2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58 #include <linux/sunrpc/clnt.h>
60 #define NFSDDBG_FACILITY NFSDDBG_PROC
63 static time_t lease_time = 90; /* default lease time */
64 static time_t user_lease_time = 90;
65 static time_t boot_time;
66 static u32 current_ownerid = 1;
67 static u32 current_fileid = 1;
68 static u32 current_delegid = 1;
70 static stateid_t zerostateid; /* bits all 0 */
71 static stateid_t onestateid; /* bits all 1 */
72 static u64 current_sessionid = 1;
74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
75 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
77 /* forward declarations */
78 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
79 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
80 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
81 static void nfs4_set_recdir(char *recdir);
85 /* Currently used for almost all code touching nfsv4 state: */
86 static DEFINE_MUTEX(client_mutex);
89 * Currently used for the del_recall_lru and file hash table. In an
90 * effort to decrease the scope of the client_mutex, this spinlock may
91 * eventually cover more:
93 static DEFINE_SPINLOCK(recall_lock);
95 static struct kmem_cache *stateowner_slab = NULL;
96 static struct kmem_cache *file_slab = NULL;
97 static struct kmem_cache *stateid_slab = NULL;
98 static struct kmem_cache *deleg_slab = NULL;
101 nfs4_lock_state(void)
103 mutex_lock(&client_mutex);
107 nfs4_unlock_state(void)
109 mutex_unlock(&client_mutex);
113 opaque_hashval(const void *ptr, int nbytes)
115 unsigned char *cptr = (unsigned char *) ptr;
125 static struct list_head del_recall_lru;
128 put_nfs4_file(struct nfs4_file *fi)
130 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
131 list_del(&fi->fi_hash);
132 spin_unlock(&recall_lock);
134 kmem_cache_free(file_slab, fi);
139 get_nfs4_file(struct nfs4_file *fi)
141 atomic_inc(&fi->fi_ref);
144 static int num_delegations;
145 unsigned int max_delegations;
148 * Open owner state (share locks)
151 /* hash tables for nfs4_stateowner */
152 #define OWNER_HASH_BITS 8
153 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
154 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
156 #define ownerid_hashval(id) \
157 ((id) & OWNER_HASH_MASK)
158 #define ownerstr_hashval(clientid, ownername) \
159 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
161 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
162 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
164 /* hash table for nfs4_file */
165 #define FILE_HASH_BITS 8
166 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
167 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
168 /* hash table for (open)nfs4_stateid */
169 #define STATEID_HASH_BITS 10
170 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
171 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
173 #define file_hashval(x) \
174 hash_ptr(x, FILE_HASH_BITS)
175 #define stateid_hashval(owner_id, file_id) \
176 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
178 static struct list_head file_hashtbl[FILE_HASH_SIZE];
179 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
181 static struct nfs4_delegation *
182 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
184 struct nfs4_delegation *dp;
185 struct nfs4_file *fp = stp->st_file;
186 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
188 dprintk("NFSD alloc_init_deleg\n");
189 if (fp->fi_had_conflict)
191 if (num_delegations > max_delegations)
193 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
197 INIT_LIST_HEAD(&dp->dl_perfile);
198 INIT_LIST_HEAD(&dp->dl_perclnt);
199 INIT_LIST_HEAD(&dp->dl_recall_lru);
204 get_file(stp->st_vfs_file);
205 dp->dl_vfs_file = stp->st_vfs_file;
207 dp->dl_ident = cb->cb_ident;
208 dp->dl_stateid.si_boot = get_seconds();
209 dp->dl_stateid.si_stateownerid = current_delegid++;
210 dp->dl_stateid.si_fileid = 0;
211 dp->dl_stateid.si_generation = 0;
212 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
214 atomic_set(&dp->dl_count, 1);
215 list_add(&dp->dl_perfile, &fp->fi_delegations);
216 list_add(&dp->dl_perclnt, &clp->cl_delegations);
221 nfs4_put_delegation(struct nfs4_delegation *dp)
223 if (atomic_dec_and_test(&dp->dl_count)) {
224 dprintk("NFSD: freeing dp %p\n",dp);
225 put_nfs4_file(dp->dl_file);
226 kmem_cache_free(deleg_slab, dp);
231 /* Remove the associated file_lock first, then remove the delegation.
232 * lease_modify() is called to remove the FS_LEASE file_lock from
233 * the i_flock list, eventually calling nfsd's lock_manager
234 * fl_release_callback.
237 nfs4_close_delegation(struct nfs4_delegation *dp)
239 struct file *filp = dp->dl_vfs_file;
241 dprintk("NFSD: close_delegation dp %p\n",dp);
242 dp->dl_vfs_file = NULL;
243 /* The following nfsd_close may not actually close the file,
244 * but we want to remove the lease in any case. */
246 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
250 /* Called under the state lock. */
252 unhash_delegation(struct nfs4_delegation *dp)
254 list_del_init(&dp->dl_perfile);
255 list_del_init(&dp->dl_perclnt);
256 spin_lock(&recall_lock);
257 list_del_init(&dp->dl_recall_lru);
258 spin_unlock(&recall_lock);
259 nfs4_close_delegation(dp);
260 nfs4_put_delegation(dp);
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS 4
269 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
272 #define clientid_hashval(id) \
273 ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
277 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278 * used in reboot/reset lease grace period processing
280 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281 * setclientid_confirmed info.
283 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
286 * client_lru holds client queue ordered by nfs4_client.cl_time
289 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290 * for last close replay.
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
303 list_del(&stp->st_hash);
304 list_del(&stp->st_perfile);
305 list_del(&stp->st_perstateowner);
308 static void free_generic_stateid(struct nfs4_stateid *stp)
310 put_nfs4_file(stp->st_file);
311 kmem_cache_free(stateid_slab, stp);
314 static void release_lock_stateid(struct nfs4_stateid *stp)
316 unhash_generic_stateid(stp);
317 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318 free_generic_stateid(stp);
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
323 struct nfs4_stateid *stp;
325 list_del(&sop->so_idhash);
326 list_del(&sop->so_strhash);
327 list_del(&sop->so_perstateid);
328 while (!list_empty(&sop->so_stateids)) {
329 stp = list_first_entry(&sop->so_stateids,
330 struct nfs4_stateid, st_perstateowner);
331 release_lock_stateid(stp);
335 static void release_lockowner(struct nfs4_stateowner *sop)
337 unhash_lockowner(sop);
338 nfs4_put_stateowner(sop);
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
344 struct nfs4_stateowner *lock_sop;
346 while (!list_empty(&open_stp->st_lockowners)) {
347 lock_sop = list_entry(open_stp->st_lockowners.next,
348 struct nfs4_stateowner, so_perstateid);
349 /* list_del(&open_stp->st_lockowners); */
350 BUG_ON(lock_sop->so_is_open_owner);
351 release_lockowner(lock_sop);
355 static void release_open_stateid(struct nfs4_stateid *stp)
357 unhash_generic_stateid(stp);
358 release_stateid_lockowners(stp);
359 nfsd_close(stp->st_vfs_file);
360 free_generic_stateid(stp);
363 static void unhash_openowner(struct nfs4_stateowner *sop)
365 struct nfs4_stateid *stp;
367 list_del(&sop->so_idhash);
368 list_del(&sop->so_strhash);
369 list_del(&sop->so_perclient);
370 list_del(&sop->so_perstateid); /* XXX: necessary? */
371 while (!list_empty(&sop->so_stateids)) {
372 stp = list_first_entry(&sop->so_stateids,
373 struct nfs4_stateid, st_perstateowner);
374 release_open_stateid(stp);
378 static void release_openowner(struct nfs4_stateowner *sop)
380 unhash_openowner(sop);
381 list_del(&sop->so_close_lru);
382 nfs4_put_stateowner(sop);
385 static DEFINE_SPINLOCK(sessionid_lock);
386 #define SESSION_HASH_SIZE 512
387 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
390 hash_sessionid(struct nfs4_sessionid *sessionid)
392 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
394 return sid->sequence % SESSION_HASH_SIZE;
398 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
400 u32 *ptr = (u32 *)(&sessionid->data[0]);
401 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
405 gen_sessionid(struct nfsd4_session *ses)
407 struct nfs4_client *clp = ses->se_client;
408 struct nfsd4_sessionid *sid;
410 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
411 sid->clientid = clp->cl_clientid;
412 sid->sequence = current_sessionid++;
417 * The protocol defines ca_maxresponssize_cached to include the size of
418 * the rpc header, but all we need to cache is the data starting after
419 * the end of the initial SEQUENCE operation--the rest we regenerate
420 * each time. Therefore we can advertise a ca_maxresponssize_cached
421 * value that is the number of bytes in our cache plus a few additional
422 * bytes. In order to stay on the safe side, and not promise more than
423 * we can cache, those additional bytes must be the minimum possible: 24
424 * bytes of rpc header (xid through accept state, with AUTH_NULL
425 * verifier), 12 for the compound header (with zero-length tag), and 44
426 * for the SEQUENCE op response:
428 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
431 * Give the client the number of ca_maxresponsesize_cached slots it
432 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
433 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
434 * than NFSD_MAX_SLOTS_PER_SESSION.
436 * If we run out of reserved DRC memory we should (up to a point)
437 * re-negotiate active sessions and reduce their slot usage to make
438 * rooom for new connections. For now we just fail the create session.
440 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
442 int mem, size = fchan->maxresp_cached;
444 if (fchan->maxreqs < 1)
447 if (size < NFSD_MIN_HDR_SEQ_SZ)
448 size = NFSD_MIN_HDR_SEQ_SZ;
449 size -= NFSD_MIN_HDR_SEQ_SZ;
450 if (size > NFSD_SLOT_CACHE_SIZE)
451 size = NFSD_SLOT_CACHE_SIZE;
453 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
454 mem = fchan->maxreqs * size;
455 if (mem > NFSD_MAX_MEM_PER_SESSION) {
456 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
457 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
458 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
459 mem = fchan->maxreqs * size;
462 spin_lock(&nfsd_drc_lock);
463 /* bound the total session drc memory ussage */
464 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
465 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
466 mem = fchan->maxreqs * size;
468 nfsd_drc_mem_used += mem;
469 spin_unlock(&nfsd_drc_lock);
471 if (fchan->maxreqs == 0)
472 return nfserr_resource;
474 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
479 * fchan holds the client values on input, and the server values on output
481 static int init_forechannel_attrs(struct svc_rqst *rqstp,
482 struct nfsd4_channel_attrs *session_fchan,
483 struct nfsd4_channel_attrs *fchan)
486 __u32 maxcount = svc_max_payload(rqstp);
488 /* headerpadsz set to zero in encode routine */
490 /* Use the client's max request and max response size if possible */
491 if (fchan->maxreq_sz > maxcount)
492 fchan->maxreq_sz = maxcount;
493 session_fchan->maxreq_sz = fchan->maxreq_sz;
495 if (fchan->maxresp_sz > maxcount)
496 fchan->maxresp_sz = maxcount;
497 session_fchan->maxresp_sz = fchan->maxresp_sz;
499 /* Use the client's maxops if possible */
500 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
501 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
502 session_fchan->maxops = fchan->maxops;
504 /* FIXME: Error means no more DRC pages so the server should
505 * recover pages from existing sessions. For now fail session
508 status = set_forechannel_drc_size(fchan);
510 session_fchan->maxresp_cached = fchan->maxresp_cached;
511 session_fchan->maxreqs = fchan->maxreqs;
513 dprintk("%s status %d\n", __func__, status);
518 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
519 struct nfsd4_create_session *cses)
521 struct nfsd4_session *new, tmp;
522 int idx, status = nfserr_resource, slotsize;
524 memset(&tmp, 0, sizeof(tmp));
526 /* FIXME: For now, we just accept the client back channel attributes. */
527 tmp.se_bchannel = cses->back_channel;
528 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
529 &cses->fore_channel);
533 /* allocate struct nfsd4_session and slot table in one piece */
534 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot);
535 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
539 memcpy(new, &tmp, sizeof(*new));
541 new->se_client = clp;
543 idx = hash_sessionid(&new->se_sessionid);
544 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
545 NFS4_MAX_SESSIONID_LEN);
547 new->se_flags = cses->flags;
548 kref_init(&new->se_ref);
549 spin_lock(&sessionid_lock);
550 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
551 list_add(&new->se_perclnt, &clp->cl_sessions);
552 spin_unlock(&sessionid_lock);
559 /* caller must hold sessionid_lock */
560 static struct nfsd4_session *
561 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
563 struct nfsd4_session *elem;
566 dump_sessionid(__func__, sessionid);
567 idx = hash_sessionid(sessionid);
568 dprintk("%s: idx is %d\n", __func__, idx);
569 /* Search in the appropriate list */
570 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
571 dump_sessionid("list traversal", &elem->se_sessionid);
572 if (!memcmp(elem->se_sessionid.data, sessionid->data,
573 NFS4_MAX_SESSIONID_LEN)) {
578 dprintk("%s: session not found\n", __func__);
582 /* caller must hold sessionid_lock */
584 unhash_session(struct nfsd4_session *ses)
586 list_del(&ses->se_hash);
587 list_del(&ses->se_perclnt);
591 release_session(struct nfsd4_session *ses)
593 spin_lock(&sessionid_lock);
595 spin_unlock(&sessionid_lock);
596 nfsd4_put_session(ses);
599 static void nfsd4_release_respages(struct page **respages, short resused);
602 free_session(struct kref *kref)
604 struct nfsd4_session *ses;
607 ses = container_of(kref, struct nfsd4_session, se_ref);
608 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
609 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
610 nfsd4_release_respages(e->ce_respages, e->ce_resused);
612 spin_lock(&nfsd_drc_lock);
613 nfsd_drc_mem_used -= ses->se_fchannel.maxreqs * NFSD_SLOT_CACHE_SIZE;
614 spin_unlock(&nfsd_drc_lock);
619 renew_client(struct nfs4_client *clp)
622 * Move client to the end to the LRU list.
624 dprintk("renewing client (clientid %08x/%08x)\n",
625 clp->cl_clientid.cl_boot,
626 clp->cl_clientid.cl_id);
627 list_move_tail(&clp->cl_lru, &client_lru);
628 clp->cl_time = get_seconds();
631 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
633 STALE_CLIENTID(clientid_t *clid)
635 if (clid->cl_boot == boot_time)
637 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
638 clid->cl_boot, clid->cl_id, boot_time);
643 * XXX Should we use a slab cache ?
644 * This type of memory management is somewhat inefficient, but we use it
645 * anyway since SETCLIENTID is not a common operation.
647 static struct nfs4_client *alloc_client(struct xdr_netobj name)
649 struct nfs4_client *clp;
651 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
654 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
655 if (clp->cl_name.data == NULL) {
659 memcpy(clp->cl_name.data, name.data, name.len);
660 clp->cl_name.len = name.len;
665 shutdown_callback_client(struct nfs4_client *clp)
667 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
671 * Callback threads take a reference on the client, so there
672 * should be no outstanding callbacks at this point.
674 clp->cl_cb_conn.cb_client = NULL;
675 rpc_shutdown_client(clnt);
677 if (clp->cl_cb_conn.cb_cred) {
678 put_rpccred(clp->cl_cb_conn.cb_cred);
679 clp->cl_cb_conn.cb_cred = NULL;
684 free_client(struct nfs4_client *clp)
686 shutdown_callback_client(clp);
687 if (clp->cl_cred.cr_group_info)
688 put_group_info(clp->cl_cred.cr_group_info);
689 kfree(clp->cl_principal);
690 kfree(clp->cl_name.data);
695 put_nfs4_client(struct nfs4_client *clp)
697 if (atomic_dec_and_test(&clp->cl_count))
702 expire_client(struct nfs4_client *clp)
704 struct nfs4_stateowner *sop;
705 struct nfs4_delegation *dp;
706 struct list_head reaplist;
708 dprintk("NFSD: expire_client cl_count %d\n",
709 atomic_read(&clp->cl_count));
711 INIT_LIST_HEAD(&reaplist);
712 spin_lock(&recall_lock);
713 while (!list_empty(&clp->cl_delegations)) {
714 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
715 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
717 list_del_init(&dp->dl_perclnt);
718 list_move(&dp->dl_recall_lru, &reaplist);
720 spin_unlock(&recall_lock);
721 while (!list_empty(&reaplist)) {
722 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
723 list_del_init(&dp->dl_recall_lru);
724 unhash_delegation(dp);
726 list_del(&clp->cl_idhash);
727 list_del(&clp->cl_strhash);
728 list_del(&clp->cl_lru);
729 while (!list_empty(&clp->cl_openowners)) {
730 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
731 release_openowner(sop);
733 while (!list_empty(&clp->cl_sessions)) {
734 struct nfsd4_session *ses;
735 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
737 release_session(ses);
739 put_nfs4_client(clp);
742 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
744 struct nfs4_client *clp;
746 clp = alloc_client(name);
749 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
750 atomic_set(&clp->cl_count, 1);
751 atomic_set(&clp->cl_cb_conn.cb_set, 0);
752 INIT_LIST_HEAD(&clp->cl_idhash);
753 INIT_LIST_HEAD(&clp->cl_strhash);
754 INIT_LIST_HEAD(&clp->cl_openowners);
755 INIT_LIST_HEAD(&clp->cl_delegations);
756 INIT_LIST_HEAD(&clp->cl_sessions);
757 INIT_LIST_HEAD(&clp->cl_lru);
761 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
763 memcpy(target->cl_verifier.data, source->data,
764 sizeof(target->cl_verifier.data));
767 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
769 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
770 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
773 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
775 target->cr_uid = source->cr_uid;
776 target->cr_gid = source->cr_gid;
777 target->cr_group_info = source->cr_group_info;
778 get_group_info(target->cr_group_info);
781 static int same_name(const char *n1, const char *n2)
783 return 0 == memcmp(n1, n2, HEXDIR_LEN);
787 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
789 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
793 same_clid(clientid_t *cl1, clientid_t *cl2)
795 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
798 /* XXX what about NGROUP */
800 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
802 return cr1->cr_uid == cr2->cr_uid;
805 static void gen_clid(struct nfs4_client *clp)
807 static u32 current_clientid = 1;
809 clp->cl_clientid.cl_boot = boot_time;
810 clp->cl_clientid.cl_id = current_clientid++;
813 static void gen_confirm(struct nfs4_client *clp)
818 p = (u32 *)clp->cl_confirm.data;
819 *p++ = get_seconds();
823 static int check_name(struct xdr_netobj name)
827 if (name.len > NFS4_OPAQUE_LIMIT) {
828 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
835 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
837 unsigned int idhashval;
839 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
840 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
841 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
842 list_add_tail(&clp->cl_lru, &client_lru);
843 clp->cl_time = get_seconds();
847 move_to_confirmed(struct nfs4_client *clp)
849 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
850 unsigned int strhashval;
852 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
853 list_del_init(&clp->cl_strhash);
854 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
855 strhashval = clientstr_hashval(clp->cl_recdir);
856 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
860 static struct nfs4_client *
861 find_confirmed_client(clientid_t *clid)
863 struct nfs4_client *clp;
864 unsigned int idhashval = clientid_hashval(clid->cl_id);
866 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
867 if (same_clid(&clp->cl_clientid, clid))
873 static struct nfs4_client *
874 find_unconfirmed_client(clientid_t *clid)
876 struct nfs4_client *clp;
877 unsigned int idhashval = clientid_hashval(clid->cl_id);
879 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
880 if (same_clid(&clp->cl_clientid, clid))
887 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
888 * parameter. Matching is based on the fact the at least one of the
889 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
891 * FIXME: we need to unify the clientid namespaces for nfsv4.x
892 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
893 * and SET_CLIENTID{,_CONFIRM}
896 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
898 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
899 return use_exchange_id == has_exchange_flags;
902 static struct nfs4_client *
903 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
904 bool use_exchange_id)
906 struct nfs4_client *clp;
908 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
909 if (same_name(clp->cl_recdir, dname) &&
910 match_clientid_establishment(clp, use_exchange_id))
916 static struct nfs4_client *
917 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
918 bool use_exchange_id)
920 struct nfs4_client *clp;
922 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
923 if (same_name(clp->cl_recdir, dname) &&
924 match_clientid_establishment(clp, use_exchange_id))
931 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
933 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
934 unsigned short expected_family;
936 /* Currently, we only support tcp and tcp6 for the callback channel */
937 if (se->se_callback_netid_len == 3 &&
938 !memcmp(se->se_callback_netid_val, "tcp", 3))
939 expected_family = AF_INET;
940 else if (se->se_callback_netid_len == 4 &&
941 !memcmp(se->se_callback_netid_val, "tcp6", 4))
942 expected_family = AF_INET6;
946 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
947 se->se_callback_addr_len,
948 (struct sockaddr *) &cb->cb_addr,
949 sizeof(cb->cb_addr));
951 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
954 if (cb->cb_addr.ss_family == AF_INET6)
955 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
957 cb->cb_minorversion = 0;
958 cb->cb_prog = se->se_callback_prog;
959 cb->cb_ident = se->se_callback_ident;
962 cb->cb_addr.ss_family = AF_UNSPEC;
964 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
965 "will not receive delegations\n",
966 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
972 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
974 struct nfsd4_compoundres *resp = rqstp->rq_resp;
976 resp->cstate.statp = statp;
980 * Dereference the result pages.
983 nfsd4_release_respages(struct page **respages, short resused)
987 dprintk("--> %s\n", __func__);
988 for (i = 0; i < resused; i++) {
991 put_page(respages[i]);
997 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1001 for (i = 0; i < count; i++) {
1002 topages[i] = frompages[i];
1005 get_page(topages[i]);
1010 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1011 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1012 * length of the XDR response is less than se_fmaxresp_cached
1013 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1014 * of the reply (e.g. readdir).
1016 * Store the base and length of the rq_req.head[0] page
1017 * of the NFSv4.1 data, just past the rpc header.
1020 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1022 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1023 struct svc_rqst *rqstp = resp->rqstp;
1024 struct kvec *resv = &rqstp->rq_res.head[0];
1026 dprintk("--> %s entry %p\n", __func__, entry);
1028 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1029 entry->ce_opcnt = resp->opcnt;
1030 entry->ce_status = resp->cstate.status;
1033 * Don't need a page to cache just the sequence operation - the slot
1037 if (nfsd4_not_cached(resp)) {
1038 entry->ce_resused = 0;
1039 entry->ce_rpchdrlen = 0;
1040 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1041 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1044 entry->ce_resused = rqstp->rq_resused;
1045 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1046 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1047 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1049 entry->ce_datav.iov_base = resp->cstate.statp;
1050 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1051 (char *)page_address(rqstp->rq_respages[0]));
1052 /* Current request rpc header length*/
1053 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1054 (char *)page_address(rqstp->rq_respages[0]);
1058 * We keep the rpc header, but take the nfs reply from the replycache.
1061 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1062 struct nfsd4_cache_entry *entry)
1064 struct svc_rqst *rqstp = resp->rqstp;
1065 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1068 /* Current request rpc header length*/
1069 len = (char *)resp->cstate.statp -
1070 (char *)page_address(rqstp->rq_respages[0]);
1071 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1072 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1073 entry->ce_datav.iov_len);
1076 /* copy the cached reply nfsd data past the current rpc header */
1077 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1078 entry->ce_datav.iov_len);
1079 resv->iov_len = len + entry->ce_datav.iov_len;
1084 * Encode the replay sequence operation from the slot values.
1085 * If cachethis is FALSE encode the uncached rep error on the next
1086 * operation which sets resp->p and increments resp->opcnt for
1087 * nfs4svc_encode_compoundres.
1091 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1092 struct nfsd4_compoundres *resp)
1094 struct nfsd4_op *op;
1095 struct nfsd4_slot *slot = resp->cstate.slot;
1097 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1098 resp->opcnt, resp->cstate.slot->sl_cache_entry.ce_cachethis);
1100 /* Encode the replayed sequence operation */
1101 op = &args->ops[resp->opcnt - 1];
1102 nfsd4_encode_operation(resp, op);
1104 /* Return nfserr_retry_uncached_rep in next operation. */
1105 if (args->opcnt > 1 && slot->sl_cache_entry.ce_cachethis == 0) {
1106 op = &args->ops[resp->opcnt++];
1107 op->status = nfserr_retry_uncached_rep;
1108 nfsd4_encode_operation(resp, op);
1114 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1115 * cached page. Replace any futher replay pages from the cache.
1118 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1119 struct nfsd4_sequence *seq)
1121 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1124 dprintk("--> %s entry %p\n", __func__, entry);
1127 * If this is just the sequence operation, we did not keep
1128 * a page in the cache entry because we can just use the
1129 * slot info stored in struct nfsd4_sequence that was checked
1130 * against the slot in nfsd4_sequence().
1132 * This occurs when seq->cachethis is FALSE, or when the client
1133 * session inactivity timer fires and a solo sequence operation
1134 * is sent (lease renewal).
1136 seq->maxslots = resp->cstate.session->se_fchannel.maxreqs;
1138 /* Either returns 0 or nfserr_retry_uncached */
1139 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1140 if (status == nfserr_retry_uncached_rep)
1143 if (!nfsd41_copy_replay_data(resp, entry)) {
1145 * Not enough room to use the replay rpc header, send the
1146 * cached header. Release all the allocated result pages.
1148 svc_free_res_pages(resp->rqstp);
1149 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1152 /* Release all but the first allocated result page */
1154 resp->rqstp->rq_resused--;
1155 svc_free_res_pages(resp->rqstp);
1157 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1158 &entry->ce_respages[1],
1159 entry->ce_resused - 1);
1162 resp->rqstp->rq_resused = entry->ce_resused;
1163 resp->opcnt = entry->ce_opcnt;
1164 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1165 status = entry->ce_status;
1171 * Set the exchange_id flags returned by the server.
1174 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1176 /* pNFS is not supported */
1177 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1179 /* Referrals are supported, Migration is not. */
1180 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1182 /* set the wire flags to return to client. */
1183 clid->flags = new->cl_exchange_flags;
1187 nfsd4_exchange_id(struct svc_rqst *rqstp,
1188 struct nfsd4_compound_state *cstate,
1189 struct nfsd4_exchange_id *exid)
1191 struct nfs4_client *unconf, *conf, *new;
1193 unsigned int strhashval;
1194 char dname[HEXDIR_LEN];
1195 char addr_str[INET6_ADDRSTRLEN];
1196 nfs4_verifier verf = exid->verifier;
1197 struct sockaddr *sa = svc_addr(rqstp);
1199 rpc_ntop(sa, addr_str, sizeof(addr_str));
1200 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1201 "ip_addr=%s flags %x, spa_how %d\n",
1202 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1203 addr_str, exid->flags, exid->spa_how);
1205 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1206 return nfserr_inval;
1208 /* Currently only support SP4_NONE */
1209 switch (exid->spa_how) {
1213 return nfserr_encr_alg_unsupp;
1215 BUG(); /* checked by xdr code */
1217 return nfserr_serverfault; /* no excuse :-/ */
1220 status = nfs4_make_rec_clidname(dname, &exid->clname);
1225 strhashval = clientstr_hashval(dname);
1230 conf = find_confirmed_client_by_str(dname, strhashval, true);
1232 if (!same_verf(&verf, &conf->cl_verifier)) {
1233 /* 18.35.4 case 8 */
1234 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1235 status = nfserr_not_same;
1238 /* Client reboot: destroy old state */
1239 expire_client(conf);
1242 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1243 /* 18.35.4 case 9 */
1244 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1245 status = nfserr_perm;
1248 expire_client(conf);
1252 * Set bit when the owner id and verifier map to an already
1253 * confirmed client id (18.35.3).
1255 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1258 * Falling into 18.35.4 case 2, possible router replay.
1259 * Leave confirmed record intact and return same result.
1261 copy_verf(conf, &verf);
1266 /* 18.35.4 case 7 */
1267 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1268 status = nfserr_noent;
1272 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1275 * Possible retry or client restart. Per 18.35.4 case 4,
1276 * a new unconfirmed record should be generated regardless
1277 * of whether any properties have changed.
1279 expire_client(unconf);
1284 new = create_client(exid->clname, dname);
1286 status = nfserr_resource;
1290 copy_verf(new, &verf);
1291 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1292 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1295 add_to_unconfirmed(new, strhashval);
1297 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1298 exid->clientid.cl_id = new->cl_clientid.cl_id;
1301 nfsd4_set_ex_flags(new, exid);
1303 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1304 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1308 nfs4_unlock_state();
1310 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1315 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1317 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1320 /* The slot is in use, and no response has been sent. */
1322 if (seqid == slot_seqid)
1323 return nfserr_jukebox;
1325 return nfserr_seq_misordered;
1328 if (likely(seqid == slot_seqid + 1))
1331 if (seqid == slot_seqid)
1332 return nfserr_replay_cache;
1334 if (seqid == 1 && (slot_seqid + 1) == 0)
1336 /* Misordered replay or misordered new request */
1337 return nfserr_seq_misordered;
1341 * Cache the create session result into the create session single DRC
1342 * slot cache by saving the xdr structure. sl_seqid has been set.
1343 * Do this for solo or embedded create session operations.
1346 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1347 struct nfsd4_clid_slot *slot, int nfserr)
1349 slot->sl_status = nfserr;
1350 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1354 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1355 struct nfsd4_clid_slot *slot)
1357 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1358 return slot->sl_status;
1362 nfsd4_create_session(struct svc_rqst *rqstp,
1363 struct nfsd4_compound_state *cstate,
1364 struct nfsd4_create_session *cr_ses)
1366 struct sockaddr *sa = svc_addr(rqstp);
1367 struct nfs4_client *conf, *unconf;
1368 struct nfsd4_clid_slot *cs_slot = NULL;
1372 unconf = find_unconfirmed_client(&cr_ses->clientid);
1373 conf = find_confirmed_client(&cr_ses->clientid);
1376 cs_slot = &conf->cl_cs_slot;
1377 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1378 if (status == nfserr_replay_cache) {
1379 dprintk("Got a create_session replay! seqid= %d\n",
1381 /* Return the cached reply status */
1382 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1384 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1385 status = nfserr_seq_misordered;
1386 dprintk("Sequence misordered!\n");
1387 dprintk("Expected seqid= %d but got seqid= %d\n",
1388 cs_slot->sl_seqid, cr_ses->seqid);
1391 cs_slot->sl_seqid++;
1392 } else if (unconf) {
1393 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1394 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1395 status = nfserr_clid_inuse;
1399 cs_slot = &unconf->cl_cs_slot;
1400 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1402 /* an unconfirmed replay returns misordered */
1403 status = nfserr_seq_misordered;
1407 cs_slot->sl_seqid++; /* from 0 to 1 */
1408 move_to_confirmed(unconf);
1411 * We do not support RDMA or persistent sessions
1413 cr_ses->flags &= ~SESSION4_PERSIST;
1414 cr_ses->flags &= ~SESSION4_RDMA;
1418 status = nfserr_stale_clientid;
1422 status = alloc_init_session(rqstp, conf, cr_ses);
1426 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1427 NFS4_MAX_SESSIONID_LEN);
1428 cr_ses->seqid = cs_slot->sl_seqid;
1431 /* cache solo and embedded create sessions under the state lock */
1432 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1434 nfs4_unlock_state();
1435 dprintk("%s returns %d\n", __func__, ntohl(status));
1440 nfsd4_destroy_session(struct svc_rqst *r,
1441 struct nfsd4_compound_state *cstate,
1442 struct nfsd4_destroy_session *sessionid)
1444 struct nfsd4_session *ses;
1445 u32 status = nfserr_badsession;
1448 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1449 * - Should we return nfserr_back_chan_busy if waiting for
1450 * callbacks on to-be-destroyed session?
1451 * - Do we need to clear any callback info from previous session?
1454 dump_sessionid(__func__, &sessionid->sessionid);
1455 spin_lock(&sessionid_lock);
1456 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1458 spin_unlock(&sessionid_lock);
1462 unhash_session(ses);
1463 spin_unlock(&sessionid_lock);
1465 /* wait for callbacks */
1466 shutdown_callback_client(ses->se_client);
1467 nfsd4_put_session(ses);
1470 dprintk("%s returns %d\n", __func__, ntohl(status));
1475 nfsd4_sequence(struct svc_rqst *rqstp,
1476 struct nfsd4_compound_state *cstate,
1477 struct nfsd4_sequence *seq)
1479 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1480 struct nfsd4_session *session;
1481 struct nfsd4_slot *slot;
1484 if (resp->opcnt != 1)
1485 return nfserr_sequence_pos;
1487 spin_lock(&sessionid_lock);
1488 status = nfserr_badsession;
1489 session = find_in_sessionid_hashtbl(&seq->sessionid);
1493 status = nfserr_badslot;
1494 if (seq->slotid >= session->se_fchannel.maxreqs)
1497 slot = &session->se_slots[seq->slotid];
1498 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1500 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1501 if (status == nfserr_replay_cache) {
1502 cstate->slot = slot;
1503 cstate->session = session;
1504 /* Return the cached reply status and set cstate->status
1505 * for nfsd4_svc_encode_compoundres processing */
1506 status = nfsd4_replay_cache_entry(resp, seq);
1507 cstate->status = nfserr_replay_cache;
1513 /* Success! bump slot seqid */
1514 slot->sl_inuse = true;
1515 slot->sl_seqid = seq->seqid;
1516 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1518 cstate->slot = slot;
1519 cstate->session = session;
1521 /* Hold a session reference until done processing the compound:
1522 * nfsd4_put_session called only if the cstate slot is set.
1524 nfsd4_get_session(session);
1526 spin_unlock(&sessionid_lock);
1527 /* Renew the clientid on success and on replay */
1528 if (cstate->session) {
1530 renew_client(session->se_client);
1531 nfs4_unlock_state();
1533 dprintk("%s: return %d\n", __func__, ntohl(status));
1538 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1539 struct nfsd4_setclientid *setclid)
1541 struct sockaddr *sa = svc_addr(rqstp);
1542 struct xdr_netobj clname = {
1543 .len = setclid->se_namelen,
1544 .data = setclid->se_name,
1546 nfs4_verifier clverifier = setclid->se_verf;
1547 unsigned int strhashval;
1548 struct nfs4_client *conf, *unconf, *new;
1551 char dname[HEXDIR_LEN];
1553 if (!check_name(clname))
1554 return nfserr_inval;
1556 status = nfs4_make_rec_clidname(dname, &clname);
1561 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1562 * We get here on a DRC miss.
1565 strhashval = clientstr_hashval(dname);
1568 conf = find_confirmed_client_by_str(dname, strhashval, false);
1570 /* RFC 3530 14.2.33 CASE 0: */
1571 status = nfserr_clid_inuse;
1572 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1573 char addr_str[INET6_ADDRSTRLEN];
1574 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1576 dprintk("NFSD: setclientid: string in use by client "
1577 "at %s\n", addr_str);
1582 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1583 * has a description of SETCLIENTID request processing consisting
1584 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1586 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1587 status = nfserr_resource;
1590 * RFC 3530 14.2.33 CASE 4:
1591 * placed first, because it is the normal case
1594 expire_client(unconf);
1595 new = create_client(clname, dname);
1599 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1601 * RFC 3530 14.2.33 CASE 1:
1602 * probable callback update
1605 /* Note this is removing unconfirmed {*x***},
1606 * which is stronger than RFC recommended {vxc**}.
1607 * This has the advantage that there is at most
1608 * one {*x***} in either list at any time.
1610 expire_client(unconf);
1612 new = create_client(clname, dname);
1615 copy_clid(new, conf);
1616 } else if (!unconf) {
1618 * RFC 3530 14.2.33 CASE 2:
1619 * probable client reboot; state will be removed if
1622 new = create_client(clname, dname);
1628 * RFC 3530 14.2.33 CASE 3:
1629 * probable client reboot; state will be removed if
1632 expire_client(unconf);
1633 new = create_client(clname, dname);
1638 copy_verf(new, &clverifier);
1639 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1640 new->cl_flavor = rqstp->rq_flavor;
1641 princ = svc_gss_principal(rqstp);
1643 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1644 if (new->cl_principal == NULL) {
1649 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1651 gen_callback(new, setclid, rpc_get_scope_id(sa));
1652 add_to_unconfirmed(new, strhashval);
1653 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1654 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1655 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1658 nfs4_unlock_state();
1664 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1665 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1666 * bullets, labeled as CASE1 - CASE4 below.
1669 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1670 struct nfsd4_compound_state *cstate,
1671 struct nfsd4_setclientid_confirm *setclientid_confirm)
1673 struct sockaddr *sa = svc_addr(rqstp);
1674 struct nfs4_client *conf, *unconf;
1675 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1676 clientid_t * clid = &setclientid_confirm->sc_clientid;
1679 if (STALE_CLIENTID(clid))
1680 return nfserr_stale_clientid;
1682 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1683 * We get here on a DRC miss.
1688 conf = find_confirmed_client(clid);
1689 unconf = find_unconfirmed_client(clid);
1691 status = nfserr_clid_inuse;
1692 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1694 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1698 * section 14.2.34 of RFC 3530 has a description of
1699 * SETCLIENTID_CONFIRM request processing consisting
1700 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1702 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1704 * RFC 3530 14.2.34 CASE 1:
1707 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1708 status = nfserr_clid_inuse;
1710 /* XXX: We just turn off callbacks until we can handle
1711 * change request correctly. */
1712 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1713 expire_client(unconf);
1717 } else if (conf && !unconf) {
1719 * RFC 3530 14.2.34 CASE 2:
1720 * probable retransmitted request; play it safe and
1723 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1724 status = nfserr_clid_inuse;
1727 } else if (!conf && unconf
1728 && same_verf(&unconf->cl_confirm, &confirm)) {
1730 * RFC 3530 14.2.34 CASE 3:
1731 * Normal case; new or rebooted client:
1733 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1734 status = nfserr_clid_inuse;
1737 clientstr_hashval(unconf->cl_recdir);
1738 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1741 nfsd4_remove_clid_dir(conf);
1742 expire_client(conf);
1744 move_to_confirmed(unconf);
1746 nfsd4_probe_callback(conf);
1749 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1750 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1753 * RFC 3530 14.2.34 CASE 4:
1754 * Client probably hasn't noticed that we rebooted yet.
1756 status = nfserr_stale_clientid;
1758 /* check that we have hit one of the cases...*/
1759 status = nfserr_clid_inuse;
1762 nfs4_unlock_state();
1766 /* OPEN Share state helper functions */
1767 static inline struct nfs4_file *
1768 alloc_init_file(struct inode *ino)
1770 struct nfs4_file *fp;
1771 unsigned int hashval = file_hashval(ino);
1773 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1775 atomic_set(&fp->fi_ref, 1);
1776 INIT_LIST_HEAD(&fp->fi_hash);
1777 INIT_LIST_HEAD(&fp->fi_stateids);
1778 INIT_LIST_HEAD(&fp->fi_delegations);
1779 spin_lock(&recall_lock);
1780 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1781 spin_unlock(&recall_lock);
1782 fp->fi_inode = igrab(ino);
1783 fp->fi_id = current_fileid++;
1784 fp->fi_had_conflict = false;
1791 nfsd4_free_slab(struct kmem_cache **slab)
1795 kmem_cache_destroy(*slab);
1800 nfsd4_free_slabs(void)
1802 nfsd4_free_slab(&stateowner_slab);
1803 nfsd4_free_slab(&file_slab);
1804 nfsd4_free_slab(&stateid_slab);
1805 nfsd4_free_slab(&deleg_slab);
1809 nfsd4_init_slabs(void)
1811 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1812 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1813 if (stateowner_slab == NULL)
1815 file_slab = kmem_cache_create("nfsd4_files",
1816 sizeof(struct nfs4_file), 0, 0, NULL);
1817 if (file_slab == NULL)
1819 stateid_slab = kmem_cache_create("nfsd4_stateids",
1820 sizeof(struct nfs4_stateid), 0, 0, NULL);
1821 if (stateid_slab == NULL)
1823 deleg_slab = kmem_cache_create("nfsd4_delegations",
1824 sizeof(struct nfs4_delegation), 0, 0, NULL);
1825 if (deleg_slab == NULL)
1830 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1835 nfs4_free_stateowner(struct kref *kref)
1837 struct nfs4_stateowner *sop =
1838 container_of(kref, struct nfs4_stateowner, so_ref);
1839 kfree(sop->so_owner.data);
1840 kmem_cache_free(stateowner_slab, sop);
1843 static inline struct nfs4_stateowner *
1844 alloc_stateowner(struct xdr_netobj *owner)
1846 struct nfs4_stateowner *sop;
1848 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1849 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1850 memcpy(sop->so_owner.data, owner->data, owner->len);
1851 sop->so_owner.len = owner->len;
1852 kref_init(&sop->so_ref);
1855 kmem_cache_free(stateowner_slab, sop);
1860 static struct nfs4_stateowner *
1861 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1862 struct nfs4_stateowner *sop;
1863 struct nfs4_replay *rp;
1864 unsigned int idhashval;
1866 if (!(sop = alloc_stateowner(&open->op_owner)))
1868 idhashval = ownerid_hashval(current_ownerid);
1869 INIT_LIST_HEAD(&sop->so_idhash);
1870 INIT_LIST_HEAD(&sop->so_strhash);
1871 INIT_LIST_HEAD(&sop->so_perclient);
1872 INIT_LIST_HEAD(&sop->so_stateids);
1873 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1874 INIT_LIST_HEAD(&sop->so_close_lru);
1876 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1877 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1878 list_add(&sop->so_perclient, &clp->cl_openowners);
1879 sop->so_is_open_owner = 1;
1880 sop->so_id = current_ownerid++;
1881 sop->so_client = clp;
1882 sop->so_seqid = open->op_seqid;
1883 sop->so_confirmed = 0;
1884 rp = &sop->so_replay;
1885 rp->rp_status = nfserr_serverfault;
1887 rp->rp_buf = rp->rp_ibuf;
1892 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1893 struct nfs4_stateowner *sop = open->op_stateowner;
1894 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1896 INIT_LIST_HEAD(&stp->st_hash);
1897 INIT_LIST_HEAD(&stp->st_perstateowner);
1898 INIT_LIST_HEAD(&stp->st_lockowners);
1899 INIT_LIST_HEAD(&stp->st_perfile);
1900 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1901 list_add(&stp->st_perstateowner, &sop->so_stateids);
1902 list_add(&stp->st_perfile, &fp->fi_stateids);
1903 stp->st_stateowner = sop;
1906 stp->st_stateid.si_boot = get_seconds();
1907 stp->st_stateid.si_stateownerid = sop->so_id;
1908 stp->st_stateid.si_fileid = fp->fi_id;
1909 stp->st_stateid.si_generation = 0;
1910 stp->st_access_bmap = 0;
1911 stp->st_deny_bmap = 0;
1912 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1913 &stp->st_access_bmap);
1914 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1915 stp->st_openstp = NULL;
1919 move_to_close_lru(struct nfs4_stateowner *sop)
1921 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1923 list_move_tail(&sop->so_close_lru, &close_lru);
1924 sop->so_time = get_seconds();
1928 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1931 return (sop->so_owner.len == owner->len) &&
1932 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1933 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1936 static struct nfs4_stateowner *
1937 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1939 struct nfs4_stateowner *so = NULL;
1941 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1942 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1948 /* search file_hashtbl[] for file */
1949 static struct nfs4_file *
1950 find_file(struct inode *ino)
1952 unsigned int hashval = file_hashval(ino);
1953 struct nfs4_file *fp;
1955 spin_lock(&recall_lock);
1956 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1957 if (fp->fi_inode == ino) {
1959 spin_unlock(&recall_lock);
1963 spin_unlock(&recall_lock);
1967 static inline int access_valid(u32 x, u32 minorversion)
1969 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1971 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1973 x &= ~NFS4_SHARE_ACCESS_MASK;
1974 if (minorversion && x) {
1975 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1977 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1979 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1986 static inline int deny_valid(u32 x)
1988 /* Note: unlike access bits, deny bits may be zero. */
1989 return x <= NFS4_SHARE_DENY_BOTH;
1993 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1994 * st_{access,deny}_bmap field of the stateid, in order to track not
1995 * only what share bits are currently in force, but also what
1996 * combinations of share bits previous opens have used. This allows us
1997 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1998 * return an error if the client attempt to downgrade to a combination
1999 * of share bits not explicable by closing some of its previous opens.
2001 * XXX: This enforcement is actually incomplete, since we don't keep
2002 * track of access/deny bit combinations; so, e.g., we allow:
2004 * OPEN allow read, deny write
2005 * OPEN allow both, deny none
2006 * DOWNGRADE allow read, deny none
2008 * which we should reject.
2011 set_access(unsigned int *access, unsigned long bmap) {
2015 for (i = 1; i < 4; i++) {
2016 if (test_bit(i, &bmap))
2022 set_deny(unsigned int *deny, unsigned long bmap) {
2026 for (i = 0; i < 4; i++) {
2027 if (test_bit(i, &bmap))
2033 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2034 unsigned int access, deny;
2036 set_access(&access, stp->st_access_bmap);
2037 set_deny(&deny, stp->st_deny_bmap);
2038 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2044 * Called to check deny when READ with all zero stateid or
2045 * WRITE with all zero or all one stateid
2048 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2050 struct inode *ino = current_fh->fh_dentry->d_inode;
2051 struct nfs4_file *fp;
2052 struct nfs4_stateid *stp;
2055 dprintk("NFSD: nfs4_share_conflict\n");
2057 fp = find_file(ino);
2060 ret = nfserr_locked;
2061 /* Search for conflicting share reservations */
2062 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2063 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2064 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2074 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2076 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2077 drop_file_write_access(filp);
2078 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2083 * Spawn a thread to perform a recall on the delegation represented
2084 * by the lease (file_lock)
2086 * Called from break_lease() with lock_kernel() held.
2087 * Note: we assume break_lease will only call this *once* for any given
2091 void nfsd_break_deleg_cb(struct file_lock *fl)
2093 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2095 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2099 /* We're assuming the state code never drops its reference
2100 * without first removing the lease. Since we're in this lease
2101 * callback (and since the lease code is serialized by the kernel
2102 * lock) we know the server hasn't removed the lease yet, we know
2103 * it's safe to take a reference: */
2104 atomic_inc(&dp->dl_count);
2105 atomic_inc(&dp->dl_client->cl_count);
2107 spin_lock(&recall_lock);
2108 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2109 spin_unlock(&recall_lock);
2111 /* only place dl_time is set. protected by lock_kernel*/
2112 dp->dl_time = get_seconds();
2115 * We don't want the locks code to timeout the lease for us;
2116 * we'll remove it ourself if the delegation isn't returned
2119 fl->fl_break_time = 0;
2121 dp->dl_file->fi_had_conflict = true;
2122 nfsd4_cb_recall(dp);
2126 * The file_lock is being reapd.
2128 * Called by locks_free_lock() with lock_kernel() held.
2131 void nfsd_release_deleg_cb(struct file_lock *fl)
2133 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2135 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2137 if (!(fl->fl_flags & FL_LEASE) || !dp)
2139 dp->dl_flock = NULL;
2143 * Set the delegation file_lock back pointer.
2145 * Called from setlease() with lock_kernel() held.
2148 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2150 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2152 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2159 * Called from setlease() with lock_kernel() held
2162 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2164 struct nfs4_delegation *onlistd =
2165 (struct nfs4_delegation *)onlist->fl_owner;
2166 struct nfs4_delegation *tryd =
2167 (struct nfs4_delegation *)try->fl_owner;
2169 if (onlist->fl_lmops != try->fl_lmops)
2172 return onlistd->dl_client == tryd->dl_client;
2177 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2180 return lease_modify(onlist, arg);
2185 static struct lock_manager_operations nfsd_lease_mng_ops = {
2186 .fl_break = nfsd_break_deleg_cb,
2187 .fl_release_private = nfsd_release_deleg_cb,
2188 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2189 .fl_mylease = nfsd_same_client_deleg_cb,
2190 .fl_change = nfsd_change_deleg_cb,
2195 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2196 struct nfsd4_open *open)
2198 clientid_t *clientid = &open->op_clientid;
2199 struct nfs4_client *clp = NULL;
2200 unsigned int strhashval;
2201 struct nfs4_stateowner *sop = NULL;
2203 if (!check_name(open->op_owner))
2204 return nfserr_inval;
2206 if (STALE_CLIENTID(&open->op_clientid))
2207 return nfserr_stale_clientid;
2209 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2210 sop = find_openstateowner_str(strhashval, open);
2211 open->op_stateowner = sop;
2213 /* Make sure the client's lease hasn't expired. */
2214 clp = find_confirmed_client(clientid);
2216 return nfserr_expired;
2219 /* When sessions are used, skip open sequenceid processing */
2220 if (nfsd4_has_session(cstate))
2222 if (!sop->so_confirmed) {
2223 /* Replace unconfirmed owners without checking for replay. */
2224 clp = sop->so_client;
2225 release_openowner(sop);
2226 open->op_stateowner = NULL;
2229 if (open->op_seqid == sop->so_seqid - 1) {
2230 if (sop->so_replay.rp_buflen)
2231 return nfserr_replay_me;
2232 /* The original OPEN failed so spectacularly
2233 * that we don't even have replay data saved!
2234 * Therefore, we have no choice but to continue
2235 * processing this OPEN; presumably, we'll
2236 * fail again for the same reason.
2238 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2241 if (open->op_seqid != sop->so_seqid)
2242 return nfserr_bad_seqid;
2244 if (open->op_stateowner == NULL) {
2245 sop = alloc_init_open_stateowner(strhashval, clp, open);
2247 return nfserr_resource;
2248 open->op_stateowner = sop;
2250 list_del_init(&sop->so_close_lru);
2251 renew_client(sop->so_client);
2255 static inline __be32
2256 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2258 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2259 return nfserr_openmode;
2264 static struct nfs4_delegation *
2265 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2267 struct nfs4_delegation *dp;
2269 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2270 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2277 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2278 struct nfs4_delegation **dp)
2281 __be32 status = nfserr_bad_stateid;
2283 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2286 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2287 RD_STATE : WR_STATE;
2288 status = nfs4_check_delegmode(*dp, flags);
2292 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2296 open->op_stateowner->so_confirmed = 1;
2301 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2303 struct nfs4_stateid *local;
2304 __be32 status = nfserr_share_denied;
2305 struct nfs4_stateowner *sop = open->op_stateowner;
2307 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2308 /* ignore lock owners */
2309 if (local->st_stateowner->so_is_open_owner == 0)
2311 /* remember if we have seen this open owner */
2312 if (local->st_stateowner == sop)
2314 /* check for conflicting share reservations */
2315 if (!test_share(local, open))
2323 static inline struct nfs4_stateid *
2324 nfs4_alloc_stateid(void)
2326 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2330 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2331 struct nfs4_delegation *dp,
2332 struct svc_fh *cur_fh, int flags)
2334 struct nfs4_stateid *stp;
2336 stp = nfs4_alloc_stateid();
2338 return nfserr_resource;
2341 get_file(dp->dl_vfs_file);
2342 stp->st_vfs_file = dp->dl_vfs_file;
2345 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2348 if (status == nfserr_dropit)
2349 status = nfserr_jukebox;
2350 kmem_cache_free(stateid_slab, stp);
2358 static inline __be32
2359 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2360 struct nfsd4_open *open)
2362 struct iattr iattr = {
2363 .ia_valid = ATTR_SIZE,
2366 if (!open->op_truncate)
2368 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2369 return nfserr_inval;
2370 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2374 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2376 struct file *filp = stp->st_vfs_file;
2377 struct inode *inode = filp->f_path.dentry->d_inode;
2378 unsigned int share_access, new_writer;
2381 set_access(&share_access, stp->st_access_bmap);
2382 new_writer = (~share_access) & open->op_share_access
2383 & NFS4_SHARE_ACCESS_WRITE;
2386 int err = get_write_access(inode);
2388 return nfserrno(err);
2389 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2391 return nfserrno(err);
2392 file_take_write(filp);
2394 status = nfsd4_truncate(rqstp, cur_fh, open);
2397 put_write_access(inode);
2400 /* remember the open */
2401 filp->f_mode |= open->op_share_access;
2402 __set_bit(open->op_share_access, &stp->st_access_bmap);
2403 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2410 nfs4_set_claim_prev(struct nfsd4_open *open)
2412 open->op_stateowner->so_confirmed = 1;
2413 open->op_stateowner->so_client->cl_firststate = 1;
2417 * Attempt to hand out a delegation.
2420 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2422 struct nfs4_delegation *dp;
2423 struct nfs4_stateowner *sop = stp->st_stateowner;
2424 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2425 struct file_lock fl, *flp = &fl;
2426 int status, flag = 0;
2428 flag = NFS4_OPEN_DELEGATE_NONE;
2429 open->op_recall = 0;
2430 switch (open->op_claim_type) {
2431 case NFS4_OPEN_CLAIM_PREVIOUS:
2432 if (!atomic_read(&cb->cb_set))
2433 open->op_recall = 1;
2434 flag = open->op_delegate_type;
2435 if (flag == NFS4_OPEN_DELEGATE_NONE)
2438 case NFS4_OPEN_CLAIM_NULL:
2439 /* Let's not give out any delegations till everyone's
2440 * had the chance to reclaim theirs.... */
2441 if (locks_in_grace())
2443 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2445 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2446 flag = NFS4_OPEN_DELEGATE_WRITE;
2448 flag = NFS4_OPEN_DELEGATE_READ;
2454 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2456 flag = NFS4_OPEN_DELEGATE_NONE;
2459 locks_init_lock(&fl);
2460 fl.fl_lmops = &nfsd_lease_mng_ops;
2461 fl.fl_flags = FL_LEASE;
2462 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2463 fl.fl_end = OFFSET_MAX;
2464 fl.fl_owner = (fl_owner_t)dp;
2465 fl.fl_file = stp->st_vfs_file;
2466 fl.fl_pid = current->tgid;
2468 /* vfs_setlease checks to see if delegation should be handed out.
2469 * the lock_manager callbacks fl_mylease and fl_change are used
2471 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2472 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2473 unhash_delegation(dp);
2474 flag = NFS4_OPEN_DELEGATE_NONE;
2478 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2480 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2481 dp->dl_stateid.si_boot,
2482 dp->dl_stateid.si_stateownerid,
2483 dp->dl_stateid.si_fileid,
2484 dp->dl_stateid.si_generation);
2486 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2487 && flag == NFS4_OPEN_DELEGATE_NONE
2488 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2489 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2490 open->op_delegate_type = flag;
2494 * called with nfs4_lock_state() held.
2497 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2499 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2500 struct nfs4_file *fp = NULL;
2501 struct inode *ino = current_fh->fh_dentry->d_inode;
2502 struct nfs4_stateid *stp = NULL;
2503 struct nfs4_delegation *dp = NULL;
2506 status = nfserr_inval;
2507 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2508 || !deny_valid(open->op_share_deny))
2511 * Lookup file; if found, lookup stateid and check open request,
2512 * and check for delegations in the process of being recalled.
2513 * If not found, create the nfs4_file struct
2515 fp = find_file(ino);
2517 if ((status = nfs4_check_open(fp, open, &stp)))
2519 status = nfs4_check_deleg(fp, open, &dp);
2523 status = nfserr_bad_stateid;
2524 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2526 status = nfserr_resource;
2527 fp = alloc_init_file(ino);
2533 * OPEN the file, or upgrade an existing OPEN.
2534 * If truncate fails, the OPEN fails.
2537 /* Stateid was found, this is an OPEN upgrade */
2538 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2541 update_stateid(&stp->st_stateid);
2543 /* Stateid was not found, this is a new OPEN */
2545 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2546 flags |= NFSD_MAY_READ;
2547 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2548 flags |= NFSD_MAY_WRITE;
2549 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2552 init_stateid(stp, fp, open);
2553 status = nfsd4_truncate(rqstp, current_fh, open);
2555 release_open_stateid(stp);
2558 if (nfsd4_has_session(&resp->cstate))
2559 update_stateid(&stp->st_stateid);
2561 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2563 if (nfsd4_has_session(&resp->cstate))
2564 open->op_stateowner->so_confirmed = 1;
2567 * Attempt to hand out a delegation. No error return, because the
2568 * OPEN succeeds even if we fail.
2570 nfs4_open_delegation(current_fh, open, stp);
2574 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2575 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2576 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2580 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2581 nfs4_set_claim_prev(open);
2583 * To finish the open response, we just need to set the rflags.
2585 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2586 if (!open->op_stateowner->so_confirmed &&
2587 !nfsd4_has_session(&resp->cstate))
2588 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2594 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2597 struct nfs4_client *clp;
2601 dprintk("process_renew(%08x/%08x): starting\n",
2602 clid->cl_boot, clid->cl_id);
2603 status = nfserr_stale_clientid;
2604 if (STALE_CLIENTID(clid))
2606 clp = find_confirmed_client(clid);
2607 status = nfserr_expired;
2609 /* We assume the client took too long to RENEW. */
2610 dprintk("nfsd4_renew: clientid not found!\n");
2614 status = nfserr_cb_path_down;
2615 if (!list_empty(&clp->cl_delegations)
2616 && !atomic_read(&clp->cl_cb_conn.cb_set))
2620 nfs4_unlock_state();
2624 struct lock_manager nfsd4_manager = {
2628 nfsd4_end_grace(void)
2630 dprintk("NFSD: end of grace period\n");
2631 nfsd4_recdir_purge_old();
2632 locks_end_grace(&nfsd4_manager);
2636 nfs4_laundromat(void)
2638 struct nfs4_client *clp;
2639 struct nfs4_stateowner *sop;
2640 struct nfs4_delegation *dp;
2641 struct list_head *pos, *next, reaplist;
2642 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2643 time_t t, clientid_val = NFSD_LEASE_TIME;
2644 time_t u, test_val = NFSD_LEASE_TIME;
2648 dprintk("NFSD: laundromat service - starting\n");
2649 if (locks_in_grace())
2651 list_for_each_safe(pos, next, &client_lru) {
2652 clp = list_entry(pos, struct nfs4_client, cl_lru);
2653 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2654 t = clp->cl_time - cutoff;
2655 if (clientid_val > t)
2659 dprintk("NFSD: purging unused client (clientid %08x)\n",
2660 clp->cl_clientid.cl_id);
2661 nfsd4_remove_clid_dir(clp);
2664 INIT_LIST_HEAD(&reaplist);
2665 spin_lock(&recall_lock);
2666 list_for_each_safe(pos, next, &del_recall_lru) {
2667 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2668 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2669 u = dp->dl_time - cutoff;
2674 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2676 list_move(&dp->dl_recall_lru, &reaplist);
2678 spin_unlock(&recall_lock);
2679 list_for_each_safe(pos, next, &reaplist) {
2680 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2681 list_del_init(&dp->dl_recall_lru);
2682 unhash_delegation(dp);
2684 test_val = NFSD_LEASE_TIME;
2685 list_for_each_safe(pos, next, &close_lru) {
2686 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2687 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2688 u = sop->so_time - cutoff;
2693 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2695 release_openowner(sop);
2697 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2698 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2699 nfs4_unlock_state();
2700 return clientid_val;
2703 static struct workqueue_struct *laundry_wq;
2704 static void laundromat_main(struct work_struct *);
2705 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2708 laundromat_main(struct work_struct *not_used)
2712 t = nfs4_laundromat();
2713 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2714 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2717 static struct nfs4_stateowner *
2718 search_close_lru(u32 st_id, int flags)
2720 struct nfs4_stateowner *local = NULL;
2722 if (flags & CLOSE_STATE) {
2723 list_for_each_entry(local, &close_lru, so_close_lru) {
2724 if (local->so_id == st_id)
2732 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2734 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2738 STALE_STATEID(stateid_t *stateid)
2740 if (time_after((unsigned long)boot_time,
2741 (unsigned long)stateid->si_boot)) {
2742 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2743 stateid->si_boot, stateid->si_stateownerid,
2744 stateid->si_fileid, stateid->si_generation);
2751 EXPIRED_STATEID(stateid_t *stateid)
2753 if (time_before((unsigned long)boot_time,
2754 ((unsigned long)stateid->si_boot)) &&
2755 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2756 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2757 stateid->si_boot, stateid->si_stateownerid,
2758 stateid->si_fileid, stateid->si_generation);
2765 stateid_error_map(stateid_t *stateid)
2767 if (STALE_STATEID(stateid))
2768 return nfserr_stale_stateid;
2769 if (EXPIRED_STATEID(stateid))
2770 return nfserr_expired;
2772 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2773 stateid->si_boot, stateid->si_stateownerid,
2774 stateid->si_fileid, stateid->si_generation);
2775 return nfserr_bad_stateid;
2779 access_permit_read(unsigned long access_bmap)
2781 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2782 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2783 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2787 access_permit_write(unsigned long access_bmap)
2789 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2790 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2794 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2796 __be32 status = nfserr_openmode;
2798 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2800 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2807 static inline __be32
2808 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2810 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2812 else if (locks_in_grace()) {
2813 /* Answer in remaining cases depends on existance of
2814 * conflicting state; so we must wait out the grace period. */
2815 return nfserr_grace;
2816 } else if (flags & WR_STATE)
2817 return nfs4_share_conflict(current_fh,
2818 NFS4_SHARE_DENY_WRITE);
2819 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2820 return nfs4_share_conflict(current_fh,
2821 NFS4_SHARE_DENY_READ);
2825 * Allow READ/WRITE during grace period on recovered state only for files
2826 * that are not able to provide mandatory locking.
2829 grace_disallows_io(struct inode *inode)
2831 return locks_in_grace() && mandatory_lock(inode);
2834 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2837 * When sessions are used the stateid generation number is ignored
2840 if ((flags & HAS_SESSION) && in->si_generation == 0)
2843 /* If the client sends us a stateid from the future, it's buggy: */
2844 if (in->si_generation > ref->si_generation)
2845 return nfserr_bad_stateid;
2847 * The following, however, can happen. For example, if the
2848 * client sends an open and some IO at the same time, the open
2849 * may bump si_generation while the IO is still in flight.
2850 * Thanks to hard links and renames, the client never knows what
2851 * file an open will affect. So it could avoid that situation
2852 * only by serializing all opens and IO from the same open
2853 * owner. To recover from the old_stateid error, the client
2854 * will just have to retry the IO:
2856 if (in->si_generation < ref->si_generation)
2857 return nfserr_old_stateid;
2862 static int is_delegation_stateid(stateid_t *stateid)
2864 return stateid->si_fileid == 0;
2868 * Checks for stateid operations
2871 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2872 stateid_t *stateid, int flags, struct file **filpp)
2874 struct nfs4_stateid *stp = NULL;
2875 struct nfs4_delegation *dp = NULL;
2876 struct svc_fh *current_fh = &cstate->current_fh;
2877 struct inode *ino = current_fh->fh_dentry->d_inode;
2883 if (grace_disallows_io(ino))
2884 return nfserr_grace;
2886 if (nfsd4_has_session(cstate))
2887 flags |= HAS_SESSION;
2889 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2890 return check_special_stateids(current_fh, stateid, flags);
2892 status = nfserr_stale_stateid;
2893 if (STALE_STATEID(stateid))
2896 status = nfserr_bad_stateid;
2897 if (is_delegation_stateid(stateid)) {
2898 dp = find_delegation_stateid(ino, stateid);
2900 status = stateid_error_map(stateid);
2903 status = check_stateid_generation(stateid, &dp->dl_stateid,
2907 status = nfs4_check_delegmode(dp, flags);
2910 renew_client(dp->dl_client);
2912 *filpp = dp->dl_vfs_file;
2913 } else { /* open or lock stateid */
2914 stp = find_stateid(stateid, flags);
2916 status = stateid_error_map(stateid);
2919 if (nfs4_check_fh(current_fh, stp))
2921 if (!stp->st_stateowner->so_confirmed)
2923 status = check_stateid_generation(stateid, &stp->st_stateid,
2927 status = nfs4_check_openmode(stp, flags);
2930 renew_client(stp->st_stateowner->so_client);
2932 *filpp = stp->st_vfs_file;
2942 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2943 RD_STATE : WR_STATE;
2947 * Checks for sequence id mutating operations.
2950 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2951 stateid_t *stateid, int flags,
2952 struct nfs4_stateowner **sopp,
2953 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2955 struct nfs4_stateid *stp;
2956 struct nfs4_stateowner *sop;
2957 struct svc_fh *current_fh = &cstate->current_fh;
2960 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2961 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2962 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2963 stateid->si_generation);
2968 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2969 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2970 return nfserr_bad_stateid;
2973 if (STALE_STATEID(stateid))
2974 return nfserr_stale_stateid;
2976 if (nfsd4_has_session(cstate))
2977 flags |= HAS_SESSION;
2980 * We return BAD_STATEID if filehandle doesn't match stateid,
2981 * the confirmed flag is incorrecly set, or the generation
2982 * number is incorrect.
2984 stp = find_stateid(stateid, flags);
2987 * Also, we should make sure this isn't just the result of
2990 sop = search_close_lru(stateid->si_stateownerid, flags);
2992 return stateid_error_map(stateid);
2998 *sopp = sop = stp->st_stateowner;
3001 clientid_t *lockclid = &lock->v.new.clientid;
3002 struct nfs4_client *clp = sop->so_client;
3006 lkflg = setlkflg(lock->lk_type);
3008 if (lock->lk_is_new) {
3009 if (!sop->so_is_open_owner)
3010 return nfserr_bad_stateid;
3011 if (!(flags & HAS_SESSION) &&
3012 !same_clid(&clp->cl_clientid, lockclid))
3013 return nfserr_bad_stateid;
3014 /* stp is the open stateid */
3015 status = nfs4_check_openmode(stp, lkflg);
3019 /* stp is the lock stateid */
3020 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3026 if (nfs4_check_fh(current_fh, stp)) {
3027 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3028 return nfserr_bad_stateid;
3032 * We now validate the seqid and stateid generation numbers.
3033 * For the moment, we ignore the possibility of
3034 * generation number wraparound.
3036 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3039 if (sop->so_confirmed && flags & CONFIRM) {
3040 dprintk("NFSD: preprocess_seqid_op: expected"
3041 " unconfirmed stateowner!\n");
3042 return nfserr_bad_stateid;
3044 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3045 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3046 " confirmed yet!\n");
3047 return nfserr_bad_stateid;
3049 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3052 renew_client(sop->so_client);
3056 if (seqid == sop->so_seqid - 1) {
3057 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3058 /* indicate replay to calling function */
3059 return nfserr_replay_me;
3061 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3062 sop->so_seqid, seqid);
3064 return nfserr_bad_seqid;
3068 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3069 struct nfsd4_open_confirm *oc)
3072 struct nfs4_stateowner *sop;
3073 struct nfs4_stateid *stp;
3075 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3076 (int)cstate->current_fh.fh_dentry->d_name.len,
3077 cstate->current_fh.fh_dentry->d_name.name);
3079 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3085 if ((status = nfs4_preprocess_seqid_op(cstate,
3086 oc->oc_seqid, &oc->oc_req_stateid,
3087 CONFIRM | OPEN_STATE,
3088 &oc->oc_stateowner, &stp, NULL)))
3091 sop = oc->oc_stateowner;
3092 sop->so_confirmed = 1;
3093 update_stateid(&stp->st_stateid);
3094 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3095 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3096 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3097 stp->st_stateid.si_boot,
3098 stp->st_stateid.si_stateownerid,
3099 stp->st_stateid.si_fileid,
3100 stp->st_stateid.si_generation);
3102 nfsd4_create_clid_dir(sop->so_client);
3104 if (oc->oc_stateowner) {
3105 nfs4_get_stateowner(oc->oc_stateowner);
3106 cstate->replay_owner = oc->oc_stateowner;
3108 nfs4_unlock_state();
3114 * unset all bits in union bitmap (bmap) that
3115 * do not exist in share (from successful OPEN_DOWNGRADE)
3118 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3121 for (i = 1; i < 4; i++) {
3122 if ((i & access) != i)
3123 __clear_bit(i, bmap);
3128 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3131 for (i = 0; i < 4; i++) {
3132 if ((i & deny) != i)
3133 __clear_bit(i, bmap);
3138 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3139 struct nfsd4_compound_state *cstate,
3140 struct nfsd4_open_downgrade *od)
3143 struct nfs4_stateid *stp;
3144 unsigned int share_access;
3146 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3147 (int)cstate->current_fh.fh_dentry->d_name.len,
3148 cstate->current_fh.fh_dentry->d_name.name);
3150 if (!access_valid(od->od_share_access, cstate->minorversion)
3151 || !deny_valid(od->od_share_deny))
3152 return nfserr_inval;
3155 if ((status = nfs4_preprocess_seqid_op(cstate,
3159 &od->od_stateowner, &stp, NULL)))
3162 status = nfserr_inval;
3163 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3164 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3165 stp->st_access_bmap, od->od_share_access);
3168 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3169 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3170 stp->st_deny_bmap, od->od_share_deny);
3173 set_access(&share_access, stp->st_access_bmap);
3174 nfs4_file_downgrade(stp->st_vfs_file,
3175 share_access & ~od->od_share_access);
3177 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3178 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3180 update_stateid(&stp->st_stateid);
3181 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3184 if (od->od_stateowner) {
3185 nfs4_get_stateowner(od->od_stateowner);
3186 cstate->replay_owner = od->od_stateowner;
3188 nfs4_unlock_state();
3193 * nfs4_unlock_state() called after encode
3196 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3197 struct nfsd4_close *close)
3200 struct nfs4_stateid *stp;
3202 dprintk("NFSD: nfsd4_close on file %.*s\n",
3203 (int)cstate->current_fh.fh_dentry->d_name.len,
3204 cstate->current_fh.fh_dentry->d_name.name);
3207 /* check close_lru for replay */
3208 if ((status = nfs4_preprocess_seqid_op(cstate,
3211 OPEN_STATE | CLOSE_STATE,
3212 &close->cl_stateowner, &stp, NULL)))
3215 update_stateid(&stp->st_stateid);
3216 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3218 /* release_stateid() calls nfsd_close() if needed */
3219 release_open_stateid(stp);
3221 /* place unused nfs4_stateowners on so_close_lru list to be
3222 * released by the laundromat service after the lease period
3223 * to enable us to handle CLOSE replay
3225 if (list_empty(&close->cl_stateowner->so_stateids))
3226 move_to_close_lru(close->cl_stateowner);
3228 if (close->cl_stateowner) {
3229 nfs4_get_stateowner(close->cl_stateowner);
3230 cstate->replay_owner = close->cl_stateowner;
3232 nfs4_unlock_state();
3237 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3238 struct nfsd4_delegreturn *dr)
3240 struct nfs4_delegation *dp;
3241 stateid_t *stateid = &dr->dr_stateid;
3242 struct inode *inode;
3246 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3248 inode = cstate->current_fh.fh_dentry->d_inode;
3250 if (nfsd4_has_session(cstate))
3251 flags |= HAS_SESSION;
3253 status = nfserr_bad_stateid;
3254 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3256 status = nfserr_stale_stateid;
3257 if (STALE_STATEID(stateid))
3259 status = nfserr_bad_stateid;
3260 if (!is_delegation_stateid(stateid))
3262 dp = find_delegation_stateid(inode, stateid);
3264 status = stateid_error_map(stateid);
3267 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3270 renew_client(dp->dl_client);
3272 unhash_delegation(dp);
3274 nfs4_unlock_state();
3281 * Lock owner state (byte-range locks)
3283 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3284 #define LOCK_HASH_BITS 8
3285 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3286 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3289 end_offset(u64 start, u64 len)
3294 return end >= start ? end: NFS4_MAX_UINT64;
3297 /* last octet in a range */
3299 last_byte_offset(u64 start, u64 len)
3305 return end > start ? end - 1: NFS4_MAX_UINT64;
3308 #define lockownerid_hashval(id) \
3309 ((id) & LOCK_HASH_MASK)
3311 static inline unsigned int
3312 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3313 struct xdr_netobj *ownername)
3315 return (file_hashval(inode) + cl_id
3316 + opaque_hashval(ownername->data, ownername->len))
3320 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3321 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3322 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3324 static struct nfs4_stateid *
3325 find_stateid(stateid_t *stid, int flags)
3327 struct nfs4_stateid *local;
3328 u32 st_id = stid->si_stateownerid;
3329 u32 f_id = stid->si_fileid;
3330 unsigned int hashval;
3332 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3333 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3334 hashval = stateid_hashval(st_id, f_id);
3335 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3336 if ((local->st_stateid.si_stateownerid == st_id) &&
3337 (local->st_stateid.si_fileid == f_id))
3342 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3343 hashval = stateid_hashval(st_id, f_id);
3344 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3345 if ((local->st_stateid.si_stateownerid == st_id) &&
3346 (local->st_stateid.si_fileid == f_id))
3353 static struct nfs4_delegation *
3354 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3356 struct nfs4_file *fp;
3357 struct nfs4_delegation *dl;
3359 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3360 stid->si_boot, stid->si_stateownerid,
3361 stid->si_fileid, stid->si_generation);
3363 fp = find_file(ino);
3366 dl = find_delegation_file(fp, stid);
3372 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3373 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3374 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3375 * locking, this prevents us from being completely protocol-compliant. The
3376 * real solution to this problem is to start using unsigned file offsets in
3377 * the VFS, but this is a very deep change!
3380 nfs4_transform_lock_offset(struct file_lock *lock)
3382 if (lock->fl_start < 0)
3383 lock->fl_start = OFFSET_MAX;
3384 if (lock->fl_end < 0)
3385 lock->fl_end = OFFSET_MAX;
3388 /* Hack!: For now, we're defining this just so we can use a pointer to it
3389 * as a unique cookie to identify our (NFSv4's) posix locks. */
3390 static struct lock_manager_operations nfsd_posix_mng_ops = {
3394 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3396 struct nfs4_stateowner *sop;
3399 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3400 sop = (struct nfs4_stateowner *) fl->fl_owner;
3401 hval = lockownerid_hashval(sop->so_id);
3402 kref_get(&sop->so_ref);
3404 deny->ld_clientid = sop->so_client->cl_clientid;
3406 deny->ld_sop = NULL;
3407 deny->ld_clientid.cl_boot = 0;
3408 deny->ld_clientid.cl_id = 0;
3410 deny->ld_start = fl->fl_start;
3411 deny->ld_length = NFS4_MAX_UINT64;
3412 if (fl->fl_end != NFS4_MAX_UINT64)
3413 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3414 deny->ld_type = NFS4_READ_LT;
3415 if (fl->fl_type != F_RDLCK)
3416 deny->ld_type = NFS4_WRITE_LT;
3419 static struct nfs4_stateowner *
3420 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3421 struct xdr_netobj *owner)
3423 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3424 struct nfs4_stateowner *op;
3426 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3427 if (same_owner_str(op, owner, clid))
3434 * Alloc a lock owner structure.
3435 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3438 * strhashval = lock_ownerstr_hashval
3441 static struct nfs4_stateowner *
3442 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3443 struct nfs4_stateowner *sop;
3444 struct nfs4_replay *rp;
3445 unsigned int idhashval;
3447 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3449 idhashval = lockownerid_hashval(current_ownerid);
3450 INIT_LIST_HEAD(&sop->so_idhash);
3451 INIT_LIST_HEAD(&sop->so_strhash);
3452 INIT_LIST_HEAD(&sop->so_perclient);
3453 INIT_LIST_HEAD(&sop->so_stateids);
3454 INIT_LIST_HEAD(&sop->so_perstateid);
3455 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3457 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3458 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3459 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3460 sop->so_is_open_owner = 0;
3461 sop->so_id = current_ownerid++;
3462 sop->so_client = clp;
3463 /* It is the openowner seqid that will be incremented in encode in the
3464 * case of new lockowners; so increment the lock seqid manually: */
3465 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3466 sop->so_confirmed = 1;
3467 rp = &sop->so_replay;
3468 rp->rp_status = nfserr_serverfault;
3470 rp->rp_buf = rp->rp_ibuf;
3474 static struct nfs4_stateid *
3475 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3477 struct nfs4_stateid *stp;
3478 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3480 stp = nfs4_alloc_stateid();
3483 INIT_LIST_HEAD(&stp->st_hash);
3484 INIT_LIST_HEAD(&stp->st_perfile);
3485 INIT_LIST_HEAD(&stp->st_perstateowner);
3486 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3487 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3488 list_add(&stp->st_perfile, &fp->fi_stateids);
3489 list_add(&stp->st_perstateowner, &sop->so_stateids);
3490 stp->st_stateowner = sop;
3493 stp->st_stateid.si_boot = get_seconds();
3494 stp->st_stateid.si_stateownerid = sop->so_id;
3495 stp->st_stateid.si_fileid = fp->fi_id;
3496 stp->st_stateid.si_generation = 0;
3497 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3498 stp->st_access_bmap = open_stp->st_access_bmap;
3499 stp->st_deny_bmap = open_stp->st_deny_bmap;
3500 stp->st_openstp = open_stp;
3507 check_lock_length(u64 offset, u64 length)
3509 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3510 LOFF_OVERFLOW(offset, length)));
3517 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3518 struct nfsd4_lock *lock)
3520 struct nfs4_stateowner *open_sop = NULL;
3521 struct nfs4_stateowner *lock_sop = NULL;
3522 struct nfs4_stateid *lock_stp;
3524 struct file_lock file_lock;
3525 struct file_lock conflock;
3527 unsigned int strhashval;
3531 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3532 (long long) lock->lk_offset,
3533 (long long) lock->lk_length);
3535 if (check_lock_length(lock->lk_offset, lock->lk_length))
3536 return nfserr_inval;
3538 if ((status = fh_verify(rqstp, &cstate->current_fh,
3539 S_IFREG, NFSD_MAY_LOCK))) {
3540 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3546 if (lock->lk_is_new) {
3548 * Client indicates that this is a new lockowner.
3549 * Use open owner and open stateid to create lock owner and
3552 struct nfs4_stateid *open_stp = NULL;
3553 struct nfs4_file *fp;
3555 status = nfserr_stale_clientid;
3556 if (!nfsd4_has_session(cstate) &&
3557 STALE_CLIENTID(&lock->lk_new_clientid))
3560 /* validate and update open stateid and open seqid */
3561 status = nfs4_preprocess_seqid_op(cstate,
3562 lock->lk_new_open_seqid,
3563 &lock->lk_new_open_stateid,
3565 &lock->lk_replay_owner, &open_stp,
3569 open_sop = lock->lk_replay_owner;
3570 /* create lockowner and lock stateid */
3571 fp = open_stp->st_file;
3572 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3573 open_sop->so_client->cl_clientid.cl_id,
3574 &lock->v.new.owner);
3575 /* XXX: Do we need to check for duplicate stateowners on
3576 * the same file, or should they just be allowed (and
3577 * create new stateids)? */
3578 status = nfserr_resource;
3579 lock_sop = alloc_init_lock_stateowner(strhashval,
3580 open_sop->so_client, open_stp, lock);
3581 if (lock_sop == NULL)
3583 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3584 if (lock_stp == NULL)
3587 /* lock (lock owner + lock stateid) already exists */
3588 status = nfs4_preprocess_seqid_op(cstate,
3589 lock->lk_old_lock_seqid,
3590 &lock->lk_old_lock_stateid,
3592 &lock->lk_replay_owner, &lock_stp, lock);
3595 lock_sop = lock->lk_replay_owner;
3597 /* lock->lk_replay_owner and lock_stp have been created or found */
3598 filp = lock_stp->st_vfs_file;
3600 status = nfserr_grace;
3601 if (locks_in_grace() && !lock->lk_reclaim)
3603 status = nfserr_no_grace;
3604 if (!locks_in_grace() && lock->lk_reclaim)
3607 locks_init_lock(&file_lock);
3608 switch (lock->lk_type) {
3611 file_lock.fl_type = F_RDLCK;
3615 case NFS4_WRITEW_LT:
3616 file_lock.fl_type = F_WRLCK;
3620 status = nfserr_inval;
3623 file_lock.fl_owner = (fl_owner_t)lock_sop;
3624 file_lock.fl_pid = current->tgid;
3625 file_lock.fl_file = filp;
3626 file_lock.fl_flags = FL_POSIX;
3627 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3629 file_lock.fl_start = lock->lk_offset;
3630 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3631 nfs4_transform_lock_offset(&file_lock);
3634 * Try to lock the file in the VFS.
3635 * Note: locks.c uses the BKL to protect the inode's lock list.
3638 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3640 case 0: /* success! */
3641 update_stateid(&lock_stp->st_stateid);
3642 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3646 case (EAGAIN): /* conflock holds conflicting lock */
3647 status = nfserr_denied;
3648 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3649 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3652 status = nfserr_deadlock;
3655 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3656 status = nfserr_resource;
3660 if (status && lock->lk_is_new && lock_sop)
3661 release_lockowner(lock_sop);
3662 if (lock->lk_replay_owner) {
3663 nfs4_get_stateowner(lock->lk_replay_owner);
3664 cstate->replay_owner = lock->lk_replay_owner;
3666 nfs4_unlock_state();
3671 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3672 * so we do a temporary open here just to get an open file to pass to
3673 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3676 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3681 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3684 err = vfs_test_lock(file, lock);
3693 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3694 struct nfsd4_lockt *lockt)
3696 struct inode *inode;
3697 struct file_lock file_lock;
3701 if (locks_in_grace())
3702 return nfserr_grace;
3704 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3705 return nfserr_inval;
3707 lockt->lt_stateowner = NULL;
3710 status = nfserr_stale_clientid;
3711 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3714 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3715 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3716 if (status == nfserr_symlink)
3717 status = nfserr_inval;
3721 inode = cstate->current_fh.fh_dentry->d_inode;
3722 locks_init_lock(&file_lock);
3723 switch (lockt->lt_type) {
3726 file_lock.fl_type = F_RDLCK;
3729 case NFS4_WRITEW_LT:
3730 file_lock.fl_type = F_WRLCK;
3733 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3734 status = nfserr_inval;
3738 lockt->lt_stateowner = find_lockstateowner_str(inode,
3739 &lockt->lt_clientid, &lockt->lt_owner);
3740 if (lockt->lt_stateowner)
3741 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3742 file_lock.fl_pid = current->tgid;
3743 file_lock.fl_flags = FL_POSIX;
3745 file_lock.fl_start = lockt->lt_offset;
3746 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3748 nfs4_transform_lock_offset(&file_lock);
3751 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3753 status = nfserrno(error);
3756 if (file_lock.fl_type != F_UNLCK) {
3757 status = nfserr_denied;
3758 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3761 nfs4_unlock_state();
3766 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3767 struct nfsd4_locku *locku)
3769 struct nfs4_stateid *stp;
3770 struct file *filp = NULL;
3771 struct file_lock file_lock;
3775 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3776 (long long) locku->lu_offset,
3777 (long long) locku->lu_length);
3779 if (check_lock_length(locku->lu_offset, locku->lu_length))
3780 return nfserr_inval;
3784 if ((status = nfs4_preprocess_seqid_op(cstate,
3788 &locku->lu_stateowner, &stp, NULL)))
3791 filp = stp->st_vfs_file;
3793 locks_init_lock(&file_lock);
3794 file_lock.fl_type = F_UNLCK;
3795 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3796 file_lock.fl_pid = current->tgid;
3797 file_lock.fl_file = filp;
3798 file_lock.fl_flags = FL_POSIX;
3799 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3800 file_lock.fl_start = locku->lu_offset;
3802 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3803 nfs4_transform_lock_offset(&file_lock);
3806 * Try to unlock the file in the VFS.
3808 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3810 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3814 * OK, unlock succeeded; the only thing left to do is update the stateid.
3816 update_stateid(&stp->st_stateid);
3817 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3820 if (locku->lu_stateowner) {
3821 nfs4_get_stateowner(locku->lu_stateowner);
3822 cstate->replay_owner = locku->lu_stateowner;
3824 nfs4_unlock_state();
3828 status = nfserrno(err);
3834 * 1: locks held by lockowner
3835 * 0: no locks held by lockowner
3838 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3840 struct file_lock **flpp;
3841 struct inode *inode = filp->f_path.dentry->d_inode;
3845 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3846 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3857 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3858 struct nfsd4_compound_state *cstate,
3859 struct nfsd4_release_lockowner *rlockowner)
3861 clientid_t *clid = &rlockowner->rl_clientid;
3862 struct nfs4_stateowner *sop;
3863 struct nfs4_stateid *stp;
3864 struct xdr_netobj *owner = &rlockowner->rl_owner;
3865 struct list_head matches;
3869 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3870 clid->cl_boot, clid->cl_id);
3872 /* XXX check for lease expiration */
3874 status = nfserr_stale_clientid;
3875 if (STALE_CLIENTID(clid))
3880 status = nfserr_locks_held;
3881 /* XXX: we're doing a linear search through all the lockowners.
3882 * Yipes! For now we'll just hope clients aren't really using
3883 * release_lockowner much, but eventually we have to fix these
3884 * data structures. */
3885 INIT_LIST_HEAD(&matches);
3886 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3887 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3888 if (!same_owner_str(sop, owner, clid))
3890 list_for_each_entry(stp, &sop->so_stateids,
3892 if (check_for_locks(stp->st_vfs_file, sop))
3894 /* Note: so_perclient unused for lockowners,
3895 * so it's OK to fool with here. */
3896 list_add(&sop->so_perclient, &matches);
3900 /* Clients probably won't expect us to return with some (but not all)
3901 * of the lockowner state released; so don't release any until all
3902 * have been checked. */
3904 while (!list_empty(&matches)) {
3905 sop = list_entry(matches.next, struct nfs4_stateowner,
3907 /* unhash_stateowner deletes so_perclient only
3908 * for openowners. */
3909 list_del(&sop->so_perclient);
3910 release_lockowner(sop);
3913 nfs4_unlock_state();
3917 static inline struct nfs4_client_reclaim *
3920 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3924 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3926 unsigned int strhashval = clientstr_hashval(name);
3927 struct nfs4_client *clp;
3929 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3934 * failure => all reset bets are off, nfserr_no_grace...
3937 nfs4_client_to_reclaim(const char *name)
3939 unsigned int strhashval;
3940 struct nfs4_client_reclaim *crp = NULL;
3942 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3943 crp = alloc_reclaim();
3946 strhashval = clientstr_hashval(name);
3947 INIT_LIST_HEAD(&crp->cr_strhash);
3948 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3949 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3950 reclaim_str_hashtbl_size++;
3955 nfs4_release_reclaim(void)
3957 struct nfs4_client_reclaim *crp = NULL;
3960 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3961 while (!list_empty(&reclaim_str_hashtbl[i])) {
3962 crp = list_entry(reclaim_str_hashtbl[i].next,
3963 struct nfs4_client_reclaim, cr_strhash);
3964 list_del(&crp->cr_strhash);
3966 reclaim_str_hashtbl_size--;
3969 BUG_ON(reclaim_str_hashtbl_size);
3973 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3974 static struct nfs4_client_reclaim *
3975 nfs4_find_reclaim_client(clientid_t *clid)
3977 unsigned int strhashval;
3978 struct nfs4_client *clp;
3979 struct nfs4_client_reclaim *crp = NULL;
3982 /* find clientid in conf_id_hashtbl */
3983 clp = find_confirmed_client(clid);
3987 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3988 clp->cl_name.len, clp->cl_name.data,
3991 /* find clp->cl_name in reclaim_str_hashtbl */
3992 strhashval = clientstr_hashval(clp->cl_recdir);
3993 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3994 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4002 * Called from OPEN. Look for clientid in reclaim list.
4005 nfs4_check_open_reclaim(clientid_t *clid)
4007 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4010 /* initialization to perform at module load time: */
4013 nfs4_state_init(void)
4017 status = nfsd4_init_slabs();
4020 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4021 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4022 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4023 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4024 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4025 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4027 for (i = 0; i < SESSION_HASH_SIZE; i++)
4028 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4029 for (i = 0; i < FILE_HASH_SIZE; i++) {
4030 INIT_LIST_HEAD(&file_hashtbl[i]);
4032 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4033 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4034 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4036 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4037 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4038 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4040 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4041 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4042 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4044 memset(&onestateid, ~0, sizeof(stateid_t));
4045 INIT_LIST_HEAD(&close_lru);
4046 INIT_LIST_HEAD(&client_lru);
4047 INIT_LIST_HEAD(&del_recall_lru);
4048 reclaim_str_hashtbl_size = 0;
4053 nfsd4_load_reboot_recovery_data(void)
4058 nfsd4_init_recdir(user_recovery_dirname);
4059 status = nfsd4_recdir_load();
4060 nfs4_unlock_state();
4062 printk("NFSD: Failure reading reboot recovery data\n");
4066 get_nfs4_grace_period(void)
4068 return max(user_lease_time, lease_time) * HZ;
4072 * Since the lifetime of a delegation isn't limited to that of an open, a
4073 * client may quite reasonably hang on to a delegation as long as it has
4074 * the inode cached. This becomes an obvious problem the first time a
4075 * client's inode cache approaches the size of the server's total memory.
4077 * For now we avoid this problem by imposing a hard limit on the number
4078 * of delegations, which varies according to the server's memory size.
4081 set_max_delegations(void)
4084 * Allow at most 4 delegations per megabyte of RAM. Quick
4085 * estimates suggest that in the worst case (where every delegation
4086 * is for a different inode), a delegation could take about 1.5K,
4087 * giving a worst case usage of about 6% of memory.
4089 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4092 /* initialization to perform when the nfsd service is started: */
4095 __nfs4_state_start(void)
4097 unsigned long grace_time;
4099 boot_time = get_seconds();
4100 grace_time = get_nfs4_grace_period();
4101 lease_time = user_lease_time;
4102 locks_start_grace(&nfsd4_manager);
4103 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4105 laundry_wq = create_singlethread_workqueue("nfsd4");
4106 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4107 set_max_delegations();
4111 nfs4_state_start(void)
4115 nfsd4_load_reboot_recovery_data();
4116 __nfs4_state_start();
4122 nfs4_lease_time(void)
4128 __nfs4_state_shutdown(void)
4131 struct nfs4_client *clp = NULL;
4132 struct nfs4_delegation *dp = NULL;
4133 struct list_head *pos, *next, reaplist;
4135 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4136 while (!list_empty(&conf_id_hashtbl[i])) {
4137 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4140 while (!list_empty(&unconf_str_hashtbl[i])) {
4141 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4145 INIT_LIST_HEAD(&reaplist);
4146 spin_lock(&recall_lock);
4147 list_for_each_safe(pos, next, &del_recall_lru) {
4148 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4149 list_move(&dp->dl_recall_lru, &reaplist);
4151 spin_unlock(&recall_lock);
4152 list_for_each_safe(pos, next, &reaplist) {
4153 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4154 list_del_init(&dp->dl_recall_lru);
4155 unhash_delegation(dp);
4158 nfsd4_shutdown_recdir();
4163 nfs4_state_shutdown(void)
4165 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4166 destroy_workqueue(laundry_wq);
4167 locks_end_grace(&nfsd4_manager);
4169 nfs4_release_reclaim();
4170 __nfs4_state_shutdown();
4171 nfs4_unlock_state();
4175 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4176 * accessed when nfsd is starting.
4179 nfs4_set_recdir(char *recdir)
4181 strcpy(user_recovery_dirname, recdir);
4185 * Change the NFSv4 recovery directory to recdir.
4188 nfs4_reset_recoverydir(char *recdir)
4193 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4197 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4198 nfs4_set_recdir(recdir);
4206 nfs4_recoverydir(void)
4208 return user_recovery_dirname;
4212 * Called when leasetime is changed.
4214 * The only way the protocol gives us to handle on-the-fly lease changes is to
4215 * simulate a reboot. Instead of doing that, we just wait till the next time
4216 * we start to register any changes in lease time. If the administrator
4217 * really wants to change the lease time *now*, they can go ahead and bring
4218 * nfsd down and then back up again after changing the lease time.
4220 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4221 * when nfsd is starting
4224 nfs4_reset_lease(time_t leasetime)
4226 user_lease_time = leasetime;