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 */
72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
73 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
75 /* forward declarations */
76 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
77 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
78 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
79 static void nfs4_set_recdir(char *recdir);
83 /* Currently used for almost all code touching nfsv4 state: */
84 static DEFINE_MUTEX(client_mutex);
87 * Currently used for the del_recall_lru and file hash table. In an
88 * effort to decrease the scope of the client_mutex, this spinlock may
89 * eventually cover more:
91 static DEFINE_SPINLOCK(recall_lock);
93 static struct kmem_cache *stateowner_slab = NULL;
94 static struct kmem_cache *file_slab = NULL;
95 static struct kmem_cache *stateid_slab = NULL;
96 static struct kmem_cache *deleg_slab = NULL;
101 mutex_lock(&client_mutex);
105 nfs4_unlock_state(void)
107 mutex_unlock(&client_mutex);
111 opaque_hashval(const void *ptr, int nbytes)
113 unsigned char *cptr = (unsigned char *) ptr;
123 static struct list_head del_recall_lru;
126 put_nfs4_file(struct nfs4_file *fi)
128 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
129 list_del(&fi->fi_hash);
130 spin_unlock(&recall_lock);
132 kmem_cache_free(file_slab, fi);
137 get_nfs4_file(struct nfs4_file *fi)
139 atomic_inc(&fi->fi_ref);
142 static int num_delegations;
143 unsigned int max_delegations;
146 * Open owner state (share locks)
149 /* hash tables for nfs4_stateowner */
150 #define OWNER_HASH_BITS 8
151 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
152 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
154 #define ownerid_hashval(id) \
155 ((id) & OWNER_HASH_MASK)
156 #define ownerstr_hashval(clientid, ownername) \
157 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
159 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
160 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
162 /* hash table for nfs4_file */
163 #define FILE_HASH_BITS 8
164 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
165 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
166 /* hash table for (open)nfs4_stateid */
167 #define STATEID_HASH_BITS 10
168 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
169 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
171 #define file_hashval(x) \
172 hash_ptr(x, FILE_HASH_BITS)
173 #define stateid_hashval(owner_id, file_id) \
174 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
176 static struct list_head file_hashtbl[FILE_HASH_SIZE];
177 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
179 static struct nfs4_delegation *
180 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
182 struct nfs4_delegation *dp;
183 struct nfs4_file *fp = stp->st_file;
184 struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
186 dprintk("NFSD alloc_init_deleg\n");
187 if (fp->fi_had_conflict)
189 if (num_delegations > max_delegations)
191 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
195 INIT_LIST_HEAD(&dp->dl_perfile);
196 INIT_LIST_HEAD(&dp->dl_perclnt);
197 INIT_LIST_HEAD(&dp->dl_recall_lru);
202 get_file(stp->st_vfs_file);
203 dp->dl_vfs_file = stp->st_vfs_file;
205 dp->dl_recall.cbr_dp = NULL;
206 dp->dl_recall.cbr_ident = cb->cb_ident;
207 dp->dl_recall.cbr_trunc = 0;
208 dp->dl_stateid.si_boot = boot_time;
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]);
404 /* caller must hold sessionid_lock */
405 static struct nfsd4_session *
406 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
408 struct nfsd4_session *elem;
411 dump_sessionid(__func__, sessionid);
412 idx = hash_sessionid(sessionid);
413 dprintk("%s: idx is %d\n", __func__, idx);
414 /* Search in the appropriate list */
415 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
416 dump_sessionid("list traversal", &elem->se_sessionid);
417 if (!memcmp(elem->se_sessionid.data, sessionid->data,
418 NFS4_MAX_SESSIONID_LEN)) {
423 dprintk("%s: session not found\n", __func__);
427 /* caller must hold sessionid_lock */
429 unhash_session(struct nfsd4_session *ses)
431 list_del(&ses->se_hash);
432 list_del(&ses->se_perclnt);
436 release_session(struct nfsd4_session *ses)
438 spin_lock(&sessionid_lock);
440 spin_unlock(&sessionid_lock);
441 nfsd4_put_session(ses);
445 free_session(struct kref *kref)
447 struct nfsd4_session *ses;
449 ses = container_of(kref, struct nfsd4_session, se_ref);
450 kfree(ses->se_slots);
455 renew_client(struct nfs4_client *clp)
458 * Move client to the end to the LRU list.
460 dprintk("renewing client (clientid %08x/%08x)\n",
461 clp->cl_clientid.cl_boot,
462 clp->cl_clientid.cl_id);
463 list_move_tail(&clp->cl_lru, &client_lru);
464 clp->cl_time = get_seconds();
467 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
469 STALE_CLIENTID(clientid_t *clid)
471 if (clid->cl_boot == boot_time)
473 dprintk("NFSD stale clientid (%08x/%08x)\n",
474 clid->cl_boot, clid->cl_id);
479 * XXX Should we use a slab cache ?
480 * This type of memory management is somewhat inefficient, but we use it
481 * anyway since SETCLIENTID is not a common operation.
483 static struct nfs4_client *alloc_client(struct xdr_netobj name)
485 struct nfs4_client *clp;
487 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
490 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
491 if (clp->cl_name.data == NULL) {
495 memcpy(clp->cl_name.data, name.data, name.len);
496 clp->cl_name.len = name.len;
501 shutdown_callback_client(struct nfs4_client *clp)
503 struct rpc_clnt *clnt = clp->cl_callback.cb_client;
507 * Callback threads take a reference on the client, so there
508 * should be no outstanding callbacks at this point.
510 clp->cl_callback.cb_client = NULL;
511 rpc_shutdown_client(clnt);
516 free_client(struct nfs4_client *clp)
518 shutdown_callback_client(clp);
519 if (clp->cl_cred.cr_group_info)
520 put_group_info(clp->cl_cred.cr_group_info);
521 kfree(clp->cl_principal);
522 kfree(clp->cl_name.data);
527 put_nfs4_client(struct nfs4_client *clp)
529 if (atomic_dec_and_test(&clp->cl_count))
534 expire_client(struct nfs4_client *clp)
536 struct nfs4_stateowner *sop;
537 struct nfs4_delegation *dp;
538 struct list_head reaplist;
540 dprintk("NFSD: expire_client cl_count %d\n",
541 atomic_read(&clp->cl_count));
543 INIT_LIST_HEAD(&reaplist);
544 spin_lock(&recall_lock);
545 while (!list_empty(&clp->cl_delegations)) {
546 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
547 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
549 list_del_init(&dp->dl_perclnt);
550 list_move(&dp->dl_recall_lru, &reaplist);
552 spin_unlock(&recall_lock);
553 while (!list_empty(&reaplist)) {
554 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
555 list_del_init(&dp->dl_recall_lru);
556 unhash_delegation(dp);
558 list_del(&clp->cl_idhash);
559 list_del(&clp->cl_strhash);
560 list_del(&clp->cl_lru);
561 while (!list_empty(&clp->cl_openowners)) {
562 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
563 release_openowner(sop);
565 while (!list_empty(&clp->cl_sessions)) {
566 struct nfsd4_session *ses;
567 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
569 release_session(ses);
571 put_nfs4_client(clp);
574 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
576 struct nfs4_client *clp;
578 clp = alloc_client(name);
581 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
582 atomic_set(&clp->cl_count, 1);
583 atomic_set(&clp->cl_callback.cb_set, 0);
584 INIT_LIST_HEAD(&clp->cl_idhash);
585 INIT_LIST_HEAD(&clp->cl_strhash);
586 INIT_LIST_HEAD(&clp->cl_openowners);
587 INIT_LIST_HEAD(&clp->cl_delegations);
588 INIT_LIST_HEAD(&clp->cl_sessions);
589 INIT_LIST_HEAD(&clp->cl_lru);
593 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
595 memcpy(target->cl_verifier.data, source->data,
596 sizeof(target->cl_verifier.data));
599 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
601 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
602 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
605 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
607 target->cr_uid = source->cr_uid;
608 target->cr_gid = source->cr_gid;
609 target->cr_group_info = source->cr_group_info;
610 get_group_info(target->cr_group_info);
613 static int same_name(const char *n1, const char *n2)
615 return 0 == memcmp(n1, n2, HEXDIR_LEN);
619 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
621 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
625 same_clid(clientid_t *cl1, clientid_t *cl2)
627 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
630 /* XXX what about NGROUP */
632 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
634 return cr1->cr_uid == cr2->cr_uid;
637 static void gen_clid(struct nfs4_client *clp)
639 static u32 current_clientid = 1;
641 clp->cl_clientid.cl_boot = boot_time;
642 clp->cl_clientid.cl_id = current_clientid++;
645 static void gen_confirm(struct nfs4_client *clp)
650 p = (u32 *)clp->cl_confirm.data;
651 *p++ = get_seconds();
655 static int check_name(struct xdr_netobj name)
659 if (name.len > NFS4_OPAQUE_LIMIT) {
660 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
667 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
669 unsigned int idhashval;
671 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
672 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
673 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
674 list_add_tail(&clp->cl_lru, &client_lru);
675 clp->cl_time = get_seconds();
679 move_to_confirmed(struct nfs4_client *clp)
681 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
682 unsigned int strhashval;
684 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
685 list_del_init(&clp->cl_strhash);
686 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
687 strhashval = clientstr_hashval(clp->cl_recdir);
688 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
692 static struct nfs4_client *
693 find_confirmed_client(clientid_t *clid)
695 struct nfs4_client *clp;
696 unsigned int idhashval = clientid_hashval(clid->cl_id);
698 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
699 if (same_clid(&clp->cl_clientid, clid))
705 static struct nfs4_client *
706 find_unconfirmed_client(clientid_t *clid)
708 struct nfs4_client *clp;
709 unsigned int idhashval = clientid_hashval(clid->cl_id);
711 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
712 if (same_clid(&clp->cl_clientid, clid))
719 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
720 * parameter. Matching is based on the fact the at least one of the
721 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
723 * FIXME: we need to unify the clientid namespaces for nfsv4.x
724 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
725 * and SET_CLIENTID{,_CONFIRM}
728 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
730 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
731 return use_exchange_id == has_exchange_flags;
734 static struct nfs4_client *
735 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
736 bool use_exchange_id)
738 struct nfs4_client *clp;
740 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
741 if (same_name(clp->cl_recdir, dname) &&
742 match_clientid_establishment(clp, use_exchange_id))
748 static struct nfs4_client *
749 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
750 bool use_exchange_id)
752 struct nfs4_client *clp;
754 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
755 if (same_name(clp->cl_recdir, dname) &&
756 match_clientid_establishment(clp, use_exchange_id))
762 /* a helper function for parse_callback */
764 parse_octet(unsigned int *lenp, char **addrp)
766 unsigned int len = *lenp;
778 if ((c < '0') || (c > '9')) {
784 n = (n * 10) + (c - '0');
795 /* parse and set the setclientid ipv4 callback address */
797 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
802 u32 addrlen = addr_len;
803 char *addr = addr_val;
808 for(i = 4; i > 0 ; i--) {
809 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
812 cbaddr |= (temp << shift);
820 for(i = 2; i > 0 ; i--) {
821 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
824 cbport |= (temp << shift);
833 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
835 struct nfs4_callback *cb = &clp->cl_callback;
837 /* Currently, we only support tcp for the callback channel */
838 if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
841 if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
842 &cb->cb_addr, &cb->cb_port)))
844 cb->cb_prog = se->se_callback_prog;
845 cb->cb_ident = se->se_callback_ident;
848 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
849 "will not receive delegations\n",
850 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
856 * Set the exchange_id flags returned by the server.
859 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
861 /* pNFS is not supported */
862 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
864 /* Referrals are supported, Migration is not. */
865 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
867 /* set the wire flags to return to client. */
868 clid->flags = new->cl_exchange_flags;
872 nfsd4_exchange_id(struct svc_rqst *rqstp,
873 struct nfsd4_compound_state *cstate,
874 struct nfsd4_exchange_id *exid)
876 struct nfs4_client *unconf, *conf, *new;
878 unsigned int strhashval;
879 char dname[HEXDIR_LEN];
880 nfs4_verifier verf = exid->verifier;
881 u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
883 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
884 " ip_addr=%u flags %x, spa_how %d\n",
885 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
886 ip_addr, exid->flags, exid->spa_how);
888 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
891 /* Currently only support SP4_NONE */
892 switch (exid->spa_how) {
896 return nfserr_encr_alg_unsupp;
898 BUG(); /* checked by xdr code */
900 return nfserr_serverfault; /* no excuse :-/ */
903 status = nfs4_make_rec_clidname(dname, &exid->clname);
908 strhashval = clientstr_hashval(dname);
913 conf = find_confirmed_client_by_str(dname, strhashval, true);
915 if (!same_verf(&verf, &conf->cl_verifier)) {
917 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
918 status = nfserr_not_same;
921 /* Client reboot: destroy old state */
925 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
927 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
928 status = nfserr_perm;
934 if (ip_addr != conf->cl_addr &&
935 !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
936 /* Client collision. 18.35.4 case 3 */
937 status = nfserr_clid_inuse;
941 * Set bit when the owner id and verifier map to an already
942 * confirmed client id (18.35.3).
944 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
947 * Falling into 18.35.4 case 2, possible router replay.
948 * Leave confirmed record intact and return same result.
950 copy_verf(conf, &verf);
955 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
956 status = nfserr_noent;
961 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
964 * Possible retry or client restart. Per 18.35.4 case 4,
965 * a new unconfirmed record should be generated regardless
966 * of whether any properties have changed.
968 expire_client(unconf);
973 new = create_client(exid->clname, dname);
975 status = nfserr_resource;
979 copy_verf(new, &verf);
980 copy_cred(&new->cl_cred, &rqstp->rq_cred);
981 new->cl_addr = ip_addr;
984 add_to_unconfirmed(new, strhashval);
986 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
987 exid->clientid.cl_id = new->cl_clientid.cl_id;
989 new->cl_seqid = exid->seqid = 1;
990 nfsd4_set_ex_flags(new, exid);
992 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
993 new->cl_seqid, new->cl_exchange_flags);
999 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1004 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1006 dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1009 /* The slot is in use, and no response has been sent. */
1010 if (slot->sl_inuse) {
1011 if (seqid == slot->sl_seqid)
1012 return nfserr_jukebox;
1014 return nfserr_seq_misordered;
1017 if (likely(seqid == slot->sl_seqid + 1))
1020 if (seqid == slot->sl_seqid)
1021 return nfserr_replay_cache;
1023 if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1025 /* Misordered replay or misordered new request */
1026 return nfserr_seq_misordered;
1030 nfsd4_create_session(struct svc_rqst *rqstp,
1031 struct nfsd4_compound_state *cstate,
1032 struct nfsd4_create_session *cr_ses)
1034 return -1; /* stub */
1038 nfsd4_destroy_session(struct svc_rqst *r,
1039 struct nfsd4_compound_state *cstate,
1040 struct nfsd4_destroy_session *sessionid)
1042 return -1; /* stub */
1046 nfsd4_sequence(struct svc_rqst *rqstp,
1047 struct nfsd4_compound_state *cstate,
1048 struct nfsd4_sequence *seq)
1050 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1051 struct nfsd4_session *session;
1052 struct nfsd4_slot *slot;
1055 if (resp->opcnt != 1)
1056 return nfserr_sequence_pos;
1058 spin_lock(&sessionid_lock);
1059 status = nfserr_badsession;
1060 session = find_in_sessionid_hashtbl(&seq->sessionid);
1064 status = nfserr_badslot;
1065 if (seq->slotid >= session->se_fnumslots)
1068 slot = &session->se_slots[seq->slotid];
1069 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1071 status = check_slot_seqid(seq->seqid, slot);
1072 if (status == nfserr_replay_cache) {
1073 cstate->slot = slot;
1074 cstate->session = session;
1080 /* Success! bump slot seqid */
1081 slot->sl_inuse = true;
1082 slot->sl_seqid = seq->seqid;
1084 cstate->slot = slot;
1085 cstate->session = session;
1088 /* Renew the clientid on success and on replay.
1089 * Hold a session reference until done processing the compound:
1090 * nfsd4_put_session called only if the cstate slot is set.
1092 renew_client(session->se_client);
1093 nfsd4_get_session(session);
1095 spin_unlock(&sessionid_lock);
1096 dprintk("%s: return %d\n", __func__, ntohl(status));
1101 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1102 struct nfsd4_setclientid *setclid)
1104 struct sockaddr_in *sin = svc_addr_in(rqstp);
1105 struct xdr_netobj clname = {
1106 .len = setclid->se_namelen,
1107 .data = setclid->se_name,
1109 nfs4_verifier clverifier = setclid->se_verf;
1110 unsigned int strhashval;
1111 struct nfs4_client *conf, *unconf, *new;
1114 char dname[HEXDIR_LEN];
1116 if (!check_name(clname))
1117 return nfserr_inval;
1119 status = nfs4_make_rec_clidname(dname, &clname);
1124 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1125 * We get here on a DRC miss.
1128 strhashval = clientstr_hashval(dname);
1131 conf = find_confirmed_client_by_str(dname, strhashval, false);
1133 /* RFC 3530 14.2.33 CASE 0: */
1134 status = nfserr_clid_inuse;
1135 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1136 dprintk("NFSD: setclientid: string in use by client"
1137 " at %pI4\n", &conf->cl_addr);
1142 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1143 * has a description of SETCLIENTID request processing consisting
1144 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1146 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1147 status = nfserr_resource;
1150 * RFC 3530 14.2.33 CASE 4:
1151 * placed first, because it is the normal case
1154 expire_client(unconf);
1155 new = create_client(clname, dname);
1159 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1161 * RFC 3530 14.2.33 CASE 1:
1162 * probable callback update
1165 /* Note this is removing unconfirmed {*x***},
1166 * which is stronger than RFC recommended {vxc**}.
1167 * This has the advantage that there is at most
1168 * one {*x***} in either list at any time.
1170 expire_client(unconf);
1172 new = create_client(clname, dname);
1175 copy_clid(new, conf);
1176 } else if (!unconf) {
1178 * RFC 3530 14.2.33 CASE 2:
1179 * probable client reboot; state will be removed if
1182 new = create_client(clname, dname);
1188 * RFC 3530 14.2.33 CASE 3:
1189 * probable client reboot; state will be removed if
1192 expire_client(unconf);
1193 new = create_client(clname, dname);
1198 copy_verf(new, &clverifier);
1199 new->cl_addr = sin->sin_addr.s_addr;
1200 new->cl_flavor = rqstp->rq_flavor;
1201 princ = svc_gss_principal(rqstp);
1203 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1204 if (new->cl_principal == NULL) {
1209 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1211 gen_callback(new, setclid);
1212 add_to_unconfirmed(new, strhashval);
1213 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1214 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1215 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1218 nfs4_unlock_state();
1224 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1225 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1226 * bullets, labeled as CASE1 - CASE4 below.
1229 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1230 struct nfsd4_compound_state *cstate,
1231 struct nfsd4_setclientid_confirm *setclientid_confirm)
1233 struct sockaddr_in *sin = svc_addr_in(rqstp);
1234 struct nfs4_client *conf, *unconf;
1235 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1236 clientid_t * clid = &setclientid_confirm->sc_clientid;
1239 if (STALE_CLIENTID(clid))
1240 return nfserr_stale_clientid;
1242 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1243 * We get here on a DRC miss.
1248 conf = find_confirmed_client(clid);
1249 unconf = find_unconfirmed_client(clid);
1251 status = nfserr_clid_inuse;
1252 if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1254 if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1258 * section 14.2.34 of RFC 3530 has a description of
1259 * SETCLIENTID_CONFIRM request processing consisting
1260 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1262 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1264 * RFC 3530 14.2.34 CASE 1:
1267 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1268 status = nfserr_clid_inuse;
1270 /* XXX: We just turn off callbacks until we can handle
1271 * change request correctly. */
1272 atomic_set(&conf->cl_callback.cb_set, 0);
1274 nfsd4_remove_clid_dir(unconf);
1275 expire_client(unconf);
1279 } else if (conf && !unconf) {
1281 * RFC 3530 14.2.34 CASE 2:
1282 * probable retransmitted request; play it safe and
1285 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1286 status = nfserr_clid_inuse;
1289 } else if (!conf && unconf
1290 && same_verf(&unconf->cl_confirm, &confirm)) {
1292 * RFC 3530 14.2.34 CASE 3:
1293 * Normal case; new or rebooted client:
1295 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1296 status = nfserr_clid_inuse;
1299 clientstr_hashval(unconf->cl_recdir);
1300 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1303 nfsd4_remove_clid_dir(conf);
1304 expire_client(conf);
1306 move_to_confirmed(unconf);
1308 nfsd4_probe_callback(conf);
1311 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1312 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1315 * RFC 3530 14.2.34 CASE 4:
1316 * Client probably hasn't noticed that we rebooted yet.
1318 status = nfserr_stale_clientid;
1320 /* check that we have hit one of the cases...*/
1321 status = nfserr_clid_inuse;
1324 nfs4_unlock_state();
1328 /* OPEN Share state helper functions */
1329 static inline struct nfs4_file *
1330 alloc_init_file(struct inode *ino)
1332 struct nfs4_file *fp;
1333 unsigned int hashval = file_hashval(ino);
1335 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1337 atomic_set(&fp->fi_ref, 1);
1338 INIT_LIST_HEAD(&fp->fi_hash);
1339 INIT_LIST_HEAD(&fp->fi_stateids);
1340 INIT_LIST_HEAD(&fp->fi_delegations);
1341 spin_lock(&recall_lock);
1342 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1343 spin_unlock(&recall_lock);
1344 fp->fi_inode = igrab(ino);
1345 fp->fi_id = current_fileid++;
1346 fp->fi_had_conflict = false;
1353 nfsd4_free_slab(struct kmem_cache **slab)
1357 kmem_cache_destroy(*slab);
1362 nfsd4_free_slabs(void)
1364 nfsd4_free_slab(&stateowner_slab);
1365 nfsd4_free_slab(&file_slab);
1366 nfsd4_free_slab(&stateid_slab);
1367 nfsd4_free_slab(&deleg_slab);
1371 nfsd4_init_slabs(void)
1373 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1374 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1375 if (stateowner_slab == NULL)
1377 file_slab = kmem_cache_create("nfsd4_files",
1378 sizeof(struct nfs4_file), 0, 0, NULL);
1379 if (file_slab == NULL)
1381 stateid_slab = kmem_cache_create("nfsd4_stateids",
1382 sizeof(struct nfs4_stateid), 0, 0, NULL);
1383 if (stateid_slab == NULL)
1385 deleg_slab = kmem_cache_create("nfsd4_delegations",
1386 sizeof(struct nfs4_delegation), 0, 0, NULL);
1387 if (deleg_slab == NULL)
1392 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1397 nfs4_free_stateowner(struct kref *kref)
1399 struct nfs4_stateowner *sop =
1400 container_of(kref, struct nfs4_stateowner, so_ref);
1401 kfree(sop->so_owner.data);
1402 kmem_cache_free(stateowner_slab, sop);
1405 static inline struct nfs4_stateowner *
1406 alloc_stateowner(struct xdr_netobj *owner)
1408 struct nfs4_stateowner *sop;
1410 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1411 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1412 memcpy(sop->so_owner.data, owner->data, owner->len);
1413 sop->so_owner.len = owner->len;
1414 kref_init(&sop->so_ref);
1417 kmem_cache_free(stateowner_slab, sop);
1422 static struct nfs4_stateowner *
1423 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1424 struct nfs4_stateowner *sop;
1425 struct nfs4_replay *rp;
1426 unsigned int idhashval;
1428 if (!(sop = alloc_stateowner(&open->op_owner)))
1430 idhashval = ownerid_hashval(current_ownerid);
1431 INIT_LIST_HEAD(&sop->so_idhash);
1432 INIT_LIST_HEAD(&sop->so_strhash);
1433 INIT_LIST_HEAD(&sop->so_perclient);
1434 INIT_LIST_HEAD(&sop->so_stateids);
1435 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1436 INIT_LIST_HEAD(&sop->so_close_lru);
1438 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1439 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1440 list_add(&sop->so_perclient, &clp->cl_openowners);
1441 sop->so_is_open_owner = 1;
1442 sop->so_id = current_ownerid++;
1443 sop->so_client = clp;
1444 sop->so_seqid = open->op_seqid;
1445 sop->so_confirmed = 0;
1446 rp = &sop->so_replay;
1447 rp->rp_status = nfserr_serverfault;
1449 rp->rp_buf = rp->rp_ibuf;
1454 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1455 struct nfs4_stateowner *sop = open->op_stateowner;
1456 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1458 INIT_LIST_HEAD(&stp->st_hash);
1459 INIT_LIST_HEAD(&stp->st_perstateowner);
1460 INIT_LIST_HEAD(&stp->st_lockowners);
1461 INIT_LIST_HEAD(&stp->st_perfile);
1462 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1463 list_add(&stp->st_perstateowner, &sop->so_stateids);
1464 list_add(&stp->st_perfile, &fp->fi_stateids);
1465 stp->st_stateowner = sop;
1468 stp->st_stateid.si_boot = boot_time;
1469 stp->st_stateid.si_stateownerid = sop->so_id;
1470 stp->st_stateid.si_fileid = fp->fi_id;
1471 stp->st_stateid.si_generation = 0;
1472 stp->st_access_bmap = 0;
1473 stp->st_deny_bmap = 0;
1474 __set_bit(open->op_share_access, &stp->st_access_bmap);
1475 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1476 stp->st_openstp = NULL;
1480 move_to_close_lru(struct nfs4_stateowner *sop)
1482 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1484 list_move_tail(&sop->so_close_lru, &close_lru);
1485 sop->so_time = get_seconds();
1489 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1492 return (sop->so_owner.len == owner->len) &&
1493 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1494 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1497 static struct nfs4_stateowner *
1498 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1500 struct nfs4_stateowner *so = NULL;
1502 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1503 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1509 /* search file_hashtbl[] for file */
1510 static struct nfs4_file *
1511 find_file(struct inode *ino)
1513 unsigned int hashval = file_hashval(ino);
1514 struct nfs4_file *fp;
1516 spin_lock(&recall_lock);
1517 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1518 if (fp->fi_inode == ino) {
1520 spin_unlock(&recall_lock);
1524 spin_unlock(&recall_lock);
1528 static inline int access_valid(u32 x)
1530 if (x < NFS4_SHARE_ACCESS_READ)
1532 if (x > NFS4_SHARE_ACCESS_BOTH)
1537 static inline int deny_valid(u32 x)
1539 /* Note: unlike access bits, deny bits may be zero. */
1540 return x <= NFS4_SHARE_DENY_BOTH;
1544 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1545 * st_{access,deny}_bmap field of the stateid, in order to track not
1546 * only what share bits are currently in force, but also what
1547 * combinations of share bits previous opens have used. This allows us
1548 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1549 * return an error if the client attempt to downgrade to a combination
1550 * of share bits not explicable by closing some of its previous opens.
1552 * XXX: This enforcement is actually incomplete, since we don't keep
1553 * track of access/deny bit combinations; so, e.g., we allow:
1555 * OPEN allow read, deny write
1556 * OPEN allow both, deny none
1557 * DOWNGRADE allow read, deny none
1559 * which we should reject.
1562 set_access(unsigned int *access, unsigned long bmap) {
1566 for (i = 1; i < 4; i++) {
1567 if (test_bit(i, &bmap))
1573 set_deny(unsigned int *deny, unsigned long bmap) {
1577 for (i = 0; i < 4; i++) {
1578 if (test_bit(i, &bmap))
1584 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1585 unsigned int access, deny;
1587 set_access(&access, stp->st_access_bmap);
1588 set_deny(&deny, stp->st_deny_bmap);
1589 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1595 * Called to check deny when READ with all zero stateid or
1596 * WRITE with all zero or all one stateid
1599 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1601 struct inode *ino = current_fh->fh_dentry->d_inode;
1602 struct nfs4_file *fp;
1603 struct nfs4_stateid *stp;
1606 dprintk("NFSD: nfs4_share_conflict\n");
1608 fp = find_file(ino);
1611 ret = nfserr_locked;
1612 /* Search for conflicting share reservations */
1613 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1614 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1615 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1625 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1627 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1628 drop_file_write_access(filp);
1629 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1634 * Recall a delegation
1637 do_recall(void *__dp)
1639 struct nfs4_delegation *dp = __dp;
1641 dp->dl_file->fi_had_conflict = true;
1642 nfsd4_cb_recall(dp);
1647 * Spawn a thread to perform a recall on the delegation represented
1648 * by the lease (file_lock)
1650 * Called from break_lease() with lock_kernel() held.
1651 * Note: we assume break_lease will only call this *once* for any given
1655 void nfsd_break_deleg_cb(struct file_lock *fl)
1657 struct nfs4_delegation *dp= (struct nfs4_delegation *)fl->fl_owner;
1658 struct task_struct *t;
1660 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1664 /* We're assuming the state code never drops its reference
1665 * without first removing the lease. Since we're in this lease
1666 * callback (and since the lease code is serialized by the kernel
1667 * lock) we know the server hasn't removed the lease yet, we know
1668 * it's safe to take a reference: */
1669 atomic_inc(&dp->dl_count);
1670 atomic_inc(&dp->dl_client->cl_count);
1672 spin_lock(&recall_lock);
1673 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1674 spin_unlock(&recall_lock);
1676 /* only place dl_time is set. protected by lock_kernel*/
1677 dp->dl_time = get_seconds();
1680 * We don't want the locks code to timeout the lease for us;
1681 * we'll remove it ourself if the delegation isn't returned
1684 fl->fl_break_time = 0;
1686 t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1688 struct nfs4_client *clp = dp->dl_client;
1690 printk(KERN_INFO "NFSD: Callback thread failed for "
1691 "for client (clientid %08x/%08x)\n",
1692 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1693 put_nfs4_client(dp->dl_client);
1694 nfs4_put_delegation(dp);
1699 * The file_lock is being reapd.
1701 * Called by locks_free_lock() with lock_kernel() held.
1704 void nfsd_release_deleg_cb(struct file_lock *fl)
1706 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1708 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1710 if (!(fl->fl_flags & FL_LEASE) || !dp)
1712 dp->dl_flock = NULL;
1716 * Set the delegation file_lock back pointer.
1718 * Called from setlease() with lock_kernel() held.
1721 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1723 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1725 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1732 * Called from setlease() with lock_kernel() held
1735 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1737 struct nfs4_delegation *onlistd =
1738 (struct nfs4_delegation *)onlist->fl_owner;
1739 struct nfs4_delegation *tryd =
1740 (struct nfs4_delegation *)try->fl_owner;
1742 if (onlist->fl_lmops != try->fl_lmops)
1745 return onlistd->dl_client == tryd->dl_client;
1750 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1753 return lease_modify(onlist, arg);
1758 static struct lock_manager_operations nfsd_lease_mng_ops = {
1759 .fl_break = nfsd_break_deleg_cb,
1760 .fl_release_private = nfsd_release_deleg_cb,
1761 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1762 .fl_mylease = nfsd_same_client_deleg_cb,
1763 .fl_change = nfsd_change_deleg_cb,
1768 nfsd4_process_open1(struct nfsd4_open *open)
1770 clientid_t *clientid = &open->op_clientid;
1771 struct nfs4_client *clp = NULL;
1772 unsigned int strhashval;
1773 struct nfs4_stateowner *sop = NULL;
1775 if (!check_name(open->op_owner))
1776 return nfserr_inval;
1778 if (STALE_CLIENTID(&open->op_clientid))
1779 return nfserr_stale_clientid;
1781 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1782 sop = find_openstateowner_str(strhashval, open);
1783 open->op_stateowner = sop;
1785 /* Make sure the client's lease hasn't expired. */
1786 clp = find_confirmed_client(clientid);
1788 return nfserr_expired;
1791 if (!sop->so_confirmed) {
1792 /* Replace unconfirmed owners without checking for replay. */
1793 clp = sop->so_client;
1794 release_openowner(sop);
1795 open->op_stateowner = NULL;
1798 if (open->op_seqid == sop->so_seqid - 1) {
1799 if (sop->so_replay.rp_buflen)
1800 return nfserr_replay_me;
1801 /* The original OPEN failed so spectacularly
1802 * that we don't even have replay data saved!
1803 * Therefore, we have no choice but to continue
1804 * processing this OPEN; presumably, we'll
1805 * fail again for the same reason.
1807 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1810 if (open->op_seqid != sop->so_seqid)
1811 return nfserr_bad_seqid;
1813 if (open->op_stateowner == NULL) {
1814 sop = alloc_init_open_stateowner(strhashval, clp, open);
1816 return nfserr_resource;
1817 open->op_stateowner = sop;
1819 list_del_init(&sop->so_close_lru);
1820 renew_client(sop->so_client);
1824 static inline __be32
1825 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1827 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1828 return nfserr_openmode;
1833 static struct nfs4_delegation *
1834 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1836 struct nfs4_delegation *dp;
1838 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1839 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1846 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1847 struct nfs4_delegation **dp)
1850 __be32 status = nfserr_bad_stateid;
1852 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1855 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1856 RD_STATE : WR_STATE;
1857 status = nfs4_check_delegmode(*dp, flags);
1861 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1865 open->op_stateowner->so_confirmed = 1;
1870 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1872 struct nfs4_stateid *local;
1873 __be32 status = nfserr_share_denied;
1874 struct nfs4_stateowner *sop = open->op_stateowner;
1876 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1877 /* ignore lock owners */
1878 if (local->st_stateowner->so_is_open_owner == 0)
1880 /* remember if we have seen this open owner */
1881 if (local->st_stateowner == sop)
1883 /* check for conflicting share reservations */
1884 if (!test_share(local, open))
1892 static inline struct nfs4_stateid *
1893 nfs4_alloc_stateid(void)
1895 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1899 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1900 struct nfs4_delegation *dp,
1901 struct svc_fh *cur_fh, int flags)
1903 struct nfs4_stateid *stp;
1905 stp = nfs4_alloc_stateid();
1907 return nfserr_resource;
1910 get_file(dp->dl_vfs_file);
1911 stp->st_vfs_file = dp->dl_vfs_file;
1914 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1917 if (status == nfserr_dropit)
1918 status = nfserr_jukebox;
1919 kmem_cache_free(stateid_slab, stp);
1927 static inline __be32
1928 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1929 struct nfsd4_open *open)
1931 struct iattr iattr = {
1932 .ia_valid = ATTR_SIZE,
1935 if (!open->op_truncate)
1937 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1938 return nfserr_inval;
1939 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1943 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1945 struct file *filp = stp->st_vfs_file;
1946 struct inode *inode = filp->f_path.dentry->d_inode;
1947 unsigned int share_access, new_writer;
1950 set_access(&share_access, stp->st_access_bmap);
1951 new_writer = (~share_access) & open->op_share_access
1952 & NFS4_SHARE_ACCESS_WRITE;
1955 int err = get_write_access(inode);
1957 return nfserrno(err);
1958 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
1960 return nfserrno(err);
1961 file_take_write(filp);
1963 status = nfsd4_truncate(rqstp, cur_fh, open);
1966 put_write_access(inode);
1969 /* remember the open */
1970 filp->f_mode |= open->op_share_access;
1971 __set_bit(open->op_share_access, &stp->st_access_bmap);
1972 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1979 nfs4_set_claim_prev(struct nfsd4_open *open)
1981 open->op_stateowner->so_confirmed = 1;
1982 open->op_stateowner->so_client->cl_firststate = 1;
1986 * Attempt to hand out a delegation.
1989 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1991 struct nfs4_delegation *dp;
1992 struct nfs4_stateowner *sop = stp->st_stateowner;
1993 struct nfs4_callback *cb = &sop->so_client->cl_callback;
1994 struct file_lock fl, *flp = &fl;
1995 int status, flag = 0;
1997 flag = NFS4_OPEN_DELEGATE_NONE;
1998 open->op_recall = 0;
1999 switch (open->op_claim_type) {
2000 case NFS4_OPEN_CLAIM_PREVIOUS:
2001 if (!atomic_read(&cb->cb_set))
2002 open->op_recall = 1;
2003 flag = open->op_delegate_type;
2004 if (flag == NFS4_OPEN_DELEGATE_NONE)
2007 case NFS4_OPEN_CLAIM_NULL:
2008 /* Let's not give out any delegations till everyone's
2009 * had the chance to reclaim theirs.... */
2010 if (locks_in_grace())
2012 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2014 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2015 flag = NFS4_OPEN_DELEGATE_WRITE;
2017 flag = NFS4_OPEN_DELEGATE_READ;
2023 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2025 flag = NFS4_OPEN_DELEGATE_NONE;
2028 locks_init_lock(&fl);
2029 fl.fl_lmops = &nfsd_lease_mng_ops;
2030 fl.fl_flags = FL_LEASE;
2031 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2032 fl.fl_end = OFFSET_MAX;
2033 fl.fl_owner = (fl_owner_t)dp;
2034 fl.fl_file = stp->st_vfs_file;
2035 fl.fl_pid = current->tgid;
2037 /* vfs_setlease checks to see if delegation should be handed out.
2038 * the lock_manager callbacks fl_mylease and fl_change are used
2040 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2041 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2042 unhash_delegation(dp);
2043 flag = NFS4_OPEN_DELEGATE_NONE;
2047 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2049 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2050 dp->dl_stateid.si_boot,
2051 dp->dl_stateid.si_stateownerid,
2052 dp->dl_stateid.si_fileid,
2053 dp->dl_stateid.si_generation);
2055 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2056 && flag == NFS4_OPEN_DELEGATE_NONE
2057 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2058 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2059 open->op_delegate_type = flag;
2063 * called with nfs4_lock_state() held.
2066 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2068 struct nfs4_file *fp = NULL;
2069 struct inode *ino = current_fh->fh_dentry->d_inode;
2070 struct nfs4_stateid *stp = NULL;
2071 struct nfs4_delegation *dp = NULL;
2074 status = nfserr_inval;
2075 if (!access_valid(open->op_share_access)
2076 || !deny_valid(open->op_share_deny))
2079 * Lookup file; if found, lookup stateid and check open request,
2080 * and check for delegations in the process of being recalled.
2081 * If not found, create the nfs4_file struct
2083 fp = find_file(ino);
2085 if ((status = nfs4_check_open(fp, open, &stp)))
2087 status = nfs4_check_deleg(fp, open, &dp);
2091 status = nfserr_bad_stateid;
2092 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2094 status = nfserr_resource;
2095 fp = alloc_init_file(ino);
2101 * OPEN the file, or upgrade an existing OPEN.
2102 * If truncate fails, the OPEN fails.
2105 /* Stateid was found, this is an OPEN upgrade */
2106 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2109 update_stateid(&stp->st_stateid);
2111 /* Stateid was not found, this is a new OPEN */
2113 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2114 flags |= NFSD_MAY_READ;
2115 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2116 flags |= NFSD_MAY_WRITE;
2117 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2120 init_stateid(stp, fp, open);
2121 status = nfsd4_truncate(rqstp, current_fh, open);
2123 release_open_stateid(stp);
2127 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2130 * Attempt to hand out a delegation. No error return, because the
2131 * OPEN succeeds even if we fail.
2133 nfs4_open_delegation(current_fh, open, stp);
2137 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2138 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2139 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2143 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2144 nfs4_set_claim_prev(open);
2146 * To finish the open response, we just need to set the rflags.
2148 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2149 if (!open->op_stateowner->so_confirmed)
2150 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2156 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2159 struct nfs4_client *clp;
2163 dprintk("process_renew(%08x/%08x): starting\n",
2164 clid->cl_boot, clid->cl_id);
2165 status = nfserr_stale_clientid;
2166 if (STALE_CLIENTID(clid))
2168 clp = find_confirmed_client(clid);
2169 status = nfserr_expired;
2171 /* We assume the client took too long to RENEW. */
2172 dprintk("nfsd4_renew: clientid not found!\n");
2176 status = nfserr_cb_path_down;
2177 if (!list_empty(&clp->cl_delegations)
2178 && !atomic_read(&clp->cl_callback.cb_set))
2182 nfs4_unlock_state();
2186 struct lock_manager nfsd4_manager = {
2190 nfsd4_end_grace(void)
2192 dprintk("NFSD: end of grace period\n");
2193 nfsd4_recdir_purge_old();
2194 locks_end_grace(&nfsd4_manager);
2198 nfs4_laundromat(void)
2200 struct nfs4_client *clp;
2201 struct nfs4_stateowner *sop;
2202 struct nfs4_delegation *dp;
2203 struct list_head *pos, *next, reaplist;
2204 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2205 time_t t, clientid_val = NFSD_LEASE_TIME;
2206 time_t u, test_val = NFSD_LEASE_TIME;
2210 dprintk("NFSD: laundromat service - starting\n");
2211 if (locks_in_grace())
2213 list_for_each_safe(pos, next, &client_lru) {
2214 clp = list_entry(pos, struct nfs4_client, cl_lru);
2215 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2216 t = clp->cl_time - cutoff;
2217 if (clientid_val > t)
2221 dprintk("NFSD: purging unused client (clientid %08x)\n",
2222 clp->cl_clientid.cl_id);
2223 nfsd4_remove_clid_dir(clp);
2226 INIT_LIST_HEAD(&reaplist);
2227 spin_lock(&recall_lock);
2228 list_for_each_safe(pos, next, &del_recall_lru) {
2229 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2230 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2231 u = dp->dl_time - cutoff;
2236 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2238 list_move(&dp->dl_recall_lru, &reaplist);
2240 spin_unlock(&recall_lock);
2241 list_for_each_safe(pos, next, &reaplist) {
2242 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2243 list_del_init(&dp->dl_recall_lru);
2244 unhash_delegation(dp);
2246 test_val = NFSD_LEASE_TIME;
2247 list_for_each_safe(pos, next, &close_lru) {
2248 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2249 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2250 u = sop->so_time - cutoff;
2255 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2257 release_openowner(sop);
2259 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2260 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2261 nfs4_unlock_state();
2262 return clientid_val;
2265 static struct workqueue_struct *laundry_wq;
2266 static void laundromat_main(struct work_struct *);
2267 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2270 laundromat_main(struct work_struct *not_used)
2274 t = nfs4_laundromat();
2275 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2276 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2279 static struct nfs4_stateowner *
2280 search_close_lru(u32 st_id, int flags)
2282 struct nfs4_stateowner *local = NULL;
2284 if (flags & CLOSE_STATE) {
2285 list_for_each_entry(local, &close_lru, so_close_lru) {
2286 if (local->so_id == st_id)
2294 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2296 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2300 STALE_STATEID(stateid_t *stateid)
2302 if (stateid->si_boot == boot_time)
2304 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2305 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2306 stateid->si_generation);
2311 access_permit_read(unsigned long access_bmap)
2313 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2314 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2315 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2319 access_permit_write(unsigned long access_bmap)
2321 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2322 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2326 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2328 __be32 status = nfserr_openmode;
2330 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2332 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2339 static inline __be32
2340 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2342 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2344 else if (locks_in_grace()) {
2345 /* Answer in remaining cases depends on existance of
2346 * conflicting state; so we must wait out the grace period. */
2347 return nfserr_grace;
2348 } else if (flags & WR_STATE)
2349 return nfs4_share_conflict(current_fh,
2350 NFS4_SHARE_DENY_WRITE);
2351 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2352 return nfs4_share_conflict(current_fh,
2353 NFS4_SHARE_DENY_READ);
2357 * Allow READ/WRITE during grace period on recovered state only for files
2358 * that are not able to provide mandatory locking.
2361 grace_disallows_io(struct inode *inode)
2363 return locks_in_grace() && mandatory_lock(inode);
2366 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
2368 /* If the client sends us a stateid from the future, it's buggy: */
2369 if (in->si_generation > ref->si_generation)
2370 return nfserr_bad_stateid;
2372 * The following, however, can happen. For example, if the
2373 * client sends an open and some IO at the same time, the open
2374 * may bump si_generation while the IO is still in flight.
2375 * Thanks to hard links and renames, the client never knows what
2376 * file an open will affect. So it could avoid that situation
2377 * only by serializing all opens and IO from the same open
2378 * owner. To recover from the old_stateid error, the client
2379 * will just have to retry the IO:
2381 if (in->si_generation < ref->si_generation)
2382 return nfserr_old_stateid;
2386 static int is_delegation_stateid(stateid_t *stateid)
2388 return stateid->si_fileid == 0;
2392 * Checks for stateid operations
2395 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2397 struct nfs4_stateid *stp = NULL;
2398 struct nfs4_delegation *dp = NULL;
2399 struct inode *ino = current_fh->fh_dentry->d_inode;
2405 if (grace_disallows_io(ino))
2406 return nfserr_grace;
2408 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2409 return check_special_stateids(current_fh, stateid, flags);
2411 status = nfserr_stale_stateid;
2412 if (STALE_STATEID(stateid))
2415 status = nfserr_bad_stateid;
2416 if (is_delegation_stateid(stateid)) {
2417 dp = find_delegation_stateid(ino, stateid);
2420 status = check_stateid_generation(stateid, &dp->dl_stateid);
2423 status = nfs4_check_delegmode(dp, flags);
2426 renew_client(dp->dl_client);
2428 *filpp = dp->dl_vfs_file;
2429 } else { /* open or lock stateid */
2430 stp = find_stateid(stateid, flags);
2433 if (nfs4_check_fh(current_fh, stp))
2435 if (!stp->st_stateowner->so_confirmed)
2437 status = check_stateid_generation(stateid, &stp->st_stateid);
2440 status = nfs4_check_openmode(stp, flags);
2443 renew_client(stp->st_stateowner->so_client);
2445 *filpp = stp->st_vfs_file;
2455 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2456 RD_STATE : WR_STATE;
2460 * Checks for sequence id mutating operations.
2463 nfs4_preprocess_seqid_op(struct svc_fh *current_fh, u32 seqid, stateid_t *stateid, int flags, struct nfs4_stateowner **sopp, struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2465 struct nfs4_stateid *stp;
2466 struct nfs4_stateowner *sop;
2469 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2470 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2471 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2472 stateid->si_generation);
2477 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2478 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2479 return nfserr_bad_stateid;
2482 if (STALE_STATEID(stateid))
2483 return nfserr_stale_stateid;
2485 * We return BAD_STATEID if filehandle doesn't match stateid,
2486 * the confirmed flag is incorrecly set, or the generation
2487 * number is incorrect.
2489 stp = find_stateid(stateid, flags);
2492 * Also, we should make sure this isn't just the result of
2495 sop = search_close_lru(stateid->si_stateownerid, flags);
2497 return nfserr_bad_stateid;
2503 *sopp = sop = stp->st_stateowner;
2506 clientid_t *lockclid = &lock->v.new.clientid;
2507 struct nfs4_client *clp = sop->so_client;
2511 lkflg = setlkflg(lock->lk_type);
2513 if (lock->lk_is_new) {
2514 if (!sop->so_is_open_owner)
2515 return nfserr_bad_stateid;
2516 if (!same_clid(&clp->cl_clientid, lockclid))
2517 return nfserr_bad_stateid;
2518 /* stp is the open stateid */
2519 status = nfs4_check_openmode(stp, lkflg);
2523 /* stp is the lock stateid */
2524 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2530 if (nfs4_check_fh(current_fh, stp)) {
2531 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2532 return nfserr_bad_stateid;
2536 * We now validate the seqid and stateid generation numbers.
2537 * For the moment, we ignore the possibility of
2538 * generation number wraparound.
2540 if (seqid != sop->so_seqid)
2543 if (sop->so_confirmed && flags & CONFIRM) {
2544 dprintk("NFSD: preprocess_seqid_op: expected"
2545 " unconfirmed stateowner!\n");
2546 return nfserr_bad_stateid;
2548 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2549 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2550 " confirmed yet!\n");
2551 return nfserr_bad_stateid;
2553 status = check_stateid_generation(stateid, &stp->st_stateid);
2556 renew_client(sop->so_client);
2560 if (seqid == sop->so_seqid - 1) {
2561 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2562 /* indicate replay to calling function */
2563 return nfserr_replay_me;
2565 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2566 sop->so_seqid, seqid);
2568 return nfserr_bad_seqid;
2572 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2573 struct nfsd4_open_confirm *oc)
2576 struct nfs4_stateowner *sop;
2577 struct nfs4_stateid *stp;
2579 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2580 (int)cstate->current_fh.fh_dentry->d_name.len,
2581 cstate->current_fh.fh_dentry->d_name.name);
2583 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2589 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2590 oc->oc_seqid, &oc->oc_req_stateid,
2591 CONFIRM | OPEN_STATE,
2592 &oc->oc_stateowner, &stp, NULL)))
2595 sop = oc->oc_stateowner;
2596 sop->so_confirmed = 1;
2597 update_stateid(&stp->st_stateid);
2598 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2599 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2600 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2601 stp->st_stateid.si_boot,
2602 stp->st_stateid.si_stateownerid,
2603 stp->st_stateid.si_fileid,
2604 stp->st_stateid.si_generation);
2606 nfsd4_create_clid_dir(sop->so_client);
2608 if (oc->oc_stateowner) {
2609 nfs4_get_stateowner(oc->oc_stateowner);
2610 cstate->replay_owner = oc->oc_stateowner;
2612 nfs4_unlock_state();
2618 * unset all bits in union bitmap (bmap) that
2619 * do not exist in share (from successful OPEN_DOWNGRADE)
2622 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2625 for (i = 1; i < 4; i++) {
2626 if ((i & access) != i)
2627 __clear_bit(i, bmap);
2632 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2635 for (i = 0; i < 4; i++) {
2636 if ((i & deny) != i)
2637 __clear_bit(i, bmap);
2642 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2643 struct nfsd4_compound_state *cstate,
2644 struct nfsd4_open_downgrade *od)
2647 struct nfs4_stateid *stp;
2648 unsigned int share_access;
2650 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2651 (int)cstate->current_fh.fh_dentry->d_name.len,
2652 cstate->current_fh.fh_dentry->d_name.name);
2654 if (!access_valid(od->od_share_access)
2655 || !deny_valid(od->od_share_deny))
2656 return nfserr_inval;
2659 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2663 &od->od_stateowner, &stp, NULL)))
2666 status = nfserr_inval;
2667 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2668 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2669 stp->st_access_bmap, od->od_share_access);
2672 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2673 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2674 stp->st_deny_bmap, od->od_share_deny);
2677 set_access(&share_access, stp->st_access_bmap);
2678 nfs4_file_downgrade(stp->st_vfs_file,
2679 share_access & ~od->od_share_access);
2681 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2682 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2684 update_stateid(&stp->st_stateid);
2685 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2688 if (od->od_stateowner) {
2689 nfs4_get_stateowner(od->od_stateowner);
2690 cstate->replay_owner = od->od_stateowner;
2692 nfs4_unlock_state();
2697 * nfs4_unlock_state() called after encode
2700 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2701 struct nfsd4_close *close)
2704 struct nfs4_stateid *stp;
2706 dprintk("NFSD: nfsd4_close on file %.*s\n",
2707 (int)cstate->current_fh.fh_dentry->d_name.len,
2708 cstate->current_fh.fh_dentry->d_name.name);
2711 /* check close_lru for replay */
2712 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2715 OPEN_STATE | CLOSE_STATE,
2716 &close->cl_stateowner, &stp, NULL)))
2719 update_stateid(&stp->st_stateid);
2720 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2722 /* release_stateid() calls nfsd_close() if needed */
2723 release_open_stateid(stp);
2725 /* place unused nfs4_stateowners on so_close_lru list to be
2726 * released by the laundromat service after the lease period
2727 * to enable us to handle CLOSE replay
2729 if (list_empty(&close->cl_stateowner->so_stateids))
2730 move_to_close_lru(close->cl_stateowner);
2732 if (close->cl_stateowner) {
2733 nfs4_get_stateowner(close->cl_stateowner);
2734 cstate->replay_owner = close->cl_stateowner;
2736 nfs4_unlock_state();
2741 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2742 struct nfsd4_delegreturn *dr)
2744 struct nfs4_delegation *dp;
2745 stateid_t *stateid = &dr->dr_stateid;
2746 struct inode *inode;
2749 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2751 inode = cstate->current_fh.fh_dentry->d_inode;
2754 status = nfserr_bad_stateid;
2755 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2757 status = nfserr_stale_stateid;
2758 if (STALE_STATEID(stateid))
2760 status = nfserr_bad_stateid;
2761 if (!is_delegation_stateid(stateid))
2763 dp = find_delegation_stateid(inode, stateid);
2766 status = check_stateid_generation(stateid, &dp->dl_stateid);
2769 renew_client(dp->dl_client);
2771 unhash_delegation(dp);
2773 nfs4_unlock_state();
2780 * Lock owner state (byte-range locks)
2782 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
2783 #define LOCK_HASH_BITS 8
2784 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
2785 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
2788 end_offset(u64 start, u64 len)
2793 return end >= start ? end: NFS4_MAX_UINT64;
2796 /* last octet in a range */
2798 last_byte_offset(u64 start, u64 len)
2804 return end > start ? end - 1: NFS4_MAX_UINT64;
2807 #define lockownerid_hashval(id) \
2808 ((id) & LOCK_HASH_MASK)
2810 static inline unsigned int
2811 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2812 struct xdr_netobj *ownername)
2814 return (file_hashval(inode) + cl_id
2815 + opaque_hashval(ownername->data, ownername->len))
2819 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2820 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2821 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2823 static struct nfs4_stateid *
2824 find_stateid(stateid_t *stid, int flags)
2826 struct nfs4_stateid *local;
2827 u32 st_id = stid->si_stateownerid;
2828 u32 f_id = stid->si_fileid;
2829 unsigned int hashval;
2831 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2832 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
2833 hashval = stateid_hashval(st_id, f_id);
2834 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2835 if ((local->st_stateid.si_stateownerid == st_id) &&
2836 (local->st_stateid.si_fileid == f_id))
2841 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
2842 hashval = stateid_hashval(st_id, f_id);
2843 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2844 if ((local->st_stateid.si_stateownerid == st_id) &&
2845 (local->st_stateid.si_fileid == f_id))
2852 static struct nfs4_delegation *
2853 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2855 struct nfs4_file *fp;
2856 struct nfs4_delegation *dl;
2858 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2859 stid->si_boot, stid->si_stateownerid,
2860 stid->si_fileid, stid->si_generation);
2862 fp = find_file(ino);
2865 dl = find_delegation_file(fp, stid);
2871 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2872 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2873 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
2874 * locking, this prevents us from being completely protocol-compliant. The
2875 * real solution to this problem is to start using unsigned file offsets in
2876 * the VFS, but this is a very deep change!
2879 nfs4_transform_lock_offset(struct file_lock *lock)
2881 if (lock->fl_start < 0)
2882 lock->fl_start = OFFSET_MAX;
2883 if (lock->fl_end < 0)
2884 lock->fl_end = OFFSET_MAX;
2887 /* Hack!: For now, we're defining this just so we can use a pointer to it
2888 * as a unique cookie to identify our (NFSv4's) posix locks. */
2889 static struct lock_manager_operations nfsd_posix_mng_ops = {
2893 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2895 struct nfs4_stateowner *sop;
2898 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2899 sop = (struct nfs4_stateowner *) fl->fl_owner;
2900 hval = lockownerid_hashval(sop->so_id);
2901 kref_get(&sop->so_ref);
2903 deny->ld_clientid = sop->so_client->cl_clientid;
2905 deny->ld_sop = NULL;
2906 deny->ld_clientid.cl_boot = 0;
2907 deny->ld_clientid.cl_id = 0;
2909 deny->ld_start = fl->fl_start;
2910 deny->ld_length = NFS4_MAX_UINT64;
2911 if (fl->fl_end != NFS4_MAX_UINT64)
2912 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2913 deny->ld_type = NFS4_READ_LT;
2914 if (fl->fl_type != F_RDLCK)
2915 deny->ld_type = NFS4_WRITE_LT;
2918 static struct nfs4_stateowner *
2919 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2920 struct xdr_netobj *owner)
2922 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2923 struct nfs4_stateowner *op;
2925 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2926 if (same_owner_str(op, owner, clid))
2933 * Alloc a lock owner structure.
2934 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2937 * strhashval = lock_ownerstr_hashval
2940 static struct nfs4_stateowner *
2941 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2942 struct nfs4_stateowner *sop;
2943 struct nfs4_replay *rp;
2944 unsigned int idhashval;
2946 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2948 idhashval = lockownerid_hashval(current_ownerid);
2949 INIT_LIST_HEAD(&sop->so_idhash);
2950 INIT_LIST_HEAD(&sop->so_strhash);
2951 INIT_LIST_HEAD(&sop->so_perclient);
2952 INIT_LIST_HEAD(&sop->so_stateids);
2953 INIT_LIST_HEAD(&sop->so_perstateid);
2954 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2956 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2957 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2958 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2959 sop->so_is_open_owner = 0;
2960 sop->so_id = current_ownerid++;
2961 sop->so_client = clp;
2962 /* It is the openowner seqid that will be incremented in encode in the
2963 * case of new lockowners; so increment the lock seqid manually: */
2964 sop->so_seqid = lock->lk_new_lock_seqid + 1;
2965 sop->so_confirmed = 1;
2966 rp = &sop->so_replay;
2967 rp->rp_status = nfserr_serverfault;
2969 rp->rp_buf = rp->rp_ibuf;
2973 static struct nfs4_stateid *
2974 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2976 struct nfs4_stateid *stp;
2977 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2979 stp = nfs4_alloc_stateid();
2982 INIT_LIST_HEAD(&stp->st_hash);
2983 INIT_LIST_HEAD(&stp->st_perfile);
2984 INIT_LIST_HEAD(&stp->st_perstateowner);
2985 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2986 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2987 list_add(&stp->st_perfile, &fp->fi_stateids);
2988 list_add(&stp->st_perstateowner, &sop->so_stateids);
2989 stp->st_stateowner = sop;
2992 stp->st_stateid.si_boot = boot_time;
2993 stp->st_stateid.si_stateownerid = sop->so_id;
2994 stp->st_stateid.si_fileid = fp->fi_id;
2995 stp->st_stateid.si_generation = 0;
2996 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2997 stp->st_access_bmap = open_stp->st_access_bmap;
2998 stp->st_deny_bmap = open_stp->st_deny_bmap;
2999 stp->st_openstp = open_stp;
3006 check_lock_length(u64 offset, u64 length)
3008 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3009 LOFF_OVERFLOW(offset, length)));
3016 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3017 struct nfsd4_lock *lock)
3019 struct nfs4_stateowner *open_sop = NULL;
3020 struct nfs4_stateowner *lock_sop = NULL;
3021 struct nfs4_stateid *lock_stp;
3023 struct file_lock file_lock;
3024 struct file_lock conflock;
3026 unsigned int strhashval;
3030 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3031 (long long) lock->lk_offset,
3032 (long long) lock->lk_length);
3034 if (check_lock_length(lock->lk_offset, lock->lk_length))
3035 return nfserr_inval;
3037 if ((status = fh_verify(rqstp, &cstate->current_fh,
3038 S_IFREG, NFSD_MAY_LOCK))) {
3039 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3045 if (lock->lk_is_new) {
3047 * Client indicates that this is a new lockowner.
3048 * Use open owner and open stateid to create lock owner and
3051 struct nfs4_stateid *open_stp = NULL;
3052 struct nfs4_file *fp;
3054 status = nfserr_stale_clientid;
3055 if (STALE_CLIENTID(&lock->lk_new_clientid))
3058 /* validate and update open stateid and open seqid */
3059 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
3060 lock->lk_new_open_seqid,
3061 &lock->lk_new_open_stateid,
3063 &lock->lk_replay_owner, &open_stp,
3067 open_sop = lock->lk_replay_owner;
3068 /* create lockowner and lock stateid */
3069 fp = open_stp->st_file;
3070 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3071 open_sop->so_client->cl_clientid.cl_id,
3072 &lock->v.new.owner);
3073 /* XXX: Do we need to check for duplicate stateowners on
3074 * the same file, or should they just be allowed (and
3075 * create new stateids)? */
3076 status = nfserr_resource;
3077 lock_sop = alloc_init_lock_stateowner(strhashval,
3078 open_sop->so_client, open_stp, lock);
3079 if (lock_sop == NULL)
3081 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3082 if (lock_stp == NULL)
3085 /* lock (lock owner + lock stateid) already exists */
3086 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
3087 lock->lk_old_lock_seqid,
3088 &lock->lk_old_lock_stateid,
3090 &lock->lk_replay_owner, &lock_stp, lock);
3093 lock_sop = lock->lk_replay_owner;
3095 /* lock->lk_replay_owner and lock_stp have been created or found */
3096 filp = lock_stp->st_vfs_file;
3098 status = nfserr_grace;
3099 if (locks_in_grace() && !lock->lk_reclaim)
3101 status = nfserr_no_grace;
3102 if (!locks_in_grace() && lock->lk_reclaim)
3105 locks_init_lock(&file_lock);
3106 switch (lock->lk_type) {
3109 file_lock.fl_type = F_RDLCK;
3113 case NFS4_WRITEW_LT:
3114 file_lock.fl_type = F_WRLCK;
3118 status = nfserr_inval;
3121 file_lock.fl_owner = (fl_owner_t)lock_sop;
3122 file_lock.fl_pid = current->tgid;
3123 file_lock.fl_file = filp;
3124 file_lock.fl_flags = FL_POSIX;
3125 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3127 file_lock.fl_start = lock->lk_offset;
3128 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3129 nfs4_transform_lock_offset(&file_lock);
3132 * Try to lock the file in the VFS.
3133 * Note: locks.c uses the BKL to protect the inode's lock list.
3136 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3138 case 0: /* success! */
3139 update_stateid(&lock_stp->st_stateid);
3140 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3144 case (EAGAIN): /* conflock holds conflicting lock */
3145 status = nfserr_denied;
3146 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3147 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3150 status = nfserr_deadlock;
3153 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3154 status = nfserr_resource;
3158 if (status && lock->lk_is_new && lock_sop)
3159 release_lockowner(lock_sop);
3160 if (lock->lk_replay_owner) {
3161 nfs4_get_stateowner(lock->lk_replay_owner);
3162 cstate->replay_owner = lock->lk_replay_owner;
3164 nfs4_unlock_state();
3169 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3170 * so we do a temporary open here just to get an open file to pass to
3171 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3174 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3179 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3182 err = vfs_test_lock(file, lock);
3191 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3192 struct nfsd4_lockt *lockt)
3194 struct inode *inode;
3195 struct file_lock file_lock;
3199 if (locks_in_grace())
3200 return nfserr_grace;
3202 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3203 return nfserr_inval;
3205 lockt->lt_stateowner = NULL;
3208 status = nfserr_stale_clientid;
3209 if (STALE_CLIENTID(&lockt->lt_clientid))
3212 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3213 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3214 if (status == nfserr_symlink)
3215 status = nfserr_inval;
3219 inode = cstate->current_fh.fh_dentry->d_inode;
3220 locks_init_lock(&file_lock);
3221 switch (lockt->lt_type) {
3224 file_lock.fl_type = F_RDLCK;
3227 case NFS4_WRITEW_LT:
3228 file_lock.fl_type = F_WRLCK;
3231 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3232 status = nfserr_inval;
3236 lockt->lt_stateowner = find_lockstateowner_str(inode,
3237 &lockt->lt_clientid, &lockt->lt_owner);
3238 if (lockt->lt_stateowner)
3239 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3240 file_lock.fl_pid = current->tgid;
3241 file_lock.fl_flags = FL_POSIX;
3243 file_lock.fl_start = lockt->lt_offset;
3244 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3246 nfs4_transform_lock_offset(&file_lock);
3249 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3251 status = nfserrno(error);
3254 if (file_lock.fl_type != F_UNLCK) {
3255 status = nfserr_denied;
3256 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3259 nfs4_unlock_state();
3264 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3265 struct nfsd4_locku *locku)
3267 struct nfs4_stateid *stp;
3268 struct file *filp = NULL;
3269 struct file_lock file_lock;
3273 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3274 (long long) locku->lu_offset,
3275 (long long) locku->lu_length);
3277 if (check_lock_length(locku->lu_offset, locku->lu_length))
3278 return nfserr_inval;
3282 if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
3286 &locku->lu_stateowner, &stp, NULL)))
3289 filp = stp->st_vfs_file;
3291 locks_init_lock(&file_lock);
3292 file_lock.fl_type = F_UNLCK;
3293 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3294 file_lock.fl_pid = current->tgid;
3295 file_lock.fl_file = filp;
3296 file_lock.fl_flags = FL_POSIX;
3297 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3298 file_lock.fl_start = locku->lu_offset;
3300 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3301 nfs4_transform_lock_offset(&file_lock);
3304 * Try to unlock the file in the VFS.
3306 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3308 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3312 * OK, unlock succeeded; the only thing left to do is update the stateid.
3314 update_stateid(&stp->st_stateid);
3315 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3318 if (locku->lu_stateowner) {
3319 nfs4_get_stateowner(locku->lu_stateowner);
3320 cstate->replay_owner = locku->lu_stateowner;
3322 nfs4_unlock_state();
3326 status = nfserrno(err);
3332 * 1: locks held by lockowner
3333 * 0: no locks held by lockowner
3336 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3338 struct file_lock **flpp;
3339 struct inode *inode = filp->f_path.dentry->d_inode;
3343 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3344 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3355 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3356 struct nfsd4_compound_state *cstate,
3357 struct nfsd4_release_lockowner *rlockowner)
3359 clientid_t *clid = &rlockowner->rl_clientid;
3360 struct nfs4_stateowner *sop;
3361 struct nfs4_stateid *stp;
3362 struct xdr_netobj *owner = &rlockowner->rl_owner;
3363 struct list_head matches;
3367 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3368 clid->cl_boot, clid->cl_id);
3370 /* XXX check for lease expiration */
3372 status = nfserr_stale_clientid;
3373 if (STALE_CLIENTID(clid))
3378 status = nfserr_locks_held;
3379 /* XXX: we're doing a linear search through all the lockowners.
3380 * Yipes! For now we'll just hope clients aren't really using
3381 * release_lockowner much, but eventually we have to fix these
3382 * data structures. */
3383 INIT_LIST_HEAD(&matches);
3384 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3385 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3386 if (!same_owner_str(sop, owner, clid))
3388 list_for_each_entry(stp, &sop->so_stateids,
3390 if (check_for_locks(stp->st_vfs_file, sop))
3392 /* Note: so_perclient unused for lockowners,
3393 * so it's OK to fool with here. */
3394 list_add(&sop->so_perclient, &matches);
3398 /* Clients probably won't expect us to return with some (but not all)
3399 * of the lockowner state released; so don't release any until all
3400 * have been checked. */
3402 while (!list_empty(&matches)) {
3403 sop = list_entry(matches.next, struct nfs4_stateowner,
3405 /* unhash_stateowner deletes so_perclient only
3406 * for openowners. */
3407 list_del(&sop->so_perclient);
3408 release_lockowner(sop);
3411 nfs4_unlock_state();
3415 static inline struct nfs4_client_reclaim *
3418 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3422 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3424 unsigned int strhashval = clientstr_hashval(name);
3425 struct nfs4_client *clp;
3427 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3432 * failure => all reset bets are off, nfserr_no_grace...
3435 nfs4_client_to_reclaim(const char *name)
3437 unsigned int strhashval;
3438 struct nfs4_client_reclaim *crp = NULL;
3440 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3441 crp = alloc_reclaim();
3444 strhashval = clientstr_hashval(name);
3445 INIT_LIST_HEAD(&crp->cr_strhash);
3446 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3447 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3448 reclaim_str_hashtbl_size++;
3453 nfs4_release_reclaim(void)
3455 struct nfs4_client_reclaim *crp = NULL;
3458 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3459 while (!list_empty(&reclaim_str_hashtbl[i])) {
3460 crp = list_entry(reclaim_str_hashtbl[i].next,
3461 struct nfs4_client_reclaim, cr_strhash);
3462 list_del(&crp->cr_strhash);
3464 reclaim_str_hashtbl_size--;
3467 BUG_ON(reclaim_str_hashtbl_size);
3471 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3472 static struct nfs4_client_reclaim *
3473 nfs4_find_reclaim_client(clientid_t *clid)
3475 unsigned int strhashval;
3476 struct nfs4_client *clp;
3477 struct nfs4_client_reclaim *crp = NULL;
3480 /* find clientid in conf_id_hashtbl */
3481 clp = find_confirmed_client(clid);
3485 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3486 clp->cl_name.len, clp->cl_name.data,
3489 /* find clp->cl_name in reclaim_str_hashtbl */
3490 strhashval = clientstr_hashval(clp->cl_recdir);
3491 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3492 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3500 * Called from OPEN. Look for clientid in reclaim list.
3503 nfs4_check_open_reclaim(clientid_t *clid)
3505 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3508 /* initialization to perform at module load time: */
3511 nfs4_state_init(void)
3515 status = nfsd4_init_slabs();
3518 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3519 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3520 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3521 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3522 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3524 for (i = 0; i < SESSION_HASH_SIZE; i++)
3525 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3526 for (i = 0; i < FILE_HASH_SIZE; i++) {
3527 INIT_LIST_HEAD(&file_hashtbl[i]);
3529 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3530 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3531 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3533 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3534 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3535 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3537 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3538 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3539 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3541 memset(&onestateid, ~0, sizeof(stateid_t));
3542 INIT_LIST_HEAD(&close_lru);
3543 INIT_LIST_HEAD(&client_lru);
3544 INIT_LIST_HEAD(&del_recall_lru);
3545 for (i = 0; i < CLIENT_HASH_SIZE; i++)
3546 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3547 reclaim_str_hashtbl_size = 0;
3552 nfsd4_load_reboot_recovery_data(void)
3557 nfsd4_init_recdir(user_recovery_dirname);
3558 status = nfsd4_recdir_load();
3559 nfs4_unlock_state();
3561 printk("NFSD: Failure reading reboot recovery data\n");
3565 get_nfs4_grace_period(void)
3567 return max(user_lease_time, lease_time) * HZ;
3571 * Since the lifetime of a delegation isn't limited to that of an open, a
3572 * client may quite reasonably hang on to a delegation as long as it has
3573 * the inode cached. This becomes an obvious problem the first time a
3574 * client's inode cache approaches the size of the server's total memory.
3576 * For now we avoid this problem by imposing a hard limit on the number
3577 * of delegations, which varies according to the server's memory size.
3580 set_max_delegations(void)
3583 * Allow at most 4 delegations per megabyte of RAM. Quick
3584 * estimates suggest that in the worst case (where every delegation
3585 * is for a different inode), a delegation could take about 1.5K,
3586 * giving a worst case usage of about 6% of memory.
3588 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3591 /* initialization to perform when the nfsd service is started: */
3594 __nfs4_state_start(void)
3596 unsigned long grace_time;
3598 boot_time = get_seconds();
3599 grace_time = get_nfs4_grace_period();
3600 lease_time = user_lease_time;
3601 locks_start_grace(&nfsd4_manager);
3602 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3604 laundry_wq = create_singlethread_workqueue("nfsd4");
3605 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3606 set_max_delegations();
3610 nfs4_state_start(void)
3614 nfsd4_load_reboot_recovery_data();
3615 __nfs4_state_start();
3621 nfs4_lease_time(void)
3627 __nfs4_state_shutdown(void)
3630 struct nfs4_client *clp = NULL;
3631 struct nfs4_delegation *dp = NULL;
3632 struct list_head *pos, *next, reaplist;
3634 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3635 while (!list_empty(&conf_id_hashtbl[i])) {
3636 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3639 while (!list_empty(&unconf_str_hashtbl[i])) {
3640 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3644 INIT_LIST_HEAD(&reaplist);
3645 spin_lock(&recall_lock);
3646 list_for_each_safe(pos, next, &del_recall_lru) {
3647 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3648 list_move(&dp->dl_recall_lru, &reaplist);
3650 spin_unlock(&recall_lock);
3651 list_for_each_safe(pos, next, &reaplist) {
3652 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3653 list_del_init(&dp->dl_recall_lru);
3654 unhash_delegation(dp);
3657 nfsd4_shutdown_recdir();
3662 nfs4_state_shutdown(void)
3664 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3665 destroy_workqueue(laundry_wq);
3666 locks_end_grace(&nfsd4_manager);
3668 nfs4_release_reclaim();
3669 __nfs4_state_shutdown();
3670 nfs4_unlock_state();
3674 * user_recovery_dirname is protected by the nfsd_mutex since it's only
3675 * accessed when nfsd is starting.
3678 nfs4_set_recdir(char *recdir)
3680 strcpy(user_recovery_dirname, recdir);
3684 * Change the NFSv4 recovery directory to recdir.
3687 nfs4_reset_recoverydir(char *recdir)
3692 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
3696 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
3697 nfs4_set_recdir(recdir);
3705 nfs4_recoverydir(void)
3707 return user_recovery_dirname;
3711 * Called when leasetime is changed.
3713 * The only way the protocol gives us to handle on-the-fly lease changes is to
3714 * simulate a reboot. Instead of doing that, we just wait till the next time
3715 * we start to register any changes in lease time. If the administrator
3716 * really wants to change the lease time *now*, they can go ahead and bring
3717 * nfsd down and then back up again after changing the lease time.
3719 * user_lease_time is protected by nfsd_mutex since it's only really accessed
3720 * when nfsd is starting
3723 nfs4_reset_lease(time_t leasetime)
3725 user_lease_time = leasetime;