2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
59 #define NFSDDBG_FACILITY NFSDDBG_PROC
62 static time_t lease_time = 90; /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
69 static stateid_t zerostateid; /* bits all 0 */
70 static stateid_t onestateid; /* bits all 1 */
71 static u64 current_sessionid = 1;
73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
74 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
76 /* forward declarations */
77 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
78 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir);
84 /* Currently used for almost all code touching nfsv4 state: */
85 static DEFINE_MUTEX(client_mutex);
88 * Currently used for the del_recall_lru and file hash table. In an
89 * effort to decrease the scope of the client_mutex, this spinlock may
90 * eventually cover more:
92 static DEFINE_SPINLOCK(recall_lock);
94 static struct kmem_cache *stateowner_slab = NULL;
95 static struct kmem_cache *file_slab = NULL;
96 static struct kmem_cache *stateid_slab = NULL;
97 static struct kmem_cache *deleg_slab = NULL;
100 nfs4_lock_state(void)
102 mutex_lock(&client_mutex);
106 nfs4_unlock_state(void)
108 mutex_unlock(&client_mutex);
112 opaque_hashval(const void *ptr, int nbytes)
114 unsigned char *cptr = (unsigned char *) ptr;
124 static struct list_head del_recall_lru;
127 put_nfs4_file(struct nfs4_file *fi)
129 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
130 list_del(&fi->fi_hash);
131 spin_unlock(&recall_lock);
133 kmem_cache_free(file_slab, fi);
138 get_nfs4_file(struct nfs4_file *fi)
140 atomic_inc(&fi->fi_ref);
143 static int num_delegations;
144 unsigned int max_delegations;
147 * Open owner state (share locks)
150 /* hash tables for nfs4_stateowner */
151 #define OWNER_HASH_BITS 8
152 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
153 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
155 #define ownerid_hashval(id) \
156 ((id) & OWNER_HASH_MASK)
157 #define ownerstr_hashval(clientid, ownername) \
158 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
160 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
161 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
163 /* hash table for nfs4_file */
164 #define FILE_HASH_BITS 8
165 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
166 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
167 /* hash table for (open)nfs4_stateid */
168 #define STATEID_HASH_BITS 10
169 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
170 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
172 #define file_hashval(x) \
173 hash_ptr(x, FILE_HASH_BITS)
174 #define stateid_hashval(owner_id, file_id) \
175 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
177 static struct list_head file_hashtbl[FILE_HASH_SIZE];
178 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
180 static struct nfs4_delegation *
181 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
183 struct nfs4_delegation *dp;
184 struct nfs4_file *fp = stp->st_file;
185 struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
187 dprintk("NFSD alloc_init_deleg\n");
188 if (fp->fi_had_conflict)
190 if (num_delegations > max_delegations)
192 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
196 INIT_LIST_HEAD(&dp->dl_perfile);
197 INIT_LIST_HEAD(&dp->dl_perclnt);
198 INIT_LIST_HEAD(&dp->dl_recall_lru);
203 get_file(stp->st_vfs_file);
204 dp->dl_vfs_file = stp->st_vfs_file;
206 dp->dl_recall.cbr_dp = NULL;
207 dp->dl_recall.cbr_ident = cb->cb_ident;
208 dp->dl_recall.cbr_trunc = 0;
209 dp->dl_stateid.si_boot = get_seconds();
210 dp->dl_stateid.si_stateownerid = current_delegid++;
211 dp->dl_stateid.si_fileid = 0;
212 dp->dl_stateid.si_generation = 0;
213 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
215 atomic_set(&dp->dl_count, 1);
216 list_add(&dp->dl_perfile, &fp->fi_delegations);
217 list_add(&dp->dl_perclnt, &clp->cl_delegations);
222 nfs4_put_delegation(struct nfs4_delegation *dp)
224 if (atomic_dec_and_test(&dp->dl_count)) {
225 dprintk("NFSD: freeing dp %p\n",dp);
226 put_nfs4_file(dp->dl_file);
227 kmem_cache_free(deleg_slab, dp);
232 /* Remove the associated file_lock first, then remove the delegation.
233 * lease_modify() is called to remove the FS_LEASE file_lock from
234 * the i_flock list, eventually calling nfsd's lock_manager
235 * fl_release_callback.
238 nfs4_close_delegation(struct nfs4_delegation *dp)
240 struct file *filp = dp->dl_vfs_file;
242 dprintk("NFSD: close_delegation dp %p\n",dp);
243 dp->dl_vfs_file = NULL;
244 /* The following nfsd_close may not actually close the file,
245 * but we want to remove the lease in any case. */
247 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
251 /* Called under the state lock. */
253 unhash_delegation(struct nfs4_delegation *dp)
255 list_del_init(&dp->dl_perfile);
256 list_del_init(&dp->dl_perclnt);
257 spin_lock(&recall_lock);
258 list_del_init(&dp->dl_recall_lru);
259 spin_unlock(&recall_lock);
260 nfs4_close_delegation(dp);
261 nfs4_put_delegation(dp);
268 /* Hash tables for nfs4_clientid state */
269 #define CLIENT_HASH_BITS 4
270 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
271 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
273 #define clientid_hashval(id) \
274 ((id) & CLIENT_HASH_MASK)
275 #define clientstr_hashval(name) \
276 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
278 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
279 * used in reboot/reset lease grace period processing
281 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
282 * setclientid_confirmed info.
284 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
287 * client_lru holds client queue ordered by nfs4_client.cl_time
290 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
291 * for last close replay.
293 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
294 static int reclaim_str_hashtbl_size = 0;
295 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
299 static struct list_head client_lru;
300 static struct list_head close_lru;
302 static void unhash_generic_stateid(struct nfs4_stateid *stp)
304 list_del(&stp->st_hash);
305 list_del(&stp->st_perfile);
306 list_del(&stp->st_perstateowner);
309 static void free_generic_stateid(struct nfs4_stateid *stp)
311 put_nfs4_file(stp->st_file);
312 kmem_cache_free(stateid_slab, stp);
315 static void release_lock_stateid(struct nfs4_stateid *stp)
317 unhash_generic_stateid(stp);
318 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
319 free_generic_stateid(stp);
322 static void unhash_lockowner(struct nfs4_stateowner *sop)
324 struct nfs4_stateid *stp;
326 list_del(&sop->so_idhash);
327 list_del(&sop->so_strhash);
328 list_del(&sop->so_perstateid);
329 while (!list_empty(&sop->so_stateids)) {
330 stp = list_first_entry(&sop->so_stateids,
331 struct nfs4_stateid, st_perstateowner);
332 release_lock_stateid(stp);
336 static void release_lockowner(struct nfs4_stateowner *sop)
338 unhash_lockowner(sop);
339 nfs4_put_stateowner(sop);
343 release_stateid_lockowners(struct nfs4_stateid *open_stp)
345 struct nfs4_stateowner *lock_sop;
347 while (!list_empty(&open_stp->st_lockowners)) {
348 lock_sop = list_entry(open_stp->st_lockowners.next,
349 struct nfs4_stateowner, so_perstateid);
350 /* list_del(&open_stp->st_lockowners); */
351 BUG_ON(lock_sop->so_is_open_owner);
352 release_lockowner(lock_sop);
356 static void release_open_stateid(struct nfs4_stateid *stp)
358 unhash_generic_stateid(stp);
359 release_stateid_lockowners(stp);
360 nfsd_close(stp->st_vfs_file);
361 free_generic_stateid(stp);
364 static void unhash_openowner(struct nfs4_stateowner *sop)
366 struct nfs4_stateid *stp;
368 list_del(&sop->so_idhash);
369 list_del(&sop->so_strhash);
370 list_del(&sop->so_perclient);
371 list_del(&sop->so_perstateid); /* XXX: necessary? */
372 while (!list_empty(&sop->so_stateids)) {
373 stp = list_first_entry(&sop->so_stateids,
374 struct nfs4_stateid, st_perstateowner);
375 release_open_stateid(stp);
379 static void release_openowner(struct nfs4_stateowner *sop)
381 unhash_openowner(sop);
382 list_del(&sop->so_close_lru);
383 nfs4_put_stateowner(sop);
386 static DEFINE_SPINLOCK(sessionid_lock);
387 #define SESSION_HASH_SIZE 512
388 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
391 hash_sessionid(struct nfs4_sessionid *sessionid)
393 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
395 return sid->sequence % SESSION_HASH_SIZE;
399 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
401 u32 *ptr = (u32 *)(&sessionid->data[0]);
402 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
406 gen_sessionid(struct nfsd4_session *ses)
408 struct nfs4_client *clp = ses->se_client;
409 struct nfsd4_sessionid *sid;
411 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
412 sid->clientid = clp->cl_clientid;
413 sid->sequence = current_sessionid++;
418 * Give the client the number of slots it requests bound by
419 * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
421 * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
422 * should (up to a point) re-negotiate active sessions and reduce their
423 * slot usage to make rooom for new connections. For now we just fail the
426 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
428 int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
430 spin_lock(&nfsd_serv->sv_lock);
431 if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
432 np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
433 nfsd_serv->sv_drc_pages_used += np;
434 spin_unlock(&nfsd_serv->sv_lock);
437 status = nfserr_resource;
440 fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
446 * fchan holds the client values on input, and the server values on output
448 static int init_forechannel_attrs(struct svc_rqst *rqstp,
449 struct nfsd4_session *session,
450 struct nfsd4_channel_attrs *fchan)
453 __u32 maxcount = svc_max_payload(rqstp);
455 /* headerpadsz set to zero in encode routine */
457 /* Use the client's max request and max response size if possible */
458 if (fchan->maxreq_sz > maxcount)
459 fchan->maxreq_sz = maxcount;
460 session->se_fmaxreq_sz = fchan->maxreq_sz;
462 if (fchan->maxresp_sz > maxcount)
463 fchan->maxresp_sz = maxcount;
464 session->se_fmaxresp_sz = fchan->maxresp_sz;
466 /* Set the max response cached size our default which is
467 * a multiple of PAGE_SIZE and small */
468 session->se_fmaxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
469 fchan->maxresp_cached = session->se_fmaxresp_cached;
471 /* Use the client's maxops if possible */
472 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
473 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
474 session->se_fmaxops = fchan->maxops;
476 /* try to use the client requested number of slots */
477 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
478 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
480 /* FIXME: Error means no more DRC pages so the server should
481 * recover pages from existing sessions. For now fail session
484 status = set_forechannel_maxreqs(fchan);
486 session->se_fnumslots = fchan->maxreqs;
491 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
492 struct nfsd4_create_session *cses)
494 struct nfsd4_session *new, tmp;
495 int idx, status = nfserr_resource, slotsize;
497 memset(&tmp, 0, sizeof(tmp));
499 /* FIXME: For now, we just accept the client back channel attributes. */
500 status = init_forechannel_attrs(rqstp, &tmp, &cses->fore_channel);
504 /* allocate struct nfsd4_session and slot table in one piece */
505 slotsize = tmp.se_fnumslots * sizeof(struct nfsd4_slot);
506 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
510 memcpy(new, &tmp, sizeof(*new));
512 new->se_client = clp;
514 idx = hash_sessionid(&new->se_sessionid);
515 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
516 NFS4_MAX_SESSIONID_LEN);
518 new->se_flags = cses->flags;
519 kref_init(&new->se_ref);
520 spin_lock(&sessionid_lock);
521 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
522 list_add(&new->se_perclnt, &clp->cl_sessions);
523 spin_unlock(&sessionid_lock);
530 /* caller must hold sessionid_lock */
531 static struct nfsd4_session *
532 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
534 struct nfsd4_session *elem;
537 dump_sessionid(__func__, sessionid);
538 idx = hash_sessionid(sessionid);
539 dprintk("%s: idx is %d\n", __func__, idx);
540 /* Search in the appropriate list */
541 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
542 dump_sessionid("list traversal", &elem->se_sessionid);
543 if (!memcmp(elem->se_sessionid.data, sessionid->data,
544 NFS4_MAX_SESSIONID_LEN)) {
549 dprintk("%s: session not found\n", __func__);
553 /* caller must hold sessionid_lock */
555 unhash_session(struct nfsd4_session *ses)
557 list_del(&ses->se_hash);
558 list_del(&ses->se_perclnt);
562 release_session(struct nfsd4_session *ses)
564 spin_lock(&sessionid_lock);
566 spin_unlock(&sessionid_lock);
567 nfsd4_put_session(ses);
570 static void nfsd4_release_respages(struct page **respages, short resused);
573 free_session(struct kref *kref)
575 struct nfsd4_session *ses;
578 ses = container_of(kref, struct nfsd4_session, se_ref);
579 for (i = 0; i < ses->se_fnumslots; i++) {
580 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
581 nfsd4_release_respages(e->ce_respages, e->ce_resused);
583 kfree(ses->se_slots);
588 renew_client(struct nfs4_client *clp)
591 * Move client to the end to the LRU list.
593 dprintk("renewing client (clientid %08x/%08x)\n",
594 clp->cl_clientid.cl_boot,
595 clp->cl_clientid.cl_id);
596 list_move_tail(&clp->cl_lru, &client_lru);
597 clp->cl_time = get_seconds();
600 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
602 STALE_CLIENTID(clientid_t *clid)
604 if (clid->cl_boot == boot_time)
606 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
607 clid->cl_boot, clid->cl_id, boot_time);
612 * XXX Should we use a slab cache ?
613 * This type of memory management is somewhat inefficient, but we use it
614 * anyway since SETCLIENTID is not a common operation.
616 static struct nfs4_client *alloc_client(struct xdr_netobj name)
618 struct nfs4_client *clp;
620 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
623 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
624 if (clp->cl_name.data == NULL) {
628 memcpy(clp->cl_name.data, name.data, name.len);
629 clp->cl_name.len = name.len;
634 shutdown_callback_client(struct nfs4_client *clp)
636 struct rpc_clnt *clnt = clp->cl_callback.cb_client;
640 * Callback threads take a reference on the client, so there
641 * should be no outstanding callbacks at this point.
643 clp->cl_callback.cb_client = NULL;
644 rpc_shutdown_client(clnt);
646 if (clp->cl_callback.cb_cred) {
647 put_rpccred(clp->cl_callback.cb_cred);
648 clp->cl_callback.cb_cred = NULL;
653 free_client(struct nfs4_client *clp)
655 shutdown_callback_client(clp);
656 nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
657 clp->cl_slot.sl_cache_entry.ce_resused);
658 if (clp->cl_cred.cr_group_info)
659 put_group_info(clp->cl_cred.cr_group_info);
660 kfree(clp->cl_principal);
661 kfree(clp->cl_name.data);
666 put_nfs4_client(struct nfs4_client *clp)
668 if (atomic_dec_and_test(&clp->cl_count))
673 expire_client(struct nfs4_client *clp)
675 struct nfs4_stateowner *sop;
676 struct nfs4_delegation *dp;
677 struct list_head reaplist;
679 dprintk("NFSD: expire_client cl_count %d\n",
680 atomic_read(&clp->cl_count));
682 INIT_LIST_HEAD(&reaplist);
683 spin_lock(&recall_lock);
684 while (!list_empty(&clp->cl_delegations)) {
685 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
686 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
688 list_del_init(&dp->dl_perclnt);
689 list_move(&dp->dl_recall_lru, &reaplist);
691 spin_unlock(&recall_lock);
692 while (!list_empty(&reaplist)) {
693 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
694 list_del_init(&dp->dl_recall_lru);
695 unhash_delegation(dp);
697 list_del(&clp->cl_idhash);
698 list_del(&clp->cl_strhash);
699 list_del(&clp->cl_lru);
700 while (!list_empty(&clp->cl_openowners)) {
701 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
702 release_openowner(sop);
704 while (!list_empty(&clp->cl_sessions)) {
705 struct nfsd4_session *ses;
706 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
708 release_session(ses);
710 put_nfs4_client(clp);
713 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
715 struct nfs4_client *clp;
717 clp = alloc_client(name);
720 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
721 atomic_set(&clp->cl_count, 1);
722 atomic_set(&clp->cl_callback.cb_set, 0);
723 INIT_LIST_HEAD(&clp->cl_idhash);
724 INIT_LIST_HEAD(&clp->cl_strhash);
725 INIT_LIST_HEAD(&clp->cl_openowners);
726 INIT_LIST_HEAD(&clp->cl_delegations);
727 INIT_LIST_HEAD(&clp->cl_sessions);
728 INIT_LIST_HEAD(&clp->cl_lru);
732 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
734 memcpy(target->cl_verifier.data, source->data,
735 sizeof(target->cl_verifier.data));
738 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
740 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
741 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
744 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
746 target->cr_uid = source->cr_uid;
747 target->cr_gid = source->cr_gid;
748 target->cr_group_info = source->cr_group_info;
749 get_group_info(target->cr_group_info);
752 static int same_name(const char *n1, const char *n2)
754 return 0 == memcmp(n1, n2, HEXDIR_LEN);
758 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
760 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
764 same_clid(clientid_t *cl1, clientid_t *cl2)
766 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
769 /* XXX what about NGROUP */
771 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
773 return cr1->cr_uid == cr2->cr_uid;
776 static void gen_clid(struct nfs4_client *clp)
778 static u32 current_clientid = 1;
780 clp->cl_clientid.cl_boot = boot_time;
781 clp->cl_clientid.cl_id = current_clientid++;
784 static void gen_confirm(struct nfs4_client *clp)
789 p = (u32 *)clp->cl_confirm.data;
790 *p++ = get_seconds();
794 static int check_name(struct xdr_netobj name)
798 if (name.len > NFS4_OPAQUE_LIMIT) {
799 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
806 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
808 unsigned int idhashval;
810 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
811 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
812 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
813 list_add_tail(&clp->cl_lru, &client_lru);
814 clp->cl_time = get_seconds();
818 move_to_confirmed(struct nfs4_client *clp)
820 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
821 unsigned int strhashval;
823 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
824 list_del_init(&clp->cl_strhash);
825 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
826 strhashval = clientstr_hashval(clp->cl_recdir);
827 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
831 static struct nfs4_client *
832 find_confirmed_client(clientid_t *clid)
834 struct nfs4_client *clp;
835 unsigned int idhashval = clientid_hashval(clid->cl_id);
837 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
838 if (same_clid(&clp->cl_clientid, clid))
844 static struct nfs4_client *
845 find_unconfirmed_client(clientid_t *clid)
847 struct nfs4_client *clp;
848 unsigned int idhashval = clientid_hashval(clid->cl_id);
850 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
851 if (same_clid(&clp->cl_clientid, clid))
858 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
859 * parameter. Matching is based on the fact the at least one of the
860 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
862 * FIXME: we need to unify the clientid namespaces for nfsv4.x
863 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
864 * and SET_CLIENTID{,_CONFIRM}
867 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
869 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
870 return use_exchange_id == has_exchange_flags;
873 static struct nfs4_client *
874 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
875 bool use_exchange_id)
877 struct nfs4_client *clp;
879 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
880 if (same_name(clp->cl_recdir, dname) &&
881 match_clientid_establishment(clp, use_exchange_id))
887 static struct nfs4_client *
888 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
889 bool use_exchange_id)
891 struct nfs4_client *clp;
893 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
894 if (same_name(clp->cl_recdir, dname) &&
895 match_clientid_establishment(clp, use_exchange_id))
901 /* a helper function for parse_callback */
903 parse_octet(unsigned int *lenp, char **addrp)
905 unsigned int len = *lenp;
917 if ((c < '0') || (c > '9')) {
923 n = (n * 10) + (c - '0');
934 /* parse and set the setclientid ipv4 callback address */
936 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
941 u32 addrlen = addr_len;
942 char *addr = addr_val;
947 for(i = 4; i > 0 ; i--) {
948 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
951 cbaddr |= (temp << shift);
959 for(i = 2; i > 0 ; i--) {
960 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
963 cbport |= (temp << shift);
972 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
974 struct nfs4_callback *cb = &clp->cl_callback;
976 /* Currently, we only support tcp for the callback channel */
977 if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
980 if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
981 &cb->cb_addr, &cb->cb_port)))
983 cb->cb_prog = se->se_callback_prog;
984 cb->cb_ident = se->se_callback_ident;
987 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
988 "will not receive delegations\n",
989 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
995 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
997 struct nfsd4_compoundres *resp = rqstp->rq_resp;
999 resp->cstate.statp = statp;
1003 * Dereference the result pages.
1006 nfsd4_release_respages(struct page **respages, short resused)
1010 dprintk("--> %s\n", __func__);
1011 for (i = 0; i < resused; i++) {
1014 put_page(respages[i]);
1020 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1024 for (i = 0; i < count; i++) {
1025 topages[i] = frompages[i];
1028 get_page(topages[i]);
1033 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1034 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1035 * length of the XDR response is less than se_fmaxresp_cached
1036 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1037 * of the reply (e.g. readdir).
1039 * Store the base and length of the rq_req.head[0] page
1040 * of the NFSv4.1 data, just past the rpc header.
1043 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1045 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1046 struct svc_rqst *rqstp = resp->rqstp;
1047 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1048 struct nfsd4_op *op = &args->ops[resp->opcnt];
1049 struct kvec *resv = &rqstp->rq_res.head[0];
1051 dprintk("--> %s entry %p\n", __func__, entry);
1053 /* Don't cache a failed OP_SEQUENCE. */
1054 if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1057 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1058 entry->ce_opcnt = resp->opcnt;
1059 entry->ce_status = resp->cstate.status;
1062 * Don't need a page to cache just the sequence operation - the slot
1066 if (nfsd4_not_cached(resp)) {
1067 entry->ce_resused = 0;
1068 entry->ce_rpchdrlen = 0;
1069 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1070 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1073 entry->ce_resused = rqstp->rq_resused;
1074 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1075 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1076 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1078 entry->ce_datav.iov_base = resp->cstate.statp;
1079 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1080 (char *)page_address(rqstp->rq_respages[0]));
1081 /* Current request rpc header length*/
1082 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1083 (char *)page_address(rqstp->rq_respages[0]);
1087 * We keep the rpc header, but take the nfs reply from the replycache.
1090 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1091 struct nfsd4_cache_entry *entry)
1093 struct svc_rqst *rqstp = resp->rqstp;
1094 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1097 /* Current request rpc header length*/
1098 len = (char *)resp->cstate.statp -
1099 (char *)page_address(rqstp->rq_respages[0]);
1100 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1101 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1102 entry->ce_datav.iov_len);
1105 /* copy the cached reply nfsd data past the current rpc header */
1106 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1107 entry->ce_datav.iov_len);
1108 resv->iov_len = len + entry->ce_datav.iov_len;
1113 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1114 * cached page. Replace any futher replay pages from the cache.
1117 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1118 struct nfsd4_sequence *seq)
1120 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1123 dprintk("--> %s entry %p\n", __func__, entry);
1126 * If this is just the sequence operation, we did not keep
1127 * a page in the cache entry because we can just use the
1128 * slot info stored in struct nfsd4_sequence that was checked
1129 * against the slot in nfsd4_sequence().
1131 * This occurs when seq->cachethis is FALSE, or when the client
1132 * session inactivity timer fires and a solo sequence operation
1133 * is sent (lease renewal).
1135 if (seq && nfsd4_not_cached(resp)) {
1136 seq->maxslots = resp->cstate.session->se_fnumslots;
1140 if (!nfsd41_copy_replay_data(resp, entry)) {
1142 * Not enough room to use the replay rpc header, send the
1143 * cached header. Release all the allocated result pages.
1145 svc_free_res_pages(resp->rqstp);
1146 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1149 /* Release all but the first allocated result page */
1151 resp->rqstp->rq_resused--;
1152 svc_free_res_pages(resp->rqstp);
1154 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1155 &entry->ce_respages[1],
1156 entry->ce_resused - 1);
1159 resp->rqstp->rq_resused = entry->ce_resused;
1160 resp->opcnt = entry->ce_opcnt;
1161 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1162 status = entry->ce_status;
1168 * Set the exchange_id flags returned by the server.
1171 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1173 /* pNFS is not supported */
1174 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1176 /* Referrals are supported, Migration is not. */
1177 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1179 /* set the wire flags to return to client. */
1180 clid->flags = new->cl_exchange_flags;
1184 nfsd4_exchange_id(struct svc_rqst *rqstp,
1185 struct nfsd4_compound_state *cstate,
1186 struct nfsd4_exchange_id *exid)
1188 struct nfs4_client *unconf, *conf, *new;
1190 unsigned int strhashval;
1191 char dname[HEXDIR_LEN];
1192 nfs4_verifier verf = exid->verifier;
1193 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1195 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1196 " ip_addr=%u flags %x, spa_how %d\n",
1197 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1198 ip_addr, exid->flags, exid->spa_how);
1200 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1201 return nfserr_inval;
1203 /* Currently only support SP4_NONE */
1204 switch (exid->spa_how) {
1208 return nfserr_encr_alg_unsupp;
1210 BUG(); /* checked by xdr code */
1212 return nfserr_serverfault; /* no excuse :-/ */
1215 status = nfs4_make_rec_clidname(dname, &exid->clname);
1220 strhashval = clientstr_hashval(dname);
1225 conf = find_confirmed_client_by_str(dname, strhashval, true);
1227 if (!same_verf(&verf, &conf->cl_verifier)) {
1228 /* 18.35.4 case 8 */
1229 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1230 status = nfserr_not_same;
1233 /* Client reboot: destroy old state */
1234 expire_client(conf);
1237 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1238 /* 18.35.4 case 9 */
1239 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1240 status = nfserr_perm;
1243 expire_client(conf);
1246 if (ip_addr != conf->cl_addr &&
1247 !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
1248 /* Client collision. 18.35.4 case 3 */
1249 status = nfserr_clid_inuse;
1253 * Set bit when the owner id and verifier map to an already
1254 * confirmed client id (18.35.3).
1256 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1259 * Falling into 18.35.4 case 2, possible router replay.
1260 * Leave confirmed record intact and return same result.
1262 copy_verf(conf, &verf);
1266 /* 18.35.4 case 7 */
1267 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1268 status = nfserr_noent;
1273 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1276 * Possible retry or client restart. Per 18.35.4 case 4,
1277 * a new unconfirmed record should be generated regardless
1278 * of whether any properties have changed.
1280 expire_client(unconf);
1285 new = create_client(exid->clname, dname);
1287 status = nfserr_resource;
1291 copy_verf(new, &verf);
1292 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1293 new->cl_addr = ip_addr;
1296 add_to_unconfirmed(new, strhashval);
1298 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1299 exid->clientid.cl_id = new->cl_clientid.cl_id;
1301 new->cl_slot.sl_seqid = 0;
1303 nfsd4_set_ex_flags(new, exid);
1305 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1306 new->cl_slot.sl_seqid, new->cl_exchange_flags);
1310 nfs4_unlock_state();
1312 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1317 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1319 dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1322 /* The slot is in use, and no response has been sent. */
1323 if (slot->sl_inuse) {
1324 if (seqid == slot->sl_seqid)
1325 return nfserr_jukebox;
1327 return nfserr_seq_misordered;
1330 if (likely(seqid == slot->sl_seqid + 1))
1333 if (seqid == slot->sl_seqid)
1334 return nfserr_replay_cache;
1336 if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1338 /* Misordered replay or misordered new request */
1339 return nfserr_seq_misordered;
1343 nfsd4_create_session(struct svc_rqst *rqstp,
1344 struct nfsd4_compound_state *cstate,
1345 struct nfsd4_create_session *cr_ses)
1347 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1348 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1349 struct nfs4_client *conf, *unconf;
1350 struct nfsd4_slot *slot = NULL;
1354 unconf = find_unconfirmed_client(&cr_ses->clientid);
1355 conf = find_confirmed_client(&cr_ses->clientid);
1358 slot = &conf->cl_slot;
1359 status = check_slot_seqid(cr_ses->seqid, slot);
1360 if (status == nfserr_replay_cache) {
1361 dprintk("Got a create_session replay! seqid= %d\n",
1363 cstate->slot = slot;
1364 cstate->status = status;
1365 /* Return the cached reply status */
1366 status = nfsd4_replay_cache_entry(resp, NULL);
1368 } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
1369 status = nfserr_seq_misordered;
1370 dprintk("Sequence misordered!\n");
1371 dprintk("Expected seqid= %d but got seqid= %d\n",
1372 slot->sl_seqid, cr_ses->seqid);
1375 conf->cl_slot.sl_seqid++;
1376 } else if (unconf) {
1377 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1378 (ip_addr != unconf->cl_addr)) {
1379 status = nfserr_clid_inuse;
1383 slot = &unconf->cl_slot;
1384 status = check_slot_seqid(cr_ses->seqid, slot);
1386 /* an unconfirmed replay returns misordered */
1387 status = nfserr_seq_misordered;
1391 slot->sl_seqid++; /* from 0 to 1 */
1392 move_to_confirmed(unconf);
1395 * We do not support RDMA or persistent sessions
1397 cr_ses->flags &= ~SESSION4_PERSIST;
1398 cr_ses->flags &= ~SESSION4_RDMA;
1402 status = nfserr_stale_clientid;
1406 status = alloc_init_session(rqstp, conf, cr_ses);
1410 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1411 NFS4_MAX_SESSIONID_LEN);
1412 cr_ses->seqid = slot->sl_seqid;
1414 slot->sl_inuse = true;
1415 cstate->slot = slot;
1416 /* Ensure a page is used for the cache */
1417 slot->sl_cache_entry.ce_cachethis = 1;
1419 nfs4_unlock_state();
1420 dprintk("%s returns %d\n", __func__, ntohl(status));
1425 nfsd4_destroy_session(struct svc_rqst *r,
1426 struct nfsd4_compound_state *cstate,
1427 struct nfsd4_destroy_session *sessionid)
1429 struct nfsd4_session *ses;
1430 u32 status = nfserr_badsession;
1433 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1434 * - Should we return nfserr_back_chan_busy if waiting for
1435 * callbacks on to-be-destroyed session?
1436 * - Do we need to clear any callback info from previous session?
1439 dump_sessionid(__func__, &sessionid->sessionid);
1440 spin_lock(&sessionid_lock);
1441 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1443 spin_unlock(&sessionid_lock);
1447 unhash_session(ses);
1448 spin_unlock(&sessionid_lock);
1450 /* wait for callbacks */
1451 shutdown_callback_client(ses->se_client);
1452 nfsd4_put_session(ses);
1455 dprintk("%s returns %d\n", __func__, ntohl(status));
1460 nfsd4_sequence(struct svc_rqst *rqstp,
1461 struct nfsd4_compound_state *cstate,
1462 struct nfsd4_sequence *seq)
1464 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1465 struct nfsd4_session *session;
1466 struct nfsd4_slot *slot;
1469 if (resp->opcnt != 1)
1470 return nfserr_sequence_pos;
1472 spin_lock(&sessionid_lock);
1473 status = nfserr_badsession;
1474 session = find_in_sessionid_hashtbl(&seq->sessionid);
1478 status = nfserr_badslot;
1479 if (seq->slotid >= session->se_fnumslots)
1482 slot = &session->se_slots[seq->slotid];
1483 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1485 status = check_slot_seqid(seq->seqid, slot);
1486 if (status == nfserr_replay_cache) {
1487 cstate->slot = slot;
1488 cstate->session = session;
1489 /* Return the cached reply status and set cstate->status
1490 * for nfsd4_svc_encode_compoundres processing */
1491 status = nfsd4_replay_cache_entry(resp, seq);
1492 cstate->status = nfserr_replay_cache;
1498 /* Success! bump slot seqid */
1499 slot->sl_inuse = true;
1500 slot->sl_seqid = seq->seqid;
1501 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1502 /* Always set the cache entry cachethis for solo sequence */
1503 if (nfsd4_is_solo_sequence(resp))
1504 slot->sl_cache_entry.ce_cachethis = 1;
1506 cstate->slot = slot;
1507 cstate->session = session;
1510 /* Renew the clientid on success and on replay.
1511 * Hold a session reference until done processing the compound:
1512 * nfsd4_put_session called only if the cstate slot is set.
1514 renew_client(session->se_client);
1515 nfsd4_get_session(session);
1517 spin_unlock(&sessionid_lock);
1518 dprintk("%s: return %d\n", __func__, ntohl(status));
1523 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1524 struct nfsd4_setclientid *setclid)
1526 struct sockaddr_in *sin = svc_addr_in(rqstp);
1527 struct xdr_netobj clname = {
1528 .len = setclid->se_namelen,
1529 .data = setclid->se_name,
1531 nfs4_verifier clverifier = setclid->se_verf;
1532 unsigned int strhashval;
1533 struct nfs4_client *conf, *unconf, *new;
1536 char dname[HEXDIR_LEN];
1538 if (!check_name(clname))
1539 return nfserr_inval;
1541 status = nfs4_make_rec_clidname(dname, &clname);
1546 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1547 * We get here on a DRC miss.
1550 strhashval = clientstr_hashval(dname);
1553 conf = find_confirmed_client_by_str(dname, strhashval, false);
1555 /* RFC 3530 14.2.33 CASE 0: */
1556 status = nfserr_clid_inuse;
1557 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1558 dprintk("NFSD: setclientid: string in use by client"
1559 " at %pI4\n", &conf->cl_addr);
1564 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1565 * has a description of SETCLIENTID request processing consisting
1566 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1568 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1569 status = nfserr_resource;
1572 * RFC 3530 14.2.33 CASE 4:
1573 * placed first, because it is the normal case
1576 expire_client(unconf);
1577 new = create_client(clname, dname);
1581 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1583 * RFC 3530 14.2.33 CASE 1:
1584 * probable callback update
1587 /* Note this is removing unconfirmed {*x***},
1588 * which is stronger than RFC recommended {vxc**}.
1589 * This has the advantage that there is at most
1590 * one {*x***} in either list at any time.
1592 expire_client(unconf);
1594 new = create_client(clname, dname);
1597 copy_clid(new, conf);
1598 } else if (!unconf) {
1600 * RFC 3530 14.2.33 CASE 2:
1601 * probable client reboot; state will be removed if
1604 new = create_client(clname, dname);
1610 * RFC 3530 14.2.33 CASE 3:
1611 * probable client reboot; state will be removed if
1614 expire_client(unconf);
1615 new = create_client(clname, dname);
1620 copy_verf(new, &clverifier);
1621 new->cl_addr = sin->sin_addr.s_addr;
1622 new->cl_flavor = rqstp->rq_flavor;
1623 princ = svc_gss_principal(rqstp);
1625 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1626 if (new->cl_principal == NULL) {
1631 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1633 gen_callback(new, setclid);
1634 add_to_unconfirmed(new, strhashval);
1635 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1636 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1637 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1640 nfs4_unlock_state();
1646 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1647 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1648 * bullets, labeled as CASE1 - CASE4 below.
1651 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1652 struct nfsd4_compound_state *cstate,
1653 struct nfsd4_setclientid_confirm *setclientid_confirm)
1655 struct sockaddr_in *sin = svc_addr_in(rqstp);
1656 struct nfs4_client *conf, *unconf;
1657 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1658 clientid_t * clid = &setclientid_confirm->sc_clientid;
1661 if (STALE_CLIENTID(clid))
1662 return nfserr_stale_clientid;
1664 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1665 * We get here on a DRC miss.
1670 conf = find_confirmed_client(clid);
1671 unconf = find_unconfirmed_client(clid);
1673 status = nfserr_clid_inuse;
1674 if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1676 if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1680 * section 14.2.34 of RFC 3530 has a description of
1681 * SETCLIENTID_CONFIRM request processing consisting
1682 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1684 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1686 * RFC 3530 14.2.34 CASE 1:
1689 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1690 status = nfserr_clid_inuse;
1692 /* XXX: We just turn off callbacks until we can handle
1693 * change request correctly. */
1694 atomic_set(&conf->cl_callback.cb_set, 0);
1695 expire_client(unconf);
1699 } else if (conf && !unconf) {
1701 * RFC 3530 14.2.34 CASE 2:
1702 * probable retransmitted request; play it safe and
1705 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1706 status = nfserr_clid_inuse;
1709 } else if (!conf && unconf
1710 && same_verf(&unconf->cl_confirm, &confirm)) {
1712 * RFC 3530 14.2.34 CASE 3:
1713 * Normal case; new or rebooted client:
1715 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1716 status = nfserr_clid_inuse;
1719 clientstr_hashval(unconf->cl_recdir);
1720 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1723 nfsd4_remove_clid_dir(conf);
1724 expire_client(conf);
1726 move_to_confirmed(unconf);
1728 nfsd4_probe_callback(conf);
1731 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1732 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1735 * RFC 3530 14.2.34 CASE 4:
1736 * Client probably hasn't noticed that we rebooted yet.
1738 status = nfserr_stale_clientid;
1740 /* check that we have hit one of the cases...*/
1741 status = nfserr_clid_inuse;
1744 nfs4_unlock_state();
1748 /* OPEN Share state helper functions */
1749 static inline struct nfs4_file *
1750 alloc_init_file(struct inode *ino)
1752 struct nfs4_file *fp;
1753 unsigned int hashval = file_hashval(ino);
1755 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1757 atomic_set(&fp->fi_ref, 1);
1758 INIT_LIST_HEAD(&fp->fi_hash);
1759 INIT_LIST_HEAD(&fp->fi_stateids);
1760 INIT_LIST_HEAD(&fp->fi_delegations);
1761 spin_lock(&recall_lock);
1762 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1763 spin_unlock(&recall_lock);
1764 fp->fi_inode = igrab(ino);
1765 fp->fi_id = current_fileid++;
1766 fp->fi_had_conflict = false;
1773 nfsd4_free_slab(struct kmem_cache **slab)
1777 kmem_cache_destroy(*slab);
1782 nfsd4_free_slabs(void)
1784 nfsd4_free_slab(&stateowner_slab);
1785 nfsd4_free_slab(&file_slab);
1786 nfsd4_free_slab(&stateid_slab);
1787 nfsd4_free_slab(&deleg_slab);
1791 nfsd4_init_slabs(void)
1793 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1794 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1795 if (stateowner_slab == NULL)
1797 file_slab = kmem_cache_create("nfsd4_files",
1798 sizeof(struct nfs4_file), 0, 0, NULL);
1799 if (file_slab == NULL)
1801 stateid_slab = kmem_cache_create("nfsd4_stateids",
1802 sizeof(struct nfs4_stateid), 0, 0, NULL);
1803 if (stateid_slab == NULL)
1805 deleg_slab = kmem_cache_create("nfsd4_delegations",
1806 sizeof(struct nfs4_delegation), 0, 0, NULL);
1807 if (deleg_slab == NULL)
1812 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1817 nfs4_free_stateowner(struct kref *kref)
1819 struct nfs4_stateowner *sop =
1820 container_of(kref, struct nfs4_stateowner, so_ref);
1821 kfree(sop->so_owner.data);
1822 kmem_cache_free(stateowner_slab, sop);
1825 static inline struct nfs4_stateowner *
1826 alloc_stateowner(struct xdr_netobj *owner)
1828 struct nfs4_stateowner *sop;
1830 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1831 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1832 memcpy(sop->so_owner.data, owner->data, owner->len);
1833 sop->so_owner.len = owner->len;
1834 kref_init(&sop->so_ref);
1837 kmem_cache_free(stateowner_slab, sop);
1842 static struct nfs4_stateowner *
1843 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1844 struct nfs4_stateowner *sop;
1845 struct nfs4_replay *rp;
1846 unsigned int idhashval;
1848 if (!(sop = alloc_stateowner(&open->op_owner)))
1850 idhashval = ownerid_hashval(current_ownerid);
1851 INIT_LIST_HEAD(&sop->so_idhash);
1852 INIT_LIST_HEAD(&sop->so_strhash);
1853 INIT_LIST_HEAD(&sop->so_perclient);
1854 INIT_LIST_HEAD(&sop->so_stateids);
1855 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1856 INIT_LIST_HEAD(&sop->so_close_lru);
1858 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1859 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1860 list_add(&sop->so_perclient, &clp->cl_openowners);
1861 sop->so_is_open_owner = 1;
1862 sop->so_id = current_ownerid++;
1863 sop->so_client = clp;
1864 sop->so_seqid = open->op_seqid;
1865 sop->so_confirmed = 0;
1866 rp = &sop->so_replay;
1867 rp->rp_status = nfserr_serverfault;
1869 rp->rp_buf = rp->rp_ibuf;
1874 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1875 struct nfs4_stateowner *sop = open->op_stateowner;
1876 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1878 INIT_LIST_HEAD(&stp->st_hash);
1879 INIT_LIST_HEAD(&stp->st_perstateowner);
1880 INIT_LIST_HEAD(&stp->st_lockowners);
1881 INIT_LIST_HEAD(&stp->st_perfile);
1882 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1883 list_add(&stp->st_perstateowner, &sop->so_stateids);
1884 list_add(&stp->st_perfile, &fp->fi_stateids);
1885 stp->st_stateowner = sop;
1888 stp->st_stateid.si_boot = get_seconds();
1889 stp->st_stateid.si_stateownerid = sop->so_id;
1890 stp->st_stateid.si_fileid = fp->fi_id;
1891 stp->st_stateid.si_generation = 0;
1892 stp->st_access_bmap = 0;
1893 stp->st_deny_bmap = 0;
1894 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1895 &stp->st_access_bmap);
1896 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1897 stp->st_openstp = NULL;
1901 move_to_close_lru(struct nfs4_stateowner *sop)
1903 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1905 list_move_tail(&sop->so_close_lru, &close_lru);
1906 sop->so_time = get_seconds();
1910 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1913 return (sop->so_owner.len == owner->len) &&
1914 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1915 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1918 static struct nfs4_stateowner *
1919 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1921 struct nfs4_stateowner *so = NULL;
1923 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1924 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1930 /* search file_hashtbl[] for file */
1931 static struct nfs4_file *
1932 find_file(struct inode *ino)
1934 unsigned int hashval = file_hashval(ino);
1935 struct nfs4_file *fp;
1937 spin_lock(&recall_lock);
1938 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1939 if (fp->fi_inode == ino) {
1941 spin_unlock(&recall_lock);
1945 spin_unlock(&recall_lock);
1949 static inline int access_valid(u32 x, u32 minorversion)
1951 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1953 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1955 x &= ~NFS4_SHARE_ACCESS_MASK;
1956 if (minorversion && x) {
1957 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1959 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1961 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1968 static inline int deny_valid(u32 x)
1970 /* Note: unlike access bits, deny bits may be zero. */
1971 return x <= NFS4_SHARE_DENY_BOTH;
1975 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1976 * st_{access,deny}_bmap field of the stateid, in order to track not
1977 * only what share bits are currently in force, but also what
1978 * combinations of share bits previous opens have used. This allows us
1979 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1980 * return an error if the client attempt to downgrade to a combination
1981 * of share bits not explicable by closing some of its previous opens.
1983 * XXX: This enforcement is actually incomplete, since we don't keep
1984 * track of access/deny bit combinations; so, e.g., we allow:
1986 * OPEN allow read, deny write
1987 * OPEN allow both, deny none
1988 * DOWNGRADE allow read, deny none
1990 * which we should reject.
1993 set_access(unsigned int *access, unsigned long bmap) {
1997 for (i = 1; i < 4; i++) {
1998 if (test_bit(i, &bmap))
2004 set_deny(unsigned int *deny, unsigned long bmap) {
2008 for (i = 0; i < 4; i++) {
2009 if (test_bit(i, &bmap))
2015 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2016 unsigned int access, deny;
2018 set_access(&access, stp->st_access_bmap);
2019 set_deny(&deny, stp->st_deny_bmap);
2020 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2026 * Called to check deny when READ with all zero stateid or
2027 * WRITE with all zero or all one stateid
2030 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2032 struct inode *ino = current_fh->fh_dentry->d_inode;
2033 struct nfs4_file *fp;
2034 struct nfs4_stateid *stp;
2037 dprintk("NFSD: nfs4_share_conflict\n");
2039 fp = find_file(ino);
2042 ret = nfserr_locked;
2043 /* Search for conflicting share reservations */
2044 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2045 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2046 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2056 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2058 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2059 drop_file_write_access(filp);
2060 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2065 * Recall a delegation
2068 do_recall(void *__dp)
2070 struct nfs4_delegation *dp = __dp;
2072 dp->dl_file->fi_had_conflict = true;
2073 nfsd4_cb_recall(dp);
2078 * Spawn a thread to perform a recall on the delegation represented
2079 * by the lease (file_lock)
2081 * Called from break_lease() with lock_kernel() held.
2082 * Note: we assume break_lease will only call this *once* for any given
2086 void nfsd_break_deleg_cb(struct file_lock *fl)
2088 struct nfs4_delegation *dp= (struct nfs4_delegation *)fl->fl_owner;
2089 struct task_struct *t;
2091 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2095 /* We're assuming the state code never drops its reference
2096 * without first removing the lease. Since we're in this lease
2097 * callback (and since the lease code is serialized by the kernel
2098 * lock) we know the server hasn't removed the lease yet, we know
2099 * it's safe to take a reference: */
2100 atomic_inc(&dp->dl_count);
2101 atomic_inc(&dp->dl_client->cl_count);
2103 spin_lock(&recall_lock);
2104 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2105 spin_unlock(&recall_lock);
2107 /* only place dl_time is set. protected by lock_kernel*/
2108 dp->dl_time = get_seconds();
2111 * We don't want the locks code to timeout the lease for us;
2112 * we'll remove it ourself if the delegation isn't returned
2115 fl->fl_break_time = 0;
2117 t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
2119 struct nfs4_client *clp = dp->dl_client;
2121 printk(KERN_INFO "NFSD: Callback thread failed for "
2122 "for client (clientid %08x/%08x)\n",
2123 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2124 put_nfs4_client(dp->dl_client);
2125 nfs4_put_delegation(dp);
2130 * The file_lock is being reapd.
2132 * Called by locks_free_lock() with lock_kernel() held.
2135 void nfsd_release_deleg_cb(struct file_lock *fl)
2137 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2139 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2141 if (!(fl->fl_flags & FL_LEASE) || !dp)
2143 dp->dl_flock = NULL;
2147 * Set the delegation file_lock back pointer.
2149 * Called from setlease() with lock_kernel() held.
2152 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2154 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2156 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2163 * Called from setlease() with lock_kernel() held
2166 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2168 struct nfs4_delegation *onlistd =
2169 (struct nfs4_delegation *)onlist->fl_owner;
2170 struct nfs4_delegation *tryd =
2171 (struct nfs4_delegation *)try->fl_owner;
2173 if (onlist->fl_lmops != try->fl_lmops)
2176 return onlistd->dl_client == tryd->dl_client;
2181 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2184 return lease_modify(onlist, arg);
2189 static struct lock_manager_operations nfsd_lease_mng_ops = {
2190 .fl_break = nfsd_break_deleg_cb,
2191 .fl_release_private = nfsd_release_deleg_cb,
2192 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2193 .fl_mylease = nfsd_same_client_deleg_cb,
2194 .fl_change = nfsd_change_deleg_cb,
2199 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2200 struct nfsd4_open *open)
2202 clientid_t *clientid = &open->op_clientid;
2203 struct nfs4_client *clp = NULL;
2204 unsigned int strhashval;
2205 struct nfs4_stateowner *sop = NULL;
2207 if (!check_name(open->op_owner))
2208 return nfserr_inval;
2210 if (STALE_CLIENTID(&open->op_clientid))
2211 return nfserr_stale_clientid;
2213 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2214 sop = find_openstateowner_str(strhashval, open);
2215 open->op_stateowner = sop;
2217 /* Make sure the client's lease hasn't expired. */
2218 clp = find_confirmed_client(clientid);
2220 return nfserr_expired;
2223 /* When sessions are used, skip open sequenceid processing */
2224 if (nfsd4_has_session(cstate))
2226 if (!sop->so_confirmed) {
2227 /* Replace unconfirmed owners without checking for replay. */
2228 clp = sop->so_client;
2229 release_openowner(sop);
2230 open->op_stateowner = NULL;
2233 if (open->op_seqid == sop->so_seqid - 1) {
2234 if (sop->so_replay.rp_buflen)
2235 return nfserr_replay_me;
2236 /* The original OPEN failed so spectacularly
2237 * that we don't even have replay data saved!
2238 * Therefore, we have no choice but to continue
2239 * processing this OPEN; presumably, we'll
2240 * fail again for the same reason.
2242 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2245 if (open->op_seqid != sop->so_seqid)
2246 return nfserr_bad_seqid;
2248 if (open->op_stateowner == NULL) {
2249 sop = alloc_init_open_stateowner(strhashval, clp, open);
2251 return nfserr_resource;
2252 open->op_stateowner = sop;
2254 list_del_init(&sop->so_close_lru);
2255 renew_client(sop->so_client);
2259 static inline __be32
2260 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2262 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2263 return nfserr_openmode;
2268 static struct nfs4_delegation *
2269 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2271 struct nfs4_delegation *dp;
2273 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2274 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2281 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2282 struct nfs4_delegation **dp)
2285 __be32 status = nfserr_bad_stateid;
2287 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2290 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2291 RD_STATE : WR_STATE;
2292 status = nfs4_check_delegmode(*dp, flags);
2296 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2300 open->op_stateowner->so_confirmed = 1;
2305 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2307 struct nfs4_stateid *local;
2308 __be32 status = nfserr_share_denied;
2309 struct nfs4_stateowner *sop = open->op_stateowner;
2311 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2312 /* ignore lock owners */
2313 if (local->st_stateowner->so_is_open_owner == 0)
2315 /* remember if we have seen this open owner */
2316 if (local->st_stateowner == sop)
2318 /* check for conflicting share reservations */
2319 if (!test_share(local, open))
2327 static inline struct nfs4_stateid *
2328 nfs4_alloc_stateid(void)
2330 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2334 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2335 struct nfs4_delegation *dp,
2336 struct svc_fh *cur_fh, int flags)
2338 struct nfs4_stateid *stp;
2340 stp = nfs4_alloc_stateid();
2342 return nfserr_resource;
2345 get_file(dp->dl_vfs_file);
2346 stp->st_vfs_file = dp->dl_vfs_file;
2349 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2352 if (status == nfserr_dropit)
2353 status = nfserr_jukebox;
2354 kmem_cache_free(stateid_slab, stp);
2362 static inline __be32
2363 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2364 struct nfsd4_open *open)
2366 struct iattr iattr = {
2367 .ia_valid = ATTR_SIZE,
2370 if (!open->op_truncate)
2372 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2373 return nfserr_inval;
2374 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2378 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2380 struct file *filp = stp->st_vfs_file;
2381 struct inode *inode = filp->f_path.dentry->d_inode;
2382 unsigned int share_access, new_writer;
2385 set_access(&share_access, stp->st_access_bmap);
2386 new_writer = (~share_access) & open->op_share_access
2387 & NFS4_SHARE_ACCESS_WRITE;
2390 int err = get_write_access(inode);
2392 return nfserrno(err);
2393 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2395 return nfserrno(err);
2396 file_take_write(filp);
2398 status = nfsd4_truncate(rqstp, cur_fh, open);
2401 put_write_access(inode);
2404 /* remember the open */
2405 filp->f_mode |= open->op_share_access;
2406 __set_bit(open->op_share_access, &stp->st_access_bmap);
2407 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2414 nfs4_set_claim_prev(struct nfsd4_open *open)
2416 open->op_stateowner->so_confirmed = 1;
2417 open->op_stateowner->so_client->cl_firststate = 1;
2421 * Attempt to hand out a delegation.
2424 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2426 struct nfs4_delegation *dp;
2427 struct nfs4_stateowner *sop = stp->st_stateowner;
2428 struct nfs4_callback *cb = &sop->so_client->cl_callback;
2429 struct file_lock fl, *flp = &fl;
2430 int status, flag = 0;
2432 flag = NFS4_OPEN_DELEGATE_NONE;
2433 open->op_recall = 0;
2434 switch (open->op_claim_type) {
2435 case NFS4_OPEN_CLAIM_PREVIOUS:
2436 if (!atomic_read(&cb->cb_set))
2437 open->op_recall = 1;
2438 flag = open->op_delegate_type;
2439 if (flag == NFS4_OPEN_DELEGATE_NONE)
2442 case NFS4_OPEN_CLAIM_NULL:
2443 /* Let's not give out any delegations till everyone's
2444 * had the chance to reclaim theirs.... */
2445 if (locks_in_grace())
2447 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2449 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2450 flag = NFS4_OPEN_DELEGATE_WRITE;
2452 flag = NFS4_OPEN_DELEGATE_READ;
2458 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2460 flag = NFS4_OPEN_DELEGATE_NONE;
2463 locks_init_lock(&fl);
2464 fl.fl_lmops = &nfsd_lease_mng_ops;
2465 fl.fl_flags = FL_LEASE;
2466 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2467 fl.fl_end = OFFSET_MAX;
2468 fl.fl_owner = (fl_owner_t)dp;
2469 fl.fl_file = stp->st_vfs_file;
2470 fl.fl_pid = current->tgid;
2472 /* vfs_setlease checks to see if delegation should be handed out.
2473 * the lock_manager callbacks fl_mylease and fl_change are used
2475 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2476 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2477 unhash_delegation(dp);
2478 flag = NFS4_OPEN_DELEGATE_NONE;
2482 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2484 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2485 dp->dl_stateid.si_boot,
2486 dp->dl_stateid.si_stateownerid,
2487 dp->dl_stateid.si_fileid,
2488 dp->dl_stateid.si_generation);
2490 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2491 && flag == NFS4_OPEN_DELEGATE_NONE
2492 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2493 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2494 open->op_delegate_type = flag;
2498 * called with nfs4_lock_state() held.
2501 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2503 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2504 struct nfs4_file *fp = NULL;
2505 struct inode *ino = current_fh->fh_dentry->d_inode;
2506 struct nfs4_stateid *stp = NULL;
2507 struct nfs4_delegation *dp = NULL;
2510 status = nfserr_inval;
2511 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2512 || !deny_valid(open->op_share_deny))
2515 * Lookup file; if found, lookup stateid and check open request,
2516 * and check for delegations in the process of being recalled.
2517 * If not found, create the nfs4_file struct
2519 fp = find_file(ino);
2521 if ((status = nfs4_check_open(fp, open, &stp)))
2523 status = nfs4_check_deleg(fp, open, &dp);
2527 status = nfserr_bad_stateid;
2528 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2530 status = nfserr_resource;
2531 fp = alloc_init_file(ino);
2537 * OPEN the file, or upgrade an existing OPEN.
2538 * If truncate fails, the OPEN fails.
2541 /* Stateid was found, this is an OPEN upgrade */
2542 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2545 update_stateid(&stp->st_stateid);
2547 /* Stateid was not found, this is a new OPEN */
2549 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2550 flags |= NFSD_MAY_READ;
2551 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2552 flags |= NFSD_MAY_WRITE;
2553 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2556 init_stateid(stp, fp, open);
2557 status = nfsd4_truncate(rqstp, current_fh, open);
2559 release_open_stateid(stp);
2562 if (nfsd4_has_session(&resp->cstate))
2563 update_stateid(&stp->st_stateid);
2565 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2567 if (nfsd4_has_session(&resp->cstate))
2568 open->op_stateowner->so_confirmed = 1;
2571 * Attempt to hand out a delegation. No error return, because the
2572 * OPEN succeeds even if we fail.
2574 nfs4_open_delegation(current_fh, open, stp);
2578 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2579 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2580 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2584 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2585 nfs4_set_claim_prev(open);
2587 * To finish the open response, we just need to set the rflags.
2589 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2590 if (!open->op_stateowner->so_confirmed &&
2591 !nfsd4_has_session(&resp->cstate))
2592 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2598 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2601 struct nfs4_client *clp;
2605 dprintk("process_renew(%08x/%08x): starting\n",
2606 clid->cl_boot, clid->cl_id);
2607 status = nfserr_stale_clientid;
2608 if (STALE_CLIENTID(clid))
2610 clp = find_confirmed_client(clid);
2611 status = nfserr_expired;
2613 /* We assume the client took too long to RENEW. */
2614 dprintk("nfsd4_renew: clientid not found!\n");
2618 status = nfserr_cb_path_down;
2619 if (!list_empty(&clp->cl_delegations)
2620 && !atomic_read(&clp->cl_callback.cb_set))
2624 nfs4_unlock_state();
2628 struct lock_manager nfsd4_manager = {
2632 nfsd4_end_grace(void)
2634 dprintk("NFSD: end of grace period\n");
2635 nfsd4_recdir_purge_old();
2636 locks_end_grace(&nfsd4_manager);
2640 nfs4_laundromat(void)
2642 struct nfs4_client *clp;
2643 struct nfs4_stateowner *sop;
2644 struct nfs4_delegation *dp;
2645 struct list_head *pos, *next, reaplist;
2646 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2647 time_t t, clientid_val = NFSD_LEASE_TIME;
2648 time_t u, test_val = NFSD_LEASE_TIME;
2652 dprintk("NFSD: laundromat service - starting\n");
2653 if (locks_in_grace())
2655 list_for_each_safe(pos, next, &client_lru) {
2656 clp = list_entry(pos, struct nfs4_client, cl_lru);
2657 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2658 t = clp->cl_time - cutoff;
2659 if (clientid_val > t)
2663 dprintk("NFSD: purging unused client (clientid %08x)\n",
2664 clp->cl_clientid.cl_id);
2665 nfsd4_remove_clid_dir(clp);
2668 INIT_LIST_HEAD(&reaplist);
2669 spin_lock(&recall_lock);
2670 list_for_each_safe(pos, next, &del_recall_lru) {
2671 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2672 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2673 u = dp->dl_time - cutoff;
2678 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2680 list_move(&dp->dl_recall_lru, &reaplist);
2682 spin_unlock(&recall_lock);
2683 list_for_each_safe(pos, next, &reaplist) {
2684 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2685 list_del_init(&dp->dl_recall_lru);
2686 unhash_delegation(dp);
2688 test_val = NFSD_LEASE_TIME;
2689 list_for_each_safe(pos, next, &close_lru) {
2690 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2691 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2692 u = sop->so_time - cutoff;
2697 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2699 release_openowner(sop);
2701 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2702 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2703 nfs4_unlock_state();
2704 return clientid_val;
2707 static struct workqueue_struct *laundry_wq;
2708 static void laundromat_main(struct work_struct *);
2709 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2712 laundromat_main(struct work_struct *not_used)
2716 t = nfs4_laundromat();
2717 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2718 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2721 static struct nfs4_stateowner *
2722 search_close_lru(u32 st_id, int flags)
2724 struct nfs4_stateowner *local = NULL;
2726 if (flags & CLOSE_STATE) {
2727 list_for_each_entry(local, &close_lru, so_close_lru) {
2728 if (local->so_id == st_id)
2736 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2738 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2742 STALE_STATEID(stateid_t *stateid)
2744 if (time_after((unsigned long)boot_time,
2745 (unsigned long)stateid->si_boot)) {
2746 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2747 stateid->si_boot, stateid->si_stateownerid,
2748 stateid->si_fileid, stateid->si_generation);
2755 EXPIRED_STATEID(stateid_t *stateid)
2757 if (time_before((unsigned long)boot_time,
2758 ((unsigned long)stateid->si_boot)) &&
2759 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2760 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2761 stateid->si_boot, stateid->si_stateownerid,
2762 stateid->si_fileid, stateid->si_generation);
2769 stateid_error_map(stateid_t *stateid)
2771 if (STALE_STATEID(stateid))
2772 return nfserr_stale_stateid;
2773 if (EXPIRED_STATEID(stateid))
2774 return nfserr_expired;
2776 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2777 stateid->si_boot, stateid->si_stateownerid,
2778 stateid->si_fileid, stateid->si_generation);
2779 return nfserr_bad_stateid;
2783 access_permit_read(unsigned long access_bmap)
2785 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2786 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2787 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2791 access_permit_write(unsigned long access_bmap)
2793 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2794 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2798 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2800 __be32 status = nfserr_openmode;
2802 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2804 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2811 static inline __be32
2812 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2814 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2816 else if (locks_in_grace()) {
2817 /* Answer in remaining cases depends on existance of
2818 * conflicting state; so we must wait out the grace period. */
2819 return nfserr_grace;
2820 } else if (flags & WR_STATE)
2821 return nfs4_share_conflict(current_fh,
2822 NFS4_SHARE_DENY_WRITE);
2823 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2824 return nfs4_share_conflict(current_fh,
2825 NFS4_SHARE_DENY_READ);
2829 * Allow READ/WRITE during grace period on recovered state only for files
2830 * that are not able to provide mandatory locking.
2833 grace_disallows_io(struct inode *inode)
2835 return locks_in_grace() && mandatory_lock(inode);
2838 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2841 * When sessions are used the stateid generation number is ignored
2844 if ((flags & HAS_SESSION) && in->si_generation == 0)
2847 /* If the client sends us a stateid from the future, it's buggy: */
2848 if (in->si_generation > ref->si_generation)
2849 return nfserr_bad_stateid;
2851 * The following, however, can happen. For example, if the
2852 * client sends an open and some IO at the same time, the open
2853 * may bump si_generation while the IO is still in flight.
2854 * Thanks to hard links and renames, the client never knows what
2855 * file an open will affect. So it could avoid that situation
2856 * only by serializing all opens and IO from the same open
2857 * owner. To recover from the old_stateid error, the client
2858 * will just have to retry the IO:
2860 if (in->si_generation < ref->si_generation)
2861 return nfserr_old_stateid;
2866 static int is_delegation_stateid(stateid_t *stateid)
2868 return stateid->si_fileid == 0;
2872 * Checks for stateid operations
2875 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2876 stateid_t *stateid, int flags, struct file **filpp)
2878 struct nfs4_stateid *stp = NULL;
2879 struct nfs4_delegation *dp = NULL;
2880 struct svc_fh *current_fh = &cstate->current_fh;
2881 struct inode *ino = current_fh->fh_dentry->d_inode;
2887 if (grace_disallows_io(ino))
2888 return nfserr_grace;
2890 if (nfsd4_has_session(cstate))
2891 flags |= HAS_SESSION;
2893 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2894 return check_special_stateids(current_fh, stateid, flags);
2896 status = nfserr_stale_stateid;
2897 if (STALE_STATEID(stateid))
2900 status = nfserr_bad_stateid;
2901 if (is_delegation_stateid(stateid)) {
2902 dp = find_delegation_stateid(ino, stateid);
2904 status = stateid_error_map(stateid);
2907 status = check_stateid_generation(stateid, &dp->dl_stateid,
2911 status = nfs4_check_delegmode(dp, flags);
2914 renew_client(dp->dl_client);
2916 *filpp = dp->dl_vfs_file;
2917 } else { /* open or lock stateid */
2918 stp = find_stateid(stateid, flags);
2920 status = stateid_error_map(stateid);
2923 if (nfs4_check_fh(current_fh, stp))
2925 if (!stp->st_stateowner->so_confirmed)
2927 status = check_stateid_generation(stateid, &stp->st_stateid,
2931 status = nfs4_check_openmode(stp, flags);
2934 renew_client(stp->st_stateowner->so_client);
2936 *filpp = stp->st_vfs_file;
2946 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2947 RD_STATE : WR_STATE;
2951 * Checks for sequence id mutating operations.
2954 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2955 stateid_t *stateid, int flags,
2956 struct nfs4_stateowner **sopp,
2957 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2959 struct nfs4_stateid *stp;
2960 struct nfs4_stateowner *sop;
2961 struct svc_fh *current_fh = &cstate->current_fh;
2964 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2965 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2966 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2967 stateid->si_generation);
2972 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2973 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2974 return nfserr_bad_stateid;
2977 if (STALE_STATEID(stateid))
2978 return nfserr_stale_stateid;
2980 if (nfsd4_has_session(cstate))
2981 flags |= HAS_SESSION;
2984 * We return BAD_STATEID if filehandle doesn't match stateid,
2985 * the confirmed flag is incorrecly set, or the generation
2986 * number is incorrect.
2988 stp = find_stateid(stateid, flags);
2991 * Also, we should make sure this isn't just the result of
2994 sop = search_close_lru(stateid->si_stateownerid, flags);
2996 return stateid_error_map(stateid);
3002 *sopp = sop = stp->st_stateowner;
3005 clientid_t *lockclid = &lock->v.new.clientid;
3006 struct nfs4_client *clp = sop->so_client;
3010 lkflg = setlkflg(lock->lk_type);
3012 if (lock->lk_is_new) {
3013 if (!sop->so_is_open_owner)
3014 return nfserr_bad_stateid;
3015 if (!(flags & HAS_SESSION) &&
3016 !same_clid(&clp->cl_clientid, lockclid))
3017 return nfserr_bad_stateid;
3018 /* stp is the open stateid */
3019 status = nfs4_check_openmode(stp, lkflg);
3023 /* stp is the lock stateid */
3024 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3030 if (nfs4_check_fh(current_fh, stp)) {
3031 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3032 return nfserr_bad_stateid;
3036 * We now validate the seqid and stateid generation numbers.
3037 * For the moment, we ignore the possibility of
3038 * generation number wraparound.
3040 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3043 if (sop->so_confirmed && flags & CONFIRM) {
3044 dprintk("NFSD: preprocess_seqid_op: expected"
3045 " unconfirmed stateowner!\n");
3046 return nfserr_bad_stateid;
3048 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3049 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3050 " confirmed yet!\n");
3051 return nfserr_bad_stateid;
3053 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3056 renew_client(sop->so_client);
3060 if (seqid == sop->so_seqid - 1) {
3061 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3062 /* indicate replay to calling function */
3063 return nfserr_replay_me;
3065 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3066 sop->so_seqid, seqid);
3068 return nfserr_bad_seqid;
3072 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3073 struct nfsd4_open_confirm *oc)
3076 struct nfs4_stateowner *sop;
3077 struct nfs4_stateid *stp;
3079 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3080 (int)cstate->current_fh.fh_dentry->d_name.len,
3081 cstate->current_fh.fh_dentry->d_name.name);
3083 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3089 if ((status = nfs4_preprocess_seqid_op(cstate,
3090 oc->oc_seqid, &oc->oc_req_stateid,
3091 CONFIRM | OPEN_STATE,
3092 &oc->oc_stateowner, &stp, NULL)))
3095 sop = oc->oc_stateowner;
3096 sop->so_confirmed = 1;
3097 update_stateid(&stp->st_stateid);
3098 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3099 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3100 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3101 stp->st_stateid.si_boot,
3102 stp->st_stateid.si_stateownerid,
3103 stp->st_stateid.si_fileid,
3104 stp->st_stateid.si_generation);
3106 nfsd4_create_clid_dir(sop->so_client);
3108 if (oc->oc_stateowner) {
3109 nfs4_get_stateowner(oc->oc_stateowner);
3110 cstate->replay_owner = oc->oc_stateowner;
3112 nfs4_unlock_state();
3118 * unset all bits in union bitmap (bmap) that
3119 * do not exist in share (from successful OPEN_DOWNGRADE)
3122 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3125 for (i = 1; i < 4; i++) {
3126 if ((i & access) != i)
3127 __clear_bit(i, bmap);
3132 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3135 for (i = 0; i < 4; i++) {
3136 if ((i & deny) != i)
3137 __clear_bit(i, bmap);
3142 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3143 struct nfsd4_compound_state *cstate,
3144 struct nfsd4_open_downgrade *od)
3147 struct nfs4_stateid *stp;
3148 unsigned int share_access;
3150 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3151 (int)cstate->current_fh.fh_dentry->d_name.len,
3152 cstate->current_fh.fh_dentry->d_name.name);
3154 if (!access_valid(od->od_share_access, cstate->minorversion)
3155 || !deny_valid(od->od_share_deny))
3156 return nfserr_inval;
3159 if ((status = nfs4_preprocess_seqid_op(cstate,
3163 &od->od_stateowner, &stp, NULL)))
3166 status = nfserr_inval;
3167 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3168 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3169 stp->st_access_bmap, od->od_share_access);
3172 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3173 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3174 stp->st_deny_bmap, od->od_share_deny);
3177 set_access(&share_access, stp->st_access_bmap);
3178 nfs4_file_downgrade(stp->st_vfs_file,
3179 share_access & ~od->od_share_access);
3181 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3182 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3184 update_stateid(&stp->st_stateid);
3185 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3188 if (od->od_stateowner) {
3189 nfs4_get_stateowner(od->od_stateowner);
3190 cstate->replay_owner = od->od_stateowner;
3192 nfs4_unlock_state();
3197 * nfs4_unlock_state() called after encode
3200 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3201 struct nfsd4_close *close)
3204 struct nfs4_stateid *stp;
3206 dprintk("NFSD: nfsd4_close on file %.*s\n",
3207 (int)cstate->current_fh.fh_dentry->d_name.len,
3208 cstate->current_fh.fh_dentry->d_name.name);
3211 /* check close_lru for replay */
3212 if ((status = nfs4_preprocess_seqid_op(cstate,
3215 OPEN_STATE | CLOSE_STATE,
3216 &close->cl_stateowner, &stp, NULL)))
3219 update_stateid(&stp->st_stateid);
3220 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3222 /* release_stateid() calls nfsd_close() if needed */
3223 release_open_stateid(stp);
3225 /* place unused nfs4_stateowners on so_close_lru list to be
3226 * released by the laundromat service after the lease period
3227 * to enable us to handle CLOSE replay
3229 if (list_empty(&close->cl_stateowner->so_stateids))
3230 move_to_close_lru(close->cl_stateowner);
3232 if (close->cl_stateowner) {
3233 nfs4_get_stateowner(close->cl_stateowner);
3234 cstate->replay_owner = close->cl_stateowner;
3236 nfs4_unlock_state();
3241 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3242 struct nfsd4_delegreturn *dr)
3244 struct nfs4_delegation *dp;
3245 stateid_t *stateid = &dr->dr_stateid;
3246 struct inode *inode;
3250 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3252 inode = cstate->current_fh.fh_dentry->d_inode;
3254 if (nfsd4_has_session(cstate))
3255 flags |= HAS_SESSION;
3257 status = nfserr_bad_stateid;
3258 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3260 status = nfserr_stale_stateid;
3261 if (STALE_STATEID(stateid))
3263 status = nfserr_bad_stateid;
3264 if (!is_delegation_stateid(stateid))
3266 dp = find_delegation_stateid(inode, stateid);
3268 status = stateid_error_map(stateid);
3271 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3274 renew_client(dp->dl_client);
3276 unhash_delegation(dp);
3278 nfs4_unlock_state();
3285 * Lock owner state (byte-range locks)
3287 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3288 #define LOCK_HASH_BITS 8
3289 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3290 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3293 end_offset(u64 start, u64 len)
3298 return end >= start ? end: NFS4_MAX_UINT64;
3301 /* last octet in a range */
3303 last_byte_offset(u64 start, u64 len)
3309 return end > start ? end - 1: NFS4_MAX_UINT64;
3312 #define lockownerid_hashval(id) \
3313 ((id) & LOCK_HASH_MASK)
3315 static inline unsigned int
3316 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3317 struct xdr_netobj *ownername)
3319 return (file_hashval(inode) + cl_id
3320 + opaque_hashval(ownername->data, ownername->len))
3324 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3325 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3326 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3328 static struct nfs4_stateid *
3329 find_stateid(stateid_t *stid, int flags)
3331 struct nfs4_stateid *local;
3332 u32 st_id = stid->si_stateownerid;
3333 u32 f_id = stid->si_fileid;
3334 unsigned int hashval;
3336 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3337 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3338 hashval = stateid_hashval(st_id, f_id);
3339 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3340 if ((local->st_stateid.si_stateownerid == st_id) &&
3341 (local->st_stateid.si_fileid == f_id))
3346 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3347 hashval = stateid_hashval(st_id, f_id);
3348 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3349 if ((local->st_stateid.si_stateownerid == st_id) &&
3350 (local->st_stateid.si_fileid == f_id))
3357 static struct nfs4_delegation *
3358 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3360 struct nfs4_file *fp;
3361 struct nfs4_delegation *dl;
3363 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3364 stid->si_boot, stid->si_stateownerid,
3365 stid->si_fileid, stid->si_generation);
3367 fp = find_file(ino);
3370 dl = find_delegation_file(fp, stid);
3376 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3377 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3378 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3379 * locking, this prevents us from being completely protocol-compliant. The
3380 * real solution to this problem is to start using unsigned file offsets in
3381 * the VFS, but this is a very deep change!
3384 nfs4_transform_lock_offset(struct file_lock *lock)
3386 if (lock->fl_start < 0)
3387 lock->fl_start = OFFSET_MAX;
3388 if (lock->fl_end < 0)
3389 lock->fl_end = OFFSET_MAX;
3392 /* Hack!: For now, we're defining this just so we can use a pointer to it
3393 * as a unique cookie to identify our (NFSv4's) posix locks. */
3394 static struct lock_manager_operations nfsd_posix_mng_ops = {
3398 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3400 struct nfs4_stateowner *sop;
3403 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3404 sop = (struct nfs4_stateowner *) fl->fl_owner;
3405 hval = lockownerid_hashval(sop->so_id);
3406 kref_get(&sop->so_ref);
3408 deny->ld_clientid = sop->so_client->cl_clientid;
3410 deny->ld_sop = NULL;
3411 deny->ld_clientid.cl_boot = 0;
3412 deny->ld_clientid.cl_id = 0;
3414 deny->ld_start = fl->fl_start;
3415 deny->ld_length = NFS4_MAX_UINT64;
3416 if (fl->fl_end != NFS4_MAX_UINT64)
3417 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3418 deny->ld_type = NFS4_READ_LT;
3419 if (fl->fl_type != F_RDLCK)
3420 deny->ld_type = NFS4_WRITE_LT;
3423 static struct nfs4_stateowner *
3424 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3425 struct xdr_netobj *owner)
3427 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3428 struct nfs4_stateowner *op;
3430 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3431 if (same_owner_str(op, owner, clid))
3438 * Alloc a lock owner structure.
3439 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3442 * strhashval = lock_ownerstr_hashval
3445 static struct nfs4_stateowner *
3446 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3447 struct nfs4_stateowner *sop;
3448 struct nfs4_replay *rp;
3449 unsigned int idhashval;
3451 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3453 idhashval = lockownerid_hashval(current_ownerid);
3454 INIT_LIST_HEAD(&sop->so_idhash);
3455 INIT_LIST_HEAD(&sop->so_strhash);
3456 INIT_LIST_HEAD(&sop->so_perclient);
3457 INIT_LIST_HEAD(&sop->so_stateids);
3458 INIT_LIST_HEAD(&sop->so_perstateid);
3459 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3461 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3462 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3463 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3464 sop->so_is_open_owner = 0;
3465 sop->so_id = current_ownerid++;
3466 sop->so_client = clp;
3467 /* It is the openowner seqid that will be incremented in encode in the
3468 * case of new lockowners; so increment the lock seqid manually: */
3469 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3470 sop->so_confirmed = 1;
3471 rp = &sop->so_replay;
3472 rp->rp_status = nfserr_serverfault;
3474 rp->rp_buf = rp->rp_ibuf;
3478 static struct nfs4_stateid *
3479 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3481 struct nfs4_stateid *stp;
3482 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3484 stp = nfs4_alloc_stateid();
3487 INIT_LIST_HEAD(&stp->st_hash);
3488 INIT_LIST_HEAD(&stp->st_perfile);
3489 INIT_LIST_HEAD(&stp->st_perstateowner);
3490 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3491 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3492 list_add(&stp->st_perfile, &fp->fi_stateids);
3493 list_add(&stp->st_perstateowner, &sop->so_stateids);
3494 stp->st_stateowner = sop;
3497 stp->st_stateid.si_boot = get_seconds();
3498 stp->st_stateid.si_stateownerid = sop->so_id;
3499 stp->st_stateid.si_fileid = fp->fi_id;
3500 stp->st_stateid.si_generation = 0;
3501 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3502 stp->st_access_bmap = open_stp->st_access_bmap;
3503 stp->st_deny_bmap = open_stp->st_deny_bmap;
3504 stp->st_openstp = open_stp;
3511 check_lock_length(u64 offset, u64 length)
3513 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3514 LOFF_OVERFLOW(offset, length)));
3521 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3522 struct nfsd4_lock *lock)
3524 struct nfs4_stateowner *open_sop = NULL;
3525 struct nfs4_stateowner *lock_sop = NULL;
3526 struct nfs4_stateid *lock_stp;
3528 struct file_lock file_lock;
3529 struct file_lock conflock;
3531 unsigned int strhashval;
3535 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3536 (long long) lock->lk_offset,
3537 (long long) lock->lk_length);
3539 if (check_lock_length(lock->lk_offset, lock->lk_length))
3540 return nfserr_inval;
3542 if ((status = fh_verify(rqstp, &cstate->current_fh,
3543 S_IFREG, NFSD_MAY_LOCK))) {
3544 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3550 if (lock->lk_is_new) {
3552 * Client indicates that this is a new lockowner.
3553 * Use open owner and open stateid to create lock owner and
3556 struct nfs4_stateid *open_stp = NULL;
3557 struct nfs4_file *fp;
3559 status = nfserr_stale_clientid;
3560 if (!nfsd4_has_session(cstate) &&
3561 STALE_CLIENTID(&lock->lk_new_clientid))
3564 /* validate and update open stateid and open seqid */
3565 status = nfs4_preprocess_seqid_op(cstate,
3566 lock->lk_new_open_seqid,
3567 &lock->lk_new_open_stateid,
3569 &lock->lk_replay_owner, &open_stp,
3573 open_sop = lock->lk_replay_owner;
3574 /* create lockowner and lock stateid */
3575 fp = open_stp->st_file;
3576 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3577 open_sop->so_client->cl_clientid.cl_id,
3578 &lock->v.new.owner);
3579 /* XXX: Do we need to check for duplicate stateowners on
3580 * the same file, or should they just be allowed (and
3581 * create new stateids)? */
3582 status = nfserr_resource;
3583 lock_sop = alloc_init_lock_stateowner(strhashval,
3584 open_sop->so_client, open_stp, lock);
3585 if (lock_sop == NULL)
3587 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3588 if (lock_stp == NULL)
3591 /* lock (lock owner + lock stateid) already exists */
3592 status = nfs4_preprocess_seqid_op(cstate,
3593 lock->lk_old_lock_seqid,
3594 &lock->lk_old_lock_stateid,
3596 &lock->lk_replay_owner, &lock_stp, lock);
3599 lock_sop = lock->lk_replay_owner;
3601 /* lock->lk_replay_owner and lock_stp have been created or found */
3602 filp = lock_stp->st_vfs_file;
3604 status = nfserr_grace;
3605 if (locks_in_grace() && !lock->lk_reclaim)
3607 status = nfserr_no_grace;
3608 if (!locks_in_grace() && lock->lk_reclaim)
3611 locks_init_lock(&file_lock);
3612 switch (lock->lk_type) {
3615 file_lock.fl_type = F_RDLCK;
3619 case NFS4_WRITEW_LT:
3620 file_lock.fl_type = F_WRLCK;
3624 status = nfserr_inval;
3627 file_lock.fl_owner = (fl_owner_t)lock_sop;
3628 file_lock.fl_pid = current->tgid;
3629 file_lock.fl_file = filp;
3630 file_lock.fl_flags = FL_POSIX;
3631 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3633 file_lock.fl_start = lock->lk_offset;
3634 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3635 nfs4_transform_lock_offset(&file_lock);
3638 * Try to lock the file in the VFS.
3639 * Note: locks.c uses the BKL to protect the inode's lock list.
3642 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3644 case 0: /* success! */
3645 update_stateid(&lock_stp->st_stateid);
3646 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3650 case (EAGAIN): /* conflock holds conflicting lock */
3651 status = nfserr_denied;
3652 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3653 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3656 status = nfserr_deadlock;
3659 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3660 status = nfserr_resource;
3664 if (status && lock->lk_is_new && lock_sop)
3665 release_lockowner(lock_sop);
3666 if (lock->lk_replay_owner) {
3667 nfs4_get_stateowner(lock->lk_replay_owner);
3668 cstate->replay_owner = lock->lk_replay_owner;
3670 nfs4_unlock_state();
3675 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3676 * so we do a temporary open here just to get an open file to pass to
3677 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3680 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3685 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3688 err = vfs_test_lock(file, lock);
3697 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3698 struct nfsd4_lockt *lockt)
3700 struct inode *inode;
3701 struct file_lock file_lock;
3705 if (locks_in_grace())
3706 return nfserr_grace;
3708 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3709 return nfserr_inval;
3711 lockt->lt_stateowner = NULL;
3714 status = nfserr_stale_clientid;
3715 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3718 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3719 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3720 if (status == nfserr_symlink)
3721 status = nfserr_inval;
3725 inode = cstate->current_fh.fh_dentry->d_inode;
3726 locks_init_lock(&file_lock);
3727 switch (lockt->lt_type) {
3730 file_lock.fl_type = F_RDLCK;
3733 case NFS4_WRITEW_LT:
3734 file_lock.fl_type = F_WRLCK;
3737 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3738 status = nfserr_inval;
3742 lockt->lt_stateowner = find_lockstateowner_str(inode,
3743 &lockt->lt_clientid, &lockt->lt_owner);
3744 if (lockt->lt_stateowner)
3745 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3746 file_lock.fl_pid = current->tgid;
3747 file_lock.fl_flags = FL_POSIX;
3749 file_lock.fl_start = lockt->lt_offset;
3750 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3752 nfs4_transform_lock_offset(&file_lock);
3755 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3757 status = nfserrno(error);
3760 if (file_lock.fl_type != F_UNLCK) {
3761 status = nfserr_denied;
3762 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3765 nfs4_unlock_state();
3770 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3771 struct nfsd4_locku *locku)
3773 struct nfs4_stateid *stp;
3774 struct file *filp = NULL;
3775 struct file_lock file_lock;
3779 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3780 (long long) locku->lu_offset,
3781 (long long) locku->lu_length);
3783 if (check_lock_length(locku->lu_offset, locku->lu_length))
3784 return nfserr_inval;
3788 if ((status = nfs4_preprocess_seqid_op(cstate,
3792 &locku->lu_stateowner, &stp, NULL)))
3795 filp = stp->st_vfs_file;
3797 locks_init_lock(&file_lock);
3798 file_lock.fl_type = F_UNLCK;
3799 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3800 file_lock.fl_pid = current->tgid;
3801 file_lock.fl_file = filp;
3802 file_lock.fl_flags = FL_POSIX;
3803 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3804 file_lock.fl_start = locku->lu_offset;
3806 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3807 nfs4_transform_lock_offset(&file_lock);
3810 * Try to unlock the file in the VFS.
3812 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3814 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3818 * OK, unlock succeeded; the only thing left to do is update the stateid.
3820 update_stateid(&stp->st_stateid);
3821 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3824 if (locku->lu_stateowner) {
3825 nfs4_get_stateowner(locku->lu_stateowner);
3826 cstate->replay_owner = locku->lu_stateowner;
3828 nfs4_unlock_state();
3832 status = nfserrno(err);
3838 * 1: locks held by lockowner
3839 * 0: no locks held by lockowner
3842 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3844 struct file_lock **flpp;
3845 struct inode *inode = filp->f_path.dentry->d_inode;
3849 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3850 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3861 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3862 struct nfsd4_compound_state *cstate,
3863 struct nfsd4_release_lockowner *rlockowner)
3865 clientid_t *clid = &rlockowner->rl_clientid;
3866 struct nfs4_stateowner *sop;
3867 struct nfs4_stateid *stp;
3868 struct xdr_netobj *owner = &rlockowner->rl_owner;
3869 struct list_head matches;
3873 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3874 clid->cl_boot, clid->cl_id);
3876 /* XXX check for lease expiration */
3878 status = nfserr_stale_clientid;
3879 if (STALE_CLIENTID(clid))
3884 status = nfserr_locks_held;
3885 /* XXX: we're doing a linear search through all the lockowners.
3886 * Yipes! For now we'll just hope clients aren't really using
3887 * release_lockowner much, but eventually we have to fix these
3888 * data structures. */
3889 INIT_LIST_HEAD(&matches);
3890 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3891 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3892 if (!same_owner_str(sop, owner, clid))
3894 list_for_each_entry(stp, &sop->so_stateids,
3896 if (check_for_locks(stp->st_vfs_file, sop))
3898 /* Note: so_perclient unused for lockowners,
3899 * so it's OK to fool with here. */
3900 list_add(&sop->so_perclient, &matches);
3904 /* Clients probably won't expect us to return with some (but not all)
3905 * of the lockowner state released; so don't release any until all
3906 * have been checked. */
3908 while (!list_empty(&matches)) {
3909 sop = list_entry(matches.next, struct nfs4_stateowner,
3911 /* unhash_stateowner deletes so_perclient only
3912 * for openowners. */
3913 list_del(&sop->so_perclient);
3914 release_lockowner(sop);
3917 nfs4_unlock_state();
3921 static inline struct nfs4_client_reclaim *
3924 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3928 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3930 unsigned int strhashval = clientstr_hashval(name);
3931 struct nfs4_client *clp;
3933 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3938 * failure => all reset bets are off, nfserr_no_grace...
3941 nfs4_client_to_reclaim(const char *name)
3943 unsigned int strhashval;
3944 struct nfs4_client_reclaim *crp = NULL;
3946 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3947 crp = alloc_reclaim();
3950 strhashval = clientstr_hashval(name);
3951 INIT_LIST_HEAD(&crp->cr_strhash);
3952 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3953 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3954 reclaim_str_hashtbl_size++;
3959 nfs4_release_reclaim(void)
3961 struct nfs4_client_reclaim *crp = NULL;
3964 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3965 while (!list_empty(&reclaim_str_hashtbl[i])) {
3966 crp = list_entry(reclaim_str_hashtbl[i].next,
3967 struct nfs4_client_reclaim, cr_strhash);
3968 list_del(&crp->cr_strhash);
3970 reclaim_str_hashtbl_size--;
3973 BUG_ON(reclaim_str_hashtbl_size);
3977 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3978 static struct nfs4_client_reclaim *
3979 nfs4_find_reclaim_client(clientid_t *clid)
3981 unsigned int strhashval;
3982 struct nfs4_client *clp;
3983 struct nfs4_client_reclaim *crp = NULL;
3986 /* find clientid in conf_id_hashtbl */
3987 clp = find_confirmed_client(clid);
3991 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3992 clp->cl_name.len, clp->cl_name.data,
3995 /* find clp->cl_name in reclaim_str_hashtbl */
3996 strhashval = clientstr_hashval(clp->cl_recdir);
3997 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3998 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4006 * Called from OPEN. Look for clientid in reclaim list.
4009 nfs4_check_open_reclaim(clientid_t *clid)
4011 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4014 /* initialization to perform at module load time: */
4017 nfs4_state_init(void)
4021 status = nfsd4_init_slabs();
4024 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4025 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4026 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4027 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4028 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4030 for (i = 0; i < SESSION_HASH_SIZE; i++)
4031 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4032 for (i = 0; i < FILE_HASH_SIZE; i++) {
4033 INIT_LIST_HEAD(&file_hashtbl[i]);
4035 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4036 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4037 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4039 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4040 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4041 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4043 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4044 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4045 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4047 memset(&onestateid, ~0, sizeof(stateid_t));
4048 INIT_LIST_HEAD(&close_lru);
4049 INIT_LIST_HEAD(&client_lru);
4050 INIT_LIST_HEAD(&del_recall_lru);
4051 for (i = 0; i < CLIENT_HASH_SIZE; i++)
4052 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4053 reclaim_str_hashtbl_size = 0;
4058 nfsd4_load_reboot_recovery_data(void)
4063 nfsd4_init_recdir(user_recovery_dirname);
4064 status = nfsd4_recdir_load();
4065 nfs4_unlock_state();
4067 printk("NFSD: Failure reading reboot recovery data\n");
4071 get_nfs4_grace_period(void)
4073 return max(user_lease_time, lease_time) * HZ;
4077 * Since the lifetime of a delegation isn't limited to that of an open, a
4078 * client may quite reasonably hang on to a delegation as long as it has
4079 * the inode cached. This becomes an obvious problem the first time a
4080 * client's inode cache approaches the size of the server's total memory.
4082 * For now we avoid this problem by imposing a hard limit on the number
4083 * of delegations, which varies according to the server's memory size.
4086 set_max_delegations(void)
4089 * Allow at most 4 delegations per megabyte of RAM. Quick
4090 * estimates suggest that in the worst case (where every delegation
4091 * is for a different inode), a delegation could take about 1.5K,
4092 * giving a worst case usage of about 6% of memory.
4094 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4097 /* initialization to perform when the nfsd service is started: */
4100 __nfs4_state_start(void)
4102 unsigned long grace_time;
4104 boot_time = get_seconds();
4105 grace_time = get_nfs4_grace_period();
4106 lease_time = user_lease_time;
4107 locks_start_grace(&nfsd4_manager);
4108 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4110 laundry_wq = create_singlethread_workqueue("nfsd4");
4111 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4112 set_max_delegations();
4116 nfs4_state_start(void)
4120 nfsd4_load_reboot_recovery_data();
4121 __nfs4_state_start();
4127 nfs4_lease_time(void)
4133 __nfs4_state_shutdown(void)
4136 struct nfs4_client *clp = NULL;
4137 struct nfs4_delegation *dp = NULL;
4138 struct list_head *pos, *next, reaplist;
4140 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4141 while (!list_empty(&conf_id_hashtbl[i])) {
4142 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4145 while (!list_empty(&unconf_str_hashtbl[i])) {
4146 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4150 INIT_LIST_HEAD(&reaplist);
4151 spin_lock(&recall_lock);
4152 list_for_each_safe(pos, next, &del_recall_lru) {
4153 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4154 list_move(&dp->dl_recall_lru, &reaplist);
4156 spin_unlock(&recall_lock);
4157 list_for_each_safe(pos, next, &reaplist) {
4158 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4159 list_del_init(&dp->dl_recall_lru);
4160 unhash_delegation(dp);
4163 nfsd4_shutdown_recdir();
4168 nfs4_state_shutdown(void)
4170 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4171 destroy_workqueue(laundry_wq);
4172 locks_end_grace(&nfsd4_manager);
4174 nfs4_release_reclaim();
4175 __nfs4_state_shutdown();
4176 nfs4_unlock_state();
4180 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4181 * accessed when nfsd is starting.
4184 nfs4_set_recdir(char *recdir)
4186 strcpy(user_recovery_dirname, recdir);
4190 * Change the NFSv4 recovery directory to recdir.
4193 nfs4_reset_recoverydir(char *recdir)
4198 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4202 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4203 nfs4_set_recdir(recdir);
4211 nfs4_recoverydir(void)
4213 return user_recovery_dirname;
4217 * Called when leasetime is changed.
4219 * The only way the protocol gives us to handle on-the-fly lease changes is to
4220 * simulate a reboot. Instead of doing that, we just wait till the next time
4221 * we start to register any changes in lease time. If the administrator
4222 * really wants to change the lease time *now*, they can go ahead and bring
4223 * nfsd down and then back up again after changing the lease time.
4225 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4226 * when nfsd is starting
4229 nfs4_reset_lease(time_t leasetime)
4231 user_lease_time = leasetime;