2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
59 #define NFSDDBG_FACILITY NFSDDBG_PROC
62 static time_t lease_time = 90; /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
69 static stateid_t zerostateid; /* bits all 0 */
70 static stateid_t onestateid; /* bits all 1 */
71 static u64 current_sessionid = 1;
73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
74 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
76 /* forward declarations */
77 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
78 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir);
84 /* Currently used for almost all code touching nfsv4 state: */
85 static DEFINE_MUTEX(client_mutex);
88 * Currently used for the del_recall_lru and file hash table. In an
89 * effort to decrease the scope of the client_mutex, this spinlock may
90 * eventually cover more:
92 static DEFINE_SPINLOCK(recall_lock);
94 static struct kmem_cache *stateowner_slab = NULL;
95 static struct kmem_cache *file_slab = NULL;
96 static struct kmem_cache *stateid_slab = NULL;
97 static struct kmem_cache *deleg_slab = NULL;
100 nfs4_lock_state(void)
102 mutex_lock(&client_mutex);
106 nfs4_unlock_state(void)
108 mutex_unlock(&client_mutex);
112 opaque_hashval(const void *ptr, int nbytes)
114 unsigned char *cptr = (unsigned char *) ptr;
124 static struct list_head del_recall_lru;
127 put_nfs4_file(struct nfs4_file *fi)
129 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
130 list_del(&fi->fi_hash);
131 spin_unlock(&recall_lock);
133 kmem_cache_free(file_slab, fi);
138 get_nfs4_file(struct nfs4_file *fi)
140 atomic_inc(&fi->fi_ref);
143 static int num_delegations;
144 unsigned int max_delegations;
147 * Open owner state (share locks)
150 /* hash tables for nfs4_stateowner */
151 #define OWNER_HASH_BITS 8
152 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
153 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
155 #define ownerid_hashval(id) \
156 ((id) & OWNER_HASH_MASK)
157 #define ownerstr_hashval(clientid, ownername) \
158 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
160 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
161 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
163 /* hash table for nfs4_file */
164 #define FILE_HASH_BITS 8
165 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
166 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
167 /* hash table for (open)nfs4_stateid */
168 #define STATEID_HASH_BITS 10
169 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
170 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
172 #define file_hashval(x) \
173 hash_ptr(x, FILE_HASH_BITS)
174 #define stateid_hashval(owner_id, file_id) \
175 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
177 static struct list_head file_hashtbl[FILE_HASH_SIZE];
178 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
180 static struct nfs4_delegation *
181 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
183 struct nfs4_delegation *dp;
184 struct nfs4_file *fp = stp->st_file;
185 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
187 dprintk("NFSD alloc_init_deleg\n");
188 if (fp->fi_had_conflict)
190 if (num_delegations > max_delegations)
192 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
196 INIT_LIST_HEAD(&dp->dl_perfile);
197 INIT_LIST_HEAD(&dp->dl_perclnt);
198 INIT_LIST_HEAD(&dp->dl_recall_lru);
203 get_file(stp->st_vfs_file);
204 dp->dl_vfs_file = stp->st_vfs_file;
206 dp->dl_ident = cb->cb_ident;
208 dp->dl_stateid.si_boot = get_seconds();
209 dp->dl_stateid.si_stateownerid = current_delegid++;
210 dp->dl_stateid.si_fileid = 0;
211 dp->dl_stateid.si_generation = 0;
212 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
214 atomic_set(&dp->dl_count, 1);
215 list_add(&dp->dl_perfile, &fp->fi_delegations);
216 list_add(&dp->dl_perclnt, &clp->cl_delegations);
221 nfs4_put_delegation(struct nfs4_delegation *dp)
223 if (atomic_dec_and_test(&dp->dl_count)) {
224 dprintk("NFSD: freeing dp %p\n",dp);
225 put_nfs4_file(dp->dl_file);
226 kmem_cache_free(deleg_slab, dp);
231 /* Remove the associated file_lock first, then remove the delegation.
232 * lease_modify() is called to remove the FS_LEASE file_lock from
233 * the i_flock list, eventually calling nfsd's lock_manager
234 * fl_release_callback.
237 nfs4_close_delegation(struct nfs4_delegation *dp)
239 struct file *filp = dp->dl_vfs_file;
241 dprintk("NFSD: close_delegation dp %p\n",dp);
242 dp->dl_vfs_file = NULL;
243 /* The following nfsd_close may not actually close the file,
244 * but we want to remove the lease in any case. */
246 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
250 /* Called under the state lock. */
252 unhash_delegation(struct nfs4_delegation *dp)
254 list_del_init(&dp->dl_perfile);
255 list_del_init(&dp->dl_perclnt);
256 spin_lock(&recall_lock);
257 list_del_init(&dp->dl_recall_lru);
258 spin_unlock(&recall_lock);
259 nfs4_close_delegation(dp);
260 nfs4_put_delegation(dp);
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS 4
269 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
272 #define clientid_hashval(id) \
273 ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
277 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278 * used in reboot/reset lease grace period processing
280 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281 * setclientid_confirmed info.
283 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
286 * client_lru holds client queue ordered by nfs4_client.cl_time
289 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290 * for last close replay.
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
303 list_del(&stp->st_hash);
304 list_del(&stp->st_perfile);
305 list_del(&stp->st_perstateowner);
308 static void free_generic_stateid(struct nfs4_stateid *stp)
310 put_nfs4_file(stp->st_file);
311 kmem_cache_free(stateid_slab, stp);
314 static void release_lock_stateid(struct nfs4_stateid *stp)
316 unhash_generic_stateid(stp);
317 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318 free_generic_stateid(stp);
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
323 struct nfs4_stateid *stp;
325 list_del(&sop->so_idhash);
326 list_del(&sop->so_strhash);
327 list_del(&sop->so_perstateid);
328 while (!list_empty(&sop->so_stateids)) {
329 stp = list_first_entry(&sop->so_stateids,
330 struct nfs4_stateid, st_perstateowner);
331 release_lock_stateid(stp);
335 static void release_lockowner(struct nfs4_stateowner *sop)
337 unhash_lockowner(sop);
338 nfs4_put_stateowner(sop);
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
344 struct nfs4_stateowner *lock_sop;
346 while (!list_empty(&open_stp->st_lockowners)) {
347 lock_sop = list_entry(open_stp->st_lockowners.next,
348 struct nfs4_stateowner, so_perstateid);
349 /* list_del(&open_stp->st_lockowners); */
350 BUG_ON(lock_sop->so_is_open_owner);
351 release_lockowner(lock_sop);
355 static void release_open_stateid(struct nfs4_stateid *stp)
357 unhash_generic_stateid(stp);
358 release_stateid_lockowners(stp);
359 nfsd_close(stp->st_vfs_file);
360 free_generic_stateid(stp);
363 static void unhash_openowner(struct nfs4_stateowner *sop)
365 struct nfs4_stateid *stp;
367 list_del(&sop->so_idhash);
368 list_del(&sop->so_strhash);
369 list_del(&sop->so_perclient);
370 list_del(&sop->so_perstateid); /* XXX: necessary? */
371 while (!list_empty(&sop->so_stateids)) {
372 stp = list_first_entry(&sop->so_stateids,
373 struct nfs4_stateid, st_perstateowner);
374 release_open_stateid(stp);
378 static void release_openowner(struct nfs4_stateowner *sop)
380 unhash_openowner(sop);
381 list_del(&sop->so_close_lru);
382 nfs4_put_stateowner(sop);
385 static DEFINE_SPINLOCK(sessionid_lock);
386 #define SESSION_HASH_SIZE 512
387 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
390 hash_sessionid(struct nfs4_sessionid *sessionid)
392 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
394 return sid->sequence % SESSION_HASH_SIZE;
398 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
400 u32 *ptr = (u32 *)(&sessionid->data[0]);
401 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
405 gen_sessionid(struct nfsd4_session *ses)
407 struct nfs4_client *clp = ses->se_client;
408 struct nfsd4_sessionid *sid;
410 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
411 sid->clientid = clp->cl_clientid;
412 sid->sequence = current_sessionid++;
417 * Give the client the number of slots it requests bound by
418 * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
420 * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
421 * should (up to a point) re-negotiate active sessions and reduce their
422 * slot usage to make rooom for new connections. For now we just fail the
425 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
427 int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
429 spin_lock(&nfsd_serv->sv_lock);
430 if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
431 np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
432 nfsd_serv->sv_drc_pages_used += np;
433 spin_unlock(&nfsd_serv->sv_lock);
436 status = nfserr_resource;
439 fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
445 * fchan holds the client values on input, and the server values on output
447 static int init_forechannel_attrs(struct svc_rqst *rqstp,
448 struct nfsd4_session *session,
449 struct nfsd4_channel_attrs *fchan)
452 __u32 maxcount = svc_max_payload(rqstp);
454 /* headerpadsz set to zero in encode routine */
456 /* Use the client's max request and max response size if possible */
457 if (fchan->maxreq_sz > maxcount)
458 fchan->maxreq_sz = maxcount;
459 session->se_fmaxreq_sz = fchan->maxreq_sz;
461 if (fchan->maxresp_sz > maxcount)
462 fchan->maxresp_sz = maxcount;
463 session->se_fmaxresp_sz = fchan->maxresp_sz;
465 /* Set the max response cached size our default which is
466 * a multiple of PAGE_SIZE and small */
467 session->se_fmaxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
468 fchan->maxresp_cached = session->se_fmaxresp_cached;
470 /* Use the client's maxops if possible */
471 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
472 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
473 session->se_fmaxops = fchan->maxops;
475 /* try to use the client requested number of slots */
476 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
477 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
479 /* FIXME: Error means no more DRC pages so the server should
480 * recover pages from existing sessions. For now fail session
483 status = set_forechannel_maxreqs(fchan);
485 session->se_fnumslots = fchan->maxreqs;
490 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
491 struct nfsd4_create_session *cses)
493 struct nfsd4_session *new, tmp;
494 int idx, status = nfserr_resource, slotsize;
496 memset(&tmp, 0, sizeof(tmp));
498 /* FIXME: For now, we just accept the client back channel attributes. */
499 status = init_forechannel_attrs(rqstp, &tmp, &cses->fore_channel);
503 /* allocate struct nfsd4_session and slot table in one piece */
504 slotsize = tmp.se_fnumslots * sizeof(struct nfsd4_slot);
505 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
509 memcpy(new, &tmp, sizeof(*new));
511 new->se_client = clp;
513 idx = hash_sessionid(&new->se_sessionid);
514 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
515 NFS4_MAX_SESSIONID_LEN);
517 new->se_flags = cses->flags;
518 kref_init(&new->se_ref);
519 spin_lock(&sessionid_lock);
520 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
521 list_add(&new->se_perclnt, &clp->cl_sessions);
522 spin_unlock(&sessionid_lock);
529 /* caller must hold sessionid_lock */
530 static struct nfsd4_session *
531 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
533 struct nfsd4_session *elem;
536 dump_sessionid(__func__, sessionid);
537 idx = hash_sessionid(sessionid);
538 dprintk("%s: idx is %d\n", __func__, idx);
539 /* Search in the appropriate list */
540 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
541 dump_sessionid("list traversal", &elem->se_sessionid);
542 if (!memcmp(elem->se_sessionid.data, sessionid->data,
543 NFS4_MAX_SESSIONID_LEN)) {
548 dprintk("%s: session not found\n", __func__);
552 /* caller must hold sessionid_lock */
554 unhash_session(struct nfsd4_session *ses)
556 list_del(&ses->se_hash);
557 list_del(&ses->se_perclnt);
561 release_session(struct nfsd4_session *ses)
563 spin_lock(&sessionid_lock);
565 spin_unlock(&sessionid_lock);
566 nfsd4_put_session(ses);
569 static void nfsd4_release_respages(struct page **respages, short resused);
572 free_session(struct kref *kref)
574 struct nfsd4_session *ses;
577 ses = container_of(kref, struct nfsd4_session, se_ref);
578 for (i = 0; i < ses->se_fnumslots; i++) {
579 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
580 nfsd4_release_respages(e->ce_respages, e->ce_resused);
582 kfree(ses->se_slots);
587 renew_client(struct nfs4_client *clp)
590 * Move client to the end to the LRU list.
592 dprintk("renewing client (clientid %08x/%08x)\n",
593 clp->cl_clientid.cl_boot,
594 clp->cl_clientid.cl_id);
595 list_move_tail(&clp->cl_lru, &client_lru);
596 clp->cl_time = get_seconds();
599 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
601 STALE_CLIENTID(clientid_t *clid)
603 if (clid->cl_boot == boot_time)
605 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
606 clid->cl_boot, clid->cl_id, boot_time);
611 * XXX Should we use a slab cache ?
612 * This type of memory management is somewhat inefficient, but we use it
613 * anyway since SETCLIENTID is not a common operation.
615 static struct nfs4_client *alloc_client(struct xdr_netobj name)
617 struct nfs4_client *clp;
619 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
622 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
623 if (clp->cl_name.data == NULL) {
627 memcpy(clp->cl_name.data, name.data, name.len);
628 clp->cl_name.len = name.len;
633 shutdown_callback_client(struct nfs4_client *clp)
635 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
639 * Callback threads take a reference on the client, so there
640 * should be no outstanding callbacks at this point.
642 clp->cl_cb_conn.cb_client = NULL;
643 rpc_shutdown_client(clnt);
645 if (clp->cl_cb_conn.cb_cred) {
646 put_rpccred(clp->cl_cb_conn.cb_cred);
647 clp->cl_cb_conn.cb_cred = NULL;
652 free_client(struct nfs4_client *clp)
654 shutdown_callback_client(clp);
655 nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
656 clp->cl_slot.sl_cache_entry.ce_resused);
657 if (clp->cl_cred.cr_group_info)
658 put_group_info(clp->cl_cred.cr_group_info);
659 kfree(clp->cl_principal);
660 kfree(clp->cl_name.data);
665 put_nfs4_client(struct nfs4_client *clp)
667 if (atomic_dec_and_test(&clp->cl_count))
672 expire_client(struct nfs4_client *clp)
674 struct nfs4_stateowner *sop;
675 struct nfs4_delegation *dp;
676 struct list_head reaplist;
678 dprintk("NFSD: expire_client cl_count %d\n",
679 atomic_read(&clp->cl_count));
681 INIT_LIST_HEAD(&reaplist);
682 spin_lock(&recall_lock);
683 while (!list_empty(&clp->cl_delegations)) {
684 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
685 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
687 list_del_init(&dp->dl_perclnt);
688 list_move(&dp->dl_recall_lru, &reaplist);
690 spin_unlock(&recall_lock);
691 while (!list_empty(&reaplist)) {
692 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
693 list_del_init(&dp->dl_recall_lru);
694 unhash_delegation(dp);
696 list_del(&clp->cl_idhash);
697 list_del(&clp->cl_strhash);
698 list_del(&clp->cl_lru);
699 while (!list_empty(&clp->cl_openowners)) {
700 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
701 release_openowner(sop);
703 while (!list_empty(&clp->cl_sessions)) {
704 struct nfsd4_session *ses;
705 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
707 release_session(ses);
709 put_nfs4_client(clp);
712 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
714 struct nfs4_client *clp;
716 clp = alloc_client(name);
719 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
720 atomic_set(&clp->cl_count, 1);
721 atomic_set(&clp->cl_cb_conn.cb_set, 0);
722 INIT_LIST_HEAD(&clp->cl_idhash);
723 INIT_LIST_HEAD(&clp->cl_strhash);
724 INIT_LIST_HEAD(&clp->cl_openowners);
725 INIT_LIST_HEAD(&clp->cl_delegations);
726 INIT_LIST_HEAD(&clp->cl_sessions);
727 INIT_LIST_HEAD(&clp->cl_lru);
731 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
733 memcpy(target->cl_verifier.data, source->data,
734 sizeof(target->cl_verifier.data));
737 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
739 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
740 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
743 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
745 target->cr_uid = source->cr_uid;
746 target->cr_gid = source->cr_gid;
747 target->cr_group_info = source->cr_group_info;
748 get_group_info(target->cr_group_info);
751 static int same_name(const char *n1, const char *n2)
753 return 0 == memcmp(n1, n2, HEXDIR_LEN);
757 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
759 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
763 same_clid(clientid_t *cl1, clientid_t *cl2)
765 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
768 /* XXX what about NGROUP */
770 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
772 return cr1->cr_uid == cr2->cr_uid;
775 static void gen_clid(struct nfs4_client *clp)
777 static u32 current_clientid = 1;
779 clp->cl_clientid.cl_boot = boot_time;
780 clp->cl_clientid.cl_id = current_clientid++;
783 static void gen_confirm(struct nfs4_client *clp)
788 p = (u32 *)clp->cl_confirm.data;
789 *p++ = get_seconds();
793 static int check_name(struct xdr_netobj name)
797 if (name.len > NFS4_OPAQUE_LIMIT) {
798 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
805 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
807 unsigned int idhashval;
809 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
810 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
811 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
812 list_add_tail(&clp->cl_lru, &client_lru);
813 clp->cl_time = get_seconds();
817 move_to_confirmed(struct nfs4_client *clp)
819 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
820 unsigned int strhashval;
822 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
823 list_del_init(&clp->cl_strhash);
824 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
825 strhashval = clientstr_hashval(clp->cl_recdir);
826 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
830 static struct nfs4_client *
831 find_confirmed_client(clientid_t *clid)
833 struct nfs4_client *clp;
834 unsigned int idhashval = clientid_hashval(clid->cl_id);
836 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
837 if (same_clid(&clp->cl_clientid, clid))
843 static struct nfs4_client *
844 find_unconfirmed_client(clientid_t *clid)
846 struct nfs4_client *clp;
847 unsigned int idhashval = clientid_hashval(clid->cl_id);
849 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
850 if (same_clid(&clp->cl_clientid, clid))
857 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
858 * parameter. Matching is based on the fact the at least one of the
859 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
861 * FIXME: we need to unify the clientid namespaces for nfsv4.x
862 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
863 * and SET_CLIENTID{,_CONFIRM}
866 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
868 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
869 return use_exchange_id == has_exchange_flags;
872 static struct nfs4_client *
873 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
874 bool use_exchange_id)
876 struct nfs4_client *clp;
878 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
879 if (same_name(clp->cl_recdir, dname) &&
880 match_clientid_establishment(clp, use_exchange_id))
886 static struct nfs4_client *
887 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
888 bool use_exchange_id)
890 struct nfs4_client *clp;
892 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
893 if (same_name(clp->cl_recdir, dname) &&
894 match_clientid_establishment(clp, use_exchange_id))
900 /* a helper function for parse_callback */
902 parse_octet(unsigned int *lenp, char **addrp)
904 unsigned int len = *lenp;
916 if ((c < '0') || (c > '9')) {
922 n = (n * 10) + (c - '0');
933 /* parse and set the setclientid ipv4 callback address */
935 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
940 u32 addrlen = addr_len;
941 char *addr = addr_val;
946 for(i = 4; i > 0 ; i--) {
947 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
950 cbaddr |= (temp << shift);
958 for(i = 2; i > 0 ; i--) {
959 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
962 cbport |= (temp << shift);
971 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
973 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
975 /* Currently, we only support tcp for the callback channel */
976 if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
979 if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
980 &cb->cb_addr, &cb->cb_port)))
982 cb->cb_prog = se->se_callback_prog;
983 cb->cb_ident = se->se_callback_ident;
986 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
987 "will not receive delegations\n",
988 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
994 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
996 struct nfsd4_compoundres *resp = rqstp->rq_resp;
998 resp->cstate.statp = statp;
1002 * Dereference the result pages.
1005 nfsd4_release_respages(struct page **respages, short resused)
1009 dprintk("--> %s\n", __func__);
1010 for (i = 0; i < resused; i++) {
1013 put_page(respages[i]);
1019 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1023 for (i = 0; i < count; i++) {
1024 topages[i] = frompages[i];
1027 get_page(topages[i]);
1032 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1033 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1034 * length of the XDR response is less than se_fmaxresp_cached
1035 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1036 * of the reply (e.g. readdir).
1038 * Store the base and length of the rq_req.head[0] page
1039 * of the NFSv4.1 data, just past the rpc header.
1042 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1044 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1045 struct svc_rqst *rqstp = resp->rqstp;
1046 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1047 struct nfsd4_op *op = &args->ops[resp->opcnt];
1048 struct kvec *resv = &rqstp->rq_res.head[0];
1050 dprintk("--> %s entry %p\n", __func__, entry);
1052 /* Don't cache a failed OP_SEQUENCE. */
1053 if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1056 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1057 entry->ce_opcnt = resp->opcnt;
1058 entry->ce_status = resp->cstate.status;
1061 * Don't need a page to cache just the sequence operation - the slot
1065 if (nfsd4_not_cached(resp)) {
1066 entry->ce_resused = 0;
1067 entry->ce_rpchdrlen = 0;
1068 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1069 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1072 entry->ce_resused = rqstp->rq_resused;
1073 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1074 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1075 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1077 entry->ce_datav.iov_base = resp->cstate.statp;
1078 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1079 (char *)page_address(rqstp->rq_respages[0]));
1080 /* Current request rpc header length*/
1081 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1082 (char *)page_address(rqstp->rq_respages[0]);
1086 * We keep the rpc header, but take the nfs reply from the replycache.
1089 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1090 struct nfsd4_cache_entry *entry)
1092 struct svc_rqst *rqstp = resp->rqstp;
1093 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1096 /* Current request rpc header length*/
1097 len = (char *)resp->cstate.statp -
1098 (char *)page_address(rqstp->rq_respages[0]);
1099 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1100 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1101 entry->ce_datav.iov_len);
1104 /* copy the cached reply nfsd data past the current rpc header */
1105 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1106 entry->ce_datav.iov_len);
1107 resv->iov_len = len + entry->ce_datav.iov_len;
1112 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1113 * cached page. Replace any futher replay pages from the cache.
1116 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1117 struct nfsd4_sequence *seq)
1119 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1122 dprintk("--> %s entry %p\n", __func__, entry);
1125 * If this is just the sequence operation, we did not keep
1126 * a page in the cache entry because we can just use the
1127 * slot info stored in struct nfsd4_sequence that was checked
1128 * against the slot in nfsd4_sequence().
1130 * This occurs when seq->cachethis is FALSE, or when the client
1131 * session inactivity timer fires and a solo sequence operation
1132 * is sent (lease renewal).
1134 if (seq && nfsd4_not_cached(resp)) {
1135 seq->maxslots = resp->cstate.session->se_fnumslots;
1139 if (!nfsd41_copy_replay_data(resp, entry)) {
1141 * Not enough room to use the replay rpc header, send the
1142 * cached header. Release all the allocated result pages.
1144 svc_free_res_pages(resp->rqstp);
1145 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1148 /* Release all but the first allocated result page */
1150 resp->rqstp->rq_resused--;
1151 svc_free_res_pages(resp->rqstp);
1153 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1154 &entry->ce_respages[1],
1155 entry->ce_resused - 1);
1158 resp->rqstp->rq_resused = entry->ce_resused;
1159 resp->opcnt = entry->ce_opcnt;
1160 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1161 status = entry->ce_status;
1167 * Set the exchange_id flags returned by the server.
1170 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1172 /* pNFS is not supported */
1173 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1175 /* Referrals are supported, Migration is not. */
1176 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1178 /* set the wire flags to return to client. */
1179 clid->flags = new->cl_exchange_flags;
1183 nfsd4_exchange_id(struct svc_rqst *rqstp,
1184 struct nfsd4_compound_state *cstate,
1185 struct nfsd4_exchange_id *exid)
1187 struct nfs4_client *unconf, *conf, *new;
1189 unsigned int strhashval;
1190 char dname[HEXDIR_LEN];
1191 nfs4_verifier verf = exid->verifier;
1192 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1194 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1195 " ip_addr=%u flags %x, spa_how %d\n",
1196 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1197 ip_addr, exid->flags, exid->spa_how);
1199 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1200 return nfserr_inval;
1202 /* Currently only support SP4_NONE */
1203 switch (exid->spa_how) {
1207 return nfserr_encr_alg_unsupp;
1209 BUG(); /* checked by xdr code */
1211 return nfserr_serverfault; /* no excuse :-/ */
1214 status = nfs4_make_rec_clidname(dname, &exid->clname);
1219 strhashval = clientstr_hashval(dname);
1224 conf = find_confirmed_client_by_str(dname, strhashval, true);
1226 if (!same_verf(&verf, &conf->cl_verifier)) {
1227 /* 18.35.4 case 8 */
1228 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1229 status = nfserr_not_same;
1232 /* Client reboot: destroy old state */
1233 expire_client(conf);
1236 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1237 /* 18.35.4 case 9 */
1238 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1239 status = nfserr_perm;
1242 expire_client(conf);
1245 if (ip_addr != conf->cl_addr &&
1246 !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
1247 /* Client collision. 18.35.4 case 3 */
1248 status = nfserr_clid_inuse;
1252 * Set bit when the owner id and verifier map to an already
1253 * confirmed client id (18.35.3).
1255 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1258 * Falling into 18.35.4 case 2, possible router replay.
1259 * Leave confirmed record intact and return same result.
1261 copy_verf(conf, &verf);
1265 /* 18.35.4 case 7 */
1266 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1267 status = nfserr_noent;
1272 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1275 * Possible retry or client restart. Per 18.35.4 case 4,
1276 * a new unconfirmed record should be generated regardless
1277 * of whether any properties have changed.
1279 expire_client(unconf);
1284 new = create_client(exid->clname, dname);
1286 status = nfserr_resource;
1290 copy_verf(new, &verf);
1291 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1292 new->cl_addr = ip_addr;
1295 add_to_unconfirmed(new, strhashval);
1297 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1298 exid->clientid.cl_id = new->cl_clientid.cl_id;
1300 new->cl_slot.sl_seqid = 0;
1302 nfsd4_set_ex_flags(new, exid);
1304 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1305 new->cl_slot.sl_seqid, new->cl_exchange_flags);
1309 nfs4_unlock_state();
1311 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1316 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1318 dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1321 /* The slot is in use, and no response has been sent. */
1322 if (slot->sl_inuse) {
1323 if (seqid == slot->sl_seqid)
1324 return nfserr_jukebox;
1326 return nfserr_seq_misordered;
1329 if (likely(seqid == slot->sl_seqid + 1))
1332 if (seqid == slot->sl_seqid)
1333 return nfserr_replay_cache;
1335 if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1337 /* Misordered replay or misordered new request */
1338 return nfserr_seq_misordered;
1342 nfsd4_create_session(struct svc_rqst *rqstp,
1343 struct nfsd4_compound_state *cstate,
1344 struct nfsd4_create_session *cr_ses)
1346 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1347 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1348 struct nfs4_client *conf, *unconf;
1349 struct nfsd4_slot *slot = NULL;
1353 unconf = find_unconfirmed_client(&cr_ses->clientid);
1354 conf = find_confirmed_client(&cr_ses->clientid);
1357 slot = &conf->cl_slot;
1358 status = check_slot_seqid(cr_ses->seqid, slot);
1359 if (status == nfserr_replay_cache) {
1360 dprintk("Got a create_session replay! seqid= %d\n",
1362 cstate->slot = slot;
1363 cstate->status = status;
1364 /* Return the cached reply status */
1365 status = nfsd4_replay_cache_entry(resp, NULL);
1367 } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
1368 status = nfserr_seq_misordered;
1369 dprintk("Sequence misordered!\n");
1370 dprintk("Expected seqid= %d but got seqid= %d\n",
1371 slot->sl_seqid, cr_ses->seqid);
1374 conf->cl_slot.sl_seqid++;
1375 } else if (unconf) {
1376 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1377 (ip_addr != unconf->cl_addr)) {
1378 status = nfserr_clid_inuse;
1382 slot = &unconf->cl_slot;
1383 status = check_slot_seqid(cr_ses->seqid, slot);
1385 /* an unconfirmed replay returns misordered */
1386 status = nfserr_seq_misordered;
1390 slot->sl_seqid++; /* from 0 to 1 */
1391 move_to_confirmed(unconf);
1394 * We do not support RDMA or persistent sessions
1396 cr_ses->flags &= ~SESSION4_PERSIST;
1397 cr_ses->flags &= ~SESSION4_RDMA;
1401 status = nfserr_stale_clientid;
1405 status = alloc_init_session(rqstp, conf, cr_ses);
1409 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1410 NFS4_MAX_SESSIONID_LEN);
1411 cr_ses->seqid = slot->sl_seqid;
1413 slot->sl_inuse = true;
1414 cstate->slot = slot;
1415 /* Ensure a page is used for the cache */
1416 slot->sl_cache_entry.ce_cachethis = 1;
1418 nfs4_unlock_state();
1419 dprintk("%s returns %d\n", __func__, ntohl(status));
1424 nfsd4_destroy_session(struct svc_rqst *r,
1425 struct nfsd4_compound_state *cstate,
1426 struct nfsd4_destroy_session *sessionid)
1428 struct nfsd4_session *ses;
1429 u32 status = nfserr_badsession;
1432 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1433 * - Should we return nfserr_back_chan_busy if waiting for
1434 * callbacks on to-be-destroyed session?
1435 * - Do we need to clear any callback info from previous session?
1438 dump_sessionid(__func__, &sessionid->sessionid);
1439 spin_lock(&sessionid_lock);
1440 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1442 spin_unlock(&sessionid_lock);
1446 unhash_session(ses);
1447 spin_unlock(&sessionid_lock);
1449 /* wait for callbacks */
1450 shutdown_callback_client(ses->se_client);
1451 nfsd4_put_session(ses);
1454 dprintk("%s returns %d\n", __func__, ntohl(status));
1459 nfsd4_sequence(struct svc_rqst *rqstp,
1460 struct nfsd4_compound_state *cstate,
1461 struct nfsd4_sequence *seq)
1463 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1464 struct nfsd4_session *session;
1465 struct nfsd4_slot *slot;
1468 if (resp->opcnt != 1)
1469 return nfserr_sequence_pos;
1471 spin_lock(&sessionid_lock);
1472 status = nfserr_badsession;
1473 session = find_in_sessionid_hashtbl(&seq->sessionid);
1477 status = nfserr_badslot;
1478 if (seq->slotid >= session->se_fnumslots)
1481 slot = &session->se_slots[seq->slotid];
1482 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1484 status = check_slot_seqid(seq->seqid, slot);
1485 if (status == nfserr_replay_cache) {
1486 cstate->slot = slot;
1487 cstate->session = session;
1488 /* Return the cached reply status and set cstate->status
1489 * for nfsd4_svc_encode_compoundres processing */
1490 status = nfsd4_replay_cache_entry(resp, seq);
1491 cstate->status = nfserr_replay_cache;
1497 /* Success! bump slot seqid */
1498 slot->sl_inuse = true;
1499 slot->sl_seqid = seq->seqid;
1500 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1501 /* Always set the cache entry cachethis for solo sequence */
1502 if (nfsd4_is_solo_sequence(resp))
1503 slot->sl_cache_entry.ce_cachethis = 1;
1505 cstate->slot = slot;
1506 cstate->session = session;
1509 /* Renew the clientid on success and on replay.
1510 * Hold a session reference until done processing the compound:
1511 * nfsd4_put_session called only if the cstate slot is set.
1513 renew_client(session->se_client);
1514 nfsd4_get_session(session);
1516 spin_unlock(&sessionid_lock);
1517 dprintk("%s: return %d\n", __func__, ntohl(status));
1522 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1523 struct nfsd4_setclientid *setclid)
1525 struct sockaddr_in *sin = svc_addr_in(rqstp);
1526 struct xdr_netobj clname = {
1527 .len = setclid->se_namelen,
1528 .data = setclid->se_name,
1530 nfs4_verifier clverifier = setclid->se_verf;
1531 unsigned int strhashval;
1532 struct nfs4_client *conf, *unconf, *new;
1535 char dname[HEXDIR_LEN];
1537 if (!check_name(clname))
1538 return nfserr_inval;
1540 status = nfs4_make_rec_clidname(dname, &clname);
1545 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1546 * We get here on a DRC miss.
1549 strhashval = clientstr_hashval(dname);
1552 conf = find_confirmed_client_by_str(dname, strhashval, false);
1554 /* RFC 3530 14.2.33 CASE 0: */
1555 status = nfserr_clid_inuse;
1556 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1557 dprintk("NFSD: setclientid: string in use by client"
1558 " at %pI4\n", &conf->cl_addr);
1563 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1564 * has a description of SETCLIENTID request processing consisting
1565 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1567 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1568 status = nfserr_resource;
1571 * RFC 3530 14.2.33 CASE 4:
1572 * placed first, because it is the normal case
1575 expire_client(unconf);
1576 new = create_client(clname, dname);
1580 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1582 * RFC 3530 14.2.33 CASE 1:
1583 * probable callback update
1586 /* Note this is removing unconfirmed {*x***},
1587 * which is stronger than RFC recommended {vxc**}.
1588 * This has the advantage that there is at most
1589 * one {*x***} in either list at any time.
1591 expire_client(unconf);
1593 new = create_client(clname, dname);
1596 copy_clid(new, conf);
1597 } else if (!unconf) {
1599 * RFC 3530 14.2.33 CASE 2:
1600 * probable client reboot; state will be removed if
1603 new = create_client(clname, dname);
1609 * RFC 3530 14.2.33 CASE 3:
1610 * probable client reboot; state will be removed if
1613 expire_client(unconf);
1614 new = create_client(clname, dname);
1619 copy_verf(new, &clverifier);
1620 new->cl_addr = sin->sin_addr.s_addr;
1621 new->cl_flavor = rqstp->rq_flavor;
1622 princ = svc_gss_principal(rqstp);
1624 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1625 if (new->cl_principal == NULL) {
1630 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1632 gen_callback(new, setclid);
1633 add_to_unconfirmed(new, strhashval);
1634 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1635 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1636 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1639 nfs4_unlock_state();
1645 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1646 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1647 * bullets, labeled as CASE1 - CASE4 below.
1650 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1651 struct nfsd4_compound_state *cstate,
1652 struct nfsd4_setclientid_confirm *setclientid_confirm)
1654 struct sockaddr_in *sin = svc_addr_in(rqstp);
1655 struct nfs4_client *conf, *unconf;
1656 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1657 clientid_t * clid = &setclientid_confirm->sc_clientid;
1660 if (STALE_CLIENTID(clid))
1661 return nfserr_stale_clientid;
1663 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1664 * We get here on a DRC miss.
1669 conf = find_confirmed_client(clid);
1670 unconf = find_unconfirmed_client(clid);
1672 status = nfserr_clid_inuse;
1673 if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1675 if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1679 * section 14.2.34 of RFC 3530 has a description of
1680 * SETCLIENTID_CONFIRM request processing consisting
1681 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1683 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1685 * RFC 3530 14.2.34 CASE 1:
1688 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1689 status = nfserr_clid_inuse;
1691 /* XXX: We just turn off callbacks until we can handle
1692 * change request correctly. */
1693 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1694 expire_client(unconf);
1698 } else if (conf && !unconf) {
1700 * RFC 3530 14.2.34 CASE 2:
1701 * probable retransmitted request; play it safe and
1704 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1705 status = nfserr_clid_inuse;
1708 } else if (!conf && unconf
1709 && same_verf(&unconf->cl_confirm, &confirm)) {
1711 * RFC 3530 14.2.34 CASE 3:
1712 * Normal case; new or rebooted client:
1714 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1715 status = nfserr_clid_inuse;
1718 clientstr_hashval(unconf->cl_recdir);
1719 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1722 nfsd4_remove_clid_dir(conf);
1723 expire_client(conf);
1725 move_to_confirmed(unconf);
1727 nfsd4_probe_callback(conf);
1730 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1731 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1734 * RFC 3530 14.2.34 CASE 4:
1735 * Client probably hasn't noticed that we rebooted yet.
1737 status = nfserr_stale_clientid;
1739 /* check that we have hit one of the cases...*/
1740 status = nfserr_clid_inuse;
1743 nfs4_unlock_state();
1747 /* OPEN Share state helper functions */
1748 static inline struct nfs4_file *
1749 alloc_init_file(struct inode *ino)
1751 struct nfs4_file *fp;
1752 unsigned int hashval = file_hashval(ino);
1754 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1756 atomic_set(&fp->fi_ref, 1);
1757 INIT_LIST_HEAD(&fp->fi_hash);
1758 INIT_LIST_HEAD(&fp->fi_stateids);
1759 INIT_LIST_HEAD(&fp->fi_delegations);
1760 spin_lock(&recall_lock);
1761 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1762 spin_unlock(&recall_lock);
1763 fp->fi_inode = igrab(ino);
1764 fp->fi_id = current_fileid++;
1765 fp->fi_had_conflict = false;
1772 nfsd4_free_slab(struct kmem_cache **slab)
1776 kmem_cache_destroy(*slab);
1781 nfsd4_free_slabs(void)
1783 nfsd4_free_slab(&stateowner_slab);
1784 nfsd4_free_slab(&file_slab);
1785 nfsd4_free_slab(&stateid_slab);
1786 nfsd4_free_slab(&deleg_slab);
1790 nfsd4_init_slabs(void)
1792 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1793 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1794 if (stateowner_slab == NULL)
1796 file_slab = kmem_cache_create("nfsd4_files",
1797 sizeof(struct nfs4_file), 0, 0, NULL);
1798 if (file_slab == NULL)
1800 stateid_slab = kmem_cache_create("nfsd4_stateids",
1801 sizeof(struct nfs4_stateid), 0, 0, NULL);
1802 if (stateid_slab == NULL)
1804 deleg_slab = kmem_cache_create("nfsd4_delegations",
1805 sizeof(struct nfs4_delegation), 0, 0, NULL);
1806 if (deleg_slab == NULL)
1811 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1816 nfs4_free_stateowner(struct kref *kref)
1818 struct nfs4_stateowner *sop =
1819 container_of(kref, struct nfs4_stateowner, so_ref);
1820 kfree(sop->so_owner.data);
1821 kmem_cache_free(stateowner_slab, sop);
1824 static inline struct nfs4_stateowner *
1825 alloc_stateowner(struct xdr_netobj *owner)
1827 struct nfs4_stateowner *sop;
1829 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1830 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1831 memcpy(sop->so_owner.data, owner->data, owner->len);
1832 sop->so_owner.len = owner->len;
1833 kref_init(&sop->so_ref);
1836 kmem_cache_free(stateowner_slab, sop);
1841 static struct nfs4_stateowner *
1842 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1843 struct nfs4_stateowner *sop;
1844 struct nfs4_replay *rp;
1845 unsigned int idhashval;
1847 if (!(sop = alloc_stateowner(&open->op_owner)))
1849 idhashval = ownerid_hashval(current_ownerid);
1850 INIT_LIST_HEAD(&sop->so_idhash);
1851 INIT_LIST_HEAD(&sop->so_strhash);
1852 INIT_LIST_HEAD(&sop->so_perclient);
1853 INIT_LIST_HEAD(&sop->so_stateids);
1854 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1855 INIT_LIST_HEAD(&sop->so_close_lru);
1857 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1858 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1859 list_add(&sop->so_perclient, &clp->cl_openowners);
1860 sop->so_is_open_owner = 1;
1861 sop->so_id = current_ownerid++;
1862 sop->so_client = clp;
1863 sop->so_seqid = open->op_seqid;
1864 sop->so_confirmed = 0;
1865 rp = &sop->so_replay;
1866 rp->rp_status = nfserr_serverfault;
1868 rp->rp_buf = rp->rp_ibuf;
1873 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1874 struct nfs4_stateowner *sop = open->op_stateowner;
1875 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1877 INIT_LIST_HEAD(&stp->st_hash);
1878 INIT_LIST_HEAD(&stp->st_perstateowner);
1879 INIT_LIST_HEAD(&stp->st_lockowners);
1880 INIT_LIST_HEAD(&stp->st_perfile);
1881 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1882 list_add(&stp->st_perstateowner, &sop->so_stateids);
1883 list_add(&stp->st_perfile, &fp->fi_stateids);
1884 stp->st_stateowner = sop;
1887 stp->st_stateid.si_boot = get_seconds();
1888 stp->st_stateid.si_stateownerid = sop->so_id;
1889 stp->st_stateid.si_fileid = fp->fi_id;
1890 stp->st_stateid.si_generation = 0;
1891 stp->st_access_bmap = 0;
1892 stp->st_deny_bmap = 0;
1893 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1894 &stp->st_access_bmap);
1895 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1896 stp->st_openstp = NULL;
1900 move_to_close_lru(struct nfs4_stateowner *sop)
1902 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1904 list_move_tail(&sop->so_close_lru, &close_lru);
1905 sop->so_time = get_seconds();
1909 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1912 return (sop->so_owner.len == owner->len) &&
1913 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1914 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1917 static struct nfs4_stateowner *
1918 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1920 struct nfs4_stateowner *so = NULL;
1922 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1923 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1929 /* search file_hashtbl[] for file */
1930 static struct nfs4_file *
1931 find_file(struct inode *ino)
1933 unsigned int hashval = file_hashval(ino);
1934 struct nfs4_file *fp;
1936 spin_lock(&recall_lock);
1937 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1938 if (fp->fi_inode == ino) {
1940 spin_unlock(&recall_lock);
1944 spin_unlock(&recall_lock);
1948 static inline int access_valid(u32 x, u32 minorversion)
1950 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1952 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1954 x &= ~NFS4_SHARE_ACCESS_MASK;
1955 if (minorversion && x) {
1956 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1958 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1960 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1967 static inline int deny_valid(u32 x)
1969 /* Note: unlike access bits, deny bits may be zero. */
1970 return x <= NFS4_SHARE_DENY_BOTH;
1974 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1975 * st_{access,deny}_bmap field of the stateid, in order to track not
1976 * only what share bits are currently in force, but also what
1977 * combinations of share bits previous opens have used. This allows us
1978 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1979 * return an error if the client attempt to downgrade to a combination
1980 * of share bits not explicable by closing some of its previous opens.
1982 * XXX: This enforcement is actually incomplete, since we don't keep
1983 * track of access/deny bit combinations; so, e.g., we allow:
1985 * OPEN allow read, deny write
1986 * OPEN allow both, deny none
1987 * DOWNGRADE allow read, deny none
1989 * which we should reject.
1992 set_access(unsigned int *access, unsigned long bmap) {
1996 for (i = 1; i < 4; i++) {
1997 if (test_bit(i, &bmap))
2003 set_deny(unsigned int *deny, unsigned long bmap) {
2007 for (i = 0; i < 4; i++) {
2008 if (test_bit(i, &bmap))
2014 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2015 unsigned int access, deny;
2017 set_access(&access, stp->st_access_bmap);
2018 set_deny(&deny, stp->st_deny_bmap);
2019 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2025 * Called to check deny when READ with all zero stateid or
2026 * WRITE with all zero or all one stateid
2029 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2031 struct inode *ino = current_fh->fh_dentry->d_inode;
2032 struct nfs4_file *fp;
2033 struct nfs4_stateid *stp;
2036 dprintk("NFSD: nfs4_share_conflict\n");
2038 fp = find_file(ino);
2041 ret = nfserr_locked;
2042 /* Search for conflicting share reservations */
2043 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2044 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2045 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2055 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2057 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2058 drop_file_write_access(filp);
2059 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2064 * Recall a delegation
2067 do_recall(void *__dp)
2069 struct nfs4_delegation *dp = __dp;
2071 dp->dl_file->fi_had_conflict = true;
2072 nfsd4_cb_recall(dp);
2077 * Spawn a thread to perform a recall on the delegation represented
2078 * by the lease (file_lock)
2080 * Called from break_lease() with lock_kernel() held.
2081 * Note: we assume break_lease will only call this *once* for any given
2085 void nfsd_break_deleg_cb(struct file_lock *fl)
2087 struct nfs4_delegation *dp= (struct nfs4_delegation *)fl->fl_owner;
2088 struct task_struct *t;
2090 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2094 /* We're assuming the state code never drops its reference
2095 * without first removing the lease. Since we're in this lease
2096 * callback (and since the lease code is serialized by the kernel
2097 * lock) we know the server hasn't removed the lease yet, we know
2098 * it's safe to take a reference: */
2099 atomic_inc(&dp->dl_count);
2100 atomic_inc(&dp->dl_client->cl_count);
2102 spin_lock(&recall_lock);
2103 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2104 spin_unlock(&recall_lock);
2106 /* only place dl_time is set. protected by lock_kernel*/
2107 dp->dl_time = get_seconds();
2110 * We don't want the locks code to timeout the lease for us;
2111 * we'll remove it ourself if the delegation isn't returned
2114 fl->fl_break_time = 0;
2116 t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
2118 struct nfs4_client *clp = dp->dl_client;
2120 printk(KERN_INFO "NFSD: Callback thread failed for "
2121 "for client (clientid %08x/%08x)\n",
2122 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2123 put_nfs4_client(dp->dl_client);
2124 nfs4_put_delegation(dp);
2129 * The file_lock is being reapd.
2131 * Called by locks_free_lock() with lock_kernel() held.
2134 void nfsd_release_deleg_cb(struct file_lock *fl)
2136 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2138 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2140 if (!(fl->fl_flags & FL_LEASE) || !dp)
2142 dp->dl_flock = NULL;
2146 * Set the delegation file_lock back pointer.
2148 * Called from setlease() with lock_kernel() held.
2151 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2153 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2155 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2162 * Called from setlease() with lock_kernel() held
2165 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2167 struct nfs4_delegation *onlistd =
2168 (struct nfs4_delegation *)onlist->fl_owner;
2169 struct nfs4_delegation *tryd =
2170 (struct nfs4_delegation *)try->fl_owner;
2172 if (onlist->fl_lmops != try->fl_lmops)
2175 return onlistd->dl_client == tryd->dl_client;
2180 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2183 return lease_modify(onlist, arg);
2188 static struct lock_manager_operations nfsd_lease_mng_ops = {
2189 .fl_break = nfsd_break_deleg_cb,
2190 .fl_release_private = nfsd_release_deleg_cb,
2191 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2192 .fl_mylease = nfsd_same_client_deleg_cb,
2193 .fl_change = nfsd_change_deleg_cb,
2198 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2199 struct nfsd4_open *open)
2201 clientid_t *clientid = &open->op_clientid;
2202 struct nfs4_client *clp = NULL;
2203 unsigned int strhashval;
2204 struct nfs4_stateowner *sop = NULL;
2206 if (!check_name(open->op_owner))
2207 return nfserr_inval;
2209 if (STALE_CLIENTID(&open->op_clientid))
2210 return nfserr_stale_clientid;
2212 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2213 sop = find_openstateowner_str(strhashval, open);
2214 open->op_stateowner = sop;
2216 /* Make sure the client's lease hasn't expired. */
2217 clp = find_confirmed_client(clientid);
2219 return nfserr_expired;
2222 /* When sessions are used, skip open sequenceid processing */
2223 if (nfsd4_has_session(cstate))
2225 if (!sop->so_confirmed) {
2226 /* Replace unconfirmed owners without checking for replay. */
2227 clp = sop->so_client;
2228 release_openowner(sop);
2229 open->op_stateowner = NULL;
2232 if (open->op_seqid == sop->so_seqid - 1) {
2233 if (sop->so_replay.rp_buflen)
2234 return nfserr_replay_me;
2235 /* The original OPEN failed so spectacularly
2236 * that we don't even have replay data saved!
2237 * Therefore, we have no choice but to continue
2238 * processing this OPEN; presumably, we'll
2239 * fail again for the same reason.
2241 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2244 if (open->op_seqid != sop->so_seqid)
2245 return nfserr_bad_seqid;
2247 if (open->op_stateowner == NULL) {
2248 sop = alloc_init_open_stateowner(strhashval, clp, open);
2250 return nfserr_resource;
2251 open->op_stateowner = sop;
2253 list_del_init(&sop->so_close_lru);
2254 renew_client(sop->so_client);
2258 static inline __be32
2259 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2261 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2262 return nfserr_openmode;
2267 static struct nfs4_delegation *
2268 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2270 struct nfs4_delegation *dp;
2272 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2273 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2280 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2281 struct nfs4_delegation **dp)
2284 __be32 status = nfserr_bad_stateid;
2286 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2289 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2290 RD_STATE : WR_STATE;
2291 status = nfs4_check_delegmode(*dp, flags);
2295 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2299 open->op_stateowner->so_confirmed = 1;
2304 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2306 struct nfs4_stateid *local;
2307 __be32 status = nfserr_share_denied;
2308 struct nfs4_stateowner *sop = open->op_stateowner;
2310 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2311 /* ignore lock owners */
2312 if (local->st_stateowner->so_is_open_owner == 0)
2314 /* remember if we have seen this open owner */
2315 if (local->st_stateowner == sop)
2317 /* check for conflicting share reservations */
2318 if (!test_share(local, open))
2326 static inline struct nfs4_stateid *
2327 nfs4_alloc_stateid(void)
2329 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2333 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2334 struct nfs4_delegation *dp,
2335 struct svc_fh *cur_fh, int flags)
2337 struct nfs4_stateid *stp;
2339 stp = nfs4_alloc_stateid();
2341 return nfserr_resource;
2344 get_file(dp->dl_vfs_file);
2345 stp->st_vfs_file = dp->dl_vfs_file;
2348 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2351 if (status == nfserr_dropit)
2352 status = nfserr_jukebox;
2353 kmem_cache_free(stateid_slab, stp);
2361 static inline __be32
2362 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2363 struct nfsd4_open *open)
2365 struct iattr iattr = {
2366 .ia_valid = ATTR_SIZE,
2369 if (!open->op_truncate)
2371 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2372 return nfserr_inval;
2373 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2377 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2379 struct file *filp = stp->st_vfs_file;
2380 struct inode *inode = filp->f_path.dentry->d_inode;
2381 unsigned int share_access, new_writer;
2384 set_access(&share_access, stp->st_access_bmap);
2385 new_writer = (~share_access) & open->op_share_access
2386 & NFS4_SHARE_ACCESS_WRITE;
2389 int err = get_write_access(inode);
2391 return nfserrno(err);
2392 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2394 return nfserrno(err);
2395 file_take_write(filp);
2397 status = nfsd4_truncate(rqstp, cur_fh, open);
2400 put_write_access(inode);
2403 /* remember the open */
2404 filp->f_mode |= open->op_share_access;
2405 __set_bit(open->op_share_access, &stp->st_access_bmap);
2406 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2413 nfs4_set_claim_prev(struct nfsd4_open *open)
2415 open->op_stateowner->so_confirmed = 1;
2416 open->op_stateowner->so_client->cl_firststate = 1;
2420 * Attempt to hand out a delegation.
2423 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2425 struct nfs4_delegation *dp;
2426 struct nfs4_stateowner *sop = stp->st_stateowner;
2427 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2428 struct file_lock fl, *flp = &fl;
2429 int status, flag = 0;
2431 flag = NFS4_OPEN_DELEGATE_NONE;
2432 open->op_recall = 0;
2433 switch (open->op_claim_type) {
2434 case NFS4_OPEN_CLAIM_PREVIOUS:
2435 if (!atomic_read(&cb->cb_set))
2436 open->op_recall = 1;
2437 flag = open->op_delegate_type;
2438 if (flag == NFS4_OPEN_DELEGATE_NONE)
2441 case NFS4_OPEN_CLAIM_NULL:
2442 /* Let's not give out any delegations till everyone's
2443 * had the chance to reclaim theirs.... */
2444 if (locks_in_grace())
2446 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2448 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2449 flag = NFS4_OPEN_DELEGATE_WRITE;
2451 flag = NFS4_OPEN_DELEGATE_READ;
2457 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2459 flag = NFS4_OPEN_DELEGATE_NONE;
2462 locks_init_lock(&fl);
2463 fl.fl_lmops = &nfsd_lease_mng_ops;
2464 fl.fl_flags = FL_LEASE;
2465 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2466 fl.fl_end = OFFSET_MAX;
2467 fl.fl_owner = (fl_owner_t)dp;
2468 fl.fl_file = stp->st_vfs_file;
2469 fl.fl_pid = current->tgid;
2471 /* vfs_setlease checks to see if delegation should be handed out.
2472 * the lock_manager callbacks fl_mylease and fl_change are used
2474 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2475 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2476 unhash_delegation(dp);
2477 flag = NFS4_OPEN_DELEGATE_NONE;
2481 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2483 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2484 dp->dl_stateid.si_boot,
2485 dp->dl_stateid.si_stateownerid,
2486 dp->dl_stateid.si_fileid,
2487 dp->dl_stateid.si_generation);
2489 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2490 && flag == NFS4_OPEN_DELEGATE_NONE
2491 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2492 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2493 open->op_delegate_type = flag;
2497 * called with nfs4_lock_state() held.
2500 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2502 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2503 struct nfs4_file *fp = NULL;
2504 struct inode *ino = current_fh->fh_dentry->d_inode;
2505 struct nfs4_stateid *stp = NULL;
2506 struct nfs4_delegation *dp = NULL;
2509 status = nfserr_inval;
2510 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2511 || !deny_valid(open->op_share_deny))
2514 * Lookup file; if found, lookup stateid and check open request,
2515 * and check for delegations in the process of being recalled.
2516 * If not found, create the nfs4_file struct
2518 fp = find_file(ino);
2520 if ((status = nfs4_check_open(fp, open, &stp)))
2522 status = nfs4_check_deleg(fp, open, &dp);
2526 status = nfserr_bad_stateid;
2527 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2529 status = nfserr_resource;
2530 fp = alloc_init_file(ino);
2536 * OPEN the file, or upgrade an existing OPEN.
2537 * If truncate fails, the OPEN fails.
2540 /* Stateid was found, this is an OPEN upgrade */
2541 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2544 update_stateid(&stp->st_stateid);
2546 /* Stateid was not found, this is a new OPEN */
2548 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2549 flags |= NFSD_MAY_READ;
2550 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2551 flags |= NFSD_MAY_WRITE;
2552 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2555 init_stateid(stp, fp, open);
2556 status = nfsd4_truncate(rqstp, current_fh, open);
2558 release_open_stateid(stp);
2561 if (nfsd4_has_session(&resp->cstate))
2562 update_stateid(&stp->st_stateid);
2564 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2566 if (nfsd4_has_session(&resp->cstate))
2567 open->op_stateowner->so_confirmed = 1;
2570 * Attempt to hand out a delegation. No error return, because the
2571 * OPEN succeeds even if we fail.
2573 nfs4_open_delegation(current_fh, open, stp);
2577 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2578 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2579 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2583 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2584 nfs4_set_claim_prev(open);
2586 * To finish the open response, we just need to set the rflags.
2588 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2589 if (!open->op_stateowner->so_confirmed &&
2590 !nfsd4_has_session(&resp->cstate))
2591 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2597 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2600 struct nfs4_client *clp;
2604 dprintk("process_renew(%08x/%08x): starting\n",
2605 clid->cl_boot, clid->cl_id);
2606 status = nfserr_stale_clientid;
2607 if (STALE_CLIENTID(clid))
2609 clp = find_confirmed_client(clid);
2610 status = nfserr_expired;
2612 /* We assume the client took too long to RENEW. */
2613 dprintk("nfsd4_renew: clientid not found!\n");
2617 status = nfserr_cb_path_down;
2618 if (!list_empty(&clp->cl_delegations)
2619 && !atomic_read(&clp->cl_cb_conn.cb_set))
2623 nfs4_unlock_state();
2627 struct lock_manager nfsd4_manager = {
2631 nfsd4_end_grace(void)
2633 dprintk("NFSD: end of grace period\n");
2634 nfsd4_recdir_purge_old();
2635 locks_end_grace(&nfsd4_manager);
2639 nfs4_laundromat(void)
2641 struct nfs4_client *clp;
2642 struct nfs4_stateowner *sop;
2643 struct nfs4_delegation *dp;
2644 struct list_head *pos, *next, reaplist;
2645 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2646 time_t t, clientid_val = NFSD_LEASE_TIME;
2647 time_t u, test_val = NFSD_LEASE_TIME;
2651 dprintk("NFSD: laundromat service - starting\n");
2652 if (locks_in_grace())
2654 list_for_each_safe(pos, next, &client_lru) {
2655 clp = list_entry(pos, struct nfs4_client, cl_lru);
2656 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2657 t = clp->cl_time - cutoff;
2658 if (clientid_val > t)
2662 dprintk("NFSD: purging unused client (clientid %08x)\n",
2663 clp->cl_clientid.cl_id);
2664 nfsd4_remove_clid_dir(clp);
2667 INIT_LIST_HEAD(&reaplist);
2668 spin_lock(&recall_lock);
2669 list_for_each_safe(pos, next, &del_recall_lru) {
2670 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2671 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2672 u = dp->dl_time - cutoff;
2677 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2679 list_move(&dp->dl_recall_lru, &reaplist);
2681 spin_unlock(&recall_lock);
2682 list_for_each_safe(pos, next, &reaplist) {
2683 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2684 list_del_init(&dp->dl_recall_lru);
2685 unhash_delegation(dp);
2687 test_val = NFSD_LEASE_TIME;
2688 list_for_each_safe(pos, next, &close_lru) {
2689 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2690 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2691 u = sop->so_time - cutoff;
2696 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2698 release_openowner(sop);
2700 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2701 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2702 nfs4_unlock_state();
2703 return clientid_val;
2706 static struct workqueue_struct *laundry_wq;
2707 static void laundromat_main(struct work_struct *);
2708 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2711 laundromat_main(struct work_struct *not_used)
2715 t = nfs4_laundromat();
2716 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2717 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2720 static struct nfs4_stateowner *
2721 search_close_lru(u32 st_id, int flags)
2723 struct nfs4_stateowner *local = NULL;
2725 if (flags & CLOSE_STATE) {
2726 list_for_each_entry(local, &close_lru, so_close_lru) {
2727 if (local->so_id == st_id)
2735 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2737 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2741 STALE_STATEID(stateid_t *stateid)
2743 if (time_after((unsigned long)boot_time,
2744 (unsigned long)stateid->si_boot)) {
2745 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2746 stateid->si_boot, stateid->si_stateownerid,
2747 stateid->si_fileid, stateid->si_generation);
2754 EXPIRED_STATEID(stateid_t *stateid)
2756 if (time_before((unsigned long)boot_time,
2757 ((unsigned long)stateid->si_boot)) &&
2758 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2759 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2760 stateid->si_boot, stateid->si_stateownerid,
2761 stateid->si_fileid, stateid->si_generation);
2768 stateid_error_map(stateid_t *stateid)
2770 if (STALE_STATEID(stateid))
2771 return nfserr_stale_stateid;
2772 if (EXPIRED_STATEID(stateid))
2773 return nfserr_expired;
2775 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2776 stateid->si_boot, stateid->si_stateownerid,
2777 stateid->si_fileid, stateid->si_generation);
2778 return nfserr_bad_stateid;
2782 access_permit_read(unsigned long access_bmap)
2784 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2785 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2786 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2790 access_permit_write(unsigned long access_bmap)
2792 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2793 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2797 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2799 __be32 status = nfserr_openmode;
2801 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2803 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2810 static inline __be32
2811 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2813 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2815 else if (locks_in_grace()) {
2816 /* Answer in remaining cases depends on existance of
2817 * conflicting state; so we must wait out the grace period. */
2818 return nfserr_grace;
2819 } else if (flags & WR_STATE)
2820 return nfs4_share_conflict(current_fh,
2821 NFS4_SHARE_DENY_WRITE);
2822 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2823 return nfs4_share_conflict(current_fh,
2824 NFS4_SHARE_DENY_READ);
2828 * Allow READ/WRITE during grace period on recovered state only for files
2829 * that are not able to provide mandatory locking.
2832 grace_disallows_io(struct inode *inode)
2834 return locks_in_grace() && mandatory_lock(inode);
2837 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2840 * When sessions are used the stateid generation number is ignored
2843 if ((flags & HAS_SESSION) && in->si_generation == 0)
2846 /* If the client sends us a stateid from the future, it's buggy: */
2847 if (in->si_generation > ref->si_generation)
2848 return nfserr_bad_stateid;
2850 * The following, however, can happen. For example, if the
2851 * client sends an open and some IO at the same time, the open
2852 * may bump si_generation while the IO is still in flight.
2853 * Thanks to hard links and renames, the client never knows what
2854 * file an open will affect. So it could avoid that situation
2855 * only by serializing all opens and IO from the same open
2856 * owner. To recover from the old_stateid error, the client
2857 * will just have to retry the IO:
2859 if (in->si_generation < ref->si_generation)
2860 return nfserr_old_stateid;
2865 static int is_delegation_stateid(stateid_t *stateid)
2867 return stateid->si_fileid == 0;
2871 * Checks for stateid operations
2874 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2875 stateid_t *stateid, int flags, struct file **filpp)
2877 struct nfs4_stateid *stp = NULL;
2878 struct nfs4_delegation *dp = NULL;
2879 struct svc_fh *current_fh = &cstate->current_fh;
2880 struct inode *ino = current_fh->fh_dentry->d_inode;
2886 if (grace_disallows_io(ino))
2887 return nfserr_grace;
2889 if (nfsd4_has_session(cstate))
2890 flags |= HAS_SESSION;
2892 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2893 return check_special_stateids(current_fh, stateid, flags);
2895 status = nfserr_stale_stateid;
2896 if (STALE_STATEID(stateid))
2899 status = nfserr_bad_stateid;
2900 if (is_delegation_stateid(stateid)) {
2901 dp = find_delegation_stateid(ino, stateid);
2903 status = stateid_error_map(stateid);
2906 status = check_stateid_generation(stateid, &dp->dl_stateid,
2910 status = nfs4_check_delegmode(dp, flags);
2913 renew_client(dp->dl_client);
2915 *filpp = dp->dl_vfs_file;
2916 } else { /* open or lock stateid */
2917 stp = find_stateid(stateid, flags);
2919 status = stateid_error_map(stateid);
2922 if (nfs4_check_fh(current_fh, stp))
2924 if (!stp->st_stateowner->so_confirmed)
2926 status = check_stateid_generation(stateid, &stp->st_stateid,
2930 status = nfs4_check_openmode(stp, flags);
2933 renew_client(stp->st_stateowner->so_client);
2935 *filpp = stp->st_vfs_file;
2945 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2946 RD_STATE : WR_STATE;
2950 * Checks for sequence id mutating operations.
2953 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2954 stateid_t *stateid, int flags,
2955 struct nfs4_stateowner **sopp,
2956 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2958 struct nfs4_stateid *stp;
2959 struct nfs4_stateowner *sop;
2960 struct svc_fh *current_fh = &cstate->current_fh;
2963 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2964 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2965 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2966 stateid->si_generation);
2971 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2972 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2973 return nfserr_bad_stateid;
2976 if (STALE_STATEID(stateid))
2977 return nfserr_stale_stateid;
2979 if (nfsd4_has_session(cstate))
2980 flags |= HAS_SESSION;
2983 * We return BAD_STATEID if filehandle doesn't match stateid,
2984 * the confirmed flag is incorrecly set, or the generation
2985 * number is incorrect.
2987 stp = find_stateid(stateid, flags);
2990 * Also, we should make sure this isn't just the result of
2993 sop = search_close_lru(stateid->si_stateownerid, flags);
2995 return stateid_error_map(stateid);
3001 *sopp = sop = stp->st_stateowner;
3004 clientid_t *lockclid = &lock->v.new.clientid;
3005 struct nfs4_client *clp = sop->so_client;
3009 lkflg = setlkflg(lock->lk_type);
3011 if (lock->lk_is_new) {
3012 if (!sop->so_is_open_owner)
3013 return nfserr_bad_stateid;
3014 if (!(flags & HAS_SESSION) &&
3015 !same_clid(&clp->cl_clientid, lockclid))
3016 return nfserr_bad_stateid;
3017 /* stp is the open stateid */
3018 status = nfs4_check_openmode(stp, lkflg);
3022 /* stp is the lock stateid */
3023 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3029 if (nfs4_check_fh(current_fh, stp)) {
3030 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3031 return nfserr_bad_stateid;
3035 * We now validate the seqid and stateid generation numbers.
3036 * For the moment, we ignore the possibility of
3037 * generation number wraparound.
3039 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3042 if (sop->so_confirmed && flags & CONFIRM) {
3043 dprintk("NFSD: preprocess_seqid_op: expected"
3044 " unconfirmed stateowner!\n");
3045 return nfserr_bad_stateid;
3047 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3048 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3049 " confirmed yet!\n");
3050 return nfserr_bad_stateid;
3052 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3055 renew_client(sop->so_client);
3059 if (seqid == sop->so_seqid - 1) {
3060 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3061 /* indicate replay to calling function */
3062 return nfserr_replay_me;
3064 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3065 sop->so_seqid, seqid);
3067 return nfserr_bad_seqid;
3071 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3072 struct nfsd4_open_confirm *oc)
3075 struct nfs4_stateowner *sop;
3076 struct nfs4_stateid *stp;
3078 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3079 (int)cstate->current_fh.fh_dentry->d_name.len,
3080 cstate->current_fh.fh_dentry->d_name.name);
3082 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3088 if ((status = nfs4_preprocess_seqid_op(cstate,
3089 oc->oc_seqid, &oc->oc_req_stateid,
3090 CONFIRM | OPEN_STATE,
3091 &oc->oc_stateowner, &stp, NULL)))
3094 sop = oc->oc_stateowner;
3095 sop->so_confirmed = 1;
3096 update_stateid(&stp->st_stateid);
3097 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3098 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3099 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3100 stp->st_stateid.si_boot,
3101 stp->st_stateid.si_stateownerid,
3102 stp->st_stateid.si_fileid,
3103 stp->st_stateid.si_generation);
3105 nfsd4_create_clid_dir(sop->so_client);
3107 if (oc->oc_stateowner) {
3108 nfs4_get_stateowner(oc->oc_stateowner);
3109 cstate->replay_owner = oc->oc_stateowner;
3111 nfs4_unlock_state();
3117 * unset all bits in union bitmap (bmap) that
3118 * do not exist in share (from successful OPEN_DOWNGRADE)
3121 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3124 for (i = 1; i < 4; i++) {
3125 if ((i & access) != i)
3126 __clear_bit(i, bmap);
3131 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3134 for (i = 0; i < 4; i++) {
3135 if ((i & deny) != i)
3136 __clear_bit(i, bmap);
3141 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3142 struct nfsd4_compound_state *cstate,
3143 struct nfsd4_open_downgrade *od)
3146 struct nfs4_stateid *stp;
3147 unsigned int share_access;
3149 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3150 (int)cstate->current_fh.fh_dentry->d_name.len,
3151 cstate->current_fh.fh_dentry->d_name.name);
3153 if (!access_valid(od->od_share_access, cstate->minorversion)
3154 || !deny_valid(od->od_share_deny))
3155 return nfserr_inval;
3158 if ((status = nfs4_preprocess_seqid_op(cstate,
3162 &od->od_stateowner, &stp, NULL)))
3165 status = nfserr_inval;
3166 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3167 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3168 stp->st_access_bmap, od->od_share_access);
3171 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3172 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3173 stp->st_deny_bmap, od->od_share_deny);
3176 set_access(&share_access, stp->st_access_bmap);
3177 nfs4_file_downgrade(stp->st_vfs_file,
3178 share_access & ~od->od_share_access);
3180 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3181 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3183 update_stateid(&stp->st_stateid);
3184 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3187 if (od->od_stateowner) {
3188 nfs4_get_stateowner(od->od_stateowner);
3189 cstate->replay_owner = od->od_stateowner;
3191 nfs4_unlock_state();
3196 * nfs4_unlock_state() called after encode
3199 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3200 struct nfsd4_close *close)
3203 struct nfs4_stateid *stp;
3205 dprintk("NFSD: nfsd4_close on file %.*s\n",
3206 (int)cstate->current_fh.fh_dentry->d_name.len,
3207 cstate->current_fh.fh_dentry->d_name.name);
3210 /* check close_lru for replay */
3211 if ((status = nfs4_preprocess_seqid_op(cstate,
3214 OPEN_STATE | CLOSE_STATE,
3215 &close->cl_stateowner, &stp, NULL)))
3218 update_stateid(&stp->st_stateid);
3219 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3221 /* release_stateid() calls nfsd_close() if needed */
3222 release_open_stateid(stp);
3224 /* place unused nfs4_stateowners on so_close_lru list to be
3225 * released by the laundromat service after the lease period
3226 * to enable us to handle CLOSE replay
3228 if (list_empty(&close->cl_stateowner->so_stateids))
3229 move_to_close_lru(close->cl_stateowner);
3231 if (close->cl_stateowner) {
3232 nfs4_get_stateowner(close->cl_stateowner);
3233 cstate->replay_owner = close->cl_stateowner;
3235 nfs4_unlock_state();
3240 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3241 struct nfsd4_delegreturn *dr)
3243 struct nfs4_delegation *dp;
3244 stateid_t *stateid = &dr->dr_stateid;
3245 struct inode *inode;
3249 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3251 inode = cstate->current_fh.fh_dentry->d_inode;
3253 if (nfsd4_has_session(cstate))
3254 flags |= HAS_SESSION;
3256 status = nfserr_bad_stateid;
3257 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3259 status = nfserr_stale_stateid;
3260 if (STALE_STATEID(stateid))
3262 status = nfserr_bad_stateid;
3263 if (!is_delegation_stateid(stateid))
3265 dp = find_delegation_stateid(inode, stateid);
3267 status = stateid_error_map(stateid);
3270 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3273 renew_client(dp->dl_client);
3275 unhash_delegation(dp);
3277 nfs4_unlock_state();
3284 * Lock owner state (byte-range locks)
3286 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3287 #define LOCK_HASH_BITS 8
3288 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3289 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3292 end_offset(u64 start, u64 len)
3297 return end >= start ? end: NFS4_MAX_UINT64;
3300 /* last octet in a range */
3302 last_byte_offset(u64 start, u64 len)
3308 return end > start ? end - 1: NFS4_MAX_UINT64;
3311 #define lockownerid_hashval(id) \
3312 ((id) & LOCK_HASH_MASK)
3314 static inline unsigned int
3315 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3316 struct xdr_netobj *ownername)
3318 return (file_hashval(inode) + cl_id
3319 + opaque_hashval(ownername->data, ownername->len))
3323 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3324 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3325 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3327 static struct nfs4_stateid *
3328 find_stateid(stateid_t *stid, int flags)
3330 struct nfs4_stateid *local;
3331 u32 st_id = stid->si_stateownerid;
3332 u32 f_id = stid->si_fileid;
3333 unsigned int hashval;
3335 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3336 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3337 hashval = stateid_hashval(st_id, f_id);
3338 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3339 if ((local->st_stateid.si_stateownerid == st_id) &&
3340 (local->st_stateid.si_fileid == f_id))
3345 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3346 hashval = stateid_hashval(st_id, f_id);
3347 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3348 if ((local->st_stateid.si_stateownerid == st_id) &&
3349 (local->st_stateid.si_fileid == f_id))
3356 static struct nfs4_delegation *
3357 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3359 struct nfs4_file *fp;
3360 struct nfs4_delegation *dl;
3362 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3363 stid->si_boot, stid->si_stateownerid,
3364 stid->si_fileid, stid->si_generation);
3366 fp = find_file(ino);
3369 dl = find_delegation_file(fp, stid);
3375 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3376 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3377 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3378 * locking, this prevents us from being completely protocol-compliant. The
3379 * real solution to this problem is to start using unsigned file offsets in
3380 * the VFS, but this is a very deep change!
3383 nfs4_transform_lock_offset(struct file_lock *lock)
3385 if (lock->fl_start < 0)
3386 lock->fl_start = OFFSET_MAX;
3387 if (lock->fl_end < 0)
3388 lock->fl_end = OFFSET_MAX;
3391 /* Hack!: For now, we're defining this just so we can use a pointer to it
3392 * as a unique cookie to identify our (NFSv4's) posix locks. */
3393 static struct lock_manager_operations nfsd_posix_mng_ops = {
3397 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3399 struct nfs4_stateowner *sop;
3402 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3403 sop = (struct nfs4_stateowner *) fl->fl_owner;
3404 hval = lockownerid_hashval(sop->so_id);
3405 kref_get(&sop->so_ref);
3407 deny->ld_clientid = sop->so_client->cl_clientid;
3409 deny->ld_sop = NULL;
3410 deny->ld_clientid.cl_boot = 0;
3411 deny->ld_clientid.cl_id = 0;
3413 deny->ld_start = fl->fl_start;
3414 deny->ld_length = NFS4_MAX_UINT64;
3415 if (fl->fl_end != NFS4_MAX_UINT64)
3416 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3417 deny->ld_type = NFS4_READ_LT;
3418 if (fl->fl_type != F_RDLCK)
3419 deny->ld_type = NFS4_WRITE_LT;
3422 static struct nfs4_stateowner *
3423 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3424 struct xdr_netobj *owner)
3426 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3427 struct nfs4_stateowner *op;
3429 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3430 if (same_owner_str(op, owner, clid))
3437 * Alloc a lock owner structure.
3438 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3441 * strhashval = lock_ownerstr_hashval
3444 static struct nfs4_stateowner *
3445 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3446 struct nfs4_stateowner *sop;
3447 struct nfs4_replay *rp;
3448 unsigned int idhashval;
3450 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3452 idhashval = lockownerid_hashval(current_ownerid);
3453 INIT_LIST_HEAD(&sop->so_idhash);
3454 INIT_LIST_HEAD(&sop->so_strhash);
3455 INIT_LIST_HEAD(&sop->so_perclient);
3456 INIT_LIST_HEAD(&sop->so_stateids);
3457 INIT_LIST_HEAD(&sop->so_perstateid);
3458 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3460 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3461 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3462 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3463 sop->so_is_open_owner = 0;
3464 sop->so_id = current_ownerid++;
3465 sop->so_client = clp;
3466 /* It is the openowner seqid that will be incremented in encode in the
3467 * case of new lockowners; so increment the lock seqid manually: */
3468 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3469 sop->so_confirmed = 1;
3470 rp = &sop->so_replay;
3471 rp->rp_status = nfserr_serverfault;
3473 rp->rp_buf = rp->rp_ibuf;
3477 static struct nfs4_stateid *
3478 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3480 struct nfs4_stateid *stp;
3481 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3483 stp = nfs4_alloc_stateid();
3486 INIT_LIST_HEAD(&stp->st_hash);
3487 INIT_LIST_HEAD(&stp->st_perfile);
3488 INIT_LIST_HEAD(&stp->st_perstateowner);
3489 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3490 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3491 list_add(&stp->st_perfile, &fp->fi_stateids);
3492 list_add(&stp->st_perstateowner, &sop->so_stateids);
3493 stp->st_stateowner = sop;
3496 stp->st_stateid.si_boot = get_seconds();
3497 stp->st_stateid.si_stateownerid = sop->so_id;
3498 stp->st_stateid.si_fileid = fp->fi_id;
3499 stp->st_stateid.si_generation = 0;
3500 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3501 stp->st_access_bmap = open_stp->st_access_bmap;
3502 stp->st_deny_bmap = open_stp->st_deny_bmap;
3503 stp->st_openstp = open_stp;
3510 check_lock_length(u64 offset, u64 length)
3512 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3513 LOFF_OVERFLOW(offset, length)));
3520 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3521 struct nfsd4_lock *lock)
3523 struct nfs4_stateowner *open_sop = NULL;
3524 struct nfs4_stateowner *lock_sop = NULL;
3525 struct nfs4_stateid *lock_stp;
3527 struct file_lock file_lock;
3528 struct file_lock conflock;
3530 unsigned int strhashval;
3534 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3535 (long long) lock->lk_offset,
3536 (long long) lock->lk_length);
3538 if (check_lock_length(lock->lk_offset, lock->lk_length))
3539 return nfserr_inval;
3541 if ((status = fh_verify(rqstp, &cstate->current_fh,
3542 S_IFREG, NFSD_MAY_LOCK))) {
3543 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3549 if (lock->lk_is_new) {
3551 * Client indicates that this is a new lockowner.
3552 * Use open owner and open stateid to create lock owner and
3555 struct nfs4_stateid *open_stp = NULL;
3556 struct nfs4_file *fp;
3558 status = nfserr_stale_clientid;
3559 if (!nfsd4_has_session(cstate) &&
3560 STALE_CLIENTID(&lock->lk_new_clientid))
3563 /* validate and update open stateid and open seqid */
3564 status = nfs4_preprocess_seqid_op(cstate,
3565 lock->lk_new_open_seqid,
3566 &lock->lk_new_open_stateid,
3568 &lock->lk_replay_owner, &open_stp,
3572 open_sop = lock->lk_replay_owner;
3573 /* create lockowner and lock stateid */
3574 fp = open_stp->st_file;
3575 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3576 open_sop->so_client->cl_clientid.cl_id,
3577 &lock->v.new.owner);
3578 /* XXX: Do we need to check for duplicate stateowners on
3579 * the same file, or should they just be allowed (and
3580 * create new stateids)? */
3581 status = nfserr_resource;
3582 lock_sop = alloc_init_lock_stateowner(strhashval,
3583 open_sop->so_client, open_stp, lock);
3584 if (lock_sop == NULL)
3586 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3587 if (lock_stp == NULL)
3590 /* lock (lock owner + lock stateid) already exists */
3591 status = nfs4_preprocess_seqid_op(cstate,
3592 lock->lk_old_lock_seqid,
3593 &lock->lk_old_lock_stateid,
3595 &lock->lk_replay_owner, &lock_stp, lock);
3598 lock_sop = lock->lk_replay_owner;
3600 /* lock->lk_replay_owner and lock_stp have been created or found */
3601 filp = lock_stp->st_vfs_file;
3603 status = nfserr_grace;
3604 if (locks_in_grace() && !lock->lk_reclaim)
3606 status = nfserr_no_grace;
3607 if (!locks_in_grace() && lock->lk_reclaim)
3610 locks_init_lock(&file_lock);
3611 switch (lock->lk_type) {
3614 file_lock.fl_type = F_RDLCK;
3618 case NFS4_WRITEW_LT:
3619 file_lock.fl_type = F_WRLCK;
3623 status = nfserr_inval;
3626 file_lock.fl_owner = (fl_owner_t)lock_sop;
3627 file_lock.fl_pid = current->tgid;
3628 file_lock.fl_file = filp;
3629 file_lock.fl_flags = FL_POSIX;
3630 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3632 file_lock.fl_start = lock->lk_offset;
3633 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3634 nfs4_transform_lock_offset(&file_lock);
3637 * Try to lock the file in the VFS.
3638 * Note: locks.c uses the BKL to protect the inode's lock list.
3641 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3643 case 0: /* success! */
3644 update_stateid(&lock_stp->st_stateid);
3645 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3649 case (EAGAIN): /* conflock holds conflicting lock */
3650 status = nfserr_denied;
3651 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3652 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3655 status = nfserr_deadlock;
3658 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3659 status = nfserr_resource;
3663 if (status && lock->lk_is_new && lock_sop)
3664 release_lockowner(lock_sop);
3665 if (lock->lk_replay_owner) {
3666 nfs4_get_stateowner(lock->lk_replay_owner);
3667 cstate->replay_owner = lock->lk_replay_owner;
3669 nfs4_unlock_state();
3674 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3675 * so we do a temporary open here just to get an open file to pass to
3676 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3679 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3684 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3687 err = vfs_test_lock(file, lock);
3696 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3697 struct nfsd4_lockt *lockt)
3699 struct inode *inode;
3700 struct file_lock file_lock;
3704 if (locks_in_grace())
3705 return nfserr_grace;
3707 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3708 return nfserr_inval;
3710 lockt->lt_stateowner = NULL;
3713 status = nfserr_stale_clientid;
3714 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3717 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3718 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3719 if (status == nfserr_symlink)
3720 status = nfserr_inval;
3724 inode = cstate->current_fh.fh_dentry->d_inode;
3725 locks_init_lock(&file_lock);
3726 switch (lockt->lt_type) {
3729 file_lock.fl_type = F_RDLCK;
3732 case NFS4_WRITEW_LT:
3733 file_lock.fl_type = F_WRLCK;
3736 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3737 status = nfserr_inval;
3741 lockt->lt_stateowner = find_lockstateowner_str(inode,
3742 &lockt->lt_clientid, &lockt->lt_owner);
3743 if (lockt->lt_stateowner)
3744 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3745 file_lock.fl_pid = current->tgid;
3746 file_lock.fl_flags = FL_POSIX;
3748 file_lock.fl_start = lockt->lt_offset;
3749 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3751 nfs4_transform_lock_offset(&file_lock);
3754 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3756 status = nfserrno(error);
3759 if (file_lock.fl_type != F_UNLCK) {
3760 status = nfserr_denied;
3761 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3764 nfs4_unlock_state();
3769 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3770 struct nfsd4_locku *locku)
3772 struct nfs4_stateid *stp;
3773 struct file *filp = NULL;
3774 struct file_lock file_lock;
3778 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3779 (long long) locku->lu_offset,
3780 (long long) locku->lu_length);
3782 if (check_lock_length(locku->lu_offset, locku->lu_length))
3783 return nfserr_inval;
3787 if ((status = nfs4_preprocess_seqid_op(cstate,
3791 &locku->lu_stateowner, &stp, NULL)))
3794 filp = stp->st_vfs_file;
3796 locks_init_lock(&file_lock);
3797 file_lock.fl_type = F_UNLCK;
3798 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3799 file_lock.fl_pid = current->tgid;
3800 file_lock.fl_file = filp;
3801 file_lock.fl_flags = FL_POSIX;
3802 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3803 file_lock.fl_start = locku->lu_offset;
3805 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3806 nfs4_transform_lock_offset(&file_lock);
3809 * Try to unlock the file in the VFS.
3811 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3813 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3817 * OK, unlock succeeded; the only thing left to do is update the stateid.
3819 update_stateid(&stp->st_stateid);
3820 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3823 if (locku->lu_stateowner) {
3824 nfs4_get_stateowner(locku->lu_stateowner);
3825 cstate->replay_owner = locku->lu_stateowner;
3827 nfs4_unlock_state();
3831 status = nfserrno(err);
3837 * 1: locks held by lockowner
3838 * 0: no locks held by lockowner
3841 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3843 struct file_lock **flpp;
3844 struct inode *inode = filp->f_path.dentry->d_inode;
3848 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3849 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3860 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3861 struct nfsd4_compound_state *cstate,
3862 struct nfsd4_release_lockowner *rlockowner)
3864 clientid_t *clid = &rlockowner->rl_clientid;
3865 struct nfs4_stateowner *sop;
3866 struct nfs4_stateid *stp;
3867 struct xdr_netobj *owner = &rlockowner->rl_owner;
3868 struct list_head matches;
3872 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3873 clid->cl_boot, clid->cl_id);
3875 /* XXX check for lease expiration */
3877 status = nfserr_stale_clientid;
3878 if (STALE_CLIENTID(clid))
3883 status = nfserr_locks_held;
3884 /* XXX: we're doing a linear search through all the lockowners.
3885 * Yipes! For now we'll just hope clients aren't really using
3886 * release_lockowner much, but eventually we have to fix these
3887 * data structures. */
3888 INIT_LIST_HEAD(&matches);
3889 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3890 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3891 if (!same_owner_str(sop, owner, clid))
3893 list_for_each_entry(stp, &sop->so_stateids,
3895 if (check_for_locks(stp->st_vfs_file, sop))
3897 /* Note: so_perclient unused for lockowners,
3898 * so it's OK to fool with here. */
3899 list_add(&sop->so_perclient, &matches);
3903 /* Clients probably won't expect us to return with some (but not all)
3904 * of the lockowner state released; so don't release any until all
3905 * have been checked. */
3907 while (!list_empty(&matches)) {
3908 sop = list_entry(matches.next, struct nfs4_stateowner,
3910 /* unhash_stateowner deletes so_perclient only
3911 * for openowners. */
3912 list_del(&sop->so_perclient);
3913 release_lockowner(sop);
3916 nfs4_unlock_state();
3920 static inline struct nfs4_client_reclaim *
3923 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3927 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3929 unsigned int strhashval = clientstr_hashval(name);
3930 struct nfs4_client *clp;
3932 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3937 * failure => all reset bets are off, nfserr_no_grace...
3940 nfs4_client_to_reclaim(const char *name)
3942 unsigned int strhashval;
3943 struct nfs4_client_reclaim *crp = NULL;
3945 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3946 crp = alloc_reclaim();
3949 strhashval = clientstr_hashval(name);
3950 INIT_LIST_HEAD(&crp->cr_strhash);
3951 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3952 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3953 reclaim_str_hashtbl_size++;
3958 nfs4_release_reclaim(void)
3960 struct nfs4_client_reclaim *crp = NULL;
3963 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3964 while (!list_empty(&reclaim_str_hashtbl[i])) {
3965 crp = list_entry(reclaim_str_hashtbl[i].next,
3966 struct nfs4_client_reclaim, cr_strhash);
3967 list_del(&crp->cr_strhash);
3969 reclaim_str_hashtbl_size--;
3972 BUG_ON(reclaim_str_hashtbl_size);
3976 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3977 static struct nfs4_client_reclaim *
3978 nfs4_find_reclaim_client(clientid_t *clid)
3980 unsigned int strhashval;
3981 struct nfs4_client *clp;
3982 struct nfs4_client_reclaim *crp = NULL;
3985 /* find clientid in conf_id_hashtbl */
3986 clp = find_confirmed_client(clid);
3990 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3991 clp->cl_name.len, clp->cl_name.data,
3994 /* find clp->cl_name in reclaim_str_hashtbl */
3995 strhashval = clientstr_hashval(clp->cl_recdir);
3996 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3997 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4005 * Called from OPEN. Look for clientid in reclaim list.
4008 nfs4_check_open_reclaim(clientid_t *clid)
4010 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4013 /* initialization to perform at module load time: */
4016 nfs4_state_init(void)
4020 status = nfsd4_init_slabs();
4023 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4024 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4025 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4026 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4027 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4029 for (i = 0; i < SESSION_HASH_SIZE; i++)
4030 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4031 for (i = 0; i < FILE_HASH_SIZE; i++) {
4032 INIT_LIST_HEAD(&file_hashtbl[i]);
4034 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4035 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4036 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4038 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4039 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4040 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4042 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4043 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4044 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4046 memset(&onestateid, ~0, sizeof(stateid_t));
4047 INIT_LIST_HEAD(&close_lru);
4048 INIT_LIST_HEAD(&client_lru);
4049 INIT_LIST_HEAD(&del_recall_lru);
4050 for (i = 0; i < CLIENT_HASH_SIZE; i++)
4051 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4052 reclaim_str_hashtbl_size = 0;
4057 nfsd4_load_reboot_recovery_data(void)
4062 nfsd4_init_recdir(user_recovery_dirname);
4063 status = nfsd4_recdir_load();
4064 nfs4_unlock_state();
4066 printk("NFSD: Failure reading reboot recovery data\n");
4070 get_nfs4_grace_period(void)
4072 return max(user_lease_time, lease_time) * HZ;
4076 * Since the lifetime of a delegation isn't limited to that of an open, a
4077 * client may quite reasonably hang on to a delegation as long as it has
4078 * the inode cached. This becomes an obvious problem the first time a
4079 * client's inode cache approaches the size of the server's total memory.
4081 * For now we avoid this problem by imposing a hard limit on the number
4082 * of delegations, which varies according to the server's memory size.
4085 set_max_delegations(void)
4088 * Allow at most 4 delegations per megabyte of RAM. Quick
4089 * estimates suggest that in the worst case (where every delegation
4090 * is for a different inode), a delegation could take about 1.5K,
4091 * giving a worst case usage of about 6% of memory.
4093 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4096 /* initialization to perform when the nfsd service is started: */
4099 __nfs4_state_start(void)
4101 unsigned long grace_time;
4103 boot_time = get_seconds();
4104 grace_time = get_nfs4_grace_period();
4105 lease_time = user_lease_time;
4106 locks_start_grace(&nfsd4_manager);
4107 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4109 laundry_wq = create_singlethread_workqueue("nfsd4");
4110 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4111 set_max_delegations();
4115 nfs4_state_start(void)
4119 nfsd4_load_reboot_recovery_data();
4120 __nfs4_state_start();
4126 nfs4_lease_time(void)
4132 __nfs4_state_shutdown(void)
4135 struct nfs4_client *clp = NULL;
4136 struct nfs4_delegation *dp = NULL;
4137 struct list_head *pos, *next, reaplist;
4139 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4140 while (!list_empty(&conf_id_hashtbl[i])) {
4141 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4144 while (!list_empty(&unconf_str_hashtbl[i])) {
4145 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4149 INIT_LIST_HEAD(&reaplist);
4150 spin_lock(&recall_lock);
4151 list_for_each_safe(pos, next, &del_recall_lru) {
4152 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4153 list_move(&dp->dl_recall_lru, &reaplist);
4155 spin_unlock(&recall_lock);
4156 list_for_each_safe(pos, next, &reaplist) {
4157 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4158 list_del_init(&dp->dl_recall_lru);
4159 unhash_delegation(dp);
4162 nfsd4_shutdown_recdir();
4167 nfs4_state_shutdown(void)
4169 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4170 destroy_workqueue(laundry_wq);
4171 locks_end_grace(&nfsd4_manager);
4173 nfs4_release_reclaim();
4174 __nfs4_state_shutdown();
4175 nfs4_unlock_state();
4179 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4180 * accessed when nfsd is starting.
4183 nfs4_set_recdir(char *recdir)
4185 strcpy(user_recovery_dirname, recdir);
4189 * Change the NFSv4 recovery directory to recdir.
4192 nfs4_reset_recoverydir(char *recdir)
4197 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4201 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4202 nfs4_set_recdir(recdir);
4210 nfs4_recoverydir(void)
4212 return user_recovery_dirname;
4216 * Called when leasetime is changed.
4218 * The only way the protocol gives us to handle on-the-fly lease changes is to
4219 * simulate a reboot. Instead of doing that, we just wait till the next time
4220 * we start to register any changes in lease time. If the administrator
4221 * really wants to change the lease time *now*, they can go ahead and bring
4222 * nfsd down and then back up again after changing the lease time.
4224 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4225 * when nfsd is starting
4228 nfs4_reset_lease(time_t leasetime)
4230 user_lease_time = leasetime;