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_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
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_ident = cb->cb_ident;
207 dp->dl_stateid.si_boot = get_seconds();
208 dp->dl_stateid.si_stateownerid = current_delegid++;
209 dp->dl_stateid.si_fileid = 0;
210 dp->dl_stateid.si_generation = 0;
211 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
213 atomic_set(&dp->dl_count, 1);
214 list_add(&dp->dl_perfile, &fp->fi_delegations);
215 list_add(&dp->dl_perclnt, &clp->cl_delegations);
220 nfs4_put_delegation(struct nfs4_delegation *dp)
222 if (atomic_dec_and_test(&dp->dl_count)) {
223 dprintk("NFSD: freeing dp %p\n",dp);
224 put_nfs4_file(dp->dl_file);
225 kmem_cache_free(deleg_slab, dp);
230 /* Remove the associated file_lock first, then remove the delegation.
231 * lease_modify() is called to remove the FS_LEASE file_lock from
232 * the i_flock list, eventually calling nfsd's lock_manager
233 * fl_release_callback.
236 nfs4_close_delegation(struct nfs4_delegation *dp)
238 struct file *filp = dp->dl_vfs_file;
240 dprintk("NFSD: close_delegation dp %p\n",dp);
241 dp->dl_vfs_file = NULL;
242 /* The following nfsd_close may not actually close the file,
243 * but we want to remove the lease in any case. */
245 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
249 /* Called under the state lock. */
251 unhash_delegation(struct nfs4_delegation *dp)
253 list_del_init(&dp->dl_perfile);
254 list_del_init(&dp->dl_perclnt);
255 spin_lock(&recall_lock);
256 list_del_init(&dp->dl_recall_lru);
257 spin_unlock(&recall_lock);
258 nfs4_close_delegation(dp);
259 nfs4_put_delegation(dp);
266 /* Hash tables for nfs4_clientid state */
267 #define CLIENT_HASH_BITS 4
268 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
269 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
271 #define clientid_hashval(id) \
272 ((id) & CLIENT_HASH_MASK)
273 #define clientstr_hashval(name) \
274 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
276 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
277 * used in reboot/reset lease grace period processing
279 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
280 * setclientid_confirmed info.
282 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
285 * client_lru holds client queue ordered by nfs4_client.cl_time
288 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
289 * for last close replay.
291 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
292 static int reclaim_str_hashtbl_size = 0;
293 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
294 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head client_lru;
298 static struct list_head close_lru;
300 static void unhash_generic_stateid(struct nfs4_stateid *stp)
302 list_del(&stp->st_hash);
303 list_del(&stp->st_perfile);
304 list_del(&stp->st_perstateowner);
307 static void free_generic_stateid(struct nfs4_stateid *stp)
309 put_nfs4_file(stp->st_file);
310 kmem_cache_free(stateid_slab, stp);
313 static void release_lock_stateid(struct nfs4_stateid *stp)
315 unhash_generic_stateid(stp);
316 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
317 free_generic_stateid(stp);
320 static void unhash_lockowner(struct nfs4_stateowner *sop)
322 struct nfs4_stateid *stp;
324 list_del(&sop->so_idhash);
325 list_del(&sop->so_strhash);
326 list_del(&sop->so_perstateid);
327 while (!list_empty(&sop->so_stateids)) {
328 stp = list_first_entry(&sop->so_stateids,
329 struct nfs4_stateid, st_perstateowner);
330 release_lock_stateid(stp);
334 static void release_lockowner(struct nfs4_stateowner *sop)
336 unhash_lockowner(sop);
337 nfs4_put_stateowner(sop);
341 release_stateid_lockowners(struct nfs4_stateid *open_stp)
343 struct nfs4_stateowner *lock_sop;
345 while (!list_empty(&open_stp->st_lockowners)) {
346 lock_sop = list_entry(open_stp->st_lockowners.next,
347 struct nfs4_stateowner, so_perstateid);
348 /* list_del(&open_stp->st_lockowners); */
349 BUG_ON(lock_sop->so_is_open_owner);
350 release_lockowner(lock_sop);
354 static void release_open_stateid(struct nfs4_stateid *stp)
356 unhash_generic_stateid(stp);
357 release_stateid_lockowners(stp);
358 nfsd_close(stp->st_vfs_file);
359 free_generic_stateid(stp);
362 static void unhash_openowner(struct nfs4_stateowner *sop)
364 struct nfs4_stateid *stp;
366 list_del(&sop->so_idhash);
367 list_del(&sop->so_strhash);
368 list_del(&sop->so_perclient);
369 list_del(&sop->so_perstateid); /* XXX: necessary? */
370 while (!list_empty(&sop->so_stateids)) {
371 stp = list_first_entry(&sop->so_stateids,
372 struct nfs4_stateid, st_perstateowner);
373 release_open_stateid(stp);
377 static void release_openowner(struct nfs4_stateowner *sop)
379 unhash_openowner(sop);
380 list_del(&sop->so_close_lru);
381 nfs4_put_stateowner(sop);
384 static DEFINE_SPINLOCK(sessionid_lock);
385 #define SESSION_HASH_SIZE 512
386 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
389 hash_sessionid(struct nfs4_sessionid *sessionid)
391 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
393 return sid->sequence % SESSION_HASH_SIZE;
397 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
399 u32 *ptr = (u32 *)(&sessionid->data[0]);
400 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
404 gen_sessionid(struct nfsd4_session *ses)
406 struct nfs4_client *clp = ses->se_client;
407 struct nfsd4_sessionid *sid;
409 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
410 sid->clientid = clp->cl_clientid;
411 sid->sequence = current_sessionid++;
416 * Give the client the number of slots it requests bound by
417 * NFSD_MAX_SLOTS_PER_SESSION and by nfsd_drc_max_mem.
419 * If we run out of reserved DRC memory we should (up to a point) re-negotiate
420 * active sessions and reduce their slot usage to make rooom for new
421 * connections. For now we just fail the create session.
423 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
427 if (fchan->maxreqs < 1)
429 else if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
430 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
432 mem = fchan->maxreqs * NFSD_SLOT_CACHE_SIZE;
434 spin_lock(&nfsd_drc_lock);
435 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem)
436 mem = ((nfsd_drc_max_mem - nfsd_drc_mem_used) /
437 NFSD_SLOT_CACHE_SIZE) * NFSD_SLOT_CACHE_SIZE;
438 nfsd_drc_mem_used += mem;
439 spin_unlock(&nfsd_drc_lock);
441 fchan->maxreqs = mem / NFSD_SLOT_CACHE_SIZE;
442 if (fchan->maxreqs == 0)
443 return nfserr_resource;
448 * fchan holds the client values on input, and the server values on output
450 static int init_forechannel_attrs(struct svc_rqst *rqstp,
451 struct nfsd4_channel_attrs *session_fchan,
452 struct nfsd4_channel_attrs *fchan)
455 __u32 maxcount = svc_max_payload(rqstp);
457 /* headerpadsz set to zero in encode routine */
459 /* Use the client's max request and max response size if possible */
460 if (fchan->maxreq_sz > maxcount)
461 fchan->maxreq_sz = maxcount;
462 session_fchan->maxreq_sz = fchan->maxreq_sz;
464 if (fchan->maxresp_sz > maxcount)
465 fchan->maxresp_sz = maxcount;
466 session_fchan->maxresp_sz = fchan->maxresp_sz;
468 session_fchan->maxresp_cached = NFSD_SLOT_CACHE_SIZE;
469 fchan->maxresp_cached = session_fchan->maxresp_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_fchan->maxops = fchan->maxops;
476 /* FIXME: Error means no more DRC pages so the server should
477 * recover pages from existing sessions. For now fail session
480 status = set_forechannel_maxreqs(fchan);
482 session_fchan->maxreqs = fchan->maxreqs;
487 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
488 struct nfsd4_create_session *cses)
490 struct nfsd4_session *new, tmp;
491 int idx, status = nfserr_resource, slotsize;
493 memset(&tmp, 0, sizeof(tmp));
495 /* FIXME: For now, we just accept the client back channel attributes. */
496 tmp.se_bchannel = cses->back_channel;
497 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
498 &cses->fore_channel);
502 /* allocate struct nfsd4_session and slot table in one piece */
503 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot);
504 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
508 memcpy(new, &tmp, sizeof(*new));
510 new->se_client = clp;
512 idx = hash_sessionid(&new->se_sessionid);
513 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
514 NFS4_MAX_SESSIONID_LEN);
516 new->se_flags = cses->flags;
517 kref_init(&new->se_ref);
518 spin_lock(&sessionid_lock);
519 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
520 list_add(&new->se_perclnt, &clp->cl_sessions);
521 spin_unlock(&sessionid_lock);
528 /* caller must hold sessionid_lock */
529 static struct nfsd4_session *
530 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
532 struct nfsd4_session *elem;
535 dump_sessionid(__func__, sessionid);
536 idx = hash_sessionid(sessionid);
537 dprintk("%s: idx is %d\n", __func__, idx);
538 /* Search in the appropriate list */
539 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
540 dump_sessionid("list traversal", &elem->se_sessionid);
541 if (!memcmp(elem->se_sessionid.data, sessionid->data,
542 NFS4_MAX_SESSIONID_LEN)) {
547 dprintk("%s: session not found\n", __func__);
551 /* caller must hold sessionid_lock */
553 unhash_session(struct nfsd4_session *ses)
555 list_del(&ses->se_hash);
556 list_del(&ses->se_perclnt);
560 release_session(struct nfsd4_session *ses)
562 spin_lock(&sessionid_lock);
564 spin_unlock(&sessionid_lock);
565 nfsd4_put_session(ses);
568 static void nfsd4_release_respages(struct page **respages, short resused);
571 free_session(struct kref *kref)
573 struct nfsd4_session *ses;
576 ses = container_of(kref, struct nfsd4_session, se_ref);
577 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
578 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
579 nfsd4_release_respages(e->ce_respages, e->ce_resused);
581 spin_lock(&nfsd_drc_lock);
582 nfsd_drc_mem_used -= ses->se_fchannel.maxreqs * NFSD_SLOT_CACHE_SIZE;
583 spin_unlock(&nfsd_drc_lock);
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_cb_conn.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_cb_conn.cb_client = NULL;
644 rpc_shutdown_client(clnt);
646 if (clp->cl_cb_conn.cb_cred) {
647 put_rpccred(clp->cl_cb_conn.cb_cred);
648 clp->cl_cb_conn.cb_cred = NULL;
653 free_client(struct nfs4_client *clp)
655 shutdown_callback_client(clp);
656 if (clp->cl_cred.cr_group_info)
657 put_group_info(clp->cl_cred.cr_group_info);
658 kfree(clp->cl_principal);
659 kfree(clp->cl_name.data);
664 put_nfs4_client(struct nfs4_client *clp)
666 if (atomic_dec_and_test(&clp->cl_count))
671 expire_client(struct nfs4_client *clp)
673 struct nfs4_stateowner *sop;
674 struct nfs4_delegation *dp;
675 struct list_head reaplist;
677 dprintk("NFSD: expire_client cl_count %d\n",
678 atomic_read(&clp->cl_count));
680 INIT_LIST_HEAD(&reaplist);
681 spin_lock(&recall_lock);
682 while (!list_empty(&clp->cl_delegations)) {
683 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
684 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
686 list_del_init(&dp->dl_perclnt);
687 list_move(&dp->dl_recall_lru, &reaplist);
689 spin_unlock(&recall_lock);
690 while (!list_empty(&reaplist)) {
691 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
692 list_del_init(&dp->dl_recall_lru);
693 unhash_delegation(dp);
695 list_del(&clp->cl_idhash);
696 list_del(&clp->cl_strhash);
697 list_del(&clp->cl_lru);
698 while (!list_empty(&clp->cl_openowners)) {
699 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
700 release_openowner(sop);
702 while (!list_empty(&clp->cl_sessions)) {
703 struct nfsd4_session *ses;
704 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
706 release_session(ses);
708 put_nfs4_client(clp);
711 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
713 struct nfs4_client *clp;
715 clp = alloc_client(name);
718 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
719 atomic_set(&clp->cl_count, 1);
720 atomic_set(&clp->cl_cb_conn.cb_set, 0);
721 INIT_LIST_HEAD(&clp->cl_idhash);
722 INIT_LIST_HEAD(&clp->cl_strhash);
723 INIT_LIST_HEAD(&clp->cl_openowners);
724 INIT_LIST_HEAD(&clp->cl_delegations);
725 INIT_LIST_HEAD(&clp->cl_sessions);
726 INIT_LIST_HEAD(&clp->cl_lru);
730 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
732 memcpy(target->cl_verifier.data, source->data,
733 sizeof(target->cl_verifier.data));
736 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
738 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
739 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
742 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
744 target->cr_uid = source->cr_uid;
745 target->cr_gid = source->cr_gid;
746 target->cr_group_info = source->cr_group_info;
747 get_group_info(target->cr_group_info);
750 static int same_name(const char *n1, const char *n2)
752 return 0 == memcmp(n1, n2, HEXDIR_LEN);
756 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
758 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
762 same_clid(clientid_t *cl1, clientid_t *cl2)
764 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
767 /* XXX what about NGROUP */
769 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
771 return cr1->cr_uid == cr2->cr_uid;
774 static void gen_clid(struct nfs4_client *clp)
776 static u32 current_clientid = 1;
778 clp->cl_clientid.cl_boot = boot_time;
779 clp->cl_clientid.cl_id = current_clientid++;
782 static void gen_confirm(struct nfs4_client *clp)
787 p = (u32 *)clp->cl_confirm.data;
788 *p++ = get_seconds();
792 static int check_name(struct xdr_netobj name)
796 if (name.len > NFS4_OPAQUE_LIMIT) {
797 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
804 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
806 unsigned int idhashval;
808 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
809 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
810 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
811 list_add_tail(&clp->cl_lru, &client_lru);
812 clp->cl_time = get_seconds();
816 move_to_confirmed(struct nfs4_client *clp)
818 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
819 unsigned int strhashval;
821 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
822 list_del_init(&clp->cl_strhash);
823 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
824 strhashval = clientstr_hashval(clp->cl_recdir);
825 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
829 static struct nfs4_client *
830 find_confirmed_client(clientid_t *clid)
832 struct nfs4_client *clp;
833 unsigned int idhashval = clientid_hashval(clid->cl_id);
835 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
836 if (same_clid(&clp->cl_clientid, clid))
842 static struct nfs4_client *
843 find_unconfirmed_client(clientid_t *clid)
845 struct nfs4_client *clp;
846 unsigned int idhashval = clientid_hashval(clid->cl_id);
848 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
849 if (same_clid(&clp->cl_clientid, clid))
856 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
857 * parameter. Matching is based on the fact the at least one of the
858 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
860 * FIXME: we need to unify the clientid namespaces for nfsv4.x
861 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
862 * and SET_CLIENTID{,_CONFIRM}
865 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
867 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
868 return use_exchange_id == has_exchange_flags;
871 static struct nfs4_client *
872 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
873 bool use_exchange_id)
875 struct nfs4_client *clp;
877 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
878 if (same_name(clp->cl_recdir, dname) &&
879 match_clientid_establishment(clp, use_exchange_id))
885 static struct nfs4_client *
886 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
887 bool use_exchange_id)
889 struct nfs4_client *clp;
891 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
892 if (same_name(clp->cl_recdir, dname) &&
893 match_clientid_establishment(clp, use_exchange_id))
899 /* a helper function for parse_callback */
901 parse_octet(unsigned int *lenp, char **addrp)
903 unsigned int len = *lenp;
915 if ((c < '0') || (c > '9')) {
921 n = (n * 10) + (c - '0');
932 /* parse and set the setclientid ipv4 callback address */
934 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
939 u32 addrlen = addr_len;
940 char *addr = addr_val;
945 for(i = 4; i > 0 ; i--) {
946 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
949 cbaddr |= (temp << shift);
957 for(i = 2; i > 0 ; i--) {
958 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
961 cbport |= (temp << shift);
970 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
972 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
974 /* Currently, we only support tcp for the callback channel */
975 if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
978 if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
979 &cb->cb_addr, &cb->cb_port)))
981 cb->cb_minorversion = 0;
982 cb->cb_prog = se->se_callback_prog;
983 cb->cb_ident = se->se_callback_ident;
986 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
987 "will not receive delegations\n",
988 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
994 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
996 struct nfsd4_compoundres *resp = rqstp->rq_resp;
998 resp->cstate.statp = statp;
1002 * Dereference the result pages.
1005 nfsd4_release_respages(struct page **respages, short resused)
1009 dprintk("--> %s\n", __func__);
1010 for (i = 0; i < resused; i++) {
1013 put_page(respages[i]);
1019 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1023 for (i = 0; i < count; i++) {
1024 topages[i] = frompages[i];
1027 get_page(topages[i]);
1032 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1033 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1034 * length of the XDR response is less than se_fmaxresp_cached
1035 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1036 * of the reply (e.g. readdir).
1038 * Store the base and length of the rq_req.head[0] page
1039 * of the NFSv4.1 data, just past the rpc header.
1042 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1044 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1045 struct svc_rqst *rqstp = resp->rqstp;
1046 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1047 struct nfsd4_op *op = &args->ops[resp->opcnt];
1048 struct kvec *resv = &rqstp->rq_res.head[0];
1050 dprintk("--> %s entry %p\n", __func__, entry);
1052 /* Don't cache a failed OP_SEQUENCE. */
1053 if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1056 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1057 entry->ce_opcnt = resp->opcnt;
1058 entry->ce_status = resp->cstate.status;
1061 * Don't need a page to cache just the sequence operation - the slot
1065 if (nfsd4_not_cached(resp)) {
1066 entry->ce_resused = 0;
1067 entry->ce_rpchdrlen = 0;
1068 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1069 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1072 entry->ce_resused = rqstp->rq_resused;
1073 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1074 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1075 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1077 entry->ce_datav.iov_base = resp->cstate.statp;
1078 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1079 (char *)page_address(rqstp->rq_respages[0]));
1080 /* Current request rpc header length*/
1081 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1082 (char *)page_address(rqstp->rq_respages[0]);
1086 * We keep the rpc header, but take the nfs reply from the replycache.
1089 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1090 struct nfsd4_cache_entry *entry)
1092 struct svc_rqst *rqstp = resp->rqstp;
1093 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1096 /* Current request rpc header length*/
1097 len = (char *)resp->cstate.statp -
1098 (char *)page_address(rqstp->rq_respages[0]);
1099 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1100 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1101 entry->ce_datav.iov_len);
1104 /* copy the cached reply nfsd data past the current rpc header */
1105 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1106 entry->ce_datav.iov_len);
1107 resv->iov_len = len + entry->ce_datav.iov_len;
1112 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1113 * cached page. Replace any futher replay pages from the cache.
1116 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1117 struct nfsd4_sequence *seq)
1119 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1122 dprintk("--> %s entry %p\n", __func__, entry);
1125 * If this is just the sequence operation, we did not keep
1126 * a page in the cache entry because we can just use the
1127 * slot info stored in struct nfsd4_sequence that was checked
1128 * against the slot in nfsd4_sequence().
1130 * This occurs when seq->cachethis is FALSE, or when the client
1131 * session inactivity timer fires and a solo sequence operation
1132 * is sent (lease renewal).
1134 if (seq && nfsd4_not_cached(resp)) {
1135 seq->maxslots = resp->cstate.session->se_fchannel.maxreqs;
1139 if (!nfsd41_copy_replay_data(resp, entry)) {
1141 * Not enough room to use the replay rpc header, send the
1142 * cached header. Release all the allocated result pages.
1144 svc_free_res_pages(resp->rqstp);
1145 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1148 /* Release all but the first allocated result page */
1150 resp->rqstp->rq_resused--;
1151 svc_free_res_pages(resp->rqstp);
1153 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1154 &entry->ce_respages[1],
1155 entry->ce_resused - 1);
1158 resp->rqstp->rq_resused = entry->ce_resused;
1159 resp->opcnt = entry->ce_opcnt;
1160 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1161 status = entry->ce_status;
1167 * Set the exchange_id flags returned by the server.
1170 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1172 /* pNFS is not supported */
1173 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1175 /* Referrals are supported, Migration is not. */
1176 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1178 /* set the wire flags to return to client. */
1179 clid->flags = new->cl_exchange_flags;
1183 nfsd4_exchange_id(struct svc_rqst *rqstp,
1184 struct nfsd4_compound_state *cstate,
1185 struct nfsd4_exchange_id *exid)
1187 struct nfs4_client *unconf, *conf, *new;
1189 unsigned int strhashval;
1190 char dname[HEXDIR_LEN];
1191 nfs4_verifier verf = exid->verifier;
1192 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1194 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1195 " ip_addr=%u flags %x, spa_how %d\n",
1196 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1197 ip_addr, exid->flags, exid->spa_how);
1199 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1200 return nfserr_inval;
1202 /* Currently only support SP4_NONE */
1203 switch (exid->spa_how) {
1207 return nfserr_encr_alg_unsupp;
1209 BUG(); /* checked by xdr code */
1211 return nfserr_serverfault; /* no excuse :-/ */
1214 status = nfs4_make_rec_clidname(dname, &exid->clname);
1219 strhashval = clientstr_hashval(dname);
1224 conf = find_confirmed_client_by_str(dname, strhashval, true);
1226 if (!same_verf(&verf, &conf->cl_verifier)) {
1227 /* 18.35.4 case 8 */
1228 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1229 status = nfserr_not_same;
1232 /* Client reboot: destroy old state */
1233 expire_client(conf);
1236 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1237 /* 18.35.4 case 9 */
1238 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1239 status = nfserr_perm;
1242 expire_client(conf);
1246 * Set bit when the owner id and verifier map to an already
1247 * confirmed client id (18.35.3).
1249 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1252 * Falling into 18.35.4 case 2, possible router replay.
1253 * Leave confirmed record intact and return same result.
1255 copy_verf(conf, &verf);
1260 /* 18.35.4 case 7 */
1261 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1262 status = nfserr_noent;
1266 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1269 * Possible retry or client restart. Per 18.35.4 case 4,
1270 * a new unconfirmed record should be generated regardless
1271 * of whether any properties have changed.
1273 expire_client(unconf);
1278 new = create_client(exid->clname, dname);
1280 status = nfserr_resource;
1284 copy_verf(new, &verf);
1285 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1286 new->cl_addr = ip_addr;
1289 add_to_unconfirmed(new, strhashval);
1291 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1292 exid->clientid.cl_id = new->cl_clientid.cl_id;
1295 nfsd4_set_ex_flags(new, exid);
1297 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1298 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1302 nfs4_unlock_state();
1304 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1309 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1311 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1314 /* The slot is in use, and no response has been sent. */
1316 if (seqid == slot_seqid)
1317 return nfserr_jukebox;
1319 return nfserr_seq_misordered;
1322 if (likely(seqid == slot_seqid + 1))
1325 if (seqid == slot_seqid)
1326 return nfserr_replay_cache;
1328 if (seqid == 1 && (slot_seqid + 1) == 0)
1330 /* Misordered replay or misordered new request */
1331 return nfserr_seq_misordered;
1335 * Cache the create session result into the create session single DRC
1336 * slot cache by saving the xdr structure. sl_seqid has been set.
1337 * Do this for solo or embedded create session operations.
1340 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1341 struct nfsd4_clid_slot *slot, int nfserr)
1343 slot->sl_status = nfserr;
1344 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1348 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1349 struct nfsd4_clid_slot *slot)
1351 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1352 return slot->sl_status;
1356 nfsd4_create_session(struct svc_rqst *rqstp,
1357 struct nfsd4_compound_state *cstate,
1358 struct nfsd4_create_session *cr_ses)
1360 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1361 struct nfs4_client *conf, *unconf;
1362 struct nfsd4_clid_slot *cs_slot = NULL;
1366 unconf = find_unconfirmed_client(&cr_ses->clientid);
1367 conf = find_confirmed_client(&cr_ses->clientid);
1370 cs_slot = &conf->cl_cs_slot;
1371 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1372 if (status == nfserr_replay_cache) {
1373 dprintk("Got a create_session replay! seqid= %d\n",
1375 /* Return the cached reply status */
1376 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1378 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1379 status = nfserr_seq_misordered;
1380 dprintk("Sequence misordered!\n");
1381 dprintk("Expected seqid= %d but got seqid= %d\n",
1382 cs_slot->sl_seqid, cr_ses->seqid);
1385 cs_slot->sl_seqid++;
1386 } else if (unconf) {
1387 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1388 (ip_addr != unconf->cl_addr)) {
1389 status = nfserr_clid_inuse;
1393 cs_slot = &unconf->cl_cs_slot;
1394 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1396 /* an unconfirmed replay returns misordered */
1397 status = nfserr_seq_misordered;
1401 cs_slot->sl_seqid++; /* from 0 to 1 */
1402 move_to_confirmed(unconf);
1405 * We do not support RDMA or persistent sessions
1407 cr_ses->flags &= ~SESSION4_PERSIST;
1408 cr_ses->flags &= ~SESSION4_RDMA;
1412 status = nfserr_stale_clientid;
1416 status = alloc_init_session(rqstp, conf, cr_ses);
1420 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1421 NFS4_MAX_SESSIONID_LEN);
1422 cr_ses->seqid = cs_slot->sl_seqid;
1425 /* cache solo and embedded create sessions under the state lock */
1426 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1428 nfs4_unlock_state();
1429 dprintk("%s returns %d\n", __func__, ntohl(status));
1434 nfsd4_destroy_session(struct svc_rqst *r,
1435 struct nfsd4_compound_state *cstate,
1436 struct nfsd4_destroy_session *sessionid)
1438 struct nfsd4_session *ses;
1439 u32 status = nfserr_badsession;
1442 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1443 * - Should we return nfserr_back_chan_busy if waiting for
1444 * callbacks on to-be-destroyed session?
1445 * - Do we need to clear any callback info from previous session?
1448 dump_sessionid(__func__, &sessionid->sessionid);
1449 spin_lock(&sessionid_lock);
1450 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1452 spin_unlock(&sessionid_lock);
1456 unhash_session(ses);
1457 spin_unlock(&sessionid_lock);
1459 /* wait for callbacks */
1460 shutdown_callback_client(ses->se_client);
1461 nfsd4_put_session(ses);
1464 dprintk("%s returns %d\n", __func__, ntohl(status));
1469 nfsd4_sequence(struct svc_rqst *rqstp,
1470 struct nfsd4_compound_state *cstate,
1471 struct nfsd4_sequence *seq)
1473 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1474 struct nfsd4_session *session;
1475 struct nfsd4_slot *slot;
1478 if (resp->opcnt != 1)
1479 return nfserr_sequence_pos;
1481 spin_lock(&sessionid_lock);
1482 status = nfserr_badsession;
1483 session = find_in_sessionid_hashtbl(&seq->sessionid);
1487 status = nfserr_badslot;
1488 if (seq->slotid >= session->se_fchannel.maxreqs)
1491 slot = &session->se_slots[seq->slotid];
1492 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1494 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1495 if (status == nfserr_replay_cache) {
1496 cstate->slot = slot;
1497 cstate->session = session;
1498 /* Return the cached reply status and set cstate->status
1499 * for nfsd4_svc_encode_compoundres processing */
1500 status = nfsd4_replay_cache_entry(resp, seq);
1501 cstate->status = nfserr_replay_cache;
1507 /* Success! bump slot seqid */
1508 slot->sl_inuse = true;
1509 slot->sl_seqid = seq->seqid;
1510 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1511 /* Always set the cache entry cachethis for solo sequence */
1512 if (nfsd4_is_solo_sequence(resp))
1513 slot->sl_cache_entry.ce_cachethis = 1;
1515 cstate->slot = slot;
1516 cstate->session = session;
1519 /* Renew the clientid on success and on replay.
1520 * Hold a session reference until done processing the compound:
1521 * nfsd4_put_session called only if the cstate slot is set.
1523 renew_client(session->se_client);
1524 nfsd4_get_session(session);
1526 spin_unlock(&sessionid_lock);
1527 dprintk("%s: return %d\n", __func__, ntohl(status));
1532 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1533 struct nfsd4_setclientid *setclid)
1535 struct sockaddr_in *sin = svc_addr_in(rqstp);
1536 struct xdr_netobj clname = {
1537 .len = setclid->se_namelen,
1538 .data = setclid->se_name,
1540 nfs4_verifier clverifier = setclid->se_verf;
1541 unsigned int strhashval;
1542 struct nfs4_client *conf, *unconf, *new;
1545 char dname[HEXDIR_LEN];
1547 if (!check_name(clname))
1548 return nfserr_inval;
1550 status = nfs4_make_rec_clidname(dname, &clname);
1555 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1556 * We get here on a DRC miss.
1559 strhashval = clientstr_hashval(dname);
1562 conf = find_confirmed_client_by_str(dname, strhashval, false);
1564 /* RFC 3530 14.2.33 CASE 0: */
1565 status = nfserr_clid_inuse;
1566 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1567 dprintk("NFSD: setclientid: string in use by client"
1568 " at %pI4\n", &conf->cl_addr);
1573 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1574 * has a description of SETCLIENTID request processing consisting
1575 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1577 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1578 status = nfserr_resource;
1581 * RFC 3530 14.2.33 CASE 4:
1582 * placed first, because it is the normal case
1585 expire_client(unconf);
1586 new = create_client(clname, dname);
1590 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1592 * RFC 3530 14.2.33 CASE 1:
1593 * probable callback update
1596 /* Note this is removing unconfirmed {*x***},
1597 * which is stronger than RFC recommended {vxc**}.
1598 * This has the advantage that there is at most
1599 * one {*x***} in either list at any time.
1601 expire_client(unconf);
1603 new = create_client(clname, dname);
1606 copy_clid(new, conf);
1607 } else if (!unconf) {
1609 * RFC 3530 14.2.33 CASE 2:
1610 * probable client reboot; state will be removed if
1613 new = create_client(clname, dname);
1619 * RFC 3530 14.2.33 CASE 3:
1620 * probable client reboot; state will be removed if
1623 expire_client(unconf);
1624 new = create_client(clname, dname);
1629 copy_verf(new, &clverifier);
1630 new->cl_addr = sin->sin_addr.s_addr;
1631 new->cl_flavor = rqstp->rq_flavor;
1632 princ = svc_gss_principal(rqstp);
1634 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1635 if (new->cl_principal == NULL) {
1640 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1642 gen_callback(new, setclid);
1643 add_to_unconfirmed(new, strhashval);
1644 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1645 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1646 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1649 nfs4_unlock_state();
1655 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1656 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1657 * bullets, labeled as CASE1 - CASE4 below.
1660 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1661 struct nfsd4_compound_state *cstate,
1662 struct nfsd4_setclientid_confirm *setclientid_confirm)
1664 struct sockaddr_in *sin = svc_addr_in(rqstp);
1665 struct nfs4_client *conf, *unconf;
1666 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1667 clientid_t * clid = &setclientid_confirm->sc_clientid;
1670 if (STALE_CLIENTID(clid))
1671 return nfserr_stale_clientid;
1673 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1674 * We get here on a DRC miss.
1679 conf = find_confirmed_client(clid);
1680 unconf = find_unconfirmed_client(clid);
1682 status = nfserr_clid_inuse;
1683 if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1685 if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1689 * section 14.2.34 of RFC 3530 has a description of
1690 * SETCLIENTID_CONFIRM request processing consisting
1691 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1693 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1695 * RFC 3530 14.2.34 CASE 1:
1698 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1699 status = nfserr_clid_inuse;
1701 /* XXX: We just turn off callbacks until we can handle
1702 * change request correctly. */
1703 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1704 expire_client(unconf);
1708 } else if (conf && !unconf) {
1710 * RFC 3530 14.2.34 CASE 2:
1711 * probable retransmitted request; play it safe and
1714 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1715 status = nfserr_clid_inuse;
1718 } else if (!conf && unconf
1719 && same_verf(&unconf->cl_confirm, &confirm)) {
1721 * RFC 3530 14.2.34 CASE 3:
1722 * Normal case; new or rebooted client:
1724 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1725 status = nfserr_clid_inuse;
1728 clientstr_hashval(unconf->cl_recdir);
1729 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1732 nfsd4_remove_clid_dir(conf);
1733 expire_client(conf);
1735 move_to_confirmed(unconf);
1737 nfsd4_probe_callback(conf);
1740 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1741 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1744 * RFC 3530 14.2.34 CASE 4:
1745 * Client probably hasn't noticed that we rebooted yet.
1747 status = nfserr_stale_clientid;
1749 /* check that we have hit one of the cases...*/
1750 status = nfserr_clid_inuse;
1753 nfs4_unlock_state();
1757 /* OPEN Share state helper functions */
1758 static inline struct nfs4_file *
1759 alloc_init_file(struct inode *ino)
1761 struct nfs4_file *fp;
1762 unsigned int hashval = file_hashval(ino);
1764 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1766 atomic_set(&fp->fi_ref, 1);
1767 INIT_LIST_HEAD(&fp->fi_hash);
1768 INIT_LIST_HEAD(&fp->fi_stateids);
1769 INIT_LIST_HEAD(&fp->fi_delegations);
1770 spin_lock(&recall_lock);
1771 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1772 spin_unlock(&recall_lock);
1773 fp->fi_inode = igrab(ino);
1774 fp->fi_id = current_fileid++;
1775 fp->fi_had_conflict = false;
1782 nfsd4_free_slab(struct kmem_cache **slab)
1786 kmem_cache_destroy(*slab);
1791 nfsd4_free_slabs(void)
1793 nfsd4_free_slab(&stateowner_slab);
1794 nfsd4_free_slab(&file_slab);
1795 nfsd4_free_slab(&stateid_slab);
1796 nfsd4_free_slab(&deleg_slab);
1800 nfsd4_init_slabs(void)
1802 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1803 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1804 if (stateowner_slab == NULL)
1806 file_slab = kmem_cache_create("nfsd4_files",
1807 sizeof(struct nfs4_file), 0, 0, NULL);
1808 if (file_slab == NULL)
1810 stateid_slab = kmem_cache_create("nfsd4_stateids",
1811 sizeof(struct nfs4_stateid), 0, 0, NULL);
1812 if (stateid_slab == NULL)
1814 deleg_slab = kmem_cache_create("nfsd4_delegations",
1815 sizeof(struct nfs4_delegation), 0, 0, NULL);
1816 if (deleg_slab == NULL)
1821 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1826 nfs4_free_stateowner(struct kref *kref)
1828 struct nfs4_stateowner *sop =
1829 container_of(kref, struct nfs4_stateowner, so_ref);
1830 kfree(sop->so_owner.data);
1831 kmem_cache_free(stateowner_slab, sop);
1834 static inline struct nfs4_stateowner *
1835 alloc_stateowner(struct xdr_netobj *owner)
1837 struct nfs4_stateowner *sop;
1839 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1840 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1841 memcpy(sop->so_owner.data, owner->data, owner->len);
1842 sop->so_owner.len = owner->len;
1843 kref_init(&sop->so_ref);
1846 kmem_cache_free(stateowner_slab, sop);
1851 static struct nfs4_stateowner *
1852 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1853 struct nfs4_stateowner *sop;
1854 struct nfs4_replay *rp;
1855 unsigned int idhashval;
1857 if (!(sop = alloc_stateowner(&open->op_owner)))
1859 idhashval = ownerid_hashval(current_ownerid);
1860 INIT_LIST_HEAD(&sop->so_idhash);
1861 INIT_LIST_HEAD(&sop->so_strhash);
1862 INIT_LIST_HEAD(&sop->so_perclient);
1863 INIT_LIST_HEAD(&sop->so_stateids);
1864 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1865 INIT_LIST_HEAD(&sop->so_close_lru);
1867 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1868 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1869 list_add(&sop->so_perclient, &clp->cl_openowners);
1870 sop->so_is_open_owner = 1;
1871 sop->so_id = current_ownerid++;
1872 sop->so_client = clp;
1873 sop->so_seqid = open->op_seqid;
1874 sop->so_confirmed = 0;
1875 rp = &sop->so_replay;
1876 rp->rp_status = nfserr_serverfault;
1878 rp->rp_buf = rp->rp_ibuf;
1883 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1884 struct nfs4_stateowner *sop = open->op_stateowner;
1885 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1887 INIT_LIST_HEAD(&stp->st_hash);
1888 INIT_LIST_HEAD(&stp->st_perstateowner);
1889 INIT_LIST_HEAD(&stp->st_lockowners);
1890 INIT_LIST_HEAD(&stp->st_perfile);
1891 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1892 list_add(&stp->st_perstateowner, &sop->so_stateids);
1893 list_add(&stp->st_perfile, &fp->fi_stateids);
1894 stp->st_stateowner = sop;
1897 stp->st_stateid.si_boot = get_seconds();
1898 stp->st_stateid.si_stateownerid = sop->so_id;
1899 stp->st_stateid.si_fileid = fp->fi_id;
1900 stp->st_stateid.si_generation = 0;
1901 stp->st_access_bmap = 0;
1902 stp->st_deny_bmap = 0;
1903 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1904 &stp->st_access_bmap);
1905 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1906 stp->st_openstp = NULL;
1910 move_to_close_lru(struct nfs4_stateowner *sop)
1912 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1914 list_move_tail(&sop->so_close_lru, &close_lru);
1915 sop->so_time = get_seconds();
1919 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1922 return (sop->so_owner.len == owner->len) &&
1923 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1924 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1927 static struct nfs4_stateowner *
1928 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1930 struct nfs4_stateowner *so = NULL;
1932 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1933 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1939 /* search file_hashtbl[] for file */
1940 static struct nfs4_file *
1941 find_file(struct inode *ino)
1943 unsigned int hashval = file_hashval(ino);
1944 struct nfs4_file *fp;
1946 spin_lock(&recall_lock);
1947 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1948 if (fp->fi_inode == ino) {
1950 spin_unlock(&recall_lock);
1954 spin_unlock(&recall_lock);
1958 static inline int access_valid(u32 x, u32 minorversion)
1960 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1962 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1964 x &= ~NFS4_SHARE_ACCESS_MASK;
1965 if (minorversion && x) {
1966 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1968 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1970 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1977 static inline int deny_valid(u32 x)
1979 /* Note: unlike access bits, deny bits may be zero. */
1980 return x <= NFS4_SHARE_DENY_BOTH;
1984 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1985 * st_{access,deny}_bmap field of the stateid, in order to track not
1986 * only what share bits are currently in force, but also what
1987 * combinations of share bits previous opens have used. This allows us
1988 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1989 * return an error if the client attempt to downgrade to a combination
1990 * of share bits not explicable by closing some of its previous opens.
1992 * XXX: This enforcement is actually incomplete, since we don't keep
1993 * track of access/deny bit combinations; so, e.g., we allow:
1995 * OPEN allow read, deny write
1996 * OPEN allow both, deny none
1997 * DOWNGRADE allow read, deny none
1999 * which we should reject.
2002 set_access(unsigned int *access, unsigned long bmap) {
2006 for (i = 1; i < 4; i++) {
2007 if (test_bit(i, &bmap))
2013 set_deny(unsigned int *deny, unsigned long bmap) {
2017 for (i = 0; i < 4; i++) {
2018 if (test_bit(i, &bmap))
2024 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2025 unsigned int access, deny;
2027 set_access(&access, stp->st_access_bmap);
2028 set_deny(&deny, stp->st_deny_bmap);
2029 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2035 * Called to check deny when READ with all zero stateid or
2036 * WRITE with all zero or all one stateid
2039 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2041 struct inode *ino = current_fh->fh_dentry->d_inode;
2042 struct nfs4_file *fp;
2043 struct nfs4_stateid *stp;
2046 dprintk("NFSD: nfs4_share_conflict\n");
2048 fp = find_file(ino);
2051 ret = nfserr_locked;
2052 /* Search for conflicting share reservations */
2053 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2054 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2055 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2065 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2067 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2068 drop_file_write_access(filp);
2069 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
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;
2086 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2090 /* We're assuming the state code never drops its reference
2091 * without first removing the lease. Since we're in this lease
2092 * callback (and since the lease code is serialized by the kernel
2093 * lock) we know the server hasn't removed the lease yet, we know
2094 * it's safe to take a reference: */
2095 atomic_inc(&dp->dl_count);
2096 atomic_inc(&dp->dl_client->cl_count);
2098 spin_lock(&recall_lock);
2099 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2100 spin_unlock(&recall_lock);
2102 /* only place dl_time is set. protected by lock_kernel*/
2103 dp->dl_time = get_seconds();
2106 * We don't want the locks code to timeout the lease for us;
2107 * we'll remove it ourself if the delegation isn't returned
2110 fl->fl_break_time = 0;
2112 dp->dl_file->fi_had_conflict = true;
2113 nfsd4_cb_recall(dp);
2117 * The file_lock is being reapd.
2119 * Called by locks_free_lock() with lock_kernel() held.
2122 void nfsd_release_deleg_cb(struct file_lock *fl)
2124 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2126 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2128 if (!(fl->fl_flags & FL_LEASE) || !dp)
2130 dp->dl_flock = NULL;
2134 * Set the delegation file_lock back pointer.
2136 * Called from setlease() with lock_kernel() held.
2139 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2141 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2143 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2150 * Called from setlease() with lock_kernel() held
2153 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2155 struct nfs4_delegation *onlistd =
2156 (struct nfs4_delegation *)onlist->fl_owner;
2157 struct nfs4_delegation *tryd =
2158 (struct nfs4_delegation *)try->fl_owner;
2160 if (onlist->fl_lmops != try->fl_lmops)
2163 return onlistd->dl_client == tryd->dl_client;
2168 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2171 return lease_modify(onlist, arg);
2176 static struct lock_manager_operations nfsd_lease_mng_ops = {
2177 .fl_break = nfsd_break_deleg_cb,
2178 .fl_release_private = nfsd_release_deleg_cb,
2179 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2180 .fl_mylease = nfsd_same_client_deleg_cb,
2181 .fl_change = nfsd_change_deleg_cb,
2186 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2187 struct nfsd4_open *open)
2189 clientid_t *clientid = &open->op_clientid;
2190 struct nfs4_client *clp = NULL;
2191 unsigned int strhashval;
2192 struct nfs4_stateowner *sop = NULL;
2194 if (!check_name(open->op_owner))
2195 return nfserr_inval;
2197 if (STALE_CLIENTID(&open->op_clientid))
2198 return nfserr_stale_clientid;
2200 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2201 sop = find_openstateowner_str(strhashval, open);
2202 open->op_stateowner = sop;
2204 /* Make sure the client's lease hasn't expired. */
2205 clp = find_confirmed_client(clientid);
2207 return nfserr_expired;
2210 /* When sessions are used, skip open sequenceid processing */
2211 if (nfsd4_has_session(cstate))
2213 if (!sop->so_confirmed) {
2214 /* Replace unconfirmed owners without checking for replay. */
2215 clp = sop->so_client;
2216 release_openowner(sop);
2217 open->op_stateowner = NULL;
2220 if (open->op_seqid == sop->so_seqid - 1) {
2221 if (sop->so_replay.rp_buflen)
2222 return nfserr_replay_me;
2223 /* The original OPEN failed so spectacularly
2224 * that we don't even have replay data saved!
2225 * Therefore, we have no choice but to continue
2226 * processing this OPEN; presumably, we'll
2227 * fail again for the same reason.
2229 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2232 if (open->op_seqid != sop->so_seqid)
2233 return nfserr_bad_seqid;
2235 if (open->op_stateowner == NULL) {
2236 sop = alloc_init_open_stateowner(strhashval, clp, open);
2238 return nfserr_resource;
2239 open->op_stateowner = sop;
2241 list_del_init(&sop->so_close_lru);
2242 renew_client(sop->so_client);
2246 static inline __be32
2247 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2249 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2250 return nfserr_openmode;
2255 static struct nfs4_delegation *
2256 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2258 struct nfs4_delegation *dp;
2260 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2261 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2268 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2269 struct nfs4_delegation **dp)
2272 __be32 status = nfserr_bad_stateid;
2274 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2277 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2278 RD_STATE : WR_STATE;
2279 status = nfs4_check_delegmode(*dp, flags);
2283 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2287 open->op_stateowner->so_confirmed = 1;
2292 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2294 struct nfs4_stateid *local;
2295 __be32 status = nfserr_share_denied;
2296 struct nfs4_stateowner *sop = open->op_stateowner;
2298 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2299 /* ignore lock owners */
2300 if (local->st_stateowner->so_is_open_owner == 0)
2302 /* remember if we have seen this open owner */
2303 if (local->st_stateowner == sop)
2305 /* check for conflicting share reservations */
2306 if (!test_share(local, open))
2314 static inline struct nfs4_stateid *
2315 nfs4_alloc_stateid(void)
2317 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2321 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2322 struct nfs4_delegation *dp,
2323 struct svc_fh *cur_fh, int flags)
2325 struct nfs4_stateid *stp;
2327 stp = nfs4_alloc_stateid();
2329 return nfserr_resource;
2332 get_file(dp->dl_vfs_file);
2333 stp->st_vfs_file = dp->dl_vfs_file;
2336 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2339 if (status == nfserr_dropit)
2340 status = nfserr_jukebox;
2341 kmem_cache_free(stateid_slab, stp);
2349 static inline __be32
2350 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2351 struct nfsd4_open *open)
2353 struct iattr iattr = {
2354 .ia_valid = ATTR_SIZE,
2357 if (!open->op_truncate)
2359 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2360 return nfserr_inval;
2361 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2365 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2367 struct file *filp = stp->st_vfs_file;
2368 struct inode *inode = filp->f_path.dentry->d_inode;
2369 unsigned int share_access, new_writer;
2372 set_access(&share_access, stp->st_access_bmap);
2373 new_writer = (~share_access) & open->op_share_access
2374 & NFS4_SHARE_ACCESS_WRITE;
2377 int err = get_write_access(inode);
2379 return nfserrno(err);
2380 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2382 return nfserrno(err);
2383 file_take_write(filp);
2385 status = nfsd4_truncate(rqstp, cur_fh, open);
2388 put_write_access(inode);
2391 /* remember the open */
2392 filp->f_mode |= open->op_share_access;
2393 __set_bit(open->op_share_access, &stp->st_access_bmap);
2394 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2401 nfs4_set_claim_prev(struct nfsd4_open *open)
2403 open->op_stateowner->so_confirmed = 1;
2404 open->op_stateowner->so_client->cl_firststate = 1;
2408 * Attempt to hand out a delegation.
2411 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2413 struct nfs4_delegation *dp;
2414 struct nfs4_stateowner *sop = stp->st_stateowner;
2415 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2416 struct file_lock fl, *flp = &fl;
2417 int status, flag = 0;
2419 flag = NFS4_OPEN_DELEGATE_NONE;
2420 open->op_recall = 0;
2421 switch (open->op_claim_type) {
2422 case NFS4_OPEN_CLAIM_PREVIOUS:
2423 if (!atomic_read(&cb->cb_set))
2424 open->op_recall = 1;
2425 flag = open->op_delegate_type;
2426 if (flag == NFS4_OPEN_DELEGATE_NONE)
2429 case NFS4_OPEN_CLAIM_NULL:
2430 /* Let's not give out any delegations till everyone's
2431 * had the chance to reclaim theirs.... */
2432 if (locks_in_grace())
2434 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2436 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2437 flag = NFS4_OPEN_DELEGATE_WRITE;
2439 flag = NFS4_OPEN_DELEGATE_READ;
2445 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2447 flag = NFS4_OPEN_DELEGATE_NONE;
2450 locks_init_lock(&fl);
2451 fl.fl_lmops = &nfsd_lease_mng_ops;
2452 fl.fl_flags = FL_LEASE;
2453 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2454 fl.fl_end = OFFSET_MAX;
2455 fl.fl_owner = (fl_owner_t)dp;
2456 fl.fl_file = stp->st_vfs_file;
2457 fl.fl_pid = current->tgid;
2459 /* vfs_setlease checks to see if delegation should be handed out.
2460 * the lock_manager callbacks fl_mylease and fl_change are used
2462 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2463 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2464 unhash_delegation(dp);
2465 flag = NFS4_OPEN_DELEGATE_NONE;
2469 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2471 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2472 dp->dl_stateid.si_boot,
2473 dp->dl_stateid.si_stateownerid,
2474 dp->dl_stateid.si_fileid,
2475 dp->dl_stateid.si_generation);
2477 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2478 && flag == NFS4_OPEN_DELEGATE_NONE
2479 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2480 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2481 open->op_delegate_type = flag;
2485 * called with nfs4_lock_state() held.
2488 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2490 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2491 struct nfs4_file *fp = NULL;
2492 struct inode *ino = current_fh->fh_dentry->d_inode;
2493 struct nfs4_stateid *stp = NULL;
2494 struct nfs4_delegation *dp = NULL;
2497 status = nfserr_inval;
2498 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2499 || !deny_valid(open->op_share_deny))
2502 * Lookup file; if found, lookup stateid and check open request,
2503 * and check for delegations in the process of being recalled.
2504 * If not found, create the nfs4_file struct
2506 fp = find_file(ino);
2508 if ((status = nfs4_check_open(fp, open, &stp)))
2510 status = nfs4_check_deleg(fp, open, &dp);
2514 status = nfserr_bad_stateid;
2515 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2517 status = nfserr_resource;
2518 fp = alloc_init_file(ino);
2524 * OPEN the file, or upgrade an existing OPEN.
2525 * If truncate fails, the OPEN fails.
2528 /* Stateid was found, this is an OPEN upgrade */
2529 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2532 update_stateid(&stp->st_stateid);
2534 /* Stateid was not found, this is a new OPEN */
2536 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2537 flags |= NFSD_MAY_READ;
2538 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2539 flags |= NFSD_MAY_WRITE;
2540 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2543 init_stateid(stp, fp, open);
2544 status = nfsd4_truncate(rqstp, current_fh, open);
2546 release_open_stateid(stp);
2549 if (nfsd4_has_session(&resp->cstate))
2550 update_stateid(&stp->st_stateid);
2552 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2554 if (nfsd4_has_session(&resp->cstate))
2555 open->op_stateowner->so_confirmed = 1;
2558 * Attempt to hand out a delegation. No error return, because the
2559 * OPEN succeeds even if we fail.
2561 nfs4_open_delegation(current_fh, open, stp);
2565 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2566 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2567 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2571 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2572 nfs4_set_claim_prev(open);
2574 * To finish the open response, we just need to set the rflags.
2576 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2577 if (!open->op_stateowner->so_confirmed &&
2578 !nfsd4_has_session(&resp->cstate))
2579 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2585 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2588 struct nfs4_client *clp;
2592 dprintk("process_renew(%08x/%08x): starting\n",
2593 clid->cl_boot, clid->cl_id);
2594 status = nfserr_stale_clientid;
2595 if (STALE_CLIENTID(clid))
2597 clp = find_confirmed_client(clid);
2598 status = nfserr_expired;
2600 /* We assume the client took too long to RENEW. */
2601 dprintk("nfsd4_renew: clientid not found!\n");
2605 status = nfserr_cb_path_down;
2606 if (!list_empty(&clp->cl_delegations)
2607 && !atomic_read(&clp->cl_cb_conn.cb_set))
2611 nfs4_unlock_state();
2615 struct lock_manager nfsd4_manager = {
2619 nfsd4_end_grace(void)
2621 dprintk("NFSD: end of grace period\n");
2622 nfsd4_recdir_purge_old();
2623 locks_end_grace(&nfsd4_manager);
2627 nfs4_laundromat(void)
2629 struct nfs4_client *clp;
2630 struct nfs4_stateowner *sop;
2631 struct nfs4_delegation *dp;
2632 struct list_head *pos, *next, reaplist;
2633 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2634 time_t t, clientid_val = NFSD_LEASE_TIME;
2635 time_t u, test_val = NFSD_LEASE_TIME;
2639 dprintk("NFSD: laundromat service - starting\n");
2640 if (locks_in_grace())
2642 list_for_each_safe(pos, next, &client_lru) {
2643 clp = list_entry(pos, struct nfs4_client, cl_lru);
2644 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2645 t = clp->cl_time - cutoff;
2646 if (clientid_val > t)
2650 dprintk("NFSD: purging unused client (clientid %08x)\n",
2651 clp->cl_clientid.cl_id);
2652 nfsd4_remove_clid_dir(clp);
2655 INIT_LIST_HEAD(&reaplist);
2656 spin_lock(&recall_lock);
2657 list_for_each_safe(pos, next, &del_recall_lru) {
2658 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2659 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2660 u = dp->dl_time - cutoff;
2665 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2667 list_move(&dp->dl_recall_lru, &reaplist);
2669 spin_unlock(&recall_lock);
2670 list_for_each_safe(pos, next, &reaplist) {
2671 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2672 list_del_init(&dp->dl_recall_lru);
2673 unhash_delegation(dp);
2675 test_val = NFSD_LEASE_TIME;
2676 list_for_each_safe(pos, next, &close_lru) {
2677 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2678 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2679 u = sop->so_time - cutoff;
2684 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2686 release_openowner(sop);
2688 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2689 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2690 nfs4_unlock_state();
2691 return clientid_val;
2694 static struct workqueue_struct *laundry_wq;
2695 static void laundromat_main(struct work_struct *);
2696 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2699 laundromat_main(struct work_struct *not_used)
2703 t = nfs4_laundromat();
2704 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2705 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2708 static struct nfs4_stateowner *
2709 search_close_lru(u32 st_id, int flags)
2711 struct nfs4_stateowner *local = NULL;
2713 if (flags & CLOSE_STATE) {
2714 list_for_each_entry(local, &close_lru, so_close_lru) {
2715 if (local->so_id == st_id)
2723 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2725 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2729 STALE_STATEID(stateid_t *stateid)
2731 if (time_after((unsigned long)boot_time,
2732 (unsigned long)stateid->si_boot)) {
2733 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2734 stateid->si_boot, stateid->si_stateownerid,
2735 stateid->si_fileid, stateid->si_generation);
2742 EXPIRED_STATEID(stateid_t *stateid)
2744 if (time_before((unsigned long)boot_time,
2745 ((unsigned long)stateid->si_boot)) &&
2746 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2747 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2748 stateid->si_boot, stateid->si_stateownerid,
2749 stateid->si_fileid, stateid->si_generation);
2756 stateid_error_map(stateid_t *stateid)
2758 if (STALE_STATEID(stateid))
2759 return nfserr_stale_stateid;
2760 if (EXPIRED_STATEID(stateid))
2761 return nfserr_expired;
2763 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2764 stateid->si_boot, stateid->si_stateownerid,
2765 stateid->si_fileid, stateid->si_generation);
2766 return nfserr_bad_stateid;
2770 access_permit_read(unsigned long access_bmap)
2772 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2773 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2774 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2778 access_permit_write(unsigned long access_bmap)
2780 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2781 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2785 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2787 __be32 status = nfserr_openmode;
2789 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2791 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2798 static inline __be32
2799 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2801 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2803 else if (locks_in_grace()) {
2804 /* Answer in remaining cases depends on existance of
2805 * conflicting state; so we must wait out the grace period. */
2806 return nfserr_grace;
2807 } else if (flags & WR_STATE)
2808 return nfs4_share_conflict(current_fh,
2809 NFS4_SHARE_DENY_WRITE);
2810 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2811 return nfs4_share_conflict(current_fh,
2812 NFS4_SHARE_DENY_READ);
2816 * Allow READ/WRITE during grace period on recovered state only for files
2817 * that are not able to provide mandatory locking.
2820 grace_disallows_io(struct inode *inode)
2822 return locks_in_grace() && mandatory_lock(inode);
2825 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2828 * When sessions are used the stateid generation number is ignored
2831 if ((flags & HAS_SESSION) && in->si_generation == 0)
2834 /* If the client sends us a stateid from the future, it's buggy: */
2835 if (in->si_generation > ref->si_generation)
2836 return nfserr_bad_stateid;
2838 * The following, however, can happen. For example, if the
2839 * client sends an open and some IO at the same time, the open
2840 * may bump si_generation while the IO is still in flight.
2841 * Thanks to hard links and renames, the client never knows what
2842 * file an open will affect. So it could avoid that situation
2843 * only by serializing all opens and IO from the same open
2844 * owner. To recover from the old_stateid error, the client
2845 * will just have to retry the IO:
2847 if (in->si_generation < ref->si_generation)
2848 return nfserr_old_stateid;
2853 static int is_delegation_stateid(stateid_t *stateid)
2855 return stateid->si_fileid == 0;
2859 * Checks for stateid operations
2862 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2863 stateid_t *stateid, int flags, struct file **filpp)
2865 struct nfs4_stateid *stp = NULL;
2866 struct nfs4_delegation *dp = NULL;
2867 struct svc_fh *current_fh = &cstate->current_fh;
2868 struct inode *ino = current_fh->fh_dentry->d_inode;
2874 if (grace_disallows_io(ino))
2875 return nfserr_grace;
2877 if (nfsd4_has_session(cstate))
2878 flags |= HAS_SESSION;
2880 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2881 return check_special_stateids(current_fh, stateid, flags);
2883 status = nfserr_stale_stateid;
2884 if (STALE_STATEID(stateid))
2887 status = nfserr_bad_stateid;
2888 if (is_delegation_stateid(stateid)) {
2889 dp = find_delegation_stateid(ino, stateid);
2891 status = stateid_error_map(stateid);
2894 status = check_stateid_generation(stateid, &dp->dl_stateid,
2898 status = nfs4_check_delegmode(dp, flags);
2901 renew_client(dp->dl_client);
2903 *filpp = dp->dl_vfs_file;
2904 } else { /* open or lock stateid */
2905 stp = find_stateid(stateid, flags);
2907 status = stateid_error_map(stateid);
2910 if (nfs4_check_fh(current_fh, stp))
2912 if (!stp->st_stateowner->so_confirmed)
2914 status = check_stateid_generation(stateid, &stp->st_stateid,
2918 status = nfs4_check_openmode(stp, flags);
2921 renew_client(stp->st_stateowner->so_client);
2923 *filpp = stp->st_vfs_file;
2933 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2934 RD_STATE : WR_STATE;
2938 * Checks for sequence id mutating operations.
2941 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2942 stateid_t *stateid, int flags,
2943 struct nfs4_stateowner **sopp,
2944 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2946 struct nfs4_stateid *stp;
2947 struct nfs4_stateowner *sop;
2948 struct svc_fh *current_fh = &cstate->current_fh;
2951 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2952 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2953 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2954 stateid->si_generation);
2959 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2960 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2961 return nfserr_bad_stateid;
2964 if (STALE_STATEID(stateid))
2965 return nfserr_stale_stateid;
2967 if (nfsd4_has_session(cstate))
2968 flags |= HAS_SESSION;
2971 * We return BAD_STATEID if filehandle doesn't match stateid,
2972 * the confirmed flag is incorrecly set, or the generation
2973 * number is incorrect.
2975 stp = find_stateid(stateid, flags);
2978 * Also, we should make sure this isn't just the result of
2981 sop = search_close_lru(stateid->si_stateownerid, flags);
2983 return stateid_error_map(stateid);
2989 *sopp = sop = stp->st_stateowner;
2992 clientid_t *lockclid = &lock->v.new.clientid;
2993 struct nfs4_client *clp = sop->so_client;
2997 lkflg = setlkflg(lock->lk_type);
2999 if (lock->lk_is_new) {
3000 if (!sop->so_is_open_owner)
3001 return nfserr_bad_stateid;
3002 if (!(flags & HAS_SESSION) &&
3003 !same_clid(&clp->cl_clientid, lockclid))
3004 return nfserr_bad_stateid;
3005 /* stp is the open stateid */
3006 status = nfs4_check_openmode(stp, lkflg);
3010 /* stp is the lock stateid */
3011 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3017 if (nfs4_check_fh(current_fh, stp)) {
3018 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3019 return nfserr_bad_stateid;
3023 * We now validate the seqid and stateid generation numbers.
3024 * For the moment, we ignore the possibility of
3025 * generation number wraparound.
3027 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3030 if (sop->so_confirmed && flags & CONFIRM) {
3031 dprintk("NFSD: preprocess_seqid_op: expected"
3032 " unconfirmed stateowner!\n");
3033 return nfserr_bad_stateid;
3035 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3036 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3037 " confirmed yet!\n");
3038 return nfserr_bad_stateid;
3040 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3043 renew_client(sop->so_client);
3047 if (seqid == sop->so_seqid - 1) {
3048 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3049 /* indicate replay to calling function */
3050 return nfserr_replay_me;
3052 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3053 sop->so_seqid, seqid);
3055 return nfserr_bad_seqid;
3059 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3060 struct nfsd4_open_confirm *oc)
3063 struct nfs4_stateowner *sop;
3064 struct nfs4_stateid *stp;
3066 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3067 (int)cstate->current_fh.fh_dentry->d_name.len,
3068 cstate->current_fh.fh_dentry->d_name.name);
3070 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3076 if ((status = nfs4_preprocess_seqid_op(cstate,
3077 oc->oc_seqid, &oc->oc_req_stateid,
3078 CONFIRM | OPEN_STATE,
3079 &oc->oc_stateowner, &stp, NULL)))
3082 sop = oc->oc_stateowner;
3083 sop->so_confirmed = 1;
3084 update_stateid(&stp->st_stateid);
3085 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3086 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3087 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3088 stp->st_stateid.si_boot,
3089 stp->st_stateid.si_stateownerid,
3090 stp->st_stateid.si_fileid,
3091 stp->st_stateid.si_generation);
3093 nfsd4_create_clid_dir(sop->so_client);
3095 if (oc->oc_stateowner) {
3096 nfs4_get_stateowner(oc->oc_stateowner);
3097 cstate->replay_owner = oc->oc_stateowner;
3099 nfs4_unlock_state();
3105 * unset all bits in union bitmap (bmap) that
3106 * do not exist in share (from successful OPEN_DOWNGRADE)
3109 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3112 for (i = 1; i < 4; i++) {
3113 if ((i & access) != i)
3114 __clear_bit(i, bmap);
3119 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3122 for (i = 0; i < 4; i++) {
3123 if ((i & deny) != i)
3124 __clear_bit(i, bmap);
3129 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3130 struct nfsd4_compound_state *cstate,
3131 struct nfsd4_open_downgrade *od)
3134 struct nfs4_stateid *stp;
3135 unsigned int share_access;
3137 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3138 (int)cstate->current_fh.fh_dentry->d_name.len,
3139 cstate->current_fh.fh_dentry->d_name.name);
3141 if (!access_valid(od->od_share_access, cstate->minorversion)
3142 || !deny_valid(od->od_share_deny))
3143 return nfserr_inval;
3146 if ((status = nfs4_preprocess_seqid_op(cstate,
3150 &od->od_stateowner, &stp, NULL)))
3153 status = nfserr_inval;
3154 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3155 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3156 stp->st_access_bmap, od->od_share_access);
3159 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3160 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3161 stp->st_deny_bmap, od->od_share_deny);
3164 set_access(&share_access, stp->st_access_bmap);
3165 nfs4_file_downgrade(stp->st_vfs_file,
3166 share_access & ~od->od_share_access);
3168 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3169 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3171 update_stateid(&stp->st_stateid);
3172 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3175 if (od->od_stateowner) {
3176 nfs4_get_stateowner(od->od_stateowner);
3177 cstate->replay_owner = od->od_stateowner;
3179 nfs4_unlock_state();
3184 * nfs4_unlock_state() called after encode
3187 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3188 struct nfsd4_close *close)
3191 struct nfs4_stateid *stp;
3193 dprintk("NFSD: nfsd4_close on file %.*s\n",
3194 (int)cstate->current_fh.fh_dentry->d_name.len,
3195 cstate->current_fh.fh_dentry->d_name.name);
3198 /* check close_lru for replay */
3199 if ((status = nfs4_preprocess_seqid_op(cstate,
3202 OPEN_STATE | CLOSE_STATE,
3203 &close->cl_stateowner, &stp, NULL)))
3206 update_stateid(&stp->st_stateid);
3207 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3209 /* release_stateid() calls nfsd_close() if needed */
3210 release_open_stateid(stp);
3212 /* place unused nfs4_stateowners on so_close_lru list to be
3213 * released by the laundromat service after the lease period
3214 * to enable us to handle CLOSE replay
3216 if (list_empty(&close->cl_stateowner->so_stateids))
3217 move_to_close_lru(close->cl_stateowner);
3219 if (close->cl_stateowner) {
3220 nfs4_get_stateowner(close->cl_stateowner);
3221 cstate->replay_owner = close->cl_stateowner;
3223 nfs4_unlock_state();
3228 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3229 struct nfsd4_delegreturn *dr)
3231 struct nfs4_delegation *dp;
3232 stateid_t *stateid = &dr->dr_stateid;
3233 struct inode *inode;
3237 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3239 inode = cstate->current_fh.fh_dentry->d_inode;
3241 if (nfsd4_has_session(cstate))
3242 flags |= HAS_SESSION;
3244 status = nfserr_bad_stateid;
3245 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3247 status = nfserr_stale_stateid;
3248 if (STALE_STATEID(stateid))
3250 status = nfserr_bad_stateid;
3251 if (!is_delegation_stateid(stateid))
3253 dp = find_delegation_stateid(inode, stateid);
3255 status = stateid_error_map(stateid);
3258 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3261 renew_client(dp->dl_client);
3263 unhash_delegation(dp);
3265 nfs4_unlock_state();
3272 * Lock owner state (byte-range locks)
3274 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3275 #define LOCK_HASH_BITS 8
3276 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3277 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3280 end_offset(u64 start, u64 len)
3285 return end >= start ? end: NFS4_MAX_UINT64;
3288 /* last octet in a range */
3290 last_byte_offset(u64 start, u64 len)
3296 return end > start ? end - 1: NFS4_MAX_UINT64;
3299 #define lockownerid_hashval(id) \
3300 ((id) & LOCK_HASH_MASK)
3302 static inline unsigned int
3303 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3304 struct xdr_netobj *ownername)
3306 return (file_hashval(inode) + cl_id
3307 + opaque_hashval(ownername->data, ownername->len))
3311 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3312 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3313 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3315 static struct nfs4_stateid *
3316 find_stateid(stateid_t *stid, int flags)
3318 struct nfs4_stateid *local;
3319 u32 st_id = stid->si_stateownerid;
3320 u32 f_id = stid->si_fileid;
3321 unsigned int hashval;
3323 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3324 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3325 hashval = stateid_hashval(st_id, f_id);
3326 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3327 if ((local->st_stateid.si_stateownerid == st_id) &&
3328 (local->st_stateid.si_fileid == f_id))
3333 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3334 hashval = stateid_hashval(st_id, f_id);
3335 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3336 if ((local->st_stateid.si_stateownerid == st_id) &&
3337 (local->st_stateid.si_fileid == f_id))
3344 static struct nfs4_delegation *
3345 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3347 struct nfs4_file *fp;
3348 struct nfs4_delegation *dl;
3350 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3351 stid->si_boot, stid->si_stateownerid,
3352 stid->si_fileid, stid->si_generation);
3354 fp = find_file(ino);
3357 dl = find_delegation_file(fp, stid);
3363 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3364 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3365 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3366 * locking, this prevents us from being completely protocol-compliant. The
3367 * real solution to this problem is to start using unsigned file offsets in
3368 * the VFS, but this is a very deep change!
3371 nfs4_transform_lock_offset(struct file_lock *lock)
3373 if (lock->fl_start < 0)
3374 lock->fl_start = OFFSET_MAX;
3375 if (lock->fl_end < 0)
3376 lock->fl_end = OFFSET_MAX;
3379 /* Hack!: For now, we're defining this just so we can use a pointer to it
3380 * as a unique cookie to identify our (NFSv4's) posix locks. */
3381 static struct lock_manager_operations nfsd_posix_mng_ops = {
3385 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3387 struct nfs4_stateowner *sop;
3390 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3391 sop = (struct nfs4_stateowner *) fl->fl_owner;
3392 hval = lockownerid_hashval(sop->so_id);
3393 kref_get(&sop->so_ref);
3395 deny->ld_clientid = sop->so_client->cl_clientid;
3397 deny->ld_sop = NULL;
3398 deny->ld_clientid.cl_boot = 0;
3399 deny->ld_clientid.cl_id = 0;
3401 deny->ld_start = fl->fl_start;
3402 deny->ld_length = NFS4_MAX_UINT64;
3403 if (fl->fl_end != NFS4_MAX_UINT64)
3404 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3405 deny->ld_type = NFS4_READ_LT;
3406 if (fl->fl_type != F_RDLCK)
3407 deny->ld_type = NFS4_WRITE_LT;
3410 static struct nfs4_stateowner *
3411 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3412 struct xdr_netobj *owner)
3414 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3415 struct nfs4_stateowner *op;
3417 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3418 if (same_owner_str(op, owner, clid))
3425 * Alloc a lock owner structure.
3426 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3429 * strhashval = lock_ownerstr_hashval
3432 static struct nfs4_stateowner *
3433 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3434 struct nfs4_stateowner *sop;
3435 struct nfs4_replay *rp;
3436 unsigned int idhashval;
3438 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3440 idhashval = lockownerid_hashval(current_ownerid);
3441 INIT_LIST_HEAD(&sop->so_idhash);
3442 INIT_LIST_HEAD(&sop->so_strhash);
3443 INIT_LIST_HEAD(&sop->so_perclient);
3444 INIT_LIST_HEAD(&sop->so_stateids);
3445 INIT_LIST_HEAD(&sop->so_perstateid);
3446 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3448 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3449 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3450 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3451 sop->so_is_open_owner = 0;
3452 sop->so_id = current_ownerid++;
3453 sop->so_client = clp;
3454 /* It is the openowner seqid that will be incremented in encode in the
3455 * case of new lockowners; so increment the lock seqid manually: */
3456 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3457 sop->so_confirmed = 1;
3458 rp = &sop->so_replay;
3459 rp->rp_status = nfserr_serverfault;
3461 rp->rp_buf = rp->rp_ibuf;
3465 static struct nfs4_stateid *
3466 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3468 struct nfs4_stateid *stp;
3469 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3471 stp = nfs4_alloc_stateid();
3474 INIT_LIST_HEAD(&stp->st_hash);
3475 INIT_LIST_HEAD(&stp->st_perfile);
3476 INIT_LIST_HEAD(&stp->st_perstateowner);
3477 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3478 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3479 list_add(&stp->st_perfile, &fp->fi_stateids);
3480 list_add(&stp->st_perstateowner, &sop->so_stateids);
3481 stp->st_stateowner = sop;
3484 stp->st_stateid.si_boot = get_seconds();
3485 stp->st_stateid.si_stateownerid = sop->so_id;
3486 stp->st_stateid.si_fileid = fp->fi_id;
3487 stp->st_stateid.si_generation = 0;
3488 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3489 stp->st_access_bmap = open_stp->st_access_bmap;
3490 stp->st_deny_bmap = open_stp->st_deny_bmap;
3491 stp->st_openstp = open_stp;
3498 check_lock_length(u64 offset, u64 length)
3500 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3501 LOFF_OVERFLOW(offset, length)));
3508 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3509 struct nfsd4_lock *lock)
3511 struct nfs4_stateowner *open_sop = NULL;
3512 struct nfs4_stateowner *lock_sop = NULL;
3513 struct nfs4_stateid *lock_stp;
3515 struct file_lock file_lock;
3516 struct file_lock conflock;
3518 unsigned int strhashval;
3522 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3523 (long long) lock->lk_offset,
3524 (long long) lock->lk_length);
3526 if (check_lock_length(lock->lk_offset, lock->lk_length))
3527 return nfserr_inval;
3529 if ((status = fh_verify(rqstp, &cstate->current_fh,
3530 S_IFREG, NFSD_MAY_LOCK))) {
3531 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3537 if (lock->lk_is_new) {
3539 * Client indicates that this is a new lockowner.
3540 * Use open owner and open stateid to create lock owner and
3543 struct nfs4_stateid *open_stp = NULL;
3544 struct nfs4_file *fp;
3546 status = nfserr_stale_clientid;
3547 if (!nfsd4_has_session(cstate) &&
3548 STALE_CLIENTID(&lock->lk_new_clientid))
3551 /* validate and update open stateid and open seqid */
3552 status = nfs4_preprocess_seqid_op(cstate,
3553 lock->lk_new_open_seqid,
3554 &lock->lk_new_open_stateid,
3556 &lock->lk_replay_owner, &open_stp,
3560 open_sop = lock->lk_replay_owner;
3561 /* create lockowner and lock stateid */
3562 fp = open_stp->st_file;
3563 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3564 open_sop->so_client->cl_clientid.cl_id,
3565 &lock->v.new.owner);
3566 /* XXX: Do we need to check for duplicate stateowners on
3567 * the same file, or should they just be allowed (and
3568 * create new stateids)? */
3569 status = nfserr_resource;
3570 lock_sop = alloc_init_lock_stateowner(strhashval,
3571 open_sop->so_client, open_stp, lock);
3572 if (lock_sop == NULL)
3574 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3575 if (lock_stp == NULL)
3578 /* lock (lock owner + lock stateid) already exists */
3579 status = nfs4_preprocess_seqid_op(cstate,
3580 lock->lk_old_lock_seqid,
3581 &lock->lk_old_lock_stateid,
3583 &lock->lk_replay_owner, &lock_stp, lock);
3586 lock_sop = lock->lk_replay_owner;
3588 /* lock->lk_replay_owner and lock_stp have been created or found */
3589 filp = lock_stp->st_vfs_file;
3591 status = nfserr_grace;
3592 if (locks_in_grace() && !lock->lk_reclaim)
3594 status = nfserr_no_grace;
3595 if (!locks_in_grace() && lock->lk_reclaim)
3598 locks_init_lock(&file_lock);
3599 switch (lock->lk_type) {
3602 file_lock.fl_type = F_RDLCK;
3606 case NFS4_WRITEW_LT:
3607 file_lock.fl_type = F_WRLCK;
3611 status = nfserr_inval;
3614 file_lock.fl_owner = (fl_owner_t)lock_sop;
3615 file_lock.fl_pid = current->tgid;
3616 file_lock.fl_file = filp;
3617 file_lock.fl_flags = FL_POSIX;
3618 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3620 file_lock.fl_start = lock->lk_offset;
3621 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3622 nfs4_transform_lock_offset(&file_lock);
3625 * Try to lock the file in the VFS.
3626 * Note: locks.c uses the BKL to protect the inode's lock list.
3629 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3631 case 0: /* success! */
3632 update_stateid(&lock_stp->st_stateid);
3633 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3637 case (EAGAIN): /* conflock holds conflicting lock */
3638 status = nfserr_denied;
3639 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3640 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3643 status = nfserr_deadlock;
3646 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3647 status = nfserr_resource;
3651 if (status && lock->lk_is_new && lock_sop)
3652 release_lockowner(lock_sop);
3653 if (lock->lk_replay_owner) {
3654 nfs4_get_stateowner(lock->lk_replay_owner);
3655 cstate->replay_owner = lock->lk_replay_owner;
3657 nfs4_unlock_state();
3662 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3663 * so we do a temporary open here just to get an open file to pass to
3664 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3667 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3672 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3675 err = vfs_test_lock(file, lock);
3684 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3685 struct nfsd4_lockt *lockt)
3687 struct inode *inode;
3688 struct file_lock file_lock;
3692 if (locks_in_grace())
3693 return nfserr_grace;
3695 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3696 return nfserr_inval;
3698 lockt->lt_stateowner = NULL;
3701 status = nfserr_stale_clientid;
3702 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3705 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3706 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3707 if (status == nfserr_symlink)
3708 status = nfserr_inval;
3712 inode = cstate->current_fh.fh_dentry->d_inode;
3713 locks_init_lock(&file_lock);
3714 switch (lockt->lt_type) {
3717 file_lock.fl_type = F_RDLCK;
3720 case NFS4_WRITEW_LT:
3721 file_lock.fl_type = F_WRLCK;
3724 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3725 status = nfserr_inval;
3729 lockt->lt_stateowner = find_lockstateowner_str(inode,
3730 &lockt->lt_clientid, &lockt->lt_owner);
3731 if (lockt->lt_stateowner)
3732 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3733 file_lock.fl_pid = current->tgid;
3734 file_lock.fl_flags = FL_POSIX;
3736 file_lock.fl_start = lockt->lt_offset;
3737 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3739 nfs4_transform_lock_offset(&file_lock);
3742 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3744 status = nfserrno(error);
3747 if (file_lock.fl_type != F_UNLCK) {
3748 status = nfserr_denied;
3749 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3752 nfs4_unlock_state();
3757 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3758 struct nfsd4_locku *locku)
3760 struct nfs4_stateid *stp;
3761 struct file *filp = NULL;
3762 struct file_lock file_lock;
3766 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3767 (long long) locku->lu_offset,
3768 (long long) locku->lu_length);
3770 if (check_lock_length(locku->lu_offset, locku->lu_length))
3771 return nfserr_inval;
3775 if ((status = nfs4_preprocess_seqid_op(cstate,
3779 &locku->lu_stateowner, &stp, NULL)))
3782 filp = stp->st_vfs_file;
3784 locks_init_lock(&file_lock);
3785 file_lock.fl_type = F_UNLCK;
3786 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3787 file_lock.fl_pid = current->tgid;
3788 file_lock.fl_file = filp;
3789 file_lock.fl_flags = FL_POSIX;
3790 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3791 file_lock.fl_start = locku->lu_offset;
3793 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3794 nfs4_transform_lock_offset(&file_lock);
3797 * Try to unlock the file in the VFS.
3799 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3801 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3805 * OK, unlock succeeded; the only thing left to do is update the stateid.
3807 update_stateid(&stp->st_stateid);
3808 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3811 if (locku->lu_stateowner) {
3812 nfs4_get_stateowner(locku->lu_stateowner);
3813 cstate->replay_owner = locku->lu_stateowner;
3815 nfs4_unlock_state();
3819 status = nfserrno(err);
3825 * 1: locks held by lockowner
3826 * 0: no locks held by lockowner
3829 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3831 struct file_lock **flpp;
3832 struct inode *inode = filp->f_path.dentry->d_inode;
3836 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3837 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3848 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3849 struct nfsd4_compound_state *cstate,
3850 struct nfsd4_release_lockowner *rlockowner)
3852 clientid_t *clid = &rlockowner->rl_clientid;
3853 struct nfs4_stateowner *sop;
3854 struct nfs4_stateid *stp;
3855 struct xdr_netobj *owner = &rlockowner->rl_owner;
3856 struct list_head matches;
3860 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3861 clid->cl_boot, clid->cl_id);
3863 /* XXX check for lease expiration */
3865 status = nfserr_stale_clientid;
3866 if (STALE_CLIENTID(clid))
3871 status = nfserr_locks_held;
3872 /* XXX: we're doing a linear search through all the lockowners.
3873 * Yipes! For now we'll just hope clients aren't really using
3874 * release_lockowner much, but eventually we have to fix these
3875 * data structures. */
3876 INIT_LIST_HEAD(&matches);
3877 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3878 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3879 if (!same_owner_str(sop, owner, clid))
3881 list_for_each_entry(stp, &sop->so_stateids,
3883 if (check_for_locks(stp->st_vfs_file, sop))
3885 /* Note: so_perclient unused for lockowners,
3886 * so it's OK to fool with here. */
3887 list_add(&sop->so_perclient, &matches);
3891 /* Clients probably won't expect us to return with some (but not all)
3892 * of the lockowner state released; so don't release any until all
3893 * have been checked. */
3895 while (!list_empty(&matches)) {
3896 sop = list_entry(matches.next, struct nfs4_stateowner,
3898 /* unhash_stateowner deletes so_perclient only
3899 * for openowners. */
3900 list_del(&sop->so_perclient);
3901 release_lockowner(sop);
3904 nfs4_unlock_state();
3908 static inline struct nfs4_client_reclaim *
3911 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3915 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3917 unsigned int strhashval = clientstr_hashval(name);
3918 struct nfs4_client *clp;
3920 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3925 * failure => all reset bets are off, nfserr_no_grace...
3928 nfs4_client_to_reclaim(const char *name)
3930 unsigned int strhashval;
3931 struct nfs4_client_reclaim *crp = NULL;
3933 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3934 crp = alloc_reclaim();
3937 strhashval = clientstr_hashval(name);
3938 INIT_LIST_HEAD(&crp->cr_strhash);
3939 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3940 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3941 reclaim_str_hashtbl_size++;
3946 nfs4_release_reclaim(void)
3948 struct nfs4_client_reclaim *crp = NULL;
3951 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3952 while (!list_empty(&reclaim_str_hashtbl[i])) {
3953 crp = list_entry(reclaim_str_hashtbl[i].next,
3954 struct nfs4_client_reclaim, cr_strhash);
3955 list_del(&crp->cr_strhash);
3957 reclaim_str_hashtbl_size--;
3960 BUG_ON(reclaim_str_hashtbl_size);
3964 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3965 static struct nfs4_client_reclaim *
3966 nfs4_find_reclaim_client(clientid_t *clid)
3968 unsigned int strhashval;
3969 struct nfs4_client *clp;
3970 struct nfs4_client_reclaim *crp = NULL;
3973 /* find clientid in conf_id_hashtbl */
3974 clp = find_confirmed_client(clid);
3978 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3979 clp->cl_name.len, clp->cl_name.data,
3982 /* find clp->cl_name in reclaim_str_hashtbl */
3983 strhashval = clientstr_hashval(clp->cl_recdir);
3984 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3985 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3993 * Called from OPEN. Look for clientid in reclaim list.
3996 nfs4_check_open_reclaim(clientid_t *clid)
3998 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4001 /* initialization to perform at module load time: */
4004 nfs4_state_init(void)
4008 status = nfsd4_init_slabs();
4011 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4012 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4013 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4014 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4015 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4016 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4018 for (i = 0; i < SESSION_HASH_SIZE; i++)
4019 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4020 for (i = 0; i < FILE_HASH_SIZE; i++) {
4021 INIT_LIST_HEAD(&file_hashtbl[i]);
4023 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4024 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4025 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4027 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4028 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4029 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4031 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4032 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4033 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4035 memset(&onestateid, ~0, sizeof(stateid_t));
4036 INIT_LIST_HEAD(&close_lru);
4037 INIT_LIST_HEAD(&client_lru);
4038 INIT_LIST_HEAD(&del_recall_lru);
4039 reclaim_str_hashtbl_size = 0;
4044 nfsd4_load_reboot_recovery_data(void)
4049 nfsd4_init_recdir(user_recovery_dirname);
4050 status = nfsd4_recdir_load();
4051 nfs4_unlock_state();
4053 printk("NFSD: Failure reading reboot recovery data\n");
4057 get_nfs4_grace_period(void)
4059 return max(user_lease_time, lease_time) * HZ;
4063 * Since the lifetime of a delegation isn't limited to that of an open, a
4064 * client may quite reasonably hang on to a delegation as long as it has
4065 * the inode cached. This becomes an obvious problem the first time a
4066 * client's inode cache approaches the size of the server's total memory.
4068 * For now we avoid this problem by imposing a hard limit on the number
4069 * of delegations, which varies according to the server's memory size.
4072 set_max_delegations(void)
4075 * Allow at most 4 delegations per megabyte of RAM. Quick
4076 * estimates suggest that in the worst case (where every delegation
4077 * is for a different inode), a delegation could take about 1.5K,
4078 * giving a worst case usage of about 6% of memory.
4080 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4083 /* initialization to perform when the nfsd service is started: */
4086 __nfs4_state_start(void)
4088 unsigned long grace_time;
4090 boot_time = get_seconds();
4091 grace_time = get_nfs4_grace_period();
4092 lease_time = user_lease_time;
4093 locks_start_grace(&nfsd4_manager);
4094 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4096 laundry_wq = create_singlethread_workqueue("nfsd4");
4097 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4098 set_max_delegations();
4102 nfs4_state_start(void)
4106 nfsd4_load_reboot_recovery_data();
4107 __nfs4_state_start();
4113 nfs4_lease_time(void)
4119 __nfs4_state_shutdown(void)
4122 struct nfs4_client *clp = NULL;
4123 struct nfs4_delegation *dp = NULL;
4124 struct list_head *pos, *next, reaplist;
4126 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4127 while (!list_empty(&conf_id_hashtbl[i])) {
4128 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4131 while (!list_empty(&unconf_str_hashtbl[i])) {
4132 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4136 INIT_LIST_HEAD(&reaplist);
4137 spin_lock(&recall_lock);
4138 list_for_each_safe(pos, next, &del_recall_lru) {
4139 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4140 list_move(&dp->dl_recall_lru, &reaplist);
4142 spin_unlock(&recall_lock);
4143 list_for_each_safe(pos, next, &reaplist) {
4144 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4145 list_del_init(&dp->dl_recall_lru);
4146 unhash_delegation(dp);
4149 nfsd4_shutdown_recdir();
4154 nfs4_state_shutdown(void)
4156 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4157 destroy_workqueue(laundry_wq);
4158 locks_end_grace(&nfsd4_manager);
4160 nfs4_release_reclaim();
4161 __nfs4_state_shutdown();
4162 nfs4_unlock_state();
4166 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4167 * accessed when nfsd is starting.
4170 nfs4_set_recdir(char *recdir)
4172 strcpy(user_recovery_dirname, recdir);
4176 * Change the NFSv4 recovery directory to recdir.
4179 nfs4_reset_recoverydir(char *recdir)
4184 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4188 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4189 nfs4_set_recdir(recdir);
4197 nfs4_recoverydir(void)
4199 return user_recovery_dirname;
4203 * Called when leasetime is changed.
4205 * The only way the protocol gives us to handle on-the-fly lease changes is to
4206 * simulate a reboot. Instead of doing that, we just wait till the next time
4207 * we start to register any changes in lease time. If the administrator
4208 * really wants to change the lease time *now*, they can go ahead and bring
4209 * nfsd down and then back up again after changing the lease time.
4211 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4212 * when nfsd is starting
4215 nfs4_reset_lease(time_t leasetime)
4217 user_lease_time = leasetime;