2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
59 #define NFSDDBG_FACILITY NFSDDBG_PROC
62 static time_t lease_time = 90; /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
69 static stateid_t zerostateid; /* bits all 0 */
70 static stateid_t onestateid; /* bits all 1 */
71 static u64 current_sessionid = 1;
73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
74 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
76 /* forward declarations */
77 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
78 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir);
84 /* Currently used for almost all code touching nfsv4 state: */
85 static DEFINE_MUTEX(client_mutex);
88 * Currently used for the del_recall_lru and file hash table. In an
89 * effort to decrease the scope of the client_mutex, this spinlock may
90 * eventually cover more:
92 static DEFINE_SPINLOCK(recall_lock);
94 static struct kmem_cache *stateowner_slab = NULL;
95 static struct kmem_cache *file_slab = NULL;
96 static struct kmem_cache *stateid_slab = NULL;
97 static struct kmem_cache *deleg_slab = NULL;
100 nfs4_lock_state(void)
102 mutex_lock(&client_mutex);
106 nfs4_unlock_state(void)
108 mutex_unlock(&client_mutex);
112 opaque_hashval(const void *ptr, int nbytes)
114 unsigned char *cptr = (unsigned char *) ptr;
124 static struct list_head del_recall_lru;
127 put_nfs4_file(struct nfs4_file *fi)
129 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
130 list_del(&fi->fi_hash);
131 spin_unlock(&recall_lock);
133 kmem_cache_free(file_slab, fi);
138 get_nfs4_file(struct nfs4_file *fi)
140 atomic_inc(&fi->fi_ref);
143 static int num_delegations;
144 unsigned int max_delegations;
147 * Open owner state (share locks)
150 /* hash tables for nfs4_stateowner */
151 #define OWNER_HASH_BITS 8
152 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
153 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
155 #define ownerid_hashval(id) \
156 ((id) & OWNER_HASH_MASK)
157 #define ownerstr_hashval(clientid, ownername) \
158 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
160 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
161 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
163 /* hash table for nfs4_file */
164 #define FILE_HASH_BITS 8
165 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
166 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
167 /* hash table for (open)nfs4_stateid */
168 #define STATEID_HASH_BITS 10
169 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
170 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
172 #define file_hashval(x) \
173 hash_ptr(x, FILE_HASH_BITS)
174 #define stateid_hashval(owner_id, file_id) \
175 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
177 static struct list_head file_hashtbl[FILE_HASH_SIZE];
178 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
180 static struct nfs4_delegation *
181 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
183 struct nfs4_delegation *dp;
184 struct nfs4_file *fp = stp->st_file;
185 struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
187 dprintk("NFSD alloc_init_deleg\n");
188 if (fp->fi_had_conflict)
190 if (num_delegations > max_delegations)
192 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
196 INIT_LIST_HEAD(&dp->dl_perfile);
197 INIT_LIST_HEAD(&dp->dl_perclnt);
198 INIT_LIST_HEAD(&dp->dl_recall_lru);
203 get_file(stp->st_vfs_file);
204 dp->dl_vfs_file = stp->st_vfs_file;
206 dp->dl_recall.cbr_dp = NULL;
207 dp->dl_recall.cbr_ident = cb->cb_ident;
208 dp->dl_recall.cbr_trunc = 0;
209 dp->dl_stateid.si_boot = get_seconds();
210 dp->dl_stateid.si_stateownerid = current_delegid++;
211 dp->dl_stateid.si_fileid = 0;
212 dp->dl_stateid.si_generation = 0;
213 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
215 atomic_set(&dp->dl_count, 1);
216 list_add(&dp->dl_perfile, &fp->fi_delegations);
217 list_add(&dp->dl_perclnt, &clp->cl_delegations);
222 nfs4_put_delegation(struct nfs4_delegation *dp)
224 if (atomic_dec_and_test(&dp->dl_count)) {
225 dprintk("NFSD: freeing dp %p\n",dp);
226 put_nfs4_file(dp->dl_file);
227 kmem_cache_free(deleg_slab, dp);
232 /* Remove the associated file_lock first, then remove the delegation.
233 * lease_modify() is called to remove the FS_LEASE file_lock from
234 * the i_flock list, eventually calling nfsd's lock_manager
235 * fl_release_callback.
238 nfs4_close_delegation(struct nfs4_delegation *dp)
240 struct file *filp = dp->dl_vfs_file;
242 dprintk("NFSD: close_delegation dp %p\n",dp);
243 dp->dl_vfs_file = NULL;
244 /* The following nfsd_close may not actually close the file,
245 * but we want to remove the lease in any case. */
247 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
251 /* Called under the state lock. */
253 unhash_delegation(struct nfs4_delegation *dp)
255 list_del_init(&dp->dl_perfile);
256 list_del_init(&dp->dl_perclnt);
257 spin_lock(&recall_lock);
258 list_del_init(&dp->dl_recall_lru);
259 spin_unlock(&recall_lock);
260 nfs4_close_delegation(dp);
261 nfs4_put_delegation(dp);
268 /* Hash tables for nfs4_clientid state */
269 #define CLIENT_HASH_BITS 4
270 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
271 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
273 #define clientid_hashval(id) \
274 ((id) & CLIENT_HASH_MASK)
275 #define clientstr_hashval(name) \
276 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
278 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
279 * used in reboot/reset lease grace period processing
281 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
282 * setclientid_confirmed info.
284 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
287 * client_lru holds client queue ordered by nfs4_client.cl_time
290 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
291 * for last close replay.
293 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
294 static int reclaim_str_hashtbl_size = 0;
295 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
299 static struct list_head client_lru;
300 static struct list_head close_lru;
302 static void unhash_generic_stateid(struct nfs4_stateid *stp)
304 list_del(&stp->st_hash);
305 list_del(&stp->st_perfile);
306 list_del(&stp->st_perstateowner);
309 static void free_generic_stateid(struct nfs4_stateid *stp)
311 put_nfs4_file(stp->st_file);
312 kmem_cache_free(stateid_slab, stp);
315 static void release_lock_stateid(struct nfs4_stateid *stp)
317 unhash_generic_stateid(stp);
318 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
319 free_generic_stateid(stp);
322 static void unhash_lockowner(struct nfs4_stateowner *sop)
324 struct nfs4_stateid *stp;
326 list_del(&sop->so_idhash);
327 list_del(&sop->so_strhash);
328 list_del(&sop->so_perstateid);
329 while (!list_empty(&sop->so_stateids)) {
330 stp = list_first_entry(&sop->so_stateids,
331 struct nfs4_stateid, st_perstateowner);
332 release_lock_stateid(stp);
336 static void release_lockowner(struct nfs4_stateowner *sop)
338 unhash_lockowner(sop);
339 nfs4_put_stateowner(sop);
343 release_stateid_lockowners(struct nfs4_stateid *open_stp)
345 struct nfs4_stateowner *lock_sop;
347 while (!list_empty(&open_stp->st_lockowners)) {
348 lock_sop = list_entry(open_stp->st_lockowners.next,
349 struct nfs4_stateowner, so_perstateid);
350 /* list_del(&open_stp->st_lockowners); */
351 BUG_ON(lock_sop->so_is_open_owner);
352 release_lockowner(lock_sop);
356 static void release_open_stateid(struct nfs4_stateid *stp)
358 unhash_generic_stateid(stp);
359 release_stateid_lockowners(stp);
360 nfsd_close(stp->st_vfs_file);
361 free_generic_stateid(stp);
364 static void unhash_openowner(struct nfs4_stateowner *sop)
366 struct nfs4_stateid *stp;
368 list_del(&sop->so_idhash);
369 list_del(&sop->so_strhash);
370 list_del(&sop->so_perclient);
371 list_del(&sop->so_perstateid); /* XXX: necessary? */
372 while (!list_empty(&sop->so_stateids)) {
373 stp = list_first_entry(&sop->so_stateids,
374 struct nfs4_stateid, st_perstateowner);
375 release_open_stateid(stp);
379 static void release_openowner(struct nfs4_stateowner *sop)
381 unhash_openowner(sop);
382 list_del(&sop->so_close_lru);
383 nfs4_put_stateowner(sop);
386 static DEFINE_SPINLOCK(sessionid_lock);
387 #define SESSION_HASH_SIZE 512
388 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
391 hash_sessionid(struct nfs4_sessionid *sessionid)
393 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
395 return sid->sequence % SESSION_HASH_SIZE;
399 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
401 u32 *ptr = (u32 *)(&sessionid->data[0]);
402 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
406 gen_sessionid(struct nfsd4_session *ses)
408 struct nfs4_client *clp = ses->se_client;
409 struct nfsd4_sessionid *sid;
411 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
412 sid->clientid = clp->cl_clientid;
413 sid->sequence = current_sessionid++;
418 * Give the client the number of slots it requests bound by
419 * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
421 * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
422 * should (up to a point) re-negotiate active sessions and reduce their
423 * slot usage to make rooom for new connections. For now we just fail the
426 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
428 int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
430 spin_lock(&nfsd_serv->sv_lock);
431 if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
432 np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
433 nfsd_serv->sv_drc_pages_used += np;
434 spin_unlock(&nfsd_serv->sv_lock);
437 status = nfserr_resource;
440 fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
446 * fchan holds the client values on input, and the server values on output
448 static int init_forechannel_attrs(struct svc_rqst *rqstp,
449 struct nfsd4_session *session,
450 struct nfsd4_channel_attrs *fchan)
453 __u32 maxcount = svc_max_payload(rqstp);
455 /* headerpadsz set to zero in encode routine */
457 /* Use the client's max request and max response size if possible */
458 if (fchan->maxreq_sz > maxcount)
459 fchan->maxreq_sz = maxcount;
460 session->se_fmaxreq_sz = fchan->maxreq_sz;
462 if (fchan->maxresp_sz > maxcount)
463 fchan->maxresp_sz = maxcount;
464 session->se_fmaxresp_sz = fchan->maxresp_sz;
466 /* Set the max response cached size our default which is
467 * a multiple of PAGE_SIZE and small */
468 session->se_fmaxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
469 fchan->maxresp_cached = session->se_fmaxresp_cached;
471 /* Use the client's maxops if possible */
472 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
473 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
474 session->se_fmaxops = fchan->maxops;
476 /* try to use the client requested number of slots */
477 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
478 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
480 /* FIXME: Error means no more DRC pages so the server should
481 * recover pages from existing sessions. For now fail session
484 status = set_forechannel_maxreqs(fchan);
486 session->se_fnumslots = fchan->maxreqs;
491 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
492 struct nfsd4_create_session *cses)
494 struct nfsd4_session *new, tmp;
495 int idx, status = nfserr_resource, slotsize;
497 memset(&tmp, 0, sizeof(tmp));
499 /* FIXME: For now, we just accept the client back channel attributes. */
500 status = init_forechannel_attrs(rqstp, &tmp, &cses->fore_channel);
504 /* allocate struct nfsd4_session and slot table in one piece */
505 slotsize = tmp.se_fnumslots * sizeof(struct nfsd4_slot);
506 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
510 memcpy(new, &tmp, sizeof(*new));
512 new->se_client = clp;
514 idx = hash_sessionid(&new->se_sessionid);
515 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
516 NFS4_MAX_SESSIONID_LEN);
518 new->se_flags = cses->flags;
519 kref_init(&new->se_ref);
520 spin_lock(&sessionid_lock);
521 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
522 list_add(&new->se_perclnt, &clp->cl_sessions);
523 spin_unlock(&sessionid_lock);
530 /* caller must hold sessionid_lock */
531 static struct nfsd4_session *
532 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
534 struct nfsd4_session *elem;
537 dump_sessionid(__func__, sessionid);
538 idx = hash_sessionid(sessionid);
539 dprintk("%s: idx is %d\n", __func__, idx);
540 /* Search in the appropriate list */
541 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
542 dump_sessionid("list traversal", &elem->se_sessionid);
543 if (!memcmp(elem->se_sessionid.data, sessionid->data,
544 NFS4_MAX_SESSIONID_LEN)) {
549 dprintk("%s: session not found\n", __func__);
553 /* caller must hold sessionid_lock */
555 unhash_session(struct nfsd4_session *ses)
557 list_del(&ses->se_hash);
558 list_del(&ses->se_perclnt);
562 release_session(struct nfsd4_session *ses)
564 spin_lock(&sessionid_lock);
566 spin_unlock(&sessionid_lock);
567 nfsd4_put_session(ses);
570 static void nfsd4_release_respages(struct page **respages, short resused);
573 free_session(struct kref *kref)
575 struct nfsd4_session *ses;
578 ses = container_of(kref, struct nfsd4_session, se_ref);
579 for (i = 0; i < ses->se_fnumslots; i++) {
580 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
581 nfsd4_release_respages(e->ce_respages, e->ce_resused);
583 kfree(ses->se_slots);
588 renew_client(struct nfs4_client *clp)
591 * Move client to the end to the LRU list.
593 dprintk("renewing client (clientid %08x/%08x)\n",
594 clp->cl_clientid.cl_boot,
595 clp->cl_clientid.cl_id);
596 list_move_tail(&clp->cl_lru, &client_lru);
597 clp->cl_time = get_seconds();
600 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
602 STALE_CLIENTID(clientid_t *clid)
604 if (clid->cl_boot == boot_time)
606 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
607 clid->cl_boot, clid->cl_id, boot_time);
612 * XXX Should we use a slab cache ?
613 * This type of memory management is somewhat inefficient, but we use it
614 * anyway since SETCLIENTID is not a common operation.
616 static struct nfs4_client *alloc_client(struct xdr_netobj name)
618 struct nfs4_client *clp;
620 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
623 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
624 if (clp->cl_name.data == NULL) {
628 memcpy(clp->cl_name.data, name.data, name.len);
629 clp->cl_name.len = name.len;
634 shutdown_callback_client(struct nfs4_client *clp)
636 struct rpc_clnt *clnt = clp->cl_callback.cb_client;
640 * Callback threads take a reference on the client, so there
641 * should be no outstanding callbacks at this point.
643 clp->cl_callback.cb_client = NULL;
644 rpc_shutdown_client(clnt);
649 free_client(struct nfs4_client *clp)
651 shutdown_callback_client(clp);
652 nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
653 clp->cl_slot.sl_cache_entry.ce_resused);
654 if (clp->cl_cred.cr_group_info)
655 put_group_info(clp->cl_cred.cr_group_info);
656 kfree(clp->cl_principal);
657 kfree(clp->cl_name.data);
662 put_nfs4_client(struct nfs4_client *clp)
664 if (atomic_dec_and_test(&clp->cl_count))
669 expire_client(struct nfs4_client *clp)
671 struct nfs4_stateowner *sop;
672 struct nfs4_delegation *dp;
673 struct list_head reaplist;
675 dprintk("NFSD: expire_client cl_count %d\n",
676 atomic_read(&clp->cl_count));
678 INIT_LIST_HEAD(&reaplist);
679 spin_lock(&recall_lock);
680 while (!list_empty(&clp->cl_delegations)) {
681 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
682 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
684 list_del_init(&dp->dl_perclnt);
685 list_move(&dp->dl_recall_lru, &reaplist);
687 spin_unlock(&recall_lock);
688 while (!list_empty(&reaplist)) {
689 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
690 list_del_init(&dp->dl_recall_lru);
691 unhash_delegation(dp);
693 list_del(&clp->cl_idhash);
694 list_del(&clp->cl_strhash);
695 list_del(&clp->cl_lru);
696 while (!list_empty(&clp->cl_openowners)) {
697 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
698 release_openowner(sop);
700 while (!list_empty(&clp->cl_sessions)) {
701 struct nfsd4_session *ses;
702 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
704 release_session(ses);
706 put_nfs4_client(clp);
709 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
711 struct nfs4_client *clp;
713 clp = alloc_client(name);
716 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
717 atomic_set(&clp->cl_count, 1);
718 atomic_set(&clp->cl_callback.cb_set, 0);
719 INIT_LIST_HEAD(&clp->cl_idhash);
720 INIT_LIST_HEAD(&clp->cl_strhash);
721 INIT_LIST_HEAD(&clp->cl_openowners);
722 INIT_LIST_HEAD(&clp->cl_delegations);
723 INIT_LIST_HEAD(&clp->cl_sessions);
724 INIT_LIST_HEAD(&clp->cl_lru);
728 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
730 memcpy(target->cl_verifier.data, source->data,
731 sizeof(target->cl_verifier.data));
734 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
736 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
737 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
740 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
742 target->cr_uid = source->cr_uid;
743 target->cr_gid = source->cr_gid;
744 target->cr_group_info = source->cr_group_info;
745 get_group_info(target->cr_group_info);
748 static int same_name(const char *n1, const char *n2)
750 return 0 == memcmp(n1, n2, HEXDIR_LEN);
754 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
756 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
760 same_clid(clientid_t *cl1, clientid_t *cl2)
762 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
765 /* XXX what about NGROUP */
767 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
769 return cr1->cr_uid == cr2->cr_uid;
772 static void gen_clid(struct nfs4_client *clp)
774 static u32 current_clientid = 1;
776 clp->cl_clientid.cl_boot = boot_time;
777 clp->cl_clientid.cl_id = current_clientid++;
780 static void gen_confirm(struct nfs4_client *clp)
785 p = (u32 *)clp->cl_confirm.data;
786 *p++ = get_seconds();
790 static int check_name(struct xdr_netobj name)
794 if (name.len > NFS4_OPAQUE_LIMIT) {
795 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
802 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
804 unsigned int idhashval;
806 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
807 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
808 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
809 list_add_tail(&clp->cl_lru, &client_lru);
810 clp->cl_time = get_seconds();
814 move_to_confirmed(struct nfs4_client *clp)
816 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
817 unsigned int strhashval;
819 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
820 list_del_init(&clp->cl_strhash);
821 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
822 strhashval = clientstr_hashval(clp->cl_recdir);
823 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
827 static struct nfs4_client *
828 find_confirmed_client(clientid_t *clid)
830 struct nfs4_client *clp;
831 unsigned int idhashval = clientid_hashval(clid->cl_id);
833 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
834 if (same_clid(&clp->cl_clientid, clid))
840 static struct nfs4_client *
841 find_unconfirmed_client(clientid_t *clid)
843 struct nfs4_client *clp;
844 unsigned int idhashval = clientid_hashval(clid->cl_id);
846 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
847 if (same_clid(&clp->cl_clientid, clid))
854 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
855 * parameter. Matching is based on the fact the at least one of the
856 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
858 * FIXME: we need to unify the clientid namespaces for nfsv4.x
859 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
860 * and SET_CLIENTID{,_CONFIRM}
863 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
865 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
866 return use_exchange_id == has_exchange_flags;
869 static struct nfs4_client *
870 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
871 bool use_exchange_id)
873 struct nfs4_client *clp;
875 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
876 if (same_name(clp->cl_recdir, dname) &&
877 match_clientid_establishment(clp, use_exchange_id))
883 static struct nfs4_client *
884 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
885 bool use_exchange_id)
887 struct nfs4_client *clp;
889 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
890 if (same_name(clp->cl_recdir, dname) &&
891 match_clientid_establishment(clp, use_exchange_id))
897 /* a helper function for parse_callback */
899 parse_octet(unsigned int *lenp, char **addrp)
901 unsigned int len = *lenp;
913 if ((c < '0') || (c > '9')) {
919 n = (n * 10) + (c - '0');
930 /* parse and set the setclientid ipv4 callback address */
932 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
937 u32 addrlen = addr_len;
938 char *addr = addr_val;
943 for(i = 4; i > 0 ; i--) {
944 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
947 cbaddr |= (temp << shift);
955 for(i = 2; i > 0 ; i--) {
956 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
959 cbport |= (temp << shift);
968 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
970 struct nfs4_callback *cb = &clp->cl_callback;
972 /* Currently, we only support tcp for the callback channel */
973 if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
976 if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
977 &cb->cb_addr, &cb->cb_port)))
979 cb->cb_prog = se->se_callback_prog;
980 cb->cb_ident = se->se_callback_ident;
983 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
984 "will not receive delegations\n",
985 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
991 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
993 struct nfsd4_compoundres *resp = rqstp->rq_resp;
995 resp->cstate.statp = statp;
999 * Dereference the result pages.
1002 nfsd4_release_respages(struct page **respages, short resused)
1006 dprintk("--> %s\n", __func__);
1007 for (i = 0; i < resused; i++) {
1010 put_page(respages[i]);
1016 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1020 for (i = 0; i < count; i++) {
1021 topages[i] = frompages[i];
1024 get_page(topages[i]);
1029 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1030 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1031 * length of the XDR response is less than se_fmaxresp_cached
1032 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1033 * of the reply (e.g. readdir).
1035 * Store the base and length of the rq_req.head[0] page
1036 * of the NFSv4.1 data, just past the rpc header.
1039 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1041 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1042 struct svc_rqst *rqstp = resp->rqstp;
1043 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1044 struct nfsd4_op *op = &args->ops[resp->opcnt];
1045 struct kvec *resv = &rqstp->rq_res.head[0];
1047 dprintk("--> %s entry %p\n", __func__, entry);
1049 /* Don't cache a failed OP_SEQUENCE. */
1050 if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1053 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1054 entry->ce_opcnt = resp->opcnt;
1055 entry->ce_status = resp->cstate.status;
1058 * Don't need a page to cache just the sequence operation - the slot
1062 if (nfsd4_not_cached(resp)) {
1063 entry->ce_resused = 0;
1064 entry->ce_rpchdrlen = 0;
1065 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1066 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1069 entry->ce_resused = rqstp->rq_resused;
1070 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1071 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1072 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1074 entry->ce_datav.iov_base = resp->cstate.statp;
1075 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1076 (char *)page_address(rqstp->rq_respages[0]));
1077 /* Current request rpc header length*/
1078 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1079 (char *)page_address(rqstp->rq_respages[0]);
1083 * We keep the rpc header, but take the nfs reply from the replycache.
1086 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1087 struct nfsd4_cache_entry *entry)
1089 struct svc_rqst *rqstp = resp->rqstp;
1090 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1093 /* Current request rpc header length*/
1094 len = (char *)resp->cstate.statp -
1095 (char *)page_address(rqstp->rq_respages[0]);
1096 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1097 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1098 entry->ce_datav.iov_len);
1101 /* copy the cached reply nfsd data past the current rpc header */
1102 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1103 entry->ce_datav.iov_len);
1104 resv->iov_len = len + entry->ce_datav.iov_len;
1109 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1110 * cached page. Replace any futher replay pages from the cache.
1113 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1114 struct nfsd4_sequence *seq)
1116 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1119 dprintk("--> %s entry %p\n", __func__, entry);
1122 * If this is just the sequence operation, we did not keep
1123 * a page in the cache entry because we can just use the
1124 * slot info stored in struct nfsd4_sequence that was checked
1125 * against the slot in nfsd4_sequence().
1127 * This occurs when seq->cachethis is FALSE, or when the client
1128 * session inactivity timer fires and a solo sequence operation
1129 * is sent (lease renewal).
1131 if (seq && nfsd4_not_cached(resp)) {
1132 seq->maxslots = resp->cstate.session->se_fnumslots;
1136 if (!nfsd41_copy_replay_data(resp, entry)) {
1138 * Not enough room to use the replay rpc header, send the
1139 * cached header. Release all the allocated result pages.
1141 svc_free_res_pages(resp->rqstp);
1142 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1145 /* Release all but the first allocated result page */
1147 resp->rqstp->rq_resused--;
1148 svc_free_res_pages(resp->rqstp);
1150 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1151 &entry->ce_respages[1],
1152 entry->ce_resused - 1);
1155 resp->rqstp->rq_resused = entry->ce_resused;
1156 resp->opcnt = entry->ce_opcnt;
1157 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1158 status = entry->ce_status;
1164 * Set the exchange_id flags returned by the server.
1167 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1169 /* pNFS is not supported */
1170 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1172 /* Referrals are supported, Migration is not. */
1173 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1175 /* set the wire flags to return to client. */
1176 clid->flags = new->cl_exchange_flags;
1180 nfsd4_exchange_id(struct svc_rqst *rqstp,
1181 struct nfsd4_compound_state *cstate,
1182 struct nfsd4_exchange_id *exid)
1184 struct nfs4_client *unconf, *conf, *new;
1186 unsigned int strhashval;
1187 char dname[HEXDIR_LEN];
1188 nfs4_verifier verf = exid->verifier;
1189 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1191 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1192 " ip_addr=%u flags %x, spa_how %d\n",
1193 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1194 ip_addr, exid->flags, exid->spa_how);
1196 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1197 return nfserr_inval;
1199 /* Currently only support SP4_NONE */
1200 switch (exid->spa_how) {
1204 return nfserr_encr_alg_unsupp;
1206 BUG(); /* checked by xdr code */
1208 return nfserr_serverfault; /* no excuse :-/ */
1211 status = nfs4_make_rec_clidname(dname, &exid->clname);
1216 strhashval = clientstr_hashval(dname);
1221 conf = find_confirmed_client_by_str(dname, strhashval, true);
1223 if (!same_verf(&verf, &conf->cl_verifier)) {
1224 /* 18.35.4 case 8 */
1225 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1226 status = nfserr_not_same;
1229 /* Client reboot: destroy old state */
1230 expire_client(conf);
1233 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1234 /* 18.35.4 case 9 */
1235 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1236 status = nfserr_perm;
1239 expire_client(conf);
1242 if (ip_addr != conf->cl_addr &&
1243 !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
1244 /* Client collision. 18.35.4 case 3 */
1245 status = nfserr_clid_inuse;
1249 * Set bit when the owner id and verifier map to an already
1250 * confirmed client id (18.35.3).
1252 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1255 * Falling into 18.35.4 case 2, possible router replay.
1256 * Leave confirmed record intact and return same result.
1258 copy_verf(conf, &verf);
1262 /* 18.35.4 case 7 */
1263 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1264 status = nfserr_noent;
1269 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1272 * Possible retry or client restart. Per 18.35.4 case 4,
1273 * a new unconfirmed record should be generated regardless
1274 * of whether any properties have changed.
1276 expire_client(unconf);
1281 new = create_client(exid->clname, dname);
1283 status = nfserr_resource;
1287 copy_verf(new, &verf);
1288 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1289 new->cl_addr = ip_addr;
1292 add_to_unconfirmed(new, strhashval);
1294 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1295 exid->clientid.cl_id = new->cl_clientid.cl_id;
1297 new->cl_slot.sl_seqid = 0;
1299 nfsd4_set_ex_flags(new, exid);
1301 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1302 new->cl_slot.sl_seqid, new->cl_exchange_flags);
1306 nfs4_unlock_state();
1308 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1313 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1315 dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1318 /* The slot is in use, and no response has been sent. */
1319 if (slot->sl_inuse) {
1320 if (seqid == slot->sl_seqid)
1321 return nfserr_jukebox;
1323 return nfserr_seq_misordered;
1326 if (likely(seqid == slot->sl_seqid + 1))
1329 if (seqid == slot->sl_seqid)
1330 return nfserr_replay_cache;
1332 if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1334 /* Misordered replay or misordered new request */
1335 return nfserr_seq_misordered;
1339 nfsd4_create_session(struct svc_rqst *rqstp,
1340 struct nfsd4_compound_state *cstate,
1341 struct nfsd4_create_session *cr_ses)
1343 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1344 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1345 struct nfs4_client *conf, *unconf;
1346 struct nfsd4_slot *slot = NULL;
1350 unconf = find_unconfirmed_client(&cr_ses->clientid);
1351 conf = find_confirmed_client(&cr_ses->clientid);
1354 slot = &conf->cl_slot;
1355 status = check_slot_seqid(cr_ses->seqid, slot);
1356 if (status == nfserr_replay_cache) {
1357 dprintk("Got a create_session replay! seqid= %d\n",
1359 cstate->slot = slot;
1360 cstate->status = status;
1361 /* Return the cached reply status */
1362 status = nfsd4_replay_cache_entry(resp, NULL);
1364 } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
1365 status = nfserr_seq_misordered;
1366 dprintk("Sequence misordered!\n");
1367 dprintk("Expected seqid= %d but got seqid= %d\n",
1368 slot->sl_seqid, cr_ses->seqid);
1371 conf->cl_slot.sl_seqid++;
1372 } else if (unconf) {
1373 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1374 (ip_addr != unconf->cl_addr)) {
1375 status = nfserr_clid_inuse;
1379 slot = &unconf->cl_slot;
1380 status = check_slot_seqid(cr_ses->seqid, slot);
1382 /* an unconfirmed replay returns misordered */
1383 status = nfserr_seq_misordered;
1387 slot->sl_seqid++; /* from 0 to 1 */
1388 move_to_confirmed(unconf);
1391 * We do not support RDMA or persistent sessions
1393 cr_ses->flags &= ~SESSION4_PERSIST;
1394 cr_ses->flags &= ~SESSION4_RDMA;
1398 status = nfserr_stale_clientid;
1402 status = alloc_init_session(rqstp, conf, cr_ses);
1406 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1407 NFS4_MAX_SESSIONID_LEN);
1408 cr_ses->seqid = slot->sl_seqid;
1410 slot->sl_inuse = true;
1411 cstate->slot = slot;
1412 /* Ensure a page is used for the cache */
1413 slot->sl_cache_entry.ce_cachethis = 1;
1415 nfs4_unlock_state();
1416 dprintk("%s returns %d\n", __func__, ntohl(status));
1421 nfsd4_destroy_session(struct svc_rqst *r,
1422 struct nfsd4_compound_state *cstate,
1423 struct nfsd4_destroy_session *sessionid)
1425 struct nfsd4_session *ses;
1426 u32 status = nfserr_badsession;
1429 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1430 * - Should we return nfserr_back_chan_busy if waiting for
1431 * callbacks on to-be-destroyed session?
1432 * - Do we need to clear any callback info from previous session?
1435 dump_sessionid(__func__, &sessionid->sessionid);
1436 spin_lock(&sessionid_lock);
1437 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1439 spin_unlock(&sessionid_lock);
1443 unhash_session(ses);
1444 spin_unlock(&sessionid_lock);
1446 /* wait for callbacks */
1447 shutdown_callback_client(ses->se_client);
1448 nfsd4_put_session(ses);
1451 dprintk("%s returns %d\n", __func__, ntohl(status));
1456 nfsd4_sequence(struct svc_rqst *rqstp,
1457 struct nfsd4_compound_state *cstate,
1458 struct nfsd4_sequence *seq)
1460 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1461 struct nfsd4_session *session;
1462 struct nfsd4_slot *slot;
1465 if (resp->opcnt != 1)
1466 return nfserr_sequence_pos;
1468 spin_lock(&sessionid_lock);
1469 status = nfserr_badsession;
1470 session = find_in_sessionid_hashtbl(&seq->sessionid);
1474 status = nfserr_badslot;
1475 if (seq->slotid >= session->se_fnumslots)
1478 slot = &session->se_slots[seq->slotid];
1479 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1481 status = check_slot_seqid(seq->seqid, slot);
1482 if (status == nfserr_replay_cache) {
1483 cstate->slot = slot;
1484 cstate->session = session;
1485 /* Return the cached reply status and set cstate->status
1486 * for nfsd4_svc_encode_compoundres processing */
1487 status = nfsd4_replay_cache_entry(resp, seq);
1488 cstate->status = nfserr_replay_cache;
1494 /* Success! bump slot seqid */
1495 slot->sl_inuse = true;
1496 slot->sl_seqid = seq->seqid;
1497 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1498 /* Always set the cache entry cachethis for solo sequence */
1499 if (nfsd4_is_solo_sequence(resp))
1500 slot->sl_cache_entry.ce_cachethis = 1;
1502 cstate->slot = slot;
1503 cstate->session = session;
1506 /* Renew the clientid on success and on replay.
1507 * Hold a session reference until done processing the compound:
1508 * nfsd4_put_session called only if the cstate slot is set.
1510 renew_client(session->se_client);
1511 nfsd4_get_session(session);
1513 spin_unlock(&sessionid_lock);
1514 dprintk("%s: return %d\n", __func__, ntohl(status));
1519 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1520 struct nfsd4_setclientid *setclid)
1522 struct sockaddr_in *sin = svc_addr_in(rqstp);
1523 struct xdr_netobj clname = {
1524 .len = setclid->se_namelen,
1525 .data = setclid->se_name,
1527 nfs4_verifier clverifier = setclid->se_verf;
1528 unsigned int strhashval;
1529 struct nfs4_client *conf, *unconf, *new;
1532 char dname[HEXDIR_LEN];
1534 if (!check_name(clname))
1535 return nfserr_inval;
1537 status = nfs4_make_rec_clidname(dname, &clname);
1542 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1543 * We get here on a DRC miss.
1546 strhashval = clientstr_hashval(dname);
1549 conf = find_confirmed_client_by_str(dname, strhashval, false);
1551 /* RFC 3530 14.2.33 CASE 0: */
1552 status = nfserr_clid_inuse;
1553 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1554 dprintk("NFSD: setclientid: string in use by client"
1555 " at %pI4\n", &conf->cl_addr);
1560 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1561 * has a description of SETCLIENTID request processing consisting
1562 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1564 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1565 status = nfserr_resource;
1568 * RFC 3530 14.2.33 CASE 4:
1569 * placed first, because it is the normal case
1572 expire_client(unconf);
1573 new = create_client(clname, dname);
1577 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1579 * RFC 3530 14.2.33 CASE 1:
1580 * probable callback update
1583 /* Note this is removing unconfirmed {*x***},
1584 * which is stronger than RFC recommended {vxc**}.
1585 * This has the advantage that there is at most
1586 * one {*x***} in either list at any time.
1588 expire_client(unconf);
1590 new = create_client(clname, dname);
1593 copy_clid(new, conf);
1594 } else if (!unconf) {
1596 * RFC 3530 14.2.33 CASE 2:
1597 * probable client reboot; state will be removed if
1600 new = create_client(clname, dname);
1606 * RFC 3530 14.2.33 CASE 3:
1607 * probable client reboot; state will be removed if
1610 expire_client(unconf);
1611 new = create_client(clname, dname);
1616 copy_verf(new, &clverifier);
1617 new->cl_addr = sin->sin_addr.s_addr;
1618 new->cl_flavor = rqstp->rq_flavor;
1619 princ = svc_gss_principal(rqstp);
1621 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1622 if (new->cl_principal == NULL) {
1627 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1629 gen_callback(new, setclid);
1630 add_to_unconfirmed(new, strhashval);
1631 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1632 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1633 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1636 nfs4_unlock_state();
1642 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1643 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1644 * bullets, labeled as CASE1 - CASE4 below.
1647 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1648 struct nfsd4_compound_state *cstate,
1649 struct nfsd4_setclientid_confirm *setclientid_confirm)
1651 struct sockaddr_in *sin = svc_addr_in(rqstp);
1652 struct nfs4_client *conf, *unconf;
1653 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1654 clientid_t * clid = &setclientid_confirm->sc_clientid;
1657 if (STALE_CLIENTID(clid))
1658 return nfserr_stale_clientid;
1660 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1661 * We get here on a DRC miss.
1666 conf = find_confirmed_client(clid);
1667 unconf = find_unconfirmed_client(clid);
1669 status = nfserr_clid_inuse;
1670 if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1672 if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1676 * section 14.2.34 of RFC 3530 has a description of
1677 * SETCLIENTID_CONFIRM request processing consisting
1678 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1680 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1682 * RFC 3530 14.2.34 CASE 1:
1685 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1686 status = nfserr_clid_inuse;
1688 /* XXX: We just turn off callbacks until we can handle
1689 * change request correctly. */
1690 atomic_set(&conf->cl_callback.cb_set, 0);
1691 expire_client(unconf);
1695 } else if (conf && !unconf) {
1697 * RFC 3530 14.2.34 CASE 2:
1698 * probable retransmitted request; play it safe and
1701 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1702 status = nfserr_clid_inuse;
1705 } else if (!conf && unconf
1706 && same_verf(&unconf->cl_confirm, &confirm)) {
1708 * RFC 3530 14.2.34 CASE 3:
1709 * Normal case; new or rebooted client:
1711 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1712 status = nfserr_clid_inuse;
1715 clientstr_hashval(unconf->cl_recdir);
1716 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1719 nfsd4_remove_clid_dir(conf);
1720 expire_client(conf);
1722 move_to_confirmed(unconf);
1724 nfsd4_probe_callback(conf);
1727 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1728 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1731 * RFC 3530 14.2.34 CASE 4:
1732 * Client probably hasn't noticed that we rebooted yet.
1734 status = nfserr_stale_clientid;
1736 /* check that we have hit one of the cases...*/
1737 status = nfserr_clid_inuse;
1740 nfs4_unlock_state();
1744 /* OPEN Share state helper functions */
1745 static inline struct nfs4_file *
1746 alloc_init_file(struct inode *ino)
1748 struct nfs4_file *fp;
1749 unsigned int hashval = file_hashval(ino);
1751 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1753 atomic_set(&fp->fi_ref, 1);
1754 INIT_LIST_HEAD(&fp->fi_hash);
1755 INIT_LIST_HEAD(&fp->fi_stateids);
1756 INIT_LIST_HEAD(&fp->fi_delegations);
1757 spin_lock(&recall_lock);
1758 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1759 spin_unlock(&recall_lock);
1760 fp->fi_inode = igrab(ino);
1761 fp->fi_id = current_fileid++;
1762 fp->fi_had_conflict = false;
1769 nfsd4_free_slab(struct kmem_cache **slab)
1773 kmem_cache_destroy(*slab);
1778 nfsd4_free_slabs(void)
1780 nfsd4_free_slab(&stateowner_slab);
1781 nfsd4_free_slab(&file_slab);
1782 nfsd4_free_slab(&stateid_slab);
1783 nfsd4_free_slab(&deleg_slab);
1787 nfsd4_init_slabs(void)
1789 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1790 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1791 if (stateowner_slab == NULL)
1793 file_slab = kmem_cache_create("nfsd4_files",
1794 sizeof(struct nfs4_file), 0, 0, NULL);
1795 if (file_slab == NULL)
1797 stateid_slab = kmem_cache_create("nfsd4_stateids",
1798 sizeof(struct nfs4_stateid), 0, 0, NULL);
1799 if (stateid_slab == NULL)
1801 deleg_slab = kmem_cache_create("nfsd4_delegations",
1802 sizeof(struct nfs4_delegation), 0, 0, NULL);
1803 if (deleg_slab == NULL)
1808 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1813 nfs4_free_stateowner(struct kref *kref)
1815 struct nfs4_stateowner *sop =
1816 container_of(kref, struct nfs4_stateowner, so_ref);
1817 kfree(sop->so_owner.data);
1818 kmem_cache_free(stateowner_slab, sop);
1821 static inline struct nfs4_stateowner *
1822 alloc_stateowner(struct xdr_netobj *owner)
1824 struct nfs4_stateowner *sop;
1826 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1827 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1828 memcpy(sop->so_owner.data, owner->data, owner->len);
1829 sop->so_owner.len = owner->len;
1830 kref_init(&sop->so_ref);
1833 kmem_cache_free(stateowner_slab, sop);
1838 static struct nfs4_stateowner *
1839 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1840 struct nfs4_stateowner *sop;
1841 struct nfs4_replay *rp;
1842 unsigned int idhashval;
1844 if (!(sop = alloc_stateowner(&open->op_owner)))
1846 idhashval = ownerid_hashval(current_ownerid);
1847 INIT_LIST_HEAD(&sop->so_idhash);
1848 INIT_LIST_HEAD(&sop->so_strhash);
1849 INIT_LIST_HEAD(&sop->so_perclient);
1850 INIT_LIST_HEAD(&sop->so_stateids);
1851 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1852 INIT_LIST_HEAD(&sop->so_close_lru);
1854 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1855 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1856 list_add(&sop->so_perclient, &clp->cl_openowners);
1857 sop->so_is_open_owner = 1;
1858 sop->so_id = current_ownerid++;
1859 sop->so_client = clp;
1860 sop->so_seqid = open->op_seqid;
1861 sop->so_confirmed = 0;
1862 rp = &sop->so_replay;
1863 rp->rp_status = nfserr_serverfault;
1865 rp->rp_buf = rp->rp_ibuf;
1870 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1871 struct nfs4_stateowner *sop = open->op_stateowner;
1872 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1874 INIT_LIST_HEAD(&stp->st_hash);
1875 INIT_LIST_HEAD(&stp->st_perstateowner);
1876 INIT_LIST_HEAD(&stp->st_lockowners);
1877 INIT_LIST_HEAD(&stp->st_perfile);
1878 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1879 list_add(&stp->st_perstateowner, &sop->so_stateids);
1880 list_add(&stp->st_perfile, &fp->fi_stateids);
1881 stp->st_stateowner = sop;
1884 stp->st_stateid.si_boot = get_seconds();
1885 stp->st_stateid.si_stateownerid = sop->so_id;
1886 stp->st_stateid.si_fileid = fp->fi_id;
1887 stp->st_stateid.si_generation = 0;
1888 stp->st_access_bmap = 0;
1889 stp->st_deny_bmap = 0;
1890 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1891 &stp->st_access_bmap);
1892 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1893 stp->st_openstp = NULL;
1897 move_to_close_lru(struct nfs4_stateowner *sop)
1899 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1901 list_move_tail(&sop->so_close_lru, &close_lru);
1902 sop->so_time = get_seconds();
1906 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1909 return (sop->so_owner.len == owner->len) &&
1910 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1911 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1914 static struct nfs4_stateowner *
1915 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1917 struct nfs4_stateowner *so = NULL;
1919 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1920 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1926 /* search file_hashtbl[] for file */
1927 static struct nfs4_file *
1928 find_file(struct inode *ino)
1930 unsigned int hashval = file_hashval(ino);
1931 struct nfs4_file *fp;
1933 spin_lock(&recall_lock);
1934 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1935 if (fp->fi_inode == ino) {
1937 spin_unlock(&recall_lock);
1941 spin_unlock(&recall_lock);
1945 static inline int access_valid(u32 x, u32 minorversion)
1947 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1949 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1951 x &= ~NFS4_SHARE_ACCESS_MASK;
1952 if (minorversion && x) {
1953 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1955 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1957 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1964 static inline int deny_valid(u32 x)
1966 /* Note: unlike access bits, deny bits may be zero. */
1967 return x <= NFS4_SHARE_DENY_BOTH;
1971 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1972 * st_{access,deny}_bmap field of the stateid, in order to track not
1973 * only what share bits are currently in force, but also what
1974 * combinations of share bits previous opens have used. This allows us
1975 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1976 * return an error if the client attempt to downgrade to a combination
1977 * of share bits not explicable by closing some of its previous opens.
1979 * XXX: This enforcement is actually incomplete, since we don't keep
1980 * track of access/deny bit combinations; so, e.g., we allow:
1982 * OPEN allow read, deny write
1983 * OPEN allow both, deny none
1984 * DOWNGRADE allow read, deny none
1986 * which we should reject.
1989 set_access(unsigned int *access, unsigned long bmap) {
1993 for (i = 1; i < 4; i++) {
1994 if (test_bit(i, &bmap))
2000 set_deny(unsigned int *deny, unsigned long bmap) {
2004 for (i = 0; i < 4; i++) {
2005 if (test_bit(i, &bmap))
2011 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2012 unsigned int access, deny;
2014 set_access(&access, stp->st_access_bmap);
2015 set_deny(&deny, stp->st_deny_bmap);
2016 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2022 * Called to check deny when READ with all zero stateid or
2023 * WRITE with all zero or all one stateid
2026 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2028 struct inode *ino = current_fh->fh_dentry->d_inode;
2029 struct nfs4_file *fp;
2030 struct nfs4_stateid *stp;
2033 dprintk("NFSD: nfs4_share_conflict\n");
2035 fp = find_file(ino);
2038 ret = nfserr_locked;
2039 /* Search for conflicting share reservations */
2040 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2041 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2042 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2052 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2054 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2055 drop_file_write_access(filp);
2056 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2061 * Recall a delegation
2064 do_recall(void *__dp)
2066 struct nfs4_delegation *dp = __dp;
2068 dp->dl_file->fi_had_conflict = true;
2069 nfsd4_cb_recall(dp);
2074 * Spawn a thread to perform a recall on the delegation represented
2075 * by the lease (file_lock)
2077 * Called from break_lease() with lock_kernel() held.
2078 * Note: we assume break_lease will only call this *once* for any given
2082 void nfsd_break_deleg_cb(struct file_lock *fl)
2084 struct nfs4_delegation *dp= (struct nfs4_delegation *)fl->fl_owner;
2085 struct task_struct *t;
2087 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2091 /* We're assuming the state code never drops its reference
2092 * without first removing the lease. Since we're in this lease
2093 * callback (and since the lease code is serialized by the kernel
2094 * lock) we know the server hasn't removed the lease yet, we know
2095 * it's safe to take a reference: */
2096 atomic_inc(&dp->dl_count);
2097 atomic_inc(&dp->dl_client->cl_count);
2099 spin_lock(&recall_lock);
2100 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2101 spin_unlock(&recall_lock);
2103 /* only place dl_time is set. protected by lock_kernel*/
2104 dp->dl_time = get_seconds();
2107 * We don't want the locks code to timeout the lease for us;
2108 * we'll remove it ourself if the delegation isn't returned
2111 fl->fl_break_time = 0;
2113 t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
2115 struct nfs4_client *clp = dp->dl_client;
2117 printk(KERN_INFO "NFSD: Callback thread failed for "
2118 "for client (clientid %08x/%08x)\n",
2119 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2120 put_nfs4_client(dp->dl_client);
2121 nfs4_put_delegation(dp);
2126 * The file_lock is being reapd.
2128 * Called by locks_free_lock() with lock_kernel() held.
2131 void nfsd_release_deleg_cb(struct file_lock *fl)
2133 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2135 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2137 if (!(fl->fl_flags & FL_LEASE) || !dp)
2139 dp->dl_flock = NULL;
2143 * Set the delegation file_lock back pointer.
2145 * Called from setlease() with lock_kernel() held.
2148 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2150 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2152 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2159 * Called from setlease() with lock_kernel() held
2162 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2164 struct nfs4_delegation *onlistd =
2165 (struct nfs4_delegation *)onlist->fl_owner;
2166 struct nfs4_delegation *tryd =
2167 (struct nfs4_delegation *)try->fl_owner;
2169 if (onlist->fl_lmops != try->fl_lmops)
2172 return onlistd->dl_client == tryd->dl_client;
2177 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2180 return lease_modify(onlist, arg);
2185 static struct lock_manager_operations nfsd_lease_mng_ops = {
2186 .fl_break = nfsd_break_deleg_cb,
2187 .fl_release_private = nfsd_release_deleg_cb,
2188 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2189 .fl_mylease = nfsd_same_client_deleg_cb,
2190 .fl_change = nfsd_change_deleg_cb,
2195 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2196 struct nfsd4_open *open)
2198 clientid_t *clientid = &open->op_clientid;
2199 struct nfs4_client *clp = NULL;
2200 unsigned int strhashval;
2201 struct nfs4_stateowner *sop = NULL;
2203 if (!check_name(open->op_owner))
2204 return nfserr_inval;
2206 if (STALE_CLIENTID(&open->op_clientid))
2207 return nfserr_stale_clientid;
2209 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2210 sop = find_openstateowner_str(strhashval, open);
2211 open->op_stateowner = sop;
2213 /* Make sure the client's lease hasn't expired. */
2214 clp = find_confirmed_client(clientid);
2216 return nfserr_expired;
2219 /* When sessions are used, skip open sequenceid processing */
2220 if (nfsd4_has_session(cstate))
2222 if (!sop->so_confirmed) {
2223 /* Replace unconfirmed owners without checking for replay. */
2224 clp = sop->so_client;
2225 release_openowner(sop);
2226 open->op_stateowner = NULL;
2229 if (open->op_seqid == sop->so_seqid - 1) {
2230 if (sop->so_replay.rp_buflen)
2231 return nfserr_replay_me;
2232 /* The original OPEN failed so spectacularly
2233 * that we don't even have replay data saved!
2234 * Therefore, we have no choice but to continue
2235 * processing this OPEN; presumably, we'll
2236 * fail again for the same reason.
2238 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2241 if (open->op_seqid != sop->so_seqid)
2242 return nfserr_bad_seqid;
2244 if (open->op_stateowner == NULL) {
2245 sop = alloc_init_open_stateowner(strhashval, clp, open);
2247 return nfserr_resource;
2248 open->op_stateowner = sop;
2250 list_del_init(&sop->so_close_lru);
2251 renew_client(sop->so_client);
2255 static inline __be32
2256 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2258 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2259 return nfserr_openmode;
2264 static struct nfs4_delegation *
2265 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2267 struct nfs4_delegation *dp;
2269 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2270 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2277 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2278 struct nfs4_delegation **dp)
2281 __be32 status = nfserr_bad_stateid;
2283 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2286 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2287 RD_STATE : WR_STATE;
2288 status = nfs4_check_delegmode(*dp, flags);
2292 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2296 open->op_stateowner->so_confirmed = 1;
2301 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2303 struct nfs4_stateid *local;
2304 __be32 status = nfserr_share_denied;
2305 struct nfs4_stateowner *sop = open->op_stateowner;
2307 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2308 /* ignore lock owners */
2309 if (local->st_stateowner->so_is_open_owner == 0)
2311 /* remember if we have seen this open owner */
2312 if (local->st_stateowner == sop)
2314 /* check for conflicting share reservations */
2315 if (!test_share(local, open))
2323 static inline struct nfs4_stateid *
2324 nfs4_alloc_stateid(void)
2326 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2330 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2331 struct nfs4_delegation *dp,
2332 struct svc_fh *cur_fh, int flags)
2334 struct nfs4_stateid *stp;
2336 stp = nfs4_alloc_stateid();
2338 return nfserr_resource;
2341 get_file(dp->dl_vfs_file);
2342 stp->st_vfs_file = dp->dl_vfs_file;
2345 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2348 if (status == nfserr_dropit)
2349 status = nfserr_jukebox;
2350 kmem_cache_free(stateid_slab, stp);
2358 static inline __be32
2359 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2360 struct nfsd4_open *open)
2362 struct iattr iattr = {
2363 .ia_valid = ATTR_SIZE,
2366 if (!open->op_truncate)
2368 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2369 return nfserr_inval;
2370 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2374 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2376 struct file *filp = stp->st_vfs_file;
2377 struct inode *inode = filp->f_path.dentry->d_inode;
2378 unsigned int share_access, new_writer;
2381 set_access(&share_access, stp->st_access_bmap);
2382 new_writer = (~share_access) & open->op_share_access
2383 & NFS4_SHARE_ACCESS_WRITE;
2386 int err = get_write_access(inode);
2388 return nfserrno(err);
2389 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2391 return nfserrno(err);
2392 file_take_write(filp);
2394 status = nfsd4_truncate(rqstp, cur_fh, open);
2397 put_write_access(inode);
2400 /* remember the open */
2401 filp->f_mode |= open->op_share_access;
2402 __set_bit(open->op_share_access, &stp->st_access_bmap);
2403 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2410 nfs4_set_claim_prev(struct nfsd4_open *open)
2412 open->op_stateowner->so_confirmed = 1;
2413 open->op_stateowner->so_client->cl_firststate = 1;
2417 * Attempt to hand out a delegation.
2420 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2422 struct nfs4_delegation *dp;
2423 struct nfs4_stateowner *sop = stp->st_stateowner;
2424 struct nfs4_callback *cb = &sop->so_client->cl_callback;
2425 struct file_lock fl, *flp = &fl;
2426 int status, flag = 0;
2428 flag = NFS4_OPEN_DELEGATE_NONE;
2429 open->op_recall = 0;
2430 switch (open->op_claim_type) {
2431 case NFS4_OPEN_CLAIM_PREVIOUS:
2432 if (!atomic_read(&cb->cb_set))
2433 open->op_recall = 1;
2434 flag = open->op_delegate_type;
2435 if (flag == NFS4_OPEN_DELEGATE_NONE)
2438 case NFS4_OPEN_CLAIM_NULL:
2439 /* Let's not give out any delegations till everyone's
2440 * had the chance to reclaim theirs.... */
2441 if (locks_in_grace())
2443 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2445 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2446 flag = NFS4_OPEN_DELEGATE_WRITE;
2448 flag = NFS4_OPEN_DELEGATE_READ;
2454 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2456 flag = NFS4_OPEN_DELEGATE_NONE;
2459 locks_init_lock(&fl);
2460 fl.fl_lmops = &nfsd_lease_mng_ops;
2461 fl.fl_flags = FL_LEASE;
2462 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2463 fl.fl_end = OFFSET_MAX;
2464 fl.fl_owner = (fl_owner_t)dp;
2465 fl.fl_file = stp->st_vfs_file;
2466 fl.fl_pid = current->tgid;
2468 /* vfs_setlease checks to see if delegation should be handed out.
2469 * the lock_manager callbacks fl_mylease and fl_change are used
2471 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2472 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2473 unhash_delegation(dp);
2474 flag = NFS4_OPEN_DELEGATE_NONE;
2478 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2480 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2481 dp->dl_stateid.si_boot,
2482 dp->dl_stateid.si_stateownerid,
2483 dp->dl_stateid.si_fileid,
2484 dp->dl_stateid.si_generation);
2486 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2487 && flag == NFS4_OPEN_DELEGATE_NONE
2488 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2489 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2490 open->op_delegate_type = flag;
2494 * called with nfs4_lock_state() held.
2497 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2499 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2500 struct nfs4_file *fp = NULL;
2501 struct inode *ino = current_fh->fh_dentry->d_inode;
2502 struct nfs4_stateid *stp = NULL;
2503 struct nfs4_delegation *dp = NULL;
2506 status = nfserr_inval;
2507 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2508 || !deny_valid(open->op_share_deny))
2511 * Lookup file; if found, lookup stateid and check open request,
2512 * and check for delegations in the process of being recalled.
2513 * If not found, create the nfs4_file struct
2515 fp = find_file(ino);
2517 if ((status = nfs4_check_open(fp, open, &stp)))
2519 status = nfs4_check_deleg(fp, open, &dp);
2523 status = nfserr_bad_stateid;
2524 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2526 status = nfserr_resource;
2527 fp = alloc_init_file(ino);
2533 * OPEN the file, or upgrade an existing OPEN.
2534 * If truncate fails, the OPEN fails.
2537 /* Stateid was found, this is an OPEN upgrade */
2538 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2541 update_stateid(&stp->st_stateid);
2543 /* Stateid was not found, this is a new OPEN */
2545 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2546 flags |= NFSD_MAY_READ;
2547 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2548 flags |= NFSD_MAY_WRITE;
2549 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2552 init_stateid(stp, fp, open);
2553 status = nfsd4_truncate(rqstp, current_fh, open);
2555 release_open_stateid(stp);
2558 if (nfsd4_has_session(&resp->cstate))
2559 update_stateid(&stp->st_stateid);
2561 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2563 if (nfsd4_has_session(&resp->cstate))
2564 open->op_stateowner->so_confirmed = 1;
2567 * Attempt to hand out a delegation. No error return, because the
2568 * OPEN succeeds even if we fail.
2570 nfs4_open_delegation(current_fh, open, stp);
2574 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2575 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2576 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2580 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2581 nfs4_set_claim_prev(open);
2583 * To finish the open response, we just need to set the rflags.
2585 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2586 if (!open->op_stateowner->so_confirmed &&
2587 !nfsd4_has_session(&resp->cstate))
2588 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2594 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2597 struct nfs4_client *clp;
2601 dprintk("process_renew(%08x/%08x): starting\n",
2602 clid->cl_boot, clid->cl_id);
2603 status = nfserr_stale_clientid;
2604 if (STALE_CLIENTID(clid))
2606 clp = find_confirmed_client(clid);
2607 status = nfserr_expired;
2609 /* We assume the client took too long to RENEW. */
2610 dprintk("nfsd4_renew: clientid not found!\n");
2614 status = nfserr_cb_path_down;
2615 if (!list_empty(&clp->cl_delegations)
2616 && !atomic_read(&clp->cl_callback.cb_set))
2620 nfs4_unlock_state();
2624 struct lock_manager nfsd4_manager = {
2628 nfsd4_end_grace(void)
2630 dprintk("NFSD: end of grace period\n");
2631 nfsd4_recdir_purge_old();
2632 locks_end_grace(&nfsd4_manager);
2636 nfs4_laundromat(void)
2638 struct nfs4_client *clp;
2639 struct nfs4_stateowner *sop;
2640 struct nfs4_delegation *dp;
2641 struct list_head *pos, *next, reaplist;
2642 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2643 time_t t, clientid_val = NFSD_LEASE_TIME;
2644 time_t u, test_val = NFSD_LEASE_TIME;
2648 dprintk("NFSD: laundromat service - starting\n");
2649 if (locks_in_grace())
2651 list_for_each_safe(pos, next, &client_lru) {
2652 clp = list_entry(pos, struct nfs4_client, cl_lru);
2653 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2654 t = clp->cl_time - cutoff;
2655 if (clientid_val > t)
2659 dprintk("NFSD: purging unused client (clientid %08x)\n",
2660 clp->cl_clientid.cl_id);
2661 nfsd4_remove_clid_dir(clp);
2664 INIT_LIST_HEAD(&reaplist);
2665 spin_lock(&recall_lock);
2666 list_for_each_safe(pos, next, &del_recall_lru) {
2667 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2668 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2669 u = dp->dl_time - cutoff;
2674 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2676 list_move(&dp->dl_recall_lru, &reaplist);
2678 spin_unlock(&recall_lock);
2679 list_for_each_safe(pos, next, &reaplist) {
2680 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2681 list_del_init(&dp->dl_recall_lru);
2682 unhash_delegation(dp);
2684 test_val = NFSD_LEASE_TIME;
2685 list_for_each_safe(pos, next, &close_lru) {
2686 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2687 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2688 u = sop->so_time - cutoff;
2693 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2695 release_openowner(sop);
2697 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2698 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2699 nfs4_unlock_state();
2700 return clientid_val;
2703 static struct workqueue_struct *laundry_wq;
2704 static void laundromat_main(struct work_struct *);
2705 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2708 laundromat_main(struct work_struct *not_used)
2712 t = nfs4_laundromat();
2713 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2714 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2717 static struct nfs4_stateowner *
2718 search_close_lru(u32 st_id, int flags)
2720 struct nfs4_stateowner *local = NULL;
2722 if (flags & CLOSE_STATE) {
2723 list_for_each_entry(local, &close_lru, so_close_lru) {
2724 if (local->so_id == st_id)
2732 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2734 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2738 STALE_STATEID(stateid_t *stateid)
2740 if (time_after((unsigned long)boot_time,
2741 (unsigned long)stateid->si_boot)) {
2742 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2743 stateid->si_boot, stateid->si_stateownerid,
2744 stateid->si_fileid, stateid->si_generation);
2751 EXPIRED_STATEID(stateid_t *stateid)
2753 if (time_before((unsigned long)boot_time,
2754 ((unsigned long)stateid->si_boot)) &&
2755 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2756 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2757 stateid->si_boot, stateid->si_stateownerid,
2758 stateid->si_fileid, stateid->si_generation);
2765 stateid_error_map(stateid_t *stateid)
2767 if (STALE_STATEID(stateid))
2768 return nfserr_stale_stateid;
2769 if (EXPIRED_STATEID(stateid))
2770 return nfserr_expired;
2772 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2773 stateid->si_boot, stateid->si_stateownerid,
2774 stateid->si_fileid, stateid->si_generation);
2775 return nfserr_bad_stateid;
2779 access_permit_read(unsigned long access_bmap)
2781 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2782 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2783 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2787 access_permit_write(unsigned long access_bmap)
2789 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2790 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2794 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2796 __be32 status = nfserr_openmode;
2798 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2800 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2807 static inline __be32
2808 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2810 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2812 else if (locks_in_grace()) {
2813 /* Answer in remaining cases depends on existance of
2814 * conflicting state; so we must wait out the grace period. */
2815 return nfserr_grace;
2816 } else if (flags & WR_STATE)
2817 return nfs4_share_conflict(current_fh,
2818 NFS4_SHARE_DENY_WRITE);
2819 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2820 return nfs4_share_conflict(current_fh,
2821 NFS4_SHARE_DENY_READ);
2825 * Allow READ/WRITE during grace period on recovered state only for files
2826 * that are not able to provide mandatory locking.
2829 grace_disallows_io(struct inode *inode)
2831 return locks_in_grace() && mandatory_lock(inode);
2834 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2837 * When sessions are used the stateid generation number is ignored
2840 if ((flags & HAS_SESSION) && in->si_generation == 0)
2843 /* If the client sends us a stateid from the future, it's buggy: */
2844 if (in->si_generation > ref->si_generation)
2845 return nfserr_bad_stateid;
2847 * The following, however, can happen. For example, if the
2848 * client sends an open and some IO at the same time, the open
2849 * may bump si_generation while the IO is still in flight.
2850 * Thanks to hard links and renames, the client never knows what
2851 * file an open will affect. So it could avoid that situation
2852 * only by serializing all opens and IO from the same open
2853 * owner. To recover from the old_stateid error, the client
2854 * will just have to retry the IO:
2856 if (in->si_generation < ref->si_generation)
2857 return nfserr_old_stateid;
2862 static int is_delegation_stateid(stateid_t *stateid)
2864 return stateid->si_fileid == 0;
2868 * Checks for stateid operations
2871 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2872 stateid_t *stateid, int flags, struct file **filpp)
2874 struct nfs4_stateid *stp = NULL;
2875 struct nfs4_delegation *dp = NULL;
2876 struct svc_fh *current_fh = &cstate->current_fh;
2877 struct inode *ino = current_fh->fh_dentry->d_inode;
2883 if (grace_disallows_io(ino))
2884 return nfserr_grace;
2886 if (nfsd4_has_session(cstate))
2887 flags |= HAS_SESSION;
2889 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2890 return check_special_stateids(current_fh, stateid, flags);
2892 status = nfserr_stale_stateid;
2893 if (STALE_STATEID(stateid))
2896 status = nfserr_bad_stateid;
2897 if (is_delegation_stateid(stateid)) {
2898 dp = find_delegation_stateid(ino, stateid);
2900 status = stateid_error_map(stateid);
2903 status = check_stateid_generation(stateid, &dp->dl_stateid,
2907 status = nfs4_check_delegmode(dp, flags);
2910 renew_client(dp->dl_client);
2912 *filpp = dp->dl_vfs_file;
2913 } else { /* open or lock stateid */
2914 stp = find_stateid(stateid, flags);
2916 status = stateid_error_map(stateid);
2919 if (nfs4_check_fh(current_fh, stp))
2921 if (!stp->st_stateowner->so_confirmed)
2923 status = check_stateid_generation(stateid, &stp->st_stateid,
2927 status = nfs4_check_openmode(stp, flags);
2930 renew_client(stp->st_stateowner->so_client);
2932 *filpp = stp->st_vfs_file;
2942 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2943 RD_STATE : WR_STATE;
2947 * Checks for sequence id mutating operations.
2950 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2951 stateid_t *stateid, int flags,
2952 struct nfs4_stateowner **sopp,
2953 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2955 struct nfs4_stateid *stp;
2956 struct nfs4_stateowner *sop;
2957 struct svc_fh *current_fh = &cstate->current_fh;
2960 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2961 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2962 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2963 stateid->si_generation);
2968 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2969 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2970 return nfserr_bad_stateid;
2973 if (STALE_STATEID(stateid))
2974 return nfserr_stale_stateid;
2976 if (nfsd4_has_session(cstate))
2977 flags |= HAS_SESSION;
2980 * We return BAD_STATEID if filehandle doesn't match stateid,
2981 * the confirmed flag is incorrecly set, or the generation
2982 * number is incorrect.
2984 stp = find_stateid(stateid, flags);
2987 * Also, we should make sure this isn't just the result of
2990 sop = search_close_lru(stateid->si_stateownerid, flags);
2992 return stateid_error_map(stateid);
2998 *sopp = sop = stp->st_stateowner;
3001 clientid_t *lockclid = &lock->v.new.clientid;
3002 struct nfs4_client *clp = sop->so_client;
3006 lkflg = setlkflg(lock->lk_type);
3008 if (lock->lk_is_new) {
3009 if (!sop->so_is_open_owner)
3010 return nfserr_bad_stateid;
3011 if (!(flags & HAS_SESSION) &&
3012 !same_clid(&clp->cl_clientid, lockclid))
3013 return nfserr_bad_stateid;
3014 /* stp is the open stateid */
3015 status = nfs4_check_openmode(stp, lkflg);
3019 /* stp is the lock stateid */
3020 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3026 if (nfs4_check_fh(current_fh, stp)) {
3027 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3028 return nfserr_bad_stateid;
3032 * We now validate the seqid and stateid generation numbers.
3033 * For the moment, we ignore the possibility of
3034 * generation number wraparound.
3036 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3039 if (sop->so_confirmed && flags & CONFIRM) {
3040 dprintk("NFSD: preprocess_seqid_op: expected"
3041 " unconfirmed stateowner!\n");
3042 return nfserr_bad_stateid;
3044 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3045 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3046 " confirmed yet!\n");
3047 return nfserr_bad_stateid;
3049 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3052 renew_client(sop->so_client);
3056 if (seqid == sop->so_seqid - 1) {
3057 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3058 /* indicate replay to calling function */
3059 return nfserr_replay_me;
3061 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3062 sop->so_seqid, seqid);
3064 return nfserr_bad_seqid;
3068 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3069 struct nfsd4_open_confirm *oc)
3072 struct nfs4_stateowner *sop;
3073 struct nfs4_stateid *stp;
3075 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3076 (int)cstate->current_fh.fh_dentry->d_name.len,
3077 cstate->current_fh.fh_dentry->d_name.name);
3079 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3085 if ((status = nfs4_preprocess_seqid_op(cstate,
3086 oc->oc_seqid, &oc->oc_req_stateid,
3087 CONFIRM | OPEN_STATE,
3088 &oc->oc_stateowner, &stp, NULL)))
3091 sop = oc->oc_stateowner;
3092 sop->so_confirmed = 1;
3093 update_stateid(&stp->st_stateid);
3094 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3095 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3096 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3097 stp->st_stateid.si_boot,
3098 stp->st_stateid.si_stateownerid,
3099 stp->st_stateid.si_fileid,
3100 stp->st_stateid.si_generation);
3102 nfsd4_create_clid_dir(sop->so_client);
3104 if (oc->oc_stateowner) {
3105 nfs4_get_stateowner(oc->oc_stateowner);
3106 cstate->replay_owner = oc->oc_stateowner;
3108 nfs4_unlock_state();
3114 * unset all bits in union bitmap (bmap) that
3115 * do not exist in share (from successful OPEN_DOWNGRADE)
3118 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3121 for (i = 1; i < 4; i++) {
3122 if ((i & access) != i)
3123 __clear_bit(i, bmap);
3128 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3131 for (i = 0; i < 4; i++) {
3132 if ((i & deny) != i)
3133 __clear_bit(i, bmap);
3138 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3139 struct nfsd4_compound_state *cstate,
3140 struct nfsd4_open_downgrade *od)
3143 struct nfs4_stateid *stp;
3144 unsigned int share_access;
3146 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3147 (int)cstate->current_fh.fh_dentry->d_name.len,
3148 cstate->current_fh.fh_dentry->d_name.name);
3150 if (!access_valid(od->od_share_access, cstate->minorversion)
3151 || !deny_valid(od->od_share_deny))
3152 return nfserr_inval;
3155 if ((status = nfs4_preprocess_seqid_op(cstate,
3159 &od->od_stateowner, &stp, NULL)))
3162 status = nfserr_inval;
3163 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3164 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3165 stp->st_access_bmap, od->od_share_access);
3168 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3169 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3170 stp->st_deny_bmap, od->od_share_deny);
3173 set_access(&share_access, stp->st_access_bmap);
3174 nfs4_file_downgrade(stp->st_vfs_file,
3175 share_access & ~od->od_share_access);
3177 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3178 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3180 update_stateid(&stp->st_stateid);
3181 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3184 if (od->od_stateowner) {
3185 nfs4_get_stateowner(od->od_stateowner);
3186 cstate->replay_owner = od->od_stateowner;
3188 nfs4_unlock_state();
3193 * nfs4_unlock_state() called after encode
3196 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3197 struct nfsd4_close *close)
3200 struct nfs4_stateid *stp;
3202 dprintk("NFSD: nfsd4_close on file %.*s\n",
3203 (int)cstate->current_fh.fh_dentry->d_name.len,
3204 cstate->current_fh.fh_dentry->d_name.name);
3207 /* check close_lru for replay */
3208 if ((status = nfs4_preprocess_seqid_op(cstate,
3211 OPEN_STATE | CLOSE_STATE,
3212 &close->cl_stateowner, &stp, NULL)))
3215 update_stateid(&stp->st_stateid);
3216 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3218 /* release_stateid() calls nfsd_close() if needed */
3219 release_open_stateid(stp);
3221 /* place unused nfs4_stateowners on so_close_lru list to be
3222 * released by the laundromat service after the lease period
3223 * to enable us to handle CLOSE replay
3225 if (list_empty(&close->cl_stateowner->so_stateids))
3226 move_to_close_lru(close->cl_stateowner);
3228 if (close->cl_stateowner) {
3229 nfs4_get_stateowner(close->cl_stateowner);
3230 cstate->replay_owner = close->cl_stateowner;
3232 nfs4_unlock_state();
3237 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3238 struct nfsd4_delegreturn *dr)
3240 struct nfs4_delegation *dp;
3241 stateid_t *stateid = &dr->dr_stateid;
3242 struct inode *inode;
3246 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3248 inode = cstate->current_fh.fh_dentry->d_inode;
3250 if (nfsd4_has_session(cstate))
3251 flags |= HAS_SESSION;
3253 status = nfserr_bad_stateid;
3254 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3256 status = nfserr_stale_stateid;
3257 if (STALE_STATEID(stateid))
3259 status = nfserr_bad_stateid;
3260 if (!is_delegation_stateid(stateid))
3262 dp = find_delegation_stateid(inode, stateid);
3264 status = stateid_error_map(stateid);
3267 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3270 renew_client(dp->dl_client);
3272 unhash_delegation(dp);
3274 nfs4_unlock_state();
3281 * Lock owner state (byte-range locks)
3283 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3284 #define LOCK_HASH_BITS 8
3285 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3286 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3289 end_offset(u64 start, u64 len)
3294 return end >= start ? end: NFS4_MAX_UINT64;
3297 /* last octet in a range */
3299 last_byte_offset(u64 start, u64 len)
3305 return end > start ? end - 1: NFS4_MAX_UINT64;
3308 #define lockownerid_hashval(id) \
3309 ((id) & LOCK_HASH_MASK)
3311 static inline unsigned int
3312 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3313 struct xdr_netobj *ownername)
3315 return (file_hashval(inode) + cl_id
3316 + opaque_hashval(ownername->data, ownername->len))
3320 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3321 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3322 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3324 static struct nfs4_stateid *
3325 find_stateid(stateid_t *stid, int flags)
3327 struct nfs4_stateid *local;
3328 u32 st_id = stid->si_stateownerid;
3329 u32 f_id = stid->si_fileid;
3330 unsigned int hashval;
3332 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3333 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3334 hashval = stateid_hashval(st_id, f_id);
3335 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3336 if ((local->st_stateid.si_stateownerid == st_id) &&
3337 (local->st_stateid.si_fileid == f_id))
3342 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3343 hashval = stateid_hashval(st_id, f_id);
3344 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3345 if ((local->st_stateid.si_stateownerid == st_id) &&
3346 (local->st_stateid.si_fileid == f_id))
3353 static struct nfs4_delegation *
3354 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3356 struct nfs4_file *fp;
3357 struct nfs4_delegation *dl;
3359 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3360 stid->si_boot, stid->si_stateownerid,
3361 stid->si_fileid, stid->si_generation);
3363 fp = find_file(ino);
3366 dl = find_delegation_file(fp, stid);
3372 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3373 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3374 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3375 * locking, this prevents us from being completely protocol-compliant. The
3376 * real solution to this problem is to start using unsigned file offsets in
3377 * the VFS, but this is a very deep change!
3380 nfs4_transform_lock_offset(struct file_lock *lock)
3382 if (lock->fl_start < 0)
3383 lock->fl_start = OFFSET_MAX;
3384 if (lock->fl_end < 0)
3385 lock->fl_end = OFFSET_MAX;
3388 /* Hack!: For now, we're defining this just so we can use a pointer to it
3389 * as a unique cookie to identify our (NFSv4's) posix locks. */
3390 static struct lock_manager_operations nfsd_posix_mng_ops = {
3394 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3396 struct nfs4_stateowner *sop;
3399 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3400 sop = (struct nfs4_stateowner *) fl->fl_owner;
3401 hval = lockownerid_hashval(sop->so_id);
3402 kref_get(&sop->so_ref);
3404 deny->ld_clientid = sop->so_client->cl_clientid;
3406 deny->ld_sop = NULL;
3407 deny->ld_clientid.cl_boot = 0;
3408 deny->ld_clientid.cl_id = 0;
3410 deny->ld_start = fl->fl_start;
3411 deny->ld_length = NFS4_MAX_UINT64;
3412 if (fl->fl_end != NFS4_MAX_UINT64)
3413 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3414 deny->ld_type = NFS4_READ_LT;
3415 if (fl->fl_type != F_RDLCK)
3416 deny->ld_type = NFS4_WRITE_LT;
3419 static struct nfs4_stateowner *
3420 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3421 struct xdr_netobj *owner)
3423 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3424 struct nfs4_stateowner *op;
3426 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3427 if (same_owner_str(op, owner, clid))
3434 * Alloc a lock owner structure.
3435 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3438 * strhashval = lock_ownerstr_hashval
3441 static struct nfs4_stateowner *
3442 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3443 struct nfs4_stateowner *sop;
3444 struct nfs4_replay *rp;
3445 unsigned int idhashval;
3447 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3449 idhashval = lockownerid_hashval(current_ownerid);
3450 INIT_LIST_HEAD(&sop->so_idhash);
3451 INIT_LIST_HEAD(&sop->so_strhash);
3452 INIT_LIST_HEAD(&sop->so_perclient);
3453 INIT_LIST_HEAD(&sop->so_stateids);
3454 INIT_LIST_HEAD(&sop->so_perstateid);
3455 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3457 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3458 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3459 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3460 sop->so_is_open_owner = 0;
3461 sop->so_id = current_ownerid++;
3462 sop->so_client = clp;
3463 /* It is the openowner seqid that will be incremented in encode in the
3464 * case of new lockowners; so increment the lock seqid manually: */
3465 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3466 sop->so_confirmed = 1;
3467 rp = &sop->so_replay;
3468 rp->rp_status = nfserr_serverfault;
3470 rp->rp_buf = rp->rp_ibuf;
3474 static struct nfs4_stateid *
3475 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3477 struct nfs4_stateid *stp;
3478 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3480 stp = nfs4_alloc_stateid();
3483 INIT_LIST_HEAD(&stp->st_hash);
3484 INIT_LIST_HEAD(&stp->st_perfile);
3485 INIT_LIST_HEAD(&stp->st_perstateowner);
3486 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3487 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3488 list_add(&stp->st_perfile, &fp->fi_stateids);
3489 list_add(&stp->st_perstateowner, &sop->so_stateids);
3490 stp->st_stateowner = sop;
3493 stp->st_stateid.si_boot = get_seconds();
3494 stp->st_stateid.si_stateownerid = sop->so_id;
3495 stp->st_stateid.si_fileid = fp->fi_id;
3496 stp->st_stateid.si_generation = 0;
3497 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3498 stp->st_access_bmap = open_stp->st_access_bmap;
3499 stp->st_deny_bmap = open_stp->st_deny_bmap;
3500 stp->st_openstp = open_stp;
3507 check_lock_length(u64 offset, u64 length)
3509 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3510 LOFF_OVERFLOW(offset, length)));
3517 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3518 struct nfsd4_lock *lock)
3520 struct nfs4_stateowner *open_sop = NULL;
3521 struct nfs4_stateowner *lock_sop = NULL;
3522 struct nfs4_stateid *lock_stp;
3524 struct file_lock file_lock;
3525 struct file_lock conflock;
3527 unsigned int strhashval;
3531 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3532 (long long) lock->lk_offset,
3533 (long long) lock->lk_length);
3535 if (check_lock_length(lock->lk_offset, lock->lk_length))
3536 return nfserr_inval;
3538 if ((status = fh_verify(rqstp, &cstate->current_fh,
3539 S_IFREG, NFSD_MAY_LOCK))) {
3540 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3546 if (lock->lk_is_new) {
3548 * Client indicates that this is a new lockowner.
3549 * Use open owner and open stateid to create lock owner and
3552 struct nfs4_stateid *open_stp = NULL;
3553 struct nfs4_file *fp;
3555 status = nfserr_stale_clientid;
3556 if (!nfsd4_has_session(cstate) &&
3557 STALE_CLIENTID(&lock->lk_new_clientid))
3560 /* validate and update open stateid and open seqid */
3561 status = nfs4_preprocess_seqid_op(cstate,
3562 lock->lk_new_open_seqid,
3563 &lock->lk_new_open_stateid,
3565 &lock->lk_replay_owner, &open_stp,
3569 open_sop = lock->lk_replay_owner;
3570 /* create lockowner and lock stateid */
3571 fp = open_stp->st_file;
3572 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3573 open_sop->so_client->cl_clientid.cl_id,
3574 &lock->v.new.owner);
3575 /* XXX: Do we need to check for duplicate stateowners on
3576 * the same file, or should they just be allowed (and
3577 * create new stateids)? */
3578 status = nfserr_resource;
3579 lock_sop = alloc_init_lock_stateowner(strhashval,
3580 open_sop->so_client, open_stp, lock);
3581 if (lock_sop == NULL)
3583 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3584 if (lock_stp == NULL)
3587 /* lock (lock owner + lock stateid) already exists */
3588 status = nfs4_preprocess_seqid_op(cstate,
3589 lock->lk_old_lock_seqid,
3590 &lock->lk_old_lock_stateid,
3592 &lock->lk_replay_owner, &lock_stp, lock);
3595 lock_sop = lock->lk_replay_owner;
3597 /* lock->lk_replay_owner and lock_stp have been created or found */
3598 filp = lock_stp->st_vfs_file;
3600 status = nfserr_grace;
3601 if (locks_in_grace() && !lock->lk_reclaim)
3603 status = nfserr_no_grace;
3604 if (!locks_in_grace() && lock->lk_reclaim)
3607 locks_init_lock(&file_lock);
3608 switch (lock->lk_type) {
3611 file_lock.fl_type = F_RDLCK;
3615 case NFS4_WRITEW_LT:
3616 file_lock.fl_type = F_WRLCK;
3620 status = nfserr_inval;
3623 file_lock.fl_owner = (fl_owner_t)lock_sop;
3624 file_lock.fl_pid = current->tgid;
3625 file_lock.fl_file = filp;
3626 file_lock.fl_flags = FL_POSIX;
3627 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3629 file_lock.fl_start = lock->lk_offset;
3630 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3631 nfs4_transform_lock_offset(&file_lock);
3634 * Try to lock the file in the VFS.
3635 * Note: locks.c uses the BKL to protect the inode's lock list.
3638 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3640 case 0: /* success! */
3641 update_stateid(&lock_stp->st_stateid);
3642 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3646 case (EAGAIN): /* conflock holds conflicting lock */
3647 status = nfserr_denied;
3648 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3649 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3652 status = nfserr_deadlock;
3655 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3656 status = nfserr_resource;
3660 if (status && lock->lk_is_new && lock_sop)
3661 release_lockowner(lock_sop);
3662 if (lock->lk_replay_owner) {
3663 nfs4_get_stateowner(lock->lk_replay_owner);
3664 cstate->replay_owner = lock->lk_replay_owner;
3666 nfs4_unlock_state();
3671 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3672 * so we do a temporary open here just to get an open file to pass to
3673 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3676 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3681 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3684 err = vfs_test_lock(file, lock);
3693 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3694 struct nfsd4_lockt *lockt)
3696 struct inode *inode;
3697 struct file_lock file_lock;
3701 if (locks_in_grace())
3702 return nfserr_grace;
3704 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3705 return nfserr_inval;
3707 lockt->lt_stateowner = NULL;
3710 status = nfserr_stale_clientid;
3711 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3714 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3715 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3716 if (status == nfserr_symlink)
3717 status = nfserr_inval;
3721 inode = cstate->current_fh.fh_dentry->d_inode;
3722 locks_init_lock(&file_lock);
3723 switch (lockt->lt_type) {
3726 file_lock.fl_type = F_RDLCK;
3729 case NFS4_WRITEW_LT:
3730 file_lock.fl_type = F_WRLCK;
3733 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3734 status = nfserr_inval;
3738 lockt->lt_stateowner = find_lockstateowner_str(inode,
3739 &lockt->lt_clientid, &lockt->lt_owner);
3740 if (lockt->lt_stateowner)
3741 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3742 file_lock.fl_pid = current->tgid;
3743 file_lock.fl_flags = FL_POSIX;
3745 file_lock.fl_start = lockt->lt_offset;
3746 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3748 nfs4_transform_lock_offset(&file_lock);
3751 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3753 status = nfserrno(error);
3756 if (file_lock.fl_type != F_UNLCK) {
3757 status = nfserr_denied;
3758 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3761 nfs4_unlock_state();
3766 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3767 struct nfsd4_locku *locku)
3769 struct nfs4_stateid *stp;
3770 struct file *filp = NULL;
3771 struct file_lock file_lock;
3775 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3776 (long long) locku->lu_offset,
3777 (long long) locku->lu_length);
3779 if (check_lock_length(locku->lu_offset, locku->lu_length))
3780 return nfserr_inval;
3784 if ((status = nfs4_preprocess_seqid_op(cstate,
3788 &locku->lu_stateowner, &stp, NULL)))
3791 filp = stp->st_vfs_file;
3793 locks_init_lock(&file_lock);
3794 file_lock.fl_type = F_UNLCK;
3795 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3796 file_lock.fl_pid = current->tgid;
3797 file_lock.fl_file = filp;
3798 file_lock.fl_flags = FL_POSIX;
3799 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3800 file_lock.fl_start = locku->lu_offset;
3802 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3803 nfs4_transform_lock_offset(&file_lock);
3806 * Try to unlock the file in the VFS.
3808 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3810 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3814 * OK, unlock succeeded; the only thing left to do is update the stateid.
3816 update_stateid(&stp->st_stateid);
3817 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3820 if (locku->lu_stateowner) {
3821 nfs4_get_stateowner(locku->lu_stateowner);
3822 cstate->replay_owner = locku->lu_stateowner;
3824 nfs4_unlock_state();
3828 status = nfserrno(err);
3834 * 1: locks held by lockowner
3835 * 0: no locks held by lockowner
3838 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3840 struct file_lock **flpp;
3841 struct inode *inode = filp->f_path.dentry->d_inode;
3845 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3846 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3857 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3858 struct nfsd4_compound_state *cstate,
3859 struct nfsd4_release_lockowner *rlockowner)
3861 clientid_t *clid = &rlockowner->rl_clientid;
3862 struct nfs4_stateowner *sop;
3863 struct nfs4_stateid *stp;
3864 struct xdr_netobj *owner = &rlockowner->rl_owner;
3865 struct list_head matches;
3869 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3870 clid->cl_boot, clid->cl_id);
3872 /* XXX check for lease expiration */
3874 status = nfserr_stale_clientid;
3875 if (STALE_CLIENTID(clid))
3880 status = nfserr_locks_held;
3881 /* XXX: we're doing a linear search through all the lockowners.
3882 * Yipes! For now we'll just hope clients aren't really using
3883 * release_lockowner much, but eventually we have to fix these
3884 * data structures. */
3885 INIT_LIST_HEAD(&matches);
3886 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3887 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3888 if (!same_owner_str(sop, owner, clid))
3890 list_for_each_entry(stp, &sop->so_stateids,
3892 if (check_for_locks(stp->st_vfs_file, sop))
3894 /* Note: so_perclient unused for lockowners,
3895 * so it's OK to fool with here. */
3896 list_add(&sop->so_perclient, &matches);
3900 /* Clients probably won't expect us to return with some (but not all)
3901 * of the lockowner state released; so don't release any until all
3902 * have been checked. */
3904 while (!list_empty(&matches)) {
3905 sop = list_entry(matches.next, struct nfs4_stateowner,
3907 /* unhash_stateowner deletes so_perclient only
3908 * for openowners. */
3909 list_del(&sop->so_perclient);
3910 release_lockowner(sop);
3913 nfs4_unlock_state();
3917 static inline struct nfs4_client_reclaim *
3920 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3924 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3926 unsigned int strhashval = clientstr_hashval(name);
3927 struct nfs4_client *clp;
3929 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3934 * failure => all reset bets are off, nfserr_no_grace...
3937 nfs4_client_to_reclaim(const char *name)
3939 unsigned int strhashval;
3940 struct nfs4_client_reclaim *crp = NULL;
3942 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3943 crp = alloc_reclaim();
3946 strhashval = clientstr_hashval(name);
3947 INIT_LIST_HEAD(&crp->cr_strhash);
3948 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3949 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3950 reclaim_str_hashtbl_size++;
3955 nfs4_release_reclaim(void)
3957 struct nfs4_client_reclaim *crp = NULL;
3960 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3961 while (!list_empty(&reclaim_str_hashtbl[i])) {
3962 crp = list_entry(reclaim_str_hashtbl[i].next,
3963 struct nfs4_client_reclaim, cr_strhash);
3964 list_del(&crp->cr_strhash);
3966 reclaim_str_hashtbl_size--;
3969 BUG_ON(reclaim_str_hashtbl_size);
3973 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3974 static struct nfs4_client_reclaim *
3975 nfs4_find_reclaim_client(clientid_t *clid)
3977 unsigned int strhashval;
3978 struct nfs4_client *clp;
3979 struct nfs4_client_reclaim *crp = NULL;
3982 /* find clientid in conf_id_hashtbl */
3983 clp = find_confirmed_client(clid);
3987 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3988 clp->cl_name.len, clp->cl_name.data,
3991 /* find clp->cl_name in reclaim_str_hashtbl */
3992 strhashval = clientstr_hashval(clp->cl_recdir);
3993 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3994 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4002 * Called from OPEN. Look for clientid in reclaim list.
4005 nfs4_check_open_reclaim(clientid_t *clid)
4007 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4010 /* initialization to perform at module load time: */
4013 nfs4_state_init(void)
4017 status = nfsd4_init_slabs();
4020 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4021 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4022 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4023 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4024 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4026 for (i = 0; i < SESSION_HASH_SIZE; i++)
4027 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4028 for (i = 0; i < FILE_HASH_SIZE; i++) {
4029 INIT_LIST_HEAD(&file_hashtbl[i]);
4031 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4032 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4033 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4035 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4036 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4037 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4039 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4040 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4041 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4043 memset(&onestateid, ~0, sizeof(stateid_t));
4044 INIT_LIST_HEAD(&close_lru);
4045 INIT_LIST_HEAD(&client_lru);
4046 INIT_LIST_HEAD(&del_recall_lru);
4047 for (i = 0; i < CLIENT_HASH_SIZE; i++)
4048 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4049 reclaim_str_hashtbl_size = 0;
4054 nfsd4_load_reboot_recovery_data(void)
4059 nfsd4_init_recdir(user_recovery_dirname);
4060 status = nfsd4_recdir_load();
4061 nfs4_unlock_state();
4063 printk("NFSD: Failure reading reboot recovery data\n");
4067 get_nfs4_grace_period(void)
4069 return max(user_lease_time, lease_time) * HZ;
4073 * Since the lifetime of a delegation isn't limited to that of an open, a
4074 * client may quite reasonably hang on to a delegation as long as it has
4075 * the inode cached. This becomes an obvious problem the first time a
4076 * client's inode cache approaches the size of the server's total memory.
4078 * For now we avoid this problem by imposing a hard limit on the number
4079 * of delegations, which varies according to the server's memory size.
4082 set_max_delegations(void)
4085 * Allow at most 4 delegations per megabyte of RAM. Quick
4086 * estimates suggest that in the worst case (where every delegation
4087 * is for a different inode), a delegation could take about 1.5K,
4088 * giving a worst case usage of about 6% of memory.
4090 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4093 /* initialization to perform when the nfsd service is started: */
4096 __nfs4_state_start(void)
4098 unsigned long grace_time;
4100 boot_time = get_seconds();
4101 grace_time = get_nfs4_grace_period();
4102 lease_time = user_lease_time;
4103 locks_start_grace(&nfsd4_manager);
4104 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4106 laundry_wq = create_singlethread_workqueue("nfsd4");
4107 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4108 set_max_delegations();
4112 nfs4_state_start(void)
4116 nfsd4_load_reboot_recovery_data();
4117 __nfs4_state_start();
4123 nfs4_lease_time(void)
4129 __nfs4_state_shutdown(void)
4132 struct nfs4_client *clp = NULL;
4133 struct nfs4_delegation *dp = NULL;
4134 struct list_head *pos, *next, reaplist;
4136 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4137 while (!list_empty(&conf_id_hashtbl[i])) {
4138 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4141 while (!list_empty(&unconf_str_hashtbl[i])) {
4142 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4146 INIT_LIST_HEAD(&reaplist);
4147 spin_lock(&recall_lock);
4148 list_for_each_safe(pos, next, &del_recall_lru) {
4149 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4150 list_move(&dp->dl_recall_lru, &reaplist);
4152 spin_unlock(&recall_lock);
4153 list_for_each_safe(pos, next, &reaplist) {
4154 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4155 list_del_init(&dp->dl_recall_lru);
4156 unhash_delegation(dp);
4159 nfsd4_shutdown_recdir();
4164 nfs4_state_shutdown(void)
4166 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4167 destroy_workqueue(laundry_wq);
4168 locks_end_grace(&nfsd4_manager);
4170 nfs4_release_reclaim();
4171 __nfs4_state_shutdown();
4172 nfs4_unlock_state();
4176 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4177 * accessed when nfsd is starting.
4180 nfs4_set_recdir(char *recdir)
4182 strcpy(user_recovery_dirname, recdir);
4186 * Change the NFSv4 recovery directory to recdir.
4189 nfs4_reset_recoverydir(char *recdir)
4194 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4198 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4199 nfs4_set_recdir(recdir);
4207 nfs4_recoverydir(void)
4209 return user_recovery_dirname;
4213 * Called when leasetime is changed.
4215 * The only way the protocol gives us to handle on-the-fly lease changes is to
4216 * simulate a reboot. Instead of doing that, we just wait till the next time
4217 * we start to register any changes in lease time. If the administrator
4218 * really wants to change the lease time *now*, they can go ahead and bring
4219 * nfsd down and then back up again after changing the lease time.
4221 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4222 * when nfsd is starting
4225 nfs4_reset_lease(time_t leasetime)
4227 user_lease_time = leasetime;