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))
901 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
903 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
904 unsigned short expected_family;
906 /* Currently, we only support tcp and tcp6 for the callback channel */
907 if (se->se_callback_netid_len == 3 &&
908 !memcmp(se->se_callback_netid_val, "tcp", 3))
909 expected_family = AF_INET;
910 else if (se->se_callback_netid_len == 4 &&
911 !memcmp(se->se_callback_netid_val, "tcp6", 4))
912 expected_family = AF_INET6;
916 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
917 se->se_callback_addr_len,
918 (struct sockaddr *) &cb->cb_addr,
919 sizeof(cb->cb_addr));
921 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
924 if (cb->cb_addr.ss_family == AF_INET6)
925 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
927 cb->cb_minorversion = 0;
928 cb->cb_prog = se->se_callback_prog;
929 cb->cb_ident = se->se_callback_ident;
932 cb->cb_addr.ss_family = AF_UNSPEC;
934 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
935 "will not receive delegations\n",
936 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
942 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
944 struct nfsd4_compoundres *resp = rqstp->rq_resp;
946 resp->cstate.statp = statp;
950 * Dereference the result pages.
953 nfsd4_release_respages(struct page **respages, short resused)
957 dprintk("--> %s\n", __func__);
958 for (i = 0; i < resused; i++) {
961 put_page(respages[i]);
967 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
971 for (i = 0; i < count; i++) {
972 topages[i] = frompages[i];
975 get_page(topages[i]);
980 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
981 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
982 * length of the XDR response is less than se_fmaxresp_cached
983 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
984 * of the reply (e.g. readdir).
986 * Store the base and length of the rq_req.head[0] page
987 * of the NFSv4.1 data, just past the rpc header.
990 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
992 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
993 struct svc_rqst *rqstp = resp->rqstp;
994 struct nfsd4_compoundargs *args = rqstp->rq_argp;
995 struct nfsd4_op *op = &args->ops[resp->opcnt];
996 struct kvec *resv = &rqstp->rq_res.head[0];
998 dprintk("--> %s entry %p\n", __func__, entry);
1000 /* Don't cache a failed OP_SEQUENCE. */
1001 if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1004 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1005 entry->ce_opcnt = resp->opcnt;
1006 entry->ce_status = resp->cstate.status;
1009 * Don't need a page to cache just the sequence operation - the slot
1013 if (nfsd4_not_cached(resp)) {
1014 entry->ce_resused = 0;
1015 entry->ce_rpchdrlen = 0;
1016 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1017 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1020 entry->ce_resused = rqstp->rq_resused;
1021 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1022 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1023 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1025 entry->ce_datav.iov_base = resp->cstate.statp;
1026 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1027 (char *)page_address(rqstp->rq_respages[0]));
1028 /* Current request rpc header length*/
1029 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1030 (char *)page_address(rqstp->rq_respages[0]);
1034 * We keep the rpc header, but take the nfs reply from the replycache.
1037 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1038 struct nfsd4_cache_entry *entry)
1040 struct svc_rqst *rqstp = resp->rqstp;
1041 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1044 /* Current request rpc header length*/
1045 len = (char *)resp->cstate.statp -
1046 (char *)page_address(rqstp->rq_respages[0]);
1047 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1048 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1049 entry->ce_datav.iov_len);
1052 /* copy the cached reply nfsd data past the current rpc header */
1053 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1054 entry->ce_datav.iov_len);
1055 resv->iov_len = len + entry->ce_datav.iov_len;
1060 * Encode the replay sequence operation from the slot values.
1061 * If cachethis is FALSE encode the uncached rep error on the next
1062 * operation which sets resp->p and increments resp->opcnt for
1063 * nfs4svc_encode_compoundres.
1067 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1068 struct nfsd4_compoundres *resp)
1070 struct nfsd4_op *op;
1071 struct nfsd4_slot *slot = resp->cstate.slot;
1073 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1074 resp->opcnt, resp->cstate.slot->sl_cache_entry.ce_cachethis);
1076 /* Encode the replayed sequence operation */
1077 op = &args->ops[resp->opcnt - 1];
1078 nfsd4_encode_operation(resp, op);
1080 /* Return nfserr_retry_uncached_rep in next operation. */
1081 if (args->opcnt > 1 && slot->sl_cache_entry.ce_cachethis == 0) {
1082 op = &args->ops[resp->opcnt++];
1083 op->status = nfserr_retry_uncached_rep;
1084 nfsd4_encode_operation(resp, op);
1090 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1091 * cached page. Replace any futher replay pages from the cache.
1094 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1095 struct nfsd4_sequence *seq)
1097 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1100 dprintk("--> %s entry %p\n", __func__, entry);
1103 * If this is just the sequence operation, we did not keep
1104 * a page in the cache entry because we can just use the
1105 * slot info stored in struct nfsd4_sequence that was checked
1106 * against the slot in nfsd4_sequence().
1108 * This occurs when seq->cachethis is FALSE, or when the client
1109 * session inactivity timer fires and a solo sequence operation
1110 * is sent (lease renewal).
1112 seq->maxslots = resp->cstate.session->se_fchannel.maxreqs;
1114 /* Either returns 0 or nfserr_retry_uncached */
1115 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1116 if (status == nfserr_retry_uncached_rep)
1119 if (!nfsd41_copy_replay_data(resp, entry)) {
1121 * Not enough room to use the replay rpc header, send the
1122 * cached header. Release all the allocated result pages.
1124 svc_free_res_pages(resp->rqstp);
1125 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1128 /* Release all but the first allocated result page */
1130 resp->rqstp->rq_resused--;
1131 svc_free_res_pages(resp->rqstp);
1133 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1134 &entry->ce_respages[1],
1135 entry->ce_resused - 1);
1138 resp->rqstp->rq_resused = entry->ce_resused;
1139 resp->opcnt = entry->ce_opcnt;
1140 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1141 status = entry->ce_status;
1147 * Set the exchange_id flags returned by the server.
1150 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1152 /* pNFS is not supported */
1153 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1155 /* Referrals are supported, Migration is not. */
1156 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1158 /* set the wire flags to return to client. */
1159 clid->flags = new->cl_exchange_flags;
1163 nfsd4_exchange_id(struct svc_rqst *rqstp,
1164 struct nfsd4_compound_state *cstate,
1165 struct nfsd4_exchange_id *exid)
1167 struct nfs4_client *unconf, *conf, *new;
1169 unsigned int strhashval;
1170 char dname[HEXDIR_LEN];
1171 char addr_str[INET6_ADDRSTRLEN];
1172 nfs4_verifier verf = exid->verifier;
1173 struct sockaddr *sa = svc_addr(rqstp);
1175 rpc_ntop(sa, addr_str, sizeof(addr_str));
1176 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1177 "ip_addr=%s flags %x, spa_how %d\n",
1178 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1179 addr_str, exid->flags, exid->spa_how);
1181 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1182 return nfserr_inval;
1184 /* Currently only support SP4_NONE */
1185 switch (exid->spa_how) {
1189 return nfserr_encr_alg_unsupp;
1191 BUG(); /* checked by xdr code */
1193 return nfserr_serverfault; /* no excuse :-/ */
1196 status = nfs4_make_rec_clidname(dname, &exid->clname);
1201 strhashval = clientstr_hashval(dname);
1206 conf = find_confirmed_client_by_str(dname, strhashval, true);
1208 if (!same_verf(&verf, &conf->cl_verifier)) {
1209 /* 18.35.4 case 8 */
1210 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1211 status = nfserr_not_same;
1214 /* Client reboot: destroy old state */
1215 expire_client(conf);
1218 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1219 /* 18.35.4 case 9 */
1220 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1221 status = nfserr_perm;
1224 expire_client(conf);
1228 * Set bit when the owner id and verifier map to an already
1229 * confirmed client id (18.35.3).
1231 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1234 * Falling into 18.35.4 case 2, possible router replay.
1235 * Leave confirmed record intact and return same result.
1237 copy_verf(conf, &verf);
1242 /* 18.35.4 case 7 */
1243 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1244 status = nfserr_noent;
1248 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1251 * Possible retry or client restart. Per 18.35.4 case 4,
1252 * a new unconfirmed record should be generated regardless
1253 * of whether any properties have changed.
1255 expire_client(unconf);
1260 new = create_client(exid->clname, dname);
1262 status = nfserr_resource;
1266 copy_verf(new, &verf);
1267 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1268 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1271 add_to_unconfirmed(new, strhashval);
1273 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1274 exid->clientid.cl_id = new->cl_clientid.cl_id;
1277 nfsd4_set_ex_flags(new, exid);
1279 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1280 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1284 nfs4_unlock_state();
1286 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1291 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1293 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1296 /* The slot is in use, and no response has been sent. */
1298 if (seqid == slot_seqid)
1299 return nfserr_jukebox;
1301 return nfserr_seq_misordered;
1304 if (likely(seqid == slot_seqid + 1))
1307 if (seqid == slot_seqid)
1308 return nfserr_replay_cache;
1310 if (seqid == 1 && (slot_seqid + 1) == 0)
1312 /* Misordered replay or misordered new request */
1313 return nfserr_seq_misordered;
1317 * Cache the create session result into the create session single DRC
1318 * slot cache by saving the xdr structure. sl_seqid has been set.
1319 * Do this for solo or embedded create session operations.
1322 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1323 struct nfsd4_clid_slot *slot, int nfserr)
1325 slot->sl_status = nfserr;
1326 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1330 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1331 struct nfsd4_clid_slot *slot)
1333 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1334 return slot->sl_status;
1338 nfsd4_create_session(struct svc_rqst *rqstp,
1339 struct nfsd4_compound_state *cstate,
1340 struct nfsd4_create_session *cr_ses)
1342 struct sockaddr *sa = svc_addr(rqstp);
1343 struct nfs4_client *conf, *unconf;
1344 struct nfsd4_clid_slot *cs_slot = NULL;
1348 unconf = find_unconfirmed_client(&cr_ses->clientid);
1349 conf = find_confirmed_client(&cr_ses->clientid);
1352 cs_slot = &conf->cl_cs_slot;
1353 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1354 if (status == nfserr_replay_cache) {
1355 dprintk("Got a create_session replay! seqid= %d\n",
1357 /* Return the cached reply status */
1358 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1360 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1361 status = nfserr_seq_misordered;
1362 dprintk("Sequence misordered!\n");
1363 dprintk("Expected seqid= %d but got seqid= %d\n",
1364 cs_slot->sl_seqid, cr_ses->seqid);
1367 cs_slot->sl_seqid++;
1368 } else if (unconf) {
1369 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1370 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1371 status = nfserr_clid_inuse;
1375 cs_slot = &unconf->cl_cs_slot;
1376 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1378 /* an unconfirmed replay returns misordered */
1379 status = nfserr_seq_misordered;
1383 cs_slot->sl_seqid++; /* from 0 to 1 */
1384 move_to_confirmed(unconf);
1387 * We do not support RDMA or persistent sessions
1389 cr_ses->flags &= ~SESSION4_PERSIST;
1390 cr_ses->flags &= ~SESSION4_RDMA;
1394 status = nfserr_stale_clientid;
1398 status = alloc_init_session(rqstp, conf, cr_ses);
1402 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1403 NFS4_MAX_SESSIONID_LEN);
1404 cr_ses->seqid = cs_slot->sl_seqid;
1407 /* cache solo and embedded create sessions under the state lock */
1408 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1410 nfs4_unlock_state();
1411 dprintk("%s returns %d\n", __func__, ntohl(status));
1416 nfsd4_destroy_session(struct svc_rqst *r,
1417 struct nfsd4_compound_state *cstate,
1418 struct nfsd4_destroy_session *sessionid)
1420 struct nfsd4_session *ses;
1421 u32 status = nfserr_badsession;
1424 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1425 * - Should we return nfserr_back_chan_busy if waiting for
1426 * callbacks on to-be-destroyed session?
1427 * - Do we need to clear any callback info from previous session?
1430 dump_sessionid(__func__, &sessionid->sessionid);
1431 spin_lock(&sessionid_lock);
1432 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1434 spin_unlock(&sessionid_lock);
1438 unhash_session(ses);
1439 spin_unlock(&sessionid_lock);
1441 /* wait for callbacks */
1442 shutdown_callback_client(ses->se_client);
1443 nfsd4_put_session(ses);
1446 dprintk("%s returns %d\n", __func__, ntohl(status));
1451 nfsd4_sequence(struct svc_rqst *rqstp,
1452 struct nfsd4_compound_state *cstate,
1453 struct nfsd4_sequence *seq)
1455 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1456 struct nfsd4_session *session;
1457 struct nfsd4_slot *slot;
1460 if (resp->opcnt != 1)
1461 return nfserr_sequence_pos;
1463 spin_lock(&sessionid_lock);
1464 status = nfserr_badsession;
1465 session = find_in_sessionid_hashtbl(&seq->sessionid);
1469 status = nfserr_badslot;
1470 if (seq->slotid >= session->se_fchannel.maxreqs)
1473 slot = &session->se_slots[seq->slotid];
1474 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1476 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1477 if (status == nfserr_replay_cache) {
1478 cstate->slot = slot;
1479 cstate->session = session;
1480 /* Return the cached reply status and set cstate->status
1481 * for nfsd4_svc_encode_compoundres processing */
1482 status = nfsd4_replay_cache_entry(resp, seq);
1483 cstate->status = nfserr_replay_cache;
1489 /* Success! bump slot seqid */
1490 slot->sl_inuse = true;
1491 slot->sl_seqid = seq->seqid;
1492 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1493 /* Always set the cache entry cachethis for solo sequence */
1494 if (nfsd4_is_solo_sequence(resp))
1495 slot->sl_cache_entry.ce_cachethis = 1;
1497 cstate->slot = slot;
1498 cstate->session = session;
1501 /* Renew the clientid on success and on replay.
1502 * Hold a session reference until done processing the compound:
1503 * nfsd4_put_session called only if the cstate slot is set.
1505 renew_client(session->se_client);
1506 nfsd4_get_session(session);
1508 spin_unlock(&sessionid_lock);
1509 dprintk("%s: return %d\n", __func__, ntohl(status));
1514 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1515 struct nfsd4_setclientid *setclid)
1517 struct sockaddr *sa = svc_addr(rqstp);
1518 struct xdr_netobj clname = {
1519 .len = setclid->se_namelen,
1520 .data = setclid->se_name,
1522 nfs4_verifier clverifier = setclid->se_verf;
1523 unsigned int strhashval;
1524 struct nfs4_client *conf, *unconf, *new;
1527 char dname[HEXDIR_LEN];
1529 if (!check_name(clname))
1530 return nfserr_inval;
1532 status = nfs4_make_rec_clidname(dname, &clname);
1537 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1538 * We get here on a DRC miss.
1541 strhashval = clientstr_hashval(dname);
1544 conf = find_confirmed_client_by_str(dname, strhashval, false);
1546 /* RFC 3530 14.2.33 CASE 0: */
1547 status = nfserr_clid_inuse;
1548 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1549 char addr_str[INET6_ADDRSTRLEN];
1550 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1552 dprintk("NFSD: setclientid: string in use by client "
1553 "at %s\n", addr_str);
1558 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1559 * has a description of SETCLIENTID request processing consisting
1560 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1562 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1563 status = nfserr_resource;
1566 * RFC 3530 14.2.33 CASE 4:
1567 * placed first, because it is the normal case
1570 expire_client(unconf);
1571 new = create_client(clname, dname);
1575 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1577 * RFC 3530 14.2.33 CASE 1:
1578 * probable callback update
1581 /* Note this is removing unconfirmed {*x***},
1582 * which is stronger than RFC recommended {vxc**}.
1583 * This has the advantage that there is at most
1584 * one {*x***} in either list at any time.
1586 expire_client(unconf);
1588 new = create_client(clname, dname);
1591 copy_clid(new, conf);
1592 } else if (!unconf) {
1594 * RFC 3530 14.2.33 CASE 2:
1595 * probable client reboot; state will be removed if
1598 new = create_client(clname, dname);
1604 * RFC 3530 14.2.33 CASE 3:
1605 * probable client reboot; state will be removed if
1608 expire_client(unconf);
1609 new = create_client(clname, dname);
1614 copy_verf(new, &clverifier);
1615 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1616 new->cl_flavor = rqstp->rq_flavor;
1617 princ = svc_gss_principal(rqstp);
1619 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1620 if (new->cl_principal == NULL) {
1625 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1627 gen_callback(new, setclid, rpc_get_scope_id(sa));
1628 add_to_unconfirmed(new, strhashval);
1629 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1630 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1631 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1634 nfs4_unlock_state();
1640 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1641 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1642 * bullets, labeled as CASE1 - CASE4 below.
1645 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1646 struct nfsd4_compound_state *cstate,
1647 struct nfsd4_setclientid_confirm *setclientid_confirm)
1649 struct sockaddr *sa = svc_addr(rqstp);
1650 struct nfs4_client *conf, *unconf;
1651 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1652 clientid_t * clid = &setclientid_confirm->sc_clientid;
1655 if (STALE_CLIENTID(clid))
1656 return nfserr_stale_clientid;
1658 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1659 * We get here on a DRC miss.
1664 conf = find_confirmed_client(clid);
1665 unconf = find_unconfirmed_client(clid);
1667 status = nfserr_clid_inuse;
1668 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1670 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1674 * section 14.2.34 of RFC 3530 has a description of
1675 * SETCLIENTID_CONFIRM request processing consisting
1676 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1678 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1680 * RFC 3530 14.2.34 CASE 1:
1683 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1684 status = nfserr_clid_inuse;
1686 /* XXX: We just turn off callbacks until we can handle
1687 * change request correctly. */
1688 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1689 expire_client(unconf);
1693 } else if (conf && !unconf) {
1695 * RFC 3530 14.2.34 CASE 2:
1696 * probable retransmitted request; play it safe and
1699 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1700 status = nfserr_clid_inuse;
1703 } else if (!conf && unconf
1704 && same_verf(&unconf->cl_confirm, &confirm)) {
1706 * RFC 3530 14.2.34 CASE 3:
1707 * Normal case; new or rebooted client:
1709 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1710 status = nfserr_clid_inuse;
1713 clientstr_hashval(unconf->cl_recdir);
1714 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1717 nfsd4_remove_clid_dir(conf);
1718 expire_client(conf);
1720 move_to_confirmed(unconf);
1722 nfsd4_probe_callback(conf);
1725 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1726 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1729 * RFC 3530 14.2.34 CASE 4:
1730 * Client probably hasn't noticed that we rebooted yet.
1732 status = nfserr_stale_clientid;
1734 /* check that we have hit one of the cases...*/
1735 status = nfserr_clid_inuse;
1738 nfs4_unlock_state();
1742 /* OPEN Share state helper functions */
1743 static inline struct nfs4_file *
1744 alloc_init_file(struct inode *ino)
1746 struct nfs4_file *fp;
1747 unsigned int hashval = file_hashval(ino);
1749 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1751 atomic_set(&fp->fi_ref, 1);
1752 INIT_LIST_HEAD(&fp->fi_hash);
1753 INIT_LIST_HEAD(&fp->fi_stateids);
1754 INIT_LIST_HEAD(&fp->fi_delegations);
1755 spin_lock(&recall_lock);
1756 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1757 spin_unlock(&recall_lock);
1758 fp->fi_inode = igrab(ino);
1759 fp->fi_id = current_fileid++;
1760 fp->fi_had_conflict = false;
1767 nfsd4_free_slab(struct kmem_cache **slab)
1771 kmem_cache_destroy(*slab);
1776 nfsd4_free_slabs(void)
1778 nfsd4_free_slab(&stateowner_slab);
1779 nfsd4_free_slab(&file_slab);
1780 nfsd4_free_slab(&stateid_slab);
1781 nfsd4_free_slab(&deleg_slab);
1785 nfsd4_init_slabs(void)
1787 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1788 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1789 if (stateowner_slab == NULL)
1791 file_slab = kmem_cache_create("nfsd4_files",
1792 sizeof(struct nfs4_file), 0, 0, NULL);
1793 if (file_slab == NULL)
1795 stateid_slab = kmem_cache_create("nfsd4_stateids",
1796 sizeof(struct nfs4_stateid), 0, 0, NULL);
1797 if (stateid_slab == NULL)
1799 deleg_slab = kmem_cache_create("nfsd4_delegations",
1800 sizeof(struct nfs4_delegation), 0, 0, NULL);
1801 if (deleg_slab == NULL)
1806 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1811 nfs4_free_stateowner(struct kref *kref)
1813 struct nfs4_stateowner *sop =
1814 container_of(kref, struct nfs4_stateowner, so_ref);
1815 kfree(sop->so_owner.data);
1816 kmem_cache_free(stateowner_slab, sop);
1819 static inline struct nfs4_stateowner *
1820 alloc_stateowner(struct xdr_netobj *owner)
1822 struct nfs4_stateowner *sop;
1824 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1825 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1826 memcpy(sop->so_owner.data, owner->data, owner->len);
1827 sop->so_owner.len = owner->len;
1828 kref_init(&sop->so_ref);
1831 kmem_cache_free(stateowner_slab, sop);
1836 static struct nfs4_stateowner *
1837 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1838 struct nfs4_stateowner *sop;
1839 struct nfs4_replay *rp;
1840 unsigned int idhashval;
1842 if (!(sop = alloc_stateowner(&open->op_owner)))
1844 idhashval = ownerid_hashval(current_ownerid);
1845 INIT_LIST_HEAD(&sop->so_idhash);
1846 INIT_LIST_HEAD(&sop->so_strhash);
1847 INIT_LIST_HEAD(&sop->so_perclient);
1848 INIT_LIST_HEAD(&sop->so_stateids);
1849 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1850 INIT_LIST_HEAD(&sop->so_close_lru);
1852 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1853 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1854 list_add(&sop->so_perclient, &clp->cl_openowners);
1855 sop->so_is_open_owner = 1;
1856 sop->so_id = current_ownerid++;
1857 sop->so_client = clp;
1858 sop->so_seqid = open->op_seqid;
1859 sop->so_confirmed = 0;
1860 rp = &sop->so_replay;
1861 rp->rp_status = nfserr_serverfault;
1863 rp->rp_buf = rp->rp_ibuf;
1868 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1869 struct nfs4_stateowner *sop = open->op_stateowner;
1870 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1872 INIT_LIST_HEAD(&stp->st_hash);
1873 INIT_LIST_HEAD(&stp->st_perstateowner);
1874 INIT_LIST_HEAD(&stp->st_lockowners);
1875 INIT_LIST_HEAD(&stp->st_perfile);
1876 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1877 list_add(&stp->st_perstateowner, &sop->so_stateids);
1878 list_add(&stp->st_perfile, &fp->fi_stateids);
1879 stp->st_stateowner = sop;
1882 stp->st_stateid.si_boot = get_seconds();
1883 stp->st_stateid.si_stateownerid = sop->so_id;
1884 stp->st_stateid.si_fileid = fp->fi_id;
1885 stp->st_stateid.si_generation = 0;
1886 stp->st_access_bmap = 0;
1887 stp->st_deny_bmap = 0;
1888 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1889 &stp->st_access_bmap);
1890 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1891 stp->st_openstp = NULL;
1895 move_to_close_lru(struct nfs4_stateowner *sop)
1897 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1899 list_move_tail(&sop->so_close_lru, &close_lru);
1900 sop->so_time = get_seconds();
1904 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1907 return (sop->so_owner.len == owner->len) &&
1908 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1909 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1912 static struct nfs4_stateowner *
1913 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1915 struct nfs4_stateowner *so = NULL;
1917 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1918 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1924 /* search file_hashtbl[] for file */
1925 static struct nfs4_file *
1926 find_file(struct inode *ino)
1928 unsigned int hashval = file_hashval(ino);
1929 struct nfs4_file *fp;
1931 spin_lock(&recall_lock);
1932 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1933 if (fp->fi_inode == ino) {
1935 spin_unlock(&recall_lock);
1939 spin_unlock(&recall_lock);
1943 static inline int access_valid(u32 x, u32 minorversion)
1945 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1947 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1949 x &= ~NFS4_SHARE_ACCESS_MASK;
1950 if (minorversion && x) {
1951 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1953 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1955 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1962 static inline int deny_valid(u32 x)
1964 /* Note: unlike access bits, deny bits may be zero. */
1965 return x <= NFS4_SHARE_DENY_BOTH;
1969 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1970 * st_{access,deny}_bmap field of the stateid, in order to track not
1971 * only what share bits are currently in force, but also what
1972 * combinations of share bits previous opens have used. This allows us
1973 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1974 * return an error if the client attempt to downgrade to a combination
1975 * of share bits not explicable by closing some of its previous opens.
1977 * XXX: This enforcement is actually incomplete, since we don't keep
1978 * track of access/deny bit combinations; so, e.g., we allow:
1980 * OPEN allow read, deny write
1981 * OPEN allow both, deny none
1982 * DOWNGRADE allow read, deny none
1984 * which we should reject.
1987 set_access(unsigned int *access, unsigned long bmap) {
1991 for (i = 1; i < 4; i++) {
1992 if (test_bit(i, &bmap))
1998 set_deny(unsigned int *deny, unsigned long bmap) {
2002 for (i = 0; i < 4; i++) {
2003 if (test_bit(i, &bmap))
2009 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2010 unsigned int access, deny;
2012 set_access(&access, stp->st_access_bmap);
2013 set_deny(&deny, stp->st_deny_bmap);
2014 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2020 * Called to check deny when READ with all zero stateid or
2021 * WRITE with all zero or all one stateid
2024 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2026 struct inode *ino = current_fh->fh_dentry->d_inode;
2027 struct nfs4_file *fp;
2028 struct nfs4_stateid *stp;
2031 dprintk("NFSD: nfs4_share_conflict\n");
2033 fp = find_file(ino);
2036 ret = nfserr_locked;
2037 /* Search for conflicting share reservations */
2038 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2039 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2040 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2050 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2052 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2053 drop_file_write_access(filp);
2054 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2059 * Spawn a thread to perform a recall on the delegation represented
2060 * by the lease (file_lock)
2062 * Called from break_lease() with lock_kernel() held.
2063 * Note: we assume break_lease will only call this *once* for any given
2067 void nfsd_break_deleg_cb(struct file_lock *fl)
2069 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2071 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2075 /* We're assuming the state code never drops its reference
2076 * without first removing the lease. Since we're in this lease
2077 * callback (and since the lease code is serialized by the kernel
2078 * lock) we know the server hasn't removed the lease yet, we know
2079 * it's safe to take a reference: */
2080 atomic_inc(&dp->dl_count);
2081 atomic_inc(&dp->dl_client->cl_count);
2083 spin_lock(&recall_lock);
2084 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2085 spin_unlock(&recall_lock);
2087 /* only place dl_time is set. protected by lock_kernel*/
2088 dp->dl_time = get_seconds();
2091 * We don't want the locks code to timeout the lease for us;
2092 * we'll remove it ourself if the delegation isn't returned
2095 fl->fl_break_time = 0;
2097 dp->dl_file->fi_had_conflict = true;
2098 nfsd4_cb_recall(dp);
2102 * The file_lock is being reapd.
2104 * Called by locks_free_lock() with lock_kernel() held.
2107 void nfsd_release_deleg_cb(struct file_lock *fl)
2109 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2111 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2113 if (!(fl->fl_flags & FL_LEASE) || !dp)
2115 dp->dl_flock = NULL;
2119 * Set the delegation file_lock back pointer.
2121 * Called from setlease() with lock_kernel() held.
2124 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2126 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2128 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2135 * Called from setlease() with lock_kernel() held
2138 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2140 struct nfs4_delegation *onlistd =
2141 (struct nfs4_delegation *)onlist->fl_owner;
2142 struct nfs4_delegation *tryd =
2143 (struct nfs4_delegation *)try->fl_owner;
2145 if (onlist->fl_lmops != try->fl_lmops)
2148 return onlistd->dl_client == tryd->dl_client;
2153 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2156 return lease_modify(onlist, arg);
2161 static struct lock_manager_operations nfsd_lease_mng_ops = {
2162 .fl_break = nfsd_break_deleg_cb,
2163 .fl_release_private = nfsd_release_deleg_cb,
2164 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2165 .fl_mylease = nfsd_same_client_deleg_cb,
2166 .fl_change = nfsd_change_deleg_cb,
2171 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2172 struct nfsd4_open *open)
2174 clientid_t *clientid = &open->op_clientid;
2175 struct nfs4_client *clp = NULL;
2176 unsigned int strhashval;
2177 struct nfs4_stateowner *sop = NULL;
2179 if (!check_name(open->op_owner))
2180 return nfserr_inval;
2182 if (STALE_CLIENTID(&open->op_clientid))
2183 return nfserr_stale_clientid;
2185 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2186 sop = find_openstateowner_str(strhashval, open);
2187 open->op_stateowner = sop;
2189 /* Make sure the client's lease hasn't expired. */
2190 clp = find_confirmed_client(clientid);
2192 return nfserr_expired;
2195 /* When sessions are used, skip open sequenceid processing */
2196 if (nfsd4_has_session(cstate))
2198 if (!sop->so_confirmed) {
2199 /* Replace unconfirmed owners without checking for replay. */
2200 clp = sop->so_client;
2201 release_openowner(sop);
2202 open->op_stateowner = NULL;
2205 if (open->op_seqid == sop->so_seqid - 1) {
2206 if (sop->so_replay.rp_buflen)
2207 return nfserr_replay_me;
2208 /* The original OPEN failed so spectacularly
2209 * that we don't even have replay data saved!
2210 * Therefore, we have no choice but to continue
2211 * processing this OPEN; presumably, we'll
2212 * fail again for the same reason.
2214 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2217 if (open->op_seqid != sop->so_seqid)
2218 return nfserr_bad_seqid;
2220 if (open->op_stateowner == NULL) {
2221 sop = alloc_init_open_stateowner(strhashval, clp, open);
2223 return nfserr_resource;
2224 open->op_stateowner = sop;
2226 list_del_init(&sop->so_close_lru);
2227 renew_client(sop->so_client);
2231 static inline __be32
2232 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2234 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2235 return nfserr_openmode;
2240 static struct nfs4_delegation *
2241 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2243 struct nfs4_delegation *dp;
2245 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2246 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2253 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2254 struct nfs4_delegation **dp)
2257 __be32 status = nfserr_bad_stateid;
2259 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2262 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2263 RD_STATE : WR_STATE;
2264 status = nfs4_check_delegmode(*dp, flags);
2268 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2272 open->op_stateowner->so_confirmed = 1;
2277 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2279 struct nfs4_stateid *local;
2280 __be32 status = nfserr_share_denied;
2281 struct nfs4_stateowner *sop = open->op_stateowner;
2283 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2284 /* ignore lock owners */
2285 if (local->st_stateowner->so_is_open_owner == 0)
2287 /* remember if we have seen this open owner */
2288 if (local->st_stateowner == sop)
2290 /* check for conflicting share reservations */
2291 if (!test_share(local, open))
2299 static inline struct nfs4_stateid *
2300 nfs4_alloc_stateid(void)
2302 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2306 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2307 struct nfs4_delegation *dp,
2308 struct svc_fh *cur_fh, int flags)
2310 struct nfs4_stateid *stp;
2312 stp = nfs4_alloc_stateid();
2314 return nfserr_resource;
2317 get_file(dp->dl_vfs_file);
2318 stp->st_vfs_file = dp->dl_vfs_file;
2321 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2324 if (status == nfserr_dropit)
2325 status = nfserr_jukebox;
2326 kmem_cache_free(stateid_slab, stp);
2334 static inline __be32
2335 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2336 struct nfsd4_open *open)
2338 struct iattr iattr = {
2339 .ia_valid = ATTR_SIZE,
2342 if (!open->op_truncate)
2344 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2345 return nfserr_inval;
2346 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2350 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2352 struct file *filp = stp->st_vfs_file;
2353 struct inode *inode = filp->f_path.dentry->d_inode;
2354 unsigned int share_access, new_writer;
2357 set_access(&share_access, stp->st_access_bmap);
2358 new_writer = (~share_access) & open->op_share_access
2359 & NFS4_SHARE_ACCESS_WRITE;
2362 int err = get_write_access(inode);
2364 return nfserrno(err);
2365 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2367 return nfserrno(err);
2368 file_take_write(filp);
2370 status = nfsd4_truncate(rqstp, cur_fh, open);
2373 put_write_access(inode);
2376 /* remember the open */
2377 filp->f_mode |= open->op_share_access;
2378 __set_bit(open->op_share_access, &stp->st_access_bmap);
2379 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2386 nfs4_set_claim_prev(struct nfsd4_open *open)
2388 open->op_stateowner->so_confirmed = 1;
2389 open->op_stateowner->so_client->cl_firststate = 1;
2393 * Attempt to hand out a delegation.
2396 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2398 struct nfs4_delegation *dp;
2399 struct nfs4_stateowner *sop = stp->st_stateowner;
2400 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2401 struct file_lock fl, *flp = &fl;
2402 int status, flag = 0;
2404 flag = NFS4_OPEN_DELEGATE_NONE;
2405 open->op_recall = 0;
2406 switch (open->op_claim_type) {
2407 case NFS4_OPEN_CLAIM_PREVIOUS:
2408 if (!atomic_read(&cb->cb_set))
2409 open->op_recall = 1;
2410 flag = open->op_delegate_type;
2411 if (flag == NFS4_OPEN_DELEGATE_NONE)
2414 case NFS4_OPEN_CLAIM_NULL:
2415 /* Let's not give out any delegations till everyone's
2416 * had the chance to reclaim theirs.... */
2417 if (locks_in_grace())
2419 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2421 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2422 flag = NFS4_OPEN_DELEGATE_WRITE;
2424 flag = NFS4_OPEN_DELEGATE_READ;
2430 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2432 flag = NFS4_OPEN_DELEGATE_NONE;
2435 locks_init_lock(&fl);
2436 fl.fl_lmops = &nfsd_lease_mng_ops;
2437 fl.fl_flags = FL_LEASE;
2438 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2439 fl.fl_end = OFFSET_MAX;
2440 fl.fl_owner = (fl_owner_t)dp;
2441 fl.fl_file = stp->st_vfs_file;
2442 fl.fl_pid = current->tgid;
2444 /* vfs_setlease checks to see if delegation should be handed out.
2445 * the lock_manager callbacks fl_mylease and fl_change are used
2447 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2448 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2449 unhash_delegation(dp);
2450 flag = NFS4_OPEN_DELEGATE_NONE;
2454 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2456 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2457 dp->dl_stateid.si_boot,
2458 dp->dl_stateid.si_stateownerid,
2459 dp->dl_stateid.si_fileid,
2460 dp->dl_stateid.si_generation);
2462 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2463 && flag == NFS4_OPEN_DELEGATE_NONE
2464 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2465 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2466 open->op_delegate_type = flag;
2470 * called with nfs4_lock_state() held.
2473 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2475 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2476 struct nfs4_file *fp = NULL;
2477 struct inode *ino = current_fh->fh_dentry->d_inode;
2478 struct nfs4_stateid *stp = NULL;
2479 struct nfs4_delegation *dp = NULL;
2482 status = nfserr_inval;
2483 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2484 || !deny_valid(open->op_share_deny))
2487 * Lookup file; if found, lookup stateid and check open request,
2488 * and check for delegations in the process of being recalled.
2489 * If not found, create the nfs4_file struct
2491 fp = find_file(ino);
2493 if ((status = nfs4_check_open(fp, open, &stp)))
2495 status = nfs4_check_deleg(fp, open, &dp);
2499 status = nfserr_bad_stateid;
2500 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2502 status = nfserr_resource;
2503 fp = alloc_init_file(ino);
2509 * OPEN the file, or upgrade an existing OPEN.
2510 * If truncate fails, the OPEN fails.
2513 /* Stateid was found, this is an OPEN upgrade */
2514 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2517 update_stateid(&stp->st_stateid);
2519 /* Stateid was not found, this is a new OPEN */
2521 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2522 flags |= NFSD_MAY_READ;
2523 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2524 flags |= NFSD_MAY_WRITE;
2525 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2528 init_stateid(stp, fp, open);
2529 status = nfsd4_truncate(rqstp, current_fh, open);
2531 release_open_stateid(stp);
2534 if (nfsd4_has_session(&resp->cstate))
2535 update_stateid(&stp->st_stateid);
2537 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2539 if (nfsd4_has_session(&resp->cstate))
2540 open->op_stateowner->so_confirmed = 1;
2543 * Attempt to hand out a delegation. No error return, because the
2544 * OPEN succeeds even if we fail.
2546 nfs4_open_delegation(current_fh, open, stp);
2550 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2551 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2552 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2556 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2557 nfs4_set_claim_prev(open);
2559 * To finish the open response, we just need to set the rflags.
2561 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2562 if (!open->op_stateowner->so_confirmed &&
2563 !nfsd4_has_session(&resp->cstate))
2564 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2570 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2573 struct nfs4_client *clp;
2577 dprintk("process_renew(%08x/%08x): starting\n",
2578 clid->cl_boot, clid->cl_id);
2579 status = nfserr_stale_clientid;
2580 if (STALE_CLIENTID(clid))
2582 clp = find_confirmed_client(clid);
2583 status = nfserr_expired;
2585 /* We assume the client took too long to RENEW. */
2586 dprintk("nfsd4_renew: clientid not found!\n");
2590 status = nfserr_cb_path_down;
2591 if (!list_empty(&clp->cl_delegations)
2592 && !atomic_read(&clp->cl_cb_conn.cb_set))
2596 nfs4_unlock_state();
2600 struct lock_manager nfsd4_manager = {
2604 nfsd4_end_grace(void)
2606 dprintk("NFSD: end of grace period\n");
2607 nfsd4_recdir_purge_old();
2608 locks_end_grace(&nfsd4_manager);
2612 nfs4_laundromat(void)
2614 struct nfs4_client *clp;
2615 struct nfs4_stateowner *sop;
2616 struct nfs4_delegation *dp;
2617 struct list_head *pos, *next, reaplist;
2618 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2619 time_t t, clientid_val = NFSD_LEASE_TIME;
2620 time_t u, test_val = NFSD_LEASE_TIME;
2624 dprintk("NFSD: laundromat service - starting\n");
2625 if (locks_in_grace())
2627 list_for_each_safe(pos, next, &client_lru) {
2628 clp = list_entry(pos, struct nfs4_client, cl_lru);
2629 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2630 t = clp->cl_time - cutoff;
2631 if (clientid_val > t)
2635 dprintk("NFSD: purging unused client (clientid %08x)\n",
2636 clp->cl_clientid.cl_id);
2637 nfsd4_remove_clid_dir(clp);
2640 INIT_LIST_HEAD(&reaplist);
2641 spin_lock(&recall_lock);
2642 list_for_each_safe(pos, next, &del_recall_lru) {
2643 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2644 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2645 u = dp->dl_time - cutoff;
2650 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2652 list_move(&dp->dl_recall_lru, &reaplist);
2654 spin_unlock(&recall_lock);
2655 list_for_each_safe(pos, next, &reaplist) {
2656 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2657 list_del_init(&dp->dl_recall_lru);
2658 unhash_delegation(dp);
2660 test_val = NFSD_LEASE_TIME;
2661 list_for_each_safe(pos, next, &close_lru) {
2662 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2663 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2664 u = sop->so_time - cutoff;
2669 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2671 release_openowner(sop);
2673 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2674 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2675 nfs4_unlock_state();
2676 return clientid_val;
2679 static struct workqueue_struct *laundry_wq;
2680 static void laundromat_main(struct work_struct *);
2681 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2684 laundromat_main(struct work_struct *not_used)
2688 t = nfs4_laundromat();
2689 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2690 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2693 static struct nfs4_stateowner *
2694 search_close_lru(u32 st_id, int flags)
2696 struct nfs4_stateowner *local = NULL;
2698 if (flags & CLOSE_STATE) {
2699 list_for_each_entry(local, &close_lru, so_close_lru) {
2700 if (local->so_id == st_id)
2708 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2710 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2714 STALE_STATEID(stateid_t *stateid)
2716 if (time_after((unsigned long)boot_time,
2717 (unsigned long)stateid->si_boot)) {
2718 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2719 stateid->si_boot, stateid->si_stateownerid,
2720 stateid->si_fileid, stateid->si_generation);
2727 EXPIRED_STATEID(stateid_t *stateid)
2729 if (time_before((unsigned long)boot_time,
2730 ((unsigned long)stateid->si_boot)) &&
2731 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2732 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2733 stateid->si_boot, stateid->si_stateownerid,
2734 stateid->si_fileid, stateid->si_generation);
2741 stateid_error_map(stateid_t *stateid)
2743 if (STALE_STATEID(stateid))
2744 return nfserr_stale_stateid;
2745 if (EXPIRED_STATEID(stateid))
2746 return nfserr_expired;
2748 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2749 stateid->si_boot, stateid->si_stateownerid,
2750 stateid->si_fileid, stateid->si_generation);
2751 return nfserr_bad_stateid;
2755 access_permit_read(unsigned long access_bmap)
2757 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2758 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2759 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2763 access_permit_write(unsigned long access_bmap)
2765 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2766 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2770 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2772 __be32 status = nfserr_openmode;
2774 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2776 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2783 static inline __be32
2784 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2786 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2788 else if (locks_in_grace()) {
2789 /* Answer in remaining cases depends on existance of
2790 * conflicting state; so we must wait out the grace period. */
2791 return nfserr_grace;
2792 } else if (flags & WR_STATE)
2793 return nfs4_share_conflict(current_fh,
2794 NFS4_SHARE_DENY_WRITE);
2795 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2796 return nfs4_share_conflict(current_fh,
2797 NFS4_SHARE_DENY_READ);
2801 * Allow READ/WRITE during grace period on recovered state only for files
2802 * that are not able to provide mandatory locking.
2805 grace_disallows_io(struct inode *inode)
2807 return locks_in_grace() && mandatory_lock(inode);
2810 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2813 * When sessions are used the stateid generation number is ignored
2816 if ((flags & HAS_SESSION) && in->si_generation == 0)
2819 /* If the client sends us a stateid from the future, it's buggy: */
2820 if (in->si_generation > ref->si_generation)
2821 return nfserr_bad_stateid;
2823 * The following, however, can happen. For example, if the
2824 * client sends an open and some IO at the same time, the open
2825 * may bump si_generation while the IO is still in flight.
2826 * Thanks to hard links and renames, the client never knows what
2827 * file an open will affect. So it could avoid that situation
2828 * only by serializing all opens and IO from the same open
2829 * owner. To recover from the old_stateid error, the client
2830 * will just have to retry the IO:
2832 if (in->si_generation < ref->si_generation)
2833 return nfserr_old_stateid;
2838 static int is_delegation_stateid(stateid_t *stateid)
2840 return stateid->si_fileid == 0;
2844 * Checks for stateid operations
2847 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2848 stateid_t *stateid, int flags, struct file **filpp)
2850 struct nfs4_stateid *stp = NULL;
2851 struct nfs4_delegation *dp = NULL;
2852 struct svc_fh *current_fh = &cstate->current_fh;
2853 struct inode *ino = current_fh->fh_dentry->d_inode;
2859 if (grace_disallows_io(ino))
2860 return nfserr_grace;
2862 if (nfsd4_has_session(cstate))
2863 flags |= HAS_SESSION;
2865 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2866 return check_special_stateids(current_fh, stateid, flags);
2868 status = nfserr_stale_stateid;
2869 if (STALE_STATEID(stateid))
2872 status = nfserr_bad_stateid;
2873 if (is_delegation_stateid(stateid)) {
2874 dp = find_delegation_stateid(ino, stateid);
2876 status = stateid_error_map(stateid);
2879 status = check_stateid_generation(stateid, &dp->dl_stateid,
2883 status = nfs4_check_delegmode(dp, flags);
2886 renew_client(dp->dl_client);
2888 *filpp = dp->dl_vfs_file;
2889 } else { /* open or lock stateid */
2890 stp = find_stateid(stateid, flags);
2892 status = stateid_error_map(stateid);
2895 if (nfs4_check_fh(current_fh, stp))
2897 if (!stp->st_stateowner->so_confirmed)
2899 status = check_stateid_generation(stateid, &stp->st_stateid,
2903 status = nfs4_check_openmode(stp, flags);
2906 renew_client(stp->st_stateowner->so_client);
2908 *filpp = stp->st_vfs_file;
2918 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2919 RD_STATE : WR_STATE;
2923 * Checks for sequence id mutating operations.
2926 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2927 stateid_t *stateid, int flags,
2928 struct nfs4_stateowner **sopp,
2929 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2931 struct nfs4_stateid *stp;
2932 struct nfs4_stateowner *sop;
2933 struct svc_fh *current_fh = &cstate->current_fh;
2936 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2937 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2938 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2939 stateid->si_generation);
2944 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2945 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2946 return nfserr_bad_stateid;
2949 if (STALE_STATEID(stateid))
2950 return nfserr_stale_stateid;
2952 if (nfsd4_has_session(cstate))
2953 flags |= HAS_SESSION;
2956 * We return BAD_STATEID if filehandle doesn't match stateid,
2957 * the confirmed flag is incorrecly set, or the generation
2958 * number is incorrect.
2960 stp = find_stateid(stateid, flags);
2963 * Also, we should make sure this isn't just the result of
2966 sop = search_close_lru(stateid->si_stateownerid, flags);
2968 return stateid_error_map(stateid);
2974 *sopp = sop = stp->st_stateowner;
2977 clientid_t *lockclid = &lock->v.new.clientid;
2978 struct nfs4_client *clp = sop->so_client;
2982 lkflg = setlkflg(lock->lk_type);
2984 if (lock->lk_is_new) {
2985 if (!sop->so_is_open_owner)
2986 return nfserr_bad_stateid;
2987 if (!(flags & HAS_SESSION) &&
2988 !same_clid(&clp->cl_clientid, lockclid))
2989 return nfserr_bad_stateid;
2990 /* stp is the open stateid */
2991 status = nfs4_check_openmode(stp, lkflg);
2995 /* stp is the lock stateid */
2996 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3002 if (nfs4_check_fh(current_fh, stp)) {
3003 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3004 return nfserr_bad_stateid;
3008 * We now validate the seqid and stateid generation numbers.
3009 * For the moment, we ignore the possibility of
3010 * generation number wraparound.
3012 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3015 if (sop->so_confirmed && flags & CONFIRM) {
3016 dprintk("NFSD: preprocess_seqid_op: expected"
3017 " unconfirmed stateowner!\n");
3018 return nfserr_bad_stateid;
3020 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3021 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3022 " confirmed yet!\n");
3023 return nfserr_bad_stateid;
3025 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3028 renew_client(sop->so_client);
3032 if (seqid == sop->so_seqid - 1) {
3033 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3034 /* indicate replay to calling function */
3035 return nfserr_replay_me;
3037 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3038 sop->so_seqid, seqid);
3040 return nfserr_bad_seqid;
3044 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3045 struct nfsd4_open_confirm *oc)
3048 struct nfs4_stateowner *sop;
3049 struct nfs4_stateid *stp;
3051 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3052 (int)cstate->current_fh.fh_dentry->d_name.len,
3053 cstate->current_fh.fh_dentry->d_name.name);
3055 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3061 if ((status = nfs4_preprocess_seqid_op(cstate,
3062 oc->oc_seqid, &oc->oc_req_stateid,
3063 CONFIRM | OPEN_STATE,
3064 &oc->oc_stateowner, &stp, NULL)))
3067 sop = oc->oc_stateowner;
3068 sop->so_confirmed = 1;
3069 update_stateid(&stp->st_stateid);
3070 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3071 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3072 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3073 stp->st_stateid.si_boot,
3074 stp->st_stateid.si_stateownerid,
3075 stp->st_stateid.si_fileid,
3076 stp->st_stateid.si_generation);
3078 nfsd4_create_clid_dir(sop->so_client);
3080 if (oc->oc_stateowner) {
3081 nfs4_get_stateowner(oc->oc_stateowner);
3082 cstate->replay_owner = oc->oc_stateowner;
3084 nfs4_unlock_state();
3090 * unset all bits in union bitmap (bmap) that
3091 * do not exist in share (from successful OPEN_DOWNGRADE)
3094 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3097 for (i = 1; i < 4; i++) {
3098 if ((i & access) != i)
3099 __clear_bit(i, bmap);
3104 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3107 for (i = 0; i < 4; i++) {
3108 if ((i & deny) != i)
3109 __clear_bit(i, bmap);
3114 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3115 struct nfsd4_compound_state *cstate,
3116 struct nfsd4_open_downgrade *od)
3119 struct nfs4_stateid *stp;
3120 unsigned int share_access;
3122 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3123 (int)cstate->current_fh.fh_dentry->d_name.len,
3124 cstate->current_fh.fh_dentry->d_name.name);
3126 if (!access_valid(od->od_share_access, cstate->minorversion)
3127 || !deny_valid(od->od_share_deny))
3128 return nfserr_inval;
3131 if ((status = nfs4_preprocess_seqid_op(cstate,
3135 &od->od_stateowner, &stp, NULL)))
3138 status = nfserr_inval;
3139 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3140 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3141 stp->st_access_bmap, od->od_share_access);
3144 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3145 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3146 stp->st_deny_bmap, od->od_share_deny);
3149 set_access(&share_access, stp->st_access_bmap);
3150 nfs4_file_downgrade(stp->st_vfs_file,
3151 share_access & ~od->od_share_access);
3153 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3154 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3156 update_stateid(&stp->st_stateid);
3157 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3160 if (od->od_stateowner) {
3161 nfs4_get_stateowner(od->od_stateowner);
3162 cstate->replay_owner = od->od_stateowner;
3164 nfs4_unlock_state();
3169 * nfs4_unlock_state() called after encode
3172 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3173 struct nfsd4_close *close)
3176 struct nfs4_stateid *stp;
3178 dprintk("NFSD: nfsd4_close on file %.*s\n",
3179 (int)cstate->current_fh.fh_dentry->d_name.len,
3180 cstate->current_fh.fh_dentry->d_name.name);
3183 /* check close_lru for replay */
3184 if ((status = nfs4_preprocess_seqid_op(cstate,
3187 OPEN_STATE | CLOSE_STATE,
3188 &close->cl_stateowner, &stp, NULL)))
3191 update_stateid(&stp->st_stateid);
3192 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3194 /* release_stateid() calls nfsd_close() if needed */
3195 release_open_stateid(stp);
3197 /* place unused nfs4_stateowners on so_close_lru list to be
3198 * released by the laundromat service after the lease period
3199 * to enable us to handle CLOSE replay
3201 if (list_empty(&close->cl_stateowner->so_stateids))
3202 move_to_close_lru(close->cl_stateowner);
3204 if (close->cl_stateowner) {
3205 nfs4_get_stateowner(close->cl_stateowner);
3206 cstate->replay_owner = close->cl_stateowner;
3208 nfs4_unlock_state();
3213 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3214 struct nfsd4_delegreturn *dr)
3216 struct nfs4_delegation *dp;
3217 stateid_t *stateid = &dr->dr_stateid;
3218 struct inode *inode;
3222 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3224 inode = cstate->current_fh.fh_dentry->d_inode;
3226 if (nfsd4_has_session(cstate))
3227 flags |= HAS_SESSION;
3229 status = nfserr_bad_stateid;
3230 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3232 status = nfserr_stale_stateid;
3233 if (STALE_STATEID(stateid))
3235 status = nfserr_bad_stateid;
3236 if (!is_delegation_stateid(stateid))
3238 dp = find_delegation_stateid(inode, stateid);
3240 status = stateid_error_map(stateid);
3243 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3246 renew_client(dp->dl_client);
3248 unhash_delegation(dp);
3250 nfs4_unlock_state();
3257 * Lock owner state (byte-range locks)
3259 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3260 #define LOCK_HASH_BITS 8
3261 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3262 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3265 end_offset(u64 start, u64 len)
3270 return end >= start ? end: NFS4_MAX_UINT64;
3273 /* last octet in a range */
3275 last_byte_offset(u64 start, u64 len)
3281 return end > start ? end - 1: NFS4_MAX_UINT64;
3284 #define lockownerid_hashval(id) \
3285 ((id) & LOCK_HASH_MASK)
3287 static inline unsigned int
3288 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3289 struct xdr_netobj *ownername)
3291 return (file_hashval(inode) + cl_id
3292 + opaque_hashval(ownername->data, ownername->len))
3296 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3297 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3298 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3300 static struct nfs4_stateid *
3301 find_stateid(stateid_t *stid, int flags)
3303 struct nfs4_stateid *local;
3304 u32 st_id = stid->si_stateownerid;
3305 u32 f_id = stid->si_fileid;
3306 unsigned int hashval;
3308 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3309 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3310 hashval = stateid_hashval(st_id, f_id);
3311 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3312 if ((local->st_stateid.si_stateownerid == st_id) &&
3313 (local->st_stateid.si_fileid == f_id))
3318 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3319 hashval = stateid_hashval(st_id, f_id);
3320 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3321 if ((local->st_stateid.si_stateownerid == st_id) &&
3322 (local->st_stateid.si_fileid == f_id))
3329 static struct nfs4_delegation *
3330 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3332 struct nfs4_file *fp;
3333 struct nfs4_delegation *dl;
3335 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3336 stid->si_boot, stid->si_stateownerid,
3337 stid->si_fileid, stid->si_generation);
3339 fp = find_file(ino);
3342 dl = find_delegation_file(fp, stid);
3348 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3349 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3350 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3351 * locking, this prevents us from being completely protocol-compliant. The
3352 * real solution to this problem is to start using unsigned file offsets in
3353 * the VFS, but this is a very deep change!
3356 nfs4_transform_lock_offset(struct file_lock *lock)
3358 if (lock->fl_start < 0)
3359 lock->fl_start = OFFSET_MAX;
3360 if (lock->fl_end < 0)
3361 lock->fl_end = OFFSET_MAX;
3364 /* Hack!: For now, we're defining this just so we can use a pointer to it
3365 * as a unique cookie to identify our (NFSv4's) posix locks. */
3366 static struct lock_manager_operations nfsd_posix_mng_ops = {
3370 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3372 struct nfs4_stateowner *sop;
3375 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3376 sop = (struct nfs4_stateowner *) fl->fl_owner;
3377 hval = lockownerid_hashval(sop->so_id);
3378 kref_get(&sop->so_ref);
3380 deny->ld_clientid = sop->so_client->cl_clientid;
3382 deny->ld_sop = NULL;
3383 deny->ld_clientid.cl_boot = 0;
3384 deny->ld_clientid.cl_id = 0;
3386 deny->ld_start = fl->fl_start;
3387 deny->ld_length = NFS4_MAX_UINT64;
3388 if (fl->fl_end != NFS4_MAX_UINT64)
3389 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3390 deny->ld_type = NFS4_READ_LT;
3391 if (fl->fl_type != F_RDLCK)
3392 deny->ld_type = NFS4_WRITE_LT;
3395 static struct nfs4_stateowner *
3396 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3397 struct xdr_netobj *owner)
3399 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3400 struct nfs4_stateowner *op;
3402 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3403 if (same_owner_str(op, owner, clid))
3410 * Alloc a lock owner structure.
3411 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3414 * strhashval = lock_ownerstr_hashval
3417 static struct nfs4_stateowner *
3418 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3419 struct nfs4_stateowner *sop;
3420 struct nfs4_replay *rp;
3421 unsigned int idhashval;
3423 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3425 idhashval = lockownerid_hashval(current_ownerid);
3426 INIT_LIST_HEAD(&sop->so_idhash);
3427 INIT_LIST_HEAD(&sop->so_strhash);
3428 INIT_LIST_HEAD(&sop->so_perclient);
3429 INIT_LIST_HEAD(&sop->so_stateids);
3430 INIT_LIST_HEAD(&sop->so_perstateid);
3431 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3433 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3434 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3435 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3436 sop->so_is_open_owner = 0;
3437 sop->so_id = current_ownerid++;
3438 sop->so_client = clp;
3439 /* It is the openowner seqid that will be incremented in encode in the
3440 * case of new lockowners; so increment the lock seqid manually: */
3441 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3442 sop->so_confirmed = 1;
3443 rp = &sop->so_replay;
3444 rp->rp_status = nfserr_serverfault;
3446 rp->rp_buf = rp->rp_ibuf;
3450 static struct nfs4_stateid *
3451 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3453 struct nfs4_stateid *stp;
3454 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3456 stp = nfs4_alloc_stateid();
3459 INIT_LIST_HEAD(&stp->st_hash);
3460 INIT_LIST_HEAD(&stp->st_perfile);
3461 INIT_LIST_HEAD(&stp->st_perstateowner);
3462 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3463 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3464 list_add(&stp->st_perfile, &fp->fi_stateids);
3465 list_add(&stp->st_perstateowner, &sop->so_stateids);
3466 stp->st_stateowner = sop;
3469 stp->st_stateid.si_boot = get_seconds();
3470 stp->st_stateid.si_stateownerid = sop->so_id;
3471 stp->st_stateid.si_fileid = fp->fi_id;
3472 stp->st_stateid.si_generation = 0;
3473 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3474 stp->st_access_bmap = open_stp->st_access_bmap;
3475 stp->st_deny_bmap = open_stp->st_deny_bmap;
3476 stp->st_openstp = open_stp;
3483 check_lock_length(u64 offset, u64 length)
3485 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3486 LOFF_OVERFLOW(offset, length)));
3493 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3494 struct nfsd4_lock *lock)
3496 struct nfs4_stateowner *open_sop = NULL;
3497 struct nfs4_stateowner *lock_sop = NULL;
3498 struct nfs4_stateid *lock_stp;
3500 struct file_lock file_lock;
3501 struct file_lock conflock;
3503 unsigned int strhashval;
3507 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3508 (long long) lock->lk_offset,
3509 (long long) lock->lk_length);
3511 if (check_lock_length(lock->lk_offset, lock->lk_length))
3512 return nfserr_inval;
3514 if ((status = fh_verify(rqstp, &cstate->current_fh,
3515 S_IFREG, NFSD_MAY_LOCK))) {
3516 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3522 if (lock->lk_is_new) {
3524 * Client indicates that this is a new lockowner.
3525 * Use open owner and open stateid to create lock owner and
3528 struct nfs4_stateid *open_stp = NULL;
3529 struct nfs4_file *fp;
3531 status = nfserr_stale_clientid;
3532 if (!nfsd4_has_session(cstate) &&
3533 STALE_CLIENTID(&lock->lk_new_clientid))
3536 /* validate and update open stateid and open seqid */
3537 status = nfs4_preprocess_seqid_op(cstate,
3538 lock->lk_new_open_seqid,
3539 &lock->lk_new_open_stateid,
3541 &lock->lk_replay_owner, &open_stp,
3545 open_sop = lock->lk_replay_owner;
3546 /* create lockowner and lock stateid */
3547 fp = open_stp->st_file;
3548 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3549 open_sop->so_client->cl_clientid.cl_id,
3550 &lock->v.new.owner);
3551 /* XXX: Do we need to check for duplicate stateowners on
3552 * the same file, or should they just be allowed (and
3553 * create new stateids)? */
3554 status = nfserr_resource;
3555 lock_sop = alloc_init_lock_stateowner(strhashval,
3556 open_sop->so_client, open_stp, lock);
3557 if (lock_sop == NULL)
3559 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3560 if (lock_stp == NULL)
3563 /* lock (lock owner + lock stateid) already exists */
3564 status = nfs4_preprocess_seqid_op(cstate,
3565 lock->lk_old_lock_seqid,
3566 &lock->lk_old_lock_stateid,
3568 &lock->lk_replay_owner, &lock_stp, lock);
3571 lock_sop = lock->lk_replay_owner;
3573 /* lock->lk_replay_owner and lock_stp have been created or found */
3574 filp = lock_stp->st_vfs_file;
3576 status = nfserr_grace;
3577 if (locks_in_grace() && !lock->lk_reclaim)
3579 status = nfserr_no_grace;
3580 if (!locks_in_grace() && lock->lk_reclaim)
3583 locks_init_lock(&file_lock);
3584 switch (lock->lk_type) {
3587 file_lock.fl_type = F_RDLCK;
3591 case NFS4_WRITEW_LT:
3592 file_lock.fl_type = F_WRLCK;
3596 status = nfserr_inval;
3599 file_lock.fl_owner = (fl_owner_t)lock_sop;
3600 file_lock.fl_pid = current->tgid;
3601 file_lock.fl_file = filp;
3602 file_lock.fl_flags = FL_POSIX;
3603 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3605 file_lock.fl_start = lock->lk_offset;
3606 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3607 nfs4_transform_lock_offset(&file_lock);
3610 * Try to lock the file in the VFS.
3611 * Note: locks.c uses the BKL to protect the inode's lock list.
3614 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3616 case 0: /* success! */
3617 update_stateid(&lock_stp->st_stateid);
3618 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3622 case (EAGAIN): /* conflock holds conflicting lock */
3623 status = nfserr_denied;
3624 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3625 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3628 status = nfserr_deadlock;
3631 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3632 status = nfserr_resource;
3636 if (status && lock->lk_is_new && lock_sop)
3637 release_lockowner(lock_sop);
3638 if (lock->lk_replay_owner) {
3639 nfs4_get_stateowner(lock->lk_replay_owner);
3640 cstate->replay_owner = lock->lk_replay_owner;
3642 nfs4_unlock_state();
3647 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3648 * so we do a temporary open here just to get an open file to pass to
3649 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3652 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3657 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3660 err = vfs_test_lock(file, lock);
3669 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3670 struct nfsd4_lockt *lockt)
3672 struct inode *inode;
3673 struct file_lock file_lock;
3677 if (locks_in_grace())
3678 return nfserr_grace;
3680 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3681 return nfserr_inval;
3683 lockt->lt_stateowner = NULL;
3686 status = nfserr_stale_clientid;
3687 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3690 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3691 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3692 if (status == nfserr_symlink)
3693 status = nfserr_inval;
3697 inode = cstate->current_fh.fh_dentry->d_inode;
3698 locks_init_lock(&file_lock);
3699 switch (lockt->lt_type) {
3702 file_lock.fl_type = F_RDLCK;
3705 case NFS4_WRITEW_LT:
3706 file_lock.fl_type = F_WRLCK;
3709 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3710 status = nfserr_inval;
3714 lockt->lt_stateowner = find_lockstateowner_str(inode,
3715 &lockt->lt_clientid, &lockt->lt_owner);
3716 if (lockt->lt_stateowner)
3717 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3718 file_lock.fl_pid = current->tgid;
3719 file_lock.fl_flags = FL_POSIX;
3721 file_lock.fl_start = lockt->lt_offset;
3722 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3724 nfs4_transform_lock_offset(&file_lock);
3727 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3729 status = nfserrno(error);
3732 if (file_lock.fl_type != F_UNLCK) {
3733 status = nfserr_denied;
3734 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3737 nfs4_unlock_state();
3742 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3743 struct nfsd4_locku *locku)
3745 struct nfs4_stateid *stp;
3746 struct file *filp = NULL;
3747 struct file_lock file_lock;
3751 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3752 (long long) locku->lu_offset,
3753 (long long) locku->lu_length);
3755 if (check_lock_length(locku->lu_offset, locku->lu_length))
3756 return nfserr_inval;
3760 if ((status = nfs4_preprocess_seqid_op(cstate,
3764 &locku->lu_stateowner, &stp, NULL)))
3767 filp = stp->st_vfs_file;
3769 locks_init_lock(&file_lock);
3770 file_lock.fl_type = F_UNLCK;
3771 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3772 file_lock.fl_pid = current->tgid;
3773 file_lock.fl_file = filp;
3774 file_lock.fl_flags = FL_POSIX;
3775 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3776 file_lock.fl_start = locku->lu_offset;
3778 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3779 nfs4_transform_lock_offset(&file_lock);
3782 * Try to unlock the file in the VFS.
3784 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3786 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3790 * OK, unlock succeeded; the only thing left to do is update the stateid.
3792 update_stateid(&stp->st_stateid);
3793 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3796 if (locku->lu_stateowner) {
3797 nfs4_get_stateowner(locku->lu_stateowner);
3798 cstate->replay_owner = locku->lu_stateowner;
3800 nfs4_unlock_state();
3804 status = nfserrno(err);
3810 * 1: locks held by lockowner
3811 * 0: no locks held by lockowner
3814 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3816 struct file_lock **flpp;
3817 struct inode *inode = filp->f_path.dentry->d_inode;
3821 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3822 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3833 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3834 struct nfsd4_compound_state *cstate,
3835 struct nfsd4_release_lockowner *rlockowner)
3837 clientid_t *clid = &rlockowner->rl_clientid;
3838 struct nfs4_stateowner *sop;
3839 struct nfs4_stateid *stp;
3840 struct xdr_netobj *owner = &rlockowner->rl_owner;
3841 struct list_head matches;
3845 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3846 clid->cl_boot, clid->cl_id);
3848 /* XXX check for lease expiration */
3850 status = nfserr_stale_clientid;
3851 if (STALE_CLIENTID(clid))
3856 status = nfserr_locks_held;
3857 /* XXX: we're doing a linear search through all the lockowners.
3858 * Yipes! For now we'll just hope clients aren't really using
3859 * release_lockowner much, but eventually we have to fix these
3860 * data structures. */
3861 INIT_LIST_HEAD(&matches);
3862 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3863 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3864 if (!same_owner_str(sop, owner, clid))
3866 list_for_each_entry(stp, &sop->so_stateids,
3868 if (check_for_locks(stp->st_vfs_file, sop))
3870 /* Note: so_perclient unused for lockowners,
3871 * so it's OK to fool with here. */
3872 list_add(&sop->so_perclient, &matches);
3876 /* Clients probably won't expect us to return with some (but not all)
3877 * of the lockowner state released; so don't release any until all
3878 * have been checked. */
3880 while (!list_empty(&matches)) {
3881 sop = list_entry(matches.next, struct nfs4_stateowner,
3883 /* unhash_stateowner deletes so_perclient only
3884 * for openowners. */
3885 list_del(&sop->so_perclient);
3886 release_lockowner(sop);
3889 nfs4_unlock_state();
3893 static inline struct nfs4_client_reclaim *
3896 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3900 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3902 unsigned int strhashval = clientstr_hashval(name);
3903 struct nfs4_client *clp;
3905 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3910 * failure => all reset bets are off, nfserr_no_grace...
3913 nfs4_client_to_reclaim(const char *name)
3915 unsigned int strhashval;
3916 struct nfs4_client_reclaim *crp = NULL;
3918 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3919 crp = alloc_reclaim();
3922 strhashval = clientstr_hashval(name);
3923 INIT_LIST_HEAD(&crp->cr_strhash);
3924 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3925 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3926 reclaim_str_hashtbl_size++;
3931 nfs4_release_reclaim(void)
3933 struct nfs4_client_reclaim *crp = NULL;
3936 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3937 while (!list_empty(&reclaim_str_hashtbl[i])) {
3938 crp = list_entry(reclaim_str_hashtbl[i].next,
3939 struct nfs4_client_reclaim, cr_strhash);
3940 list_del(&crp->cr_strhash);
3942 reclaim_str_hashtbl_size--;
3945 BUG_ON(reclaim_str_hashtbl_size);
3949 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3950 static struct nfs4_client_reclaim *
3951 nfs4_find_reclaim_client(clientid_t *clid)
3953 unsigned int strhashval;
3954 struct nfs4_client *clp;
3955 struct nfs4_client_reclaim *crp = NULL;
3958 /* find clientid in conf_id_hashtbl */
3959 clp = find_confirmed_client(clid);
3963 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3964 clp->cl_name.len, clp->cl_name.data,
3967 /* find clp->cl_name in reclaim_str_hashtbl */
3968 strhashval = clientstr_hashval(clp->cl_recdir);
3969 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3970 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3978 * Called from OPEN. Look for clientid in reclaim list.
3981 nfs4_check_open_reclaim(clientid_t *clid)
3983 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3986 /* initialization to perform at module load time: */
3989 nfs4_state_init(void)
3993 status = nfsd4_init_slabs();
3996 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3997 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3998 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3999 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4000 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4001 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4003 for (i = 0; i < SESSION_HASH_SIZE; i++)
4004 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4005 for (i = 0; i < FILE_HASH_SIZE; i++) {
4006 INIT_LIST_HEAD(&file_hashtbl[i]);
4008 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4009 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4010 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4012 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4013 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4014 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4016 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4017 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4018 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4020 memset(&onestateid, ~0, sizeof(stateid_t));
4021 INIT_LIST_HEAD(&close_lru);
4022 INIT_LIST_HEAD(&client_lru);
4023 INIT_LIST_HEAD(&del_recall_lru);
4024 reclaim_str_hashtbl_size = 0;
4029 nfsd4_load_reboot_recovery_data(void)
4034 nfsd4_init_recdir(user_recovery_dirname);
4035 status = nfsd4_recdir_load();
4036 nfs4_unlock_state();
4038 printk("NFSD: Failure reading reboot recovery data\n");
4042 get_nfs4_grace_period(void)
4044 return max(user_lease_time, lease_time) * HZ;
4048 * Since the lifetime of a delegation isn't limited to that of an open, a
4049 * client may quite reasonably hang on to a delegation as long as it has
4050 * the inode cached. This becomes an obvious problem the first time a
4051 * client's inode cache approaches the size of the server's total memory.
4053 * For now we avoid this problem by imposing a hard limit on the number
4054 * of delegations, which varies according to the server's memory size.
4057 set_max_delegations(void)
4060 * Allow at most 4 delegations per megabyte of RAM. Quick
4061 * estimates suggest that in the worst case (where every delegation
4062 * is for a different inode), a delegation could take about 1.5K,
4063 * giving a worst case usage of about 6% of memory.
4065 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4068 /* initialization to perform when the nfsd service is started: */
4071 __nfs4_state_start(void)
4073 unsigned long grace_time;
4075 boot_time = get_seconds();
4076 grace_time = get_nfs4_grace_period();
4077 lease_time = user_lease_time;
4078 locks_start_grace(&nfsd4_manager);
4079 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4081 laundry_wq = create_singlethread_workqueue("nfsd4");
4082 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4083 set_max_delegations();
4087 nfs4_state_start(void)
4091 nfsd4_load_reboot_recovery_data();
4092 __nfs4_state_start();
4098 nfs4_lease_time(void)
4104 __nfs4_state_shutdown(void)
4107 struct nfs4_client *clp = NULL;
4108 struct nfs4_delegation *dp = NULL;
4109 struct list_head *pos, *next, reaplist;
4111 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4112 while (!list_empty(&conf_id_hashtbl[i])) {
4113 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4116 while (!list_empty(&unconf_str_hashtbl[i])) {
4117 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4121 INIT_LIST_HEAD(&reaplist);
4122 spin_lock(&recall_lock);
4123 list_for_each_safe(pos, next, &del_recall_lru) {
4124 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4125 list_move(&dp->dl_recall_lru, &reaplist);
4127 spin_unlock(&recall_lock);
4128 list_for_each_safe(pos, next, &reaplist) {
4129 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4130 list_del_init(&dp->dl_recall_lru);
4131 unhash_delegation(dp);
4134 nfsd4_shutdown_recdir();
4139 nfs4_state_shutdown(void)
4141 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4142 destroy_workqueue(laundry_wq);
4143 locks_end_grace(&nfsd4_manager);
4145 nfs4_release_reclaim();
4146 __nfs4_state_shutdown();
4147 nfs4_unlock_state();
4151 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4152 * accessed when nfsd is starting.
4155 nfs4_set_recdir(char *recdir)
4157 strcpy(user_recovery_dirname, recdir);
4161 * Change the NFSv4 recovery directory to recdir.
4164 nfs4_reset_recoverydir(char *recdir)
4169 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4173 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4174 nfs4_set_recdir(recdir);
4182 nfs4_recoverydir(void)
4184 return user_recovery_dirname;
4188 * Called when leasetime is changed.
4190 * The only way the protocol gives us to handle on-the-fly lease changes is to
4191 * simulate a reboot. Instead of doing that, we just wait till the next time
4192 * we start to register any changes in lease time. If the administrator
4193 * really wants to change the lease time *now*, they can go ahead and bring
4194 * nfsd down and then back up again after changing the lease time.
4196 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4197 * when nfsd is starting
4200 nfs4_reset_lease(time_t leasetime)
4202 user_lease_time = leasetime;