2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
36 #include <linux/smp_lock.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
45 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 time_t nfsd4_lease = 90; /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static stateid_t zerostateid; /* bits all 0 */
55 static stateid_t onestateid; /* bits all 1 */
56 static u64 current_sessionid = 1;
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
73 * Currently used for the del_recall_lru and file hash table. In an
74 * effort to decrease the scope of the client_mutex, this spinlock may
75 * eventually cover more:
77 static DEFINE_SPINLOCK(recall_lock);
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
87 mutex_lock(&client_mutex);
91 nfs4_unlock_state(void)
93 mutex_unlock(&client_mutex);
97 opaque_hashval(const void *ptr, int nbytes)
99 unsigned char *cptr = (unsigned char *) ptr;
109 static struct list_head del_recall_lru;
112 put_nfs4_file(struct nfs4_file *fi)
114 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115 list_del(&fi->fi_hash);
116 spin_unlock(&recall_lock);
118 kmem_cache_free(file_slab, fi);
123 get_nfs4_file(struct nfs4_file *fi)
125 atomic_inc(&fi->fi_ref);
128 static int num_delegations;
129 unsigned int max_delegations;
132 * Open owner state (share locks)
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS 8
137 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
140 #define ownerid_hashval(id) \
141 ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS 8
150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS 10
154 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
157 #define file_hashval(x) \
158 hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id) \
160 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
167 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168 atomic_inc(&fp->fi_access[oflag]);
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
173 if (oflag == O_RDWR) {
174 __nfs4_file_get_access(fp, O_RDONLY);
175 __nfs4_file_get_access(fp, O_WRONLY);
177 __nfs4_file_get_access(fp, oflag);
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
182 if (fp->fi_fds[oflag]) {
183 fput(fp->fi_fds[oflag]);
184 fp->fi_fds[oflag] = NULL;
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
190 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191 nfs4_file_put_fd(fp, O_RDWR);
192 nfs4_file_put_fd(fp, oflag);
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
198 if (oflag == O_RDWR) {
199 __nfs4_file_put_access(fp, O_RDONLY);
200 __nfs4_file_put_access(fp, O_WRONLY);
202 __nfs4_file_put_access(fp, oflag);
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
208 struct nfs4_delegation *dp;
209 struct nfs4_file *fp = stp->st_file;
211 dprintk("NFSD alloc_init_deleg\n");
213 * Major work on the lease subsystem (for example, to support
214 * calbacks on stat) will be required before we can support
215 * write delegations properly.
217 if (type != NFS4_OPEN_DELEGATE_READ)
219 if (fp->fi_had_conflict)
221 if (num_delegations > max_delegations)
223 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
227 INIT_LIST_HEAD(&dp->dl_perfile);
228 INIT_LIST_HEAD(&dp->dl_perclnt);
229 INIT_LIST_HEAD(&dp->dl_recall_lru);
233 nfs4_file_get_access(fp, O_RDONLY);
236 dp->dl_stateid.si_boot = boot_time;
237 dp->dl_stateid.si_stateownerid = current_delegid++;
238 dp->dl_stateid.si_fileid = 0;
239 dp->dl_stateid.si_generation = 0;
240 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
242 atomic_set(&dp->dl_count, 1);
243 list_add(&dp->dl_perfile, &fp->fi_delegations);
244 list_add(&dp->dl_perclnt, &clp->cl_delegations);
245 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
250 nfs4_put_delegation(struct nfs4_delegation *dp)
252 if (atomic_dec_and_test(&dp->dl_count)) {
253 dprintk("NFSD: freeing dp %p\n",dp);
254 put_nfs4_file(dp->dl_file);
255 kmem_cache_free(deleg_slab, dp);
260 /* Remove the associated file_lock first, then remove the delegation.
261 * lease_modify() is called to remove the FS_LEASE file_lock from
262 * the i_flock list, eventually calling nfsd's lock_manager
263 * fl_release_callback.
266 nfs4_close_delegation(struct nfs4_delegation *dp)
268 struct file *filp = find_readable_file(dp->dl_file);
270 dprintk("NFSD: close_delegation dp %p\n",dp);
272 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
273 nfs4_file_put_access(dp->dl_file, O_RDONLY);
276 /* Called under the state lock. */
278 unhash_delegation(struct nfs4_delegation *dp)
280 list_del_init(&dp->dl_perfile);
281 list_del_init(&dp->dl_perclnt);
282 spin_lock(&recall_lock);
283 list_del_init(&dp->dl_recall_lru);
284 spin_unlock(&recall_lock);
285 nfs4_close_delegation(dp);
286 nfs4_put_delegation(dp);
293 /* client_lock protects the client lru list and session hash table */
294 static DEFINE_SPINLOCK(client_lock);
296 /* Hash tables for nfs4_clientid state */
297 #define CLIENT_HASH_BITS 4
298 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
299 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
301 #define clientid_hashval(id) \
302 ((id) & CLIENT_HASH_MASK)
303 #define clientstr_hashval(name) \
304 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
306 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
307 * used in reboot/reset lease grace period processing
309 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
310 * setclientid_confirmed info.
312 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
315 * client_lru holds client queue ordered by nfs4_client.cl_time
318 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
319 * for last close replay.
321 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
322 static int reclaim_str_hashtbl_size = 0;
323 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
324 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
325 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
326 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
327 static struct list_head client_lru;
328 static struct list_head close_lru;
330 static void unhash_generic_stateid(struct nfs4_stateid *stp)
332 list_del(&stp->st_hash);
333 list_del(&stp->st_perfile);
334 list_del(&stp->st_perstateowner);
337 static void free_generic_stateid(struct nfs4_stateid *stp)
339 put_nfs4_file(stp->st_file);
340 kmem_cache_free(stateid_slab, stp);
343 static void release_lock_stateid(struct nfs4_stateid *stp)
347 unhash_generic_stateid(stp);
348 file = find_any_file(stp->st_file);
350 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
351 free_generic_stateid(stp);
354 static void unhash_lockowner(struct nfs4_stateowner *sop)
356 struct nfs4_stateid *stp;
358 list_del(&sop->so_idhash);
359 list_del(&sop->so_strhash);
360 list_del(&sop->so_perstateid);
361 while (!list_empty(&sop->so_stateids)) {
362 stp = list_first_entry(&sop->so_stateids,
363 struct nfs4_stateid, st_perstateowner);
364 release_lock_stateid(stp);
368 static void release_lockowner(struct nfs4_stateowner *sop)
370 unhash_lockowner(sop);
371 nfs4_put_stateowner(sop);
375 release_stateid_lockowners(struct nfs4_stateid *open_stp)
377 struct nfs4_stateowner *lock_sop;
379 while (!list_empty(&open_stp->st_lockowners)) {
380 lock_sop = list_entry(open_stp->st_lockowners.next,
381 struct nfs4_stateowner, so_perstateid);
382 /* list_del(&open_stp->st_lockowners); */
383 BUG_ON(lock_sop->so_is_open_owner);
384 release_lockowner(lock_sop);
389 * We store the NONE, READ, WRITE, and BOTH bits separately in the
390 * st_{access,deny}_bmap field of the stateid, in order to track not
391 * only what share bits are currently in force, but also what
392 * combinations of share bits previous opens have used. This allows us
393 * to enforce the recommendation of rfc 3530 14.2.19 that the server
394 * return an error if the client attempt to downgrade to a combination
395 * of share bits not explicable by closing some of its previous opens.
397 * XXX: This enforcement is actually incomplete, since we don't keep
398 * track of access/deny bit combinations; so, e.g., we allow:
400 * OPEN allow read, deny write
401 * OPEN allow both, deny none
402 * DOWNGRADE allow read, deny none
404 * which we should reject.
407 set_access(unsigned int *access, unsigned long bmap) {
411 for (i = 1; i < 4; i++) {
412 if (test_bit(i, &bmap))
418 set_deny(unsigned int *deny, unsigned long bmap) {
422 for (i = 0; i < 4; i++) {
423 if (test_bit(i, &bmap))
429 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
430 unsigned int access, deny;
432 set_access(&access, stp->st_access_bmap);
433 set_deny(&deny, stp->st_deny_bmap);
434 if ((access & open->op_share_deny) || (deny & open->op_share_access))
439 static int nfs4_access_to_omode(u32 access)
441 switch (access & NFS4_SHARE_ACCESS_BOTH) {
442 case NFS4_SHARE_ACCESS_READ:
444 case NFS4_SHARE_ACCESS_WRITE:
446 case NFS4_SHARE_ACCESS_BOTH:
452 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
456 set_access(&access, stp->st_access_bmap);
457 return nfs4_access_to_omode(access);
460 static void release_open_stateid(struct nfs4_stateid *stp)
462 int oflag = nfs4_access_bmap_to_omode(stp);
464 unhash_generic_stateid(stp);
465 release_stateid_lockowners(stp);
466 nfs4_file_put_access(stp->st_file, oflag);
467 free_generic_stateid(stp);
470 static void unhash_openowner(struct nfs4_stateowner *sop)
472 struct nfs4_stateid *stp;
474 list_del(&sop->so_idhash);
475 list_del(&sop->so_strhash);
476 list_del(&sop->so_perclient);
477 list_del(&sop->so_perstateid); /* XXX: necessary? */
478 while (!list_empty(&sop->so_stateids)) {
479 stp = list_first_entry(&sop->so_stateids,
480 struct nfs4_stateid, st_perstateowner);
481 release_open_stateid(stp);
485 static void release_openowner(struct nfs4_stateowner *sop)
487 unhash_openowner(sop);
488 list_del(&sop->so_close_lru);
489 nfs4_put_stateowner(sop);
492 #define SESSION_HASH_SIZE 512
493 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
496 hash_sessionid(struct nfs4_sessionid *sessionid)
498 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
500 return sid->sequence % SESSION_HASH_SIZE;
504 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
506 u32 *ptr = (u32 *)(&sessionid->data[0]);
507 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
511 gen_sessionid(struct nfsd4_session *ses)
513 struct nfs4_client *clp = ses->se_client;
514 struct nfsd4_sessionid *sid;
516 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
517 sid->clientid = clp->cl_clientid;
518 sid->sequence = current_sessionid++;
523 * The protocol defines ca_maxresponssize_cached to include the size of
524 * the rpc header, but all we need to cache is the data starting after
525 * the end of the initial SEQUENCE operation--the rest we regenerate
526 * each time. Therefore we can advertise a ca_maxresponssize_cached
527 * value that is the number of bytes in our cache plus a few additional
528 * bytes. In order to stay on the safe side, and not promise more than
529 * we can cache, those additional bytes must be the minimum possible: 24
530 * bytes of rpc header (xid through accept state, with AUTH_NULL
531 * verifier), 12 for the compound header (with zero-length tag), and 44
532 * for the SEQUENCE op response:
534 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
537 free_session_slots(struct nfsd4_session *ses)
541 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
542 kfree(ses->se_slots[i]);
546 * We don't actually need to cache the rpc and session headers, so we
547 * can allocate a little less for each slot:
549 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
551 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
554 static int nfsd4_sanitize_slot_size(u32 size)
556 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
557 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
563 * XXX: If we run out of reserved DRC memory we could (up to a point)
564 * re-negotiate active sessions and reduce their slot usage to make
565 * rooom for new connections. For now we just fail the create session.
567 static int nfsd4_get_drc_mem(int slotsize, u32 num)
571 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
573 spin_lock(&nfsd_drc_lock);
574 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
575 nfsd_drc_max_mem - nfsd_drc_mem_used);
576 num = min_t(int, num, avail / slotsize);
577 nfsd_drc_mem_used += num * slotsize;
578 spin_unlock(&nfsd_drc_lock);
583 static void nfsd4_put_drc_mem(int slotsize, int num)
585 spin_lock(&nfsd_drc_lock);
586 nfsd_drc_mem_used -= slotsize * num;
587 spin_unlock(&nfsd_drc_lock);
590 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
592 struct nfsd4_session *new;
595 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
596 + sizeof(struct nfsd4_session) > PAGE_SIZE);
597 mem = numslots * sizeof(struct nfsd4_slot *);
599 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
602 /* allocate each struct nfsd4_slot and data cache in one piece */
603 for (i = 0; i < numslots; i++) {
604 mem = sizeof(struct nfsd4_slot) + slotsize;
605 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
606 if (!new->se_slots[i])
612 kfree(new->se_slots[i]);
617 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
619 u32 maxrpc = nfsd_serv->sv_max_mesg;
621 new->maxreqs = numslots;
622 new->maxresp_cached = slotsize + NFSD_MIN_HDR_SEQ_SZ;
623 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
624 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
625 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
628 static __be32 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
630 struct nfsd4_session *new;
631 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
632 int numslots, slotsize;
636 * Note decreasing slot size below client's request may
637 * make it difficult for client to function correctly, whereas
638 * decreasing the number of slots will (just?) affect
639 * performance. When short on memory we therefore prefer to
640 * decrease number of slots instead of their size.
642 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
643 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
645 new = alloc_session(slotsize, numslots);
647 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
648 return nfserr_jukebox;
650 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
652 new->se_client = clp;
654 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
655 NFS4_MAX_SESSIONID_LEN);
657 new->se_flags = cses->flags;
658 kref_init(&new->se_ref);
659 idx = hash_sessionid(&new->se_sessionid);
660 spin_lock(&client_lock);
661 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
662 list_add(&new->se_perclnt, &clp->cl_sessions);
663 spin_unlock(&client_lock);
668 /* caller must hold client_lock */
669 static struct nfsd4_session *
670 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
672 struct nfsd4_session *elem;
675 dump_sessionid(__func__, sessionid);
676 idx = hash_sessionid(sessionid);
677 /* Search in the appropriate list */
678 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
679 if (!memcmp(elem->se_sessionid.data, sessionid->data,
680 NFS4_MAX_SESSIONID_LEN)) {
685 dprintk("%s: session not found\n", __func__);
689 /* caller must hold client_lock */
691 unhash_session(struct nfsd4_session *ses)
693 list_del(&ses->se_hash);
694 list_del(&ses->se_perclnt);
698 free_session(struct kref *kref)
700 struct nfsd4_session *ses;
703 ses = container_of(kref, struct nfsd4_session, se_ref);
704 spin_lock(&nfsd_drc_lock);
705 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
706 nfsd_drc_mem_used -= mem;
707 spin_unlock(&nfsd_drc_lock);
708 free_session_slots(ses);
712 /* must be called under the client_lock */
714 renew_client_locked(struct nfs4_client *clp)
716 if (is_client_expired(clp)) {
717 dprintk("%s: client (clientid %08x/%08x) already expired\n",
719 clp->cl_clientid.cl_boot,
720 clp->cl_clientid.cl_id);
725 * Move client to the end to the LRU list.
727 dprintk("renewing client (clientid %08x/%08x)\n",
728 clp->cl_clientid.cl_boot,
729 clp->cl_clientid.cl_id);
730 list_move_tail(&clp->cl_lru, &client_lru);
731 clp->cl_time = get_seconds();
735 renew_client(struct nfs4_client *clp)
737 spin_lock(&client_lock);
738 renew_client_locked(clp);
739 spin_unlock(&client_lock);
742 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
744 STALE_CLIENTID(clientid_t *clid)
746 if (clid->cl_boot == boot_time)
748 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
749 clid->cl_boot, clid->cl_id, boot_time);
754 * XXX Should we use a slab cache ?
755 * This type of memory management is somewhat inefficient, but we use it
756 * anyway since SETCLIENTID is not a common operation.
758 static struct nfs4_client *alloc_client(struct xdr_netobj name)
760 struct nfs4_client *clp;
762 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
765 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
766 if (clp->cl_name.data == NULL) {
770 memcpy(clp->cl_name.data, name.data, name.len);
771 clp->cl_name.len = name.len;
776 free_client(struct nfs4_client *clp)
778 if (clp->cl_cred.cr_group_info)
779 put_group_info(clp->cl_cred.cr_group_info);
780 kfree(clp->cl_principal);
781 kfree(clp->cl_name.data);
786 release_session_client(struct nfsd4_session *session)
788 struct nfs4_client *clp = session->se_client;
790 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
792 if (is_client_expired(clp)) {
794 session->se_client = NULL;
796 renew_client_locked(clp);
797 spin_unlock(&client_lock);
800 /* must be called under the client_lock */
802 unhash_client_locked(struct nfs4_client *clp)
804 mark_client_expired(clp);
805 list_del(&clp->cl_lru);
806 while (!list_empty(&clp->cl_sessions)) {
807 struct nfsd4_session *ses;
808 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
811 nfsd4_put_session(ses);
816 expire_client(struct nfs4_client *clp)
818 struct nfs4_stateowner *sop;
819 struct nfs4_delegation *dp;
820 struct list_head reaplist;
822 INIT_LIST_HEAD(&reaplist);
823 spin_lock(&recall_lock);
824 while (!list_empty(&clp->cl_delegations)) {
825 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
826 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
828 list_del_init(&dp->dl_perclnt);
829 list_move(&dp->dl_recall_lru, &reaplist);
831 spin_unlock(&recall_lock);
832 while (!list_empty(&reaplist)) {
833 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
834 list_del_init(&dp->dl_recall_lru);
835 unhash_delegation(dp);
837 while (!list_empty(&clp->cl_openowners)) {
838 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
839 release_openowner(sop);
841 nfsd4_shutdown_callback(clp);
842 if (clp->cl_cb_conn.cb_xprt)
843 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
844 list_del(&clp->cl_idhash);
845 list_del(&clp->cl_strhash);
846 spin_lock(&client_lock);
847 unhash_client_locked(clp);
848 if (atomic_read(&clp->cl_refcount) == 0)
850 spin_unlock(&client_lock);
853 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
855 memcpy(target->cl_verifier.data, source->data,
856 sizeof(target->cl_verifier.data));
859 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
861 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
862 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
865 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
867 target->cr_uid = source->cr_uid;
868 target->cr_gid = source->cr_gid;
869 target->cr_group_info = source->cr_group_info;
870 get_group_info(target->cr_group_info);
873 static int same_name(const char *n1, const char *n2)
875 return 0 == memcmp(n1, n2, HEXDIR_LEN);
879 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
881 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
885 same_clid(clientid_t *cl1, clientid_t *cl2)
887 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
890 /* XXX what about NGROUP */
892 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
894 return cr1->cr_uid == cr2->cr_uid;
897 static void gen_clid(struct nfs4_client *clp)
899 static u32 current_clientid = 1;
901 clp->cl_clientid.cl_boot = boot_time;
902 clp->cl_clientid.cl_id = current_clientid++;
905 static void gen_confirm(struct nfs4_client *clp)
910 p = (u32 *)clp->cl_confirm.data;
911 *p++ = get_seconds();
915 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
916 struct svc_rqst *rqstp, nfs4_verifier *verf)
918 struct nfs4_client *clp;
919 struct sockaddr *sa = svc_addr(rqstp);
922 clp = alloc_client(name);
926 princ = svc_gss_principal(rqstp);
928 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
929 if (clp->cl_principal == NULL) {
935 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
936 atomic_set(&clp->cl_refcount, 0);
937 atomic_set(&clp->cl_cb_set, 0);
938 INIT_LIST_HEAD(&clp->cl_idhash);
939 INIT_LIST_HEAD(&clp->cl_strhash);
940 INIT_LIST_HEAD(&clp->cl_openowners);
941 INIT_LIST_HEAD(&clp->cl_delegations);
942 INIT_LIST_HEAD(&clp->cl_sessions);
943 INIT_LIST_HEAD(&clp->cl_lru);
944 spin_lock_init(&clp->cl_lock);
945 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
946 clp->cl_time = get_seconds();
947 clear_bit(0, &clp->cl_cb_slot_busy);
948 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
949 copy_verf(clp, verf);
950 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
951 clp->cl_flavor = rqstp->rq_flavor;
952 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
958 static int check_name(struct xdr_netobj name)
962 if (name.len > NFS4_OPAQUE_LIMIT) {
963 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
970 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
972 unsigned int idhashval;
974 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
975 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
976 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
981 move_to_confirmed(struct nfs4_client *clp)
983 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
984 unsigned int strhashval;
986 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
987 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
988 strhashval = clientstr_hashval(clp->cl_recdir);
989 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
993 static struct nfs4_client *
994 find_confirmed_client(clientid_t *clid)
996 struct nfs4_client *clp;
997 unsigned int idhashval = clientid_hashval(clid->cl_id);
999 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1000 if (same_clid(&clp->cl_clientid, clid))
1006 static struct nfs4_client *
1007 find_unconfirmed_client(clientid_t *clid)
1009 struct nfs4_client *clp;
1010 unsigned int idhashval = clientid_hashval(clid->cl_id);
1012 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1013 if (same_clid(&clp->cl_clientid, clid))
1020 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
1021 * parameter. Matching is based on the fact the at least one of the
1022 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
1024 * FIXME: we need to unify the clientid namespaces for nfsv4.x
1025 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
1026 * and SET_CLIENTID{,_CONFIRM}
1029 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
1031 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
1032 return use_exchange_id == has_exchange_flags;
1035 static struct nfs4_client *
1036 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
1037 bool use_exchange_id)
1039 struct nfs4_client *clp;
1041 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1042 if (same_name(clp->cl_recdir, dname) &&
1043 match_clientid_establishment(clp, use_exchange_id))
1049 static struct nfs4_client *
1050 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
1051 bool use_exchange_id)
1053 struct nfs4_client *clp;
1055 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1056 if (same_name(clp->cl_recdir, dname) &&
1057 match_clientid_establishment(clp, use_exchange_id))
1064 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1066 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1067 unsigned short expected_family;
1069 /* Currently, we only support tcp and tcp6 for the callback channel */
1070 if (se->se_callback_netid_len == 3 &&
1071 !memcmp(se->se_callback_netid_val, "tcp", 3))
1072 expected_family = AF_INET;
1073 else if (se->se_callback_netid_len == 4 &&
1074 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1075 expected_family = AF_INET6;
1079 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1080 se->se_callback_addr_len,
1081 (struct sockaddr *)&conn->cb_addr,
1082 sizeof(conn->cb_addr));
1084 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1087 if (conn->cb_addr.ss_family == AF_INET6)
1088 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1090 conn->cb_minorversion = 0;
1091 conn->cb_prog = se->se_callback_prog;
1092 conn->cb_ident = se->se_callback_ident;
1095 conn->cb_addr.ss_family = AF_UNSPEC;
1096 conn->cb_addrlen = 0;
1097 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1098 "will not receive delegations\n",
1099 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1105 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1108 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1110 struct nfsd4_slot *slot = resp->cstate.slot;
1113 dprintk("--> %s slot %p\n", __func__, slot);
1115 slot->sl_opcnt = resp->opcnt;
1116 slot->sl_status = resp->cstate.status;
1118 if (nfsd4_not_cached(resp)) {
1119 slot->sl_datalen = 0;
1122 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1123 base = (char *)resp->cstate.datap -
1124 (char *)resp->xbuf->head[0].iov_base;
1125 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1127 WARN("%s: sessions DRC could not cache compound\n", __func__);
1132 * Encode the replay sequence operation from the slot values.
1133 * If cachethis is FALSE encode the uncached rep error on the next
1134 * operation which sets resp->p and increments resp->opcnt for
1135 * nfs4svc_encode_compoundres.
1139 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1140 struct nfsd4_compoundres *resp)
1142 struct nfsd4_op *op;
1143 struct nfsd4_slot *slot = resp->cstate.slot;
1145 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1146 resp->opcnt, resp->cstate.slot->sl_cachethis);
1148 /* Encode the replayed sequence operation */
1149 op = &args->ops[resp->opcnt - 1];
1150 nfsd4_encode_operation(resp, op);
1152 /* Return nfserr_retry_uncached_rep in next operation. */
1153 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1154 op = &args->ops[resp->opcnt++];
1155 op->status = nfserr_retry_uncached_rep;
1156 nfsd4_encode_operation(resp, op);
1162 * The sequence operation is not cached because we can use the slot and
1166 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1167 struct nfsd4_sequence *seq)
1169 struct nfsd4_slot *slot = resp->cstate.slot;
1172 dprintk("--> %s slot %p\n", __func__, slot);
1174 /* Either returns 0 or nfserr_retry_uncached */
1175 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1176 if (status == nfserr_retry_uncached_rep)
1179 /* The sequence operation has been encoded, cstate->datap set. */
1180 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1182 resp->opcnt = slot->sl_opcnt;
1183 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1184 status = slot->sl_status;
1190 * Set the exchange_id flags returned by the server.
1193 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1195 /* pNFS is not supported */
1196 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1198 /* Referrals are supported, Migration is not. */
1199 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1201 /* set the wire flags to return to client. */
1202 clid->flags = new->cl_exchange_flags;
1206 nfsd4_exchange_id(struct svc_rqst *rqstp,
1207 struct nfsd4_compound_state *cstate,
1208 struct nfsd4_exchange_id *exid)
1210 struct nfs4_client *unconf, *conf, *new;
1212 unsigned int strhashval;
1213 char dname[HEXDIR_LEN];
1214 char addr_str[INET6_ADDRSTRLEN];
1215 nfs4_verifier verf = exid->verifier;
1216 struct sockaddr *sa = svc_addr(rqstp);
1218 rpc_ntop(sa, addr_str, sizeof(addr_str));
1219 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1220 "ip_addr=%s flags %x, spa_how %d\n",
1221 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1222 addr_str, exid->flags, exid->spa_how);
1224 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1225 return nfserr_inval;
1227 /* Currently only support SP4_NONE */
1228 switch (exid->spa_how) {
1232 return nfserr_encr_alg_unsupp;
1234 BUG(); /* checked by xdr code */
1236 return nfserr_serverfault; /* no excuse :-/ */
1239 status = nfs4_make_rec_clidname(dname, &exid->clname);
1244 strhashval = clientstr_hashval(dname);
1249 conf = find_confirmed_client_by_str(dname, strhashval, true);
1251 if (!same_verf(&verf, &conf->cl_verifier)) {
1252 /* 18.35.4 case 8 */
1253 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1254 status = nfserr_not_same;
1257 /* Client reboot: destroy old state */
1258 expire_client(conf);
1261 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1262 /* 18.35.4 case 9 */
1263 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1264 status = nfserr_perm;
1267 expire_client(conf);
1271 * Set bit when the owner id and verifier map to an already
1272 * confirmed client id (18.35.3).
1274 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1277 * Falling into 18.35.4 case 2, possible router replay.
1278 * Leave confirmed record intact and return same result.
1280 copy_verf(conf, &verf);
1285 /* 18.35.4 case 7 */
1286 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1287 status = nfserr_noent;
1291 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1294 * Possible retry or client restart. Per 18.35.4 case 4,
1295 * a new unconfirmed record should be generated regardless
1296 * of whether any properties have changed.
1298 expire_client(unconf);
1303 new = create_client(exid->clname, dname, rqstp, &verf);
1305 status = nfserr_jukebox;
1310 add_to_unconfirmed(new, strhashval);
1312 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1313 exid->clientid.cl_id = new->cl_clientid.cl_id;
1316 nfsd4_set_ex_flags(new, exid);
1318 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1319 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1323 nfs4_unlock_state();
1325 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1330 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1332 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1335 /* The slot is in use, and no response has been sent. */
1337 if (seqid == slot_seqid)
1338 return nfserr_jukebox;
1340 return nfserr_seq_misordered;
1343 if (likely(seqid == slot_seqid + 1))
1346 if (seqid == slot_seqid)
1347 return nfserr_replay_cache;
1349 if (seqid == 1 && (slot_seqid + 1) == 0)
1351 /* Misordered replay or misordered new request */
1352 return nfserr_seq_misordered;
1356 * Cache the create session result into the create session single DRC
1357 * slot cache by saving the xdr structure. sl_seqid has been set.
1358 * Do this for solo or embedded create session operations.
1361 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1362 struct nfsd4_clid_slot *slot, int nfserr)
1364 slot->sl_status = nfserr;
1365 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1369 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1370 struct nfsd4_clid_slot *slot)
1372 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1373 return slot->sl_status;
1377 nfsd4_create_session(struct svc_rqst *rqstp,
1378 struct nfsd4_compound_state *cstate,
1379 struct nfsd4_create_session *cr_ses)
1381 struct sockaddr *sa = svc_addr(rqstp);
1382 struct nfs4_client *conf, *unconf;
1383 struct nfsd4_clid_slot *cs_slot = NULL;
1387 unconf = find_unconfirmed_client(&cr_ses->clientid);
1388 conf = find_confirmed_client(&cr_ses->clientid);
1391 cs_slot = &conf->cl_cs_slot;
1392 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1393 if (status == nfserr_replay_cache) {
1394 dprintk("Got a create_session replay! seqid= %d\n",
1396 /* Return the cached reply status */
1397 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1399 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1400 status = nfserr_seq_misordered;
1401 dprintk("Sequence misordered!\n");
1402 dprintk("Expected seqid= %d but got seqid= %d\n",
1403 cs_slot->sl_seqid, cr_ses->seqid);
1406 cs_slot->sl_seqid++;
1407 } else if (unconf) {
1408 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1409 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1410 status = nfserr_clid_inuse;
1414 cs_slot = &unconf->cl_cs_slot;
1415 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1417 /* an unconfirmed replay returns misordered */
1418 status = nfserr_seq_misordered;
1422 cs_slot->sl_seqid++; /* from 0 to 1 */
1423 move_to_confirmed(unconf);
1425 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1426 unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1427 svc_xprt_get(rqstp->rq_xprt);
1429 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1431 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1432 unconf->cl_cb_conn.cb_minorversion =
1433 cstate->minorversion;
1434 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1435 unconf->cl_cb_seq_nr = 1;
1436 nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1440 status = nfserr_stale_clientid;
1445 * We do not support RDMA or persistent sessions
1447 cr_ses->flags &= ~SESSION4_PERSIST;
1448 cr_ses->flags &= ~SESSION4_RDMA;
1450 status = alloc_init_session(rqstp, conf, cr_ses);
1454 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1455 NFS4_MAX_SESSIONID_LEN);
1456 cr_ses->seqid = cs_slot->sl_seqid;
1459 /* cache solo and embedded create sessions under the state lock */
1460 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1462 nfs4_unlock_state();
1463 dprintk("%s returns %d\n", __func__, ntohl(status));
1467 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1469 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1470 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1472 return argp->opcnt == resp->opcnt;
1475 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1479 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1483 nfsd4_destroy_session(struct svc_rqst *r,
1484 struct nfsd4_compound_state *cstate,
1485 struct nfsd4_destroy_session *sessionid)
1487 struct nfsd4_session *ses;
1488 u32 status = nfserr_badsession;
1491 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1492 * - Should we return nfserr_back_chan_busy if waiting for
1493 * callbacks on to-be-destroyed session?
1494 * - Do we need to clear any callback info from previous session?
1497 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1498 if (!nfsd4_last_compound_op(r))
1499 return nfserr_not_only_op;
1501 dump_sessionid(__func__, &sessionid->sessionid);
1502 spin_lock(&client_lock);
1503 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1505 spin_unlock(&client_lock);
1509 unhash_session(ses);
1510 spin_unlock(&client_lock);
1513 /* wait for callbacks */
1514 nfsd4_shutdown_callback(ses->se_client);
1515 nfs4_unlock_state();
1516 nfsd4_put_session(ses);
1519 dprintk("%s returns %d\n", __func__, ntohl(status));
1524 nfsd4_sequence(struct svc_rqst *rqstp,
1525 struct nfsd4_compound_state *cstate,
1526 struct nfsd4_sequence *seq)
1528 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1529 struct nfsd4_session *session;
1530 struct nfsd4_slot *slot;
1533 if (resp->opcnt != 1)
1534 return nfserr_sequence_pos;
1536 spin_lock(&client_lock);
1537 status = nfserr_badsession;
1538 session = find_in_sessionid_hashtbl(&seq->sessionid);
1542 status = nfserr_badslot;
1543 if (seq->slotid >= session->se_fchannel.maxreqs)
1546 slot = session->se_slots[seq->slotid];
1547 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1549 /* We do not negotiate the number of slots yet, so set the
1550 * maxslots to the session maxreqs which is used to encode
1551 * sr_highest_slotid and the sr_target_slot id to maxslots */
1552 seq->maxslots = session->se_fchannel.maxreqs;
1554 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1555 if (status == nfserr_replay_cache) {
1556 cstate->slot = slot;
1557 cstate->session = session;
1558 /* Return the cached reply status and set cstate->status
1559 * for nfsd4_proc_compound processing */
1560 status = nfsd4_replay_cache_entry(resp, seq);
1561 cstate->status = nfserr_replay_cache;
1567 /* Success! bump slot seqid */
1568 slot->sl_inuse = true;
1569 slot->sl_seqid = seq->seqid;
1570 slot->sl_cachethis = seq->cachethis;
1572 cstate->slot = slot;
1573 cstate->session = session;
1576 /* Hold a session reference until done processing the compound. */
1577 if (cstate->session) {
1578 nfsd4_get_session(cstate->session);
1579 atomic_inc(&session->se_client->cl_refcount);
1581 spin_unlock(&client_lock);
1582 dprintk("%s: return %d\n", __func__, ntohl(status));
1587 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1589 if (rc->rca_one_fs) {
1590 if (!cstate->current_fh.fh_dentry)
1591 return nfserr_nofilehandle;
1593 * We don't take advantage of the rca_one_fs case.
1594 * That's OK, it's optional, we can safely ignore it.
1599 if (is_client_expired(cstate->session->se_client)) {
1600 nfs4_unlock_state();
1602 * The following error isn't really legal.
1603 * But we only get here if the client just explicitly
1604 * destroyed the client. Surely it no longer cares what
1605 * error it gets back on an operation for the dead
1608 return nfserr_stale_clientid;
1610 nfsd4_create_clid_dir(cstate->session->se_client);
1611 nfs4_unlock_state();
1616 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1617 struct nfsd4_setclientid *setclid)
1619 struct sockaddr *sa = svc_addr(rqstp);
1620 struct xdr_netobj clname = {
1621 .len = setclid->se_namelen,
1622 .data = setclid->se_name,
1624 nfs4_verifier clverifier = setclid->se_verf;
1625 unsigned int strhashval;
1626 struct nfs4_client *conf, *unconf, *new;
1628 char dname[HEXDIR_LEN];
1630 if (!check_name(clname))
1631 return nfserr_inval;
1633 status = nfs4_make_rec_clidname(dname, &clname);
1638 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1639 * We get here on a DRC miss.
1642 strhashval = clientstr_hashval(dname);
1645 conf = find_confirmed_client_by_str(dname, strhashval, false);
1647 /* RFC 3530 14.2.33 CASE 0: */
1648 status = nfserr_clid_inuse;
1649 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1650 char addr_str[INET6_ADDRSTRLEN];
1651 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1653 dprintk("NFSD: setclientid: string in use by client "
1654 "at %s\n", addr_str);
1659 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1660 * has a description of SETCLIENTID request processing consisting
1661 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1663 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1664 status = nfserr_resource;
1667 * RFC 3530 14.2.33 CASE 4:
1668 * placed first, because it is the normal case
1671 expire_client(unconf);
1672 new = create_client(clname, dname, rqstp, &clverifier);
1676 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1678 * RFC 3530 14.2.33 CASE 1:
1679 * probable callback update
1682 /* Note this is removing unconfirmed {*x***},
1683 * which is stronger than RFC recommended {vxc**}.
1684 * This has the advantage that there is at most
1685 * one {*x***} in either list at any time.
1687 expire_client(unconf);
1689 new = create_client(clname, dname, rqstp, &clverifier);
1692 copy_clid(new, conf);
1693 } else if (!unconf) {
1695 * RFC 3530 14.2.33 CASE 2:
1696 * probable client reboot; state will be removed if
1699 new = create_client(clname, dname, rqstp, &clverifier);
1705 * RFC 3530 14.2.33 CASE 3:
1706 * probable client reboot; state will be removed if
1709 expire_client(unconf);
1710 new = create_client(clname, dname, rqstp, &clverifier);
1715 gen_callback(new, setclid, rpc_get_scope_id(sa));
1716 add_to_unconfirmed(new, strhashval);
1717 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1718 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1719 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1722 nfs4_unlock_state();
1728 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1729 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1730 * bullets, labeled as CASE1 - CASE4 below.
1733 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1734 struct nfsd4_compound_state *cstate,
1735 struct nfsd4_setclientid_confirm *setclientid_confirm)
1737 struct sockaddr *sa = svc_addr(rqstp);
1738 struct nfs4_client *conf, *unconf;
1739 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1740 clientid_t * clid = &setclientid_confirm->sc_clientid;
1743 if (STALE_CLIENTID(clid))
1744 return nfserr_stale_clientid;
1746 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1747 * We get here on a DRC miss.
1752 conf = find_confirmed_client(clid);
1753 unconf = find_unconfirmed_client(clid);
1755 status = nfserr_clid_inuse;
1756 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1758 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1762 * section 14.2.34 of RFC 3530 has a description of
1763 * SETCLIENTID_CONFIRM request processing consisting
1764 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1766 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1768 * RFC 3530 14.2.34 CASE 1:
1771 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1772 status = nfserr_clid_inuse;
1774 atomic_set(&conf->cl_cb_set, 0);
1775 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1776 expire_client(unconf);
1780 } else if (conf && !unconf) {
1782 * RFC 3530 14.2.34 CASE 2:
1783 * probable retransmitted request; play it safe and
1786 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1787 status = nfserr_clid_inuse;
1790 } else if (!conf && unconf
1791 && same_verf(&unconf->cl_confirm, &confirm)) {
1793 * RFC 3530 14.2.34 CASE 3:
1794 * Normal case; new or rebooted client:
1796 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1797 status = nfserr_clid_inuse;
1800 clientstr_hashval(unconf->cl_recdir);
1801 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1804 nfsd4_remove_clid_dir(conf);
1805 expire_client(conf);
1807 move_to_confirmed(unconf);
1809 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1812 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1813 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1816 * RFC 3530 14.2.34 CASE 4:
1817 * Client probably hasn't noticed that we rebooted yet.
1819 status = nfserr_stale_clientid;
1821 /* check that we have hit one of the cases...*/
1822 status = nfserr_clid_inuse;
1825 nfs4_unlock_state();
1829 /* OPEN Share state helper functions */
1830 static inline struct nfs4_file *
1831 alloc_init_file(struct inode *ino)
1833 struct nfs4_file *fp;
1834 unsigned int hashval = file_hashval(ino);
1836 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1838 atomic_set(&fp->fi_ref, 1);
1839 INIT_LIST_HEAD(&fp->fi_hash);
1840 INIT_LIST_HEAD(&fp->fi_stateids);
1841 INIT_LIST_HEAD(&fp->fi_delegations);
1842 fp->fi_inode = igrab(ino);
1843 fp->fi_id = current_fileid++;
1844 fp->fi_had_conflict = false;
1845 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
1846 memset(fp->fi_access, 0, sizeof(fp->fi_access));
1847 spin_lock(&recall_lock);
1848 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1849 spin_unlock(&recall_lock);
1856 nfsd4_free_slab(struct kmem_cache **slab)
1860 kmem_cache_destroy(*slab);
1865 nfsd4_free_slabs(void)
1867 nfsd4_free_slab(&stateowner_slab);
1868 nfsd4_free_slab(&file_slab);
1869 nfsd4_free_slab(&stateid_slab);
1870 nfsd4_free_slab(&deleg_slab);
1874 nfsd4_init_slabs(void)
1876 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1877 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1878 if (stateowner_slab == NULL)
1880 file_slab = kmem_cache_create("nfsd4_files",
1881 sizeof(struct nfs4_file), 0, 0, NULL);
1882 if (file_slab == NULL)
1884 stateid_slab = kmem_cache_create("nfsd4_stateids",
1885 sizeof(struct nfs4_stateid), 0, 0, NULL);
1886 if (stateid_slab == NULL)
1888 deleg_slab = kmem_cache_create("nfsd4_delegations",
1889 sizeof(struct nfs4_delegation), 0, 0, NULL);
1890 if (deleg_slab == NULL)
1895 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1900 nfs4_free_stateowner(struct kref *kref)
1902 struct nfs4_stateowner *sop =
1903 container_of(kref, struct nfs4_stateowner, so_ref);
1904 kfree(sop->so_owner.data);
1905 kmem_cache_free(stateowner_slab, sop);
1908 static inline struct nfs4_stateowner *
1909 alloc_stateowner(struct xdr_netobj *owner)
1911 struct nfs4_stateowner *sop;
1913 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1914 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1915 memcpy(sop->so_owner.data, owner->data, owner->len);
1916 sop->so_owner.len = owner->len;
1917 kref_init(&sop->so_ref);
1920 kmem_cache_free(stateowner_slab, sop);
1925 static struct nfs4_stateowner *
1926 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1927 struct nfs4_stateowner *sop;
1928 struct nfs4_replay *rp;
1929 unsigned int idhashval;
1931 if (!(sop = alloc_stateowner(&open->op_owner)))
1933 idhashval = ownerid_hashval(current_ownerid);
1934 INIT_LIST_HEAD(&sop->so_idhash);
1935 INIT_LIST_HEAD(&sop->so_strhash);
1936 INIT_LIST_HEAD(&sop->so_perclient);
1937 INIT_LIST_HEAD(&sop->so_stateids);
1938 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1939 INIT_LIST_HEAD(&sop->so_close_lru);
1941 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1942 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1943 list_add(&sop->so_perclient, &clp->cl_openowners);
1944 sop->so_is_open_owner = 1;
1945 sop->so_id = current_ownerid++;
1946 sop->so_client = clp;
1947 sop->so_seqid = open->op_seqid;
1948 sop->so_confirmed = 0;
1949 rp = &sop->so_replay;
1950 rp->rp_status = nfserr_serverfault;
1952 rp->rp_buf = rp->rp_ibuf;
1957 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1958 struct nfs4_stateowner *sop = open->op_stateowner;
1959 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1961 INIT_LIST_HEAD(&stp->st_hash);
1962 INIT_LIST_HEAD(&stp->st_perstateowner);
1963 INIT_LIST_HEAD(&stp->st_lockowners);
1964 INIT_LIST_HEAD(&stp->st_perfile);
1965 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1966 list_add(&stp->st_perstateowner, &sop->so_stateids);
1967 list_add(&stp->st_perfile, &fp->fi_stateids);
1968 stp->st_stateowner = sop;
1971 stp->st_stateid.si_boot = boot_time;
1972 stp->st_stateid.si_stateownerid = sop->so_id;
1973 stp->st_stateid.si_fileid = fp->fi_id;
1974 stp->st_stateid.si_generation = 0;
1975 stp->st_access_bmap = 0;
1976 stp->st_deny_bmap = 0;
1977 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1978 &stp->st_access_bmap);
1979 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1980 stp->st_openstp = NULL;
1984 move_to_close_lru(struct nfs4_stateowner *sop)
1986 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1988 list_move_tail(&sop->so_close_lru, &close_lru);
1989 sop->so_time = get_seconds();
1993 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1996 return (sop->so_owner.len == owner->len) &&
1997 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1998 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2001 static struct nfs4_stateowner *
2002 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2004 struct nfs4_stateowner *so = NULL;
2006 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2007 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2013 /* search file_hashtbl[] for file */
2014 static struct nfs4_file *
2015 find_file(struct inode *ino)
2017 unsigned int hashval = file_hashval(ino);
2018 struct nfs4_file *fp;
2020 spin_lock(&recall_lock);
2021 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2022 if (fp->fi_inode == ino) {
2024 spin_unlock(&recall_lock);
2028 spin_unlock(&recall_lock);
2032 static inline int access_valid(u32 x, u32 minorversion)
2034 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2036 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2038 x &= ~NFS4_SHARE_ACCESS_MASK;
2039 if (minorversion && x) {
2040 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2042 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2044 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2051 static inline int deny_valid(u32 x)
2053 /* Note: unlike access bits, deny bits may be zero. */
2054 return x <= NFS4_SHARE_DENY_BOTH;
2058 * Called to check deny when READ with all zero stateid or
2059 * WRITE with all zero or all one stateid
2062 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2064 struct inode *ino = current_fh->fh_dentry->d_inode;
2065 struct nfs4_file *fp;
2066 struct nfs4_stateid *stp;
2069 dprintk("NFSD: nfs4_share_conflict\n");
2071 fp = find_file(ino);
2074 ret = nfserr_locked;
2075 /* Search for conflicting share reservations */
2076 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2077 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2078 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2088 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2090 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2091 nfs4_file_put_access(fp, O_WRONLY);
2092 if (share_access & NFS4_SHARE_ACCESS_READ)
2093 nfs4_file_put_access(fp, O_RDONLY);
2097 * Spawn a thread to perform a recall on the delegation represented
2098 * by the lease (file_lock)
2100 * Called from break_lease() with lock_kernel() held.
2101 * Note: we assume break_lease will only call this *once* for any given
2105 void nfsd_break_deleg_cb(struct file_lock *fl)
2107 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2109 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2113 /* We're assuming the state code never drops its reference
2114 * without first removing the lease. Since we're in this lease
2115 * callback (and since the lease code is serialized by the kernel
2116 * lock) we know the server hasn't removed the lease yet, we know
2117 * it's safe to take a reference: */
2118 atomic_inc(&dp->dl_count);
2120 spin_lock(&recall_lock);
2121 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2122 spin_unlock(&recall_lock);
2124 /* only place dl_time is set. protected by lock_kernel*/
2125 dp->dl_time = get_seconds();
2128 * We don't want the locks code to timeout the lease for us;
2129 * we'll remove it ourself if the delegation isn't returned
2132 fl->fl_break_time = 0;
2134 dp->dl_file->fi_had_conflict = true;
2135 nfsd4_cb_recall(dp);
2139 * The file_lock is being reapd.
2141 * Called by locks_free_lock() with lock_kernel() held.
2144 void nfsd_release_deleg_cb(struct file_lock *fl)
2146 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2148 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2150 if (!(fl->fl_flags & FL_LEASE) || !dp)
2152 dp->dl_flock = NULL;
2156 * Set the delegation file_lock back pointer.
2158 * Called from setlease() with lock_kernel() held.
2161 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2163 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2165 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2172 * Called from setlease() with lock_kernel() held
2175 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2177 struct nfs4_delegation *onlistd =
2178 (struct nfs4_delegation *)onlist->fl_owner;
2179 struct nfs4_delegation *tryd =
2180 (struct nfs4_delegation *)try->fl_owner;
2182 if (onlist->fl_lmops != try->fl_lmops)
2185 return onlistd->dl_client == tryd->dl_client;
2190 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2193 return lease_modify(onlist, arg);
2198 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2199 .fl_break = nfsd_break_deleg_cb,
2200 .fl_release_private = nfsd_release_deleg_cb,
2201 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2202 .fl_mylease = nfsd_same_client_deleg_cb,
2203 .fl_change = nfsd_change_deleg_cb,
2208 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2209 struct nfsd4_open *open)
2211 clientid_t *clientid = &open->op_clientid;
2212 struct nfs4_client *clp = NULL;
2213 unsigned int strhashval;
2214 struct nfs4_stateowner *sop = NULL;
2216 if (!check_name(open->op_owner))
2217 return nfserr_inval;
2219 if (STALE_CLIENTID(&open->op_clientid))
2220 return nfserr_stale_clientid;
2222 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2223 sop = find_openstateowner_str(strhashval, open);
2224 open->op_stateowner = sop;
2226 /* Make sure the client's lease hasn't expired. */
2227 clp = find_confirmed_client(clientid);
2229 return nfserr_expired;
2232 /* When sessions are used, skip open sequenceid processing */
2233 if (nfsd4_has_session(cstate))
2235 if (!sop->so_confirmed) {
2236 /* Replace unconfirmed owners without checking for replay. */
2237 clp = sop->so_client;
2238 release_openowner(sop);
2239 open->op_stateowner = NULL;
2242 if (open->op_seqid == sop->so_seqid - 1) {
2243 if (sop->so_replay.rp_buflen)
2244 return nfserr_replay_me;
2245 /* The original OPEN failed so spectacularly
2246 * that we don't even have replay data saved!
2247 * Therefore, we have no choice but to continue
2248 * processing this OPEN; presumably, we'll
2249 * fail again for the same reason.
2251 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2254 if (open->op_seqid != sop->so_seqid)
2255 return nfserr_bad_seqid;
2257 if (open->op_stateowner == NULL) {
2258 sop = alloc_init_open_stateowner(strhashval, clp, open);
2260 return nfserr_resource;
2261 open->op_stateowner = sop;
2263 list_del_init(&sop->so_close_lru);
2264 renew_client(sop->so_client);
2268 static inline __be32
2269 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2271 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2272 return nfserr_openmode;
2277 static struct nfs4_delegation *
2278 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2280 struct nfs4_delegation *dp;
2282 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2283 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2289 int share_access_to_flags(u32 share_access)
2291 share_access &= ~NFS4_SHARE_WANT_MASK;
2293 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2297 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2298 struct nfs4_delegation **dp)
2301 __be32 status = nfserr_bad_stateid;
2303 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2306 flags = share_access_to_flags(open->op_share_access);
2307 status = nfs4_check_delegmode(*dp, flags);
2311 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2315 open->op_stateowner->so_confirmed = 1;
2320 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2322 struct nfs4_stateid *local;
2323 __be32 status = nfserr_share_denied;
2324 struct nfs4_stateowner *sop = open->op_stateowner;
2326 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2327 /* ignore lock owners */
2328 if (local->st_stateowner->so_is_open_owner == 0)
2330 /* remember if we have seen this open owner */
2331 if (local->st_stateowner == sop)
2333 /* check for conflicting share reservations */
2334 if (!test_share(local, open))
2342 static inline struct nfs4_stateid *
2343 nfs4_alloc_stateid(void)
2345 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2348 static inline int nfs4_access_to_access(u32 nfs4_access)
2352 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2353 flags |= NFSD_MAY_READ;
2354 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2355 flags |= NFSD_MAY_WRITE;
2359 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2360 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2363 int oflag = nfs4_access_to_omode(nfs4_access);
2364 int access = nfs4_access_to_access(nfs4_access);
2366 if (!fp->fi_fds[oflag]) {
2367 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2368 &fp->fi_fds[oflag]);
2369 if (status == nfserr_dropit)
2370 status = nfserr_jukebox;
2374 nfs4_file_get_access(fp, oflag);
2380 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2381 struct nfs4_file *fp, struct svc_fh *cur_fh,
2382 struct nfsd4_open *open)
2384 struct nfs4_stateid *stp;
2387 stp = nfs4_alloc_stateid();
2389 return nfserr_resource;
2391 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2393 kmem_cache_free(stateid_slab, stp);
2400 static inline __be32
2401 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2402 struct nfsd4_open *open)
2404 struct iattr iattr = {
2405 .ia_valid = ATTR_SIZE,
2408 if (!open->op_truncate)
2410 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2411 return nfserr_inval;
2412 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2416 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2418 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2422 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2424 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2428 status = nfsd4_truncate(rqstp, cur_fh, open);
2431 int oflag = nfs4_access_to_omode(new_access);
2432 nfs4_file_put_access(fp, oflag);
2436 /* remember the open */
2437 __set_bit(op_share_access, &stp->st_access_bmap);
2438 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2445 nfs4_set_claim_prev(struct nfsd4_open *open)
2447 open->op_stateowner->so_confirmed = 1;
2448 open->op_stateowner->so_client->cl_firststate = 1;
2452 * Attempt to hand out a delegation.
2455 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2457 struct nfs4_delegation *dp;
2458 struct nfs4_stateowner *sop = stp->st_stateowner;
2459 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2460 struct file_lock fl, *flp = &fl;
2461 int status, flag = 0;
2463 flag = NFS4_OPEN_DELEGATE_NONE;
2464 open->op_recall = 0;
2465 switch (open->op_claim_type) {
2466 case NFS4_OPEN_CLAIM_PREVIOUS:
2468 open->op_recall = 1;
2469 flag = open->op_delegate_type;
2470 if (flag == NFS4_OPEN_DELEGATE_NONE)
2473 case NFS4_OPEN_CLAIM_NULL:
2474 /* Let's not give out any delegations till everyone's
2475 * had the chance to reclaim theirs.... */
2476 if (locks_in_grace())
2478 if (!cb_up || !sop->so_confirmed)
2480 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2481 flag = NFS4_OPEN_DELEGATE_WRITE;
2483 flag = NFS4_OPEN_DELEGATE_READ;
2489 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2491 flag = NFS4_OPEN_DELEGATE_NONE;
2494 locks_init_lock(&fl);
2495 fl.fl_lmops = &nfsd_lease_mng_ops;
2496 fl.fl_flags = FL_LEASE;
2497 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2498 fl.fl_end = OFFSET_MAX;
2499 fl.fl_owner = (fl_owner_t)dp;
2500 fl.fl_file = find_readable_file(stp->st_file);
2501 BUG_ON(!fl.fl_file);
2502 fl.fl_pid = current->tgid;
2504 /* vfs_setlease checks to see if delegation should be handed out.
2505 * the lock_manager callbacks fl_mylease and fl_change are used
2507 if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
2508 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2509 unhash_delegation(dp);
2510 flag = NFS4_OPEN_DELEGATE_NONE;
2514 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2516 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2517 STATEID_VAL(&dp->dl_stateid));
2519 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2520 && flag == NFS4_OPEN_DELEGATE_NONE
2521 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2522 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2523 open->op_delegate_type = flag;
2527 * called with nfs4_lock_state() held.
2530 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2532 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2533 struct nfs4_file *fp = NULL;
2534 struct inode *ino = current_fh->fh_dentry->d_inode;
2535 struct nfs4_stateid *stp = NULL;
2536 struct nfs4_delegation *dp = NULL;
2539 status = nfserr_inval;
2540 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2541 || !deny_valid(open->op_share_deny))
2544 * Lookup file; if found, lookup stateid and check open request,
2545 * and check for delegations in the process of being recalled.
2546 * If not found, create the nfs4_file struct
2548 fp = find_file(ino);
2550 if ((status = nfs4_check_open(fp, open, &stp)))
2552 status = nfs4_check_deleg(fp, open, &dp);
2556 status = nfserr_bad_stateid;
2557 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2559 status = nfserr_resource;
2560 fp = alloc_init_file(ino);
2566 * OPEN the file, or upgrade an existing OPEN.
2567 * If truncate fails, the OPEN fails.
2570 /* Stateid was found, this is an OPEN upgrade */
2571 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2574 update_stateid(&stp->st_stateid);
2576 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2579 init_stateid(stp, fp, open);
2580 status = nfsd4_truncate(rqstp, current_fh, open);
2582 release_open_stateid(stp);
2585 if (nfsd4_has_session(&resp->cstate))
2586 update_stateid(&stp->st_stateid);
2588 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2590 if (nfsd4_has_session(&resp->cstate))
2591 open->op_stateowner->so_confirmed = 1;
2594 * Attempt to hand out a delegation. No error return, because the
2595 * OPEN succeeds even if we fail.
2597 nfs4_open_delegation(current_fh, open, stp);
2601 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2602 STATEID_VAL(&stp->st_stateid));
2606 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2607 nfs4_set_claim_prev(open);
2609 * To finish the open response, we just need to set the rflags.
2611 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2612 if (!open->op_stateowner->so_confirmed &&
2613 !nfsd4_has_session(&resp->cstate))
2614 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2620 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2623 struct nfs4_client *clp;
2627 dprintk("process_renew(%08x/%08x): starting\n",
2628 clid->cl_boot, clid->cl_id);
2629 status = nfserr_stale_clientid;
2630 if (STALE_CLIENTID(clid))
2632 clp = find_confirmed_client(clid);
2633 status = nfserr_expired;
2635 /* We assume the client took too long to RENEW. */
2636 dprintk("nfsd4_renew: clientid not found!\n");
2640 status = nfserr_cb_path_down;
2641 if (!list_empty(&clp->cl_delegations)
2642 && !atomic_read(&clp->cl_cb_set))
2646 nfs4_unlock_state();
2650 struct lock_manager nfsd4_manager = {
2654 nfsd4_end_grace(void)
2656 dprintk("NFSD: end of grace period\n");
2657 nfsd4_recdir_purge_old();
2658 locks_end_grace(&nfsd4_manager);
2660 * Now that every NFSv4 client has had the chance to recover and
2661 * to see the (possibly new, possibly shorter) lease time, we
2662 * can safely set the next grace time to the current lease time:
2664 nfsd4_grace = nfsd4_lease;
2668 nfs4_laundromat(void)
2670 struct nfs4_client *clp;
2671 struct nfs4_stateowner *sop;
2672 struct nfs4_delegation *dp;
2673 struct list_head *pos, *next, reaplist;
2674 time_t cutoff = get_seconds() - nfsd4_lease;
2675 time_t t, clientid_val = nfsd4_lease;
2676 time_t u, test_val = nfsd4_lease;
2680 dprintk("NFSD: laundromat service - starting\n");
2681 if (locks_in_grace())
2683 INIT_LIST_HEAD(&reaplist);
2684 spin_lock(&client_lock);
2685 list_for_each_safe(pos, next, &client_lru) {
2686 clp = list_entry(pos, struct nfs4_client, cl_lru);
2687 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2688 t = clp->cl_time - cutoff;
2689 if (clientid_val > t)
2693 if (atomic_read(&clp->cl_refcount)) {
2694 dprintk("NFSD: client in use (clientid %08x)\n",
2695 clp->cl_clientid.cl_id);
2698 unhash_client_locked(clp);
2699 list_add(&clp->cl_lru, &reaplist);
2701 spin_unlock(&client_lock);
2702 list_for_each_safe(pos, next, &reaplist) {
2703 clp = list_entry(pos, struct nfs4_client, cl_lru);
2704 dprintk("NFSD: purging unused client (clientid %08x)\n",
2705 clp->cl_clientid.cl_id);
2706 nfsd4_remove_clid_dir(clp);
2709 spin_lock(&recall_lock);
2710 list_for_each_safe(pos, next, &del_recall_lru) {
2711 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2712 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2713 u = dp->dl_time - cutoff;
2718 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2720 list_move(&dp->dl_recall_lru, &reaplist);
2722 spin_unlock(&recall_lock);
2723 list_for_each_safe(pos, next, &reaplist) {
2724 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2725 list_del_init(&dp->dl_recall_lru);
2726 unhash_delegation(dp);
2728 test_val = nfsd4_lease;
2729 list_for_each_safe(pos, next, &close_lru) {
2730 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2731 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2732 u = sop->so_time - cutoff;
2737 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2739 release_openowner(sop);
2741 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2742 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2743 nfs4_unlock_state();
2744 return clientid_val;
2747 static struct workqueue_struct *laundry_wq;
2748 static void laundromat_main(struct work_struct *);
2749 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2752 laundromat_main(struct work_struct *not_used)
2756 t = nfs4_laundromat();
2757 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2758 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2761 static struct nfs4_stateowner *
2762 search_close_lru(u32 st_id, int flags)
2764 struct nfs4_stateowner *local = NULL;
2766 if (flags & CLOSE_STATE) {
2767 list_for_each_entry(local, &close_lru, so_close_lru) {
2768 if (local->so_id == st_id)
2776 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2778 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2782 STALE_STATEID(stateid_t *stateid)
2784 if (stateid->si_boot == boot_time)
2786 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2787 STATEID_VAL(stateid));
2792 access_permit_read(unsigned long access_bmap)
2794 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2795 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2796 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2800 access_permit_write(unsigned long access_bmap)
2802 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2803 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2807 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2809 __be32 status = nfserr_openmode;
2811 /* For lock stateid's, we test the parent open, not the lock: */
2812 if (stp->st_openstp)
2813 stp = stp->st_openstp;
2814 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2816 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2823 static inline __be32
2824 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2826 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2828 else if (locks_in_grace()) {
2829 /* Answer in remaining cases depends on existance of
2830 * conflicting state; so we must wait out the grace period. */
2831 return nfserr_grace;
2832 } else if (flags & WR_STATE)
2833 return nfs4_share_conflict(current_fh,
2834 NFS4_SHARE_DENY_WRITE);
2835 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2836 return nfs4_share_conflict(current_fh,
2837 NFS4_SHARE_DENY_READ);
2841 * Allow READ/WRITE during grace period on recovered state only for files
2842 * that are not able to provide mandatory locking.
2845 grace_disallows_io(struct inode *inode)
2847 return locks_in_grace() && mandatory_lock(inode);
2850 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2853 * When sessions are used the stateid generation number is ignored
2856 if ((flags & HAS_SESSION) && in->si_generation == 0)
2859 /* If the client sends us a stateid from the future, it's buggy: */
2860 if (in->si_generation > ref->si_generation)
2861 return nfserr_bad_stateid;
2863 * The following, however, can happen. For example, if the
2864 * client sends an open and some IO at the same time, the open
2865 * may bump si_generation while the IO is still in flight.
2866 * Thanks to hard links and renames, the client never knows what
2867 * file an open will affect. So it could avoid that situation
2868 * only by serializing all opens and IO from the same open
2869 * owner. To recover from the old_stateid error, the client
2870 * will just have to retry the IO:
2872 if (in->si_generation < ref->si_generation)
2873 return nfserr_old_stateid;
2878 static int is_delegation_stateid(stateid_t *stateid)
2880 return stateid->si_fileid == 0;
2884 * Checks for stateid operations
2887 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2888 stateid_t *stateid, int flags, struct file **filpp)
2890 struct nfs4_stateid *stp = NULL;
2891 struct nfs4_delegation *dp = NULL;
2892 struct svc_fh *current_fh = &cstate->current_fh;
2893 struct inode *ino = current_fh->fh_dentry->d_inode;
2899 if (grace_disallows_io(ino))
2900 return nfserr_grace;
2902 if (nfsd4_has_session(cstate))
2903 flags |= HAS_SESSION;
2905 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2906 return check_special_stateids(current_fh, stateid, flags);
2908 status = nfserr_stale_stateid;
2909 if (STALE_STATEID(stateid))
2912 status = nfserr_bad_stateid;
2913 if (is_delegation_stateid(stateid)) {
2914 dp = find_delegation_stateid(ino, stateid);
2917 status = check_stateid_generation(stateid, &dp->dl_stateid,
2921 status = nfs4_check_delegmode(dp, flags);
2924 renew_client(dp->dl_client);
2926 *filpp = find_readable_file(dp->dl_file);
2928 } else { /* open or lock stateid */
2929 stp = find_stateid(stateid, flags);
2932 if (nfs4_check_fh(current_fh, stp))
2934 if (!stp->st_stateowner->so_confirmed)
2936 status = check_stateid_generation(stateid, &stp->st_stateid,
2940 status = nfs4_check_openmode(stp, flags);
2943 renew_client(stp->st_stateowner->so_client);
2945 if (flags & RD_STATE)
2946 *filpp = find_readable_file(stp->st_file);
2948 *filpp = find_writeable_file(stp->st_file);
2959 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2960 RD_STATE : WR_STATE;
2964 * Checks for sequence id mutating operations.
2967 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2968 stateid_t *stateid, int flags,
2969 struct nfs4_stateowner **sopp,
2970 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2972 struct nfs4_stateid *stp;
2973 struct nfs4_stateowner *sop;
2974 struct svc_fh *current_fh = &cstate->current_fh;
2977 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2978 seqid, STATEID_VAL(stateid));
2983 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2984 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2985 return nfserr_bad_stateid;
2988 if (STALE_STATEID(stateid))
2989 return nfserr_stale_stateid;
2991 if (nfsd4_has_session(cstate))
2992 flags |= HAS_SESSION;
2995 * We return BAD_STATEID if filehandle doesn't match stateid,
2996 * the confirmed flag is incorrecly set, or the generation
2997 * number is incorrect.
2999 stp = find_stateid(stateid, flags);
3002 * Also, we should make sure this isn't just the result of
3005 sop = search_close_lru(stateid->si_stateownerid, flags);
3007 return nfserr_bad_stateid;
3013 *sopp = sop = stp->st_stateowner;
3016 clientid_t *lockclid = &lock->v.new.clientid;
3017 struct nfs4_client *clp = sop->so_client;
3021 lkflg = setlkflg(lock->lk_type);
3023 if (lock->lk_is_new) {
3024 if (!sop->so_is_open_owner)
3025 return nfserr_bad_stateid;
3026 if (!(flags & HAS_SESSION) &&
3027 !same_clid(&clp->cl_clientid, lockclid))
3028 return nfserr_bad_stateid;
3029 /* stp is the open stateid */
3030 status = nfs4_check_openmode(stp, lkflg);
3034 /* stp is the lock stateid */
3035 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3041 if (nfs4_check_fh(current_fh, stp)) {
3042 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3043 return nfserr_bad_stateid;
3047 * We now validate the seqid and stateid generation numbers.
3048 * For the moment, we ignore the possibility of
3049 * generation number wraparound.
3051 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3054 if (sop->so_confirmed && flags & CONFIRM) {
3055 dprintk("NFSD: preprocess_seqid_op: expected"
3056 " unconfirmed stateowner!\n");
3057 return nfserr_bad_stateid;
3059 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3060 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3061 " confirmed yet!\n");
3062 return nfserr_bad_stateid;
3064 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3067 renew_client(sop->so_client);
3071 if (seqid == sop->so_seqid - 1) {
3072 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3073 /* indicate replay to calling function */
3074 return nfserr_replay_me;
3076 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3077 sop->so_seqid, seqid);
3079 return nfserr_bad_seqid;
3083 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3084 struct nfsd4_open_confirm *oc)
3087 struct nfs4_stateowner *sop;
3088 struct nfs4_stateid *stp;
3090 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3091 (int)cstate->current_fh.fh_dentry->d_name.len,
3092 cstate->current_fh.fh_dentry->d_name.name);
3094 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3100 if ((status = nfs4_preprocess_seqid_op(cstate,
3101 oc->oc_seqid, &oc->oc_req_stateid,
3102 CONFIRM | OPEN_STATE,
3103 &oc->oc_stateowner, &stp, NULL)))
3106 sop = oc->oc_stateowner;
3107 sop->so_confirmed = 1;
3108 update_stateid(&stp->st_stateid);
3109 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3110 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3111 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3113 nfsd4_create_clid_dir(sop->so_client);
3115 if (oc->oc_stateowner) {
3116 nfs4_get_stateowner(oc->oc_stateowner);
3117 cstate->replay_owner = oc->oc_stateowner;
3119 nfs4_unlock_state();
3125 * unset all bits in union bitmap (bmap) that
3126 * do not exist in share (from successful OPEN_DOWNGRADE)
3129 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3132 for (i = 1; i < 4; i++) {
3133 if ((i & access) != i)
3134 __clear_bit(i, bmap);
3139 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3142 for (i = 0; i < 4; i++) {
3143 if ((i & deny) != i)
3144 __clear_bit(i, bmap);
3149 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3150 struct nfsd4_compound_state *cstate,
3151 struct nfsd4_open_downgrade *od)
3154 struct nfs4_stateid *stp;
3155 unsigned int share_access;
3157 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3158 (int)cstate->current_fh.fh_dentry->d_name.len,
3159 cstate->current_fh.fh_dentry->d_name.name);
3161 if (!access_valid(od->od_share_access, cstate->minorversion)
3162 || !deny_valid(od->od_share_deny))
3163 return nfserr_inval;
3166 if ((status = nfs4_preprocess_seqid_op(cstate,
3170 &od->od_stateowner, &stp, NULL)))
3173 status = nfserr_inval;
3174 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3175 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3176 stp->st_access_bmap, od->od_share_access);
3179 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3180 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3181 stp->st_deny_bmap, od->od_share_deny);
3184 set_access(&share_access, stp->st_access_bmap);
3185 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3187 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3188 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3190 update_stateid(&stp->st_stateid);
3191 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3194 if (od->od_stateowner) {
3195 nfs4_get_stateowner(od->od_stateowner);
3196 cstate->replay_owner = od->od_stateowner;
3198 nfs4_unlock_state();
3203 * nfs4_unlock_state() called after encode
3206 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3207 struct nfsd4_close *close)
3210 struct nfs4_stateid *stp;
3212 dprintk("NFSD: nfsd4_close on file %.*s\n",
3213 (int)cstate->current_fh.fh_dentry->d_name.len,
3214 cstate->current_fh.fh_dentry->d_name.name);
3217 /* check close_lru for replay */
3218 if ((status = nfs4_preprocess_seqid_op(cstate,
3221 OPEN_STATE | CLOSE_STATE,
3222 &close->cl_stateowner, &stp, NULL)))
3225 update_stateid(&stp->st_stateid);
3226 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3228 /* release_stateid() calls nfsd_close() if needed */
3229 release_open_stateid(stp);
3231 /* place unused nfs4_stateowners on so_close_lru list to be
3232 * released by the laundromat service after the lease period
3233 * to enable us to handle CLOSE replay
3235 if (list_empty(&close->cl_stateowner->so_stateids))
3236 move_to_close_lru(close->cl_stateowner);
3238 if (close->cl_stateowner) {
3239 nfs4_get_stateowner(close->cl_stateowner);
3240 cstate->replay_owner = close->cl_stateowner;
3242 nfs4_unlock_state();
3247 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3248 struct nfsd4_delegreturn *dr)
3250 struct nfs4_delegation *dp;
3251 stateid_t *stateid = &dr->dr_stateid;
3252 struct inode *inode;
3256 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3258 inode = cstate->current_fh.fh_dentry->d_inode;
3260 if (nfsd4_has_session(cstate))
3261 flags |= HAS_SESSION;
3263 status = nfserr_bad_stateid;
3264 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3266 status = nfserr_stale_stateid;
3267 if (STALE_STATEID(stateid))
3269 status = nfserr_bad_stateid;
3270 if (!is_delegation_stateid(stateid))
3272 dp = find_delegation_stateid(inode, stateid);
3275 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3278 renew_client(dp->dl_client);
3280 unhash_delegation(dp);
3282 nfs4_unlock_state();
3289 * Lock owner state (byte-range locks)
3291 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3292 #define LOCK_HASH_BITS 8
3293 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3294 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3297 end_offset(u64 start, u64 len)
3302 return end >= start ? end: NFS4_MAX_UINT64;
3305 /* last octet in a range */
3307 last_byte_offset(u64 start, u64 len)
3313 return end > start ? end - 1: NFS4_MAX_UINT64;
3316 #define lockownerid_hashval(id) \
3317 ((id) & LOCK_HASH_MASK)
3319 static inline unsigned int
3320 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3321 struct xdr_netobj *ownername)
3323 return (file_hashval(inode) + cl_id
3324 + opaque_hashval(ownername->data, ownername->len))
3328 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3329 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3330 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3332 static struct nfs4_stateid *
3333 find_stateid(stateid_t *stid, int flags)
3335 struct nfs4_stateid *local;
3336 u32 st_id = stid->si_stateownerid;
3337 u32 f_id = stid->si_fileid;
3338 unsigned int hashval;
3340 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3341 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3342 hashval = stateid_hashval(st_id, f_id);
3343 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3344 if ((local->st_stateid.si_stateownerid == st_id) &&
3345 (local->st_stateid.si_fileid == f_id))
3350 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3351 hashval = stateid_hashval(st_id, f_id);
3352 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3353 if ((local->st_stateid.si_stateownerid == st_id) &&
3354 (local->st_stateid.si_fileid == f_id))
3361 static struct nfs4_delegation *
3362 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3364 struct nfs4_file *fp;
3365 struct nfs4_delegation *dl;
3367 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3370 fp = find_file(ino);
3373 dl = find_delegation_file(fp, stid);
3379 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3380 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3381 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3382 * locking, this prevents us from being completely protocol-compliant. The
3383 * real solution to this problem is to start using unsigned file offsets in
3384 * the VFS, but this is a very deep change!
3387 nfs4_transform_lock_offset(struct file_lock *lock)
3389 if (lock->fl_start < 0)
3390 lock->fl_start = OFFSET_MAX;
3391 if (lock->fl_end < 0)
3392 lock->fl_end = OFFSET_MAX;
3395 /* Hack!: For now, we're defining this just so we can use a pointer to it
3396 * as a unique cookie to identify our (NFSv4's) posix locks. */
3397 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3401 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3403 struct nfs4_stateowner *sop;
3405 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3406 sop = (struct nfs4_stateowner *) fl->fl_owner;
3407 kref_get(&sop->so_ref);
3409 deny->ld_clientid = sop->so_client->cl_clientid;
3411 deny->ld_sop = NULL;
3412 deny->ld_clientid.cl_boot = 0;
3413 deny->ld_clientid.cl_id = 0;
3415 deny->ld_start = fl->fl_start;
3416 deny->ld_length = NFS4_MAX_UINT64;
3417 if (fl->fl_end != NFS4_MAX_UINT64)
3418 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3419 deny->ld_type = NFS4_READ_LT;
3420 if (fl->fl_type != F_RDLCK)
3421 deny->ld_type = NFS4_WRITE_LT;
3424 static struct nfs4_stateowner *
3425 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3426 struct xdr_netobj *owner)
3428 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3429 struct nfs4_stateowner *op;
3431 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3432 if (same_owner_str(op, owner, clid))
3439 * Alloc a lock owner structure.
3440 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3443 * strhashval = lock_ownerstr_hashval
3446 static struct nfs4_stateowner *
3447 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3448 struct nfs4_stateowner *sop;
3449 struct nfs4_replay *rp;
3450 unsigned int idhashval;
3452 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3454 idhashval = lockownerid_hashval(current_ownerid);
3455 INIT_LIST_HEAD(&sop->so_idhash);
3456 INIT_LIST_HEAD(&sop->so_strhash);
3457 INIT_LIST_HEAD(&sop->so_perclient);
3458 INIT_LIST_HEAD(&sop->so_stateids);
3459 INIT_LIST_HEAD(&sop->so_perstateid);
3460 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3462 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3463 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3464 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3465 sop->so_is_open_owner = 0;
3466 sop->so_id = current_ownerid++;
3467 sop->so_client = clp;
3468 /* It is the openowner seqid that will be incremented in encode in the
3469 * case of new lockowners; so increment the lock seqid manually: */
3470 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3471 sop->so_confirmed = 1;
3472 rp = &sop->so_replay;
3473 rp->rp_status = nfserr_serverfault;
3475 rp->rp_buf = rp->rp_ibuf;
3479 static struct nfs4_stateid *
3480 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3482 struct nfs4_stateid *stp;
3483 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3485 stp = nfs4_alloc_stateid();
3488 INIT_LIST_HEAD(&stp->st_hash);
3489 INIT_LIST_HEAD(&stp->st_perfile);
3490 INIT_LIST_HEAD(&stp->st_perstateowner);
3491 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3492 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3493 list_add(&stp->st_perfile, &fp->fi_stateids);
3494 list_add(&stp->st_perstateowner, &sop->so_stateids);
3495 stp->st_stateowner = sop;
3498 stp->st_stateid.si_boot = boot_time;
3499 stp->st_stateid.si_stateownerid = sop->so_id;
3500 stp->st_stateid.si_fileid = fp->fi_id;
3501 stp->st_stateid.si_generation = 0;
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;
3526 struct nfs4_file *fp;
3527 struct file *filp = NULL;
3528 struct file_lock file_lock;
3529 struct file_lock conflock;
3531 unsigned int strhashval;
3535 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3536 (long long) lock->lk_offset,
3537 (long long) lock->lk_length);
3539 if (check_lock_length(lock->lk_offset, lock->lk_length))
3540 return nfserr_inval;
3542 if ((status = fh_verify(rqstp, &cstate->current_fh,
3543 S_IFREG, NFSD_MAY_LOCK))) {
3544 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3550 if (lock->lk_is_new) {
3552 * Client indicates that this is a new lockowner.
3553 * Use open owner and open stateid to create lock owner and
3556 struct nfs4_stateid *open_stp = NULL;
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;
3599 fp = lock_stp->st_file;
3601 /* lock->lk_replay_owner and lock_stp have been created or found */
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 if (find_readable_file(lock_stp->st_file)) {
3615 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3616 filp = find_readable_file(lock_stp->st_file);
3618 file_lock.fl_type = F_RDLCK;
3622 case NFS4_WRITEW_LT:
3623 if (find_writeable_file(lock_stp->st_file)) {
3624 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3625 filp = find_writeable_file(lock_stp->st_file);
3627 file_lock.fl_type = F_WRLCK;
3631 status = nfserr_inval;
3635 status = nfserr_openmode;
3638 file_lock.fl_owner = (fl_owner_t)lock_sop;
3639 file_lock.fl_pid = current->tgid;
3640 file_lock.fl_file = filp;
3641 file_lock.fl_flags = FL_POSIX;
3642 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3644 file_lock.fl_start = lock->lk_offset;
3645 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3646 nfs4_transform_lock_offset(&file_lock);
3649 * Try to lock the file in the VFS.
3650 * Note: locks.c uses the BKL to protect the inode's lock list.
3653 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3655 case 0: /* success! */
3656 update_stateid(&lock_stp->st_stateid);
3657 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3661 case (EAGAIN): /* conflock holds conflicting lock */
3662 status = nfserr_denied;
3663 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3664 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3667 status = nfserr_deadlock;
3670 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3671 status = nfserr_resource;
3675 if (status && lock->lk_is_new && lock_sop)
3676 release_lockowner(lock_sop);
3677 if (lock->lk_replay_owner) {
3678 nfs4_get_stateowner(lock->lk_replay_owner);
3679 cstate->replay_owner = lock->lk_replay_owner;
3681 nfs4_unlock_state();
3686 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3687 * so we do a temporary open here just to get an open file to pass to
3688 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3691 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3696 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3699 err = vfs_test_lock(file, lock);
3708 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3709 struct nfsd4_lockt *lockt)
3711 struct inode *inode;
3712 struct file_lock file_lock;
3716 if (locks_in_grace())
3717 return nfserr_grace;
3719 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3720 return nfserr_inval;
3722 lockt->lt_stateowner = NULL;
3725 status = nfserr_stale_clientid;
3726 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3729 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3730 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3731 if (status == nfserr_symlink)
3732 status = nfserr_inval;
3736 inode = cstate->current_fh.fh_dentry->d_inode;
3737 locks_init_lock(&file_lock);
3738 switch (lockt->lt_type) {
3741 file_lock.fl_type = F_RDLCK;
3744 case NFS4_WRITEW_LT:
3745 file_lock.fl_type = F_WRLCK;
3748 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3749 status = nfserr_inval;
3753 lockt->lt_stateowner = find_lockstateowner_str(inode,
3754 &lockt->lt_clientid, &lockt->lt_owner);
3755 if (lockt->lt_stateowner)
3756 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3757 file_lock.fl_pid = current->tgid;
3758 file_lock.fl_flags = FL_POSIX;
3760 file_lock.fl_start = lockt->lt_offset;
3761 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3763 nfs4_transform_lock_offset(&file_lock);
3766 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3768 status = nfserrno(error);
3771 if (file_lock.fl_type != F_UNLCK) {
3772 status = nfserr_denied;
3773 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3776 nfs4_unlock_state();
3781 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3782 struct nfsd4_locku *locku)
3784 struct nfs4_stateid *stp;
3785 struct file *filp = NULL;
3786 struct file_lock file_lock;
3790 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3791 (long long) locku->lu_offset,
3792 (long long) locku->lu_length);
3794 if (check_lock_length(locku->lu_offset, locku->lu_length))
3795 return nfserr_inval;
3799 if ((status = nfs4_preprocess_seqid_op(cstate,
3803 &locku->lu_stateowner, &stp, NULL)))
3806 filp = find_any_file(stp->st_file);
3808 status = nfserr_lock_range;
3812 locks_init_lock(&file_lock);
3813 file_lock.fl_type = F_UNLCK;
3814 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3815 file_lock.fl_pid = current->tgid;
3816 file_lock.fl_file = filp;
3817 file_lock.fl_flags = FL_POSIX;
3818 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3819 file_lock.fl_start = locku->lu_offset;
3821 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3822 nfs4_transform_lock_offset(&file_lock);
3825 * Try to unlock the file in the VFS.
3827 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3829 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3833 * OK, unlock succeeded; the only thing left to do is update the stateid.
3835 update_stateid(&stp->st_stateid);
3836 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3839 if (locku->lu_stateowner) {
3840 nfs4_get_stateowner(locku->lu_stateowner);
3841 cstate->replay_owner = locku->lu_stateowner;
3843 nfs4_unlock_state();
3847 status = nfserrno(err);
3853 * 1: locks held by lockowner
3854 * 0: no locks held by lockowner
3857 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
3859 struct file_lock **flpp;
3860 struct inode *inode = filp->fi_inode;
3864 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3865 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3876 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3877 struct nfsd4_compound_state *cstate,
3878 struct nfsd4_release_lockowner *rlockowner)
3880 clientid_t *clid = &rlockowner->rl_clientid;
3881 struct nfs4_stateowner *sop;
3882 struct nfs4_stateid *stp;
3883 struct xdr_netobj *owner = &rlockowner->rl_owner;
3884 struct list_head matches;
3888 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3889 clid->cl_boot, clid->cl_id);
3891 /* XXX check for lease expiration */
3893 status = nfserr_stale_clientid;
3894 if (STALE_CLIENTID(clid))
3899 status = nfserr_locks_held;
3900 /* XXX: we're doing a linear search through all the lockowners.
3901 * Yipes! For now we'll just hope clients aren't really using
3902 * release_lockowner much, but eventually we have to fix these
3903 * data structures. */
3904 INIT_LIST_HEAD(&matches);
3905 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3906 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3907 if (!same_owner_str(sop, owner, clid))
3909 list_for_each_entry(stp, &sop->so_stateids,
3911 if (check_for_locks(stp->st_file, sop))
3913 /* Note: so_perclient unused for lockowners,
3914 * so it's OK to fool with here. */
3915 list_add(&sop->so_perclient, &matches);
3919 /* Clients probably won't expect us to return with some (but not all)
3920 * of the lockowner state released; so don't release any until all
3921 * have been checked. */
3923 while (!list_empty(&matches)) {
3924 sop = list_entry(matches.next, struct nfs4_stateowner,
3926 /* unhash_stateowner deletes so_perclient only
3927 * for openowners. */
3928 list_del(&sop->so_perclient);
3929 release_lockowner(sop);
3932 nfs4_unlock_state();
3936 static inline struct nfs4_client_reclaim *
3939 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3943 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3945 unsigned int strhashval = clientstr_hashval(name);
3946 struct nfs4_client *clp;
3948 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3953 * failure => all reset bets are off, nfserr_no_grace...
3956 nfs4_client_to_reclaim(const char *name)
3958 unsigned int strhashval;
3959 struct nfs4_client_reclaim *crp = NULL;
3961 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3962 crp = alloc_reclaim();
3965 strhashval = clientstr_hashval(name);
3966 INIT_LIST_HEAD(&crp->cr_strhash);
3967 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3968 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3969 reclaim_str_hashtbl_size++;
3974 nfs4_release_reclaim(void)
3976 struct nfs4_client_reclaim *crp = NULL;
3979 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3980 while (!list_empty(&reclaim_str_hashtbl[i])) {
3981 crp = list_entry(reclaim_str_hashtbl[i].next,
3982 struct nfs4_client_reclaim, cr_strhash);
3983 list_del(&crp->cr_strhash);
3985 reclaim_str_hashtbl_size--;
3988 BUG_ON(reclaim_str_hashtbl_size);
3992 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3993 static struct nfs4_client_reclaim *
3994 nfs4_find_reclaim_client(clientid_t *clid)
3996 unsigned int strhashval;
3997 struct nfs4_client *clp;
3998 struct nfs4_client_reclaim *crp = NULL;
4001 /* find clientid in conf_id_hashtbl */
4002 clp = find_confirmed_client(clid);
4006 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4007 clp->cl_name.len, clp->cl_name.data,
4010 /* find clp->cl_name in reclaim_str_hashtbl */
4011 strhashval = clientstr_hashval(clp->cl_recdir);
4012 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4013 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4021 * Called from OPEN. Look for clientid in reclaim list.
4024 nfs4_check_open_reclaim(clientid_t *clid)
4026 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4029 /* initialization to perform at module load time: */
4032 nfs4_state_init(void)
4036 status = nfsd4_init_slabs();
4039 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4040 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4041 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4042 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4043 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4044 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4046 for (i = 0; i < SESSION_HASH_SIZE; i++)
4047 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4048 for (i = 0; i < FILE_HASH_SIZE; i++) {
4049 INIT_LIST_HEAD(&file_hashtbl[i]);
4051 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4052 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4053 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4055 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4056 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4057 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4059 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4060 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4061 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4063 memset(&onestateid, ~0, sizeof(stateid_t));
4064 INIT_LIST_HEAD(&close_lru);
4065 INIT_LIST_HEAD(&client_lru);
4066 INIT_LIST_HEAD(&del_recall_lru);
4067 reclaim_str_hashtbl_size = 0;
4072 nfsd4_load_reboot_recovery_data(void)
4077 nfsd4_init_recdir(user_recovery_dirname);
4078 status = nfsd4_recdir_load();
4079 nfs4_unlock_state();
4081 printk("NFSD: Failure reading reboot recovery data\n");
4085 * Since the lifetime of a delegation isn't limited to that of an open, a
4086 * client may quite reasonably hang on to a delegation as long as it has
4087 * the inode cached. This becomes an obvious problem the first time a
4088 * client's inode cache approaches the size of the server's total memory.
4090 * For now we avoid this problem by imposing a hard limit on the number
4091 * of delegations, which varies according to the server's memory size.
4094 set_max_delegations(void)
4097 * Allow at most 4 delegations per megabyte of RAM. Quick
4098 * estimates suggest that in the worst case (where every delegation
4099 * is for a different inode), a delegation could take about 1.5K,
4100 * giving a worst case usage of about 6% of memory.
4102 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4105 /* initialization to perform when the nfsd service is started: */
4108 __nfs4_state_start(void)
4112 boot_time = get_seconds();
4113 locks_start_grace(&nfsd4_manager);
4114 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4116 ret = set_callback_cred();
4119 laundry_wq = create_singlethread_workqueue("nfsd4");
4120 if (laundry_wq == NULL)
4122 ret = nfsd4_create_callback_queue();
4124 goto out_free_laundry;
4125 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4126 set_max_delegations();
4129 destroy_workqueue(laundry_wq);
4134 nfs4_state_start(void)
4136 nfsd4_load_reboot_recovery_data();
4137 return __nfs4_state_start();
4141 __nfs4_state_shutdown(void)
4144 struct nfs4_client *clp = NULL;
4145 struct nfs4_delegation *dp = NULL;
4146 struct list_head *pos, *next, reaplist;
4148 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4149 while (!list_empty(&conf_id_hashtbl[i])) {
4150 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4153 while (!list_empty(&unconf_str_hashtbl[i])) {
4154 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4158 INIT_LIST_HEAD(&reaplist);
4159 spin_lock(&recall_lock);
4160 list_for_each_safe(pos, next, &del_recall_lru) {
4161 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4162 list_move(&dp->dl_recall_lru, &reaplist);
4164 spin_unlock(&recall_lock);
4165 list_for_each_safe(pos, next, &reaplist) {
4166 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4167 list_del_init(&dp->dl_recall_lru);
4168 unhash_delegation(dp);
4171 nfsd4_shutdown_recdir();
4175 nfs4_state_shutdown(void)
4177 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4178 destroy_workqueue(laundry_wq);
4179 locks_end_grace(&nfsd4_manager);
4181 nfs4_release_reclaim();
4182 __nfs4_state_shutdown();
4183 nfs4_unlock_state();
4184 nfsd4_destroy_callback_queue();
4188 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4189 * accessed when nfsd is starting.
4192 nfs4_set_recdir(char *recdir)
4194 strcpy(user_recovery_dirname, recdir);
4198 * Change the NFSv4 recovery directory to recdir.
4201 nfs4_reset_recoverydir(char *recdir)
4206 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4210 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4211 nfs4_set_recdir(recdir);
4219 nfs4_recoverydir(void)
4221 return user_recovery_dirname;