174
3175static int is_open_stateid(struct nfs4_stateid *stateid)
3176{
3177 return stateid->st_openstp == NULL;
3178}
3179
3180__be32 nfs4_validate_stateid(stateid_t *stateid, int flags)
3181{
3182 struct nfs4_stateid *stp = NULL;
3183 __be32 status = nfserr_stale_stateid;
3184
3185 if (STALE_STATEID(stateid))
3186 goto out;
3187
3188 status = nfserr_expired;
3189 stp = search_for_stateid(stateid);
3190 if (!stp)
3191 goto out;
3192 status = nfserr_bad_stateid;
3193
3194 if (!stp->st_stateowner->so_confirmed)
3195 goto out;
3196
3197 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3198 if (status)
3199 goto out;
3200
3201 status = nfs_ok;
3202out:
3203 return status;
3204}
3205
3206/*
3207* Checks for stateid operations
3208*/
3209__be32
3210nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3211 stateid_t *stateid, int flags, struct file **filpp)
3212{
3213 struct nfs4_stateid *stp = NULL;
3214 struct nfs4_delegation *dp = NULL;
3215 struct svc_fh *current_fh = &cstate->current_fh;
3216 struct inode *ino = current_fh->fh_dentry->d_inode;
3217 __be32 status;
3218
3219 if (filpp)
3220 *filpp = NULL;
3221
3222 if (grace_disallows_io(ino))
3223 return nfserr_grace;
3224
3225 if (nfsd4_has_session(cstate))
3226 flags |= HAS_SESSION;
3227
3228 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3229 return check_special_stateids(current_fh, stateid, flags);
3230
3231 status = nfserr_stale_stateid;
3232 if (STALE_STATEID(stateid))
3233 goto out;
3234
3235 /*
3236 * We assume that any stateid that has the current boot time,
3237 * but that we can't find, is expired:
3238 */
3239 status = nfserr_expired;
3240 if (is_delegation_stateid(stateid)) {
3241 dp = find_delegation_stateid(ino, stateid);
3242 if (!dp)
3243 goto out;
3244 status = check_stateid_generation(stateid, &dp->dl_stateid,
3245 flags);
3246 if (status)
3247 goto out;
3248 status = nfs4_check_delegmode(dp, flags);
3249 if (status)
3250 goto out;
3251 renew_client(dp->dl_client);
3252 if (filpp) {
3253 *filpp = dp->dl_file->fi_deleg_file;
3254 BUG_ON(!*filpp);
3255 }
3256 } else { /* open or lock stateid */
3257 stp = find_stateid(stateid, flags);
3258 if (!stp)
3259 goto out;
3260 status = nfserr_bad_stateid;
3261 if (nfs4_check_fh(current_fh, stp))
3262 goto out;
3263 if (!stp->st_stateowner->so_confirmed)
3264 goto out;
3265 status = check_stateid_generation(stateid, &stp->st_stateid,
3266 flags);
3267 if (status)
3268 goto out;
3269 status = nfs4_check_openmode(stp, flags);
3270 if (status)
3271 goto out;
3272 renew_client(stp->st_stateowner->so_client);
3273 if (filpp) {
3274 if (flags & RD_STATE)
3275 *filpp = find_readable_file(stp->st_file);
3276 else
3277 *filpp = find_writeable_file(stp->st_file);
3278 }
3279 }
3280 status = nfs_ok;
3281out:
3282 return status;
3283}
3284
3285static __be32
3286nfsd4_free_delegation_stateid(stateid_t *stateid)
3287{
3288 struct nfs4_delegation *dp = search_for_delegation(stateid);
3289 if (dp)
3290 return nfserr_locks_held;
3291 return nfserr_bad_stateid;
3292}
3293
3294static __be32
3295nfsd4_free_lock_stateid(struct nfs4_stateid *stp)
3296{
3297 if (check_for_locks(stp->st_file, stp->st_stateowner))
3298 return nfserr_locks_held;
3299 release_lock_stateid(stp);
3300 return nfs_ok;
3301}
3302
3303/*
3304 * Test if the stateid is valid
3305 */
3306__be32
3307nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3308 struct nfsd4_test_stateid *test_stateid)
3309{
3310 test_stateid->ts_has_session = nfsd4_has_session(cstate);
3311 return nfs_ok;
3312}
3313
3314/*
3315 * Free a state id
3316 */
3317__be32
3318nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3319 struct nfsd4_free_stateid *free_stateid)
3320{
3321 stateid_t *stateid = &free_stateid->fr_stateid;
3322 struct nfs4_stateid *stp;
3323 __be32 ret;
3324
3325 nfs4_lock_state();
3326 if (is_delegation_stateid(stateid)) {
3327 ret = nfsd4_free_delegation_stateid(stateid);
3328 goto out;
3329 }
3330
3331 stp = search_for_stateid(stateid);
3332 if (!stp) {
3333 ret = nfserr_bad_stateid;
3334 goto out;
3335 }
3336 if (stateid->si_generation != 0) {
3337 if (stateid->si_generation < stp->st_stateid.si_generation) {
3338 ret = nfserr_old_stateid;
3339 goto out;
3340 }
3341 if (stateid->si_generation > stp->st_stateid.si_generation) {
3342 ret = nfserr_bad_stateid;
3343 goto out;
3344 }
3345 }
3346
3347 if (is_open_stateid(stp)) {
3348 ret = nfserr_locks_held;
3349 goto out;
3350 } else {
3351 ret = nfsd4_free_lock_stateid(stp);
3352 goto out;
3353 }
3354
3355out:
3356 nfs4_unlock_state();
3357 return ret;
3358}
3359
3360static inline int
3361setlkflg (int type)
3362{
3363 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3364 RD_STATE : WR_STATE;
3365}
3366
3367/*
3368 * Checks for sequence id mutating operations.
3369 */
3370static __be32
3371nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3372 stateid_t *stateid, int flags,
3373 struct nfs4_stateowner **sopp,
3374 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3375{
3376 struct nfs4_stateid *stp;
3377 struct nfs4_stateowner *sop;
3378 struct svc_fh *current_fh = &cstate->current_fh;
3379 __be32 status;
3380
3381 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3382 seqid, STATEID_VAL(stateid));
3383
3384 *stpp = NULL;
3385 *sopp = NULL;
3386
3387 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3388 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3389 return nfserr_bad_stateid;
3390 }
3391
3392 if (STALE_STATEID(stateid))
3393 return nfserr_stale_stateid;
3394
3395 if (nfsd4_has_session(cstate))
3396 flags |= HAS_SESSION;
3397
3398 /*
3399 * We return BAD_STATEID if filehandle doesn't match stateid,
3400 * the confirmed flag is incorrecly set, or the generation
3401 * number is incorrect.
3402 */
3403 stp = find_stateid(stateid, flags);
3404 if (stp == NULL) {
3405 /*
3406 * Also, we should make sure this isn't just the result of
3407 * a replayed close:
3408 */
3409 sop = search_close_lru(stateid->si_stateownerid, flags);
3410 /* It's not stale; let's assume it's expired: */
3411 if (sop == NULL)
3412 return nfserr_expired;
3413 *sopp = sop;
3414 goto check_replay;
3415 }
3416
3417 *stpp = stp;
3418 *sopp = sop = stp->st_stateowner;
3419
3420 if (lock) {
3421 clientid_t *lockclid = &lock->v.new.clientid;
3422 struct nfs4_client *clp = sop->so_client;
3423 int lkflg = 0;
3424 __be32 status;
3425
3426 lkflg = setlkflg(lock->lk_type);
3427
3428 if (lock->lk_is_new) {
3429 if (!sop->so_is_open_owner)
3430 return nfserr_bad_stateid;
3431 if (!(flags & HAS_SESSION) &&
3432 !same_clid(&clp->cl_clientid, lockclid))
3433 return nfserr_bad_stateid;
3434 /* stp is the open stateid */
3435 status = nfs4_check_openmode(stp, lkflg);
3436 if (status)
3437 return status;
3438 } else {
3439 /* stp is the lock stateid */
3440 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3441 if (status)
3442 return status;
3443 }
3444 }
3445
3446 if (nfs4_check_fh(current_fh, stp)) {
3447 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3448 return nfserr_bad_stateid;
3449 }
3450
3451 /*
3452 * We now validate the seqid and stateid generation numbers.
3453 * For the moment, we ignore the possibility of
3454 * generation number wraparound.
3455 */
3456 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3457 goto check_replay;
3458
3459 if (sop->so_confirmed && flags & CONFIRM) {
3460 dprintk("NFSD: preprocess_seqid_op: expected"
3461 " unconfirmed stateowner!\n");
3462 return nfserr_bad_stateid;
3463 }
3464 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3465 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3466 " confirmed yet!\n");
3467 return nfserr_bad_stateid;
3468 }
3469 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3470 if (status)
3471 return status;
3472 renew_client(sop->so_client);
3473 return nfs_ok;
3474
3475check_replay:
3476 if (seqid == sop->so_seqid - 1) {
3477 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3478 /* indicate replay to calling function */
3479 return nfserr_replay_me;
3480 }
3481 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3482 sop->so_seqid, seqid);
3483 *sopp = NULL;
3484 return nfserr_bad_seqid;
3485}
3486
3487__be32
3488nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3489 struct nfsd4_open_confirm *oc)
3490{
3491 __be32 status;
3492 struct nfs4_stateowner *sop;
3493 struct nfs4_stateid *stp;
3494
3495 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3496 (int)cstate->current_fh.fh_dentry->d_name.len,
3497 cstate->current_fh.fh_dentry->d_name.name);
3498
3499 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3500 if (status)
3501 return status;
3502
3503 nfs4_lock_state();
3504
3505 if ((status = nfs4_preprocess_seqid_op(cstate,
3506 oc->oc_seqid, &oc->oc_req_stateid,
3507 CONFIRM | OPEN_STATE,
3508 &oc->oc_stateowner, &stp, NULL)))
3509 goto out;
3510
3511 sop = oc->oc_stateowner;
3512 sop->so_confirmed = 1;
3513 update_stateid(&stp->st_stateid);
3514 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3515 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3516 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3517
3518 nfsd4_create_clid_dir(sop->so_client);
3519out:
3520 if (oc->oc_stateowner) {
3521 nfs4_get_stateowner(oc->oc_stateowner);
3522 cstate->replay_owner = oc->oc_stateowner;
3523 }
3524 nfs4_unlock_state();
3525 return status;
3526}
3527
3528static inline void nfs4_file_downgrade(struct nfs4_stateid *stp, unsigned int to_access)
3529{
3530 int i;
3531
3532 for (i = 1; i < 4; i++) {
3533 if (test_bit(i, &stp->st_access_bmap) && !(i & to_access)) {
3534 nfs4_file_put_access(stp->st_file, i);
3535 __clear_bit(i, &stp->st_access_bmap);
3536 }
3537 }
3538}
3539
3540static void
3541reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3542{
3543 int i;
3544 for (i = 0; i < 4; i++) {
3545 if ((i & deny) != i)
3546 __clear_bit(i, bmap);
3547 }
3548}
3549
3550__be32
3551nfsd4_open_downgrade(struct svc_rqst *rqstp,
3552 struct nfsd4_compound_state *cstate,
3553 struct nfsd4_open_downgrade *od)
3554{
3555 __be32 status;
3556 struct nfs4_stateid *stp;
3557
3558 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3559 (int)cstate->current_fh.fh_dentry->d_name.len,
3560 cstate->current_fh.fh_dentry->d_name.name);
3561
3562 if (!access_valid(od->od_share_access, cstate->minorversion)
3563 || !deny_valid(od->od_share_deny))
3564 return nfserr_inval;
3565
3566 nfs4_lock_state();
3567 if ((status = nfs4_preprocess_seqid_op(cstate,
3568 od->od_seqid,
3569 &od->od_stateid,
3570 OPEN_STATE,
3571 &od->od_stateowner, &stp, NULL)))
3572 goto out;
3573
3574 status = nfserr_inval;
3575 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3576 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3577 stp->st_access_bmap, od->od_share_access);
3578 goto out;
3579 }
3580 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3581 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3582 stp->st_deny_bmap, od->od_share_deny);
3583 goto out;
3584 }
3585 nfs4_file_downgrade(stp, od->od_share_access);
3586
3587 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3588
3589 update_stateid(&stp->st_stateid);
3590 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3591 status = nfs_ok;
3592out:
3593 if (od->od_stateowner) {
3594 nfs4_get_stateowner(od->od_stateowner);
3595 cstate->replay_owner = od->od_stateowner;
3596 }
3597 nfs4_unlock_state();
3598 return status;
3599}
3600
3601/*
3602 * nfs4_unlock_state() called after encode
3603 */
3604__be32
3605nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3606 struct nfsd4_close *close)
3607{
3608 __be32 status;
3609 struct nfs4_stateid *stp;
3610
3611 dprintk("NFSD: nfsd4_close on file %.*s\n",
3612 (int)cstate->current_fh.fh_dentry->d_name.len,
3613 cstate->current_fh.fh_dentry->d_name.name);
3614
3615 nfs4_lock_state();
3616 /* check close_lru for replay */
3617 if ((status = nfs4_preprocess_seqid_op(cstate,
3618 close->cl_seqid,
3619 &close->cl_stateid,
3620 OPEN_STATE | CLOSE_STATE,
3621 &close->cl_stateowner, &stp, NULL)))
3622 goto out;
3623 status = nfs_ok;
3624 update_stateid(&stp->st_stateid);
3625 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3626
3627 /* release_stateid() calls nfsd_close() if needed */
3628 release_open_stateid(stp);
3629
3630 /* place unused nfs4_stateowners on so_close_lru list to be
3631 * released by the laundromat service after the lease period
3632 * to enable us to handle CLOSE replay
3633 */
3634 if (list_empty(&close->cl_stateowner->so_stateids))
3635 move_to_close_lru(close->cl_stateowner);
3636out:
3637 if (close->cl_stateowner) {
3638 nfs4_get_stateowner(close->cl_stateowner);
3639 cstate->replay_owner = close->cl_stateowner;
3640 }
3641 nfs4_unlock_state();
3642 return status;
3643}
3644
3645__be32
3646nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3647 struct nfsd4_delegreturn *dr)
3648{
3649 struct nfs4_delegation *dp;
3650 stateid_t *stateid = &dr->dr_stateid;
3651 struct inode *inode;
3652 __be32 status;
3653 int flags = 0;
3654
3655 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3656 return status;
3657 inode = cstate->current_fh.fh_dentry->d_inode;
3658
3659 if (nfsd4_has_session(cstate))
3660 flags |= HAS_SESSION;
3661 nfs4_lock_state();
3662 status = nfserr_bad_stateid;
3663 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3664 goto out;
3665 status = nfserr_stale_stateid;
3666 if (STALE_STATEID(stateid))
3667 goto out;
3668 status = nfserr_bad_stateid;
3669 if (!is_delegation_stateid(stateid))
3670 goto out;
3671 status = nfserr_expired;
3672 dp = find_delegation_stateid(inode, stateid);
3673 if (!dp)
3674 goto out;
3675 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3676 if (status)
3677 goto out;
3678 renew_client(dp->dl_client);
3679
3680 unhash_delegation(dp);
3681out:
3682 nfs4_unlock_state();
3683
3684 return status;
3685}
3686
3687
3688/*
3689 * Lock owner state (byte-range locks)
3690 */
3691#define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3692#define LOCK_HASH_BITS 8
3693#define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3694#define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3695
3696static inline u64
3697end_offset(u64 start, u64 len)
3698{
3699 u64 end;
3700
3701 end = start + len;
3702 return end >= start ? end: NFS4_MAX_UINT64;
3703}
3704
3705/* last octet in a range */
3706static inline u64
3707last_byte_offset(u64 start, u64 len)
3708{
3709 u64 end;
3710
3711 BUG_ON(!len);
3712 end = start + len;
3713 return end > start ? end - 1: NFS4_MAX_UINT64;
3714}
3715
3716#define lockownerid_hashval(id) \
3717 ((id) & LOCK_HASH_MASK)
3718
3719static inline unsigned int
3720lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3721 struct xdr_netobj *ownername)
3722{
3723 return (file_hashval(inode) + cl_id
3724 + opaque_hashval(ownername->data, ownername->len))
3725 & LOCK_HASH_MASK;
3726}
3727
3728static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3729static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3730static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3731
3732static int
3733same_stateid(stateid_t *id_one, stateid_t *id_two)
3734{
3735 if (id_one->si_stateownerid != id_two->si_stateownerid)
3736 return 0;
3737 return id_one->si_fileid == id_two->si_fileid;
3738}
3739
3740static struct nfs4_stateid *
3741find_stateid(stateid_t *stid, int flags)
3742{
3743 struct nfs4_stateid *local;
3744 u32 st_id = stid->si_stateownerid;
3745 u32 f_id = stid->si_fileid;
3746 unsigned int hashval;
3747
3748 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3749 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3750 hashval = stateid_hashval(st_id, f_id);
3751 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3752 if ((local->st_stateid.si_stateownerid == st_id) &&
3753 (local->st_stateid.si_fileid == f_id))
3754 return local;
3755 }
3756 }
3757
3758 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3759 hashval = stateid_hashval(st_id, f_id);
3760 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3761 if ((local->st_stateid.si_stateownerid == st_id) &&
3762 (local->st_stateid.si_fileid == f_id))
3763 return local;
3764 }
3765 }
3766 return NULL;
3767}
3768
3769static struct nfs4_stateid *
3770search_for_stateid(stateid_t *stid)
3771{
3772 struct nfs4_stateid *local;
3773 unsigned int hashval = stateid_hashval(stid->si_stateownerid, stid->si_fileid);
3774
3775 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3776 if (same_stateid(&local->st_stateid, stid))
3777 return local;
3778 }
3779
3780 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3781 if (same_stateid(&local->st_stateid, stid))
3782 return local;
3783 }
3784 return NULL;
3785}
3786
3787static struct nfs4_delegation *
3788search_for_delegation(stateid_t *stid)
3789{
3790 struct nfs4_file *fp;
3791 struct nfs4_delegation *dp;
3792 struct list_head *pos;
3793 int i;
3794
3795 for (i = 0; i < FILE_HASH_SIZE; i++) {
3796 list_for_each_entry(fp, &file_hashtbl[i], fi_hash) {
3797 list_for_each(pos, &fp->fi_delegations) {
3798 dp = list_entry(pos, struct nfs4_delegation, dl_perfile);
3799 if (same_stateid(&dp->dl_stateid, stid))
3800 return dp;
3801 }
3802 }
3803 }
3804 return NULL;
3805}
3806
3807static struct nfs4_delegation *
3808find_delegation_stateid(struct inode *ino, stateid_t *stid)
3809{
3810 struct nfs4_file *fp;
3811 struct nfs4_delegation *dl;
3812
3813 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3814 STATEID_VAL(stid));
3815
3816 fp = find_file(ino);
3817 if (!fp)
3818 return NULL;
3819 dl = find_delegation_file(fp, stid);
3820 put_nfs4_file(fp);
3821 return dl;
3822}
3823
3824/*
3825 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3826 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3827 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3828 * locking, this prevents us from being completely protocol-compliant. The
3829 * real solution to this problem is to start using unsigned file offsets in
3830 * the VFS, but this is a very deep change!
3831 */
3832static inline void
3833nfs4_transform_lock_offset(struct file_lock *lock)
3834{
3835 if (lock->fl_start < 0)
3836 lock->fl_start = OFFSET_MAX;
3837 if (lock->fl_end < 0)
3838 lock->fl_end = OFFSET_MAX;
3839}
3840
3841/* Hack!: For now, we're defining this just so we can use a pointer to it
3842 * as a unique cookie to identify our (NFSv4's) posix locks. */
3843static const struct lock_manager_operations nfsd_posix_mng_ops = {
3844};
3845
3846static inline void
3847nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3848{
3849 struct nfs4_stateowner *sop;
3850
3851 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3852 sop = (struct nfs4_stateowner *) fl->fl_owner;
3853 kref_get(&sop->so_ref);
3854 deny->ld_sop = sop;
3855 deny->ld_clientid = sop->so_client->cl_clientid;
3856 } else {
3857 deny->ld_sop = NULL;
3858 deny->ld_clientid.cl_boot = 0;
3859 deny->ld_clientid.cl_id = 0;
3860 }
3861 deny->ld_start = fl->fl_start;
3862 deny->ld_length = NFS4_MAX_UINT64;
3863 if (fl->fl_end != NFS4_MAX_UINT64)
3864 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3865 deny->ld_type = NFS4_READ_LT;
3866 if (fl->fl_type != F_RDLCK)
3867 deny->ld_type = NFS4_WRITE_LT;
3868}
3869
3870static struct nfs4_stateowner *
3871find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3872 struct xdr_netobj *owner)
3873{
3874 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3875 struct nfs4_stateowner *op;
3876
3877 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3878 if (same_owner_str(op, owner, clid))
3879 return op;
3880 }
3881 return NULL;
3882}
3883
3884/*
3885 * Alloc a lock owner structure.
3886 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3887 * occurred.
3888 *
3889 * strhashval = lock_ownerstr_hashval
3890 */
3891
3892static struct nfs4_stateowner *
3893alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3894 struct nfs4_stateowner *sop;
3895 struct nfs4_replay *rp;
3896 unsigned int idhashval;
3897
3898 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3899 return NULL;
3900 idhashval = lockownerid_hashval(current_ownerid);
3901 INIT_LIST_HEAD(&sop->so_idhash);
3902 INIT_LIST_HEAD(&sop->so_strhash);
3903 INIT_LIST_HEAD(&sop->so_perclient);
3904 INIT_LIST_HEAD(&sop->so_stateids);
3905 INIT_LIST_HEAD(&sop->so_perstateid);
3906 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3907 sop->so_time = 0;
3908 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3909 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3910 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3911 sop->so_is_open_owner = 0;
3912 sop->so_id = current_ownerid++;
3913 sop->so_client = clp;
3914 /* It is the openowner seqid that will be incremented in encode in the
3915 * case of new lockowners; so increment the lock seqid manually: */
3916 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3917 sop->so_confirmed = 1;
3918 rp = &sop->so_replay;
3919 rp->rp_status = nfserr_serverfault;
3920 rp->rp_buflen = 0;
3921 rp->rp_buf = rp->rp_ibuf;
3922 return sop;
3923}
3924
3925static struct nfs4_stateid *
3926alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3927{
3928 struct nfs4_stateid *stp;
3929 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3930
3931 stp = nfs4_alloc_stateid();
3932 if (stp == NULL)
3933 goto out;
3934 INIT_LIST_HEAD(&stp->st_hash);
3935 INIT_LIST_HEAD(&stp->st_perfile);
3936 INIT_LIST_HEAD(&stp->st_perstateowner);
3937 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3938 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3939 list_add(&stp->st_perfile, &fp->fi_stateids);
3940 list_add(&stp->st_perstateowner, &sop->so_stateids);
3941 stp->st_stateowner = sop;
3942 get_nfs4_file(fp);
3943 stp->st_file = fp;
3944 stp->st_stateid.si_boot = boot_time;
3945 stp->st_stateid.si_stateownerid = sop->so_id;
3946 stp->st_stateid.si_fileid = fp->fi_id;
3947 stp->st_stateid.si_generation = 0;
3948 stp->st_access_bmap = 0;
3949 stp->st_deny_bmap = open_stp->st_deny_bmap;
3950 stp->st_openstp = open_stp;
3951
3952out:
3953 return stp;
3954}
3955
3956static int
3957check_lock_length(u64 offset, u64 length)
3958{
3959 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3960 LOFF_OVERFLOW(offset, length)));
3961}
3962
3963static void get_lock_access(struct nfs4_stateid *lock_stp, u32 access)
3964{
3965 struct nfs4_file *fp = lock_stp->st_file;
3966 int oflag = nfs4_access_to_omode(access);
3967
3968 if (test_bit(access, &lock_stp->st_access_bmap))
3969 return;
3970 nfs4_file_get_access(fp, oflag);
3971 __set_bit(access, &lock_stp->st_access_bmap);
3972}
3973
3974/*
3975 * LOCK operation
3976 */
3977__be32
3978nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3979 struct nfsd4_lock *lock)
3980{
3981 struct nfs4_stateowner *open_sop = NULL;
3982 struct nfs4_stateowner *lock_sop = NULL;
3983 struct nfs4_stateid *lock_stp;
3984 struct nfs4_file *fp;
3985 struct file *filp = NULL;
3986 struct file_lock file_lock;
3987 struct file_lock conflock;
3988 __be32 status = 0;
3989 unsigned int strhashval;
3990 int err;
3991
3992 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3993 (long long) lock->lk_offset,
3994 (long long) lock->lk_length);
3995
3996 if (check_lock_length(lock->lk_offset, lock->lk_length))
3997 return nfserr_inval;
3998
3999 if ((status = fh_verify(rqstp, &cstate->current_fh,
4000 S_IFREG, NFSD_MAY_LOCK))) {
4001 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4002 return status;
4003 }
4004
4005 nfs4_lock_state();
4006
4007 if (lock->lk_is_new) {
4008 /*
4009 * Client indicates that this is a new lockowner.
4010 * Use open owner and open stateid to create lock owner and
4011 * lock stateid.
4012 */
4013 struct nfs4_stateid *open_stp = NULL;
4014
4015 status = nfserr_stale_clientid;
4016 if (!nfsd4_has_session(cstate) &&
4017 STALE_CLIENTID(&lock->lk_new_clientid))
4018 goto out;
4019
4020 /* validate and update open stateid and open seqid */
4021 status = nfs4_preprocess_seqid_op(cstate,
4022 lock->lk_new_open_seqid,
4023 &lock->lk_new_open_stateid,
4024 OPEN_STATE,
4025 &lock->lk_replay_owner, &open_stp,
4026 lock);
4027 if (status)
4028 goto out;
4029 open_sop = lock->lk_replay_owner;
4030 /* create lockowner and lock stateid */
4031 fp = open_stp->st_file;
4032 strhashval = lock_ownerstr_hashval(fp->fi_inode,
4033 open_sop->so_client->cl_clientid.cl_id,
4034 &lock->v.new.owner);
4035 /* XXX: Do we need to check for duplicate stateowners on
4036 * the same file, or should they just be allowed (and
4037 * create new stateids)? */
4038 status = nfserr_resource;
4039 lock_sop = alloc_init_lock_stateowner(strhashval,
4040 open_sop->so_client, open_stp, lock);
4041 if (lock_sop == NULL)
4042 goto out;
4043 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
4044 if (lock_stp == NULL)
4045 goto out;
4046 } else {
4047 /* lock (lock owner + lock stateid) already exists */
4048 status = nfs4_preprocess_seqid_op(cstate,
4049 lock->lk_old_lock_seqid,
4050 &lock->lk_old_lock_stateid,
4051 LOCK_STATE,
4052 &lock->lk_replay_owner, &lock_stp, lock);
4053 if (status)
4054 goto out;
4055 lock_sop = lock->lk_replay_owner;
4056 fp = lock_stp->st_file;
4057 }
4058 /* lock->lk_replay_owner and lock_stp have been created or found */
4059
4060 status = nfserr_grace;
4061 if (locks_in_grace() && !lock->lk_reclaim)
4062 goto out;
4063 status = nfserr_no_grace;
4064 if (!locks_in_grace() && lock->lk_reclaim)
4065 goto out;
4066
4067 locks_init_lock(&file_lock);
4068 switch (lock->lk_type) {
4069 case NFS4_READ_LT:
4070 case NFS4_READW_LT:
4071 filp = find_readable_file(lock_stp->st_file);
4072 if (filp)
4073 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4074 file_lock.fl_type = F_RDLCK;
4075 break;
4076 case NFS4_WRITE_LT:
4077 case NFS4_WRITEW_LT:
4078 filp = find_writeable_file(lock_stp->st_file);
4079 if (filp)
4080 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4081 file_lock.fl_type = F_WRLCK;
4082 break;
4083 default:
4084 status = nfserr_inval;
4085 goto out;
4086 }
4087 if (!filp) {
4088 status = nfserr_openmode;
4089 goto out;
4090 }
4091 file_lock.fl_owner = (fl_owner_t)lock_sop;
4092 file_lock.fl_pid = current->tgid;
4093 file_lock.fl_file = filp;
4094 file_lock.fl_flags = FL_POSIX;
4095 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4096
4097 file_lock.fl_start = lock->lk_offset;
4098 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4099 nfs4_transform_lock_offset(&file_lock);
4100
4101 /*
4102 * Try to lock the file in the VFS.
4103 * Note: locks.c uses the BKL to protect the inode's lock list.
4104 */
4105
4106 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4107 switch (-err) {
4108 case 0: /* success! */
4109 update_stateid(&lock_stp->st_stateid);
4110 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
4111 sizeof(stateid_t));
4112 status = 0;
4113 break;
4114 case (EAGAIN): /* conflock holds conflicting lock */
4115 status = nfserr_denied;
4116 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4117 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4118 break;
4119 case (EDEADLK):
4120 status = nfserr_deadlock;
4121 break;
4122 default:
4123 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4124 status = nfserr_resource;
4125 break;
4126 }
4127out:
4128 if (status && lock->lk_is_new && lock_sop)
4129 release_lockowner(lock_sop);
4130 if (lock->lk_replay_owner) {
4131 nfs4_get_stateowner(lock->lk_replay_owner);
4132 cstate->replay_owner = lock->lk_replay_owner;
4133 }
4134 nfs4_unlock_state();
4135 return status;
4136}
4137
4138/*
4139 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4140 * so we do a temporary open here just to get an open file to pass to
4141 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
4142 * inode operation.)
4143 */
4144static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4145{
4146 struct file *file;
4147 int err;
4148
4149 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4150 if (err)
4151 return err;
4152 err = vfs_test_lock(file, lock);
4153 nfsd_close(file);
4154 return err;
4155}
4156
4157/*
4158 * LOCKT operation
4159 */
4160__be32
4161nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4162 struct nfsd4_lockt *lockt)
4163{
4164 struct inode *inode;
4165 struct file_lock file_lock;
4166 int error;
4167 __be32 status;
4168
4169 if (locks_in_grace())
4170 return nfserr_grace;
4171
4172 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4173 return nfserr_inval;
4174
4175 lockt->lt_stateowner = NULL;
4176 nfs4_lock_state();
4177
4178 status = nfserr_stale_clientid;
4179 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4180 goto out;
4181
4182 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
4183 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
4184 if (status == nfserr_symlink)
4185 status = nfserr_inval;
4186 goto out;
4187 }
4188
4189 inode = cstate->current_fh.fh_dentry->d_inode;
4190 locks_init_lock(&file_lock);
4191 switch (lockt->lt_type) {
4192 case NFS4_READ_LT:
4193 case NFS4_READW_LT:
4194 file_lock.fl_type = F_RDLCK;
4195 break;
4196 case NFS4_WRITE_LT:
4197 case NFS4_WRITEW_LT:
4198 file_lock.fl_type = F_WRLCK;
4199 break;
4200 default:
4201 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4202 status = nfserr_inval;
4203 goto out;
4204 }
4205
4206 lockt->lt_stateowner = find_lockstateowner_str(inode,
4207 &lockt->lt_clientid, &lockt->lt_owner);
4208 if (lockt->lt_stateowner)
4209 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
4210 file_lock.fl_pid = current->tgid;
4211 file_lock.fl_flags = FL_POSIX;
4212
4213 file_lock.fl_start = lockt->lt_offset;
4214 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4215
4216 nfs4_transform_lock_offset(&file_lock);
4217
4218 status = nfs_ok;
4219 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4220 if (error) {
4221 status = nfserrno(error);
4222 goto out;
4223 }
4224 if (file_lock.fl_type != F_UNLCK) {
4225 status = nfserr_denied;
4226 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4227 }
4228out:
4229 nfs4_unlock_state();
4230 return status;
4231}
4232
4233__be32
4234nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4235 struct nfsd4_locku *locku)
4236{
4237 struct nfs4_stateid *stp;
4238 struct file *filp = NULL;
4239 struct file_lock file_lock;
4240 __be32 status;
4241 int err;
4242
4243 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4244 (long long) locku->lu_offset,
4245 (long long) locku->lu_length);
4246
4247 if (check_lock_length(locku->lu_offset, locku->lu_length))
4248 return nfserr_inval;
4249
4250 nfs4_lock_state();
4251
4252 if ((status = nfs4_preprocess_seqid_op(cstate,
4253 locku->lu_seqid,
4254 &locku->lu_stateid,
4255 LOCK_STATE,
4256 &locku->lu_stateowner, &stp, NULL)))
4257 goto out;
4258
4259 filp = find_any_file(stp->st_file);
4260 if (!filp) {
4261 status = nfserr_lock_range;
4262 goto out;
4263 }
4264 BUG_ON(!filp);
4265 locks_init_lock(&file_lock);
4266 file_lock.fl_type = F_UNLCK;
4267 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
4268 file_lock.fl_pid = current->tgid;
4269 file_lock.fl_file = filp;
4270 file_lock.fl_flags = FL_POSIX;
4271 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4272 file_lock.fl_start = locku->lu_offset;
4273
4274 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4275 nfs4_transform_lock_offset(&file_lock);
4276
4277 /*
4278 * Try to unlock the file in the VFS.
4279 */
4280 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4281 if (err) {
4282 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4283 goto out_nfserr;
4284 }
4285 /*
4286 * OK, unlock succeeded; the only thing left to do is update the stateid.
4287 */
4288 update_stateid(&stp->st_stateid);
4289 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4290
4291out:
4292 if (locku->lu_stateowner) {
4293 nfs4_get_stateowner(locku->lu_stateowner);
4294 cstate->replay_owner = locku->lu_stateowner;
4295 }
4296 nfs4_unlock_state();
4297 return status;
4298
4299out_nfserr:
4300 status = nfserrno(err);
4301 goto out;
4302}
4303
4304/*
4305 * returns
4306 * 1: locks held by lockowner
4307 * 0: no locks held by lockowner
4308 */
4309static int
4310check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4311{
4312 struct file_lock **flpp;
4313 struct inode *inode = filp->fi_inode;
4314 int status = 0;
4315
4316 lock_flocks();
4317 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4318 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4319 status = 1;
4320 goto out;
4321 }
4322 }
4323out:
4324 unlock_flocks();
4325 return status;
4326}
4327
4328__be32
4329nfsd4_release_lockowner(struct svc_rqst *rqstp,
4330 struct nfsd4_compound_state *cstate,
4331 struct nfsd4_release_lockowner *rlockowner)
4332{
4333 clientid_t *clid = &rlockowner->rl_clientid;
4334 struct nfs4_stateowner *sop;
4335 struct nfs4_stateid *stp;
4336 struct xdr_netobj *owner = &rlockowner->rl_owner;
4337 struct list_head matches;
4338 int i;
4339 __be32 status;
4340
4341 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4342 clid->cl_boot, clid->cl_id);
4343
4344 /* XXX check for lease expiration */
4345
4346 status = nfserr_stale_clientid;
4347 if (STALE_CLIENTID(clid))
4348 return status;
4349
4350 nfs4_lock_state();
4351
4352 status = nfserr_locks_held;
4353 /* XXX: we're doing a linear search through all the lockowners.
4354 * Yipes! For now we'll just hope clients aren't really using
4355 * release_lockowner much, but eventually we have to fix these
4356 * data structures. */
4357 INIT_LIST_HEAD(&matches);
4358 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4359 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4360 if (!same_owner_str(sop, owner, clid))
4361 continue;
4362 list_for_each_entry(stp, &sop->so_stateids,
4363 st_perstateowner) {
4364 if (check_for_locks(stp->st_file, sop))
4365 goto out;
4366 /* Note: so_perclient unused for lockowners,
4367 * so it's OK to fool with here. */
4368 list_add(&sop->so_perclient, &matches);
4369 }
4370 }
4371 }
4372 /* Clients probably won't expect us to return with some (but not all)
4373 * of the lockowner state released; so don't release any until all
4374 * have been checked. */
4375 status = nfs_ok;
4376 while (!list_empty(&matches)) {
4377 sop = list_entry(matches.next, struct nfs4_stateowner,
4378 so_perclient);
4379 /* unhash_stateowner deletes so_perclient only
4380 * for openowners. */
4381 list_del(&sop->so_perclient);
4382 release_lockowner(sop);
4383 }
4384out:
4385 nfs4_unlock_state();
4386 return status;
4387}
4388
4389static inline struct nfs4_client_reclaim *
4390alloc_reclaim(void)
4391{
4392 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4393}
4394
4395int
4396nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4397{
4398 unsigned int strhashval = clientstr_hashval(name);
4399 struct nfs4_client *clp;
4400
4401 clp = find_confirmed_client_by_str(name, strhashval);
4402 return clp ? 1 : 0;
4403}
4404
4405/*
4406 * failure => all reset bets are off, nfserr_no_grace...
4407 */
4408int
4409nfs4_client_to_reclaim(const char *name)
4410{
4411 unsigned int strhashval;
4412 struct nfs4_client_reclaim *crp = NULL;
4413
4414 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4415 crp = alloc_reclaim();
4416 if (!crp)
4417 return 0;
4418 strhashval = clientstr_hashval(name);
4419 INIT_LIST_HEAD(&crp->cr_strhash);
4420 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4421 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4422 reclaim_str_hashtbl_size++;
4423 return 1;
4424}
4425
4426static void
4427nfs4_release_reclaim(void)
4428{
4429 struct nfs4_client_reclaim *crp = NULL;
4430 int i;
4431
4432 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4433 while (!list_empty(&reclaim_str_hashtbl[i])) {
4434 crp = list_entry(reclaim_str_hashtbl[i].next,
4435 struct nfs4_client_reclaim, cr_strhash);
4436 list_del(&crp->cr_strhash);
4437 kfree(crp);
4438 reclaim_str_hashtbl_size--;
4439 }
4440 }
4441 BUG_ON(reclaim_str_hashtbl_size);
4442}
4443
4444/*
4445 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4446static struct nfs4_client_reclaim *
4447nfs4_find_reclaim_client(clientid_t *clid)
4448{
4449 unsigned int strhashval;
4450 struct nfs4_client *clp;
4451 struct nfs4_client_reclaim *crp = NULL;
4452
4453
4454 /* find clientid in conf_id_hashtbl */
4455 clp = find_confirmed_client(clid);
4456 if (clp == NULL)
4457 return NULL;
4458
4459 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4460 clp->cl_name.len, clp->cl_name.data,
4461 clp->cl_recdir);
4462
4463 /* find clp->cl_name in reclaim_str_hashtbl */
4464 strhashval = clientstr_hashval(clp->cl_recdir);
4465 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4466 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4467 return crp;
4468 }
4469 }
4470 return NULL;
4471}
4472
4473/*
4474* Called from OPEN. Look for clientid in reclaim list.
4475*/
4476__be32
4477nfs4_check_open_reclaim(clientid_t *clid)
4478{
4479 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4480}
4481
4482/* initialization to perform at module load time: */
4483
4484int
4485nfs4_state_init(void)
4486{
4487 int i, status;
4488
4489 status = nfsd4_init_slabs();
4490 if (status)
4491 return status;
4492 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4493 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4494 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4495 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4496 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4497 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4498 }
4499 for (i = 0; i < SESSION_HASH_SIZE; i++)
4500 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4501 for (i = 0; i < FILE_HASH_SIZE; i++) {
4502 INIT_LIST_HEAD(&file_hashtbl[i]);
4503 }
4504 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4505 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4506 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4507 }
4508 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4509 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4510 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4511 }
4512 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4513 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4514 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4515 }
4516 memset(&onestateid, ~0, sizeof(stateid_t));
4517 INIT_LIST_HEAD(&close_lru);
4518 INIT_LIST_HEAD(&client_lru);
4519 INIT_LIST_HEAD(&del_recall_lru);
4520 reclaim_str_hashtbl_size = 0;
4521 return 0;
4522}
4523
4524static void
4525nfsd4_load_reboot_recovery_data(void)
4526{
4527 int status;
4528
4529 nfs4_lock_state();
4530 nfsd4_init_recdir(user_recovery_dirname);
4531 status = nfsd4_recdir_load();
4532 nfs4_unlock_state();
4533 if (status)
4534 printk("NFSD: Failure reading reboot recovery data\n");
4535}
4536
4537/*
4538 * Since the lifetime of a delegation isn't limited to that of an open, a
4539 * client may quite reasonably hang on to a delegation as long as it has
4540 * the inode cached. This becomes an obvious problem the first time a
4541 * client's inode cache approaches the size of the server's total memory.
4542 *
4543 * For now we avoid this problem by imposing a hard limit on the number
4544 * of delegations, which varies according to the server's memory size.
4545 */
4546static void
4547set_max_delegations(void)
4548{
4549 /*
4550 * Allow at most 4 delegations per megabyte of RAM. Quick
4551 * estimates suggest that in the worst case (where every delegation
4552 * is for a different inode), a delegation could take about 1.5K,
4553 * giving a worst case usage of about 6% of memory.
4554 */
4555 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4556}
4557
4558/* initialization to perform when the nfsd service is started: */
4559
4560static int
4561__nfs4_state_start(void)
4562{
4563 int ret;
4564
4565 boot_time = get_seconds();
4566 locks_start_grace(&nfsd4_manager);
4567 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4568 nfsd4_grace);
4569 ret = set_callback_cred();
4570 if (ret)
4571 return -ENOMEM;
4572 laundry_wq = create_singlethread_workqueue("nfsd4");
4573 if (laundry_wq == NULL)
4574 return -ENOMEM;
4575 ret = nfsd4_create_callback_queue();
4576 if (ret)
4577 goto out_free_laundry;
4578 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4579 set_max_delegations();
4580 return 0;
4581out_free_laundry:
4582 destroy_workqueue(laundry_wq);
4583 return ret;
4584}
4585
4586int
4587nfs4_state_start(void)
4588{
4589 nfsd4_load_reboot_recovery_data();
4590 return __nfs4_state_start();
4591}
4592
4593static void
4594__nfs4_state_shutdown(void)
4595{
4596 int i;
4597 struct nfs4_client *clp = NULL;
4598 struct nfs4_delegation *dp = NULL;
4599 struct list_head *pos, *next, reaplist;
4600
4601 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4602 while (!list_empty(&conf_id_hashtbl[i])) {
4603 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4604 expire_client(clp);
4605 }
4606 while (!list_empty(&unconf_str_hashtbl[i])) {
4607 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4608 expire_client(clp);
4609 }
4610 }
4611 INIT_LIST_HEAD(&reaplist);
4612 spin_lock(&recall_lock);
4613 list_for_each_safe(pos, next, &del_recall_lru) {
4614 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4615 list_move(&dp->dl_recall_lru, &reaplist);
4616 }
4617 spin_unlock(&recall_lock);
4618 list_for_each_safe(pos, next, &reaplist) {
4619 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4620 list_del_init(&dp->dl_recall_lru);
4621 unhash_delegation(dp);
4622 }
4623
4624 nfsd4_shutdown_recdir();
4625}
4626
4627void
4628nfs4_state_shutdown(void)
4629{
4630 cancel_delayed_work_sync(&laundromat_work);
4631 destroy_workqueue(laundry_wq);
4632 locks_end_grace(&nfsd4_manager);
4633 nfs4_lock_state();
4634 nfs4_release_reclaim();
4635 __nfs4_state_shutdown();
4636 nfs4_unlock_state();
4637 nfsd4_destroy_callback_queue();
4638}
4639
4640/*
4641 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4642 * accessed when nfsd is starting.
4643 */
4644static void
4645nfs4_set_recdir(char *recdir)
4646{
4647 strcpy(user_recovery_dirname, recdir);
4648}
4649
4650/*
4651 * Change the NFSv4 recovery directory to recdir.
4652 */
4653int
4654nfs4_reset_recoverydir(char *recdir)
4655{
4656 int status;
4657 struct path path;
4658
4659 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4660 if (status)
4661 return status;
4662 status = -ENOTDIR;
4663 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4664 nfs4_set_recdir(recdir);
4665 status = 0;
4666 }
4667 path_put(&path);
4668 return status;
4669}
4670
4671char *
4672nfs4_recoverydir(void)
4673{
4674 return user_recovery_dirname;
4675}
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/nfs_ssc.h>
48#include "xdr4.h"
49#include "xdr4cb.h"
50#include "vfs.h"
51#include "current_stateid.h"
52
53#include "netns.h"
54#include "pnfs.h"
55#include "filecache.h"
56#include "trace.h"
57
58#define NFSDDBG_FACILITY NFSDDBG_PROC
59
60#define all_ones {{~0,~0},~0}
61static const stateid_t one_stateid = {
62 .si_generation = ~0,
63 .si_opaque = all_ones,
64};
65static const stateid_t zero_stateid = {
66 /* all fields zero */
67};
68static const stateid_t currentstateid = {
69 .si_generation = 1,
70};
71static const stateid_t close_stateid = {
72 .si_generation = 0xffffffffU,
73};
74
75static u64 current_sessionid = 1;
76
77#define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
78#define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
79#define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
80#define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
81
82/* forward declarations */
83static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
84static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
85void nfsd4_end_grace(struct nfsd_net *nn);
86static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
87
88/* Locking: */
89
90/*
91 * Currently used for the del_recall_lru and file hash table. In an
92 * effort to decrease the scope of the client_mutex, this spinlock may
93 * eventually cover more:
94 */
95static DEFINE_SPINLOCK(state_lock);
96
97enum nfsd4_st_mutex_lock_subclass {
98 OPEN_STATEID_MUTEX = 0,
99 LOCK_STATEID_MUTEX = 1,
100};
101
102/*
103 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
104 * the refcount on the open stateid to drop.
105 */
106static DECLARE_WAIT_QUEUE_HEAD(close_wq);
107
108/*
109 * A waitqueue where a writer to clients/#/ctl destroying a client can
110 * wait for cl_rpc_users to drop to 0 and then for the client to be
111 * unhashed.
112 */
113static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
114
115static struct kmem_cache *client_slab;
116static struct kmem_cache *openowner_slab;
117static struct kmem_cache *lockowner_slab;
118static struct kmem_cache *file_slab;
119static struct kmem_cache *stateid_slab;
120static struct kmem_cache *deleg_slab;
121static struct kmem_cache *odstate_slab;
122
123static void free_session(struct nfsd4_session *);
124
125static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
126static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
127
128static bool is_session_dead(struct nfsd4_session *ses)
129{
130 return ses->se_flags & NFS4_SESSION_DEAD;
131}
132
133static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
134{
135 if (atomic_read(&ses->se_ref) > ref_held_by_me)
136 return nfserr_jukebox;
137 ses->se_flags |= NFS4_SESSION_DEAD;
138 return nfs_ok;
139}
140
141static bool is_client_expired(struct nfs4_client *clp)
142{
143 return clp->cl_time == 0;
144}
145
146static __be32 get_client_locked(struct nfs4_client *clp)
147{
148 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
149
150 lockdep_assert_held(&nn->client_lock);
151
152 if (is_client_expired(clp))
153 return nfserr_expired;
154 atomic_inc(&clp->cl_rpc_users);
155 return nfs_ok;
156}
157
158/* must be called under the client_lock */
159static inline void
160renew_client_locked(struct nfs4_client *clp)
161{
162 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
163
164 if (is_client_expired(clp)) {
165 WARN_ON(1);
166 printk("%s: client (clientid %08x/%08x) already expired\n",
167 __func__,
168 clp->cl_clientid.cl_boot,
169 clp->cl_clientid.cl_id);
170 return;
171 }
172
173 list_move_tail(&clp->cl_lru, &nn->client_lru);
174 clp->cl_time = ktime_get_boottime_seconds();
175}
176
177static void put_client_renew_locked(struct nfs4_client *clp)
178{
179 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
180
181 lockdep_assert_held(&nn->client_lock);
182
183 if (!atomic_dec_and_test(&clp->cl_rpc_users))
184 return;
185 if (!is_client_expired(clp))
186 renew_client_locked(clp);
187 else
188 wake_up_all(&expiry_wq);
189}
190
191static void put_client_renew(struct nfs4_client *clp)
192{
193 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
194
195 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
196 return;
197 if (!is_client_expired(clp))
198 renew_client_locked(clp);
199 else
200 wake_up_all(&expiry_wq);
201 spin_unlock(&nn->client_lock);
202}
203
204static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
205{
206 __be32 status;
207
208 if (is_session_dead(ses))
209 return nfserr_badsession;
210 status = get_client_locked(ses->se_client);
211 if (status)
212 return status;
213 atomic_inc(&ses->se_ref);
214 return nfs_ok;
215}
216
217static void nfsd4_put_session_locked(struct nfsd4_session *ses)
218{
219 struct nfs4_client *clp = ses->se_client;
220 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
221
222 lockdep_assert_held(&nn->client_lock);
223
224 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
225 free_session(ses);
226 put_client_renew_locked(clp);
227}
228
229static void nfsd4_put_session(struct nfsd4_session *ses)
230{
231 struct nfs4_client *clp = ses->se_client;
232 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
233
234 spin_lock(&nn->client_lock);
235 nfsd4_put_session_locked(ses);
236 spin_unlock(&nn->client_lock);
237}
238
239static struct nfsd4_blocked_lock *
240find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
241 struct nfsd_net *nn)
242{
243 struct nfsd4_blocked_lock *cur, *found = NULL;
244
245 spin_lock(&nn->blocked_locks_lock);
246 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
247 if (fh_match(fh, &cur->nbl_fh)) {
248 list_del_init(&cur->nbl_list);
249 list_del_init(&cur->nbl_lru);
250 found = cur;
251 break;
252 }
253 }
254 spin_unlock(&nn->blocked_locks_lock);
255 if (found)
256 locks_delete_block(&found->nbl_lock);
257 return found;
258}
259
260static struct nfsd4_blocked_lock *
261find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
262 struct nfsd_net *nn)
263{
264 struct nfsd4_blocked_lock *nbl;
265
266 nbl = find_blocked_lock(lo, fh, nn);
267 if (!nbl) {
268 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
269 if (nbl) {
270 INIT_LIST_HEAD(&nbl->nbl_list);
271 INIT_LIST_HEAD(&nbl->nbl_lru);
272 fh_copy_shallow(&nbl->nbl_fh, fh);
273 locks_init_lock(&nbl->nbl_lock);
274 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
275 &nfsd4_cb_notify_lock_ops,
276 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
277 }
278 }
279 return nbl;
280}
281
282static void
283free_blocked_lock(struct nfsd4_blocked_lock *nbl)
284{
285 locks_delete_block(&nbl->nbl_lock);
286 locks_release_private(&nbl->nbl_lock);
287 kfree(nbl);
288}
289
290static void
291remove_blocked_locks(struct nfs4_lockowner *lo)
292{
293 struct nfs4_client *clp = lo->lo_owner.so_client;
294 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
295 struct nfsd4_blocked_lock *nbl;
296 LIST_HEAD(reaplist);
297
298 /* Dequeue all blocked locks */
299 spin_lock(&nn->blocked_locks_lock);
300 while (!list_empty(&lo->lo_blocked)) {
301 nbl = list_first_entry(&lo->lo_blocked,
302 struct nfsd4_blocked_lock,
303 nbl_list);
304 list_del_init(&nbl->nbl_list);
305 list_move(&nbl->nbl_lru, &reaplist);
306 }
307 spin_unlock(&nn->blocked_locks_lock);
308
309 /* Now free them */
310 while (!list_empty(&reaplist)) {
311 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
312 nbl_lru);
313 list_del_init(&nbl->nbl_lru);
314 free_blocked_lock(nbl);
315 }
316}
317
318static void
319nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
320{
321 struct nfsd4_blocked_lock *nbl = container_of(cb,
322 struct nfsd4_blocked_lock, nbl_cb);
323 locks_delete_block(&nbl->nbl_lock);
324}
325
326static int
327nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
328{
329 /*
330 * Since this is just an optimization, we don't try very hard if it
331 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
332 * just quit trying on anything else.
333 */
334 switch (task->tk_status) {
335 case -NFS4ERR_DELAY:
336 rpc_delay(task, 1 * HZ);
337 return 0;
338 default:
339 return 1;
340 }
341}
342
343static void
344nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
345{
346 struct nfsd4_blocked_lock *nbl = container_of(cb,
347 struct nfsd4_blocked_lock, nbl_cb);
348
349 free_blocked_lock(nbl);
350}
351
352static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
353 .prepare = nfsd4_cb_notify_lock_prepare,
354 .done = nfsd4_cb_notify_lock_done,
355 .release = nfsd4_cb_notify_lock_release,
356};
357
358/*
359 * We store the NONE, READ, WRITE, and BOTH bits separately in the
360 * st_{access,deny}_bmap field of the stateid, in order to track not
361 * only what share bits are currently in force, but also what
362 * combinations of share bits previous opens have used. This allows us
363 * to enforce the recommendation of rfc 3530 14.2.19 that the server
364 * return an error if the client attempt to downgrade to a combination
365 * of share bits not explicable by closing some of its previous opens.
366 *
367 * XXX: This enforcement is actually incomplete, since we don't keep
368 * track of access/deny bit combinations; so, e.g., we allow:
369 *
370 * OPEN allow read, deny write
371 * OPEN allow both, deny none
372 * DOWNGRADE allow read, deny none
373 *
374 * which we should reject.
375 */
376static unsigned int
377bmap_to_share_mode(unsigned long bmap)
378{
379 int i;
380 unsigned int access = 0;
381
382 for (i = 1; i < 4; i++) {
383 if (test_bit(i, &bmap))
384 access |= i;
385 }
386 return access;
387}
388
389/* set share access for a given stateid */
390static inline void
391set_access(u32 access, struct nfs4_ol_stateid *stp)
392{
393 unsigned char mask = 1 << access;
394
395 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
396 stp->st_access_bmap |= mask;
397}
398
399/* clear share access for a given stateid */
400static inline void
401clear_access(u32 access, struct nfs4_ol_stateid *stp)
402{
403 unsigned char mask = 1 << access;
404
405 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
406 stp->st_access_bmap &= ~mask;
407}
408
409/* test whether a given stateid has access */
410static inline bool
411test_access(u32 access, struct nfs4_ol_stateid *stp)
412{
413 unsigned char mask = 1 << access;
414
415 return (bool)(stp->st_access_bmap & mask);
416}
417
418/* set share deny for a given stateid */
419static inline void
420set_deny(u32 deny, struct nfs4_ol_stateid *stp)
421{
422 unsigned char mask = 1 << deny;
423
424 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
425 stp->st_deny_bmap |= mask;
426}
427
428/* clear share deny for a given stateid */
429static inline void
430clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
431{
432 unsigned char mask = 1 << deny;
433
434 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
435 stp->st_deny_bmap &= ~mask;
436}
437
438/* test whether a given stateid is denying specific access */
439static inline bool
440test_deny(u32 deny, struct nfs4_ol_stateid *stp)
441{
442 unsigned char mask = 1 << deny;
443
444 return (bool)(stp->st_deny_bmap & mask);
445}
446
447static int nfs4_access_to_omode(u32 access)
448{
449 switch (access & NFS4_SHARE_ACCESS_BOTH) {
450 case NFS4_SHARE_ACCESS_READ:
451 return O_RDONLY;
452 case NFS4_SHARE_ACCESS_WRITE:
453 return O_WRONLY;
454 case NFS4_SHARE_ACCESS_BOTH:
455 return O_RDWR;
456 }
457 WARN_ON_ONCE(1);
458 return O_RDONLY;
459}
460
461static inline int
462access_permit_read(struct nfs4_ol_stateid *stp)
463{
464 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
465 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
466 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
467}
468
469static inline int
470access_permit_write(struct nfs4_ol_stateid *stp)
471{
472 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
473 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
474}
475
476static inline struct nfs4_stateowner *
477nfs4_get_stateowner(struct nfs4_stateowner *sop)
478{
479 atomic_inc(&sop->so_count);
480 return sop;
481}
482
483static int
484same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
485{
486 return (sop->so_owner.len == owner->len) &&
487 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
488}
489
490static struct nfs4_openowner *
491find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
492 struct nfs4_client *clp)
493{
494 struct nfs4_stateowner *so;
495
496 lockdep_assert_held(&clp->cl_lock);
497
498 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
499 so_strhash) {
500 if (!so->so_is_open_owner)
501 continue;
502 if (same_owner_str(so, &open->op_owner))
503 return openowner(nfs4_get_stateowner(so));
504 }
505 return NULL;
506}
507
508static struct nfs4_openowner *
509find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
510 struct nfs4_client *clp)
511{
512 struct nfs4_openowner *oo;
513
514 spin_lock(&clp->cl_lock);
515 oo = find_openstateowner_str_locked(hashval, open, clp);
516 spin_unlock(&clp->cl_lock);
517 return oo;
518}
519
520static inline u32
521opaque_hashval(const void *ptr, int nbytes)
522{
523 unsigned char *cptr = (unsigned char *) ptr;
524
525 u32 x = 0;
526 while (nbytes--) {
527 x *= 37;
528 x += *cptr++;
529 }
530 return x;
531}
532
533static void nfsd4_free_file_rcu(struct rcu_head *rcu)
534{
535 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
536
537 kmem_cache_free(file_slab, fp);
538}
539
540void
541put_nfs4_file(struct nfs4_file *fi)
542{
543 might_lock(&state_lock);
544
545 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
546 hlist_del_rcu(&fi->fi_hash);
547 spin_unlock(&state_lock);
548 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
549 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
550 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
551 }
552}
553
554static struct nfsd_file *
555__nfs4_get_fd(struct nfs4_file *f, int oflag)
556{
557 if (f->fi_fds[oflag])
558 return nfsd_file_get(f->fi_fds[oflag]);
559 return NULL;
560}
561
562static struct nfsd_file *
563find_writeable_file_locked(struct nfs4_file *f)
564{
565 struct nfsd_file *ret;
566
567 lockdep_assert_held(&f->fi_lock);
568
569 ret = __nfs4_get_fd(f, O_WRONLY);
570 if (!ret)
571 ret = __nfs4_get_fd(f, O_RDWR);
572 return ret;
573}
574
575static struct nfsd_file *
576find_writeable_file(struct nfs4_file *f)
577{
578 struct nfsd_file *ret;
579
580 spin_lock(&f->fi_lock);
581 ret = find_writeable_file_locked(f);
582 spin_unlock(&f->fi_lock);
583
584 return ret;
585}
586
587static struct nfsd_file *
588find_readable_file_locked(struct nfs4_file *f)
589{
590 struct nfsd_file *ret;
591
592 lockdep_assert_held(&f->fi_lock);
593
594 ret = __nfs4_get_fd(f, O_RDONLY);
595 if (!ret)
596 ret = __nfs4_get_fd(f, O_RDWR);
597 return ret;
598}
599
600static struct nfsd_file *
601find_readable_file(struct nfs4_file *f)
602{
603 struct nfsd_file *ret;
604
605 spin_lock(&f->fi_lock);
606 ret = find_readable_file_locked(f);
607 spin_unlock(&f->fi_lock);
608
609 return ret;
610}
611
612struct nfsd_file *
613find_any_file(struct nfs4_file *f)
614{
615 struct nfsd_file *ret;
616
617 if (!f)
618 return NULL;
619 spin_lock(&f->fi_lock);
620 ret = __nfs4_get_fd(f, O_RDWR);
621 if (!ret) {
622 ret = __nfs4_get_fd(f, O_WRONLY);
623 if (!ret)
624 ret = __nfs4_get_fd(f, O_RDONLY);
625 }
626 spin_unlock(&f->fi_lock);
627 return ret;
628}
629
630static struct nfsd_file *find_deleg_file(struct nfs4_file *f)
631{
632 struct nfsd_file *ret = NULL;
633
634 spin_lock(&f->fi_lock);
635 if (f->fi_deleg_file)
636 ret = nfsd_file_get(f->fi_deleg_file);
637 spin_unlock(&f->fi_lock);
638 return ret;
639}
640
641static atomic_long_t num_delegations;
642unsigned long max_delegations;
643
644/*
645 * Open owner state (share locks)
646 */
647
648/* hash tables for lock and open owners */
649#define OWNER_HASH_BITS 8
650#define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
651#define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
652
653static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
654{
655 unsigned int ret;
656
657 ret = opaque_hashval(ownername->data, ownername->len);
658 return ret & OWNER_HASH_MASK;
659}
660
661/* hash table for nfs4_file */
662#define FILE_HASH_BITS 8
663#define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
664
665static unsigned int file_hashval(struct svc_fh *fh)
666{
667 struct inode *inode = d_inode(fh->fh_dentry);
668
669 /* XXX: why not (here & in file cache) use inode? */
670 return (unsigned int)hash_long(inode->i_ino, FILE_HASH_BITS);
671}
672
673static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
674
675static void
676__nfs4_file_get_access(struct nfs4_file *fp, u32 access)
677{
678 lockdep_assert_held(&fp->fi_lock);
679
680 if (access & NFS4_SHARE_ACCESS_WRITE)
681 atomic_inc(&fp->fi_access[O_WRONLY]);
682 if (access & NFS4_SHARE_ACCESS_READ)
683 atomic_inc(&fp->fi_access[O_RDONLY]);
684}
685
686static __be32
687nfs4_file_get_access(struct nfs4_file *fp, u32 access)
688{
689 lockdep_assert_held(&fp->fi_lock);
690
691 /* Does this access mode make sense? */
692 if (access & ~NFS4_SHARE_ACCESS_BOTH)
693 return nfserr_inval;
694
695 /* Does it conflict with a deny mode already set? */
696 if ((access & fp->fi_share_deny) != 0)
697 return nfserr_share_denied;
698
699 __nfs4_file_get_access(fp, access);
700 return nfs_ok;
701}
702
703static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
704{
705 /* Common case is that there is no deny mode. */
706 if (deny) {
707 /* Does this deny mode make sense? */
708 if (deny & ~NFS4_SHARE_DENY_BOTH)
709 return nfserr_inval;
710
711 if ((deny & NFS4_SHARE_DENY_READ) &&
712 atomic_read(&fp->fi_access[O_RDONLY]))
713 return nfserr_share_denied;
714
715 if ((deny & NFS4_SHARE_DENY_WRITE) &&
716 atomic_read(&fp->fi_access[O_WRONLY]))
717 return nfserr_share_denied;
718 }
719 return nfs_ok;
720}
721
722static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
723{
724 might_lock(&fp->fi_lock);
725
726 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
727 struct nfsd_file *f1 = NULL;
728 struct nfsd_file *f2 = NULL;
729
730 swap(f1, fp->fi_fds[oflag]);
731 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
732 swap(f2, fp->fi_fds[O_RDWR]);
733 spin_unlock(&fp->fi_lock);
734 if (f1)
735 nfsd_file_put(f1);
736 if (f2)
737 nfsd_file_put(f2);
738 }
739}
740
741static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
742{
743 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
744
745 if (access & NFS4_SHARE_ACCESS_WRITE)
746 __nfs4_file_put_access(fp, O_WRONLY);
747 if (access & NFS4_SHARE_ACCESS_READ)
748 __nfs4_file_put_access(fp, O_RDONLY);
749}
750
751/*
752 * Allocate a new open/delegation state counter. This is needed for
753 * pNFS for proper return on close semantics.
754 *
755 * Note that we only allocate it for pNFS-enabled exports, otherwise
756 * all pointers to struct nfs4_clnt_odstate are always NULL.
757 */
758static struct nfs4_clnt_odstate *
759alloc_clnt_odstate(struct nfs4_client *clp)
760{
761 struct nfs4_clnt_odstate *co;
762
763 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
764 if (co) {
765 co->co_client = clp;
766 refcount_set(&co->co_odcount, 1);
767 }
768 return co;
769}
770
771static void
772hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
773{
774 struct nfs4_file *fp = co->co_file;
775
776 lockdep_assert_held(&fp->fi_lock);
777 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
778}
779
780static inline void
781get_clnt_odstate(struct nfs4_clnt_odstate *co)
782{
783 if (co)
784 refcount_inc(&co->co_odcount);
785}
786
787static void
788put_clnt_odstate(struct nfs4_clnt_odstate *co)
789{
790 struct nfs4_file *fp;
791
792 if (!co)
793 return;
794
795 fp = co->co_file;
796 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
797 list_del(&co->co_perfile);
798 spin_unlock(&fp->fi_lock);
799
800 nfsd4_return_all_file_layouts(co->co_client, fp);
801 kmem_cache_free(odstate_slab, co);
802 }
803}
804
805static struct nfs4_clnt_odstate *
806find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
807{
808 struct nfs4_clnt_odstate *co;
809 struct nfs4_client *cl;
810
811 if (!new)
812 return NULL;
813
814 cl = new->co_client;
815
816 spin_lock(&fp->fi_lock);
817 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
818 if (co->co_client == cl) {
819 get_clnt_odstate(co);
820 goto out;
821 }
822 }
823 co = new;
824 co->co_file = fp;
825 hash_clnt_odstate_locked(new);
826out:
827 spin_unlock(&fp->fi_lock);
828 return co;
829}
830
831struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
832 void (*sc_free)(struct nfs4_stid *))
833{
834 struct nfs4_stid *stid;
835 int new_id;
836
837 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
838 if (!stid)
839 return NULL;
840
841 idr_preload(GFP_KERNEL);
842 spin_lock(&cl->cl_lock);
843 /* Reserving 0 for start of file in nfsdfs "states" file: */
844 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
845 spin_unlock(&cl->cl_lock);
846 idr_preload_end();
847 if (new_id < 0)
848 goto out_free;
849
850 stid->sc_free = sc_free;
851 stid->sc_client = cl;
852 stid->sc_stateid.si_opaque.so_id = new_id;
853 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
854 /* Will be incremented before return to client: */
855 refcount_set(&stid->sc_count, 1);
856 spin_lock_init(&stid->sc_lock);
857 INIT_LIST_HEAD(&stid->sc_cp_list);
858
859 /*
860 * It shouldn't be a problem to reuse an opaque stateid value.
861 * I don't think it is for 4.1. But with 4.0 I worry that, for
862 * example, a stray write retransmission could be accepted by
863 * the server when it should have been rejected. Therefore,
864 * adopt a trick from the sctp code to attempt to maximize the
865 * amount of time until an id is reused, by ensuring they always
866 * "increase" (mod INT_MAX):
867 */
868 return stid;
869out_free:
870 kmem_cache_free(slab, stid);
871 return NULL;
872}
873
874/*
875 * Create a unique stateid_t to represent each COPY.
876 */
877static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
878 unsigned char sc_type)
879{
880 int new_id;
881
882 stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
883 stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
884 stid->sc_type = sc_type;
885
886 idr_preload(GFP_KERNEL);
887 spin_lock(&nn->s2s_cp_lock);
888 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
889 stid->stid.si_opaque.so_id = new_id;
890 stid->stid.si_generation = 1;
891 spin_unlock(&nn->s2s_cp_lock);
892 idr_preload_end();
893 if (new_id < 0)
894 return 0;
895 return 1;
896}
897
898int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
899{
900 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
901}
902
903struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
904 struct nfs4_stid *p_stid)
905{
906 struct nfs4_cpntf_state *cps;
907
908 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
909 if (!cps)
910 return NULL;
911 cps->cpntf_time = ktime_get_boottime_seconds();
912 refcount_set(&cps->cp_stateid.sc_count, 1);
913 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
914 goto out_free;
915 spin_lock(&nn->s2s_cp_lock);
916 list_add(&cps->cp_list, &p_stid->sc_cp_list);
917 spin_unlock(&nn->s2s_cp_lock);
918 return cps;
919out_free:
920 kfree(cps);
921 return NULL;
922}
923
924void nfs4_free_copy_state(struct nfsd4_copy *copy)
925{
926 struct nfsd_net *nn;
927
928 WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
929 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
930 spin_lock(&nn->s2s_cp_lock);
931 idr_remove(&nn->s2s_cp_stateids,
932 copy->cp_stateid.stid.si_opaque.so_id);
933 spin_unlock(&nn->s2s_cp_lock);
934}
935
936static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
937{
938 struct nfs4_cpntf_state *cps;
939 struct nfsd_net *nn;
940
941 nn = net_generic(net, nfsd_net_id);
942 spin_lock(&nn->s2s_cp_lock);
943 while (!list_empty(&stid->sc_cp_list)) {
944 cps = list_first_entry(&stid->sc_cp_list,
945 struct nfs4_cpntf_state, cp_list);
946 _free_cpntf_state_locked(nn, cps);
947 }
948 spin_unlock(&nn->s2s_cp_lock);
949}
950
951static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
952{
953 struct nfs4_stid *stid;
954
955 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
956 if (!stid)
957 return NULL;
958
959 return openlockstateid(stid);
960}
961
962static void nfs4_free_deleg(struct nfs4_stid *stid)
963{
964 kmem_cache_free(deleg_slab, stid);
965 atomic_long_dec(&num_delegations);
966}
967
968/*
969 * When we recall a delegation, we should be careful not to hand it
970 * out again straight away.
971 * To ensure this we keep a pair of bloom filters ('new' and 'old')
972 * in which the filehandles of recalled delegations are "stored".
973 * If a filehandle appear in either filter, a delegation is blocked.
974 * When a delegation is recalled, the filehandle is stored in the "new"
975 * filter.
976 * Every 30 seconds we swap the filters and clear the "new" one,
977 * unless both are empty of course.
978 *
979 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
980 * low 3 bytes as hash-table indices.
981 *
982 * 'blocked_delegations_lock', which is always taken in block_delegations(),
983 * is used to manage concurrent access. Testing does not need the lock
984 * except when swapping the two filters.
985 */
986static DEFINE_SPINLOCK(blocked_delegations_lock);
987static struct bloom_pair {
988 int entries, old_entries;
989 time64_t swap_time;
990 int new; /* index into 'set' */
991 DECLARE_BITMAP(set[2], 256);
992} blocked_delegations;
993
994static int delegation_blocked(struct knfsd_fh *fh)
995{
996 u32 hash;
997 struct bloom_pair *bd = &blocked_delegations;
998
999 if (bd->entries == 0)
1000 return 0;
1001 if (ktime_get_seconds() - bd->swap_time > 30) {
1002 spin_lock(&blocked_delegations_lock);
1003 if (ktime_get_seconds() - bd->swap_time > 30) {
1004 bd->entries -= bd->old_entries;
1005 bd->old_entries = bd->entries;
1006 memset(bd->set[bd->new], 0,
1007 sizeof(bd->set[0]));
1008 bd->new = 1-bd->new;
1009 bd->swap_time = ktime_get_seconds();
1010 }
1011 spin_unlock(&blocked_delegations_lock);
1012 }
1013 hash = jhash(&fh->fh_base, fh->fh_size, 0);
1014 if (test_bit(hash&255, bd->set[0]) &&
1015 test_bit((hash>>8)&255, bd->set[0]) &&
1016 test_bit((hash>>16)&255, bd->set[0]))
1017 return 1;
1018
1019 if (test_bit(hash&255, bd->set[1]) &&
1020 test_bit((hash>>8)&255, bd->set[1]) &&
1021 test_bit((hash>>16)&255, bd->set[1]))
1022 return 1;
1023
1024 return 0;
1025}
1026
1027static void block_delegations(struct knfsd_fh *fh)
1028{
1029 u32 hash;
1030 struct bloom_pair *bd = &blocked_delegations;
1031
1032 hash = jhash(&fh->fh_base, fh->fh_size, 0);
1033
1034 spin_lock(&blocked_delegations_lock);
1035 __set_bit(hash&255, bd->set[bd->new]);
1036 __set_bit((hash>>8)&255, bd->set[bd->new]);
1037 __set_bit((hash>>16)&255, bd->set[bd->new]);
1038 if (bd->entries == 0)
1039 bd->swap_time = ktime_get_seconds();
1040 bd->entries += 1;
1041 spin_unlock(&blocked_delegations_lock);
1042}
1043
1044static struct nfs4_delegation *
1045alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1046 struct svc_fh *current_fh,
1047 struct nfs4_clnt_odstate *odstate)
1048{
1049 struct nfs4_delegation *dp;
1050 long n;
1051
1052 dprintk("NFSD alloc_init_deleg\n");
1053 n = atomic_long_inc_return(&num_delegations);
1054 if (n < 0 || n > max_delegations)
1055 goto out_dec;
1056 if (delegation_blocked(¤t_fh->fh_handle))
1057 goto out_dec;
1058 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
1059 if (dp == NULL)
1060 goto out_dec;
1061
1062 /*
1063 * delegation seqid's are never incremented. The 4.1 special
1064 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1065 * 0 anyway just for consistency and use 1:
1066 */
1067 dp->dl_stid.sc_stateid.si_generation = 1;
1068 INIT_LIST_HEAD(&dp->dl_perfile);
1069 INIT_LIST_HEAD(&dp->dl_perclnt);
1070 INIT_LIST_HEAD(&dp->dl_recall_lru);
1071 dp->dl_clnt_odstate = odstate;
1072 get_clnt_odstate(odstate);
1073 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
1074 dp->dl_retries = 1;
1075 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1076 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1077 get_nfs4_file(fp);
1078 dp->dl_stid.sc_file = fp;
1079 return dp;
1080out_dec:
1081 atomic_long_dec(&num_delegations);
1082 return NULL;
1083}
1084
1085void
1086nfs4_put_stid(struct nfs4_stid *s)
1087{
1088 struct nfs4_file *fp = s->sc_file;
1089 struct nfs4_client *clp = s->sc_client;
1090
1091 might_lock(&clp->cl_lock);
1092
1093 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1094 wake_up_all(&close_wq);
1095 return;
1096 }
1097 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1098 nfs4_free_cpntf_statelist(clp->net, s);
1099 spin_unlock(&clp->cl_lock);
1100 s->sc_free(s);
1101 if (fp)
1102 put_nfs4_file(fp);
1103}
1104
1105void
1106nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1107{
1108 stateid_t *src = &stid->sc_stateid;
1109
1110 spin_lock(&stid->sc_lock);
1111 if (unlikely(++src->si_generation == 0))
1112 src->si_generation = 1;
1113 memcpy(dst, src, sizeof(*dst));
1114 spin_unlock(&stid->sc_lock);
1115}
1116
1117static void put_deleg_file(struct nfs4_file *fp)
1118{
1119 struct nfsd_file *nf = NULL;
1120
1121 spin_lock(&fp->fi_lock);
1122 if (--fp->fi_delegees == 0)
1123 swap(nf, fp->fi_deleg_file);
1124 spin_unlock(&fp->fi_lock);
1125
1126 if (nf)
1127 nfsd_file_put(nf);
1128}
1129
1130static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1131{
1132 struct nfs4_file *fp = dp->dl_stid.sc_file;
1133 struct nfsd_file *nf = fp->fi_deleg_file;
1134
1135 WARN_ON_ONCE(!fp->fi_delegees);
1136
1137 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1138 put_deleg_file(fp);
1139}
1140
1141static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1142{
1143 put_clnt_odstate(dp->dl_clnt_odstate);
1144 nfs4_unlock_deleg_lease(dp);
1145 nfs4_put_stid(&dp->dl_stid);
1146}
1147
1148void nfs4_unhash_stid(struct nfs4_stid *s)
1149{
1150 s->sc_type = 0;
1151}
1152
1153/**
1154 * nfs4_delegation_exists - Discover if this delegation already exists
1155 * @clp: a pointer to the nfs4_client we're granting a delegation to
1156 * @fp: a pointer to the nfs4_file we're granting a delegation on
1157 *
1158 * Return:
1159 * On success: true iff an existing delegation is found
1160 */
1161
1162static bool
1163nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1164{
1165 struct nfs4_delegation *searchdp = NULL;
1166 struct nfs4_client *searchclp = NULL;
1167
1168 lockdep_assert_held(&state_lock);
1169 lockdep_assert_held(&fp->fi_lock);
1170
1171 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1172 searchclp = searchdp->dl_stid.sc_client;
1173 if (clp == searchclp) {
1174 return true;
1175 }
1176 }
1177 return false;
1178}
1179
1180/**
1181 * hash_delegation_locked - Add a delegation to the appropriate lists
1182 * @dp: a pointer to the nfs4_delegation we are adding.
1183 * @fp: a pointer to the nfs4_file we're granting a delegation on
1184 *
1185 * Return:
1186 * On success: NULL if the delegation was successfully hashed.
1187 *
1188 * On error: -EAGAIN if one was previously granted to this
1189 * nfs4_client for this nfs4_file. Delegation is not hashed.
1190 *
1191 */
1192
1193static int
1194hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1195{
1196 struct nfs4_client *clp = dp->dl_stid.sc_client;
1197
1198 lockdep_assert_held(&state_lock);
1199 lockdep_assert_held(&fp->fi_lock);
1200
1201 if (nfs4_delegation_exists(clp, fp))
1202 return -EAGAIN;
1203 refcount_inc(&dp->dl_stid.sc_count);
1204 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1205 list_add(&dp->dl_perfile, &fp->fi_delegations);
1206 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1207 return 0;
1208}
1209
1210static bool
1211unhash_delegation_locked(struct nfs4_delegation *dp)
1212{
1213 struct nfs4_file *fp = dp->dl_stid.sc_file;
1214
1215 lockdep_assert_held(&state_lock);
1216
1217 if (list_empty(&dp->dl_perfile))
1218 return false;
1219
1220 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1221 /* Ensure that deleg break won't try to requeue it */
1222 ++dp->dl_time;
1223 spin_lock(&fp->fi_lock);
1224 list_del_init(&dp->dl_perclnt);
1225 list_del_init(&dp->dl_recall_lru);
1226 list_del_init(&dp->dl_perfile);
1227 spin_unlock(&fp->fi_lock);
1228 return true;
1229}
1230
1231static void destroy_delegation(struct nfs4_delegation *dp)
1232{
1233 bool unhashed;
1234
1235 spin_lock(&state_lock);
1236 unhashed = unhash_delegation_locked(dp);
1237 spin_unlock(&state_lock);
1238 if (unhashed)
1239 destroy_unhashed_deleg(dp);
1240}
1241
1242static void revoke_delegation(struct nfs4_delegation *dp)
1243{
1244 struct nfs4_client *clp = dp->dl_stid.sc_client;
1245
1246 WARN_ON(!list_empty(&dp->dl_recall_lru));
1247
1248 if (clp->cl_minorversion) {
1249 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1250 refcount_inc(&dp->dl_stid.sc_count);
1251 spin_lock(&clp->cl_lock);
1252 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1253 spin_unlock(&clp->cl_lock);
1254 }
1255 destroy_unhashed_deleg(dp);
1256}
1257
1258/*
1259 * SETCLIENTID state
1260 */
1261
1262static unsigned int clientid_hashval(u32 id)
1263{
1264 return id & CLIENT_HASH_MASK;
1265}
1266
1267static unsigned int clientstr_hashval(struct xdr_netobj name)
1268{
1269 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1270}
1271
1272/*
1273 * A stateid that had a deny mode associated with it is being released
1274 * or downgraded. Recalculate the deny mode on the file.
1275 */
1276static void
1277recalculate_deny_mode(struct nfs4_file *fp)
1278{
1279 struct nfs4_ol_stateid *stp;
1280
1281 spin_lock(&fp->fi_lock);
1282 fp->fi_share_deny = 0;
1283 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1284 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1285 spin_unlock(&fp->fi_lock);
1286}
1287
1288static void
1289reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1290{
1291 int i;
1292 bool change = false;
1293
1294 for (i = 1; i < 4; i++) {
1295 if ((i & deny) != i) {
1296 change = true;
1297 clear_deny(i, stp);
1298 }
1299 }
1300
1301 /* Recalculate per-file deny mode if there was a change */
1302 if (change)
1303 recalculate_deny_mode(stp->st_stid.sc_file);
1304}
1305
1306/* release all access and file references for a given stateid */
1307static void
1308release_all_access(struct nfs4_ol_stateid *stp)
1309{
1310 int i;
1311 struct nfs4_file *fp = stp->st_stid.sc_file;
1312
1313 if (fp && stp->st_deny_bmap != 0)
1314 recalculate_deny_mode(fp);
1315
1316 for (i = 1; i < 4; i++) {
1317 if (test_access(i, stp))
1318 nfs4_file_put_access(stp->st_stid.sc_file, i);
1319 clear_access(i, stp);
1320 }
1321}
1322
1323static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1324{
1325 kfree(sop->so_owner.data);
1326 sop->so_ops->so_free(sop);
1327}
1328
1329static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1330{
1331 struct nfs4_client *clp = sop->so_client;
1332
1333 might_lock(&clp->cl_lock);
1334
1335 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1336 return;
1337 sop->so_ops->so_unhash(sop);
1338 spin_unlock(&clp->cl_lock);
1339 nfs4_free_stateowner(sop);
1340}
1341
1342static bool
1343nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1344{
1345 return list_empty(&stp->st_perfile);
1346}
1347
1348static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1349{
1350 struct nfs4_file *fp = stp->st_stid.sc_file;
1351
1352 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1353
1354 if (list_empty(&stp->st_perfile))
1355 return false;
1356
1357 spin_lock(&fp->fi_lock);
1358 list_del_init(&stp->st_perfile);
1359 spin_unlock(&fp->fi_lock);
1360 list_del(&stp->st_perstateowner);
1361 return true;
1362}
1363
1364static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1365{
1366 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1367
1368 put_clnt_odstate(stp->st_clnt_odstate);
1369 release_all_access(stp);
1370 if (stp->st_stateowner)
1371 nfs4_put_stateowner(stp->st_stateowner);
1372 kmem_cache_free(stateid_slab, stid);
1373}
1374
1375static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1376{
1377 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1378 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1379 struct nfsd_file *nf;
1380
1381 nf = find_any_file(stp->st_stid.sc_file);
1382 if (nf) {
1383 get_file(nf->nf_file);
1384 filp_close(nf->nf_file, (fl_owner_t)lo);
1385 nfsd_file_put(nf);
1386 }
1387 nfs4_free_ol_stateid(stid);
1388}
1389
1390/*
1391 * Put the persistent reference to an already unhashed generic stateid, while
1392 * holding the cl_lock. If it's the last reference, then put it onto the
1393 * reaplist for later destruction.
1394 */
1395static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1396 struct list_head *reaplist)
1397{
1398 struct nfs4_stid *s = &stp->st_stid;
1399 struct nfs4_client *clp = s->sc_client;
1400
1401 lockdep_assert_held(&clp->cl_lock);
1402
1403 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1404
1405 if (!refcount_dec_and_test(&s->sc_count)) {
1406 wake_up_all(&close_wq);
1407 return;
1408 }
1409
1410 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1411 list_add(&stp->st_locks, reaplist);
1412}
1413
1414static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1415{
1416 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1417
1418 if (!unhash_ol_stateid(stp))
1419 return false;
1420 list_del_init(&stp->st_locks);
1421 nfs4_unhash_stid(&stp->st_stid);
1422 return true;
1423}
1424
1425static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1426{
1427 struct nfs4_client *clp = stp->st_stid.sc_client;
1428 bool unhashed;
1429
1430 spin_lock(&clp->cl_lock);
1431 unhashed = unhash_lock_stateid(stp);
1432 spin_unlock(&clp->cl_lock);
1433 if (unhashed)
1434 nfs4_put_stid(&stp->st_stid);
1435}
1436
1437static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1438{
1439 struct nfs4_client *clp = lo->lo_owner.so_client;
1440
1441 lockdep_assert_held(&clp->cl_lock);
1442
1443 list_del_init(&lo->lo_owner.so_strhash);
1444}
1445
1446/*
1447 * Free a list of generic stateids that were collected earlier after being
1448 * fully unhashed.
1449 */
1450static void
1451free_ol_stateid_reaplist(struct list_head *reaplist)
1452{
1453 struct nfs4_ol_stateid *stp;
1454 struct nfs4_file *fp;
1455
1456 might_sleep();
1457
1458 while (!list_empty(reaplist)) {
1459 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1460 st_locks);
1461 list_del(&stp->st_locks);
1462 fp = stp->st_stid.sc_file;
1463 stp->st_stid.sc_free(&stp->st_stid);
1464 if (fp)
1465 put_nfs4_file(fp);
1466 }
1467}
1468
1469static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1470 struct list_head *reaplist)
1471{
1472 struct nfs4_ol_stateid *stp;
1473
1474 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1475
1476 while (!list_empty(&open_stp->st_locks)) {
1477 stp = list_entry(open_stp->st_locks.next,
1478 struct nfs4_ol_stateid, st_locks);
1479 WARN_ON(!unhash_lock_stateid(stp));
1480 put_ol_stateid_locked(stp, reaplist);
1481 }
1482}
1483
1484static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1485 struct list_head *reaplist)
1486{
1487 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1488
1489 if (!unhash_ol_stateid(stp))
1490 return false;
1491 release_open_stateid_locks(stp, reaplist);
1492 return true;
1493}
1494
1495static void release_open_stateid(struct nfs4_ol_stateid *stp)
1496{
1497 LIST_HEAD(reaplist);
1498
1499 spin_lock(&stp->st_stid.sc_client->cl_lock);
1500 if (unhash_open_stateid(stp, &reaplist))
1501 put_ol_stateid_locked(stp, &reaplist);
1502 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1503 free_ol_stateid_reaplist(&reaplist);
1504}
1505
1506static void unhash_openowner_locked(struct nfs4_openowner *oo)
1507{
1508 struct nfs4_client *clp = oo->oo_owner.so_client;
1509
1510 lockdep_assert_held(&clp->cl_lock);
1511
1512 list_del_init(&oo->oo_owner.so_strhash);
1513 list_del_init(&oo->oo_perclient);
1514}
1515
1516static void release_last_closed_stateid(struct nfs4_openowner *oo)
1517{
1518 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1519 nfsd_net_id);
1520 struct nfs4_ol_stateid *s;
1521
1522 spin_lock(&nn->client_lock);
1523 s = oo->oo_last_closed_stid;
1524 if (s) {
1525 list_del_init(&oo->oo_close_lru);
1526 oo->oo_last_closed_stid = NULL;
1527 }
1528 spin_unlock(&nn->client_lock);
1529 if (s)
1530 nfs4_put_stid(&s->st_stid);
1531}
1532
1533static void release_openowner(struct nfs4_openowner *oo)
1534{
1535 struct nfs4_ol_stateid *stp;
1536 struct nfs4_client *clp = oo->oo_owner.so_client;
1537 struct list_head reaplist;
1538
1539 INIT_LIST_HEAD(&reaplist);
1540
1541 spin_lock(&clp->cl_lock);
1542 unhash_openowner_locked(oo);
1543 while (!list_empty(&oo->oo_owner.so_stateids)) {
1544 stp = list_first_entry(&oo->oo_owner.so_stateids,
1545 struct nfs4_ol_stateid, st_perstateowner);
1546 if (unhash_open_stateid(stp, &reaplist))
1547 put_ol_stateid_locked(stp, &reaplist);
1548 }
1549 spin_unlock(&clp->cl_lock);
1550 free_ol_stateid_reaplist(&reaplist);
1551 release_last_closed_stateid(oo);
1552 nfs4_put_stateowner(&oo->oo_owner);
1553}
1554
1555static inline int
1556hash_sessionid(struct nfs4_sessionid *sessionid)
1557{
1558 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1559
1560 return sid->sequence % SESSION_HASH_SIZE;
1561}
1562
1563#ifdef CONFIG_SUNRPC_DEBUG
1564static inline void
1565dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1566{
1567 u32 *ptr = (u32 *)(&sessionid->data[0]);
1568 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1569}
1570#else
1571static inline void
1572dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1573{
1574}
1575#endif
1576
1577/*
1578 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1579 * won't be used for replay.
1580 */
1581void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1582{
1583 struct nfs4_stateowner *so = cstate->replay_owner;
1584
1585 if (nfserr == nfserr_replay_me)
1586 return;
1587
1588 if (!seqid_mutating_err(ntohl(nfserr))) {
1589 nfsd4_cstate_clear_replay(cstate);
1590 return;
1591 }
1592 if (!so)
1593 return;
1594 if (so->so_is_open_owner)
1595 release_last_closed_stateid(openowner(so));
1596 so->so_seqid++;
1597 return;
1598}
1599
1600static void
1601gen_sessionid(struct nfsd4_session *ses)
1602{
1603 struct nfs4_client *clp = ses->se_client;
1604 struct nfsd4_sessionid *sid;
1605
1606 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1607 sid->clientid = clp->cl_clientid;
1608 sid->sequence = current_sessionid++;
1609 sid->reserved = 0;
1610}
1611
1612/*
1613 * The protocol defines ca_maxresponssize_cached to include the size of
1614 * the rpc header, but all we need to cache is the data starting after
1615 * the end of the initial SEQUENCE operation--the rest we regenerate
1616 * each time. Therefore we can advertise a ca_maxresponssize_cached
1617 * value that is the number of bytes in our cache plus a few additional
1618 * bytes. In order to stay on the safe side, and not promise more than
1619 * we can cache, those additional bytes must be the minimum possible: 24
1620 * bytes of rpc header (xid through accept state, with AUTH_NULL
1621 * verifier), 12 for the compound header (with zero-length tag), and 44
1622 * for the SEQUENCE op response:
1623 */
1624#define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1625
1626static void
1627free_session_slots(struct nfsd4_session *ses)
1628{
1629 int i;
1630
1631 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1632 free_svc_cred(&ses->se_slots[i]->sl_cred);
1633 kfree(ses->se_slots[i]);
1634 }
1635}
1636
1637/*
1638 * We don't actually need to cache the rpc and session headers, so we
1639 * can allocate a little less for each slot:
1640 */
1641static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1642{
1643 u32 size;
1644
1645 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1646 size = 0;
1647 else
1648 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1649 return size + sizeof(struct nfsd4_slot);
1650}
1651
1652/*
1653 * XXX: If we run out of reserved DRC memory we could (up to a point)
1654 * re-negotiate active sessions and reduce their slot usage to make
1655 * room for new connections. For now we just fail the create session.
1656 */
1657static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1658{
1659 u32 slotsize = slot_bytes(ca);
1660 u32 num = ca->maxreqs;
1661 unsigned long avail, total_avail;
1662 unsigned int scale_factor;
1663
1664 spin_lock(&nfsd_drc_lock);
1665 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1666 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1667 else
1668 /* We have handed out more space than we chose in
1669 * set_max_drc() to allow. That isn't really a
1670 * problem as long as that doesn't make us think we
1671 * have lots more due to integer overflow.
1672 */
1673 total_avail = 0;
1674 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1675 /*
1676 * Never use more than a fraction of the remaining memory,
1677 * unless it's the only way to give this client a slot.
1678 * The chosen fraction is either 1/8 or 1/number of threads,
1679 * whichever is smaller. This ensures there are adequate
1680 * slots to support multiple clients per thread.
1681 * Give the client one slot even if that would require
1682 * over-allocation--it is better than failure.
1683 */
1684 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1685
1686 avail = clamp_t(unsigned long, avail, slotsize,
1687 total_avail/scale_factor);
1688 num = min_t(int, num, avail / slotsize);
1689 num = max_t(int, num, 1);
1690 nfsd_drc_mem_used += num * slotsize;
1691 spin_unlock(&nfsd_drc_lock);
1692
1693 return num;
1694}
1695
1696static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1697{
1698 int slotsize = slot_bytes(ca);
1699
1700 spin_lock(&nfsd_drc_lock);
1701 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1702 spin_unlock(&nfsd_drc_lock);
1703}
1704
1705static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1706 struct nfsd4_channel_attrs *battrs)
1707{
1708 int numslots = fattrs->maxreqs;
1709 int slotsize = slot_bytes(fattrs);
1710 struct nfsd4_session *new;
1711 int mem, i;
1712
1713 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1714 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1715 mem = numslots * sizeof(struct nfsd4_slot *);
1716
1717 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1718 if (!new)
1719 return NULL;
1720 /* allocate each struct nfsd4_slot and data cache in one piece */
1721 for (i = 0; i < numslots; i++) {
1722 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1723 if (!new->se_slots[i])
1724 goto out_free;
1725 }
1726
1727 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1728 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1729
1730 return new;
1731out_free:
1732 while (i--)
1733 kfree(new->se_slots[i]);
1734 kfree(new);
1735 return NULL;
1736}
1737
1738static void free_conn(struct nfsd4_conn *c)
1739{
1740 svc_xprt_put(c->cn_xprt);
1741 kfree(c);
1742}
1743
1744static void nfsd4_conn_lost(struct svc_xpt_user *u)
1745{
1746 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1747 struct nfs4_client *clp = c->cn_session->se_client;
1748
1749 trace_nfsd_cb_lost(clp);
1750
1751 spin_lock(&clp->cl_lock);
1752 if (!list_empty(&c->cn_persession)) {
1753 list_del(&c->cn_persession);
1754 free_conn(c);
1755 }
1756 nfsd4_probe_callback(clp);
1757 spin_unlock(&clp->cl_lock);
1758}
1759
1760static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1761{
1762 struct nfsd4_conn *conn;
1763
1764 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1765 if (!conn)
1766 return NULL;
1767 svc_xprt_get(rqstp->rq_xprt);
1768 conn->cn_xprt = rqstp->rq_xprt;
1769 conn->cn_flags = flags;
1770 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1771 return conn;
1772}
1773
1774static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1775{
1776 conn->cn_session = ses;
1777 list_add(&conn->cn_persession, &ses->se_conns);
1778}
1779
1780static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1781{
1782 struct nfs4_client *clp = ses->se_client;
1783
1784 spin_lock(&clp->cl_lock);
1785 __nfsd4_hash_conn(conn, ses);
1786 spin_unlock(&clp->cl_lock);
1787}
1788
1789static int nfsd4_register_conn(struct nfsd4_conn *conn)
1790{
1791 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1792 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1793}
1794
1795static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1796{
1797 int ret;
1798
1799 nfsd4_hash_conn(conn, ses);
1800 ret = nfsd4_register_conn(conn);
1801 if (ret)
1802 /* oops; xprt is already down: */
1803 nfsd4_conn_lost(&conn->cn_xpt_user);
1804 /* We may have gained or lost a callback channel: */
1805 nfsd4_probe_callback_sync(ses->se_client);
1806}
1807
1808static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1809{
1810 u32 dir = NFS4_CDFC4_FORE;
1811
1812 if (cses->flags & SESSION4_BACK_CHAN)
1813 dir |= NFS4_CDFC4_BACK;
1814 return alloc_conn(rqstp, dir);
1815}
1816
1817/* must be called under client_lock */
1818static void nfsd4_del_conns(struct nfsd4_session *s)
1819{
1820 struct nfs4_client *clp = s->se_client;
1821 struct nfsd4_conn *c;
1822
1823 spin_lock(&clp->cl_lock);
1824 while (!list_empty(&s->se_conns)) {
1825 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1826 list_del_init(&c->cn_persession);
1827 spin_unlock(&clp->cl_lock);
1828
1829 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1830 free_conn(c);
1831
1832 spin_lock(&clp->cl_lock);
1833 }
1834 spin_unlock(&clp->cl_lock);
1835}
1836
1837static void __free_session(struct nfsd4_session *ses)
1838{
1839 free_session_slots(ses);
1840 kfree(ses);
1841}
1842
1843static void free_session(struct nfsd4_session *ses)
1844{
1845 nfsd4_del_conns(ses);
1846 nfsd4_put_drc_mem(&ses->se_fchannel);
1847 __free_session(ses);
1848}
1849
1850static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1851{
1852 int idx;
1853 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1854
1855 new->se_client = clp;
1856 gen_sessionid(new);
1857
1858 INIT_LIST_HEAD(&new->se_conns);
1859
1860 new->se_cb_seq_nr = 1;
1861 new->se_flags = cses->flags;
1862 new->se_cb_prog = cses->callback_prog;
1863 new->se_cb_sec = cses->cb_sec;
1864 atomic_set(&new->se_ref, 0);
1865 idx = hash_sessionid(&new->se_sessionid);
1866 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1867 spin_lock(&clp->cl_lock);
1868 list_add(&new->se_perclnt, &clp->cl_sessions);
1869 spin_unlock(&clp->cl_lock);
1870
1871 {
1872 struct sockaddr *sa = svc_addr(rqstp);
1873 /*
1874 * This is a little silly; with sessions there's no real
1875 * use for the callback address. Use the peer address
1876 * as a reasonable default for now, but consider fixing
1877 * the rpc client not to require an address in the
1878 * future:
1879 */
1880 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1881 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1882 }
1883}
1884
1885/* caller must hold client_lock */
1886static struct nfsd4_session *
1887__find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1888{
1889 struct nfsd4_session *elem;
1890 int idx;
1891 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1892
1893 lockdep_assert_held(&nn->client_lock);
1894
1895 dump_sessionid(__func__, sessionid);
1896 idx = hash_sessionid(sessionid);
1897 /* Search in the appropriate list */
1898 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1899 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1900 NFS4_MAX_SESSIONID_LEN)) {
1901 return elem;
1902 }
1903 }
1904
1905 dprintk("%s: session not found\n", __func__);
1906 return NULL;
1907}
1908
1909static struct nfsd4_session *
1910find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1911 __be32 *ret)
1912{
1913 struct nfsd4_session *session;
1914 __be32 status = nfserr_badsession;
1915
1916 session = __find_in_sessionid_hashtbl(sessionid, net);
1917 if (!session)
1918 goto out;
1919 status = nfsd4_get_session_locked(session);
1920 if (status)
1921 session = NULL;
1922out:
1923 *ret = status;
1924 return session;
1925}
1926
1927/* caller must hold client_lock */
1928static void
1929unhash_session(struct nfsd4_session *ses)
1930{
1931 struct nfs4_client *clp = ses->se_client;
1932 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1933
1934 lockdep_assert_held(&nn->client_lock);
1935
1936 list_del(&ses->se_hash);
1937 spin_lock(&ses->se_client->cl_lock);
1938 list_del(&ses->se_perclnt);
1939 spin_unlock(&ses->se_client->cl_lock);
1940}
1941
1942/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1943static int
1944STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1945{
1946 /*
1947 * We're assuming the clid was not given out from a boot
1948 * precisely 2^32 (about 136 years) before this one. That seems
1949 * a safe assumption:
1950 */
1951 if (clid->cl_boot == (u32)nn->boot_time)
1952 return 0;
1953 trace_nfsd_clid_stale(clid);
1954 return 1;
1955}
1956
1957/*
1958 * XXX Should we use a slab cache ?
1959 * This type of memory management is somewhat inefficient, but we use it
1960 * anyway since SETCLIENTID is not a common operation.
1961 */
1962static struct nfs4_client *alloc_client(struct xdr_netobj name)
1963{
1964 struct nfs4_client *clp;
1965 int i;
1966
1967 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1968 if (clp == NULL)
1969 return NULL;
1970 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1971 if (clp->cl_name.data == NULL)
1972 goto err_no_name;
1973 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1974 sizeof(struct list_head),
1975 GFP_KERNEL);
1976 if (!clp->cl_ownerstr_hashtbl)
1977 goto err_no_hashtbl;
1978 for (i = 0; i < OWNER_HASH_SIZE; i++)
1979 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1980 INIT_LIST_HEAD(&clp->cl_sessions);
1981 idr_init(&clp->cl_stateids);
1982 atomic_set(&clp->cl_rpc_users, 0);
1983 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1984 INIT_LIST_HEAD(&clp->cl_idhash);
1985 INIT_LIST_HEAD(&clp->cl_openowners);
1986 INIT_LIST_HEAD(&clp->cl_delegations);
1987 INIT_LIST_HEAD(&clp->cl_lru);
1988 INIT_LIST_HEAD(&clp->cl_revoked);
1989#ifdef CONFIG_NFSD_PNFS
1990 INIT_LIST_HEAD(&clp->cl_lo_states);
1991#endif
1992 INIT_LIST_HEAD(&clp->async_copies);
1993 spin_lock_init(&clp->async_lock);
1994 spin_lock_init(&clp->cl_lock);
1995 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1996 return clp;
1997err_no_hashtbl:
1998 kfree(clp->cl_name.data);
1999err_no_name:
2000 kmem_cache_free(client_slab, clp);
2001 return NULL;
2002}
2003
2004static void __free_client(struct kref *k)
2005{
2006 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2007 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2008
2009 free_svc_cred(&clp->cl_cred);
2010 kfree(clp->cl_ownerstr_hashtbl);
2011 kfree(clp->cl_name.data);
2012 kfree(clp->cl_nii_domain.data);
2013 kfree(clp->cl_nii_name.data);
2014 idr_destroy(&clp->cl_stateids);
2015 kmem_cache_free(client_slab, clp);
2016}
2017
2018static void drop_client(struct nfs4_client *clp)
2019{
2020 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2021}
2022
2023static void
2024free_client(struct nfs4_client *clp)
2025{
2026 while (!list_empty(&clp->cl_sessions)) {
2027 struct nfsd4_session *ses;
2028 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2029 se_perclnt);
2030 list_del(&ses->se_perclnt);
2031 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2032 free_session(ses);
2033 }
2034 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2035 if (clp->cl_nfsd_dentry) {
2036 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2037 clp->cl_nfsd_dentry = NULL;
2038 wake_up_all(&expiry_wq);
2039 }
2040 drop_client(clp);
2041}
2042
2043/* must be called under the client_lock */
2044static void
2045unhash_client_locked(struct nfs4_client *clp)
2046{
2047 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2048 struct nfsd4_session *ses;
2049
2050 lockdep_assert_held(&nn->client_lock);
2051
2052 /* Mark the client as expired! */
2053 clp->cl_time = 0;
2054 /* Make it invisible */
2055 if (!list_empty(&clp->cl_idhash)) {
2056 list_del_init(&clp->cl_idhash);
2057 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2058 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2059 else
2060 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2061 }
2062 list_del_init(&clp->cl_lru);
2063 spin_lock(&clp->cl_lock);
2064 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2065 list_del_init(&ses->se_hash);
2066 spin_unlock(&clp->cl_lock);
2067}
2068
2069static void
2070unhash_client(struct nfs4_client *clp)
2071{
2072 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2073
2074 spin_lock(&nn->client_lock);
2075 unhash_client_locked(clp);
2076 spin_unlock(&nn->client_lock);
2077}
2078
2079static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2080{
2081 if (atomic_read(&clp->cl_rpc_users))
2082 return nfserr_jukebox;
2083 unhash_client_locked(clp);
2084 return nfs_ok;
2085}
2086
2087static void
2088__destroy_client(struct nfs4_client *clp)
2089{
2090 int i;
2091 struct nfs4_openowner *oo;
2092 struct nfs4_delegation *dp;
2093 struct list_head reaplist;
2094
2095 INIT_LIST_HEAD(&reaplist);
2096 spin_lock(&state_lock);
2097 while (!list_empty(&clp->cl_delegations)) {
2098 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2099 WARN_ON(!unhash_delegation_locked(dp));
2100 list_add(&dp->dl_recall_lru, &reaplist);
2101 }
2102 spin_unlock(&state_lock);
2103 while (!list_empty(&reaplist)) {
2104 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2105 list_del_init(&dp->dl_recall_lru);
2106 destroy_unhashed_deleg(dp);
2107 }
2108 while (!list_empty(&clp->cl_revoked)) {
2109 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2110 list_del_init(&dp->dl_recall_lru);
2111 nfs4_put_stid(&dp->dl_stid);
2112 }
2113 while (!list_empty(&clp->cl_openowners)) {
2114 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2115 nfs4_get_stateowner(&oo->oo_owner);
2116 release_openowner(oo);
2117 }
2118 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2119 struct nfs4_stateowner *so, *tmp;
2120
2121 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2122 so_strhash) {
2123 /* Should be no openowners at this point */
2124 WARN_ON_ONCE(so->so_is_open_owner);
2125 remove_blocked_locks(lockowner(so));
2126 }
2127 }
2128 nfsd4_return_all_client_layouts(clp);
2129 nfsd4_shutdown_copy(clp);
2130 nfsd4_shutdown_callback(clp);
2131 if (clp->cl_cb_conn.cb_xprt)
2132 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2133 free_client(clp);
2134 wake_up_all(&expiry_wq);
2135}
2136
2137static void
2138destroy_client(struct nfs4_client *clp)
2139{
2140 unhash_client(clp);
2141 __destroy_client(clp);
2142}
2143
2144static void inc_reclaim_complete(struct nfs4_client *clp)
2145{
2146 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2147
2148 if (!nn->track_reclaim_completes)
2149 return;
2150 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2151 return;
2152 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2153 nn->reclaim_str_hashtbl_size) {
2154 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2155 clp->net->ns.inum);
2156 nfsd4_end_grace(nn);
2157 }
2158}
2159
2160static void expire_client(struct nfs4_client *clp)
2161{
2162 unhash_client(clp);
2163 nfsd4_client_record_remove(clp);
2164 __destroy_client(clp);
2165}
2166
2167static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2168{
2169 memcpy(target->cl_verifier.data, source->data,
2170 sizeof(target->cl_verifier.data));
2171}
2172
2173static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2174{
2175 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2176 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2177}
2178
2179static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2180{
2181 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2182 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2183 GFP_KERNEL);
2184 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2185 if ((source->cr_principal && !target->cr_principal) ||
2186 (source->cr_raw_principal && !target->cr_raw_principal) ||
2187 (source->cr_targ_princ && !target->cr_targ_princ))
2188 return -ENOMEM;
2189
2190 target->cr_flavor = source->cr_flavor;
2191 target->cr_uid = source->cr_uid;
2192 target->cr_gid = source->cr_gid;
2193 target->cr_group_info = source->cr_group_info;
2194 get_group_info(target->cr_group_info);
2195 target->cr_gss_mech = source->cr_gss_mech;
2196 if (source->cr_gss_mech)
2197 gss_mech_get(source->cr_gss_mech);
2198 return 0;
2199}
2200
2201static int
2202compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2203{
2204 if (o1->len < o2->len)
2205 return -1;
2206 if (o1->len > o2->len)
2207 return 1;
2208 return memcmp(o1->data, o2->data, o1->len);
2209}
2210
2211static int
2212same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2213{
2214 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2215}
2216
2217static int
2218same_clid(clientid_t *cl1, clientid_t *cl2)
2219{
2220 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2221}
2222
2223static bool groups_equal(struct group_info *g1, struct group_info *g2)
2224{
2225 int i;
2226
2227 if (g1->ngroups != g2->ngroups)
2228 return false;
2229 for (i=0; i<g1->ngroups; i++)
2230 if (!gid_eq(g1->gid[i], g2->gid[i]))
2231 return false;
2232 return true;
2233}
2234
2235/*
2236 * RFC 3530 language requires clid_inuse be returned when the
2237 * "principal" associated with a requests differs from that previously
2238 * used. We use uid, gid's, and gss principal string as our best
2239 * approximation. We also don't want to allow non-gss use of a client
2240 * established using gss: in theory cr_principal should catch that
2241 * change, but in practice cr_principal can be null even in the gss case
2242 * since gssd doesn't always pass down a principal string.
2243 */
2244static bool is_gss_cred(struct svc_cred *cr)
2245{
2246 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2247 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2248}
2249
2250
2251static bool
2252same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2253{
2254 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2255 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2256 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2257 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2258 return false;
2259 /* XXX: check that cr_targ_princ fields match ? */
2260 if (cr1->cr_principal == cr2->cr_principal)
2261 return true;
2262 if (!cr1->cr_principal || !cr2->cr_principal)
2263 return false;
2264 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2265}
2266
2267static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2268{
2269 struct svc_cred *cr = &rqstp->rq_cred;
2270 u32 service;
2271
2272 if (!cr->cr_gss_mech)
2273 return false;
2274 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2275 return service == RPC_GSS_SVC_INTEGRITY ||
2276 service == RPC_GSS_SVC_PRIVACY;
2277}
2278
2279bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2280{
2281 struct svc_cred *cr = &rqstp->rq_cred;
2282
2283 if (!cl->cl_mach_cred)
2284 return true;
2285 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2286 return false;
2287 if (!svc_rqst_integrity_protected(rqstp))
2288 return false;
2289 if (cl->cl_cred.cr_raw_principal)
2290 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2291 cr->cr_raw_principal);
2292 if (!cr->cr_principal)
2293 return false;
2294 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2295}
2296
2297static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2298{
2299 __be32 verf[2];
2300
2301 /*
2302 * This is opaque to client, so no need to byte-swap. Use
2303 * __force to keep sparse happy
2304 */
2305 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2306 verf[1] = (__force __be32)nn->clverifier_counter++;
2307 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2308}
2309
2310static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2311{
2312 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2313 clp->cl_clientid.cl_id = nn->clientid_counter++;
2314 gen_confirm(clp, nn);
2315}
2316
2317static struct nfs4_stid *
2318find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2319{
2320 struct nfs4_stid *ret;
2321
2322 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2323 if (!ret || !ret->sc_type)
2324 return NULL;
2325 return ret;
2326}
2327
2328static struct nfs4_stid *
2329find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2330{
2331 struct nfs4_stid *s;
2332
2333 spin_lock(&cl->cl_lock);
2334 s = find_stateid_locked(cl, t);
2335 if (s != NULL) {
2336 if (typemask & s->sc_type)
2337 refcount_inc(&s->sc_count);
2338 else
2339 s = NULL;
2340 }
2341 spin_unlock(&cl->cl_lock);
2342 return s;
2343}
2344
2345static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2346{
2347 struct nfsdfs_client *nc;
2348 nc = get_nfsdfs_client(inode);
2349 if (!nc)
2350 return NULL;
2351 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2352}
2353
2354static void seq_quote_mem(struct seq_file *m, char *data, int len)
2355{
2356 seq_printf(m, "\"");
2357 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2358 seq_printf(m, "\"");
2359}
2360
2361static const char *cb_state2str(int state)
2362{
2363 switch (state) {
2364 case NFSD4_CB_UP:
2365 return "UP";
2366 case NFSD4_CB_UNKNOWN:
2367 return "UNKNOWN";
2368 case NFSD4_CB_DOWN:
2369 return "DOWN";
2370 case NFSD4_CB_FAULT:
2371 return "FAULT";
2372 }
2373 return "UNDEFINED";
2374}
2375
2376static int client_info_show(struct seq_file *m, void *v)
2377{
2378 struct inode *inode = m->private;
2379 struct nfs4_client *clp;
2380 u64 clid;
2381
2382 clp = get_nfsdfs_clp(inode);
2383 if (!clp)
2384 return -ENXIO;
2385 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2386 seq_printf(m, "clientid: 0x%llx\n", clid);
2387 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2388 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2389 seq_puts(m, "status: confirmed\n");
2390 else
2391 seq_puts(m, "status: unconfirmed\n");
2392 seq_printf(m, "name: ");
2393 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2394 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2395 if (clp->cl_nii_domain.data) {
2396 seq_printf(m, "Implementation domain: ");
2397 seq_quote_mem(m, clp->cl_nii_domain.data,
2398 clp->cl_nii_domain.len);
2399 seq_printf(m, "\nImplementation name: ");
2400 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2401 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2402 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2403 }
2404 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2405 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2406 drop_client(clp);
2407
2408 return 0;
2409}
2410
2411static int client_info_open(struct inode *inode, struct file *file)
2412{
2413 return single_open(file, client_info_show, inode);
2414}
2415
2416static const struct file_operations client_info_fops = {
2417 .open = client_info_open,
2418 .read = seq_read,
2419 .llseek = seq_lseek,
2420 .release = single_release,
2421};
2422
2423static void *states_start(struct seq_file *s, loff_t *pos)
2424 __acquires(&clp->cl_lock)
2425{
2426 struct nfs4_client *clp = s->private;
2427 unsigned long id = *pos;
2428 void *ret;
2429
2430 spin_lock(&clp->cl_lock);
2431 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2432 *pos = id;
2433 return ret;
2434}
2435
2436static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2437{
2438 struct nfs4_client *clp = s->private;
2439 unsigned long id = *pos;
2440 void *ret;
2441
2442 id = *pos;
2443 id++;
2444 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2445 *pos = id;
2446 return ret;
2447}
2448
2449static void states_stop(struct seq_file *s, void *v)
2450 __releases(&clp->cl_lock)
2451{
2452 struct nfs4_client *clp = s->private;
2453
2454 spin_unlock(&clp->cl_lock);
2455}
2456
2457static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2458{
2459 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2460}
2461
2462static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2463{
2464 struct inode *inode = f->nf_inode;
2465
2466 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2467 MAJOR(inode->i_sb->s_dev),
2468 MINOR(inode->i_sb->s_dev),
2469 inode->i_ino);
2470}
2471
2472static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2473{
2474 seq_printf(s, "owner: ");
2475 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2476}
2477
2478static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2479{
2480 seq_printf(s, "0x%.8x", stid->si_generation);
2481 seq_printf(s, "%12phN", &stid->si_opaque);
2482}
2483
2484static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2485{
2486 struct nfs4_ol_stateid *ols;
2487 struct nfs4_file *nf;
2488 struct nfsd_file *file;
2489 struct nfs4_stateowner *oo;
2490 unsigned int access, deny;
2491
2492 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2493 return 0; /* XXX: or SEQ_SKIP? */
2494 ols = openlockstateid(st);
2495 oo = ols->st_stateowner;
2496 nf = st->sc_file;
2497 file = find_any_file(nf);
2498 if (!file)
2499 return 0;
2500
2501 seq_printf(s, "- ");
2502 nfs4_show_stateid(s, &st->sc_stateid);
2503 seq_printf(s, ": { type: open, ");
2504
2505 access = bmap_to_share_mode(ols->st_access_bmap);
2506 deny = bmap_to_share_mode(ols->st_deny_bmap);
2507
2508 seq_printf(s, "access: %s%s, ",
2509 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2510 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2511 seq_printf(s, "deny: %s%s, ",
2512 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2513 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2514
2515 nfs4_show_superblock(s, file);
2516 seq_printf(s, ", ");
2517 nfs4_show_fname(s, file);
2518 seq_printf(s, ", ");
2519 nfs4_show_owner(s, oo);
2520 seq_printf(s, " }\n");
2521 nfsd_file_put(file);
2522
2523 return 0;
2524}
2525
2526static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2527{
2528 struct nfs4_ol_stateid *ols;
2529 struct nfs4_file *nf;
2530 struct nfsd_file *file;
2531 struct nfs4_stateowner *oo;
2532
2533 ols = openlockstateid(st);
2534 oo = ols->st_stateowner;
2535 nf = st->sc_file;
2536 file = find_any_file(nf);
2537 if (!file)
2538 return 0;
2539
2540 seq_printf(s, "- ");
2541 nfs4_show_stateid(s, &st->sc_stateid);
2542 seq_printf(s, ": { type: lock, ");
2543
2544 /*
2545 * Note: a lock stateid isn't really the same thing as a lock,
2546 * it's the locking state held by one owner on a file, and there
2547 * may be multiple (or no) lock ranges associated with it.
2548 * (Same for the matter is true of open stateids.)
2549 */
2550
2551 nfs4_show_superblock(s, file);
2552 /* XXX: open stateid? */
2553 seq_printf(s, ", ");
2554 nfs4_show_fname(s, file);
2555 seq_printf(s, ", ");
2556 nfs4_show_owner(s, oo);
2557 seq_printf(s, " }\n");
2558 nfsd_file_put(file);
2559
2560 return 0;
2561}
2562
2563static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2564{
2565 struct nfs4_delegation *ds;
2566 struct nfs4_file *nf;
2567 struct nfsd_file *file;
2568
2569 ds = delegstateid(st);
2570 nf = st->sc_file;
2571 file = find_deleg_file(nf);
2572 if (!file)
2573 return 0;
2574
2575 seq_printf(s, "- ");
2576 nfs4_show_stateid(s, &st->sc_stateid);
2577 seq_printf(s, ": { type: deleg, ");
2578
2579 /* Kinda dead code as long as we only support read delegs: */
2580 seq_printf(s, "access: %s, ",
2581 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2582
2583 /* XXX: lease time, whether it's being recalled. */
2584
2585 nfs4_show_superblock(s, file);
2586 seq_printf(s, ", ");
2587 nfs4_show_fname(s, file);
2588 seq_printf(s, " }\n");
2589 nfsd_file_put(file);
2590
2591 return 0;
2592}
2593
2594static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2595{
2596 struct nfs4_layout_stateid *ls;
2597 struct nfsd_file *file;
2598
2599 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2600 file = ls->ls_file;
2601
2602 seq_printf(s, "- ");
2603 nfs4_show_stateid(s, &st->sc_stateid);
2604 seq_printf(s, ": { type: layout, ");
2605
2606 /* XXX: What else would be useful? */
2607
2608 nfs4_show_superblock(s, file);
2609 seq_printf(s, ", ");
2610 nfs4_show_fname(s, file);
2611 seq_printf(s, " }\n");
2612
2613 return 0;
2614}
2615
2616static int states_show(struct seq_file *s, void *v)
2617{
2618 struct nfs4_stid *st = v;
2619
2620 switch (st->sc_type) {
2621 case NFS4_OPEN_STID:
2622 return nfs4_show_open(s, st);
2623 case NFS4_LOCK_STID:
2624 return nfs4_show_lock(s, st);
2625 case NFS4_DELEG_STID:
2626 return nfs4_show_deleg(s, st);
2627 case NFS4_LAYOUT_STID:
2628 return nfs4_show_layout(s, st);
2629 default:
2630 return 0; /* XXX: or SEQ_SKIP? */
2631 }
2632 /* XXX: copy stateids? */
2633}
2634
2635static struct seq_operations states_seq_ops = {
2636 .start = states_start,
2637 .next = states_next,
2638 .stop = states_stop,
2639 .show = states_show
2640};
2641
2642static int client_states_open(struct inode *inode, struct file *file)
2643{
2644 struct seq_file *s;
2645 struct nfs4_client *clp;
2646 int ret;
2647
2648 clp = get_nfsdfs_clp(inode);
2649 if (!clp)
2650 return -ENXIO;
2651
2652 ret = seq_open(file, &states_seq_ops);
2653 if (ret)
2654 return ret;
2655 s = file->private_data;
2656 s->private = clp;
2657 return 0;
2658}
2659
2660static int client_opens_release(struct inode *inode, struct file *file)
2661{
2662 struct seq_file *m = file->private_data;
2663 struct nfs4_client *clp = m->private;
2664
2665 /* XXX: alternatively, we could get/drop in seq start/stop */
2666 drop_client(clp);
2667 return 0;
2668}
2669
2670static const struct file_operations client_states_fops = {
2671 .open = client_states_open,
2672 .read = seq_read,
2673 .llseek = seq_lseek,
2674 .release = client_opens_release,
2675};
2676
2677/*
2678 * Normally we refuse to destroy clients that are in use, but here the
2679 * administrator is telling us to just do it. We also want to wait
2680 * so the caller has a guarantee that the client's locks are gone by
2681 * the time the write returns:
2682 */
2683static void force_expire_client(struct nfs4_client *clp)
2684{
2685 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2686 bool already_expired;
2687
2688 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2689
2690 spin_lock(&nn->client_lock);
2691 clp->cl_time = 0;
2692 spin_unlock(&nn->client_lock);
2693
2694 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2695 spin_lock(&nn->client_lock);
2696 already_expired = list_empty(&clp->cl_lru);
2697 if (!already_expired)
2698 unhash_client_locked(clp);
2699 spin_unlock(&nn->client_lock);
2700
2701 if (!already_expired)
2702 expire_client(clp);
2703 else
2704 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2705}
2706
2707static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2708 size_t size, loff_t *pos)
2709{
2710 char *data;
2711 struct nfs4_client *clp;
2712
2713 data = simple_transaction_get(file, buf, size);
2714 if (IS_ERR(data))
2715 return PTR_ERR(data);
2716 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2717 return -EINVAL;
2718 clp = get_nfsdfs_clp(file_inode(file));
2719 if (!clp)
2720 return -ENXIO;
2721 force_expire_client(clp);
2722 drop_client(clp);
2723 return 7;
2724}
2725
2726static const struct file_operations client_ctl_fops = {
2727 .write = client_ctl_write,
2728 .release = simple_transaction_release,
2729};
2730
2731static const struct tree_descr client_files[] = {
2732 [0] = {"info", &client_info_fops, S_IRUSR},
2733 [1] = {"states", &client_states_fops, S_IRUSR},
2734 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2735 [3] = {""},
2736};
2737
2738static struct nfs4_client *create_client(struct xdr_netobj name,
2739 struct svc_rqst *rqstp, nfs4_verifier *verf)
2740{
2741 struct nfs4_client *clp;
2742 struct sockaddr *sa = svc_addr(rqstp);
2743 int ret;
2744 struct net *net = SVC_NET(rqstp);
2745 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2746 struct dentry *dentries[ARRAY_SIZE(client_files)];
2747
2748 clp = alloc_client(name);
2749 if (clp == NULL)
2750 return NULL;
2751
2752 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2753 if (ret) {
2754 free_client(clp);
2755 return NULL;
2756 }
2757 gen_clid(clp, nn);
2758 kref_init(&clp->cl_nfsdfs.cl_ref);
2759 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2760 clp->cl_time = ktime_get_boottime_seconds();
2761 clear_bit(0, &clp->cl_cb_slot_busy);
2762 copy_verf(clp, verf);
2763 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2764 clp->cl_cb_session = NULL;
2765 clp->net = net;
2766 clp->cl_nfsd_dentry = nfsd_client_mkdir(
2767 nn, &clp->cl_nfsdfs,
2768 clp->cl_clientid.cl_id - nn->clientid_base,
2769 client_files, dentries);
2770 clp->cl_nfsd_info_dentry = dentries[0];
2771 if (!clp->cl_nfsd_dentry) {
2772 free_client(clp);
2773 return NULL;
2774 }
2775 return clp;
2776}
2777
2778static void
2779add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2780{
2781 struct rb_node **new = &(root->rb_node), *parent = NULL;
2782 struct nfs4_client *clp;
2783
2784 while (*new) {
2785 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2786 parent = *new;
2787
2788 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2789 new = &((*new)->rb_left);
2790 else
2791 new = &((*new)->rb_right);
2792 }
2793
2794 rb_link_node(&new_clp->cl_namenode, parent, new);
2795 rb_insert_color(&new_clp->cl_namenode, root);
2796}
2797
2798static struct nfs4_client *
2799find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2800{
2801 int cmp;
2802 struct rb_node *node = root->rb_node;
2803 struct nfs4_client *clp;
2804
2805 while (node) {
2806 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2807 cmp = compare_blob(&clp->cl_name, name);
2808 if (cmp > 0)
2809 node = node->rb_left;
2810 else if (cmp < 0)
2811 node = node->rb_right;
2812 else
2813 return clp;
2814 }
2815 return NULL;
2816}
2817
2818static void
2819add_to_unconfirmed(struct nfs4_client *clp)
2820{
2821 unsigned int idhashval;
2822 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2823
2824 lockdep_assert_held(&nn->client_lock);
2825
2826 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2827 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2828 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2829 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2830 renew_client_locked(clp);
2831}
2832
2833static void
2834move_to_confirmed(struct nfs4_client *clp)
2835{
2836 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2837 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2838
2839 lockdep_assert_held(&nn->client_lock);
2840
2841 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2842 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2843 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2844 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2845 trace_nfsd_clid_confirmed(&clp->cl_clientid);
2846 renew_client_locked(clp);
2847}
2848
2849static struct nfs4_client *
2850find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2851{
2852 struct nfs4_client *clp;
2853 unsigned int idhashval = clientid_hashval(clid->cl_id);
2854
2855 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2856 if (same_clid(&clp->cl_clientid, clid)) {
2857 if ((bool)clp->cl_minorversion != sessions)
2858 return NULL;
2859 renew_client_locked(clp);
2860 return clp;
2861 }
2862 }
2863 return NULL;
2864}
2865
2866static struct nfs4_client *
2867find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2868{
2869 struct list_head *tbl = nn->conf_id_hashtbl;
2870
2871 lockdep_assert_held(&nn->client_lock);
2872 return find_client_in_id_table(tbl, clid, sessions);
2873}
2874
2875static struct nfs4_client *
2876find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2877{
2878 struct list_head *tbl = nn->unconf_id_hashtbl;
2879
2880 lockdep_assert_held(&nn->client_lock);
2881 return find_client_in_id_table(tbl, clid, sessions);
2882}
2883
2884static bool clp_used_exchangeid(struct nfs4_client *clp)
2885{
2886 return clp->cl_exchange_flags != 0;
2887}
2888
2889static struct nfs4_client *
2890find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2891{
2892 lockdep_assert_held(&nn->client_lock);
2893 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2894}
2895
2896static struct nfs4_client *
2897find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2898{
2899 lockdep_assert_held(&nn->client_lock);
2900 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2901}
2902
2903static void
2904gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2905{
2906 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2907 struct sockaddr *sa = svc_addr(rqstp);
2908 u32 scopeid = rpc_get_scope_id(sa);
2909 unsigned short expected_family;
2910
2911 /* Currently, we only support tcp and tcp6 for the callback channel */
2912 if (se->se_callback_netid_len == 3 &&
2913 !memcmp(se->se_callback_netid_val, "tcp", 3))
2914 expected_family = AF_INET;
2915 else if (se->se_callback_netid_len == 4 &&
2916 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2917 expected_family = AF_INET6;
2918 else
2919 goto out_err;
2920
2921 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2922 se->se_callback_addr_len,
2923 (struct sockaddr *)&conn->cb_addr,
2924 sizeof(conn->cb_addr));
2925
2926 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2927 goto out_err;
2928
2929 if (conn->cb_addr.ss_family == AF_INET6)
2930 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2931
2932 conn->cb_prog = se->se_callback_prog;
2933 conn->cb_ident = se->se_callback_ident;
2934 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2935 trace_nfsd_cb_args(clp, conn);
2936 return;
2937out_err:
2938 conn->cb_addr.ss_family = AF_UNSPEC;
2939 conn->cb_addrlen = 0;
2940 trace_nfsd_cb_nodelegs(clp);
2941 return;
2942}
2943
2944/*
2945 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2946 */
2947static void
2948nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2949{
2950 struct xdr_buf *buf = resp->xdr->buf;
2951 struct nfsd4_slot *slot = resp->cstate.slot;
2952 unsigned int base;
2953
2954 dprintk("--> %s slot %p\n", __func__, slot);
2955
2956 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2957 slot->sl_opcnt = resp->opcnt;
2958 slot->sl_status = resp->cstate.status;
2959 free_svc_cred(&slot->sl_cred);
2960 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2961
2962 if (!nfsd4_cache_this(resp)) {
2963 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2964 return;
2965 }
2966 slot->sl_flags |= NFSD4_SLOT_CACHED;
2967
2968 base = resp->cstate.data_offset;
2969 slot->sl_datalen = buf->len - base;
2970 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2971 WARN(1, "%s: sessions DRC could not cache compound\n",
2972 __func__);
2973 return;
2974}
2975
2976/*
2977 * Encode the replay sequence operation from the slot values.
2978 * If cachethis is FALSE encode the uncached rep error on the next
2979 * operation which sets resp->p and increments resp->opcnt for
2980 * nfs4svc_encode_compoundres.
2981 *
2982 */
2983static __be32
2984nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2985 struct nfsd4_compoundres *resp)
2986{
2987 struct nfsd4_op *op;
2988 struct nfsd4_slot *slot = resp->cstate.slot;
2989
2990 /* Encode the replayed sequence operation */
2991 op = &args->ops[resp->opcnt - 1];
2992 nfsd4_encode_operation(resp, op);
2993
2994 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2995 return op->status;
2996 if (args->opcnt == 1) {
2997 /*
2998 * The original operation wasn't a solo sequence--we
2999 * always cache those--so this retry must not match the
3000 * original:
3001 */
3002 op->status = nfserr_seq_false_retry;
3003 } else {
3004 op = &args->ops[resp->opcnt++];
3005 op->status = nfserr_retry_uncached_rep;
3006 nfsd4_encode_operation(resp, op);
3007 }
3008 return op->status;
3009}
3010
3011/*
3012 * The sequence operation is not cached because we can use the slot and
3013 * session values.
3014 */
3015static __be32
3016nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3017 struct nfsd4_sequence *seq)
3018{
3019 struct nfsd4_slot *slot = resp->cstate.slot;
3020 struct xdr_stream *xdr = resp->xdr;
3021 __be32 *p;
3022 __be32 status;
3023
3024 dprintk("--> %s slot %p\n", __func__, slot);
3025
3026 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3027 if (status)
3028 return status;
3029
3030 p = xdr_reserve_space(xdr, slot->sl_datalen);
3031 if (!p) {
3032 WARN_ON_ONCE(1);
3033 return nfserr_serverfault;
3034 }
3035 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3036 xdr_commit_encode(xdr);
3037
3038 resp->opcnt = slot->sl_opcnt;
3039 return slot->sl_status;
3040}
3041
3042/*
3043 * Set the exchange_id flags returned by the server.
3044 */
3045static void
3046nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3047{
3048#ifdef CONFIG_NFSD_PNFS
3049 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3050#else
3051 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3052#endif
3053
3054 /* Referrals are supported, Migration is not. */
3055 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3056
3057 /* set the wire flags to return to client. */
3058 clid->flags = new->cl_exchange_flags;
3059}
3060
3061static bool client_has_openowners(struct nfs4_client *clp)
3062{
3063 struct nfs4_openowner *oo;
3064
3065 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3066 if (!list_empty(&oo->oo_owner.so_stateids))
3067 return true;
3068 }
3069 return false;
3070}
3071
3072static bool client_has_state(struct nfs4_client *clp)
3073{
3074 return client_has_openowners(clp)
3075#ifdef CONFIG_NFSD_PNFS
3076 || !list_empty(&clp->cl_lo_states)
3077#endif
3078 || !list_empty(&clp->cl_delegations)
3079 || !list_empty(&clp->cl_sessions)
3080 || !list_empty(&clp->async_copies);
3081}
3082
3083static __be32 copy_impl_id(struct nfs4_client *clp,
3084 struct nfsd4_exchange_id *exid)
3085{
3086 if (!exid->nii_domain.data)
3087 return 0;
3088 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3089 if (!clp->cl_nii_domain.data)
3090 return nfserr_jukebox;
3091 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3092 if (!clp->cl_nii_name.data)
3093 return nfserr_jukebox;
3094 clp->cl_nii_time = exid->nii_time;
3095 return 0;
3096}
3097
3098__be32
3099nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3100 union nfsd4_op_u *u)
3101{
3102 struct nfsd4_exchange_id *exid = &u->exchange_id;
3103 struct nfs4_client *conf, *new;
3104 struct nfs4_client *unconf = NULL;
3105 __be32 status;
3106 char addr_str[INET6_ADDRSTRLEN];
3107 nfs4_verifier verf = exid->verifier;
3108 struct sockaddr *sa = svc_addr(rqstp);
3109 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3110 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3111
3112 rpc_ntop(sa, addr_str, sizeof(addr_str));
3113 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3114 "ip_addr=%s flags %x, spa_how %u\n",
3115 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3116 addr_str, exid->flags, exid->spa_how);
3117
3118 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3119 return nfserr_inval;
3120
3121 new = create_client(exid->clname, rqstp, &verf);
3122 if (new == NULL)
3123 return nfserr_jukebox;
3124 status = copy_impl_id(new, exid);
3125 if (status)
3126 goto out_nolock;
3127
3128 switch (exid->spa_how) {
3129 case SP4_MACH_CRED:
3130 exid->spo_must_enforce[0] = 0;
3131 exid->spo_must_enforce[1] = (
3132 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3133 1 << (OP_EXCHANGE_ID - 32) |
3134 1 << (OP_CREATE_SESSION - 32) |
3135 1 << (OP_DESTROY_SESSION - 32) |
3136 1 << (OP_DESTROY_CLIENTID - 32));
3137
3138 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3139 1 << (OP_OPEN_DOWNGRADE) |
3140 1 << (OP_LOCKU) |
3141 1 << (OP_DELEGRETURN));
3142
3143 exid->spo_must_allow[1] &= (
3144 1 << (OP_TEST_STATEID - 32) |
3145 1 << (OP_FREE_STATEID - 32));
3146 if (!svc_rqst_integrity_protected(rqstp)) {
3147 status = nfserr_inval;
3148 goto out_nolock;
3149 }
3150 /*
3151 * Sometimes userspace doesn't give us a principal.
3152 * Which is a bug, really. Anyway, we can't enforce
3153 * MACH_CRED in that case, better to give up now:
3154 */
3155 if (!new->cl_cred.cr_principal &&
3156 !new->cl_cred.cr_raw_principal) {
3157 status = nfserr_serverfault;
3158 goto out_nolock;
3159 }
3160 new->cl_mach_cred = true;
3161 break;
3162 case SP4_NONE:
3163 break;
3164 default: /* checked by xdr code */
3165 WARN_ON_ONCE(1);
3166 fallthrough;
3167 case SP4_SSV:
3168 status = nfserr_encr_alg_unsupp;
3169 goto out_nolock;
3170 }
3171
3172 /* Cases below refer to rfc 5661 section 18.35.4: */
3173 spin_lock(&nn->client_lock);
3174 conf = find_confirmed_client_by_name(&exid->clname, nn);
3175 if (conf) {
3176 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3177 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3178
3179 if (update) {
3180 if (!clp_used_exchangeid(conf)) { /* buggy client */
3181 status = nfserr_inval;
3182 goto out;
3183 }
3184 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3185 status = nfserr_wrong_cred;
3186 goto out;
3187 }
3188 if (!creds_match) { /* case 9 */
3189 status = nfserr_perm;
3190 goto out;
3191 }
3192 if (!verfs_match) { /* case 8 */
3193 status = nfserr_not_same;
3194 goto out;
3195 }
3196 /* case 6 */
3197 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3198 trace_nfsd_clid_confirmed_r(conf);
3199 goto out_copy;
3200 }
3201 if (!creds_match) { /* case 3 */
3202 if (client_has_state(conf)) {
3203 status = nfserr_clid_inuse;
3204 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3205 goto out;
3206 }
3207 goto out_new;
3208 }
3209 if (verfs_match) { /* case 2 */
3210 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3211 trace_nfsd_clid_confirmed_r(conf);
3212 goto out_copy;
3213 }
3214 /* case 5, client reboot */
3215 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3216 conf = NULL;
3217 goto out_new;
3218 }
3219
3220 if (update) { /* case 7 */
3221 status = nfserr_noent;
3222 goto out;
3223 }
3224
3225 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3226 if (unconf) /* case 4, possible retry or client restart */
3227 unhash_client_locked(unconf);
3228
3229 /* case 1, new owner ID */
3230 trace_nfsd_clid_fresh(new);
3231
3232out_new:
3233 if (conf) {
3234 status = mark_client_expired_locked(conf);
3235 if (status)
3236 goto out;
3237 trace_nfsd_clid_replaced(&conf->cl_clientid);
3238 }
3239 new->cl_minorversion = cstate->minorversion;
3240 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3241 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3242
3243 add_to_unconfirmed(new);
3244 swap(new, conf);
3245out_copy:
3246 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3247 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3248
3249 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3250 nfsd4_set_ex_flags(conf, exid);
3251
3252 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3253 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3254 status = nfs_ok;
3255
3256out:
3257 spin_unlock(&nn->client_lock);
3258out_nolock:
3259 if (new)
3260 expire_client(new);
3261 if (unconf) {
3262 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3263 expire_client(unconf);
3264 }
3265 return status;
3266}
3267
3268static __be32
3269check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3270{
3271 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3272 slot_seqid);
3273
3274 /* The slot is in use, and no response has been sent. */
3275 if (slot_inuse) {
3276 if (seqid == slot_seqid)
3277 return nfserr_jukebox;
3278 else
3279 return nfserr_seq_misordered;
3280 }
3281 /* Note unsigned 32-bit arithmetic handles wraparound: */
3282 if (likely(seqid == slot_seqid + 1))
3283 return nfs_ok;
3284 if (seqid == slot_seqid)
3285 return nfserr_replay_cache;
3286 return nfserr_seq_misordered;
3287}
3288
3289/*
3290 * Cache the create session result into the create session single DRC
3291 * slot cache by saving the xdr structure. sl_seqid has been set.
3292 * Do this for solo or embedded create session operations.
3293 */
3294static void
3295nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3296 struct nfsd4_clid_slot *slot, __be32 nfserr)
3297{
3298 slot->sl_status = nfserr;
3299 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3300}
3301
3302static __be32
3303nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3304 struct nfsd4_clid_slot *slot)
3305{
3306 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3307 return slot->sl_status;
3308}
3309
3310#define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3311 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3312 1 + /* MIN tag is length with zero, only length */ \
3313 3 + /* version, opcount, opcode */ \
3314 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3315 /* seqid, slotID, slotID, cache */ \
3316 4 ) * sizeof(__be32))
3317
3318#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3319 2 + /* verifier: AUTH_NULL, length 0 */\
3320 1 + /* status */ \
3321 1 + /* MIN tag is length with zero, only length */ \
3322 3 + /* opcount, opcode, opstatus*/ \
3323 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3324 /* seqid, slotID, slotID, slotID, status */ \
3325 5 ) * sizeof(__be32))
3326
3327static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3328{
3329 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3330
3331 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3332 return nfserr_toosmall;
3333 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3334 return nfserr_toosmall;
3335 ca->headerpadsz = 0;
3336 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3337 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3338 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3339 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3340 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3341 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3342 /*
3343 * Note decreasing slot size below client's request may make it
3344 * difficult for client to function correctly, whereas
3345 * decreasing the number of slots will (just?) affect
3346 * performance. When short on memory we therefore prefer to
3347 * decrease number of slots instead of their size. Clients that
3348 * request larger slots than they need will get poor results:
3349 * Note that we always allow at least one slot, because our
3350 * accounting is soft and provides no guarantees either way.
3351 */
3352 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3353
3354 return nfs_ok;
3355}
3356
3357/*
3358 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3359 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3360 */
3361#define RPC_MAX_HEADER_WITH_AUTH_SYS \
3362 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3363
3364#define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3365 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3366
3367#define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3368 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3369#define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3370 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3371 sizeof(__be32))
3372
3373static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3374{
3375 ca->headerpadsz = 0;
3376
3377 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3378 return nfserr_toosmall;
3379 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3380 return nfserr_toosmall;
3381 ca->maxresp_cached = 0;
3382 if (ca->maxops < 2)
3383 return nfserr_toosmall;
3384
3385 return nfs_ok;
3386}
3387
3388static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3389{
3390 switch (cbs->flavor) {
3391 case RPC_AUTH_NULL:
3392 case RPC_AUTH_UNIX:
3393 return nfs_ok;
3394 default:
3395 /*
3396 * GSS case: the spec doesn't allow us to return this
3397 * error. But it also doesn't allow us not to support
3398 * GSS.
3399 * I'd rather this fail hard than return some error the
3400 * client might think it can already handle:
3401 */
3402 return nfserr_encr_alg_unsupp;
3403 }
3404}
3405
3406__be32
3407nfsd4_create_session(struct svc_rqst *rqstp,
3408 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3409{
3410 struct nfsd4_create_session *cr_ses = &u->create_session;
3411 struct sockaddr *sa = svc_addr(rqstp);
3412 struct nfs4_client *conf, *unconf;
3413 struct nfs4_client *old = NULL;
3414 struct nfsd4_session *new;
3415 struct nfsd4_conn *conn;
3416 struct nfsd4_clid_slot *cs_slot = NULL;
3417 __be32 status = 0;
3418 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3419
3420 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3421 return nfserr_inval;
3422 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3423 if (status)
3424 return status;
3425 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3426 if (status)
3427 return status;
3428 status = check_backchannel_attrs(&cr_ses->back_channel);
3429 if (status)
3430 goto out_release_drc_mem;
3431 status = nfserr_jukebox;
3432 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3433 if (!new)
3434 goto out_release_drc_mem;
3435 conn = alloc_conn_from_crses(rqstp, cr_ses);
3436 if (!conn)
3437 goto out_free_session;
3438
3439 spin_lock(&nn->client_lock);
3440 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3441 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3442 WARN_ON_ONCE(conf && unconf);
3443
3444 if (conf) {
3445 status = nfserr_wrong_cred;
3446 if (!nfsd4_mach_creds_match(conf, rqstp))
3447 goto out_free_conn;
3448 cs_slot = &conf->cl_cs_slot;
3449 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3450 if (status) {
3451 if (status == nfserr_replay_cache)
3452 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3453 goto out_free_conn;
3454 }
3455 } else if (unconf) {
3456 status = nfserr_clid_inuse;
3457 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3458 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3459 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3460 goto out_free_conn;
3461 }
3462 status = nfserr_wrong_cred;
3463 if (!nfsd4_mach_creds_match(unconf, rqstp))
3464 goto out_free_conn;
3465 cs_slot = &unconf->cl_cs_slot;
3466 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3467 if (status) {
3468 /* an unconfirmed replay returns misordered */
3469 status = nfserr_seq_misordered;
3470 goto out_free_conn;
3471 }
3472 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3473 if (old) {
3474 status = mark_client_expired_locked(old);
3475 if (status) {
3476 old = NULL;
3477 goto out_free_conn;
3478 }
3479 trace_nfsd_clid_replaced(&old->cl_clientid);
3480 }
3481 move_to_confirmed(unconf);
3482 conf = unconf;
3483 } else {
3484 status = nfserr_stale_clientid;
3485 goto out_free_conn;
3486 }
3487 status = nfs_ok;
3488 /* Persistent sessions are not supported */
3489 cr_ses->flags &= ~SESSION4_PERSIST;
3490 /* Upshifting from TCP to RDMA is not supported */
3491 cr_ses->flags &= ~SESSION4_RDMA;
3492
3493 init_session(rqstp, new, conf, cr_ses);
3494 nfsd4_get_session_locked(new);
3495
3496 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3497 NFS4_MAX_SESSIONID_LEN);
3498 cs_slot->sl_seqid++;
3499 cr_ses->seqid = cs_slot->sl_seqid;
3500
3501 /* cache solo and embedded create sessions under the client_lock */
3502 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3503 spin_unlock(&nn->client_lock);
3504 if (conf == unconf)
3505 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3506 /* init connection and backchannel */
3507 nfsd4_init_conn(rqstp, conn, new);
3508 nfsd4_put_session(new);
3509 if (old)
3510 expire_client(old);
3511 return status;
3512out_free_conn:
3513 spin_unlock(&nn->client_lock);
3514 free_conn(conn);
3515 if (old)
3516 expire_client(old);
3517out_free_session:
3518 __free_session(new);
3519out_release_drc_mem:
3520 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3521 return status;
3522}
3523
3524static __be32 nfsd4_map_bcts_dir(u32 *dir)
3525{
3526 switch (*dir) {
3527 case NFS4_CDFC4_FORE:
3528 case NFS4_CDFC4_BACK:
3529 return nfs_ok;
3530 case NFS4_CDFC4_FORE_OR_BOTH:
3531 case NFS4_CDFC4_BACK_OR_BOTH:
3532 *dir = NFS4_CDFC4_BOTH;
3533 return nfs_ok;
3534 }
3535 return nfserr_inval;
3536}
3537
3538__be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3539 struct nfsd4_compound_state *cstate,
3540 union nfsd4_op_u *u)
3541{
3542 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3543 struct nfsd4_session *session = cstate->session;
3544 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3545 __be32 status;
3546
3547 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3548 if (status)
3549 return status;
3550 spin_lock(&nn->client_lock);
3551 session->se_cb_prog = bc->bc_cb_program;
3552 session->se_cb_sec = bc->bc_cb_sec;
3553 spin_unlock(&nn->client_lock);
3554
3555 nfsd4_probe_callback(session->se_client);
3556
3557 return nfs_ok;
3558}
3559
3560static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3561{
3562 struct nfsd4_conn *c;
3563
3564 list_for_each_entry(c, &s->se_conns, cn_persession) {
3565 if (c->cn_xprt == xpt) {
3566 return c;
3567 }
3568 }
3569 return NULL;
3570}
3571
3572static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3573 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3574{
3575 struct nfs4_client *clp = session->se_client;
3576 struct svc_xprt *xpt = rqst->rq_xprt;
3577 struct nfsd4_conn *c;
3578 __be32 status;
3579
3580 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3581 spin_lock(&clp->cl_lock);
3582 c = __nfsd4_find_conn(xpt, session);
3583 if (!c)
3584 status = nfserr_noent;
3585 else if (req == c->cn_flags)
3586 status = nfs_ok;
3587 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3588 c->cn_flags != NFS4_CDFC4_BACK)
3589 status = nfs_ok;
3590 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3591 c->cn_flags != NFS4_CDFC4_FORE)
3592 status = nfs_ok;
3593 else
3594 status = nfserr_inval;
3595 spin_unlock(&clp->cl_lock);
3596 if (status == nfs_ok && conn)
3597 *conn = c;
3598 return status;
3599}
3600
3601__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3602 struct nfsd4_compound_state *cstate,
3603 union nfsd4_op_u *u)
3604{
3605 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3606 __be32 status;
3607 struct nfsd4_conn *conn;
3608 struct nfsd4_session *session;
3609 struct net *net = SVC_NET(rqstp);
3610 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3611
3612 if (!nfsd4_last_compound_op(rqstp))
3613 return nfserr_not_only_op;
3614 spin_lock(&nn->client_lock);
3615 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3616 spin_unlock(&nn->client_lock);
3617 if (!session)
3618 goto out_no_session;
3619 status = nfserr_wrong_cred;
3620 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3621 goto out;
3622 status = nfsd4_match_existing_connection(rqstp, session,
3623 bcts->dir, &conn);
3624 if (status == nfs_ok) {
3625 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
3626 bcts->dir == NFS4_CDFC4_BACK)
3627 conn->cn_flags |= NFS4_CDFC4_BACK;
3628 nfsd4_probe_callback(session->se_client);
3629 goto out;
3630 }
3631 if (status == nfserr_inval)
3632 goto out;
3633 status = nfsd4_map_bcts_dir(&bcts->dir);
3634 if (status)
3635 goto out;
3636 conn = alloc_conn(rqstp, bcts->dir);
3637 status = nfserr_jukebox;
3638 if (!conn)
3639 goto out;
3640 nfsd4_init_conn(rqstp, conn, session);
3641 status = nfs_ok;
3642out:
3643 nfsd4_put_session(session);
3644out_no_session:
3645 return status;
3646}
3647
3648static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3649{
3650 if (!cstate->session)
3651 return false;
3652 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3653}
3654
3655__be32
3656nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3657 union nfsd4_op_u *u)
3658{
3659 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3660 struct nfsd4_session *ses;
3661 __be32 status;
3662 int ref_held_by_me = 0;
3663 struct net *net = SVC_NET(r);
3664 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3665
3666 status = nfserr_not_only_op;
3667 if (nfsd4_compound_in_session(cstate, sessionid)) {
3668 if (!nfsd4_last_compound_op(r))
3669 goto out;
3670 ref_held_by_me++;
3671 }
3672 dump_sessionid(__func__, sessionid);
3673 spin_lock(&nn->client_lock);
3674 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3675 if (!ses)
3676 goto out_client_lock;
3677 status = nfserr_wrong_cred;
3678 if (!nfsd4_mach_creds_match(ses->se_client, r))
3679 goto out_put_session;
3680 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3681 if (status)
3682 goto out_put_session;
3683 unhash_session(ses);
3684 spin_unlock(&nn->client_lock);
3685
3686 nfsd4_probe_callback_sync(ses->se_client);
3687
3688 spin_lock(&nn->client_lock);
3689 status = nfs_ok;
3690out_put_session:
3691 nfsd4_put_session_locked(ses);
3692out_client_lock:
3693 spin_unlock(&nn->client_lock);
3694out:
3695 return status;
3696}
3697
3698static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3699{
3700 struct nfs4_client *clp = ses->se_client;
3701 struct nfsd4_conn *c;
3702 __be32 status = nfs_ok;
3703 int ret;
3704
3705 spin_lock(&clp->cl_lock);
3706 c = __nfsd4_find_conn(new->cn_xprt, ses);
3707 if (c)
3708 goto out_free;
3709 status = nfserr_conn_not_bound_to_session;
3710 if (clp->cl_mach_cred)
3711 goto out_free;
3712 __nfsd4_hash_conn(new, ses);
3713 spin_unlock(&clp->cl_lock);
3714 ret = nfsd4_register_conn(new);
3715 if (ret)
3716 /* oops; xprt is already down: */
3717 nfsd4_conn_lost(&new->cn_xpt_user);
3718 return nfs_ok;
3719out_free:
3720 spin_unlock(&clp->cl_lock);
3721 free_conn(new);
3722 return status;
3723}
3724
3725static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3726{
3727 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3728
3729 return args->opcnt > session->se_fchannel.maxops;
3730}
3731
3732static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3733 struct nfsd4_session *session)
3734{
3735 struct xdr_buf *xb = &rqstp->rq_arg;
3736
3737 return xb->len > session->se_fchannel.maxreq_sz;
3738}
3739
3740static bool replay_matches_cache(struct svc_rqst *rqstp,
3741 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3742{
3743 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3744
3745 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3746 (bool)seq->cachethis)
3747 return false;
3748 /*
3749 * If there's an error then the reply can have fewer ops than
3750 * the call.
3751 */
3752 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3753 return false;
3754 /*
3755 * But if we cached a reply with *more* ops than the call you're
3756 * sending us now, then this new call is clearly not really a
3757 * replay of the old one:
3758 */
3759 if (slot->sl_opcnt > argp->opcnt)
3760 return false;
3761 /* This is the only check explicitly called by spec: */
3762 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3763 return false;
3764 /*
3765 * There may be more comparisons we could actually do, but the
3766 * spec doesn't require us to catch every case where the calls
3767 * don't match (that would require caching the call as well as
3768 * the reply), so we don't bother.
3769 */
3770 return true;
3771}
3772
3773__be32
3774nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3775 union nfsd4_op_u *u)
3776{
3777 struct nfsd4_sequence *seq = &u->sequence;
3778 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3779 struct xdr_stream *xdr = resp->xdr;
3780 struct nfsd4_session *session;
3781 struct nfs4_client *clp;
3782 struct nfsd4_slot *slot;
3783 struct nfsd4_conn *conn;
3784 __be32 status;
3785 int buflen;
3786 struct net *net = SVC_NET(rqstp);
3787 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3788
3789 if (resp->opcnt != 1)
3790 return nfserr_sequence_pos;
3791
3792 /*
3793 * Will be either used or freed by nfsd4_sequence_check_conn
3794 * below.
3795 */
3796 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3797 if (!conn)
3798 return nfserr_jukebox;
3799
3800 spin_lock(&nn->client_lock);
3801 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3802 if (!session)
3803 goto out_no_session;
3804 clp = session->se_client;
3805
3806 status = nfserr_too_many_ops;
3807 if (nfsd4_session_too_many_ops(rqstp, session))
3808 goto out_put_session;
3809
3810 status = nfserr_req_too_big;
3811 if (nfsd4_request_too_big(rqstp, session))
3812 goto out_put_session;
3813
3814 status = nfserr_badslot;
3815 if (seq->slotid >= session->se_fchannel.maxreqs)
3816 goto out_put_session;
3817
3818 slot = session->se_slots[seq->slotid];
3819 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3820
3821 /* We do not negotiate the number of slots yet, so set the
3822 * maxslots to the session maxreqs which is used to encode
3823 * sr_highest_slotid and the sr_target_slot id to maxslots */
3824 seq->maxslots = session->se_fchannel.maxreqs;
3825
3826 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3827 slot->sl_flags & NFSD4_SLOT_INUSE);
3828 if (status == nfserr_replay_cache) {
3829 status = nfserr_seq_misordered;
3830 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3831 goto out_put_session;
3832 status = nfserr_seq_false_retry;
3833 if (!replay_matches_cache(rqstp, seq, slot))
3834 goto out_put_session;
3835 cstate->slot = slot;
3836 cstate->session = session;
3837 cstate->clp = clp;
3838 /* Return the cached reply status and set cstate->status
3839 * for nfsd4_proc_compound processing */
3840 status = nfsd4_replay_cache_entry(resp, seq);
3841 cstate->status = nfserr_replay_cache;
3842 goto out;
3843 }
3844 if (status)
3845 goto out_put_session;
3846
3847 status = nfsd4_sequence_check_conn(conn, session);
3848 conn = NULL;
3849 if (status)
3850 goto out_put_session;
3851
3852 buflen = (seq->cachethis) ?
3853 session->se_fchannel.maxresp_cached :
3854 session->se_fchannel.maxresp_sz;
3855 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3856 nfserr_rep_too_big;
3857 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3858 goto out_put_session;
3859 svc_reserve(rqstp, buflen);
3860
3861 status = nfs_ok;
3862 /* Success! bump slot seqid */
3863 slot->sl_seqid = seq->seqid;
3864 slot->sl_flags |= NFSD4_SLOT_INUSE;
3865 if (seq->cachethis)
3866 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3867 else
3868 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3869
3870 cstate->slot = slot;
3871 cstate->session = session;
3872 cstate->clp = clp;
3873
3874out:
3875 switch (clp->cl_cb_state) {
3876 case NFSD4_CB_DOWN:
3877 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3878 break;
3879 case NFSD4_CB_FAULT:
3880 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3881 break;
3882 default:
3883 seq->status_flags = 0;
3884 }
3885 if (!list_empty(&clp->cl_revoked))
3886 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3887out_no_session:
3888 if (conn)
3889 free_conn(conn);
3890 spin_unlock(&nn->client_lock);
3891 return status;
3892out_put_session:
3893 nfsd4_put_session_locked(session);
3894 goto out_no_session;
3895}
3896
3897void
3898nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3899{
3900 struct nfsd4_compound_state *cs = &resp->cstate;
3901
3902 if (nfsd4_has_session(cs)) {
3903 if (cs->status != nfserr_replay_cache) {
3904 nfsd4_store_cache_entry(resp);
3905 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3906 }
3907 /* Drop session reference that was taken in nfsd4_sequence() */
3908 nfsd4_put_session(cs->session);
3909 } else if (cs->clp)
3910 put_client_renew(cs->clp);
3911}
3912
3913__be32
3914nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3915 struct nfsd4_compound_state *cstate,
3916 union nfsd4_op_u *u)
3917{
3918 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3919 struct nfs4_client *conf, *unconf;
3920 struct nfs4_client *clp = NULL;
3921 __be32 status = 0;
3922 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3923
3924 spin_lock(&nn->client_lock);
3925 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3926 conf = find_confirmed_client(&dc->clientid, true, nn);
3927 WARN_ON_ONCE(conf && unconf);
3928
3929 if (conf) {
3930 if (client_has_state(conf)) {
3931 status = nfserr_clientid_busy;
3932 goto out;
3933 }
3934 status = mark_client_expired_locked(conf);
3935 if (status)
3936 goto out;
3937 clp = conf;
3938 } else if (unconf)
3939 clp = unconf;
3940 else {
3941 status = nfserr_stale_clientid;
3942 goto out;
3943 }
3944 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3945 clp = NULL;
3946 status = nfserr_wrong_cred;
3947 goto out;
3948 }
3949 trace_nfsd_clid_destroyed(&clp->cl_clientid);
3950 unhash_client_locked(clp);
3951out:
3952 spin_unlock(&nn->client_lock);
3953 if (clp)
3954 expire_client(clp);
3955 return status;
3956}
3957
3958__be32
3959nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3960 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3961{
3962 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3963 struct nfs4_client *clp = cstate->clp;
3964 __be32 status = 0;
3965
3966 if (rc->rca_one_fs) {
3967 if (!cstate->current_fh.fh_dentry)
3968 return nfserr_nofilehandle;
3969 /*
3970 * We don't take advantage of the rca_one_fs case.
3971 * That's OK, it's optional, we can safely ignore it.
3972 */
3973 return nfs_ok;
3974 }
3975
3976 status = nfserr_complete_already;
3977 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
3978 goto out;
3979
3980 status = nfserr_stale_clientid;
3981 if (is_client_expired(clp))
3982 /*
3983 * The following error isn't really legal.
3984 * But we only get here if the client just explicitly
3985 * destroyed the client. Surely it no longer cares what
3986 * error it gets back on an operation for the dead
3987 * client.
3988 */
3989 goto out;
3990
3991 status = nfs_ok;
3992 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
3993 nfsd4_client_record_create(clp);
3994 inc_reclaim_complete(clp);
3995out:
3996 return status;
3997}
3998
3999__be32
4000nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4001 union nfsd4_op_u *u)
4002{
4003 struct nfsd4_setclientid *setclid = &u->setclientid;
4004 struct xdr_netobj clname = setclid->se_name;
4005 nfs4_verifier clverifier = setclid->se_verf;
4006 struct nfs4_client *conf, *new;
4007 struct nfs4_client *unconf = NULL;
4008 __be32 status;
4009 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4010
4011 new = create_client(clname, rqstp, &clverifier);
4012 if (new == NULL)
4013 return nfserr_jukebox;
4014 spin_lock(&nn->client_lock);
4015 conf = find_confirmed_client_by_name(&clname, nn);
4016 if (conf && client_has_state(conf)) {
4017 status = nfserr_clid_inuse;
4018 if (clp_used_exchangeid(conf))
4019 goto out;
4020 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4021 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4022 goto out;
4023 }
4024 }
4025 unconf = find_unconfirmed_client_by_name(&clname, nn);
4026 if (unconf)
4027 unhash_client_locked(unconf);
4028 if (conf) {
4029 if (same_verf(&conf->cl_verifier, &clverifier)) {
4030 copy_clid(new, conf);
4031 gen_confirm(new, nn);
4032 } else
4033 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4034 &clverifier);
4035 } else
4036 trace_nfsd_clid_fresh(new);
4037 new->cl_minorversion = 0;
4038 gen_callback(new, setclid, rqstp);
4039 add_to_unconfirmed(new);
4040 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4041 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4042 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4043 new = NULL;
4044 status = nfs_ok;
4045out:
4046 spin_unlock(&nn->client_lock);
4047 if (new)
4048 free_client(new);
4049 if (unconf) {
4050 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4051 expire_client(unconf);
4052 }
4053 return status;
4054}
4055
4056__be32
4057nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4058 struct nfsd4_compound_state *cstate,
4059 union nfsd4_op_u *u)
4060{
4061 struct nfsd4_setclientid_confirm *setclientid_confirm =
4062 &u->setclientid_confirm;
4063 struct nfs4_client *conf, *unconf;
4064 struct nfs4_client *old = NULL;
4065 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4066 clientid_t * clid = &setclientid_confirm->sc_clientid;
4067 __be32 status;
4068 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4069
4070 if (STALE_CLIENTID(clid, nn))
4071 return nfserr_stale_clientid;
4072
4073 spin_lock(&nn->client_lock);
4074 conf = find_confirmed_client(clid, false, nn);
4075 unconf = find_unconfirmed_client(clid, false, nn);
4076 /*
4077 * We try hard to give out unique clientid's, so if we get an
4078 * attempt to confirm the same clientid with a different cred,
4079 * the client may be buggy; this should never happen.
4080 *
4081 * Nevertheless, RFC 7530 recommends INUSE for this case:
4082 */
4083 status = nfserr_clid_inuse;
4084 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4085 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4086 goto out;
4087 }
4088 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4089 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4090 goto out;
4091 }
4092 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4093 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4094 status = nfs_ok;
4095 } else
4096 status = nfserr_stale_clientid;
4097 goto out;
4098 }
4099 status = nfs_ok;
4100 if (conf) {
4101 old = unconf;
4102 unhash_client_locked(old);
4103 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4104 } else {
4105 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4106 if (old) {
4107 status = nfserr_clid_inuse;
4108 if (client_has_state(old)
4109 && !same_creds(&unconf->cl_cred,
4110 &old->cl_cred))
4111 goto out;
4112 status = mark_client_expired_locked(old);
4113 if (status) {
4114 old = NULL;
4115 goto out;
4116 }
4117 trace_nfsd_clid_replaced(&old->cl_clientid);
4118 }
4119 move_to_confirmed(unconf);
4120 conf = unconf;
4121 }
4122 get_client_locked(conf);
4123 spin_unlock(&nn->client_lock);
4124 if (conf == unconf)
4125 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4126 nfsd4_probe_callback(conf);
4127 spin_lock(&nn->client_lock);
4128 put_client_renew_locked(conf);
4129out:
4130 spin_unlock(&nn->client_lock);
4131 if (old)
4132 expire_client(old);
4133 return status;
4134}
4135
4136static struct nfs4_file *nfsd4_alloc_file(void)
4137{
4138 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4139}
4140
4141/* OPEN Share state helper functions */
4142static void nfsd4_init_file(struct svc_fh *fh, unsigned int hashval,
4143 struct nfs4_file *fp)
4144{
4145 lockdep_assert_held(&state_lock);
4146
4147 refcount_set(&fp->fi_ref, 1);
4148 spin_lock_init(&fp->fi_lock);
4149 INIT_LIST_HEAD(&fp->fi_stateids);
4150 INIT_LIST_HEAD(&fp->fi_delegations);
4151 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4152 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4153 fp->fi_deleg_file = NULL;
4154 fp->fi_had_conflict = false;
4155 fp->fi_share_deny = 0;
4156 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4157 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4158 fp->fi_aliased = false;
4159 fp->fi_inode = d_inode(fh->fh_dentry);
4160#ifdef CONFIG_NFSD_PNFS
4161 INIT_LIST_HEAD(&fp->fi_lo_states);
4162 atomic_set(&fp->fi_lo_recalls, 0);
4163#endif
4164 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4165}
4166
4167void
4168nfsd4_free_slabs(void)
4169{
4170 kmem_cache_destroy(client_slab);
4171 kmem_cache_destroy(openowner_slab);
4172 kmem_cache_destroy(lockowner_slab);
4173 kmem_cache_destroy(file_slab);
4174 kmem_cache_destroy(stateid_slab);
4175 kmem_cache_destroy(deleg_slab);
4176 kmem_cache_destroy(odstate_slab);
4177}
4178
4179int
4180nfsd4_init_slabs(void)
4181{
4182 client_slab = kmem_cache_create("nfsd4_clients",
4183 sizeof(struct nfs4_client), 0, 0, NULL);
4184 if (client_slab == NULL)
4185 goto out;
4186 openowner_slab = kmem_cache_create("nfsd4_openowners",
4187 sizeof(struct nfs4_openowner), 0, 0, NULL);
4188 if (openowner_slab == NULL)
4189 goto out_free_client_slab;
4190 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4191 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4192 if (lockowner_slab == NULL)
4193 goto out_free_openowner_slab;
4194 file_slab = kmem_cache_create("nfsd4_files",
4195 sizeof(struct nfs4_file), 0, 0, NULL);
4196 if (file_slab == NULL)
4197 goto out_free_lockowner_slab;
4198 stateid_slab = kmem_cache_create("nfsd4_stateids",
4199 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4200 if (stateid_slab == NULL)
4201 goto out_free_file_slab;
4202 deleg_slab = kmem_cache_create("nfsd4_delegations",
4203 sizeof(struct nfs4_delegation), 0, 0, NULL);
4204 if (deleg_slab == NULL)
4205 goto out_free_stateid_slab;
4206 odstate_slab = kmem_cache_create("nfsd4_odstate",
4207 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4208 if (odstate_slab == NULL)
4209 goto out_free_deleg_slab;
4210 return 0;
4211
4212out_free_deleg_slab:
4213 kmem_cache_destroy(deleg_slab);
4214out_free_stateid_slab:
4215 kmem_cache_destroy(stateid_slab);
4216out_free_file_slab:
4217 kmem_cache_destroy(file_slab);
4218out_free_lockowner_slab:
4219 kmem_cache_destroy(lockowner_slab);
4220out_free_openowner_slab:
4221 kmem_cache_destroy(openowner_slab);
4222out_free_client_slab:
4223 kmem_cache_destroy(client_slab);
4224out:
4225 return -ENOMEM;
4226}
4227
4228static void init_nfs4_replay(struct nfs4_replay *rp)
4229{
4230 rp->rp_status = nfserr_serverfault;
4231 rp->rp_buflen = 0;
4232 rp->rp_buf = rp->rp_ibuf;
4233 mutex_init(&rp->rp_mutex);
4234}
4235
4236static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4237 struct nfs4_stateowner *so)
4238{
4239 if (!nfsd4_has_session(cstate)) {
4240 mutex_lock(&so->so_replay.rp_mutex);
4241 cstate->replay_owner = nfs4_get_stateowner(so);
4242 }
4243}
4244
4245void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4246{
4247 struct nfs4_stateowner *so = cstate->replay_owner;
4248
4249 if (so != NULL) {
4250 cstate->replay_owner = NULL;
4251 mutex_unlock(&so->so_replay.rp_mutex);
4252 nfs4_put_stateowner(so);
4253 }
4254}
4255
4256static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4257{
4258 struct nfs4_stateowner *sop;
4259
4260 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4261 if (!sop)
4262 return NULL;
4263
4264 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4265 if (!sop->so_owner.data) {
4266 kmem_cache_free(slab, sop);
4267 return NULL;
4268 }
4269
4270 INIT_LIST_HEAD(&sop->so_stateids);
4271 sop->so_client = clp;
4272 init_nfs4_replay(&sop->so_replay);
4273 atomic_set(&sop->so_count, 1);
4274 return sop;
4275}
4276
4277static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4278{
4279 lockdep_assert_held(&clp->cl_lock);
4280
4281 list_add(&oo->oo_owner.so_strhash,
4282 &clp->cl_ownerstr_hashtbl[strhashval]);
4283 list_add(&oo->oo_perclient, &clp->cl_openowners);
4284}
4285
4286static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4287{
4288 unhash_openowner_locked(openowner(so));
4289}
4290
4291static void nfs4_free_openowner(struct nfs4_stateowner *so)
4292{
4293 struct nfs4_openowner *oo = openowner(so);
4294
4295 kmem_cache_free(openowner_slab, oo);
4296}
4297
4298static const struct nfs4_stateowner_operations openowner_ops = {
4299 .so_unhash = nfs4_unhash_openowner,
4300 .so_free = nfs4_free_openowner,
4301};
4302
4303static struct nfs4_ol_stateid *
4304nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4305{
4306 struct nfs4_ol_stateid *local, *ret = NULL;
4307 struct nfs4_openowner *oo = open->op_openowner;
4308
4309 lockdep_assert_held(&fp->fi_lock);
4310
4311 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4312 /* ignore lock owners */
4313 if (local->st_stateowner->so_is_open_owner == 0)
4314 continue;
4315 if (local->st_stateowner != &oo->oo_owner)
4316 continue;
4317 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4318 ret = local;
4319 refcount_inc(&ret->st_stid.sc_count);
4320 break;
4321 }
4322 }
4323 return ret;
4324}
4325
4326static __be32
4327nfsd4_verify_open_stid(struct nfs4_stid *s)
4328{
4329 __be32 ret = nfs_ok;
4330
4331 switch (s->sc_type) {
4332 default:
4333 break;
4334 case 0:
4335 case NFS4_CLOSED_STID:
4336 case NFS4_CLOSED_DELEG_STID:
4337 ret = nfserr_bad_stateid;
4338 break;
4339 case NFS4_REVOKED_DELEG_STID:
4340 ret = nfserr_deleg_revoked;
4341 }
4342 return ret;
4343}
4344
4345/* Lock the stateid st_mutex, and deal with races with CLOSE */
4346static __be32
4347nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4348{
4349 __be32 ret;
4350
4351 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4352 ret = nfsd4_verify_open_stid(&stp->st_stid);
4353 if (ret != nfs_ok)
4354 mutex_unlock(&stp->st_mutex);
4355 return ret;
4356}
4357
4358static struct nfs4_ol_stateid *
4359nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4360{
4361 struct nfs4_ol_stateid *stp;
4362 for (;;) {
4363 spin_lock(&fp->fi_lock);
4364 stp = nfsd4_find_existing_open(fp, open);
4365 spin_unlock(&fp->fi_lock);
4366 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4367 break;
4368 nfs4_put_stid(&stp->st_stid);
4369 }
4370 return stp;
4371}
4372
4373static struct nfs4_openowner *
4374alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4375 struct nfsd4_compound_state *cstate)
4376{
4377 struct nfs4_client *clp = cstate->clp;
4378 struct nfs4_openowner *oo, *ret;
4379
4380 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4381 if (!oo)
4382 return NULL;
4383 oo->oo_owner.so_ops = &openowner_ops;
4384 oo->oo_owner.so_is_open_owner = 1;
4385 oo->oo_owner.so_seqid = open->op_seqid;
4386 oo->oo_flags = 0;
4387 if (nfsd4_has_session(cstate))
4388 oo->oo_flags |= NFS4_OO_CONFIRMED;
4389 oo->oo_time = 0;
4390 oo->oo_last_closed_stid = NULL;
4391 INIT_LIST_HEAD(&oo->oo_close_lru);
4392 spin_lock(&clp->cl_lock);
4393 ret = find_openstateowner_str_locked(strhashval, open, clp);
4394 if (ret == NULL) {
4395 hash_openowner(oo, clp, strhashval);
4396 ret = oo;
4397 } else
4398 nfs4_free_stateowner(&oo->oo_owner);
4399
4400 spin_unlock(&clp->cl_lock);
4401 return ret;
4402}
4403
4404static struct nfs4_ol_stateid *
4405init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4406{
4407
4408 struct nfs4_openowner *oo = open->op_openowner;
4409 struct nfs4_ol_stateid *retstp = NULL;
4410 struct nfs4_ol_stateid *stp;
4411
4412 stp = open->op_stp;
4413 /* We are moving these outside of the spinlocks to avoid the warnings */
4414 mutex_init(&stp->st_mutex);
4415 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4416
4417retry:
4418 spin_lock(&oo->oo_owner.so_client->cl_lock);
4419 spin_lock(&fp->fi_lock);
4420
4421 retstp = nfsd4_find_existing_open(fp, open);
4422 if (retstp)
4423 goto out_unlock;
4424
4425 open->op_stp = NULL;
4426 refcount_inc(&stp->st_stid.sc_count);
4427 stp->st_stid.sc_type = NFS4_OPEN_STID;
4428 INIT_LIST_HEAD(&stp->st_locks);
4429 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4430 get_nfs4_file(fp);
4431 stp->st_stid.sc_file = fp;
4432 stp->st_access_bmap = 0;
4433 stp->st_deny_bmap = 0;
4434 stp->st_openstp = NULL;
4435 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4436 list_add(&stp->st_perfile, &fp->fi_stateids);
4437
4438out_unlock:
4439 spin_unlock(&fp->fi_lock);
4440 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4441 if (retstp) {
4442 /* Handle races with CLOSE */
4443 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4444 nfs4_put_stid(&retstp->st_stid);
4445 goto retry;
4446 }
4447 /* To keep mutex tracking happy */
4448 mutex_unlock(&stp->st_mutex);
4449 stp = retstp;
4450 }
4451 return stp;
4452}
4453
4454/*
4455 * In the 4.0 case we need to keep the owners around a little while to handle
4456 * CLOSE replay. We still do need to release any file access that is held by
4457 * them before returning however.
4458 */
4459static void
4460move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4461{
4462 struct nfs4_ol_stateid *last;
4463 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4464 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4465 nfsd_net_id);
4466
4467 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4468
4469 /*
4470 * We know that we hold one reference via nfsd4_close, and another
4471 * "persistent" reference for the client. If the refcount is higher
4472 * than 2, then there are still calls in progress that are using this
4473 * stateid. We can't put the sc_file reference until they are finished.
4474 * Wait for the refcount to drop to 2. Since it has been unhashed,
4475 * there should be no danger of the refcount going back up again at
4476 * this point.
4477 */
4478 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4479
4480 release_all_access(s);
4481 if (s->st_stid.sc_file) {
4482 put_nfs4_file(s->st_stid.sc_file);
4483 s->st_stid.sc_file = NULL;
4484 }
4485
4486 spin_lock(&nn->client_lock);
4487 last = oo->oo_last_closed_stid;
4488 oo->oo_last_closed_stid = s;
4489 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4490 oo->oo_time = ktime_get_boottime_seconds();
4491 spin_unlock(&nn->client_lock);
4492 if (last)
4493 nfs4_put_stid(&last->st_stid);
4494}
4495
4496/* search file_hashtbl[] for file */
4497static struct nfs4_file *
4498find_file_locked(struct svc_fh *fh, unsigned int hashval)
4499{
4500 struct nfs4_file *fp;
4501
4502 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4503 lockdep_is_held(&state_lock)) {
4504 if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4505 if (refcount_inc_not_zero(&fp->fi_ref))
4506 return fp;
4507 }
4508 }
4509 return NULL;
4510}
4511
4512static struct nfs4_file *insert_file(struct nfs4_file *new, struct svc_fh *fh,
4513 unsigned int hashval)
4514{
4515 struct nfs4_file *fp;
4516 struct nfs4_file *ret = NULL;
4517 bool alias_found = false;
4518
4519 spin_lock(&state_lock);
4520 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4521 lockdep_is_held(&state_lock)) {
4522 if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4523 if (refcount_inc_not_zero(&fp->fi_ref))
4524 ret = fp;
4525 } else if (d_inode(fh->fh_dentry) == fp->fi_inode)
4526 fp->fi_aliased = alias_found = true;
4527 }
4528 if (likely(ret == NULL)) {
4529 nfsd4_init_file(fh, hashval, new);
4530 new->fi_aliased = alias_found;
4531 ret = new;
4532 }
4533 spin_unlock(&state_lock);
4534 return ret;
4535}
4536
4537static struct nfs4_file * find_file(struct svc_fh *fh)
4538{
4539 struct nfs4_file *fp;
4540 unsigned int hashval = file_hashval(fh);
4541
4542 rcu_read_lock();
4543 fp = find_file_locked(fh, hashval);
4544 rcu_read_unlock();
4545 return fp;
4546}
4547
4548static struct nfs4_file *
4549find_or_add_file(struct nfs4_file *new, struct svc_fh *fh)
4550{
4551 struct nfs4_file *fp;
4552 unsigned int hashval = file_hashval(fh);
4553
4554 rcu_read_lock();
4555 fp = find_file_locked(fh, hashval);
4556 rcu_read_unlock();
4557 if (fp)
4558 return fp;
4559
4560 return insert_file(new, fh, hashval);
4561}
4562
4563/*
4564 * Called to check deny when READ with all zero stateid or
4565 * WRITE with all zero or all one stateid
4566 */
4567static __be32
4568nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4569{
4570 struct nfs4_file *fp;
4571 __be32 ret = nfs_ok;
4572
4573 fp = find_file(current_fh);
4574 if (!fp)
4575 return ret;
4576 /* Check for conflicting share reservations */
4577 spin_lock(&fp->fi_lock);
4578 if (fp->fi_share_deny & deny_type)
4579 ret = nfserr_locked;
4580 spin_unlock(&fp->fi_lock);
4581 put_nfs4_file(fp);
4582 return ret;
4583}
4584
4585static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4586{
4587 struct nfs4_delegation *dp = cb_to_delegation(cb);
4588 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4589 nfsd_net_id);
4590
4591 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4592
4593 /*
4594 * We can't do this in nfsd_break_deleg_cb because it is
4595 * already holding inode->i_lock.
4596 *
4597 * If the dl_time != 0, then we know that it has already been
4598 * queued for a lease break. Don't queue it again.
4599 */
4600 spin_lock(&state_lock);
4601 if (dp->dl_time == 0) {
4602 dp->dl_time = ktime_get_boottime_seconds();
4603 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4604 }
4605 spin_unlock(&state_lock);
4606}
4607
4608static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4609 struct rpc_task *task)
4610{
4611 struct nfs4_delegation *dp = cb_to_delegation(cb);
4612
4613 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4614 dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4615 return 1;
4616
4617 switch (task->tk_status) {
4618 case 0:
4619 return 1;
4620 case -NFS4ERR_DELAY:
4621 rpc_delay(task, 2 * HZ);
4622 return 0;
4623 case -EBADHANDLE:
4624 case -NFS4ERR_BAD_STATEID:
4625 /*
4626 * Race: client probably got cb_recall before open reply
4627 * granting delegation.
4628 */
4629 if (dp->dl_retries--) {
4630 rpc_delay(task, 2 * HZ);
4631 return 0;
4632 }
4633 fallthrough;
4634 default:
4635 return 1;
4636 }
4637}
4638
4639static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4640{
4641 struct nfs4_delegation *dp = cb_to_delegation(cb);
4642
4643 nfs4_put_stid(&dp->dl_stid);
4644}
4645
4646static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4647 .prepare = nfsd4_cb_recall_prepare,
4648 .done = nfsd4_cb_recall_done,
4649 .release = nfsd4_cb_recall_release,
4650};
4651
4652static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4653{
4654 /*
4655 * We're assuming the state code never drops its reference
4656 * without first removing the lease. Since we're in this lease
4657 * callback (and since the lease code is serialized by the
4658 * i_lock) we know the server hasn't removed the lease yet, and
4659 * we know it's safe to take a reference.
4660 */
4661 refcount_inc(&dp->dl_stid.sc_count);
4662 nfsd4_run_cb(&dp->dl_recall);
4663}
4664
4665/* Called from break_lease() with i_lock held. */
4666static bool
4667nfsd_break_deleg_cb(struct file_lock *fl)
4668{
4669 bool ret = false;
4670 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4671 struct nfs4_file *fp = dp->dl_stid.sc_file;
4672
4673 trace_nfsd_cb_recall(&dp->dl_stid);
4674
4675 /*
4676 * We don't want the locks code to timeout the lease for us;
4677 * we'll remove it ourself if a delegation isn't returned
4678 * in time:
4679 */
4680 fl->fl_break_time = 0;
4681
4682 spin_lock(&fp->fi_lock);
4683 fp->fi_had_conflict = true;
4684 nfsd_break_one_deleg(dp);
4685 spin_unlock(&fp->fi_lock);
4686 return ret;
4687}
4688
4689static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4690{
4691 struct nfs4_delegation *dl = fl->fl_owner;
4692 struct svc_rqst *rqst;
4693 struct nfs4_client *clp;
4694
4695 if (!i_am_nfsd())
4696 return NULL;
4697 rqst = kthread_data(current);
4698 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4699 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4700 return NULL;
4701 clp = *(rqst->rq_lease_breaker);
4702 return dl->dl_stid.sc_client == clp;
4703}
4704
4705static int
4706nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4707 struct list_head *dispose)
4708{
4709 if (arg & F_UNLCK)
4710 return lease_modify(onlist, arg, dispose);
4711 else
4712 return -EAGAIN;
4713}
4714
4715static const struct lock_manager_operations nfsd_lease_mng_ops = {
4716 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4717 .lm_break = nfsd_break_deleg_cb,
4718 .lm_change = nfsd_change_deleg_cb,
4719};
4720
4721static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4722{
4723 if (nfsd4_has_session(cstate))
4724 return nfs_ok;
4725 if (seqid == so->so_seqid - 1)
4726 return nfserr_replay_me;
4727 if (seqid == so->so_seqid)
4728 return nfs_ok;
4729 return nfserr_bad_seqid;
4730}
4731
4732static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
4733 struct nfsd_net *nn)
4734{
4735 struct nfs4_client *found;
4736
4737 spin_lock(&nn->client_lock);
4738 found = find_confirmed_client(clid, sessions, nn);
4739 if (found)
4740 atomic_inc(&found->cl_rpc_users);
4741 spin_unlock(&nn->client_lock);
4742 return found;
4743}
4744
4745static __be32 set_client(clientid_t *clid,
4746 struct nfsd4_compound_state *cstate,
4747 struct nfsd_net *nn)
4748{
4749 if (cstate->clp) {
4750 if (!same_clid(&cstate->clp->cl_clientid, clid))
4751 return nfserr_stale_clientid;
4752 return nfs_ok;
4753 }
4754 if (STALE_CLIENTID(clid, nn))
4755 return nfserr_stale_clientid;
4756 /*
4757 * We're in the 4.0 case (otherwise the SEQUENCE op would have
4758 * set cstate->clp), so session = false:
4759 */
4760 cstate->clp = lookup_clientid(clid, false, nn);
4761 if (!cstate->clp)
4762 return nfserr_expired;
4763 return nfs_ok;
4764}
4765
4766__be32
4767nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4768 struct nfsd4_open *open, struct nfsd_net *nn)
4769{
4770 clientid_t *clientid = &open->op_clientid;
4771 struct nfs4_client *clp = NULL;
4772 unsigned int strhashval;
4773 struct nfs4_openowner *oo = NULL;
4774 __be32 status;
4775
4776 /*
4777 * In case we need it later, after we've already created the
4778 * file and don't want to risk a further failure:
4779 */
4780 open->op_file = nfsd4_alloc_file();
4781 if (open->op_file == NULL)
4782 return nfserr_jukebox;
4783
4784 status = set_client(clientid, cstate, nn);
4785 if (status)
4786 return status;
4787 clp = cstate->clp;
4788
4789 strhashval = ownerstr_hashval(&open->op_owner);
4790 oo = find_openstateowner_str(strhashval, open, clp);
4791 open->op_openowner = oo;
4792 if (!oo) {
4793 goto new_owner;
4794 }
4795 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4796 /* Replace unconfirmed owners without checking for replay. */
4797 release_openowner(oo);
4798 open->op_openowner = NULL;
4799 goto new_owner;
4800 }
4801 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4802 if (status)
4803 return status;
4804 goto alloc_stateid;
4805new_owner:
4806 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4807 if (oo == NULL)
4808 return nfserr_jukebox;
4809 open->op_openowner = oo;
4810alloc_stateid:
4811 open->op_stp = nfs4_alloc_open_stateid(clp);
4812 if (!open->op_stp)
4813 return nfserr_jukebox;
4814
4815 if (nfsd4_has_session(cstate) &&
4816 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4817 open->op_odstate = alloc_clnt_odstate(clp);
4818 if (!open->op_odstate)
4819 return nfserr_jukebox;
4820 }
4821
4822 return nfs_ok;
4823}
4824
4825static inline __be32
4826nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4827{
4828 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4829 return nfserr_openmode;
4830 else
4831 return nfs_ok;
4832}
4833
4834static int share_access_to_flags(u32 share_access)
4835{
4836 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4837}
4838
4839static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4840{
4841 struct nfs4_stid *ret;
4842
4843 ret = find_stateid_by_type(cl, s,
4844 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4845 if (!ret)
4846 return NULL;
4847 return delegstateid(ret);
4848}
4849
4850static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4851{
4852 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4853 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4854}
4855
4856static __be32
4857nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4858 struct nfs4_delegation **dp)
4859{
4860 int flags;
4861 __be32 status = nfserr_bad_stateid;
4862 struct nfs4_delegation *deleg;
4863
4864 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4865 if (deleg == NULL)
4866 goto out;
4867 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4868 nfs4_put_stid(&deleg->dl_stid);
4869 if (cl->cl_minorversion)
4870 status = nfserr_deleg_revoked;
4871 goto out;
4872 }
4873 flags = share_access_to_flags(open->op_share_access);
4874 status = nfs4_check_delegmode(deleg, flags);
4875 if (status) {
4876 nfs4_put_stid(&deleg->dl_stid);
4877 goto out;
4878 }
4879 *dp = deleg;
4880out:
4881 if (!nfsd4_is_deleg_cur(open))
4882 return nfs_ok;
4883 if (status)
4884 return status;
4885 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4886 return nfs_ok;
4887}
4888
4889static inline int nfs4_access_to_access(u32 nfs4_access)
4890{
4891 int flags = 0;
4892
4893 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4894 flags |= NFSD_MAY_READ;
4895 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4896 flags |= NFSD_MAY_WRITE;
4897 return flags;
4898}
4899
4900static inline __be32
4901nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4902 struct nfsd4_open *open)
4903{
4904 struct iattr iattr = {
4905 .ia_valid = ATTR_SIZE,
4906 .ia_size = 0,
4907 };
4908 if (!open->op_truncate)
4909 return 0;
4910 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4911 return nfserr_inval;
4912 return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4913}
4914
4915static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4916 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4917 struct nfsd4_open *open)
4918{
4919 struct nfsd_file *nf = NULL;
4920 __be32 status;
4921 int oflag = nfs4_access_to_omode(open->op_share_access);
4922 int access = nfs4_access_to_access(open->op_share_access);
4923 unsigned char old_access_bmap, old_deny_bmap;
4924
4925 spin_lock(&fp->fi_lock);
4926
4927 /*
4928 * Are we trying to set a deny mode that would conflict with
4929 * current access?
4930 */
4931 status = nfs4_file_check_deny(fp, open->op_share_deny);
4932 if (status != nfs_ok) {
4933 spin_unlock(&fp->fi_lock);
4934 goto out;
4935 }
4936
4937 /* set access to the file */
4938 status = nfs4_file_get_access(fp, open->op_share_access);
4939 if (status != nfs_ok) {
4940 spin_unlock(&fp->fi_lock);
4941 goto out;
4942 }
4943
4944 /* Set access bits in stateid */
4945 old_access_bmap = stp->st_access_bmap;
4946 set_access(open->op_share_access, stp);
4947
4948 /* Set new deny mask */
4949 old_deny_bmap = stp->st_deny_bmap;
4950 set_deny(open->op_share_deny, stp);
4951 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4952
4953 if (!fp->fi_fds[oflag]) {
4954 spin_unlock(&fp->fi_lock);
4955 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4956 if (status)
4957 goto out_put_access;
4958 spin_lock(&fp->fi_lock);
4959 if (!fp->fi_fds[oflag]) {
4960 fp->fi_fds[oflag] = nf;
4961 nf = NULL;
4962 }
4963 }
4964 spin_unlock(&fp->fi_lock);
4965 if (nf)
4966 nfsd_file_put(nf);
4967
4968 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
4969 access));
4970 if (status)
4971 goto out_put_access;
4972
4973 status = nfsd4_truncate(rqstp, cur_fh, open);
4974 if (status)
4975 goto out_put_access;
4976out:
4977 return status;
4978out_put_access:
4979 stp->st_access_bmap = old_access_bmap;
4980 nfs4_file_put_access(fp, open->op_share_access);
4981 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4982 goto out;
4983}
4984
4985static __be32
4986nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4987{
4988 __be32 status;
4989 unsigned char old_deny_bmap = stp->st_deny_bmap;
4990
4991 if (!test_access(open->op_share_access, stp))
4992 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4993
4994 /* test and set deny mode */
4995 spin_lock(&fp->fi_lock);
4996 status = nfs4_file_check_deny(fp, open->op_share_deny);
4997 if (status == nfs_ok) {
4998 set_deny(open->op_share_deny, stp);
4999 fp->fi_share_deny |=
5000 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5001 }
5002 spin_unlock(&fp->fi_lock);
5003
5004 if (status != nfs_ok)
5005 return status;
5006
5007 status = nfsd4_truncate(rqstp, cur_fh, open);
5008 if (status != nfs_ok)
5009 reset_union_bmap_deny(old_deny_bmap, stp);
5010 return status;
5011}
5012
5013/* Should we give out recallable state?: */
5014static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5015{
5016 if (clp->cl_cb_state == NFSD4_CB_UP)
5017 return true;
5018 /*
5019 * In the sessions case, since we don't have to establish a
5020 * separate connection for callbacks, we assume it's OK
5021 * until we hear otherwise:
5022 */
5023 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5024}
5025
5026static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5027 int flag)
5028{
5029 struct file_lock *fl;
5030
5031 fl = locks_alloc_lock();
5032 if (!fl)
5033 return NULL;
5034 fl->fl_lmops = &nfsd_lease_mng_ops;
5035 fl->fl_flags = FL_DELEG;
5036 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5037 fl->fl_end = OFFSET_MAX;
5038 fl->fl_owner = (fl_owner_t)dp;
5039 fl->fl_pid = current->tgid;
5040 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5041 return fl;
5042}
5043
5044static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5045 struct nfs4_file *fp)
5046{
5047 struct nfs4_ol_stateid *st;
5048 struct file *f = fp->fi_deleg_file->nf_file;
5049 struct inode *ino = locks_inode(f);
5050 int writes;
5051
5052 writes = atomic_read(&ino->i_writecount);
5053 if (!writes)
5054 return 0;
5055 /*
5056 * There could be multiple filehandles (hence multiple
5057 * nfs4_files) referencing this file, but that's not too
5058 * common; let's just give up in that case rather than
5059 * trying to go look up all the clients using that other
5060 * nfs4_file as well:
5061 */
5062 if (fp->fi_aliased)
5063 return -EAGAIN;
5064 /*
5065 * If there's a close in progress, make sure that we see it
5066 * clear any fi_fds[] entries before we see it decrement
5067 * i_writecount:
5068 */
5069 smp_mb__after_atomic();
5070
5071 if (fp->fi_fds[O_WRONLY])
5072 writes--;
5073 if (fp->fi_fds[O_RDWR])
5074 writes--;
5075 if (writes > 0)
5076 return -EAGAIN; /* There may be non-NFSv4 writers */
5077 /*
5078 * It's possible there are non-NFSv4 write opens in progress,
5079 * but if they haven't incremented i_writecount yet then they
5080 * also haven't called break lease yet; so, they'll break this
5081 * lease soon enough. So, all that's left to check for is NFSv4
5082 * opens:
5083 */
5084 spin_lock(&fp->fi_lock);
5085 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5086 if (st->st_openstp == NULL /* it's an open */ &&
5087 access_permit_write(st) &&
5088 st->st_stid.sc_client != clp) {
5089 spin_unlock(&fp->fi_lock);
5090 return -EAGAIN;
5091 }
5092 }
5093 spin_unlock(&fp->fi_lock);
5094 /*
5095 * There's a small chance that we could be racing with another
5096 * NFSv4 open. However, any open that hasn't added itself to
5097 * the fi_stateids list also hasn't called break_lease yet; so,
5098 * they'll break this lease soon enough.
5099 */
5100 return 0;
5101}
5102
5103static struct nfs4_delegation *
5104nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
5105 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
5106{
5107 int status = 0;
5108 struct nfs4_delegation *dp;
5109 struct nfsd_file *nf;
5110 struct file_lock *fl;
5111
5112 /*
5113 * The fi_had_conflict and nfs_get_existing_delegation checks
5114 * here are just optimizations; we'll need to recheck them at
5115 * the end:
5116 */
5117 if (fp->fi_had_conflict)
5118 return ERR_PTR(-EAGAIN);
5119
5120 nf = find_readable_file(fp);
5121 if (!nf) {
5122 /*
5123 * We probably could attempt another open and get a read
5124 * delegation, but for now, don't bother until the
5125 * client actually sends us one.
5126 */
5127 return ERR_PTR(-EAGAIN);
5128 }
5129 spin_lock(&state_lock);
5130 spin_lock(&fp->fi_lock);
5131 if (nfs4_delegation_exists(clp, fp))
5132 status = -EAGAIN;
5133 else if (!fp->fi_deleg_file) {
5134 fp->fi_deleg_file = nf;
5135 /* increment early to prevent fi_deleg_file from being
5136 * cleared */
5137 fp->fi_delegees = 1;
5138 nf = NULL;
5139 } else
5140 fp->fi_delegees++;
5141 spin_unlock(&fp->fi_lock);
5142 spin_unlock(&state_lock);
5143 if (nf)
5144 nfsd_file_put(nf);
5145 if (status)
5146 return ERR_PTR(status);
5147
5148 status = -ENOMEM;
5149 dp = alloc_init_deleg(clp, fp, fh, odstate);
5150 if (!dp)
5151 goto out_delegees;
5152
5153 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5154 if (!fl)
5155 goto out_clnt_odstate;
5156
5157 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5158 if (fl)
5159 locks_free_lock(fl);
5160 if (status)
5161 goto out_clnt_odstate;
5162 status = nfsd4_check_conflicting_opens(clp, fp);
5163 if (status)
5164 goto out_unlock;
5165
5166 spin_lock(&state_lock);
5167 spin_lock(&fp->fi_lock);
5168 if (fp->fi_had_conflict)
5169 status = -EAGAIN;
5170 else
5171 status = hash_delegation_locked(dp, fp);
5172 spin_unlock(&fp->fi_lock);
5173 spin_unlock(&state_lock);
5174
5175 if (status)
5176 goto out_unlock;
5177
5178 return dp;
5179out_unlock:
5180 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5181out_clnt_odstate:
5182 put_clnt_odstate(dp->dl_clnt_odstate);
5183 nfs4_put_stid(&dp->dl_stid);
5184out_delegees:
5185 put_deleg_file(fp);
5186 return ERR_PTR(status);
5187}
5188
5189static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5190{
5191 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5192 if (status == -EAGAIN)
5193 open->op_why_no_deleg = WND4_CONTENTION;
5194 else {
5195 open->op_why_no_deleg = WND4_RESOURCE;
5196 switch (open->op_deleg_want) {
5197 case NFS4_SHARE_WANT_READ_DELEG:
5198 case NFS4_SHARE_WANT_WRITE_DELEG:
5199 case NFS4_SHARE_WANT_ANY_DELEG:
5200 break;
5201 case NFS4_SHARE_WANT_CANCEL:
5202 open->op_why_no_deleg = WND4_CANCELLED;
5203 break;
5204 case NFS4_SHARE_WANT_NO_DELEG:
5205 WARN_ON_ONCE(1);
5206 }
5207 }
5208}
5209
5210/*
5211 * Attempt to hand out a delegation.
5212 *
5213 * Note we don't support write delegations, and won't until the vfs has
5214 * proper support for them.
5215 */
5216static void
5217nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
5218 struct nfs4_ol_stateid *stp)
5219{
5220 struct nfs4_delegation *dp;
5221 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5222 struct nfs4_client *clp = stp->st_stid.sc_client;
5223 int cb_up;
5224 int status = 0;
5225
5226 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5227 open->op_recall = 0;
5228 switch (open->op_claim_type) {
5229 case NFS4_OPEN_CLAIM_PREVIOUS:
5230 if (!cb_up)
5231 open->op_recall = 1;
5232 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5233 goto out_no_deleg;
5234 break;
5235 case NFS4_OPEN_CLAIM_NULL:
5236 case NFS4_OPEN_CLAIM_FH:
5237 /*
5238 * Let's not give out any delegations till everyone's
5239 * had the chance to reclaim theirs, *and* until
5240 * NLM locks have all been reclaimed:
5241 */
5242 if (locks_in_grace(clp->net))
5243 goto out_no_deleg;
5244 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5245 goto out_no_deleg;
5246 break;
5247 default:
5248 goto out_no_deleg;
5249 }
5250 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5251 if (IS_ERR(dp))
5252 goto out_no_deleg;
5253
5254 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5255
5256 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5257 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5258 nfs4_put_stid(&dp->dl_stid);
5259 return;
5260out_no_deleg:
5261 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5262 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5263 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5264 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5265 open->op_recall = 1;
5266 }
5267
5268 /* 4.1 client asking for a delegation? */
5269 if (open->op_deleg_want)
5270 nfsd4_open_deleg_none_ext(open, status);
5271 return;
5272}
5273
5274static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5275 struct nfs4_delegation *dp)
5276{
5277 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5278 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5279 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5280 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5281 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5282 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5283 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5284 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5285 }
5286 /* Otherwise the client must be confused wanting a delegation
5287 * it already has, therefore we don't return
5288 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5289 */
5290}
5291
5292__be32
5293nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5294{
5295 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5296 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5297 struct nfs4_file *fp = NULL;
5298 struct nfs4_ol_stateid *stp = NULL;
5299 struct nfs4_delegation *dp = NULL;
5300 __be32 status;
5301 bool new_stp = false;
5302
5303 /*
5304 * Lookup file; if found, lookup stateid and check open request,
5305 * and check for delegations in the process of being recalled.
5306 * If not found, create the nfs4_file struct
5307 */
5308 fp = find_or_add_file(open->op_file, current_fh);
5309 if (fp != open->op_file) {
5310 status = nfs4_check_deleg(cl, open, &dp);
5311 if (status)
5312 goto out;
5313 stp = nfsd4_find_and_lock_existing_open(fp, open);
5314 } else {
5315 open->op_file = NULL;
5316 status = nfserr_bad_stateid;
5317 if (nfsd4_is_deleg_cur(open))
5318 goto out;
5319 }
5320
5321 if (!stp) {
5322 stp = init_open_stateid(fp, open);
5323 if (!open->op_stp)
5324 new_stp = true;
5325 }
5326
5327 /*
5328 * OPEN the file, or upgrade an existing OPEN.
5329 * If truncate fails, the OPEN fails.
5330 *
5331 * stp is already locked.
5332 */
5333 if (!new_stp) {
5334 /* Stateid was found, this is an OPEN upgrade */
5335 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5336 if (status) {
5337 mutex_unlock(&stp->st_mutex);
5338 goto out;
5339 }
5340 } else {
5341 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5342 if (status) {
5343 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5344 release_open_stateid(stp);
5345 mutex_unlock(&stp->st_mutex);
5346 goto out;
5347 }
5348
5349 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5350 open->op_odstate);
5351 if (stp->st_clnt_odstate == open->op_odstate)
5352 open->op_odstate = NULL;
5353 }
5354
5355 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5356 mutex_unlock(&stp->st_mutex);
5357
5358 if (nfsd4_has_session(&resp->cstate)) {
5359 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5360 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5361 open->op_why_no_deleg = WND4_NOT_WANTED;
5362 goto nodeleg;
5363 }
5364 }
5365
5366 /*
5367 * Attempt to hand out a delegation. No error return, because the
5368 * OPEN succeeds even if we fail.
5369 */
5370 nfs4_open_delegation(current_fh, open, stp);
5371nodeleg:
5372 status = nfs_ok;
5373 trace_nfsd_open(&stp->st_stid.sc_stateid);
5374out:
5375 /* 4.1 client trying to upgrade/downgrade delegation? */
5376 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5377 open->op_deleg_want)
5378 nfsd4_deleg_xgrade_none_ext(open, dp);
5379
5380 if (fp)
5381 put_nfs4_file(fp);
5382 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5383 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5384 /*
5385 * To finish the open response, we just need to set the rflags.
5386 */
5387 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5388 if (nfsd4_has_session(&resp->cstate))
5389 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5390 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5391 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5392
5393 if (dp)
5394 nfs4_put_stid(&dp->dl_stid);
5395 if (stp)
5396 nfs4_put_stid(&stp->st_stid);
5397
5398 return status;
5399}
5400
5401void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5402 struct nfsd4_open *open)
5403{
5404 if (open->op_openowner) {
5405 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5406
5407 nfsd4_cstate_assign_replay(cstate, so);
5408 nfs4_put_stateowner(so);
5409 }
5410 if (open->op_file)
5411 kmem_cache_free(file_slab, open->op_file);
5412 if (open->op_stp)
5413 nfs4_put_stid(&open->op_stp->st_stid);
5414 if (open->op_odstate)
5415 kmem_cache_free(odstate_slab, open->op_odstate);
5416}
5417
5418__be32
5419nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5420 union nfsd4_op_u *u)
5421{
5422 clientid_t *clid = &u->renew;
5423 struct nfs4_client *clp;
5424 __be32 status;
5425 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5426
5427 trace_nfsd_clid_renew(clid);
5428 status = set_client(clid, cstate, nn);
5429 if (status)
5430 return status;
5431 clp = cstate->clp;
5432 if (!list_empty(&clp->cl_delegations)
5433 && clp->cl_cb_state != NFSD4_CB_UP)
5434 return nfserr_cb_path_down;
5435 return nfs_ok;
5436}
5437
5438void
5439nfsd4_end_grace(struct nfsd_net *nn)
5440{
5441 /* do nothing if grace period already ended */
5442 if (nn->grace_ended)
5443 return;
5444
5445 trace_nfsd_grace_complete(nn);
5446 nn->grace_ended = true;
5447 /*
5448 * If the server goes down again right now, an NFSv4
5449 * client will still be allowed to reclaim after it comes back up,
5450 * even if it hasn't yet had a chance to reclaim state this time.
5451 *
5452 */
5453 nfsd4_record_grace_done(nn);
5454 /*
5455 * At this point, NFSv4 clients can still reclaim. But if the
5456 * server crashes, any that have not yet reclaimed will be out
5457 * of luck on the next boot.
5458 *
5459 * (NFSv4.1+ clients are considered to have reclaimed once they
5460 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5461 * have reclaimed after their first OPEN.)
5462 */
5463 locks_end_grace(&nn->nfsd4_manager);
5464 /*
5465 * At this point, and once lockd and/or any other containers
5466 * exit their grace period, further reclaims will fail and
5467 * regular locking can resume.
5468 */
5469}
5470
5471/*
5472 * If we've waited a lease period but there are still clients trying to
5473 * reclaim, wait a little longer to give them a chance to finish.
5474 */
5475static bool clients_still_reclaiming(struct nfsd_net *nn)
5476{
5477 time64_t double_grace_period_end = nn->boot_time +
5478 2 * nn->nfsd4_lease;
5479
5480 if (nn->track_reclaim_completes &&
5481 atomic_read(&nn->nr_reclaim_complete) ==
5482 nn->reclaim_str_hashtbl_size)
5483 return false;
5484 if (!nn->somebody_reclaimed)
5485 return false;
5486 nn->somebody_reclaimed = false;
5487 /*
5488 * If we've given them *two* lease times to reclaim, and they're
5489 * still not done, give up:
5490 */
5491 if (ktime_get_boottime_seconds() > double_grace_period_end)
5492 return false;
5493 return true;
5494}
5495
5496struct laundry_time {
5497 time64_t cutoff;
5498 time64_t new_timeo;
5499};
5500
5501static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5502{
5503 time64_t time_remaining;
5504
5505 if (last_refresh < lt->cutoff)
5506 return true;
5507 time_remaining = last_refresh - lt->cutoff;
5508 lt->new_timeo = min(lt->new_timeo, time_remaining);
5509 return false;
5510}
5511
5512#ifdef CONFIG_NFSD_V4_2_INTER_SSC
5513void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5514{
5515 spin_lock_init(&nn->nfsd_ssc_lock);
5516 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5517 init_waitqueue_head(&nn->nfsd_ssc_waitq);
5518}
5519EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5520
5521/*
5522 * This is called when nfsd is being shutdown, after all inter_ssc
5523 * cleanup were done, to destroy the ssc delayed unmount list.
5524 */
5525static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5526{
5527 struct nfsd4_ssc_umount_item *ni = NULL;
5528 struct nfsd4_ssc_umount_item *tmp;
5529
5530 spin_lock(&nn->nfsd_ssc_lock);
5531 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5532 list_del(&ni->nsui_list);
5533 spin_unlock(&nn->nfsd_ssc_lock);
5534 mntput(ni->nsui_vfsmount);
5535 kfree(ni);
5536 spin_lock(&nn->nfsd_ssc_lock);
5537 }
5538 spin_unlock(&nn->nfsd_ssc_lock);
5539}
5540
5541static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5542{
5543 bool do_wakeup = false;
5544 struct nfsd4_ssc_umount_item *ni = 0;
5545 struct nfsd4_ssc_umount_item *tmp;
5546
5547 spin_lock(&nn->nfsd_ssc_lock);
5548 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5549 if (time_after(jiffies, ni->nsui_expire)) {
5550 if (refcount_read(&ni->nsui_refcnt) > 1)
5551 continue;
5552
5553 /* mark being unmount */
5554 ni->nsui_busy = true;
5555 spin_unlock(&nn->nfsd_ssc_lock);
5556 mntput(ni->nsui_vfsmount);
5557 spin_lock(&nn->nfsd_ssc_lock);
5558
5559 /* waiters need to start from begin of list */
5560 list_del(&ni->nsui_list);
5561 kfree(ni);
5562
5563 /* wakeup ssc_connect waiters */
5564 do_wakeup = true;
5565 continue;
5566 }
5567 break;
5568 }
5569 if (do_wakeup)
5570 wake_up_all(&nn->nfsd_ssc_waitq);
5571 spin_unlock(&nn->nfsd_ssc_lock);
5572}
5573#endif
5574
5575static time64_t
5576nfs4_laundromat(struct nfsd_net *nn)
5577{
5578 struct nfs4_client *clp;
5579 struct nfs4_openowner *oo;
5580 struct nfs4_delegation *dp;
5581 struct nfs4_ol_stateid *stp;
5582 struct nfsd4_blocked_lock *nbl;
5583 struct list_head *pos, *next, reaplist;
5584 struct laundry_time lt = {
5585 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
5586 .new_timeo = nn->nfsd4_lease
5587 };
5588 struct nfs4_cpntf_state *cps;
5589 copy_stateid_t *cps_t;
5590 int i;
5591
5592 if (clients_still_reclaiming(nn)) {
5593 lt.new_timeo = 0;
5594 goto out;
5595 }
5596 nfsd4_end_grace(nn);
5597 INIT_LIST_HEAD(&reaplist);
5598
5599 spin_lock(&nn->s2s_cp_lock);
5600 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5601 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5602 if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5603 state_expired(<, cps->cpntf_time))
5604 _free_cpntf_state_locked(nn, cps);
5605 }
5606 spin_unlock(&nn->s2s_cp_lock);
5607
5608 spin_lock(&nn->client_lock);
5609 list_for_each_safe(pos, next, &nn->client_lru) {
5610 clp = list_entry(pos, struct nfs4_client, cl_lru);
5611 if (!state_expired(<, clp->cl_time))
5612 break;
5613 if (mark_client_expired_locked(clp))
5614 continue;
5615 list_add(&clp->cl_lru, &reaplist);
5616 }
5617 spin_unlock(&nn->client_lock);
5618 list_for_each_safe(pos, next, &reaplist) {
5619 clp = list_entry(pos, struct nfs4_client, cl_lru);
5620 trace_nfsd_clid_purged(&clp->cl_clientid);
5621 list_del_init(&clp->cl_lru);
5622 expire_client(clp);
5623 }
5624 spin_lock(&state_lock);
5625 list_for_each_safe(pos, next, &nn->del_recall_lru) {
5626 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5627 if (!state_expired(<, dp->dl_time))
5628 break;
5629 WARN_ON(!unhash_delegation_locked(dp));
5630 list_add(&dp->dl_recall_lru, &reaplist);
5631 }
5632 spin_unlock(&state_lock);
5633 while (!list_empty(&reaplist)) {
5634 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5635 dl_recall_lru);
5636 list_del_init(&dp->dl_recall_lru);
5637 revoke_delegation(dp);
5638 }
5639
5640 spin_lock(&nn->client_lock);
5641 while (!list_empty(&nn->close_lru)) {
5642 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5643 oo_close_lru);
5644 if (!state_expired(<, oo->oo_time))
5645 break;
5646 list_del_init(&oo->oo_close_lru);
5647 stp = oo->oo_last_closed_stid;
5648 oo->oo_last_closed_stid = NULL;
5649 spin_unlock(&nn->client_lock);
5650 nfs4_put_stid(&stp->st_stid);
5651 spin_lock(&nn->client_lock);
5652 }
5653 spin_unlock(&nn->client_lock);
5654
5655 /*
5656 * It's possible for a client to try and acquire an already held lock
5657 * that is being held for a long time, and then lose interest in it.
5658 * So, we clean out any un-revisited request after a lease period
5659 * under the assumption that the client is no longer interested.
5660 *
5661 * RFC5661, sec. 9.6 states that the client must not rely on getting
5662 * notifications and must continue to poll for locks, even when the
5663 * server supports them. Thus this shouldn't lead to clients blocking
5664 * indefinitely once the lock does become free.
5665 */
5666 BUG_ON(!list_empty(&reaplist));
5667 spin_lock(&nn->blocked_locks_lock);
5668 while (!list_empty(&nn->blocked_locks_lru)) {
5669 nbl = list_first_entry(&nn->blocked_locks_lru,
5670 struct nfsd4_blocked_lock, nbl_lru);
5671 if (!state_expired(<, nbl->nbl_time))
5672 break;
5673 list_move(&nbl->nbl_lru, &reaplist);
5674 list_del_init(&nbl->nbl_list);
5675 }
5676 spin_unlock(&nn->blocked_locks_lock);
5677
5678 while (!list_empty(&reaplist)) {
5679 nbl = list_first_entry(&reaplist,
5680 struct nfsd4_blocked_lock, nbl_lru);
5681 list_del_init(&nbl->nbl_lru);
5682 free_blocked_lock(nbl);
5683 }
5684#ifdef CONFIG_NFSD_V4_2_INTER_SSC
5685 /* service the server-to-server copy delayed unmount list */
5686 nfsd4_ssc_expire_umount(nn);
5687#endif
5688out:
5689 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5690}
5691
5692static struct workqueue_struct *laundry_wq;
5693static void laundromat_main(struct work_struct *);
5694
5695static void
5696laundromat_main(struct work_struct *laundry)
5697{
5698 time64_t t;
5699 struct delayed_work *dwork = to_delayed_work(laundry);
5700 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5701 laundromat_work);
5702
5703 t = nfs4_laundromat(nn);
5704 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5705}
5706
5707static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5708{
5709 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5710 return nfserr_bad_stateid;
5711 return nfs_ok;
5712}
5713
5714static
5715__be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5716{
5717 __be32 status = nfserr_openmode;
5718
5719 /* For lock stateid's, we test the parent open, not the lock: */
5720 if (stp->st_openstp)
5721 stp = stp->st_openstp;
5722 if ((flags & WR_STATE) && !access_permit_write(stp))
5723 goto out;
5724 if ((flags & RD_STATE) && !access_permit_read(stp))
5725 goto out;
5726 status = nfs_ok;
5727out:
5728 return status;
5729}
5730
5731static inline __be32
5732check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5733{
5734 if (ONE_STATEID(stateid) && (flags & RD_STATE))
5735 return nfs_ok;
5736 else if (opens_in_grace(net)) {
5737 /* Answer in remaining cases depends on existence of
5738 * conflicting state; so we must wait out the grace period. */
5739 return nfserr_grace;
5740 } else if (flags & WR_STATE)
5741 return nfs4_share_conflict(current_fh,
5742 NFS4_SHARE_DENY_WRITE);
5743 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5744 return nfs4_share_conflict(current_fh,
5745 NFS4_SHARE_DENY_READ);
5746}
5747
5748/*
5749 * Allow READ/WRITE during grace period on recovered state only for files
5750 * that are not able to provide mandatory locking.
5751 */
5752static inline int
5753grace_disallows_io(struct net *net, struct inode *inode)
5754{
5755 return opens_in_grace(net) && mandatory_lock(inode);
5756}
5757
5758static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5759{
5760 /*
5761 * When sessions are used the stateid generation number is ignored
5762 * when it is zero.
5763 */
5764 if (has_session && in->si_generation == 0)
5765 return nfs_ok;
5766
5767 if (in->si_generation == ref->si_generation)
5768 return nfs_ok;
5769
5770 /* If the client sends us a stateid from the future, it's buggy: */
5771 if (nfsd4_stateid_generation_after(in, ref))
5772 return nfserr_bad_stateid;
5773 /*
5774 * However, we could see a stateid from the past, even from a
5775 * non-buggy client. For example, if the client sends a lock
5776 * while some IO is outstanding, the lock may bump si_generation
5777 * while the IO is still in flight. The client could avoid that
5778 * situation by waiting for responses on all the IO requests,
5779 * but better performance may result in retrying IO that
5780 * receives an old_stateid error if requests are rarely
5781 * reordered in flight:
5782 */
5783 return nfserr_old_stateid;
5784}
5785
5786static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5787{
5788 __be32 ret;
5789
5790 spin_lock(&s->sc_lock);
5791 ret = nfsd4_verify_open_stid(s);
5792 if (ret == nfs_ok)
5793 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5794 spin_unlock(&s->sc_lock);
5795 return ret;
5796}
5797
5798static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5799{
5800 if (ols->st_stateowner->so_is_open_owner &&
5801 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5802 return nfserr_bad_stateid;
5803 return nfs_ok;
5804}
5805
5806static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5807{
5808 struct nfs4_stid *s;
5809 __be32 status = nfserr_bad_stateid;
5810
5811 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5812 CLOSE_STATEID(stateid))
5813 return status;
5814 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
5815 return status;
5816 spin_lock(&cl->cl_lock);
5817 s = find_stateid_locked(cl, stateid);
5818 if (!s)
5819 goto out_unlock;
5820 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5821 if (status)
5822 goto out_unlock;
5823 switch (s->sc_type) {
5824 case NFS4_DELEG_STID:
5825 status = nfs_ok;
5826 break;
5827 case NFS4_REVOKED_DELEG_STID:
5828 status = nfserr_deleg_revoked;
5829 break;
5830 case NFS4_OPEN_STID:
5831 case NFS4_LOCK_STID:
5832 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5833 break;
5834 default:
5835 printk("unknown stateid type %x\n", s->sc_type);
5836 fallthrough;
5837 case NFS4_CLOSED_STID:
5838 case NFS4_CLOSED_DELEG_STID:
5839 status = nfserr_bad_stateid;
5840 }
5841out_unlock:
5842 spin_unlock(&cl->cl_lock);
5843 return status;
5844}
5845
5846__be32
5847nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5848 stateid_t *stateid, unsigned char typemask,
5849 struct nfs4_stid **s, struct nfsd_net *nn)
5850{
5851 __be32 status;
5852 bool return_revoked = false;
5853
5854 /*
5855 * only return revoked delegations if explicitly asked.
5856 * otherwise we report revoked or bad_stateid status.
5857 */
5858 if (typemask & NFS4_REVOKED_DELEG_STID)
5859 return_revoked = true;
5860 else if (typemask & NFS4_DELEG_STID)
5861 typemask |= NFS4_REVOKED_DELEG_STID;
5862
5863 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5864 CLOSE_STATEID(stateid))
5865 return nfserr_bad_stateid;
5866 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
5867 if (status == nfserr_stale_clientid) {
5868 if (cstate->session)
5869 return nfserr_bad_stateid;
5870 return nfserr_stale_stateid;
5871 }
5872 if (status)
5873 return status;
5874 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5875 if (!*s)
5876 return nfserr_bad_stateid;
5877 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5878 nfs4_put_stid(*s);
5879 if (cstate->minorversion)
5880 return nfserr_deleg_revoked;
5881 return nfserr_bad_stateid;
5882 }
5883 return nfs_ok;
5884}
5885
5886static struct nfsd_file *
5887nfs4_find_file(struct nfs4_stid *s, int flags)
5888{
5889 if (!s)
5890 return NULL;
5891
5892 switch (s->sc_type) {
5893 case NFS4_DELEG_STID:
5894 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5895 return NULL;
5896 return nfsd_file_get(s->sc_file->fi_deleg_file);
5897 case NFS4_OPEN_STID:
5898 case NFS4_LOCK_STID:
5899 if (flags & RD_STATE)
5900 return find_readable_file(s->sc_file);
5901 else
5902 return find_writeable_file(s->sc_file);
5903 }
5904
5905 return NULL;
5906}
5907
5908static __be32
5909nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5910{
5911 __be32 status;
5912
5913 status = nfsd4_check_openowner_confirmed(ols);
5914 if (status)
5915 return status;
5916 return nfs4_check_openmode(ols, flags);
5917}
5918
5919static __be32
5920nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5921 struct nfsd_file **nfp, int flags)
5922{
5923 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5924 struct nfsd_file *nf;
5925 __be32 status;
5926
5927 nf = nfs4_find_file(s, flags);
5928 if (nf) {
5929 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5930 acc | NFSD_MAY_OWNER_OVERRIDE);
5931 if (status) {
5932 nfsd_file_put(nf);
5933 goto out;
5934 }
5935 } else {
5936 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5937 if (status)
5938 return status;
5939 }
5940 *nfp = nf;
5941out:
5942 return status;
5943}
5944static void
5945_free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5946{
5947 WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
5948 if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
5949 return;
5950 list_del(&cps->cp_list);
5951 idr_remove(&nn->s2s_cp_stateids,
5952 cps->cp_stateid.stid.si_opaque.so_id);
5953 kfree(cps);
5954}
5955/*
5956 * A READ from an inter server to server COPY will have a
5957 * copy stateid. Look up the copy notify stateid from the
5958 * idr structure and take a reference on it.
5959 */
5960__be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5961 struct nfs4_client *clp,
5962 struct nfs4_cpntf_state **cps)
5963{
5964 copy_stateid_t *cps_t;
5965 struct nfs4_cpntf_state *state = NULL;
5966
5967 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
5968 return nfserr_bad_stateid;
5969 spin_lock(&nn->s2s_cp_lock);
5970 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
5971 if (cps_t) {
5972 state = container_of(cps_t, struct nfs4_cpntf_state,
5973 cp_stateid);
5974 if (state->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID) {
5975 state = NULL;
5976 goto unlock;
5977 }
5978 if (!clp)
5979 refcount_inc(&state->cp_stateid.sc_count);
5980 else
5981 _free_cpntf_state_locked(nn, state);
5982 }
5983unlock:
5984 spin_unlock(&nn->s2s_cp_lock);
5985 if (!state)
5986 return nfserr_bad_stateid;
5987 if (!clp && state)
5988 *cps = state;
5989 return 0;
5990}
5991
5992static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5993 struct nfs4_stid **stid)
5994{
5995 __be32 status;
5996 struct nfs4_cpntf_state *cps = NULL;
5997 struct nfs4_client *found;
5998
5999 status = manage_cpntf_state(nn, st, NULL, &cps);
6000 if (status)
6001 return status;
6002
6003 cps->cpntf_time = ktime_get_boottime_seconds();
6004
6005 status = nfserr_expired;
6006 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6007 if (!found)
6008 goto out;
6009
6010 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6011 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
6012 if (*stid)
6013 status = nfs_ok;
6014 else
6015 status = nfserr_bad_stateid;
6016
6017 put_client_renew(found);
6018out:
6019 nfs4_put_cpntf_state(nn, cps);
6020 return status;
6021}
6022
6023void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6024{
6025 spin_lock(&nn->s2s_cp_lock);
6026 _free_cpntf_state_locked(nn, cps);
6027 spin_unlock(&nn->s2s_cp_lock);
6028}
6029
6030/*
6031 * Checks for stateid operations
6032 */
6033__be32
6034nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6035 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6036 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6037 struct nfs4_stid **cstid)
6038{
6039 struct inode *ino = d_inode(fhp->fh_dentry);
6040 struct net *net = SVC_NET(rqstp);
6041 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6042 struct nfs4_stid *s = NULL;
6043 __be32 status;
6044
6045 if (nfp)
6046 *nfp = NULL;
6047
6048 if (grace_disallows_io(net, ino))
6049 return nfserr_grace;
6050
6051 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6052 status = check_special_stateids(net, fhp, stateid, flags);
6053 goto done;
6054 }
6055
6056 status = nfsd4_lookup_stateid(cstate, stateid,
6057 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
6058 &s, nn);
6059 if (status == nfserr_bad_stateid)
6060 status = find_cpntf_state(nn, stateid, &s);
6061 if (status)
6062 return status;
6063 status = nfsd4_stid_check_stateid_generation(stateid, s,
6064 nfsd4_has_session(cstate));
6065 if (status)
6066 goto out;
6067
6068 switch (s->sc_type) {
6069 case NFS4_DELEG_STID:
6070 status = nfs4_check_delegmode(delegstateid(s), flags);
6071 break;
6072 case NFS4_OPEN_STID:
6073 case NFS4_LOCK_STID:
6074 status = nfs4_check_olstateid(openlockstateid(s), flags);
6075 break;
6076 default:
6077 status = nfserr_bad_stateid;
6078 break;
6079 }
6080 if (status)
6081 goto out;
6082 status = nfs4_check_fh(fhp, s);
6083
6084done:
6085 if (status == nfs_ok && nfp)
6086 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
6087out:
6088 if (s) {
6089 if (!status && cstid)
6090 *cstid = s;
6091 else
6092 nfs4_put_stid(s);
6093 }
6094 return status;
6095}
6096
6097/*
6098 * Test if the stateid is valid
6099 */
6100__be32
6101nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6102 union nfsd4_op_u *u)
6103{
6104 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
6105 struct nfsd4_test_stateid_id *stateid;
6106 struct nfs4_client *cl = cstate->clp;
6107
6108 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
6109 stateid->ts_id_status =
6110 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
6111
6112 return nfs_ok;
6113}
6114
6115static __be32
6116nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
6117{
6118 struct nfs4_ol_stateid *stp = openlockstateid(s);
6119 __be32 ret;
6120
6121 ret = nfsd4_lock_ol_stateid(stp);
6122 if (ret)
6123 goto out_put_stid;
6124
6125 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6126 if (ret)
6127 goto out;
6128
6129 ret = nfserr_locks_held;
6130 if (check_for_locks(stp->st_stid.sc_file,
6131 lockowner(stp->st_stateowner)))
6132 goto out;
6133
6134 release_lock_stateid(stp);
6135 ret = nfs_ok;
6136
6137out:
6138 mutex_unlock(&stp->st_mutex);
6139out_put_stid:
6140 nfs4_put_stid(s);
6141 return ret;
6142}
6143
6144__be32
6145nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6146 union nfsd4_op_u *u)
6147{
6148 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
6149 stateid_t *stateid = &free_stateid->fr_stateid;
6150 struct nfs4_stid *s;
6151 struct nfs4_delegation *dp;
6152 struct nfs4_client *cl = cstate->clp;
6153 __be32 ret = nfserr_bad_stateid;
6154
6155 spin_lock(&cl->cl_lock);
6156 s = find_stateid_locked(cl, stateid);
6157 if (!s)
6158 goto out_unlock;
6159 spin_lock(&s->sc_lock);
6160 switch (s->sc_type) {
6161 case NFS4_DELEG_STID:
6162 ret = nfserr_locks_held;
6163 break;
6164 case NFS4_OPEN_STID:
6165 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6166 if (ret)
6167 break;
6168 ret = nfserr_locks_held;
6169 break;
6170 case NFS4_LOCK_STID:
6171 spin_unlock(&s->sc_lock);
6172 refcount_inc(&s->sc_count);
6173 spin_unlock(&cl->cl_lock);
6174 ret = nfsd4_free_lock_stateid(stateid, s);
6175 goto out;
6176 case NFS4_REVOKED_DELEG_STID:
6177 spin_unlock(&s->sc_lock);
6178 dp = delegstateid(s);
6179 list_del_init(&dp->dl_recall_lru);
6180 spin_unlock(&cl->cl_lock);
6181 nfs4_put_stid(s);
6182 ret = nfs_ok;
6183 goto out;
6184 /* Default falls through and returns nfserr_bad_stateid */
6185 }
6186 spin_unlock(&s->sc_lock);
6187out_unlock:
6188 spin_unlock(&cl->cl_lock);
6189out:
6190 return ret;
6191}
6192
6193static inline int
6194setlkflg (int type)
6195{
6196 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6197 RD_STATE : WR_STATE;
6198}
6199
6200static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6201{
6202 struct svc_fh *current_fh = &cstate->current_fh;
6203 struct nfs4_stateowner *sop = stp->st_stateowner;
6204 __be32 status;
6205
6206 status = nfsd4_check_seqid(cstate, sop, seqid);
6207 if (status)
6208 return status;
6209 status = nfsd4_lock_ol_stateid(stp);
6210 if (status != nfs_ok)
6211 return status;
6212 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6213 if (status == nfs_ok)
6214 status = nfs4_check_fh(current_fh, &stp->st_stid);
6215 if (status != nfs_ok)
6216 mutex_unlock(&stp->st_mutex);
6217 return status;
6218}
6219
6220/*
6221 * Checks for sequence id mutating operations.
6222 */
6223static __be32
6224nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6225 stateid_t *stateid, char typemask,
6226 struct nfs4_ol_stateid **stpp,
6227 struct nfsd_net *nn)
6228{
6229 __be32 status;
6230 struct nfs4_stid *s;
6231 struct nfs4_ol_stateid *stp = NULL;
6232
6233 trace_nfsd_preprocess(seqid, stateid);
6234
6235 *stpp = NULL;
6236 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6237 if (status)
6238 return status;
6239 stp = openlockstateid(s);
6240 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6241
6242 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6243 if (!status)
6244 *stpp = stp;
6245 else
6246 nfs4_put_stid(&stp->st_stid);
6247 return status;
6248}
6249
6250static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6251 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6252{
6253 __be32 status;
6254 struct nfs4_openowner *oo;
6255 struct nfs4_ol_stateid *stp;
6256
6257 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6258 NFS4_OPEN_STID, &stp, nn);
6259 if (status)
6260 return status;
6261 oo = openowner(stp->st_stateowner);
6262 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6263 mutex_unlock(&stp->st_mutex);
6264 nfs4_put_stid(&stp->st_stid);
6265 return nfserr_bad_stateid;
6266 }
6267 *stpp = stp;
6268 return nfs_ok;
6269}
6270
6271__be32
6272nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6273 union nfsd4_op_u *u)
6274{
6275 struct nfsd4_open_confirm *oc = &u->open_confirm;
6276 __be32 status;
6277 struct nfs4_openowner *oo;
6278 struct nfs4_ol_stateid *stp;
6279 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6280
6281 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6282 cstate->current_fh.fh_dentry);
6283
6284 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6285 if (status)
6286 return status;
6287
6288 status = nfs4_preprocess_seqid_op(cstate,
6289 oc->oc_seqid, &oc->oc_req_stateid,
6290 NFS4_OPEN_STID, &stp, nn);
6291 if (status)
6292 goto out;
6293 oo = openowner(stp->st_stateowner);
6294 status = nfserr_bad_stateid;
6295 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6296 mutex_unlock(&stp->st_mutex);
6297 goto put_stateid;
6298 }
6299 oo->oo_flags |= NFS4_OO_CONFIRMED;
6300 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6301 mutex_unlock(&stp->st_mutex);
6302 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6303 nfsd4_client_record_create(oo->oo_owner.so_client);
6304 status = nfs_ok;
6305put_stateid:
6306 nfs4_put_stid(&stp->st_stid);
6307out:
6308 nfsd4_bump_seqid(cstate, status);
6309 return status;
6310}
6311
6312static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6313{
6314 if (!test_access(access, stp))
6315 return;
6316 nfs4_file_put_access(stp->st_stid.sc_file, access);
6317 clear_access(access, stp);
6318}
6319
6320static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6321{
6322 switch (to_access) {
6323 case NFS4_SHARE_ACCESS_READ:
6324 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6325 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6326 break;
6327 case NFS4_SHARE_ACCESS_WRITE:
6328 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6329 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6330 break;
6331 case NFS4_SHARE_ACCESS_BOTH:
6332 break;
6333 default:
6334 WARN_ON_ONCE(1);
6335 }
6336}
6337
6338__be32
6339nfsd4_open_downgrade(struct svc_rqst *rqstp,
6340 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6341{
6342 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6343 __be32 status;
6344 struct nfs4_ol_stateid *stp;
6345 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6346
6347 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6348 cstate->current_fh.fh_dentry);
6349
6350 /* We don't yet support WANT bits: */
6351 if (od->od_deleg_want)
6352 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6353 od->od_deleg_want);
6354
6355 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6356 &od->od_stateid, &stp, nn);
6357 if (status)
6358 goto out;
6359 status = nfserr_inval;
6360 if (!test_access(od->od_share_access, stp)) {
6361 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6362 stp->st_access_bmap, od->od_share_access);
6363 goto put_stateid;
6364 }
6365 if (!test_deny(od->od_share_deny, stp)) {
6366 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6367 stp->st_deny_bmap, od->od_share_deny);
6368 goto put_stateid;
6369 }
6370 nfs4_stateid_downgrade(stp, od->od_share_access);
6371 reset_union_bmap_deny(od->od_share_deny, stp);
6372 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6373 status = nfs_ok;
6374put_stateid:
6375 mutex_unlock(&stp->st_mutex);
6376 nfs4_put_stid(&stp->st_stid);
6377out:
6378 nfsd4_bump_seqid(cstate, status);
6379 return status;
6380}
6381
6382static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6383{
6384 struct nfs4_client *clp = s->st_stid.sc_client;
6385 bool unhashed;
6386 LIST_HEAD(reaplist);
6387
6388 spin_lock(&clp->cl_lock);
6389 unhashed = unhash_open_stateid(s, &reaplist);
6390
6391 if (clp->cl_minorversion) {
6392 if (unhashed)
6393 put_ol_stateid_locked(s, &reaplist);
6394 spin_unlock(&clp->cl_lock);
6395 free_ol_stateid_reaplist(&reaplist);
6396 } else {
6397 spin_unlock(&clp->cl_lock);
6398 free_ol_stateid_reaplist(&reaplist);
6399 if (unhashed)
6400 move_to_close_lru(s, clp->net);
6401 }
6402}
6403
6404/*
6405 * nfs4_unlock_state() called after encode
6406 */
6407__be32
6408nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6409 union nfsd4_op_u *u)
6410{
6411 struct nfsd4_close *close = &u->close;
6412 __be32 status;
6413 struct nfs4_ol_stateid *stp;
6414 struct net *net = SVC_NET(rqstp);
6415 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6416
6417 dprintk("NFSD: nfsd4_close on file %pd\n",
6418 cstate->current_fh.fh_dentry);
6419
6420 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6421 &close->cl_stateid,
6422 NFS4_OPEN_STID|NFS4_CLOSED_STID,
6423 &stp, nn);
6424 nfsd4_bump_seqid(cstate, status);
6425 if (status)
6426 goto out;
6427
6428 stp->st_stid.sc_type = NFS4_CLOSED_STID;
6429
6430 /*
6431 * Technically we don't _really_ have to increment or copy it, since
6432 * it should just be gone after this operation and we clobber the
6433 * copied value below, but we continue to do so here just to ensure
6434 * that racing ops see that there was a state change.
6435 */
6436 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6437
6438 nfsd4_close_open_stateid(stp);
6439 mutex_unlock(&stp->st_mutex);
6440
6441 /* v4.1+ suggests that we send a special stateid in here, since the
6442 * clients should just ignore this anyway. Since this is not useful
6443 * for v4.0 clients either, we set it to the special close_stateid
6444 * universally.
6445 *
6446 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6447 */
6448 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6449
6450 /* put reference from nfs4_preprocess_seqid_op */
6451 nfs4_put_stid(&stp->st_stid);
6452out:
6453 return status;
6454}
6455
6456__be32
6457nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6458 union nfsd4_op_u *u)
6459{
6460 struct nfsd4_delegreturn *dr = &u->delegreturn;
6461 struct nfs4_delegation *dp;
6462 stateid_t *stateid = &dr->dr_stateid;
6463 struct nfs4_stid *s;
6464 __be32 status;
6465 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6466
6467 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6468 return status;
6469
6470 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6471 if (status)
6472 goto out;
6473 dp = delegstateid(s);
6474 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6475 if (status)
6476 goto put_stateid;
6477
6478 destroy_delegation(dp);
6479put_stateid:
6480 nfs4_put_stid(&dp->dl_stid);
6481out:
6482 return status;
6483}
6484
6485/* last octet in a range */
6486static inline u64
6487last_byte_offset(u64 start, u64 len)
6488{
6489 u64 end;
6490
6491 WARN_ON_ONCE(!len);
6492 end = start + len;
6493 return end > start ? end - 1: NFS4_MAX_UINT64;
6494}
6495
6496/*
6497 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6498 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6499 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
6500 * locking, this prevents us from being completely protocol-compliant. The
6501 * real solution to this problem is to start using unsigned file offsets in
6502 * the VFS, but this is a very deep change!
6503 */
6504static inline void
6505nfs4_transform_lock_offset(struct file_lock *lock)
6506{
6507 if (lock->fl_start < 0)
6508 lock->fl_start = OFFSET_MAX;
6509 if (lock->fl_end < 0)
6510 lock->fl_end = OFFSET_MAX;
6511}
6512
6513static fl_owner_t
6514nfsd4_fl_get_owner(fl_owner_t owner)
6515{
6516 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6517
6518 nfs4_get_stateowner(&lo->lo_owner);
6519 return owner;
6520}
6521
6522static void
6523nfsd4_fl_put_owner(fl_owner_t owner)
6524{
6525 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6526
6527 if (lo)
6528 nfs4_put_stateowner(&lo->lo_owner);
6529}
6530
6531static void
6532nfsd4_lm_notify(struct file_lock *fl)
6533{
6534 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
6535 struct net *net = lo->lo_owner.so_client->net;
6536 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6537 struct nfsd4_blocked_lock *nbl = container_of(fl,
6538 struct nfsd4_blocked_lock, nbl_lock);
6539 bool queue = false;
6540
6541 /* An empty list means that something else is going to be using it */
6542 spin_lock(&nn->blocked_locks_lock);
6543 if (!list_empty(&nbl->nbl_list)) {
6544 list_del_init(&nbl->nbl_list);
6545 list_del_init(&nbl->nbl_lru);
6546 queue = true;
6547 }
6548 spin_unlock(&nn->blocked_locks_lock);
6549
6550 if (queue) {
6551 trace_nfsd_cb_notify_lock(lo, nbl);
6552 nfsd4_run_cb(&nbl->nbl_cb);
6553 }
6554}
6555
6556static const struct lock_manager_operations nfsd_posix_mng_ops = {
6557 .lm_notify = nfsd4_lm_notify,
6558 .lm_get_owner = nfsd4_fl_get_owner,
6559 .lm_put_owner = nfsd4_fl_put_owner,
6560};
6561
6562static inline void
6563nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6564{
6565 struct nfs4_lockowner *lo;
6566
6567 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6568 lo = (struct nfs4_lockowner *) fl->fl_owner;
6569 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6570 GFP_KERNEL);
6571 if (!deny->ld_owner.data)
6572 /* We just don't care that much */
6573 goto nevermind;
6574 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6575 } else {
6576nevermind:
6577 deny->ld_owner.len = 0;
6578 deny->ld_owner.data = NULL;
6579 deny->ld_clientid.cl_boot = 0;
6580 deny->ld_clientid.cl_id = 0;
6581 }
6582 deny->ld_start = fl->fl_start;
6583 deny->ld_length = NFS4_MAX_UINT64;
6584 if (fl->fl_end != NFS4_MAX_UINT64)
6585 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6586 deny->ld_type = NFS4_READ_LT;
6587 if (fl->fl_type != F_RDLCK)
6588 deny->ld_type = NFS4_WRITE_LT;
6589}
6590
6591static struct nfs4_lockowner *
6592find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6593{
6594 unsigned int strhashval = ownerstr_hashval(owner);
6595 struct nfs4_stateowner *so;
6596
6597 lockdep_assert_held(&clp->cl_lock);
6598
6599 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6600 so_strhash) {
6601 if (so->so_is_open_owner)
6602 continue;
6603 if (same_owner_str(so, owner))
6604 return lockowner(nfs4_get_stateowner(so));
6605 }
6606 return NULL;
6607}
6608
6609static struct nfs4_lockowner *
6610find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6611{
6612 struct nfs4_lockowner *lo;
6613
6614 spin_lock(&clp->cl_lock);
6615 lo = find_lockowner_str_locked(clp, owner);
6616 spin_unlock(&clp->cl_lock);
6617 return lo;
6618}
6619
6620static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6621{
6622 unhash_lockowner_locked(lockowner(sop));
6623}
6624
6625static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6626{
6627 struct nfs4_lockowner *lo = lockowner(sop);
6628
6629 kmem_cache_free(lockowner_slab, lo);
6630}
6631
6632static const struct nfs4_stateowner_operations lockowner_ops = {
6633 .so_unhash = nfs4_unhash_lockowner,
6634 .so_free = nfs4_free_lockowner,
6635};
6636
6637/*
6638 * Alloc a lock owner structure.
6639 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6640 * occurred.
6641 *
6642 * strhashval = ownerstr_hashval
6643 */
6644static struct nfs4_lockowner *
6645alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6646 struct nfs4_ol_stateid *open_stp,
6647 struct nfsd4_lock *lock)
6648{
6649 struct nfs4_lockowner *lo, *ret;
6650
6651 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6652 if (!lo)
6653 return NULL;
6654 INIT_LIST_HEAD(&lo->lo_blocked);
6655 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6656 lo->lo_owner.so_is_open_owner = 0;
6657 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6658 lo->lo_owner.so_ops = &lockowner_ops;
6659 spin_lock(&clp->cl_lock);
6660 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6661 if (ret == NULL) {
6662 list_add(&lo->lo_owner.so_strhash,
6663 &clp->cl_ownerstr_hashtbl[strhashval]);
6664 ret = lo;
6665 } else
6666 nfs4_free_stateowner(&lo->lo_owner);
6667
6668 spin_unlock(&clp->cl_lock);
6669 return ret;
6670}
6671
6672static struct nfs4_ol_stateid *
6673find_lock_stateid(const struct nfs4_lockowner *lo,
6674 const struct nfs4_ol_stateid *ost)
6675{
6676 struct nfs4_ol_stateid *lst;
6677
6678 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
6679
6680 /* If ost is not hashed, ost->st_locks will not be valid */
6681 if (!nfs4_ol_stateid_unhashed(ost))
6682 list_for_each_entry(lst, &ost->st_locks, st_locks) {
6683 if (lst->st_stateowner == &lo->lo_owner) {
6684 refcount_inc(&lst->st_stid.sc_count);
6685 return lst;
6686 }
6687 }
6688 return NULL;
6689}
6690
6691static struct nfs4_ol_stateid *
6692init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6693 struct nfs4_file *fp, struct inode *inode,
6694 struct nfs4_ol_stateid *open_stp)
6695{
6696 struct nfs4_client *clp = lo->lo_owner.so_client;
6697 struct nfs4_ol_stateid *retstp;
6698
6699 mutex_init(&stp->st_mutex);
6700 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6701retry:
6702 spin_lock(&clp->cl_lock);
6703 if (nfs4_ol_stateid_unhashed(open_stp))
6704 goto out_close;
6705 retstp = find_lock_stateid(lo, open_stp);
6706 if (retstp)
6707 goto out_found;
6708 refcount_inc(&stp->st_stid.sc_count);
6709 stp->st_stid.sc_type = NFS4_LOCK_STID;
6710 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6711 get_nfs4_file(fp);
6712 stp->st_stid.sc_file = fp;
6713 stp->st_access_bmap = 0;
6714 stp->st_deny_bmap = open_stp->st_deny_bmap;
6715 stp->st_openstp = open_stp;
6716 spin_lock(&fp->fi_lock);
6717 list_add(&stp->st_locks, &open_stp->st_locks);
6718 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6719 list_add(&stp->st_perfile, &fp->fi_stateids);
6720 spin_unlock(&fp->fi_lock);
6721 spin_unlock(&clp->cl_lock);
6722 return stp;
6723out_found:
6724 spin_unlock(&clp->cl_lock);
6725 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6726 nfs4_put_stid(&retstp->st_stid);
6727 goto retry;
6728 }
6729 /* To keep mutex tracking happy */
6730 mutex_unlock(&stp->st_mutex);
6731 return retstp;
6732out_close:
6733 spin_unlock(&clp->cl_lock);
6734 mutex_unlock(&stp->st_mutex);
6735 return NULL;
6736}
6737
6738static struct nfs4_ol_stateid *
6739find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6740 struct inode *inode, struct nfs4_ol_stateid *ost,
6741 bool *new)
6742{
6743 struct nfs4_stid *ns = NULL;
6744 struct nfs4_ol_stateid *lst;
6745 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6746 struct nfs4_client *clp = oo->oo_owner.so_client;
6747
6748 *new = false;
6749 spin_lock(&clp->cl_lock);
6750 lst = find_lock_stateid(lo, ost);
6751 spin_unlock(&clp->cl_lock);
6752 if (lst != NULL) {
6753 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6754 goto out;
6755 nfs4_put_stid(&lst->st_stid);
6756 }
6757 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6758 if (ns == NULL)
6759 return NULL;
6760
6761 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6762 if (lst == openlockstateid(ns))
6763 *new = true;
6764 else
6765 nfs4_put_stid(ns);
6766out:
6767 return lst;
6768}
6769
6770static int
6771check_lock_length(u64 offset, u64 length)
6772{
6773 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6774 (length > ~offset)));
6775}
6776
6777static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6778{
6779 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6780
6781 lockdep_assert_held(&fp->fi_lock);
6782
6783 if (test_access(access, lock_stp))
6784 return;
6785 __nfs4_file_get_access(fp, access);
6786 set_access(access, lock_stp);
6787}
6788
6789static __be32
6790lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6791 struct nfs4_ol_stateid *ost,
6792 struct nfsd4_lock *lock,
6793 struct nfs4_ol_stateid **plst, bool *new)
6794{
6795 __be32 status;
6796 struct nfs4_file *fi = ost->st_stid.sc_file;
6797 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6798 struct nfs4_client *cl = oo->oo_owner.so_client;
6799 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6800 struct nfs4_lockowner *lo;
6801 struct nfs4_ol_stateid *lst;
6802 unsigned int strhashval;
6803
6804 lo = find_lockowner_str(cl, &lock->lk_new_owner);
6805 if (!lo) {
6806 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6807 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6808 if (lo == NULL)
6809 return nfserr_jukebox;
6810 } else {
6811 /* with an existing lockowner, seqids must be the same */
6812 status = nfserr_bad_seqid;
6813 if (!cstate->minorversion &&
6814 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6815 goto out;
6816 }
6817
6818 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6819 if (lst == NULL) {
6820 status = nfserr_jukebox;
6821 goto out;
6822 }
6823
6824 status = nfs_ok;
6825 *plst = lst;
6826out:
6827 nfs4_put_stateowner(&lo->lo_owner);
6828 return status;
6829}
6830
6831/*
6832 * LOCK operation
6833 */
6834__be32
6835nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6836 union nfsd4_op_u *u)
6837{
6838 struct nfsd4_lock *lock = &u->lock;
6839 struct nfs4_openowner *open_sop = NULL;
6840 struct nfs4_lockowner *lock_sop = NULL;
6841 struct nfs4_ol_stateid *lock_stp = NULL;
6842 struct nfs4_ol_stateid *open_stp = NULL;
6843 struct nfs4_file *fp;
6844 struct nfsd_file *nf = NULL;
6845 struct nfsd4_blocked_lock *nbl = NULL;
6846 struct file_lock *file_lock = NULL;
6847 struct file_lock *conflock = NULL;
6848 struct super_block *sb;
6849 __be32 status = 0;
6850 int lkflg;
6851 int err;
6852 bool new = false;
6853 unsigned char fl_type;
6854 unsigned int fl_flags = FL_POSIX;
6855 struct net *net = SVC_NET(rqstp);
6856 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6857
6858 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6859 (long long) lock->lk_offset,
6860 (long long) lock->lk_length);
6861
6862 if (check_lock_length(lock->lk_offset, lock->lk_length))
6863 return nfserr_inval;
6864
6865 if ((status = fh_verify(rqstp, &cstate->current_fh,
6866 S_IFREG, NFSD_MAY_LOCK))) {
6867 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6868 return status;
6869 }
6870 sb = cstate->current_fh.fh_dentry->d_sb;
6871
6872 if (lock->lk_is_new) {
6873 if (nfsd4_has_session(cstate))
6874 /* See rfc 5661 18.10.3: given clientid is ignored: */
6875 memcpy(&lock->lk_new_clientid,
6876 &cstate->clp->cl_clientid,
6877 sizeof(clientid_t));
6878
6879 /* validate and update open stateid and open seqid */
6880 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6881 lock->lk_new_open_seqid,
6882 &lock->lk_new_open_stateid,
6883 &open_stp, nn);
6884 if (status)
6885 goto out;
6886 mutex_unlock(&open_stp->st_mutex);
6887 open_sop = openowner(open_stp->st_stateowner);
6888 status = nfserr_bad_stateid;
6889 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6890 &lock->lk_new_clientid))
6891 goto out;
6892 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6893 &lock_stp, &new);
6894 } else {
6895 status = nfs4_preprocess_seqid_op(cstate,
6896 lock->lk_old_lock_seqid,
6897 &lock->lk_old_lock_stateid,
6898 NFS4_LOCK_STID, &lock_stp, nn);
6899 }
6900 if (status)
6901 goto out;
6902 lock_sop = lockowner(lock_stp->st_stateowner);
6903
6904 lkflg = setlkflg(lock->lk_type);
6905 status = nfs4_check_openmode(lock_stp, lkflg);
6906 if (status)
6907 goto out;
6908
6909 status = nfserr_grace;
6910 if (locks_in_grace(net) && !lock->lk_reclaim)
6911 goto out;
6912 status = nfserr_no_grace;
6913 if (!locks_in_grace(net) && lock->lk_reclaim)
6914 goto out;
6915
6916 fp = lock_stp->st_stid.sc_file;
6917 switch (lock->lk_type) {
6918 case NFS4_READW_LT:
6919 if (nfsd4_has_session(cstate) &&
6920 !(sb->s_export_op->flags & EXPORT_OP_SYNC_LOCKS))
6921 fl_flags |= FL_SLEEP;
6922 fallthrough;
6923 case NFS4_READ_LT:
6924 spin_lock(&fp->fi_lock);
6925 nf = find_readable_file_locked(fp);
6926 if (nf)
6927 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6928 spin_unlock(&fp->fi_lock);
6929 fl_type = F_RDLCK;
6930 break;
6931 case NFS4_WRITEW_LT:
6932 if (nfsd4_has_session(cstate) &&
6933 !(sb->s_export_op->flags & EXPORT_OP_SYNC_LOCKS))
6934 fl_flags |= FL_SLEEP;
6935 fallthrough;
6936 case NFS4_WRITE_LT:
6937 spin_lock(&fp->fi_lock);
6938 nf = find_writeable_file_locked(fp);
6939 if (nf)
6940 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6941 spin_unlock(&fp->fi_lock);
6942 fl_type = F_WRLCK;
6943 break;
6944 default:
6945 status = nfserr_inval;
6946 goto out;
6947 }
6948
6949 if (!nf) {
6950 status = nfserr_openmode;
6951 goto out;
6952 }
6953
6954 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6955 if (!nbl) {
6956 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6957 status = nfserr_jukebox;
6958 goto out;
6959 }
6960
6961 file_lock = &nbl->nbl_lock;
6962 file_lock->fl_type = fl_type;
6963 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6964 file_lock->fl_pid = current->tgid;
6965 file_lock->fl_file = nf->nf_file;
6966 file_lock->fl_flags = fl_flags;
6967 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6968 file_lock->fl_start = lock->lk_offset;
6969 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6970 nfs4_transform_lock_offset(file_lock);
6971
6972 conflock = locks_alloc_lock();
6973 if (!conflock) {
6974 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6975 status = nfserr_jukebox;
6976 goto out;
6977 }
6978
6979 if (fl_flags & FL_SLEEP) {
6980 nbl->nbl_time = ktime_get_boottime_seconds();
6981 spin_lock(&nn->blocked_locks_lock);
6982 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6983 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6984 spin_unlock(&nn->blocked_locks_lock);
6985 }
6986
6987 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6988 switch (err) {
6989 case 0: /* success! */
6990 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6991 status = 0;
6992 if (lock->lk_reclaim)
6993 nn->somebody_reclaimed = true;
6994 break;
6995 case FILE_LOCK_DEFERRED:
6996 nbl = NULL;
6997 fallthrough;
6998 case -EAGAIN: /* conflock holds conflicting lock */
6999 status = nfserr_denied;
7000 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
7001 nfs4_set_lock_denied(conflock, &lock->lk_denied);
7002 break;
7003 case -EDEADLK:
7004 status = nfserr_deadlock;
7005 break;
7006 default:
7007 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
7008 status = nfserrno(err);
7009 break;
7010 }
7011out:
7012 if (nbl) {
7013 /* dequeue it if we queued it before */
7014 if (fl_flags & FL_SLEEP) {
7015 spin_lock(&nn->blocked_locks_lock);
7016 list_del_init(&nbl->nbl_list);
7017 list_del_init(&nbl->nbl_lru);
7018 spin_unlock(&nn->blocked_locks_lock);
7019 }
7020 free_blocked_lock(nbl);
7021 }
7022 if (nf)
7023 nfsd_file_put(nf);
7024 if (lock_stp) {
7025 /* Bump seqid manually if the 4.0 replay owner is openowner */
7026 if (cstate->replay_owner &&
7027 cstate->replay_owner != &lock_sop->lo_owner &&
7028 seqid_mutating_err(ntohl(status)))
7029 lock_sop->lo_owner.so_seqid++;
7030
7031 /*
7032 * If this is a new, never-before-used stateid, and we are
7033 * returning an error, then just go ahead and release it.
7034 */
7035 if (status && new)
7036 release_lock_stateid(lock_stp);
7037
7038 mutex_unlock(&lock_stp->st_mutex);
7039
7040 nfs4_put_stid(&lock_stp->st_stid);
7041 }
7042 if (open_stp)
7043 nfs4_put_stid(&open_stp->st_stid);
7044 nfsd4_bump_seqid(cstate, status);
7045 if (conflock)
7046 locks_free_lock(conflock);
7047 return status;
7048}
7049
7050/*
7051 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
7052 * so we do a temporary open here just to get an open file to pass to
7053 * vfs_test_lock.
7054 */
7055static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
7056{
7057 struct nfsd_file *nf;
7058 __be32 err;
7059
7060 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
7061 if (err)
7062 return err;
7063 fh_lock(fhp); /* to block new leases till after test_lock: */
7064 err = nfserrno(nfsd_open_break_lease(fhp->fh_dentry->d_inode,
7065 NFSD_MAY_READ));
7066 if (err)
7067 goto out;
7068 lock->fl_file = nf->nf_file;
7069 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
7070 lock->fl_file = NULL;
7071out:
7072 fh_unlock(fhp);
7073 nfsd_file_put(nf);
7074 return err;
7075}
7076
7077/*
7078 * LOCKT operation
7079 */
7080__be32
7081nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7082 union nfsd4_op_u *u)
7083{
7084 struct nfsd4_lockt *lockt = &u->lockt;
7085 struct file_lock *file_lock = NULL;
7086 struct nfs4_lockowner *lo = NULL;
7087 __be32 status;
7088 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7089
7090 if (locks_in_grace(SVC_NET(rqstp)))
7091 return nfserr_grace;
7092
7093 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
7094 return nfserr_inval;
7095
7096 if (!nfsd4_has_session(cstate)) {
7097 status = set_client(&lockt->lt_clientid, cstate, nn);
7098 if (status)
7099 goto out;
7100 }
7101
7102 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7103 goto out;
7104
7105 file_lock = locks_alloc_lock();
7106 if (!file_lock) {
7107 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7108 status = nfserr_jukebox;
7109 goto out;
7110 }
7111
7112 switch (lockt->lt_type) {
7113 case NFS4_READ_LT:
7114 case NFS4_READW_LT:
7115 file_lock->fl_type = F_RDLCK;
7116 break;
7117 case NFS4_WRITE_LT:
7118 case NFS4_WRITEW_LT:
7119 file_lock->fl_type = F_WRLCK;
7120 break;
7121 default:
7122 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
7123 status = nfserr_inval;
7124 goto out;
7125 }
7126
7127 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
7128 if (lo)
7129 file_lock->fl_owner = (fl_owner_t)lo;
7130 file_lock->fl_pid = current->tgid;
7131 file_lock->fl_flags = FL_POSIX;
7132
7133 file_lock->fl_start = lockt->lt_offset;
7134 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
7135
7136 nfs4_transform_lock_offset(file_lock);
7137
7138 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
7139 if (status)
7140 goto out;
7141
7142 if (file_lock->fl_type != F_UNLCK) {
7143 status = nfserr_denied;
7144 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
7145 }
7146out:
7147 if (lo)
7148 nfs4_put_stateowner(&lo->lo_owner);
7149 if (file_lock)
7150 locks_free_lock(file_lock);
7151 return status;
7152}
7153
7154__be32
7155nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7156 union nfsd4_op_u *u)
7157{
7158 struct nfsd4_locku *locku = &u->locku;
7159 struct nfs4_ol_stateid *stp;
7160 struct nfsd_file *nf = NULL;
7161 struct file_lock *file_lock = NULL;
7162 __be32 status;
7163 int err;
7164 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7165
7166 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
7167 (long long) locku->lu_offset,
7168 (long long) locku->lu_length);
7169
7170 if (check_lock_length(locku->lu_offset, locku->lu_length))
7171 return nfserr_inval;
7172
7173 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
7174 &locku->lu_stateid, NFS4_LOCK_STID,
7175 &stp, nn);
7176 if (status)
7177 goto out;
7178 nf = find_any_file(stp->st_stid.sc_file);
7179 if (!nf) {
7180 status = nfserr_lock_range;
7181 goto put_stateid;
7182 }
7183 file_lock = locks_alloc_lock();
7184 if (!file_lock) {
7185 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7186 status = nfserr_jukebox;
7187 goto put_file;
7188 }
7189
7190 file_lock->fl_type = F_UNLCK;
7191 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7192 file_lock->fl_pid = current->tgid;
7193 file_lock->fl_file = nf->nf_file;
7194 file_lock->fl_flags = FL_POSIX;
7195 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7196 file_lock->fl_start = locku->lu_offset;
7197
7198 file_lock->fl_end = last_byte_offset(locku->lu_offset,
7199 locku->lu_length);
7200 nfs4_transform_lock_offset(file_lock);
7201
7202 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7203 if (err) {
7204 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7205 goto out_nfserr;
7206 }
7207 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7208put_file:
7209 nfsd_file_put(nf);
7210put_stateid:
7211 mutex_unlock(&stp->st_mutex);
7212 nfs4_put_stid(&stp->st_stid);
7213out:
7214 nfsd4_bump_seqid(cstate, status);
7215 if (file_lock)
7216 locks_free_lock(file_lock);
7217 return status;
7218
7219out_nfserr:
7220 status = nfserrno(err);
7221 goto put_file;
7222}
7223
7224/*
7225 * returns
7226 * true: locks held by lockowner
7227 * false: no locks held by lockowner
7228 */
7229static bool
7230check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7231{
7232 struct file_lock *fl;
7233 int status = false;
7234 struct nfsd_file *nf = find_any_file(fp);
7235 struct inode *inode;
7236 struct file_lock_context *flctx;
7237
7238 if (!nf) {
7239 /* Any valid lock stateid should have some sort of access */
7240 WARN_ON_ONCE(1);
7241 return status;
7242 }
7243
7244 inode = locks_inode(nf->nf_file);
7245 flctx = inode->i_flctx;
7246
7247 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7248 spin_lock(&flctx->flc_lock);
7249 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7250 if (fl->fl_owner == (fl_owner_t)lowner) {
7251 status = true;
7252 break;
7253 }
7254 }
7255 spin_unlock(&flctx->flc_lock);
7256 }
7257 nfsd_file_put(nf);
7258 return status;
7259}
7260
7261__be32
7262nfsd4_release_lockowner(struct svc_rqst *rqstp,
7263 struct nfsd4_compound_state *cstate,
7264 union nfsd4_op_u *u)
7265{
7266 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7267 clientid_t *clid = &rlockowner->rl_clientid;
7268 struct nfs4_stateowner *sop;
7269 struct nfs4_lockowner *lo = NULL;
7270 struct nfs4_ol_stateid *stp;
7271 struct xdr_netobj *owner = &rlockowner->rl_owner;
7272 unsigned int hashval = ownerstr_hashval(owner);
7273 __be32 status;
7274 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7275 struct nfs4_client *clp;
7276 LIST_HEAD (reaplist);
7277
7278 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7279 clid->cl_boot, clid->cl_id);
7280
7281 status = set_client(clid, cstate, nn);
7282 if (status)
7283 return status;
7284
7285 clp = cstate->clp;
7286 /* Find the matching lock stateowner */
7287 spin_lock(&clp->cl_lock);
7288 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
7289 so_strhash) {
7290
7291 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
7292 continue;
7293
7294 /* see if there are still any locks associated with it */
7295 lo = lockowner(sop);
7296 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
7297 if (check_for_locks(stp->st_stid.sc_file, lo)) {
7298 status = nfserr_locks_held;
7299 spin_unlock(&clp->cl_lock);
7300 return status;
7301 }
7302 }
7303
7304 nfs4_get_stateowner(sop);
7305 break;
7306 }
7307 if (!lo) {
7308 spin_unlock(&clp->cl_lock);
7309 return status;
7310 }
7311
7312 unhash_lockowner_locked(lo);
7313 while (!list_empty(&lo->lo_owner.so_stateids)) {
7314 stp = list_first_entry(&lo->lo_owner.so_stateids,
7315 struct nfs4_ol_stateid,
7316 st_perstateowner);
7317 WARN_ON(!unhash_lock_stateid(stp));
7318 put_ol_stateid_locked(stp, &reaplist);
7319 }
7320 spin_unlock(&clp->cl_lock);
7321 free_ol_stateid_reaplist(&reaplist);
7322 remove_blocked_locks(lo);
7323 nfs4_put_stateowner(&lo->lo_owner);
7324
7325 return status;
7326}
7327
7328static inline struct nfs4_client_reclaim *
7329alloc_reclaim(void)
7330{
7331 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7332}
7333
7334bool
7335nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7336{
7337 struct nfs4_client_reclaim *crp;
7338
7339 crp = nfsd4_find_reclaim_client(name, nn);
7340 return (crp && crp->cr_clp);
7341}
7342
7343/*
7344 * failure => all reset bets are off, nfserr_no_grace...
7345 *
7346 * The caller is responsible for freeing name.data if NULL is returned (it
7347 * will be freed in nfs4_remove_reclaim_record in the normal case).
7348 */
7349struct nfs4_client_reclaim *
7350nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7351 struct nfsd_net *nn)
7352{
7353 unsigned int strhashval;
7354 struct nfs4_client_reclaim *crp;
7355
7356 crp = alloc_reclaim();
7357 if (crp) {
7358 strhashval = clientstr_hashval(name);
7359 INIT_LIST_HEAD(&crp->cr_strhash);
7360 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7361 crp->cr_name.data = name.data;
7362 crp->cr_name.len = name.len;
7363 crp->cr_princhash.data = princhash.data;
7364 crp->cr_princhash.len = princhash.len;
7365 crp->cr_clp = NULL;
7366 nn->reclaim_str_hashtbl_size++;
7367 }
7368 return crp;
7369}
7370
7371void
7372nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7373{
7374 list_del(&crp->cr_strhash);
7375 kfree(crp->cr_name.data);
7376 kfree(crp->cr_princhash.data);
7377 kfree(crp);
7378 nn->reclaim_str_hashtbl_size--;
7379}
7380
7381void
7382nfs4_release_reclaim(struct nfsd_net *nn)
7383{
7384 struct nfs4_client_reclaim *crp = NULL;
7385 int i;
7386
7387 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7388 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7389 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7390 struct nfs4_client_reclaim, cr_strhash);
7391 nfs4_remove_reclaim_record(crp, nn);
7392 }
7393 }
7394 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7395}
7396
7397/*
7398 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7399struct nfs4_client_reclaim *
7400nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7401{
7402 unsigned int strhashval;
7403 struct nfs4_client_reclaim *crp = NULL;
7404
7405 strhashval = clientstr_hashval(name);
7406 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7407 if (compare_blob(&crp->cr_name, &name) == 0) {
7408 return crp;
7409 }
7410 }
7411 return NULL;
7412}
7413
7414__be32
7415nfs4_check_open_reclaim(struct nfs4_client *clp)
7416{
7417 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
7418 return nfserr_no_grace;
7419
7420 if (nfsd4_client_record_check(clp))
7421 return nfserr_reclaim_bad;
7422
7423 return nfs_ok;
7424}
7425
7426/*
7427 * Since the lifetime of a delegation isn't limited to that of an open, a
7428 * client may quite reasonably hang on to a delegation as long as it has
7429 * the inode cached. This becomes an obvious problem the first time a
7430 * client's inode cache approaches the size of the server's total memory.
7431 *
7432 * For now we avoid this problem by imposing a hard limit on the number
7433 * of delegations, which varies according to the server's memory size.
7434 */
7435static void
7436set_max_delegations(void)
7437{
7438 /*
7439 * Allow at most 4 delegations per megabyte of RAM. Quick
7440 * estimates suggest that in the worst case (where every delegation
7441 * is for a different inode), a delegation could take about 1.5K,
7442 * giving a worst case usage of about 6% of memory.
7443 */
7444 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7445}
7446
7447static int nfs4_state_create_net(struct net *net)
7448{
7449 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7450 int i;
7451
7452 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7453 sizeof(struct list_head),
7454 GFP_KERNEL);
7455 if (!nn->conf_id_hashtbl)
7456 goto err;
7457 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7458 sizeof(struct list_head),
7459 GFP_KERNEL);
7460 if (!nn->unconf_id_hashtbl)
7461 goto err_unconf_id;
7462 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7463 sizeof(struct list_head),
7464 GFP_KERNEL);
7465 if (!nn->sessionid_hashtbl)
7466 goto err_sessionid;
7467
7468 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7469 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7470 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7471 }
7472 for (i = 0; i < SESSION_HASH_SIZE; i++)
7473 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7474 nn->conf_name_tree = RB_ROOT;
7475 nn->unconf_name_tree = RB_ROOT;
7476 nn->boot_time = ktime_get_real_seconds();
7477 nn->grace_ended = false;
7478 nn->nfsd4_manager.block_opens = true;
7479 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7480 INIT_LIST_HEAD(&nn->client_lru);
7481 INIT_LIST_HEAD(&nn->close_lru);
7482 INIT_LIST_HEAD(&nn->del_recall_lru);
7483 spin_lock_init(&nn->client_lock);
7484 spin_lock_init(&nn->s2s_cp_lock);
7485 idr_init(&nn->s2s_cp_stateids);
7486
7487 spin_lock_init(&nn->blocked_locks_lock);
7488 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7489
7490 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7491 get_net(net);
7492
7493 return 0;
7494
7495err_sessionid:
7496 kfree(nn->unconf_id_hashtbl);
7497err_unconf_id:
7498 kfree(nn->conf_id_hashtbl);
7499err:
7500 return -ENOMEM;
7501}
7502
7503static void
7504nfs4_state_destroy_net(struct net *net)
7505{
7506 int i;
7507 struct nfs4_client *clp = NULL;
7508 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7509
7510 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7511 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7512 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7513 destroy_client(clp);
7514 }
7515 }
7516
7517 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7518
7519 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7520 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7521 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7522 destroy_client(clp);
7523 }
7524 }
7525
7526 kfree(nn->sessionid_hashtbl);
7527 kfree(nn->unconf_id_hashtbl);
7528 kfree(nn->conf_id_hashtbl);
7529 put_net(net);
7530}
7531
7532int
7533nfs4_state_start_net(struct net *net)
7534{
7535 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7536 int ret;
7537
7538 ret = nfs4_state_create_net(net);
7539 if (ret)
7540 return ret;
7541 locks_start_grace(net, &nn->nfsd4_manager);
7542 nfsd4_client_tracking_init(net);
7543 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7544 goto skip_grace;
7545 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
7546 nn->nfsd4_grace, net->ns.inum);
7547 trace_nfsd_grace_start(nn);
7548 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7549 return 0;
7550
7551skip_grace:
7552 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7553 net->ns.inum);
7554 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7555 nfsd4_end_grace(nn);
7556 return 0;
7557}
7558
7559/* initialization to perform when the nfsd service is started: */
7560
7561int
7562nfs4_state_start(void)
7563{
7564 int ret;
7565
7566 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7567 if (laundry_wq == NULL) {
7568 ret = -ENOMEM;
7569 goto out;
7570 }
7571 ret = nfsd4_create_callback_queue();
7572 if (ret)
7573 goto out_free_laundry;
7574
7575 set_max_delegations();
7576 return 0;
7577
7578out_free_laundry:
7579 destroy_workqueue(laundry_wq);
7580out:
7581 return ret;
7582}
7583
7584void
7585nfs4_state_shutdown_net(struct net *net)
7586{
7587 struct nfs4_delegation *dp = NULL;
7588 struct list_head *pos, *next, reaplist;
7589 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7590
7591 cancel_delayed_work_sync(&nn->laundromat_work);
7592 locks_end_grace(&nn->nfsd4_manager);
7593
7594 INIT_LIST_HEAD(&reaplist);
7595 spin_lock(&state_lock);
7596 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7597 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7598 WARN_ON(!unhash_delegation_locked(dp));
7599 list_add(&dp->dl_recall_lru, &reaplist);
7600 }
7601 spin_unlock(&state_lock);
7602 list_for_each_safe(pos, next, &reaplist) {
7603 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7604 list_del_init(&dp->dl_recall_lru);
7605 destroy_unhashed_deleg(dp);
7606 }
7607
7608 nfsd4_client_tracking_exit(net);
7609 nfs4_state_destroy_net(net);
7610#ifdef CONFIG_NFSD_V4_2_INTER_SSC
7611 nfsd4_ssc_shutdown_umount(nn);
7612#endif
7613}
7614
7615void
7616nfs4_state_shutdown(void)
7617{
7618 destroy_workqueue(laundry_wq);
7619 nfsd4_destroy_callback_queue();
7620}
7621
7622static void
7623get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7624{
7625 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
7626 CURRENT_STATEID(stateid))
7627 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7628}
7629
7630static void
7631put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7632{
7633 if (cstate->minorversion) {
7634 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7635 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7636 }
7637}
7638
7639void
7640clear_current_stateid(struct nfsd4_compound_state *cstate)
7641{
7642 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7643}
7644
7645/*
7646 * functions to set current state id
7647 */
7648void
7649nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7650 union nfsd4_op_u *u)
7651{
7652 put_stateid(cstate, &u->open_downgrade.od_stateid);
7653}
7654
7655void
7656nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7657 union nfsd4_op_u *u)
7658{
7659 put_stateid(cstate, &u->open.op_stateid);
7660}
7661
7662void
7663nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7664 union nfsd4_op_u *u)
7665{
7666 put_stateid(cstate, &u->close.cl_stateid);
7667}
7668
7669void
7670nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7671 union nfsd4_op_u *u)
7672{
7673 put_stateid(cstate, &u->lock.lk_resp_stateid);
7674}
7675
7676/*
7677 * functions to consume current state id
7678 */
7679
7680void
7681nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7682 union nfsd4_op_u *u)
7683{
7684 get_stateid(cstate, &u->open_downgrade.od_stateid);
7685}
7686
7687void
7688nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7689 union nfsd4_op_u *u)
7690{
7691 get_stateid(cstate, &u->delegreturn.dr_stateid);
7692}
7693
7694void
7695nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7696 union nfsd4_op_u *u)
7697{
7698 get_stateid(cstate, &u->free_stateid.fr_stateid);
7699}
7700
7701void
7702nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7703 union nfsd4_op_u *u)
7704{
7705 get_stateid(cstate, &u->setattr.sa_stateid);
7706}
7707
7708void
7709nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7710 union nfsd4_op_u *u)
7711{
7712 get_stateid(cstate, &u->close.cl_stateid);
7713}
7714
7715void
7716nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7717 union nfsd4_op_u *u)
7718{
7719 get_stateid(cstate, &u->locku.lu_stateid);
7720}
7721
7722void
7723nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7724 union nfsd4_op_u *u)
7725{
7726 get_stateid(cstate, &u->read.rd_stateid);
7727}
7728
7729void
7730nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7731 union nfsd4_op_u *u)
7732{
7733 get_stateid(cstate, &u->write.wr_stateid);
7734}