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 * Give the client the number of slots it requests bound by
418 * NFSD_MAX_SLOTS_PER_SESSION and by nfsd_drc_max_mem.
420 * If we run out of reserved DRC memory we should (up to a point) re-negotiate
421 * active sessions and reduce their slot usage to make rooom for new
422 * connections. For now we just fail the create session.
424 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
428 if (fchan->maxreqs < 1)
430 else if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
431 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
433 mem = fchan->maxreqs * NFSD_SLOT_CACHE_SIZE;
435 spin_lock(&nfsd_drc_lock);
436 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem)
437 mem = ((nfsd_drc_max_mem - nfsd_drc_mem_used) /
438 NFSD_SLOT_CACHE_SIZE) * NFSD_SLOT_CACHE_SIZE;
439 nfsd_drc_mem_used += mem;
440 spin_unlock(&nfsd_drc_lock);
442 fchan->maxreqs = mem / NFSD_SLOT_CACHE_SIZE;
443 if (fchan->maxreqs == 0)
444 return nfserr_resource;
449 * fchan holds the client values on input, and the server values on output
451 static int init_forechannel_attrs(struct svc_rqst *rqstp,
452 struct nfsd4_channel_attrs *session_fchan,
453 struct nfsd4_channel_attrs *fchan)
456 __u32 maxcount = svc_max_payload(rqstp);
458 /* headerpadsz set to zero in encode routine */
460 /* Use the client's max request and max response size if possible */
461 if (fchan->maxreq_sz > maxcount)
462 fchan->maxreq_sz = maxcount;
463 session_fchan->maxreq_sz = fchan->maxreq_sz;
465 if (fchan->maxresp_sz > maxcount)
466 fchan->maxresp_sz = maxcount;
467 session_fchan->maxresp_sz = fchan->maxresp_sz;
469 session_fchan->maxresp_cached = NFSD_SLOT_CACHE_SIZE;
470 fchan->maxresp_cached = session_fchan->maxresp_cached;
472 /* Use the client's maxops if possible */
473 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
474 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
475 session_fchan->maxops = fchan->maxops;
477 /* FIXME: Error means no more DRC pages so the server should
478 * recover pages from existing sessions. For now fail session
481 status = set_forechannel_maxreqs(fchan);
483 session_fchan->maxreqs = fchan->maxreqs;
488 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
489 struct nfsd4_create_session *cses)
491 struct nfsd4_session *new, tmp;
492 int idx, status = nfserr_resource, slotsize;
494 memset(&tmp, 0, sizeof(tmp));
496 /* FIXME: For now, we just accept the client back channel attributes. */
497 tmp.se_bchannel = cses->back_channel;
498 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
499 &cses->fore_channel);
503 /* allocate struct nfsd4_session and slot table in one piece */
504 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot);
505 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
509 memcpy(new, &tmp, sizeof(*new));
511 new->se_client = clp;
513 idx = hash_sessionid(&new->se_sessionid);
514 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
515 NFS4_MAX_SESSIONID_LEN);
517 new->se_flags = cses->flags;
518 kref_init(&new->se_ref);
519 spin_lock(&sessionid_lock);
520 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
521 list_add(&new->se_perclnt, &clp->cl_sessions);
522 spin_unlock(&sessionid_lock);
529 /* caller must hold sessionid_lock */
530 static struct nfsd4_session *
531 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
533 struct nfsd4_session *elem;
536 dump_sessionid(__func__, sessionid);
537 idx = hash_sessionid(sessionid);
538 dprintk("%s: idx is %d\n", __func__, idx);
539 /* Search in the appropriate list */
540 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
541 dump_sessionid("list traversal", &elem->se_sessionid);
542 if (!memcmp(elem->se_sessionid.data, sessionid->data,
543 NFS4_MAX_SESSIONID_LEN)) {
548 dprintk("%s: session not found\n", __func__);
552 /* caller must hold sessionid_lock */
554 unhash_session(struct nfsd4_session *ses)
556 list_del(&ses->se_hash);
557 list_del(&ses->se_perclnt);
561 release_session(struct nfsd4_session *ses)
563 spin_lock(&sessionid_lock);
565 spin_unlock(&sessionid_lock);
566 nfsd4_put_session(ses);
569 static void nfsd4_release_respages(struct page **respages, short resused);
572 free_session(struct kref *kref)
574 struct nfsd4_session *ses;
577 ses = container_of(kref, struct nfsd4_session, se_ref);
578 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
579 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
580 nfsd4_release_respages(e->ce_respages, e->ce_resused);
582 spin_lock(&nfsd_drc_lock);
583 nfsd_drc_mem_used -= ses->se_fchannel.maxreqs * NFSD_SLOT_CACHE_SIZE;
584 spin_unlock(&nfsd_drc_lock);
589 renew_client(struct nfs4_client *clp)
592 * Move client to the end to the LRU list.
594 dprintk("renewing client (clientid %08x/%08x)\n",
595 clp->cl_clientid.cl_boot,
596 clp->cl_clientid.cl_id);
597 list_move_tail(&clp->cl_lru, &client_lru);
598 clp->cl_time = get_seconds();
601 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
603 STALE_CLIENTID(clientid_t *clid)
605 if (clid->cl_boot == boot_time)
607 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
608 clid->cl_boot, clid->cl_id, boot_time);
613 * XXX Should we use a slab cache ?
614 * This type of memory management is somewhat inefficient, but we use it
615 * anyway since SETCLIENTID is not a common operation.
617 static struct nfs4_client *alloc_client(struct xdr_netobj name)
619 struct nfs4_client *clp;
621 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
624 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
625 if (clp->cl_name.data == NULL) {
629 memcpy(clp->cl_name.data, name.data, name.len);
630 clp->cl_name.len = name.len;
635 shutdown_callback_client(struct nfs4_client *clp)
637 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
641 * Callback threads take a reference on the client, so there
642 * should be no outstanding callbacks at this point.
644 clp->cl_cb_conn.cb_client = NULL;
645 rpc_shutdown_client(clnt);
647 if (clp->cl_cb_conn.cb_cred) {
648 put_rpccred(clp->cl_cb_conn.cb_cred);
649 clp->cl_cb_conn.cb_cred = NULL;
654 free_client(struct nfs4_client *clp)
656 shutdown_callback_client(clp);
657 if (clp->cl_cred.cr_group_info)
658 put_group_info(clp->cl_cred.cr_group_info);
659 kfree(clp->cl_principal);
660 kfree(clp->cl_name.data);
665 put_nfs4_client(struct nfs4_client *clp)
667 if (atomic_dec_and_test(&clp->cl_count))
672 expire_client(struct nfs4_client *clp)
674 struct nfs4_stateowner *sop;
675 struct nfs4_delegation *dp;
676 struct list_head reaplist;
678 dprintk("NFSD: expire_client cl_count %d\n",
679 atomic_read(&clp->cl_count));
681 INIT_LIST_HEAD(&reaplist);
682 spin_lock(&recall_lock);
683 while (!list_empty(&clp->cl_delegations)) {
684 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
685 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
687 list_del_init(&dp->dl_perclnt);
688 list_move(&dp->dl_recall_lru, &reaplist);
690 spin_unlock(&recall_lock);
691 while (!list_empty(&reaplist)) {
692 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
693 list_del_init(&dp->dl_recall_lru);
694 unhash_delegation(dp);
696 list_del(&clp->cl_idhash);
697 list_del(&clp->cl_strhash);
698 list_del(&clp->cl_lru);
699 while (!list_empty(&clp->cl_openowners)) {
700 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
701 release_openowner(sop);
703 while (!list_empty(&clp->cl_sessions)) {
704 struct nfsd4_session *ses;
705 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
707 release_session(ses);
709 put_nfs4_client(clp);
712 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
714 struct nfs4_client *clp;
716 clp = alloc_client(name);
719 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
720 atomic_set(&clp->cl_count, 1);
721 atomic_set(&clp->cl_cb_conn.cb_set, 0);
722 INIT_LIST_HEAD(&clp->cl_idhash);
723 INIT_LIST_HEAD(&clp->cl_strhash);
724 INIT_LIST_HEAD(&clp->cl_openowners);
725 INIT_LIST_HEAD(&clp->cl_delegations);
726 INIT_LIST_HEAD(&clp->cl_sessions);
727 INIT_LIST_HEAD(&clp->cl_lru);
731 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
733 memcpy(target->cl_verifier.data, source->data,
734 sizeof(target->cl_verifier.data));
737 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
739 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
740 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
743 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
745 target->cr_uid = source->cr_uid;
746 target->cr_gid = source->cr_gid;
747 target->cr_group_info = source->cr_group_info;
748 get_group_info(target->cr_group_info);
751 static int same_name(const char *n1, const char *n2)
753 return 0 == memcmp(n1, n2, HEXDIR_LEN);
757 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
759 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
763 same_clid(clientid_t *cl1, clientid_t *cl2)
765 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
768 /* XXX what about NGROUP */
770 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
772 return cr1->cr_uid == cr2->cr_uid;
775 static void gen_clid(struct nfs4_client *clp)
777 static u32 current_clientid = 1;
779 clp->cl_clientid.cl_boot = boot_time;
780 clp->cl_clientid.cl_id = current_clientid++;
783 static void gen_confirm(struct nfs4_client *clp)
788 p = (u32 *)clp->cl_confirm.data;
789 *p++ = get_seconds();
793 static int check_name(struct xdr_netobj name)
797 if (name.len > NFS4_OPAQUE_LIMIT) {
798 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
805 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
807 unsigned int idhashval;
809 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
810 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
811 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
812 list_add_tail(&clp->cl_lru, &client_lru);
813 clp->cl_time = get_seconds();
817 move_to_confirmed(struct nfs4_client *clp)
819 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
820 unsigned int strhashval;
822 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
823 list_del_init(&clp->cl_strhash);
824 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
825 strhashval = clientstr_hashval(clp->cl_recdir);
826 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
830 static struct nfs4_client *
831 find_confirmed_client(clientid_t *clid)
833 struct nfs4_client *clp;
834 unsigned int idhashval = clientid_hashval(clid->cl_id);
836 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
837 if (same_clid(&clp->cl_clientid, clid))
843 static struct nfs4_client *
844 find_unconfirmed_client(clientid_t *clid)
846 struct nfs4_client *clp;
847 unsigned int idhashval = clientid_hashval(clid->cl_id);
849 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
850 if (same_clid(&clp->cl_clientid, clid))
857 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
858 * parameter. Matching is based on the fact the at least one of the
859 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
861 * FIXME: we need to unify the clientid namespaces for nfsv4.x
862 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
863 * and SET_CLIENTID{,_CONFIRM}
866 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
868 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
869 return use_exchange_id == has_exchange_flags;
872 static struct nfs4_client *
873 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
874 bool use_exchange_id)
876 struct nfs4_client *clp;
878 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
879 if (same_name(clp->cl_recdir, dname) &&
880 match_clientid_establishment(clp, use_exchange_id))
886 static struct nfs4_client *
887 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
888 bool use_exchange_id)
890 struct nfs4_client *clp;
892 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
893 if (same_name(clp->cl_recdir, dname) &&
894 match_clientid_establishment(clp, use_exchange_id))
900 /* a helper function for parse_callback */
902 parse_octet(unsigned int *lenp, char **addrp)
904 unsigned int len = *lenp;
916 if ((c < '0') || (c > '9')) {
922 n = (n * 10) + (c - '0');
933 /* parse and set the setclientid ipv4 callback address */
935 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
940 u32 addrlen = addr_len;
941 char *addr = addr_val;
946 for(i = 4; i > 0 ; i--) {
947 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
950 cbaddr |= (temp << shift);
958 for(i = 2; i > 0 ; i--) {
959 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
962 cbport |= (temp << shift);
971 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
973 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
975 /* Currently, we only support tcp for the callback channel */
976 if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
979 if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
980 &cb->cb_addr, &cb->cb_port)))
982 cb->cb_minorversion = 0;
983 cb->cb_prog = se->se_callback_prog;
984 cb->cb_ident = se->se_callback_ident;
987 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
988 "will not receive delegations\n",
989 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
995 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
997 struct nfsd4_compoundres *resp = rqstp->rq_resp;
999 resp->cstate.statp = statp;
1003 * Dereference the result pages.
1006 nfsd4_release_respages(struct page **respages, short resused)
1010 dprintk("--> %s\n", __func__);
1011 for (i = 0; i < resused; i++) {
1014 put_page(respages[i]);
1020 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1024 for (i = 0; i < count; i++) {
1025 topages[i] = frompages[i];
1028 get_page(topages[i]);
1033 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1034 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1035 * length of the XDR response is less than se_fmaxresp_cached
1036 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1037 * of the reply (e.g. readdir).
1039 * Store the base and length of the rq_req.head[0] page
1040 * of the NFSv4.1 data, just past the rpc header.
1043 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1045 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1046 struct svc_rqst *rqstp = resp->rqstp;
1047 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1048 struct nfsd4_op *op = &args->ops[resp->opcnt];
1049 struct kvec *resv = &rqstp->rq_res.head[0];
1051 dprintk("--> %s entry %p\n", __func__, entry);
1053 /* Don't cache a failed OP_SEQUENCE. */
1054 if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1057 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1058 entry->ce_opcnt = resp->opcnt;
1059 entry->ce_status = resp->cstate.status;
1062 * Don't need a page to cache just the sequence operation - the slot
1066 if (nfsd4_not_cached(resp)) {
1067 entry->ce_resused = 0;
1068 entry->ce_rpchdrlen = 0;
1069 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1070 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1073 entry->ce_resused = rqstp->rq_resused;
1074 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1075 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1076 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1078 entry->ce_datav.iov_base = resp->cstate.statp;
1079 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1080 (char *)page_address(rqstp->rq_respages[0]));
1081 /* Current request rpc header length*/
1082 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1083 (char *)page_address(rqstp->rq_respages[0]);
1087 * We keep the rpc header, but take the nfs reply from the replycache.
1090 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1091 struct nfsd4_cache_entry *entry)
1093 struct svc_rqst *rqstp = resp->rqstp;
1094 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1097 /* Current request rpc header length*/
1098 len = (char *)resp->cstate.statp -
1099 (char *)page_address(rqstp->rq_respages[0]);
1100 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1101 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1102 entry->ce_datav.iov_len);
1105 /* copy the cached reply nfsd data past the current rpc header */
1106 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1107 entry->ce_datav.iov_len);
1108 resv->iov_len = len + entry->ce_datav.iov_len;
1113 * Encode the replay sequence operation from the slot values.
1114 * If cachethis is FALSE encode the uncached rep error on the next
1115 * operation which sets resp->p and increments resp->opcnt for
1116 * nfs4svc_encode_compoundres.
1120 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1121 struct nfsd4_compoundres *resp)
1123 struct nfsd4_op *op;
1124 struct nfsd4_slot *slot = resp->cstate.slot;
1126 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1127 resp->opcnt, resp->cstate.slot->sl_cache_entry.ce_cachethis);
1129 /* Encode the replayed sequence operation */
1130 op = &args->ops[resp->opcnt - 1];
1131 nfsd4_encode_operation(resp, op);
1133 /* Return nfserr_retry_uncached_rep in next operation. */
1134 if (args->opcnt > 1 && slot->sl_cache_entry.ce_cachethis == 0) {
1135 op = &args->ops[resp->opcnt++];
1136 op->status = nfserr_retry_uncached_rep;
1137 nfsd4_encode_operation(resp, op);
1143 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1144 * cached page. Replace any futher replay pages from the cache.
1147 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1148 struct nfsd4_sequence *seq)
1150 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1153 dprintk("--> %s entry %p\n", __func__, entry);
1156 * If this is just the sequence operation, we did not keep
1157 * a page in the cache entry because we can just use the
1158 * slot info stored in struct nfsd4_sequence that was checked
1159 * against the slot in nfsd4_sequence().
1161 * This occurs when seq->cachethis is FALSE, or when the client
1162 * session inactivity timer fires and a solo sequence operation
1163 * is sent (lease renewal).
1165 seq->maxslots = resp->cstate.session->se_fchannel.maxreqs;
1167 /* Either returns 0 or nfserr_retry_uncached */
1168 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1169 if (status == nfserr_retry_uncached_rep)
1172 if (!nfsd41_copy_replay_data(resp, entry)) {
1174 * Not enough room to use the replay rpc header, send the
1175 * cached header. Release all the allocated result pages.
1177 svc_free_res_pages(resp->rqstp);
1178 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1181 /* Release all but the first allocated result page */
1183 resp->rqstp->rq_resused--;
1184 svc_free_res_pages(resp->rqstp);
1186 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1187 &entry->ce_respages[1],
1188 entry->ce_resused - 1);
1191 resp->rqstp->rq_resused = entry->ce_resused;
1192 resp->opcnt = entry->ce_opcnt;
1193 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1194 status = entry->ce_status;
1200 * Set the exchange_id flags returned by the server.
1203 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1205 /* pNFS is not supported */
1206 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1208 /* Referrals are supported, Migration is not. */
1209 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1211 /* set the wire flags to return to client. */
1212 clid->flags = new->cl_exchange_flags;
1216 nfsd4_exchange_id(struct svc_rqst *rqstp,
1217 struct nfsd4_compound_state *cstate,
1218 struct nfsd4_exchange_id *exid)
1220 struct nfs4_client *unconf, *conf, *new;
1222 unsigned int strhashval;
1223 char dname[HEXDIR_LEN];
1224 char addr_str[INET6_ADDRSTRLEN];
1225 nfs4_verifier verf = exid->verifier;
1226 struct sockaddr *sa = svc_addr(rqstp);
1228 rpc_ntop(sa, addr_str, sizeof(addr_str));
1229 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1230 "ip_addr=%s flags %x, spa_how %d\n",
1231 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1232 addr_str, exid->flags, exid->spa_how);
1234 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1235 return nfserr_inval;
1237 /* Currently only support SP4_NONE */
1238 switch (exid->spa_how) {
1242 return nfserr_encr_alg_unsupp;
1244 BUG(); /* checked by xdr code */
1246 return nfserr_serverfault; /* no excuse :-/ */
1249 status = nfs4_make_rec_clidname(dname, &exid->clname);
1254 strhashval = clientstr_hashval(dname);
1259 conf = find_confirmed_client_by_str(dname, strhashval, true);
1261 if (!same_verf(&verf, &conf->cl_verifier)) {
1262 /* 18.35.4 case 8 */
1263 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1264 status = nfserr_not_same;
1267 /* Client reboot: destroy old state */
1268 expire_client(conf);
1271 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1272 /* 18.35.4 case 9 */
1273 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1274 status = nfserr_perm;
1277 expire_client(conf);
1281 * Set bit when the owner id and verifier map to an already
1282 * confirmed client id (18.35.3).
1284 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1287 * Falling into 18.35.4 case 2, possible router replay.
1288 * Leave confirmed record intact and return same result.
1290 copy_verf(conf, &verf);
1295 /* 18.35.4 case 7 */
1296 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1297 status = nfserr_noent;
1301 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1304 * Possible retry or client restart. Per 18.35.4 case 4,
1305 * a new unconfirmed record should be generated regardless
1306 * of whether any properties have changed.
1308 expire_client(unconf);
1313 new = create_client(exid->clname, dname);
1315 status = nfserr_resource;
1319 copy_verf(new, &verf);
1320 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1321 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1324 add_to_unconfirmed(new, strhashval);
1326 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1327 exid->clientid.cl_id = new->cl_clientid.cl_id;
1330 nfsd4_set_ex_flags(new, exid);
1332 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1333 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1337 nfs4_unlock_state();
1339 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1344 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1346 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1349 /* The slot is in use, and no response has been sent. */
1351 if (seqid == slot_seqid)
1352 return nfserr_jukebox;
1354 return nfserr_seq_misordered;
1357 if (likely(seqid == slot_seqid + 1))
1360 if (seqid == slot_seqid)
1361 return nfserr_replay_cache;
1363 if (seqid == 1 && (slot_seqid + 1) == 0)
1365 /* Misordered replay or misordered new request */
1366 return nfserr_seq_misordered;
1370 * Cache the create session result into the create session single DRC
1371 * slot cache by saving the xdr structure. sl_seqid has been set.
1372 * Do this for solo or embedded create session operations.
1375 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1376 struct nfsd4_clid_slot *slot, int nfserr)
1378 slot->sl_status = nfserr;
1379 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1383 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1384 struct nfsd4_clid_slot *slot)
1386 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1387 return slot->sl_status;
1391 nfsd4_create_session(struct svc_rqst *rqstp,
1392 struct nfsd4_compound_state *cstate,
1393 struct nfsd4_create_session *cr_ses)
1395 struct sockaddr *sa = svc_addr(rqstp);
1396 struct nfs4_client *conf, *unconf;
1397 struct nfsd4_clid_slot *cs_slot = NULL;
1401 unconf = find_unconfirmed_client(&cr_ses->clientid);
1402 conf = find_confirmed_client(&cr_ses->clientid);
1405 cs_slot = &conf->cl_cs_slot;
1406 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1407 if (status == nfserr_replay_cache) {
1408 dprintk("Got a create_session replay! seqid= %d\n",
1410 /* Return the cached reply status */
1411 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1413 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1414 status = nfserr_seq_misordered;
1415 dprintk("Sequence misordered!\n");
1416 dprintk("Expected seqid= %d but got seqid= %d\n",
1417 cs_slot->sl_seqid, cr_ses->seqid);
1420 cs_slot->sl_seqid++;
1421 } else if (unconf) {
1422 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1423 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1424 status = nfserr_clid_inuse;
1428 cs_slot = &unconf->cl_cs_slot;
1429 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1431 /* an unconfirmed replay returns misordered */
1432 status = nfserr_seq_misordered;
1436 cs_slot->sl_seqid++; /* from 0 to 1 */
1437 move_to_confirmed(unconf);
1440 * We do not support RDMA or persistent sessions
1442 cr_ses->flags &= ~SESSION4_PERSIST;
1443 cr_ses->flags &= ~SESSION4_RDMA;
1447 status = nfserr_stale_clientid;
1451 status = alloc_init_session(rqstp, conf, cr_ses);
1455 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1456 NFS4_MAX_SESSIONID_LEN);
1457 cr_ses->seqid = cs_slot->sl_seqid;
1460 /* cache solo and embedded create sessions under the state lock */
1461 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1463 nfs4_unlock_state();
1464 dprintk("%s returns %d\n", __func__, ntohl(status));
1469 nfsd4_destroy_session(struct svc_rqst *r,
1470 struct nfsd4_compound_state *cstate,
1471 struct nfsd4_destroy_session *sessionid)
1473 struct nfsd4_session *ses;
1474 u32 status = nfserr_badsession;
1477 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1478 * - Should we return nfserr_back_chan_busy if waiting for
1479 * callbacks on to-be-destroyed session?
1480 * - Do we need to clear any callback info from previous session?
1483 dump_sessionid(__func__, &sessionid->sessionid);
1484 spin_lock(&sessionid_lock);
1485 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1487 spin_unlock(&sessionid_lock);
1491 unhash_session(ses);
1492 spin_unlock(&sessionid_lock);
1494 /* wait for callbacks */
1495 shutdown_callback_client(ses->se_client);
1496 nfsd4_put_session(ses);
1499 dprintk("%s returns %d\n", __func__, ntohl(status));
1504 nfsd4_sequence(struct svc_rqst *rqstp,
1505 struct nfsd4_compound_state *cstate,
1506 struct nfsd4_sequence *seq)
1508 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1509 struct nfsd4_session *session;
1510 struct nfsd4_slot *slot;
1513 if (resp->opcnt != 1)
1514 return nfserr_sequence_pos;
1516 spin_lock(&sessionid_lock);
1517 status = nfserr_badsession;
1518 session = find_in_sessionid_hashtbl(&seq->sessionid);
1522 status = nfserr_badslot;
1523 if (seq->slotid >= session->se_fchannel.maxreqs)
1526 slot = &session->se_slots[seq->slotid];
1527 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1529 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1530 if (status == nfserr_replay_cache) {
1531 cstate->slot = slot;
1532 cstate->session = session;
1533 /* Return the cached reply status and set cstate->status
1534 * for nfsd4_svc_encode_compoundres processing */
1535 status = nfsd4_replay_cache_entry(resp, seq);
1536 cstate->status = nfserr_replay_cache;
1542 /* Success! bump slot seqid */
1543 slot->sl_inuse = true;
1544 slot->sl_seqid = seq->seqid;
1545 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1546 /* Always set the cache entry cachethis for solo sequence */
1547 if (nfsd4_is_solo_sequence(resp))
1548 slot->sl_cache_entry.ce_cachethis = 1;
1550 cstate->slot = slot;
1551 cstate->session = session;
1554 /* Renew the clientid on success and on replay.
1555 * Hold a session reference until done processing the compound:
1556 * nfsd4_put_session called only if the cstate slot is set.
1558 renew_client(session->se_client);
1559 nfsd4_get_session(session);
1561 spin_unlock(&sessionid_lock);
1562 dprintk("%s: return %d\n", __func__, ntohl(status));
1567 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1568 struct nfsd4_setclientid *setclid)
1570 struct sockaddr *sa = svc_addr(rqstp);
1571 struct xdr_netobj clname = {
1572 .len = setclid->se_namelen,
1573 .data = setclid->se_name,
1575 nfs4_verifier clverifier = setclid->se_verf;
1576 unsigned int strhashval;
1577 struct nfs4_client *conf, *unconf, *new;
1580 char dname[HEXDIR_LEN];
1582 if (!check_name(clname))
1583 return nfserr_inval;
1585 status = nfs4_make_rec_clidname(dname, &clname);
1590 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1591 * We get here on a DRC miss.
1594 strhashval = clientstr_hashval(dname);
1597 conf = find_confirmed_client_by_str(dname, strhashval, false);
1599 /* RFC 3530 14.2.33 CASE 0: */
1600 status = nfserr_clid_inuse;
1601 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1602 char addr_str[INET6_ADDRSTRLEN];
1603 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1605 dprintk("NFSD: setclientid: string in use by client "
1606 "at %s\n", addr_str);
1611 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1612 * has a description of SETCLIENTID request processing consisting
1613 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1615 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1616 status = nfserr_resource;
1619 * RFC 3530 14.2.33 CASE 4:
1620 * placed first, because it is the normal case
1623 expire_client(unconf);
1624 new = create_client(clname, dname);
1628 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1630 * RFC 3530 14.2.33 CASE 1:
1631 * probable callback update
1634 /* Note this is removing unconfirmed {*x***},
1635 * which is stronger than RFC recommended {vxc**}.
1636 * This has the advantage that there is at most
1637 * one {*x***} in either list at any time.
1639 expire_client(unconf);
1641 new = create_client(clname, dname);
1644 copy_clid(new, conf);
1645 } else if (!unconf) {
1647 * RFC 3530 14.2.33 CASE 2:
1648 * probable client reboot; state will be removed if
1651 new = create_client(clname, dname);
1657 * RFC 3530 14.2.33 CASE 3:
1658 * probable client reboot; state will be removed if
1661 expire_client(unconf);
1662 new = create_client(clname, dname);
1667 copy_verf(new, &clverifier);
1668 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1669 new->cl_flavor = rqstp->rq_flavor;
1670 princ = svc_gss_principal(rqstp);
1672 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1673 if (new->cl_principal == NULL) {
1678 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1680 gen_callback(new, setclid);
1681 add_to_unconfirmed(new, strhashval);
1682 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1683 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1684 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1687 nfs4_unlock_state();
1693 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1694 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1695 * bullets, labeled as CASE1 - CASE4 below.
1698 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1699 struct nfsd4_compound_state *cstate,
1700 struct nfsd4_setclientid_confirm *setclientid_confirm)
1702 struct sockaddr *sa = svc_addr(rqstp);
1703 struct nfs4_client *conf, *unconf;
1704 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1705 clientid_t * clid = &setclientid_confirm->sc_clientid;
1708 if (STALE_CLIENTID(clid))
1709 return nfserr_stale_clientid;
1711 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1712 * We get here on a DRC miss.
1717 conf = find_confirmed_client(clid);
1718 unconf = find_unconfirmed_client(clid);
1720 status = nfserr_clid_inuse;
1721 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1723 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1727 * section 14.2.34 of RFC 3530 has a description of
1728 * SETCLIENTID_CONFIRM request processing consisting
1729 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1731 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1733 * RFC 3530 14.2.34 CASE 1:
1736 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1737 status = nfserr_clid_inuse;
1739 /* XXX: We just turn off callbacks until we can handle
1740 * change request correctly. */
1741 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1742 expire_client(unconf);
1746 } else if (conf && !unconf) {
1748 * RFC 3530 14.2.34 CASE 2:
1749 * probable retransmitted request; play it safe and
1752 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1753 status = nfserr_clid_inuse;
1756 } else if (!conf && unconf
1757 && same_verf(&unconf->cl_confirm, &confirm)) {
1759 * RFC 3530 14.2.34 CASE 3:
1760 * Normal case; new or rebooted client:
1762 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1763 status = nfserr_clid_inuse;
1766 clientstr_hashval(unconf->cl_recdir);
1767 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1770 nfsd4_remove_clid_dir(conf);
1771 expire_client(conf);
1773 move_to_confirmed(unconf);
1775 nfsd4_probe_callback(conf);
1778 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1779 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1782 * RFC 3530 14.2.34 CASE 4:
1783 * Client probably hasn't noticed that we rebooted yet.
1785 status = nfserr_stale_clientid;
1787 /* check that we have hit one of the cases...*/
1788 status = nfserr_clid_inuse;
1791 nfs4_unlock_state();
1795 /* OPEN Share state helper functions */
1796 static inline struct nfs4_file *
1797 alloc_init_file(struct inode *ino)
1799 struct nfs4_file *fp;
1800 unsigned int hashval = file_hashval(ino);
1802 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1804 atomic_set(&fp->fi_ref, 1);
1805 INIT_LIST_HEAD(&fp->fi_hash);
1806 INIT_LIST_HEAD(&fp->fi_stateids);
1807 INIT_LIST_HEAD(&fp->fi_delegations);
1808 spin_lock(&recall_lock);
1809 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1810 spin_unlock(&recall_lock);
1811 fp->fi_inode = igrab(ino);
1812 fp->fi_id = current_fileid++;
1813 fp->fi_had_conflict = false;
1820 nfsd4_free_slab(struct kmem_cache **slab)
1824 kmem_cache_destroy(*slab);
1829 nfsd4_free_slabs(void)
1831 nfsd4_free_slab(&stateowner_slab);
1832 nfsd4_free_slab(&file_slab);
1833 nfsd4_free_slab(&stateid_slab);
1834 nfsd4_free_slab(&deleg_slab);
1838 nfsd4_init_slabs(void)
1840 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1841 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1842 if (stateowner_slab == NULL)
1844 file_slab = kmem_cache_create("nfsd4_files",
1845 sizeof(struct nfs4_file), 0, 0, NULL);
1846 if (file_slab == NULL)
1848 stateid_slab = kmem_cache_create("nfsd4_stateids",
1849 sizeof(struct nfs4_stateid), 0, 0, NULL);
1850 if (stateid_slab == NULL)
1852 deleg_slab = kmem_cache_create("nfsd4_delegations",
1853 sizeof(struct nfs4_delegation), 0, 0, NULL);
1854 if (deleg_slab == NULL)
1859 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1864 nfs4_free_stateowner(struct kref *kref)
1866 struct nfs4_stateowner *sop =
1867 container_of(kref, struct nfs4_stateowner, so_ref);
1868 kfree(sop->so_owner.data);
1869 kmem_cache_free(stateowner_slab, sop);
1872 static inline struct nfs4_stateowner *
1873 alloc_stateowner(struct xdr_netobj *owner)
1875 struct nfs4_stateowner *sop;
1877 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1878 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1879 memcpy(sop->so_owner.data, owner->data, owner->len);
1880 sop->so_owner.len = owner->len;
1881 kref_init(&sop->so_ref);
1884 kmem_cache_free(stateowner_slab, sop);
1889 static struct nfs4_stateowner *
1890 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1891 struct nfs4_stateowner *sop;
1892 struct nfs4_replay *rp;
1893 unsigned int idhashval;
1895 if (!(sop = alloc_stateowner(&open->op_owner)))
1897 idhashval = ownerid_hashval(current_ownerid);
1898 INIT_LIST_HEAD(&sop->so_idhash);
1899 INIT_LIST_HEAD(&sop->so_strhash);
1900 INIT_LIST_HEAD(&sop->so_perclient);
1901 INIT_LIST_HEAD(&sop->so_stateids);
1902 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1903 INIT_LIST_HEAD(&sop->so_close_lru);
1905 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1906 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1907 list_add(&sop->so_perclient, &clp->cl_openowners);
1908 sop->so_is_open_owner = 1;
1909 sop->so_id = current_ownerid++;
1910 sop->so_client = clp;
1911 sop->so_seqid = open->op_seqid;
1912 sop->so_confirmed = 0;
1913 rp = &sop->so_replay;
1914 rp->rp_status = nfserr_serverfault;
1916 rp->rp_buf = rp->rp_ibuf;
1921 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1922 struct nfs4_stateowner *sop = open->op_stateowner;
1923 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1925 INIT_LIST_HEAD(&stp->st_hash);
1926 INIT_LIST_HEAD(&stp->st_perstateowner);
1927 INIT_LIST_HEAD(&stp->st_lockowners);
1928 INIT_LIST_HEAD(&stp->st_perfile);
1929 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1930 list_add(&stp->st_perstateowner, &sop->so_stateids);
1931 list_add(&stp->st_perfile, &fp->fi_stateids);
1932 stp->st_stateowner = sop;
1935 stp->st_stateid.si_boot = get_seconds();
1936 stp->st_stateid.si_stateownerid = sop->so_id;
1937 stp->st_stateid.si_fileid = fp->fi_id;
1938 stp->st_stateid.si_generation = 0;
1939 stp->st_access_bmap = 0;
1940 stp->st_deny_bmap = 0;
1941 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1942 &stp->st_access_bmap);
1943 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1944 stp->st_openstp = NULL;
1948 move_to_close_lru(struct nfs4_stateowner *sop)
1950 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1952 list_move_tail(&sop->so_close_lru, &close_lru);
1953 sop->so_time = get_seconds();
1957 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1960 return (sop->so_owner.len == owner->len) &&
1961 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1962 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1965 static struct nfs4_stateowner *
1966 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1968 struct nfs4_stateowner *so = NULL;
1970 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1971 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1977 /* search file_hashtbl[] for file */
1978 static struct nfs4_file *
1979 find_file(struct inode *ino)
1981 unsigned int hashval = file_hashval(ino);
1982 struct nfs4_file *fp;
1984 spin_lock(&recall_lock);
1985 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1986 if (fp->fi_inode == ino) {
1988 spin_unlock(&recall_lock);
1992 spin_unlock(&recall_lock);
1996 static inline int access_valid(u32 x, u32 minorversion)
1998 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2000 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2002 x &= ~NFS4_SHARE_ACCESS_MASK;
2003 if (minorversion && x) {
2004 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2006 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2008 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2015 static inline int deny_valid(u32 x)
2017 /* Note: unlike access bits, deny bits may be zero. */
2018 return x <= NFS4_SHARE_DENY_BOTH;
2022 * We store the NONE, READ, WRITE, and BOTH bits separately in the
2023 * st_{access,deny}_bmap field of the stateid, in order to track not
2024 * only what share bits are currently in force, but also what
2025 * combinations of share bits previous opens have used. This allows us
2026 * to enforce the recommendation of rfc 3530 14.2.19 that the server
2027 * return an error if the client attempt to downgrade to a combination
2028 * of share bits not explicable by closing some of its previous opens.
2030 * XXX: This enforcement is actually incomplete, since we don't keep
2031 * track of access/deny bit combinations; so, e.g., we allow:
2033 * OPEN allow read, deny write
2034 * OPEN allow both, deny none
2035 * DOWNGRADE allow read, deny none
2037 * which we should reject.
2040 set_access(unsigned int *access, unsigned long bmap) {
2044 for (i = 1; i < 4; i++) {
2045 if (test_bit(i, &bmap))
2051 set_deny(unsigned int *deny, unsigned long bmap) {
2055 for (i = 0; i < 4; i++) {
2056 if (test_bit(i, &bmap))
2062 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2063 unsigned int access, deny;
2065 set_access(&access, stp->st_access_bmap);
2066 set_deny(&deny, stp->st_deny_bmap);
2067 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2073 * Called to check deny when READ with all zero stateid or
2074 * WRITE with all zero or all one stateid
2077 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2079 struct inode *ino = current_fh->fh_dentry->d_inode;
2080 struct nfs4_file *fp;
2081 struct nfs4_stateid *stp;
2084 dprintk("NFSD: nfs4_share_conflict\n");
2086 fp = find_file(ino);
2089 ret = nfserr_locked;
2090 /* Search for conflicting share reservations */
2091 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2092 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2093 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2103 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2105 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2106 drop_file_write_access(filp);
2107 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2112 * Spawn a thread to perform a recall on the delegation represented
2113 * by the lease (file_lock)
2115 * Called from break_lease() with lock_kernel() held.
2116 * Note: we assume break_lease will only call this *once* for any given
2120 void nfsd_break_deleg_cb(struct file_lock *fl)
2122 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2124 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2128 /* We're assuming the state code never drops its reference
2129 * without first removing the lease. Since we're in this lease
2130 * callback (and since the lease code is serialized by the kernel
2131 * lock) we know the server hasn't removed the lease yet, we know
2132 * it's safe to take a reference: */
2133 atomic_inc(&dp->dl_count);
2134 atomic_inc(&dp->dl_client->cl_count);
2136 spin_lock(&recall_lock);
2137 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2138 spin_unlock(&recall_lock);
2140 /* only place dl_time is set. protected by lock_kernel*/
2141 dp->dl_time = get_seconds();
2144 * We don't want the locks code to timeout the lease for us;
2145 * we'll remove it ourself if the delegation isn't returned
2148 fl->fl_break_time = 0;
2150 dp->dl_file->fi_had_conflict = true;
2151 nfsd4_cb_recall(dp);
2155 * The file_lock is being reapd.
2157 * Called by locks_free_lock() with lock_kernel() held.
2160 void nfsd_release_deleg_cb(struct file_lock *fl)
2162 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2164 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2166 if (!(fl->fl_flags & FL_LEASE) || !dp)
2168 dp->dl_flock = NULL;
2172 * Set the delegation file_lock back pointer.
2174 * Called from setlease() with lock_kernel() held.
2177 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2179 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2181 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2188 * Called from setlease() with lock_kernel() held
2191 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2193 struct nfs4_delegation *onlistd =
2194 (struct nfs4_delegation *)onlist->fl_owner;
2195 struct nfs4_delegation *tryd =
2196 (struct nfs4_delegation *)try->fl_owner;
2198 if (onlist->fl_lmops != try->fl_lmops)
2201 return onlistd->dl_client == tryd->dl_client;
2206 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2209 return lease_modify(onlist, arg);
2214 static struct lock_manager_operations nfsd_lease_mng_ops = {
2215 .fl_break = nfsd_break_deleg_cb,
2216 .fl_release_private = nfsd_release_deleg_cb,
2217 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2218 .fl_mylease = nfsd_same_client_deleg_cb,
2219 .fl_change = nfsd_change_deleg_cb,
2224 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2225 struct nfsd4_open *open)
2227 clientid_t *clientid = &open->op_clientid;
2228 struct nfs4_client *clp = NULL;
2229 unsigned int strhashval;
2230 struct nfs4_stateowner *sop = NULL;
2232 if (!check_name(open->op_owner))
2233 return nfserr_inval;
2235 if (STALE_CLIENTID(&open->op_clientid))
2236 return nfserr_stale_clientid;
2238 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2239 sop = find_openstateowner_str(strhashval, open);
2240 open->op_stateowner = sop;
2242 /* Make sure the client's lease hasn't expired. */
2243 clp = find_confirmed_client(clientid);
2245 return nfserr_expired;
2248 /* When sessions are used, skip open sequenceid processing */
2249 if (nfsd4_has_session(cstate))
2251 if (!sop->so_confirmed) {
2252 /* Replace unconfirmed owners without checking for replay. */
2253 clp = sop->so_client;
2254 release_openowner(sop);
2255 open->op_stateowner = NULL;
2258 if (open->op_seqid == sop->so_seqid - 1) {
2259 if (sop->so_replay.rp_buflen)
2260 return nfserr_replay_me;
2261 /* The original OPEN failed so spectacularly
2262 * that we don't even have replay data saved!
2263 * Therefore, we have no choice but to continue
2264 * processing this OPEN; presumably, we'll
2265 * fail again for the same reason.
2267 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2270 if (open->op_seqid != sop->so_seqid)
2271 return nfserr_bad_seqid;
2273 if (open->op_stateowner == NULL) {
2274 sop = alloc_init_open_stateowner(strhashval, clp, open);
2276 return nfserr_resource;
2277 open->op_stateowner = sop;
2279 list_del_init(&sop->so_close_lru);
2280 renew_client(sop->so_client);
2284 static inline __be32
2285 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2287 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2288 return nfserr_openmode;
2293 static struct nfs4_delegation *
2294 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2296 struct nfs4_delegation *dp;
2298 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2299 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2306 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2307 struct nfs4_delegation **dp)
2310 __be32 status = nfserr_bad_stateid;
2312 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2315 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2316 RD_STATE : WR_STATE;
2317 status = nfs4_check_delegmode(*dp, flags);
2321 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2325 open->op_stateowner->so_confirmed = 1;
2330 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2332 struct nfs4_stateid *local;
2333 __be32 status = nfserr_share_denied;
2334 struct nfs4_stateowner *sop = open->op_stateowner;
2336 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2337 /* ignore lock owners */
2338 if (local->st_stateowner->so_is_open_owner == 0)
2340 /* remember if we have seen this open owner */
2341 if (local->st_stateowner == sop)
2343 /* check for conflicting share reservations */
2344 if (!test_share(local, open))
2352 static inline struct nfs4_stateid *
2353 nfs4_alloc_stateid(void)
2355 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2359 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2360 struct nfs4_delegation *dp,
2361 struct svc_fh *cur_fh, int flags)
2363 struct nfs4_stateid *stp;
2365 stp = nfs4_alloc_stateid();
2367 return nfserr_resource;
2370 get_file(dp->dl_vfs_file);
2371 stp->st_vfs_file = dp->dl_vfs_file;
2374 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2377 if (status == nfserr_dropit)
2378 status = nfserr_jukebox;
2379 kmem_cache_free(stateid_slab, stp);
2387 static inline __be32
2388 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2389 struct nfsd4_open *open)
2391 struct iattr iattr = {
2392 .ia_valid = ATTR_SIZE,
2395 if (!open->op_truncate)
2397 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2398 return nfserr_inval;
2399 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2403 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2405 struct file *filp = stp->st_vfs_file;
2406 struct inode *inode = filp->f_path.dentry->d_inode;
2407 unsigned int share_access, new_writer;
2410 set_access(&share_access, stp->st_access_bmap);
2411 new_writer = (~share_access) & open->op_share_access
2412 & NFS4_SHARE_ACCESS_WRITE;
2415 int err = get_write_access(inode);
2417 return nfserrno(err);
2418 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2420 return nfserrno(err);
2421 file_take_write(filp);
2423 status = nfsd4_truncate(rqstp, cur_fh, open);
2426 put_write_access(inode);
2429 /* remember the open */
2430 filp->f_mode |= open->op_share_access;
2431 __set_bit(open->op_share_access, &stp->st_access_bmap);
2432 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2439 nfs4_set_claim_prev(struct nfsd4_open *open)
2441 open->op_stateowner->so_confirmed = 1;
2442 open->op_stateowner->so_client->cl_firststate = 1;
2446 * Attempt to hand out a delegation.
2449 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2451 struct nfs4_delegation *dp;
2452 struct nfs4_stateowner *sop = stp->st_stateowner;
2453 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2454 struct file_lock fl, *flp = &fl;
2455 int status, flag = 0;
2457 flag = NFS4_OPEN_DELEGATE_NONE;
2458 open->op_recall = 0;
2459 switch (open->op_claim_type) {
2460 case NFS4_OPEN_CLAIM_PREVIOUS:
2461 if (!atomic_read(&cb->cb_set))
2462 open->op_recall = 1;
2463 flag = open->op_delegate_type;
2464 if (flag == NFS4_OPEN_DELEGATE_NONE)
2467 case NFS4_OPEN_CLAIM_NULL:
2468 /* Let's not give out any delegations till everyone's
2469 * had the chance to reclaim theirs.... */
2470 if (locks_in_grace())
2472 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2474 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2475 flag = NFS4_OPEN_DELEGATE_WRITE;
2477 flag = NFS4_OPEN_DELEGATE_READ;
2483 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2485 flag = NFS4_OPEN_DELEGATE_NONE;
2488 locks_init_lock(&fl);
2489 fl.fl_lmops = &nfsd_lease_mng_ops;
2490 fl.fl_flags = FL_LEASE;
2491 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2492 fl.fl_end = OFFSET_MAX;
2493 fl.fl_owner = (fl_owner_t)dp;
2494 fl.fl_file = stp->st_vfs_file;
2495 fl.fl_pid = current->tgid;
2497 /* vfs_setlease checks to see if delegation should be handed out.
2498 * the lock_manager callbacks fl_mylease and fl_change are used
2500 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2501 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2502 unhash_delegation(dp);
2503 flag = NFS4_OPEN_DELEGATE_NONE;
2507 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2509 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2510 dp->dl_stateid.si_boot,
2511 dp->dl_stateid.si_stateownerid,
2512 dp->dl_stateid.si_fileid,
2513 dp->dl_stateid.si_generation);
2515 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2516 && flag == NFS4_OPEN_DELEGATE_NONE
2517 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2518 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2519 open->op_delegate_type = flag;
2523 * called with nfs4_lock_state() held.
2526 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2528 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2529 struct nfs4_file *fp = NULL;
2530 struct inode *ino = current_fh->fh_dentry->d_inode;
2531 struct nfs4_stateid *stp = NULL;
2532 struct nfs4_delegation *dp = NULL;
2535 status = nfserr_inval;
2536 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2537 || !deny_valid(open->op_share_deny))
2540 * Lookup file; if found, lookup stateid and check open request,
2541 * and check for delegations in the process of being recalled.
2542 * If not found, create the nfs4_file struct
2544 fp = find_file(ino);
2546 if ((status = nfs4_check_open(fp, open, &stp)))
2548 status = nfs4_check_deleg(fp, open, &dp);
2552 status = nfserr_bad_stateid;
2553 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2555 status = nfserr_resource;
2556 fp = alloc_init_file(ino);
2562 * OPEN the file, or upgrade an existing OPEN.
2563 * If truncate fails, the OPEN fails.
2566 /* Stateid was found, this is an OPEN upgrade */
2567 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2570 update_stateid(&stp->st_stateid);
2572 /* Stateid was not found, this is a new OPEN */
2574 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2575 flags |= NFSD_MAY_READ;
2576 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2577 flags |= NFSD_MAY_WRITE;
2578 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2581 init_stateid(stp, fp, open);
2582 status = nfsd4_truncate(rqstp, current_fh, open);
2584 release_open_stateid(stp);
2587 if (nfsd4_has_session(&resp->cstate))
2588 update_stateid(&stp->st_stateid);
2590 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2592 if (nfsd4_has_session(&resp->cstate))
2593 open->op_stateowner->so_confirmed = 1;
2596 * Attempt to hand out a delegation. No error return, because the
2597 * OPEN succeeds even if we fail.
2599 nfs4_open_delegation(current_fh, open, stp);
2603 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2604 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2605 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2609 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2610 nfs4_set_claim_prev(open);
2612 * To finish the open response, we just need to set the rflags.
2614 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2615 if (!open->op_stateowner->so_confirmed &&
2616 !nfsd4_has_session(&resp->cstate))
2617 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2623 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2626 struct nfs4_client *clp;
2630 dprintk("process_renew(%08x/%08x): starting\n",
2631 clid->cl_boot, clid->cl_id);
2632 status = nfserr_stale_clientid;
2633 if (STALE_CLIENTID(clid))
2635 clp = find_confirmed_client(clid);
2636 status = nfserr_expired;
2638 /* We assume the client took too long to RENEW. */
2639 dprintk("nfsd4_renew: clientid not found!\n");
2643 status = nfserr_cb_path_down;
2644 if (!list_empty(&clp->cl_delegations)
2645 && !atomic_read(&clp->cl_cb_conn.cb_set))
2649 nfs4_unlock_state();
2653 struct lock_manager nfsd4_manager = {
2657 nfsd4_end_grace(void)
2659 dprintk("NFSD: end of grace period\n");
2660 nfsd4_recdir_purge_old();
2661 locks_end_grace(&nfsd4_manager);
2665 nfs4_laundromat(void)
2667 struct nfs4_client *clp;
2668 struct nfs4_stateowner *sop;
2669 struct nfs4_delegation *dp;
2670 struct list_head *pos, *next, reaplist;
2671 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2672 time_t t, clientid_val = NFSD_LEASE_TIME;
2673 time_t u, test_val = NFSD_LEASE_TIME;
2677 dprintk("NFSD: laundromat service - starting\n");
2678 if (locks_in_grace())
2680 list_for_each_safe(pos, next, &client_lru) {
2681 clp = list_entry(pos, struct nfs4_client, cl_lru);
2682 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2683 t = clp->cl_time - cutoff;
2684 if (clientid_val > t)
2688 dprintk("NFSD: purging unused client (clientid %08x)\n",
2689 clp->cl_clientid.cl_id);
2690 nfsd4_remove_clid_dir(clp);
2693 INIT_LIST_HEAD(&reaplist);
2694 spin_lock(&recall_lock);
2695 list_for_each_safe(pos, next, &del_recall_lru) {
2696 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2697 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2698 u = dp->dl_time - cutoff;
2703 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2705 list_move(&dp->dl_recall_lru, &reaplist);
2707 spin_unlock(&recall_lock);
2708 list_for_each_safe(pos, next, &reaplist) {
2709 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2710 list_del_init(&dp->dl_recall_lru);
2711 unhash_delegation(dp);
2713 test_val = NFSD_LEASE_TIME;
2714 list_for_each_safe(pos, next, &close_lru) {
2715 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2716 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2717 u = sop->so_time - cutoff;
2722 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2724 release_openowner(sop);
2726 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2727 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2728 nfs4_unlock_state();
2729 return clientid_val;
2732 static struct workqueue_struct *laundry_wq;
2733 static void laundromat_main(struct work_struct *);
2734 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2737 laundromat_main(struct work_struct *not_used)
2741 t = nfs4_laundromat();
2742 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2743 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2746 static struct nfs4_stateowner *
2747 search_close_lru(u32 st_id, int flags)
2749 struct nfs4_stateowner *local = NULL;
2751 if (flags & CLOSE_STATE) {
2752 list_for_each_entry(local, &close_lru, so_close_lru) {
2753 if (local->so_id == st_id)
2761 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2763 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2767 STALE_STATEID(stateid_t *stateid)
2769 if (time_after((unsigned long)boot_time,
2770 (unsigned long)stateid->si_boot)) {
2771 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2772 stateid->si_boot, stateid->si_stateownerid,
2773 stateid->si_fileid, stateid->si_generation);
2780 EXPIRED_STATEID(stateid_t *stateid)
2782 if (time_before((unsigned long)boot_time,
2783 ((unsigned long)stateid->si_boot)) &&
2784 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2785 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2786 stateid->si_boot, stateid->si_stateownerid,
2787 stateid->si_fileid, stateid->si_generation);
2794 stateid_error_map(stateid_t *stateid)
2796 if (STALE_STATEID(stateid))
2797 return nfserr_stale_stateid;
2798 if (EXPIRED_STATEID(stateid))
2799 return nfserr_expired;
2801 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2802 stateid->si_boot, stateid->si_stateownerid,
2803 stateid->si_fileid, stateid->si_generation);
2804 return nfserr_bad_stateid;
2808 access_permit_read(unsigned long access_bmap)
2810 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2811 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2812 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2816 access_permit_write(unsigned long access_bmap)
2818 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2819 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2823 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2825 __be32 status = nfserr_openmode;
2827 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2829 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2836 static inline __be32
2837 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2839 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2841 else if (locks_in_grace()) {
2842 /* Answer in remaining cases depends on existance of
2843 * conflicting state; so we must wait out the grace period. */
2844 return nfserr_grace;
2845 } else if (flags & WR_STATE)
2846 return nfs4_share_conflict(current_fh,
2847 NFS4_SHARE_DENY_WRITE);
2848 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2849 return nfs4_share_conflict(current_fh,
2850 NFS4_SHARE_DENY_READ);
2854 * Allow READ/WRITE during grace period on recovered state only for files
2855 * that are not able to provide mandatory locking.
2858 grace_disallows_io(struct inode *inode)
2860 return locks_in_grace() && mandatory_lock(inode);
2863 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2866 * When sessions are used the stateid generation number is ignored
2869 if ((flags & HAS_SESSION) && in->si_generation == 0)
2872 /* If the client sends us a stateid from the future, it's buggy: */
2873 if (in->si_generation > ref->si_generation)
2874 return nfserr_bad_stateid;
2876 * The following, however, can happen. For example, if the
2877 * client sends an open and some IO at the same time, the open
2878 * may bump si_generation while the IO is still in flight.
2879 * Thanks to hard links and renames, the client never knows what
2880 * file an open will affect. So it could avoid that situation
2881 * only by serializing all opens and IO from the same open
2882 * owner. To recover from the old_stateid error, the client
2883 * will just have to retry the IO:
2885 if (in->si_generation < ref->si_generation)
2886 return nfserr_old_stateid;
2891 static int is_delegation_stateid(stateid_t *stateid)
2893 return stateid->si_fileid == 0;
2897 * Checks for stateid operations
2900 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2901 stateid_t *stateid, int flags, struct file **filpp)
2903 struct nfs4_stateid *stp = NULL;
2904 struct nfs4_delegation *dp = NULL;
2905 struct svc_fh *current_fh = &cstate->current_fh;
2906 struct inode *ino = current_fh->fh_dentry->d_inode;
2912 if (grace_disallows_io(ino))
2913 return nfserr_grace;
2915 if (nfsd4_has_session(cstate))
2916 flags |= HAS_SESSION;
2918 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2919 return check_special_stateids(current_fh, stateid, flags);
2921 status = nfserr_stale_stateid;
2922 if (STALE_STATEID(stateid))
2925 status = nfserr_bad_stateid;
2926 if (is_delegation_stateid(stateid)) {
2927 dp = find_delegation_stateid(ino, stateid);
2929 status = stateid_error_map(stateid);
2932 status = check_stateid_generation(stateid, &dp->dl_stateid,
2936 status = nfs4_check_delegmode(dp, flags);
2939 renew_client(dp->dl_client);
2941 *filpp = dp->dl_vfs_file;
2942 } else { /* open or lock stateid */
2943 stp = find_stateid(stateid, flags);
2945 status = stateid_error_map(stateid);
2948 if (nfs4_check_fh(current_fh, stp))
2950 if (!stp->st_stateowner->so_confirmed)
2952 status = check_stateid_generation(stateid, &stp->st_stateid,
2956 status = nfs4_check_openmode(stp, flags);
2959 renew_client(stp->st_stateowner->so_client);
2961 *filpp = stp->st_vfs_file;
2971 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2972 RD_STATE : WR_STATE;
2976 * Checks for sequence id mutating operations.
2979 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2980 stateid_t *stateid, int flags,
2981 struct nfs4_stateowner **sopp,
2982 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2984 struct nfs4_stateid *stp;
2985 struct nfs4_stateowner *sop;
2986 struct svc_fh *current_fh = &cstate->current_fh;
2989 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2990 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2991 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2992 stateid->si_generation);
2997 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2998 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2999 return nfserr_bad_stateid;
3002 if (STALE_STATEID(stateid))
3003 return nfserr_stale_stateid;
3005 if (nfsd4_has_session(cstate))
3006 flags |= HAS_SESSION;
3009 * We return BAD_STATEID if filehandle doesn't match stateid,
3010 * the confirmed flag is incorrecly set, or the generation
3011 * number is incorrect.
3013 stp = find_stateid(stateid, flags);
3016 * Also, we should make sure this isn't just the result of
3019 sop = search_close_lru(stateid->si_stateownerid, flags);
3021 return stateid_error_map(stateid);
3027 *sopp = sop = stp->st_stateowner;
3030 clientid_t *lockclid = &lock->v.new.clientid;
3031 struct nfs4_client *clp = sop->so_client;
3035 lkflg = setlkflg(lock->lk_type);
3037 if (lock->lk_is_new) {
3038 if (!sop->so_is_open_owner)
3039 return nfserr_bad_stateid;
3040 if (!(flags & HAS_SESSION) &&
3041 !same_clid(&clp->cl_clientid, lockclid))
3042 return nfserr_bad_stateid;
3043 /* stp is the open stateid */
3044 status = nfs4_check_openmode(stp, lkflg);
3048 /* stp is the lock stateid */
3049 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3055 if (nfs4_check_fh(current_fh, stp)) {
3056 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3057 return nfserr_bad_stateid;
3061 * We now validate the seqid and stateid generation numbers.
3062 * For the moment, we ignore the possibility of
3063 * generation number wraparound.
3065 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3068 if (sop->so_confirmed && flags & CONFIRM) {
3069 dprintk("NFSD: preprocess_seqid_op: expected"
3070 " unconfirmed stateowner!\n");
3071 return nfserr_bad_stateid;
3073 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3074 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3075 " confirmed yet!\n");
3076 return nfserr_bad_stateid;
3078 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3081 renew_client(sop->so_client);
3085 if (seqid == sop->so_seqid - 1) {
3086 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3087 /* indicate replay to calling function */
3088 return nfserr_replay_me;
3090 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3091 sop->so_seqid, seqid);
3093 return nfserr_bad_seqid;
3097 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3098 struct nfsd4_open_confirm *oc)
3101 struct nfs4_stateowner *sop;
3102 struct nfs4_stateid *stp;
3104 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3105 (int)cstate->current_fh.fh_dentry->d_name.len,
3106 cstate->current_fh.fh_dentry->d_name.name);
3108 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3114 if ((status = nfs4_preprocess_seqid_op(cstate,
3115 oc->oc_seqid, &oc->oc_req_stateid,
3116 CONFIRM | OPEN_STATE,
3117 &oc->oc_stateowner, &stp, NULL)))
3120 sop = oc->oc_stateowner;
3121 sop->so_confirmed = 1;
3122 update_stateid(&stp->st_stateid);
3123 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3124 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3125 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3126 stp->st_stateid.si_boot,
3127 stp->st_stateid.si_stateownerid,
3128 stp->st_stateid.si_fileid,
3129 stp->st_stateid.si_generation);
3131 nfsd4_create_clid_dir(sop->so_client);
3133 if (oc->oc_stateowner) {
3134 nfs4_get_stateowner(oc->oc_stateowner);
3135 cstate->replay_owner = oc->oc_stateowner;
3137 nfs4_unlock_state();
3143 * unset all bits in union bitmap (bmap) that
3144 * do not exist in share (from successful OPEN_DOWNGRADE)
3147 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3150 for (i = 1; i < 4; i++) {
3151 if ((i & access) != i)
3152 __clear_bit(i, bmap);
3157 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3160 for (i = 0; i < 4; i++) {
3161 if ((i & deny) != i)
3162 __clear_bit(i, bmap);
3167 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3168 struct nfsd4_compound_state *cstate,
3169 struct nfsd4_open_downgrade *od)
3172 struct nfs4_stateid *stp;
3173 unsigned int share_access;
3175 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3176 (int)cstate->current_fh.fh_dentry->d_name.len,
3177 cstate->current_fh.fh_dentry->d_name.name);
3179 if (!access_valid(od->od_share_access, cstate->minorversion)
3180 || !deny_valid(od->od_share_deny))
3181 return nfserr_inval;
3184 if ((status = nfs4_preprocess_seqid_op(cstate,
3188 &od->od_stateowner, &stp, NULL)))
3191 status = nfserr_inval;
3192 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3193 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3194 stp->st_access_bmap, od->od_share_access);
3197 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3198 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3199 stp->st_deny_bmap, od->od_share_deny);
3202 set_access(&share_access, stp->st_access_bmap);
3203 nfs4_file_downgrade(stp->st_vfs_file,
3204 share_access & ~od->od_share_access);
3206 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3207 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3209 update_stateid(&stp->st_stateid);
3210 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3213 if (od->od_stateowner) {
3214 nfs4_get_stateowner(od->od_stateowner);
3215 cstate->replay_owner = od->od_stateowner;
3217 nfs4_unlock_state();
3222 * nfs4_unlock_state() called after encode
3225 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3226 struct nfsd4_close *close)
3229 struct nfs4_stateid *stp;
3231 dprintk("NFSD: nfsd4_close on file %.*s\n",
3232 (int)cstate->current_fh.fh_dentry->d_name.len,
3233 cstate->current_fh.fh_dentry->d_name.name);
3236 /* check close_lru for replay */
3237 if ((status = nfs4_preprocess_seqid_op(cstate,
3240 OPEN_STATE | CLOSE_STATE,
3241 &close->cl_stateowner, &stp, NULL)))
3244 update_stateid(&stp->st_stateid);
3245 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3247 /* release_stateid() calls nfsd_close() if needed */
3248 release_open_stateid(stp);
3250 /* place unused nfs4_stateowners on so_close_lru list to be
3251 * released by the laundromat service after the lease period
3252 * to enable us to handle CLOSE replay
3254 if (list_empty(&close->cl_stateowner->so_stateids))
3255 move_to_close_lru(close->cl_stateowner);
3257 if (close->cl_stateowner) {
3258 nfs4_get_stateowner(close->cl_stateowner);
3259 cstate->replay_owner = close->cl_stateowner;
3261 nfs4_unlock_state();
3266 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3267 struct nfsd4_delegreturn *dr)
3269 struct nfs4_delegation *dp;
3270 stateid_t *stateid = &dr->dr_stateid;
3271 struct inode *inode;
3275 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3277 inode = cstate->current_fh.fh_dentry->d_inode;
3279 if (nfsd4_has_session(cstate))
3280 flags |= HAS_SESSION;
3282 status = nfserr_bad_stateid;
3283 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3285 status = nfserr_stale_stateid;
3286 if (STALE_STATEID(stateid))
3288 status = nfserr_bad_stateid;
3289 if (!is_delegation_stateid(stateid))
3291 dp = find_delegation_stateid(inode, stateid);
3293 status = stateid_error_map(stateid);
3296 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3299 renew_client(dp->dl_client);
3301 unhash_delegation(dp);
3303 nfs4_unlock_state();
3310 * Lock owner state (byte-range locks)
3312 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3313 #define LOCK_HASH_BITS 8
3314 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3315 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3318 end_offset(u64 start, u64 len)
3323 return end >= start ? end: NFS4_MAX_UINT64;
3326 /* last octet in a range */
3328 last_byte_offset(u64 start, u64 len)
3334 return end > start ? end - 1: NFS4_MAX_UINT64;
3337 #define lockownerid_hashval(id) \
3338 ((id) & LOCK_HASH_MASK)
3340 static inline unsigned int
3341 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3342 struct xdr_netobj *ownername)
3344 return (file_hashval(inode) + cl_id
3345 + opaque_hashval(ownername->data, ownername->len))
3349 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3350 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3351 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3353 static struct nfs4_stateid *
3354 find_stateid(stateid_t *stid, int flags)
3356 struct nfs4_stateid *local;
3357 u32 st_id = stid->si_stateownerid;
3358 u32 f_id = stid->si_fileid;
3359 unsigned int hashval;
3361 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3362 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3363 hashval = stateid_hashval(st_id, f_id);
3364 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3365 if ((local->st_stateid.si_stateownerid == st_id) &&
3366 (local->st_stateid.si_fileid == f_id))
3371 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3372 hashval = stateid_hashval(st_id, f_id);
3373 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3374 if ((local->st_stateid.si_stateownerid == st_id) &&
3375 (local->st_stateid.si_fileid == f_id))
3382 static struct nfs4_delegation *
3383 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3385 struct nfs4_file *fp;
3386 struct nfs4_delegation *dl;
3388 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3389 stid->si_boot, stid->si_stateownerid,
3390 stid->si_fileid, stid->si_generation);
3392 fp = find_file(ino);
3395 dl = find_delegation_file(fp, stid);
3401 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3402 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3403 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3404 * locking, this prevents us from being completely protocol-compliant. The
3405 * real solution to this problem is to start using unsigned file offsets in
3406 * the VFS, but this is a very deep change!
3409 nfs4_transform_lock_offset(struct file_lock *lock)
3411 if (lock->fl_start < 0)
3412 lock->fl_start = OFFSET_MAX;
3413 if (lock->fl_end < 0)
3414 lock->fl_end = OFFSET_MAX;
3417 /* Hack!: For now, we're defining this just so we can use a pointer to it
3418 * as a unique cookie to identify our (NFSv4's) posix locks. */
3419 static struct lock_manager_operations nfsd_posix_mng_ops = {
3423 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3425 struct nfs4_stateowner *sop;
3428 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3429 sop = (struct nfs4_stateowner *) fl->fl_owner;
3430 hval = lockownerid_hashval(sop->so_id);
3431 kref_get(&sop->so_ref);
3433 deny->ld_clientid = sop->so_client->cl_clientid;
3435 deny->ld_sop = NULL;
3436 deny->ld_clientid.cl_boot = 0;
3437 deny->ld_clientid.cl_id = 0;
3439 deny->ld_start = fl->fl_start;
3440 deny->ld_length = NFS4_MAX_UINT64;
3441 if (fl->fl_end != NFS4_MAX_UINT64)
3442 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3443 deny->ld_type = NFS4_READ_LT;
3444 if (fl->fl_type != F_RDLCK)
3445 deny->ld_type = NFS4_WRITE_LT;
3448 static struct nfs4_stateowner *
3449 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3450 struct xdr_netobj *owner)
3452 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3453 struct nfs4_stateowner *op;
3455 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3456 if (same_owner_str(op, owner, clid))
3463 * Alloc a lock owner structure.
3464 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3467 * strhashval = lock_ownerstr_hashval
3470 static struct nfs4_stateowner *
3471 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3472 struct nfs4_stateowner *sop;
3473 struct nfs4_replay *rp;
3474 unsigned int idhashval;
3476 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3478 idhashval = lockownerid_hashval(current_ownerid);
3479 INIT_LIST_HEAD(&sop->so_idhash);
3480 INIT_LIST_HEAD(&sop->so_strhash);
3481 INIT_LIST_HEAD(&sop->so_perclient);
3482 INIT_LIST_HEAD(&sop->so_stateids);
3483 INIT_LIST_HEAD(&sop->so_perstateid);
3484 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3486 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3487 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3488 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3489 sop->so_is_open_owner = 0;
3490 sop->so_id = current_ownerid++;
3491 sop->so_client = clp;
3492 /* It is the openowner seqid that will be incremented in encode in the
3493 * case of new lockowners; so increment the lock seqid manually: */
3494 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3495 sop->so_confirmed = 1;
3496 rp = &sop->so_replay;
3497 rp->rp_status = nfserr_serverfault;
3499 rp->rp_buf = rp->rp_ibuf;
3503 static struct nfs4_stateid *
3504 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3506 struct nfs4_stateid *stp;
3507 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3509 stp = nfs4_alloc_stateid();
3512 INIT_LIST_HEAD(&stp->st_hash);
3513 INIT_LIST_HEAD(&stp->st_perfile);
3514 INIT_LIST_HEAD(&stp->st_perstateowner);
3515 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3516 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3517 list_add(&stp->st_perfile, &fp->fi_stateids);
3518 list_add(&stp->st_perstateowner, &sop->so_stateids);
3519 stp->st_stateowner = sop;
3522 stp->st_stateid.si_boot = get_seconds();
3523 stp->st_stateid.si_stateownerid = sop->so_id;
3524 stp->st_stateid.si_fileid = fp->fi_id;
3525 stp->st_stateid.si_generation = 0;
3526 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3527 stp->st_access_bmap = open_stp->st_access_bmap;
3528 stp->st_deny_bmap = open_stp->st_deny_bmap;
3529 stp->st_openstp = open_stp;
3536 check_lock_length(u64 offset, u64 length)
3538 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3539 LOFF_OVERFLOW(offset, length)));
3546 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3547 struct nfsd4_lock *lock)
3549 struct nfs4_stateowner *open_sop = NULL;
3550 struct nfs4_stateowner *lock_sop = NULL;
3551 struct nfs4_stateid *lock_stp;
3553 struct file_lock file_lock;
3554 struct file_lock conflock;
3556 unsigned int strhashval;
3560 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3561 (long long) lock->lk_offset,
3562 (long long) lock->lk_length);
3564 if (check_lock_length(lock->lk_offset, lock->lk_length))
3565 return nfserr_inval;
3567 if ((status = fh_verify(rqstp, &cstate->current_fh,
3568 S_IFREG, NFSD_MAY_LOCK))) {
3569 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3575 if (lock->lk_is_new) {
3577 * Client indicates that this is a new lockowner.
3578 * Use open owner and open stateid to create lock owner and
3581 struct nfs4_stateid *open_stp = NULL;
3582 struct nfs4_file *fp;
3584 status = nfserr_stale_clientid;
3585 if (!nfsd4_has_session(cstate) &&
3586 STALE_CLIENTID(&lock->lk_new_clientid))
3589 /* validate and update open stateid and open seqid */
3590 status = nfs4_preprocess_seqid_op(cstate,
3591 lock->lk_new_open_seqid,
3592 &lock->lk_new_open_stateid,
3594 &lock->lk_replay_owner, &open_stp,
3598 open_sop = lock->lk_replay_owner;
3599 /* create lockowner and lock stateid */
3600 fp = open_stp->st_file;
3601 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3602 open_sop->so_client->cl_clientid.cl_id,
3603 &lock->v.new.owner);
3604 /* XXX: Do we need to check for duplicate stateowners on
3605 * the same file, or should they just be allowed (and
3606 * create new stateids)? */
3607 status = nfserr_resource;
3608 lock_sop = alloc_init_lock_stateowner(strhashval,
3609 open_sop->so_client, open_stp, lock);
3610 if (lock_sop == NULL)
3612 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3613 if (lock_stp == NULL)
3616 /* lock (lock owner + lock stateid) already exists */
3617 status = nfs4_preprocess_seqid_op(cstate,
3618 lock->lk_old_lock_seqid,
3619 &lock->lk_old_lock_stateid,
3621 &lock->lk_replay_owner, &lock_stp, lock);
3624 lock_sop = lock->lk_replay_owner;
3626 /* lock->lk_replay_owner and lock_stp have been created or found */
3627 filp = lock_stp->st_vfs_file;
3629 status = nfserr_grace;
3630 if (locks_in_grace() && !lock->lk_reclaim)
3632 status = nfserr_no_grace;
3633 if (!locks_in_grace() && lock->lk_reclaim)
3636 locks_init_lock(&file_lock);
3637 switch (lock->lk_type) {
3640 file_lock.fl_type = F_RDLCK;
3644 case NFS4_WRITEW_LT:
3645 file_lock.fl_type = F_WRLCK;
3649 status = nfserr_inval;
3652 file_lock.fl_owner = (fl_owner_t)lock_sop;
3653 file_lock.fl_pid = current->tgid;
3654 file_lock.fl_file = filp;
3655 file_lock.fl_flags = FL_POSIX;
3656 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3658 file_lock.fl_start = lock->lk_offset;
3659 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3660 nfs4_transform_lock_offset(&file_lock);
3663 * Try to lock the file in the VFS.
3664 * Note: locks.c uses the BKL to protect the inode's lock list.
3667 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3669 case 0: /* success! */
3670 update_stateid(&lock_stp->st_stateid);
3671 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3675 case (EAGAIN): /* conflock holds conflicting lock */
3676 status = nfserr_denied;
3677 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3678 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3681 status = nfserr_deadlock;
3684 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3685 status = nfserr_resource;
3689 if (status && lock->lk_is_new && lock_sop)
3690 release_lockowner(lock_sop);
3691 if (lock->lk_replay_owner) {
3692 nfs4_get_stateowner(lock->lk_replay_owner);
3693 cstate->replay_owner = lock->lk_replay_owner;
3695 nfs4_unlock_state();
3700 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3701 * so we do a temporary open here just to get an open file to pass to
3702 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3705 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3710 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3713 err = vfs_test_lock(file, lock);
3722 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3723 struct nfsd4_lockt *lockt)
3725 struct inode *inode;
3726 struct file_lock file_lock;
3730 if (locks_in_grace())
3731 return nfserr_grace;
3733 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3734 return nfserr_inval;
3736 lockt->lt_stateowner = NULL;
3739 status = nfserr_stale_clientid;
3740 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3743 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3744 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3745 if (status == nfserr_symlink)
3746 status = nfserr_inval;
3750 inode = cstate->current_fh.fh_dentry->d_inode;
3751 locks_init_lock(&file_lock);
3752 switch (lockt->lt_type) {
3755 file_lock.fl_type = F_RDLCK;
3758 case NFS4_WRITEW_LT:
3759 file_lock.fl_type = F_WRLCK;
3762 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3763 status = nfserr_inval;
3767 lockt->lt_stateowner = find_lockstateowner_str(inode,
3768 &lockt->lt_clientid, &lockt->lt_owner);
3769 if (lockt->lt_stateowner)
3770 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3771 file_lock.fl_pid = current->tgid;
3772 file_lock.fl_flags = FL_POSIX;
3774 file_lock.fl_start = lockt->lt_offset;
3775 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3777 nfs4_transform_lock_offset(&file_lock);
3780 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3782 status = nfserrno(error);
3785 if (file_lock.fl_type != F_UNLCK) {
3786 status = nfserr_denied;
3787 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3790 nfs4_unlock_state();
3795 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3796 struct nfsd4_locku *locku)
3798 struct nfs4_stateid *stp;
3799 struct file *filp = NULL;
3800 struct file_lock file_lock;
3804 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3805 (long long) locku->lu_offset,
3806 (long long) locku->lu_length);
3808 if (check_lock_length(locku->lu_offset, locku->lu_length))
3809 return nfserr_inval;
3813 if ((status = nfs4_preprocess_seqid_op(cstate,
3817 &locku->lu_stateowner, &stp, NULL)))
3820 filp = stp->st_vfs_file;
3822 locks_init_lock(&file_lock);
3823 file_lock.fl_type = F_UNLCK;
3824 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3825 file_lock.fl_pid = current->tgid;
3826 file_lock.fl_file = filp;
3827 file_lock.fl_flags = FL_POSIX;
3828 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3829 file_lock.fl_start = locku->lu_offset;
3831 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3832 nfs4_transform_lock_offset(&file_lock);
3835 * Try to unlock the file in the VFS.
3837 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3839 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3843 * OK, unlock succeeded; the only thing left to do is update the stateid.
3845 update_stateid(&stp->st_stateid);
3846 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3849 if (locku->lu_stateowner) {
3850 nfs4_get_stateowner(locku->lu_stateowner);
3851 cstate->replay_owner = locku->lu_stateowner;
3853 nfs4_unlock_state();
3857 status = nfserrno(err);
3863 * 1: locks held by lockowner
3864 * 0: no locks held by lockowner
3867 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3869 struct file_lock **flpp;
3870 struct inode *inode = filp->f_path.dentry->d_inode;
3874 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3875 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3886 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3887 struct nfsd4_compound_state *cstate,
3888 struct nfsd4_release_lockowner *rlockowner)
3890 clientid_t *clid = &rlockowner->rl_clientid;
3891 struct nfs4_stateowner *sop;
3892 struct nfs4_stateid *stp;
3893 struct xdr_netobj *owner = &rlockowner->rl_owner;
3894 struct list_head matches;
3898 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3899 clid->cl_boot, clid->cl_id);
3901 /* XXX check for lease expiration */
3903 status = nfserr_stale_clientid;
3904 if (STALE_CLIENTID(clid))
3909 status = nfserr_locks_held;
3910 /* XXX: we're doing a linear search through all the lockowners.
3911 * Yipes! For now we'll just hope clients aren't really using
3912 * release_lockowner much, but eventually we have to fix these
3913 * data structures. */
3914 INIT_LIST_HEAD(&matches);
3915 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3916 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3917 if (!same_owner_str(sop, owner, clid))
3919 list_for_each_entry(stp, &sop->so_stateids,
3921 if (check_for_locks(stp->st_vfs_file, sop))
3923 /* Note: so_perclient unused for lockowners,
3924 * so it's OK to fool with here. */
3925 list_add(&sop->so_perclient, &matches);
3929 /* Clients probably won't expect us to return with some (but not all)
3930 * of the lockowner state released; so don't release any until all
3931 * have been checked. */
3933 while (!list_empty(&matches)) {
3934 sop = list_entry(matches.next, struct nfs4_stateowner,
3936 /* unhash_stateowner deletes so_perclient only
3937 * for openowners. */
3938 list_del(&sop->so_perclient);
3939 release_lockowner(sop);
3942 nfs4_unlock_state();
3946 static inline struct nfs4_client_reclaim *
3949 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3953 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3955 unsigned int strhashval = clientstr_hashval(name);
3956 struct nfs4_client *clp;
3958 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3963 * failure => all reset bets are off, nfserr_no_grace...
3966 nfs4_client_to_reclaim(const char *name)
3968 unsigned int strhashval;
3969 struct nfs4_client_reclaim *crp = NULL;
3971 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3972 crp = alloc_reclaim();
3975 strhashval = clientstr_hashval(name);
3976 INIT_LIST_HEAD(&crp->cr_strhash);
3977 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3978 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3979 reclaim_str_hashtbl_size++;
3984 nfs4_release_reclaim(void)
3986 struct nfs4_client_reclaim *crp = NULL;
3989 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3990 while (!list_empty(&reclaim_str_hashtbl[i])) {
3991 crp = list_entry(reclaim_str_hashtbl[i].next,
3992 struct nfs4_client_reclaim, cr_strhash);
3993 list_del(&crp->cr_strhash);
3995 reclaim_str_hashtbl_size--;
3998 BUG_ON(reclaim_str_hashtbl_size);
4002 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4003 static struct nfs4_client_reclaim *
4004 nfs4_find_reclaim_client(clientid_t *clid)
4006 unsigned int strhashval;
4007 struct nfs4_client *clp;
4008 struct nfs4_client_reclaim *crp = NULL;
4011 /* find clientid in conf_id_hashtbl */
4012 clp = find_confirmed_client(clid);
4016 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4017 clp->cl_name.len, clp->cl_name.data,
4020 /* find clp->cl_name in reclaim_str_hashtbl */
4021 strhashval = clientstr_hashval(clp->cl_recdir);
4022 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4023 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4031 * Called from OPEN. Look for clientid in reclaim list.
4034 nfs4_check_open_reclaim(clientid_t *clid)
4036 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4039 /* initialization to perform at module load time: */
4042 nfs4_state_init(void)
4046 status = nfsd4_init_slabs();
4049 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4050 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4051 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4052 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4053 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4054 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4056 for (i = 0; i < SESSION_HASH_SIZE; i++)
4057 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4058 for (i = 0; i < FILE_HASH_SIZE; i++) {
4059 INIT_LIST_HEAD(&file_hashtbl[i]);
4061 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4062 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4063 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4065 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4066 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4067 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4069 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4070 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4071 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4073 memset(&onestateid, ~0, sizeof(stateid_t));
4074 INIT_LIST_HEAD(&close_lru);
4075 INIT_LIST_HEAD(&client_lru);
4076 INIT_LIST_HEAD(&del_recall_lru);
4077 reclaim_str_hashtbl_size = 0;
4082 nfsd4_load_reboot_recovery_data(void)
4087 nfsd4_init_recdir(user_recovery_dirname);
4088 status = nfsd4_recdir_load();
4089 nfs4_unlock_state();
4091 printk("NFSD: Failure reading reboot recovery data\n");
4095 get_nfs4_grace_period(void)
4097 return max(user_lease_time, lease_time) * HZ;
4101 * Since the lifetime of a delegation isn't limited to that of an open, a
4102 * client may quite reasonably hang on to a delegation as long as it has
4103 * the inode cached. This becomes an obvious problem the first time a
4104 * client's inode cache approaches the size of the server's total memory.
4106 * For now we avoid this problem by imposing a hard limit on the number
4107 * of delegations, which varies according to the server's memory size.
4110 set_max_delegations(void)
4113 * Allow at most 4 delegations per megabyte of RAM. Quick
4114 * estimates suggest that in the worst case (where every delegation
4115 * is for a different inode), a delegation could take about 1.5K,
4116 * giving a worst case usage of about 6% of memory.
4118 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4121 /* initialization to perform when the nfsd service is started: */
4124 __nfs4_state_start(void)
4126 unsigned long grace_time;
4128 boot_time = get_seconds();
4129 grace_time = get_nfs4_grace_period();
4130 lease_time = user_lease_time;
4131 locks_start_grace(&nfsd4_manager);
4132 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4134 laundry_wq = create_singlethread_workqueue("nfsd4");
4135 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4136 set_max_delegations();
4140 nfs4_state_start(void)
4144 nfsd4_load_reboot_recovery_data();
4145 __nfs4_state_start();
4151 nfs4_lease_time(void)
4157 __nfs4_state_shutdown(void)
4160 struct nfs4_client *clp = NULL;
4161 struct nfs4_delegation *dp = NULL;
4162 struct list_head *pos, *next, reaplist;
4164 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4165 while (!list_empty(&conf_id_hashtbl[i])) {
4166 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4169 while (!list_empty(&unconf_str_hashtbl[i])) {
4170 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4174 INIT_LIST_HEAD(&reaplist);
4175 spin_lock(&recall_lock);
4176 list_for_each_safe(pos, next, &del_recall_lru) {
4177 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4178 list_move(&dp->dl_recall_lru, &reaplist);
4180 spin_unlock(&recall_lock);
4181 list_for_each_safe(pos, next, &reaplist) {
4182 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4183 list_del_init(&dp->dl_recall_lru);
4184 unhash_delegation(dp);
4187 nfsd4_shutdown_recdir();
4192 nfs4_state_shutdown(void)
4194 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4195 destroy_workqueue(laundry_wq);
4196 locks_end_grace(&nfsd4_manager);
4198 nfs4_release_reclaim();
4199 __nfs4_state_shutdown();
4200 nfs4_unlock_state();
4204 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4205 * accessed when nfsd is starting.
4208 nfs4_set_recdir(char *recdir)
4210 strcpy(user_recovery_dirname, recdir);
4214 * Change the NFSv4 recovery directory to recdir.
4217 nfs4_reset_recoverydir(char *recdir)
4222 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4226 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4227 nfs4_set_recdir(recdir);
4235 nfs4_recoverydir(void)
4237 return user_recovery_dirname;
4241 * Called when leasetime is changed.
4243 * The only way the protocol gives us to handle on-the-fly lease changes is to
4244 * simulate a reboot. Instead of doing that, we just wait till the next time
4245 * we start to register any changes in lease time. If the administrator
4246 * really wants to change the lease time *now*, they can go ahead and bring
4247 * nfsd down and then back up again after changing the lease time.
4249 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4250 * when nfsd is starting
4253 nfs4_reset_lease(time_t leasetime)
4255 user_lease_time = leasetime;