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)
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 cb->cb_minorversion = 0;
925 cb->cb_prog = se->se_callback_prog;
926 cb->cb_ident = se->se_callback_ident;
929 cb->cb_addr.ss_family = AF_UNSPEC;
931 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
932 "will not receive delegations\n",
933 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
939 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
941 struct nfsd4_compoundres *resp = rqstp->rq_resp;
943 resp->cstate.statp = statp;
947 * Dereference the result pages.
950 nfsd4_release_respages(struct page **respages, short resused)
954 dprintk("--> %s\n", __func__);
955 for (i = 0; i < resused; i++) {
958 put_page(respages[i]);
964 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
968 for (i = 0; i < count; i++) {
969 topages[i] = frompages[i];
972 get_page(topages[i]);
977 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
978 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
979 * length of the XDR response is less than se_fmaxresp_cached
980 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
981 * of the reply (e.g. readdir).
983 * Store the base and length of the rq_req.head[0] page
984 * of the NFSv4.1 data, just past the rpc header.
987 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
989 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
990 struct svc_rqst *rqstp = resp->rqstp;
991 struct nfsd4_compoundargs *args = rqstp->rq_argp;
992 struct nfsd4_op *op = &args->ops[resp->opcnt];
993 struct kvec *resv = &rqstp->rq_res.head[0];
995 dprintk("--> %s entry %p\n", __func__, entry);
997 /* Don't cache a failed OP_SEQUENCE. */
998 if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1001 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1002 entry->ce_opcnt = resp->opcnt;
1003 entry->ce_status = resp->cstate.status;
1006 * Don't need a page to cache just the sequence operation - the slot
1010 if (nfsd4_not_cached(resp)) {
1011 entry->ce_resused = 0;
1012 entry->ce_rpchdrlen = 0;
1013 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1014 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1017 entry->ce_resused = rqstp->rq_resused;
1018 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1019 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1020 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1022 entry->ce_datav.iov_base = resp->cstate.statp;
1023 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1024 (char *)page_address(rqstp->rq_respages[0]));
1025 /* Current request rpc header length*/
1026 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1027 (char *)page_address(rqstp->rq_respages[0]);
1031 * We keep the rpc header, but take the nfs reply from the replycache.
1034 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1035 struct nfsd4_cache_entry *entry)
1037 struct svc_rqst *rqstp = resp->rqstp;
1038 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1041 /* Current request rpc header length*/
1042 len = (char *)resp->cstate.statp -
1043 (char *)page_address(rqstp->rq_respages[0]);
1044 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1045 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1046 entry->ce_datav.iov_len);
1049 /* copy the cached reply nfsd data past the current rpc header */
1050 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1051 entry->ce_datav.iov_len);
1052 resv->iov_len = len + entry->ce_datav.iov_len;
1057 * Encode the replay sequence operation from the slot values.
1058 * If cachethis is FALSE encode the uncached rep error on the next
1059 * operation which sets resp->p and increments resp->opcnt for
1060 * nfs4svc_encode_compoundres.
1064 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1065 struct nfsd4_compoundres *resp)
1067 struct nfsd4_op *op;
1068 struct nfsd4_slot *slot = resp->cstate.slot;
1070 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1071 resp->opcnt, resp->cstate.slot->sl_cache_entry.ce_cachethis);
1073 /* Encode the replayed sequence operation */
1074 op = &args->ops[resp->opcnt - 1];
1075 nfsd4_encode_operation(resp, op);
1077 /* Return nfserr_retry_uncached_rep in next operation. */
1078 if (args->opcnt > 1 && slot->sl_cache_entry.ce_cachethis == 0) {
1079 op = &args->ops[resp->opcnt++];
1080 op->status = nfserr_retry_uncached_rep;
1081 nfsd4_encode_operation(resp, op);
1087 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1088 * cached page. Replace any futher replay pages from the cache.
1091 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1092 struct nfsd4_sequence *seq)
1094 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1097 dprintk("--> %s entry %p\n", __func__, entry);
1100 * If this is just the sequence operation, we did not keep
1101 * a page in the cache entry because we can just use the
1102 * slot info stored in struct nfsd4_sequence that was checked
1103 * against the slot in nfsd4_sequence().
1105 * This occurs when seq->cachethis is FALSE, or when the client
1106 * session inactivity timer fires and a solo sequence operation
1107 * is sent (lease renewal).
1109 seq->maxslots = resp->cstate.session->se_fchannel.maxreqs;
1111 /* Either returns 0 or nfserr_retry_uncached */
1112 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1113 if (status == nfserr_retry_uncached_rep)
1116 if (!nfsd41_copy_replay_data(resp, entry)) {
1118 * Not enough room to use the replay rpc header, send the
1119 * cached header. Release all the allocated result pages.
1121 svc_free_res_pages(resp->rqstp);
1122 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1125 /* Release all but the first allocated result page */
1127 resp->rqstp->rq_resused--;
1128 svc_free_res_pages(resp->rqstp);
1130 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1131 &entry->ce_respages[1],
1132 entry->ce_resused - 1);
1135 resp->rqstp->rq_resused = entry->ce_resused;
1136 resp->opcnt = entry->ce_opcnt;
1137 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1138 status = entry->ce_status;
1144 * Set the exchange_id flags returned by the server.
1147 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1149 /* pNFS is not supported */
1150 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1152 /* Referrals are supported, Migration is not. */
1153 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1155 /* set the wire flags to return to client. */
1156 clid->flags = new->cl_exchange_flags;
1160 nfsd4_exchange_id(struct svc_rqst *rqstp,
1161 struct nfsd4_compound_state *cstate,
1162 struct nfsd4_exchange_id *exid)
1164 struct nfs4_client *unconf, *conf, *new;
1166 unsigned int strhashval;
1167 char dname[HEXDIR_LEN];
1168 char addr_str[INET6_ADDRSTRLEN];
1169 nfs4_verifier verf = exid->verifier;
1170 struct sockaddr *sa = svc_addr(rqstp);
1172 rpc_ntop(sa, addr_str, sizeof(addr_str));
1173 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1174 "ip_addr=%s flags %x, spa_how %d\n",
1175 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1176 addr_str, exid->flags, exid->spa_how);
1178 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1179 return nfserr_inval;
1181 /* Currently only support SP4_NONE */
1182 switch (exid->spa_how) {
1186 return nfserr_encr_alg_unsupp;
1188 BUG(); /* checked by xdr code */
1190 return nfserr_serverfault; /* no excuse :-/ */
1193 status = nfs4_make_rec_clidname(dname, &exid->clname);
1198 strhashval = clientstr_hashval(dname);
1203 conf = find_confirmed_client_by_str(dname, strhashval, true);
1205 if (!same_verf(&verf, &conf->cl_verifier)) {
1206 /* 18.35.4 case 8 */
1207 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1208 status = nfserr_not_same;
1211 /* Client reboot: destroy old state */
1212 expire_client(conf);
1215 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1216 /* 18.35.4 case 9 */
1217 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1218 status = nfserr_perm;
1221 expire_client(conf);
1225 * Set bit when the owner id and verifier map to an already
1226 * confirmed client id (18.35.3).
1228 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1231 * Falling into 18.35.4 case 2, possible router replay.
1232 * Leave confirmed record intact and return same result.
1234 copy_verf(conf, &verf);
1239 /* 18.35.4 case 7 */
1240 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1241 status = nfserr_noent;
1245 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1248 * Possible retry or client restart. Per 18.35.4 case 4,
1249 * a new unconfirmed record should be generated regardless
1250 * of whether any properties have changed.
1252 expire_client(unconf);
1257 new = create_client(exid->clname, dname);
1259 status = nfserr_resource;
1263 copy_verf(new, &verf);
1264 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1265 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1268 add_to_unconfirmed(new, strhashval);
1270 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1271 exid->clientid.cl_id = new->cl_clientid.cl_id;
1274 nfsd4_set_ex_flags(new, exid);
1276 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1277 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1281 nfs4_unlock_state();
1283 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1288 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1290 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1293 /* The slot is in use, and no response has been sent. */
1295 if (seqid == slot_seqid)
1296 return nfserr_jukebox;
1298 return nfserr_seq_misordered;
1301 if (likely(seqid == slot_seqid + 1))
1304 if (seqid == slot_seqid)
1305 return nfserr_replay_cache;
1307 if (seqid == 1 && (slot_seqid + 1) == 0)
1309 /* Misordered replay or misordered new request */
1310 return nfserr_seq_misordered;
1314 * Cache the create session result into the create session single DRC
1315 * slot cache by saving the xdr structure. sl_seqid has been set.
1316 * Do this for solo or embedded create session operations.
1319 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1320 struct nfsd4_clid_slot *slot, int nfserr)
1322 slot->sl_status = nfserr;
1323 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1327 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1328 struct nfsd4_clid_slot *slot)
1330 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1331 return slot->sl_status;
1335 nfsd4_create_session(struct svc_rqst *rqstp,
1336 struct nfsd4_compound_state *cstate,
1337 struct nfsd4_create_session *cr_ses)
1339 struct sockaddr *sa = svc_addr(rqstp);
1340 struct nfs4_client *conf, *unconf;
1341 struct nfsd4_clid_slot *cs_slot = NULL;
1345 unconf = find_unconfirmed_client(&cr_ses->clientid);
1346 conf = find_confirmed_client(&cr_ses->clientid);
1349 cs_slot = &conf->cl_cs_slot;
1350 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1351 if (status == nfserr_replay_cache) {
1352 dprintk("Got a create_session replay! seqid= %d\n",
1354 /* Return the cached reply status */
1355 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1357 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1358 status = nfserr_seq_misordered;
1359 dprintk("Sequence misordered!\n");
1360 dprintk("Expected seqid= %d but got seqid= %d\n",
1361 cs_slot->sl_seqid, cr_ses->seqid);
1364 cs_slot->sl_seqid++;
1365 } else if (unconf) {
1366 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1367 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1368 status = nfserr_clid_inuse;
1372 cs_slot = &unconf->cl_cs_slot;
1373 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1375 /* an unconfirmed replay returns misordered */
1376 status = nfserr_seq_misordered;
1380 cs_slot->sl_seqid++; /* from 0 to 1 */
1381 move_to_confirmed(unconf);
1384 * We do not support RDMA or persistent sessions
1386 cr_ses->flags &= ~SESSION4_PERSIST;
1387 cr_ses->flags &= ~SESSION4_RDMA;
1391 status = nfserr_stale_clientid;
1395 status = alloc_init_session(rqstp, conf, cr_ses);
1399 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1400 NFS4_MAX_SESSIONID_LEN);
1401 cr_ses->seqid = cs_slot->sl_seqid;
1404 /* cache solo and embedded create sessions under the state lock */
1405 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1407 nfs4_unlock_state();
1408 dprintk("%s returns %d\n", __func__, ntohl(status));
1413 nfsd4_destroy_session(struct svc_rqst *r,
1414 struct nfsd4_compound_state *cstate,
1415 struct nfsd4_destroy_session *sessionid)
1417 struct nfsd4_session *ses;
1418 u32 status = nfserr_badsession;
1421 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1422 * - Should we return nfserr_back_chan_busy if waiting for
1423 * callbacks on to-be-destroyed session?
1424 * - Do we need to clear any callback info from previous session?
1427 dump_sessionid(__func__, &sessionid->sessionid);
1428 spin_lock(&sessionid_lock);
1429 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1431 spin_unlock(&sessionid_lock);
1435 unhash_session(ses);
1436 spin_unlock(&sessionid_lock);
1438 /* wait for callbacks */
1439 shutdown_callback_client(ses->se_client);
1440 nfsd4_put_session(ses);
1443 dprintk("%s returns %d\n", __func__, ntohl(status));
1448 nfsd4_sequence(struct svc_rqst *rqstp,
1449 struct nfsd4_compound_state *cstate,
1450 struct nfsd4_sequence *seq)
1452 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1453 struct nfsd4_session *session;
1454 struct nfsd4_slot *slot;
1457 if (resp->opcnt != 1)
1458 return nfserr_sequence_pos;
1460 spin_lock(&sessionid_lock);
1461 status = nfserr_badsession;
1462 session = find_in_sessionid_hashtbl(&seq->sessionid);
1466 status = nfserr_badslot;
1467 if (seq->slotid >= session->se_fchannel.maxreqs)
1470 slot = &session->se_slots[seq->slotid];
1471 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1473 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1474 if (status == nfserr_replay_cache) {
1475 cstate->slot = slot;
1476 cstate->session = session;
1477 /* Return the cached reply status and set cstate->status
1478 * for nfsd4_svc_encode_compoundres processing */
1479 status = nfsd4_replay_cache_entry(resp, seq);
1480 cstate->status = nfserr_replay_cache;
1486 /* Success! bump slot seqid */
1487 slot->sl_inuse = true;
1488 slot->sl_seqid = seq->seqid;
1489 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1490 /* Always set the cache entry cachethis for solo sequence */
1491 if (nfsd4_is_solo_sequence(resp))
1492 slot->sl_cache_entry.ce_cachethis = 1;
1494 cstate->slot = slot;
1495 cstate->session = session;
1498 /* Renew the clientid on success and on replay.
1499 * Hold a session reference until done processing the compound:
1500 * nfsd4_put_session called only if the cstate slot is set.
1502 renew_client(session->se_client);
1503 nfsd4_get_session(session);
1505 spin_unlock(&sessionid_lock);
1506 dprintk("%s: return %d\n", __func__, ntohl(status));
1511 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1512 struct nfsd4_setclientid *setclid)
1514 struct sockaddr *sa = svc_addr(rqstp);
1515 struct xdr_netobj clname = {
1516 .len = setclid->se_namelen,
1517 .data = setclid->se_name,
1519 nfs4_verifier clverifier = setclid->se_verf;
1520 unsigned int strhashval;
1521 struct nfs4_client *conf, *unconf, *new;
1524 char dname[HEXDIR_LEN];
1526 if (!check_name(clname))
1527 return nfserr_inval;
1529 status = nfs4_make_rec_clidname(dname, &clname);
1534 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1535 * We get here on a DRC miss.
1538 strhashval = clientstr_hashval(dname);
1541 conf = find_confirmed_client_by_str(dname, strhashval, false);
1543 /* RFC 3530 14.2.33 CASE 0: */
1544 status = nfserr_clid_inuse;
1545 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1546 char addr_str[INET6_ADDRSTRLEN];
1547 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1549 dprintk("NFSD: setclientid: string in use by client "
1550 "at %s\n", addr_str);
1555 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1556 * has a description of SETCLIENTID request processing consisting
1557 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1559 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1560 status = nfserr_resource;
1563 * RFC 3530 14.2.33 CASE 4:
1564 * placed first, because it is the normal case
1567 expire_client(unconf);
1568 new = create_client(clname, dname);
1572 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1574 * RFC 3530 14.2.33 CASE 1:
1575 * probable callback update
1578 /* Note this is removing unconfirmed {*x***},
1579 * which is stronger than RFC recommended {vxc**}.
1580 * This has the advantage that there is at most
1581 * one {*x***} in either list at any time.
1583 expire_client(unconf);
1585 new = create_client(clname, dname);
1588 copy_clid(new, conf);
1589 } else if (!unconf) {
1591 * RFC 3530 14.2.33 CASE 2:
1592 * probable client reboot; state will be removed if
1595 new = create_client(clname, dname);
1601 * RFC 3530 14.2.33 CASE 3:
1602 * probable client reboot; state will be removed if
1605 expire_client(unconf);
1606 new = create_client(clname, dname);
1611 copy_verf(new, &clverifier);
1612 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1613 new->cl_flavor = rqstp->rq_flavor;
1614 princ = svc_gss_principal(rqstp);
1616 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1617 if (new->cl_principal == NULL) {
1622 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1624 gen_callback(new, setclid);
1625 add_to_unconfirmed(new, strhashval);
1626 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1627 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1628 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1631 nfs4_unlock_state();
1637 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1638 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1639 * bullets, labeled as CASE1 - CASE4 below.
1642 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1643 struct nfsd4_compound_state *cstate,
1644 struct nfsd4_setclientid_confirm *setclientid_confirm)
1646 struct sockaddr *sa = svc_addr(rqstp);
1647 struct nfs4_client *conf, *unconf;
1648 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1649 clientid_t * clid = &setclientid_confirm->sc_clientid;
1652 if (STALE_CLIENTID(clid))
1653 return nfserr_stale_clientid;
1655 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1656 * We get here on a DRC miss.
1661 conf = find_confirmed_client(clid);
1662 unconf = find_unconfirmed_client(clid);
1664 status = nfserr_clid_inuse;
1665 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1667 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1671 * section 14.2.34 of RFC 3530 has a description of
1672 * SETCLIENTID_CONFIRM request processing consisting
1673 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1675 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1677 * RFC 3530 14.2.34 CASE 1:
1680 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1681 status = nfserr_clid_inuse;
1683 /* XXX: We just turn off callbacks until we can handle
1684 * change request correctly. */
1685 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1686 expire_client(unconf);
1690 } else if (conf && !unconf) {
1692 * RFC 3530 14.2.34 CASE 2:
1693 * probable retransmitted request; play it safe and
1696 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1697 status = nfserr_clid_inuse;
1700 } else if (!conf && unconf
1701 && same_verf(&unconf->cl_confirm, &confirm)) {
1703 * RFC 3530 14.2.34 CASE 3:
1704 * Normal case; new or rebooted client:
1706 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1707 status = nfserr_clid_inuse;
1710 clientstr_hashval(unconf->cl_recdir);
1711 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1714 nfsd4_remove_clid_dir(conf);
1715 expire_client(conf);
1717 move_to_confirmed(unconf);
1719 nfsd4_probe_callback(conf);
1722 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1723 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1726 * RFC 3530 14.2.34 CASE 4:
1727 * Client probably hasn't noticed that we rebooted yet.
1729 status = nfserr_stale_clientid;
1731 /* check that we have hit one of the cases...*/
1732 status = nfserr_clid_inuse;
1735 nfs4_unlock_state();
1739 /* OPEN Share state helper functions */
1740 static inline struct nfs4_file *
1741 alloc_init_file(struct inode *ino)
1743 struct nfs4_file *fp;
1744 unsigned int hashval = file_hashval(ino);
1746 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1748 atomic_set(&fp->fi_ref, 1);
1749 INIT_LIST_HEAD(&fp->fi_hash);
1750 INIT_LIST_HEAD(&fp->fi_stateids);
1751 INIT_LIST_HEAD(&fp->fi_delegations);
1752 spin_lock(&recall_lock);
1753 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1754 spin_unlock(&recall_lock);
1755 fp->fi_inode = igrab(ino);
1756 fp->fi_id = current_fileid++;
1757 fp->fi_had_conflict = false;
1764 nfsd4_free_slab(struct kmem_cache **slab)
1768 kmem_cache_destroy(*slab);
1773 nfsd4_free_slabs(void)
1775 nfsd4_free_slab(&stateowner_slab);
1776 nfsd4_free_slab(&file_slab);
1777 nfsd4_free_slab(&stateid_slab);
1778 nfsd4_free_slab(&deleg_slab);
1782 nfsd4_init_slabs(void)
1784 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1785 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1786 if (stateowner_slab == NULL)
1788 file_slab = kmem_cache_create("nfsd4_files",
1789 sizeof(struct nfs4_file), 0, 0, NULL);
1790 if (file_slab == NULL)
1792 stateid_slab = kmem_cache_create("nfsd4_stateids",
1793 sizeof(struct nfs4_stateid), 0, 0, NULL);
1794 if (stateid_slab == NULL)
1796 deleg_slab = kmem_cache_create("nfsd4_delegations",
1797 sizeof(struct nfs4_delegation), 0, 0, NULL);
1798 if (deleg_slab == NULL)
1803 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1808 nfs4_free_stateowner(struct kref *kref)
1810 struct nfs4_stateowner *sop =
1811 container_of(kref, struct nfs4_stateowner, so_ref);
1812 kfree(sop->so_owner.data);
1813 kmem_cache_free(stateowner_slab, sop);
1816 static inline struct nfs4_stateowner *
1817 alloc_stateowner(struct xdr_netobj *owner)
1819 struct nfs4_stateowner *sop;
1821 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1822 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1823 memcpy(sop->so_owner.data, owner->data, owner->len);
1824 sop->so_owner.len = owner->len;
1825 kref_init(&sop->so_ref);
1828 kmem_cache_free(stateowner_slab, sop);
1833 static struct nfs4_stateowner *
1834 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1835 struct nfs4_stateowner *sop;
1836 struct nfs4_replay *rp;
1837 unsigned int idhashval;
1839 if (!(sop = alloc_stateowner(&open->op_owner)))
1841 idhashval = ownerid_hashval(current_ownerid);
1842 INIT_LIST_HEAD(&sop->so_idhash);
1843 INIT_LIST_HEAD(&sop->so_strhash);
1844 INIT_LIST_HEAD(&sop->so_perclient);
1845 INIT_LIST_HEAD(&sop->so_stateids);
1846 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1847 INIT_LIST_HEAD(&sop->so_close_lru);
1849 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1850 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1851 list_add(&sop->so_perclient, &clp->cl_openowners);
1852 sop->so_is_open_owner = 1;
1853 sop->so_id = current_ownerid++;
1854 sop->so_client = clp;
1855 sop->so_seqid = open->op_seqid;
1856 sop->so_confirmed = 0;
1857 rp = &sop->so_replay;
1858 rp->rp_status = nfserr_serverfault;
1860 rp->rp_buf = rp->rp_ibuf;
1865 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1866 struct nfs4_stateowner *sop = open->op_stateowner;
1867 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1869 INIT_LIST_HEAD(&stp->st_hash);
1870 INIT_LIST_HEAD(&stp->st_perstateowner);
1871 INIT_LIST_HEAD(&stp->st_lockowners);
1872 INIT_LIST_HEAD(&stp->st_perfile);
1873 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1874 list_add(&stp->st_perstateowner, &sop->so_stateids);
1875 list_add(&stp->st_perfile, &fp->fi_stateids);
1876 stp->st_stateowner = sop;
1879 stp->st_stateid.si_boot = get_seconds();
1880 stp->st_stateid.si_stateownerid = sop->so_id;
1881 stp->st_stateid.si_fileid = fp->fi_id;
1882 stp->st_stateid.si_generation = 0;
1883 stp->st_access_bmap = 0;
1884 stp->st_deny_bmap = 0;
1885 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1886 &stp->st_access_bmap);
1887 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1888 stp->st_openstp = NULL;
1892 move_to_close_lru(struct nfs4_stateowner *sop)
1894 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1896 list_move_tail(&sop->so_close_lru, &close_lru);
1897 sop->so_time = get_seconds();
1901 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1904 return (sop->so_owner.len == owner->len) &&
1905 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1906 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1909 static struct nfs4_stateowner *
1910 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1912 struct nfs4_stateowner *so = NULL;
1914 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1915 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1921 /* search file_hashtbl[] for file */
1922 static struct nfs4_file *
1923 find_file(struct inode *ino)
1925 unsigned int hashval = file_hashval(ino);
1926 struct nfs4_file *fp;
1928 spin_lock(&recall_lock);
1929 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1930 if (fp->fi_inode == ino) {
1932 spin_unlock(&recall_lock);
1936 spin_unlock(&recall_lock);
1940 static inline int access_valid(u32 x, u32 minorversion)
1942 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1944 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1946 x &= ~NFS4_SHARE_ACCESS_MASK;
1947 if (minorversion && x) {
1948 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1950 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1952 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1959 static inline int deny_valid(u32 x)
1961 /* Note: unlike access bits, deny bits may be zero. */
1962 return x <= NFS4_SHARE_DENY_BOTH;
1966 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1967 * st_{access,deny}_bmap field of the stateid, in order to track not
1968 * only what share bits are currently in force, but also what
1969 * combinations of share bits previous opens have used. This allows us
1970 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1971 * return an error if the client attempt to downgrade to a combination
1972 * of share bits not explicable by closing some of its previous opens.
1974 * XXX: This enforcement is actually incomplete, since we don't keep
1975 * track of access/deny bit combinations; so, e.g., we allow:
1977 * OPEN allow read, deny write
1978 * OPEN allow both, deny none
1979 * DOWNGRADE allow read, deny none
1981 * which we should reject.
1984 set_access(unsigned int *access, unsigned long bmap) {
1988 for (i = 1; i < 4; i++) {
1989 if (test_bit(i, &bmap))
1995 set_deny(unsigned int *deny, unsigned long bmap) {
1999 for (i = 0; i < 4; i++) {
2000 if (test_bit(i, &bmap))
2006 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2007 unsigned int access, deny;
2009 set_access(&access, stp->st_access_bmap);
2010 set_deny(&deny, stp->st_deny_bmap);
2011 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2017 * Called to check deny when READ with all zero stateid or
2018 * WRITE with all zero or all one stateid
2021 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2023 struct inode *ino = current_fh->fh_dentry->d_inode;
2024 struct nfs4_file *fp;
2025 struct nfs4_stateid *stp;
2028 dprintk("NFSD: nfs4_share_conflict\n");
2030 fp = find_file(ino);
2033 ret = nfserr_locked;
2034 /* Search for conflicting share reservations */
2035 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2036 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2037 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2047 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2049 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2050 drop_file_write_access(filp);
2051 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2056 * Spawn a thread to perform a recall on the delegation represented
2057 * by the lease (file_lock)
2059 * Called from break_lease() with lock_kernel() held.
2060 * Note: we assume break_lease will only call this *once* for any given
2064 void nfsd_break_deleg_cb(struct file_lock *fl)
2066 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2068 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2072 /* We're assuming the state code never drops its reference
2073 * without first removing the lease. Since we're in this lease
2074 * callback (and since the lease code is serialized by the kernel
2075 * lock) we know the server hasn't removed the lease yet, we know
2076 * it's safe to take a reference: */
2077 atomic_inc(&dp->dl_count);
2078 atomic_inc(&dp->dl_client->cl_count);
2080 spin_lock(&recall_lock);
2081 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2082 spin_unlock(&recall_lock);
2084 /* only place dl_time is set. protected by lock_kernel*/
2085 dp->dl_time = get_seconds();
2088 * We don't want the locks code to timeout the lease for us;
2089 * we'll remove it ourself if the delegation isn't returned
2092 fl->fl_break_time = 0;
2094 dp->dl_file->fi_had_conflict = true;
2095 nfsd4_cb_recall(dp);
2099 * The file_lock is being reapd.
2101 * Called by locks_free_lock() with lock_kernel() held.
2104 void nfsd_release_deleg_cb(struct file_lock *fl)
2106 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2108 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2110 if (!(fl->fl_flags & FL_LEASE) || !dp)
2112 dp->dl_flock = NULL;
2116 * Set the delegation file_lock back pointer.
2118 * Called from setlease() with lock_kernel() held.
2121 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2123 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2125 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2132 * Called from setlease() with lock_kernel() held
2135 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2137 struct nfs4_delegation *onlistd =
2138 (struct nfs4_delegation *)onlist->fl_owner;
2139 struct nfs4_delegation *tryd =
2140 (struct nfs4_delegation *)try->fl_owner;
2142 if (onlist->fl_lmops != try->fl_lmops)
2145 return onlistd->dl_client == tryd->dl_client;
2150 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2153 return lease_modify(onlist, arg);
2158 static struct lock_manager_operations nfsd_lease_mng_ops = {
2159 .fl_break = nfsd_break_deleg_cb,
2160 .fl_release_private = nfsd_release_deleg_cb,
2161 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2162 .fl_mylease = nfsd_same_client_deleg_cb,
2163 .fl_change = nfsd_change_deleg_cb,
2168 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2169 struct nfsd4_open *open)
2171 clientid_t *clientid = &open->op_clientid;
2172 struct nfs4_client *clp = NULL;
2173 unsigned int strhashval;
2174 struct nfs4_stateowner *sop = NULL;
2176 if (!check_name(open->op_owner))
2177 return nfserr_inval;
2179 if (STALE_CLIENTID(&open->op_clientid))
2180 return nfserr_stale_clientid;
2182 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2183 sop = find_openstateowner_str(strhashval, open);
2184 open->op_stateowner = sop;
2186 /* Make sure the client's lease hasn't expired. */
2187 clp = find_confirmed_client(clientid);
2189 return nfserr_expired;
2192 /* When sessions are used, skip open sequenceid processing */
2193 if (nfsd4_has_session(cstate))
2195 if (!sop->so_confirmed) {
2196 /* Replace unconfirmed owners without checking for replay. */
2197 clp = sop->so_client;
2198 release_openowner(sop);
2199 open->op_stateowner = NULL;
2202 if (open->op_seqid == sop->so_seqid - 1) {
2203 if (sop->so_replay.rp_buflen)
2204 return nfserr_replay_me;
2205 /* The original OPEN failed so spectacularly
2206 * that we don't even have replay data saved!
2207 * Therefore, we have no choice but to continue
2208 * processing this OPEN; presumably, we'll
2209 * fail again for the same reason.
2211 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2214 if (open->op_seqid != sop->so_seqid)
2215 return nfserr_bad_seqid;
2217 if (open->op_stateowner == NULL) {
2218 sop = alloc_init_open_stateowner(strhashval, clp, open);
2220 return nfserr_resource;
2221 open->op_stateowner = sop;
2223 list_del_init(&sop->so_close_lru);
2224 renew_client(sop->so_client);
2228 static inline __be32
2229 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2231 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2232 return nfserr_openmode;
2237 static struct nfs4_delegation *
2238 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2240 struct nfs4_delegation *dp;
2242 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2243 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2250 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2251 struct nfs4_delegation **dp)
2254 __be32 status = nfserr_bad_stateid;
2256 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2259 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2260 RD_STATE : WR_STATE;
2261 status = nfs4_check_delegmode(*dp, flags);
2265 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2269 open->op_stateowner->so_confirmed = 1;
2274 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2276 struct nfs4_stateid *local;
2277 __be32 status = nfserr_share_denied;
2278 struct nfs4_stateowner *sop = open->op_stateowner;
2280 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2281 /* ignore lock owners */
2282 if (local->st_stateowner->so_is_open_owner == 0)
2284 /* remember if we have seen this open owner */
2285 if (local->st_stateowner == sop)
2287 /* check for conflicting share reservations */
2288 if (!test_share(local, open))
2296 static inline struct nfs4_stateid *
2297 nfs4_alloc_stateid(void)
2299 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2303 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2304 struct nfs4_delegation *dp,
2305 struct svc_fh *cur_fh, int flags)
2307 struct nfs4_stateid *stp;
2309 stp = nfs4_alloc_stateid();
2311 return nfserr_resource;
2314 get_file(dp->dl_vfs_file);
2315 stp->st_vfs_file = dp->dl_vfs_file;
2318 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2321 if (status == nfserr_dropit)
2322 status = nfserr_jukebox;
2323 kmem_cache_free(stateid_slab, stp);
2331 static inline __be32
2332 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2333 struct nfsd4_open *open)
2335 struct iattr iattr = {
2336 .ia_valid = ATTR_SIZE,
2339 if (!open->op_truncate)
2341 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2342 return nfserr_inval;
2343 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2347 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2349 struct file *filp = stp->st_vfs_file;
2350 struct inode *inode = filp->f_path.dentry->d_inode;
2351 unsigned int share_access, new_writer;
2354 set_access(&share_access, stp->st_access_bmap);
2355 new_writer = (~share_access) & open->op_share_access
2356 & NFS4_SHARE_ACCESS_WRITE;
2359 int err = get_write_access(inode);
2361 return nfserrno(err);
2362 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2364 return nfserrno(err);
2365 file_take_write(filp);
2367 status = nfsd4_truncate(rqstp, cur_fh, open);
2370 put_write_access(inode);
2373 /* remember the open */
2374 filp->f_mode |= open->op_share_access;
2375 __set_bit(open->op_share_access, &stp->st_access_bmap);
2376 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2383 nfs4_set_claim_prev(struct nfsd4_open *open)
2385 open->op_stateowner->so_confirmed = 1;
2386 open->op_stateowner->so_client->cl_firststate = 1;
2390 * Attempt to hand out a delegation.
2393 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2395 struct nfs4_delegation *dp;
2396 struct nfs4_stateowner *sop = stp->st_stateowner;
2397 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2398 struct file_lock fl, *flp = &fl;
2399 int status, flag = 0;
2401 flag = NFS4_OPEN_DELEGATE_NONE;
2402 open->op_recall = 0;
2403 switch (open->op_claim_type) {
2404 case NFS4_OPEN_CLAIM_PREVIOUS:
2405 if (!atomic_read(&cb->cb_set))
2406 open->op_recall = 1;
2407 flag = open->op_delegate_type;
2408 if (flag == NFS4_OPEN_DELEGATE_NONE)
2411 case NFS4_OPEN_CLAIM_NULL:
2412 /* Let's not give out any delegations till everyone's
2413 * had the chance to reclaim theirs.... */
2414 if (locks_in_grace())
2416 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2418 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2419 flag = NFS4_OPEN_DELEGATE_WRITE;
2421 flag = NFS4_OPEN_DELEGATE_READ;
2427 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2429 flag = NFS4_OPEN_DELEGATE_NONE;
2432 locks_init_lock(&fl);
2433 fl.fl_lmops = &nfsd_lease_mng_ops;
2434 fl.fl_flags = FL_LEASE;
2435 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2436 fl.fl_end = OFFSET_MAX;
2437 fl.fl_owner = (fl_owner_t)dp;
2438 fl.fl_file = stp->st_vfs_file;
2439 fl.fl_pid = current->tgid;
2441 /* vfs_setlease checks to see if delegation should be handed out.
2442 * the lock_manager callbacks fl_mylease and fl_change are used
2444 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2445 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2446 unhash_delegation(dp);
2447 flag = NFS4_OPEN_DELEGATE_NONE;
2451 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2453 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2454 dp->dl_stateid.si_boot,
2455 dp->dl_stateid.si_stateownerid,
2456 dp->dl_stateid.si_fileid,
2457 dp->dl_stateid.si_generation);
2459 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2460 && flag == NFS4_OPEN_DELEGATE_NONE
2461 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2462 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2463 open->op_delegate_type = flag;
2467 * called with nfs4_lock_state() held.
2470 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2472 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2473 struct nfs4_file *fp = NULL;
2474 struct inode *ino = current_fh->fh_dentry->d_inode;
2475 struct nfs4_stateid *stp = NULL;
2476 struct nfs4_delegation *dp = NULL;
2479 status = nfserr_inval;
2480 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2481 || !deny_valid(open->op_share_deny))
2484 * Lookup file; if found, lookup stateid and check open request,
2485 * and check for delegations in the process of being recalled.
2486 * If not found, create the nfs4_file struct
2488 fp = find_file(ino);
2490 if ((status = nfs4_check_open(fp, open, &stp)))
2492 status = nfs4_check_deleg(fp, open, &dp);
2496 status = nfserr_bad_stateid;
2497 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2499 status = nfserr_resource;
2500 fp = alloc_init_file(ino);
2506 * OPEN the file, or upgrade an existing OPEN.
2507 * If truncate fails, the OPEN fails.
2510 /* Stateid was found, this is an OPEN upgrade */
2511 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2514 update_stateid(&stp->st_stateid);
2516 /* Stateid was not found, this is a new OPEN */
2518 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2519 flags |= NFSD_MAY_READ;
2520 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2521 flags |= NFSD_MAY_WRITE;
2522 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2525 init_stateid(stp, fp, open);
2526 status = nfsd4_truncate(rqstp, current_fh, open);
2528 release_open_stateid(stp);
2531 if (nfsd4_has_session(&resp->cstate))
2532 update_stateid(&stp->st_stateid);
2534 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2536 if (nfsd4_has_session(&resp->cstate))
2537 open->op_stateowner->so_confirmed = 1;
2540 * Attempt to hand out a delegation. No error return, because the
2541 * OPEN succeeds even if we fail.
2543 nfs4_open_delegation(current_fh, open, stp);
2547 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2548 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2549 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2553 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2554 nfs4_set_claim_prev(open);
2556 * To finish the open response, we just need to set the rflags.
2558 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2559 if (!open->op_stateowner->so_confirmed &&
2560 !nfsd4_has_session(&resp->cstate))
2561 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2567 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2570 struct nfs4_client *clp;
2574 dprintk("process_renew(%08x/%08x): starting\n",
2575 clid->cl_boot, clid->cl_id);
2576 status = nfserr_stale_clientid;
2577 if (STALE_CLIENTID(clid))
2579 clp = find_confirmed_client(clid);
2580 status = nfserr_expired;
2582 /* We assume the client took too long to RENEW. */
2583 dprintk("nfsd4_renew: clientid not found!\n");
2587 status = nfserr_cb_path_down;
2588 if (!list_empty(&clp->cl_delegations)
2589 && !atomic_read(&clp->cl_cb_conn.cb_set))
2593 nfs4_unlock_state();
2597 struct lock_manager nfsd4_manager = {
2601 nfsd4_end_grace(void)
2603 dprintk("NFSD: end of grace period\n");
2604 nfsd4_recdir_purge_old();
2605 locks_end_grace(&nfsd4_manager);
2609 nfs4_laundromat(void)
2611 struct nfs4_client *clp;
2612 struct nfs4_stateowner *sop;
2613 struct nfs4_delegation *dp;
2614 struct list_head *pos, *next, reaplist;
2615 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2616 time_t t, clientid_val = NFSD_LEASE_TIME;
2617 time_t u, test_val = NFSD_LEASE_TIME;
2621 dprintk("NFSD: laundromat service - starting\n");
2622 if (locks_in_grace())
2624 list_for_each_safe(pos, next, &client_lru) {
2625 clp = list_entry(pos, struct nfs4_client, cl_lru);
2626 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2627 t = clp->cl_time - cutoff;
2628 if (clientid_val > t)
2632 dprintk("NFSD: purging unused client (clientid %08x)\n",
2633 clp->cl_clientid.cl_id);
2634 nfsd4_remove_clid_dir(clp);
2637 INIT_LIST_HEAD(&reaplist);
2638 spin_lock(&recall_lock);
2639 list_for_each_safe(pos, next, &del_recall_lru) {
2640 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2641 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2642 u = dp->dl_time - cutoff;
2647 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2649 list_move(&dp->dl_recall_lru, &reaplist);
2651 spin_unlock(&recall_lock);
2652 list_for_each_safe(pos, next, &reaplist) {
2653 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2654 list_del_init(&dp->dl_recall_lru);
2655 unhash_delegation(dp);
2657 test_val = NFSD_LEASE_TIME;
2658 list_for_each_safe(pos, next, &close_lru) {
2659 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2660 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2661 u = sop->so_time - cutoff;
2666 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2668 release_openowner(sop);
2670 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2671 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2672 nfs4_unlock_state();
2673 return clientid_val;
2676 static struct workqueue_struct *laundry_wq;
2677 static void laundromat_main(struct work_struct *);
2678 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2681 laundromat_main(struct work_struct *not_used)
2685 t = nfs4_laundromat();
2686 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2687 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2690 static struct nfs4_stateowner *
2691 search_close_lru(u32 st_id, int flags)
2693 struct nfs4_stateowner *local = NULL;
2695 if (flags & CLOSE_STATE) {
2696 list_for_each_entry(local, &close_lru, so_close_lru) {
2697 if (local->so_id == st_id)
2705 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2707 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2711 STALE_STATEID(stateid_t *stateid)
2713 if (time_after((unsigned long)boot_time,
2714 (unsigned long)stateid->si_boot)) {
2715 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2716 stateid->si_boot, stateid->si_stateownerid,
2717 stateid->si_fileid, stateid->si_generation);
2724 EXPIRED_STATEID(stateid_t *stateid)
2726 if (time_before((unsigned long)boot_time,
2727 ((unsigned long)stateid->si_boot)) &&
2728 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2729 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2730 stateid->si_boot, stateid->si_stateownerid,
2731 stateid->si_fileid, stateid->si_generation);
2738 stateid_error_map(stateid_t *stateid)
2740 if (STALE_STATEID(stateid))
2741 return nfserr_stale_stateid;
2742 if (EXPIRED_STATEID(stateid))
2743 return nfserr_expired;
2745 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2746 stateid->si_boot, stateid->si_stateownerid,
2747 stateid->si_fileid, stateid->si_generation);
2748 return nfserr_bad_stateid;
2752 access_permit_read(unsigned long access_bmap)
2754 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2755 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2756 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2760 access_permit_write(unsigned long access_bmap)
2762 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2763 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2767 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2769 __be32 status = nfserr_openmode;
2771 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2773 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2780 static inline __be32
2781 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2783 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2785 else if (locks_in_grace()) {
2786 /* Answer in remaining cases depends on existance of
2787 * conflicting state; so we must wait out the grace period. */
2788 return nfserr_grace;
2789 } else if (flags & WR_STATE)
2790 return nfs4_share_conflict(current_fh,
2791 NFS4_SHARE_DENY_WRITE);
2792 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2793 return nfs4_share_conflict(current_fh,
2794 NFS4_SHARE_DENY_READ);
2798 * Allow READ/WRITE during grace period on recovered state only for files
2799 * that are not able to provide mandatory locking.
2802 grace_disallows_io(struct inode *inode)
2804 return locks_in_grace() && mandatory_lock(inode);
2807 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2810 * When sessions are used the stateid generation number is ignored
2813 if ((flags & HAS_SESSION) && in->si_generation == 0)
2816 /* If the client sends us a stateid from the future, it's buggy: */
2817 if (in->si_generation > ref->si_generation)
2818 return nfserr_bad_stateid;
2820 * The following, however, can happen. For example, if the
2821 * client sends an open and some IO at the same time, the open
2822 * may bump si_generation while the IO is still in flight.
2823 * Thanks to hard links and renames, the client never knows what
2824 * file an open will affect. So it could avoid that situation
2825 * only by serializing all opens and IO from the same open
2826 * owner. To recover from the old_stateid error, the client
2827 * will just have to retry the IO:
2829 if (in->si_generation < ref->si_generation)
2830 return nfserr_old_stateid;
2835 static int is_delegation_stateid(stateid_t *stateid)
2837 return stateid->si_fileid == 0;
2841 * Checks for stateid operations
2844 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2845 stateid_t *stateid, int flags, struct file **filpp)
2847 struct nfs4_stateid *stp = NULL;
2848 struct nfs4_delegation *dp = NULL;
2849 struct svc_fh *current_fh = &cstate->current_fh;
2850 struct inode *ino = current_fh->fh_dentry->d_inode;
2856 if (grace_disallows_io(ino))
2857 return nfserr_grace;
2859 if (nfsd4_has_session(cstate))
2860 flags |= HAS_SESSION;
2862 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2863 return check_special_stateids(current_fh, stateid, flags);
2865 status = nfserr_stale_stateid;
2866 if (STALE_STATEID(stateid))
2869 status = nfserr_bad_stateid;
2870 if (is_delegation_stateid(stateid)) {
2871 dp = find_delegation_stateid(ino, stateid);
2873 status = stateid_error_map(stateid);
2876 status = check_stateid_generation(stateid, &dp->dl_stateid,
2880 status = nfs4_check_delegmode(dp, flags);
2883 renew_client(dp->dl_client);
2885 *filpp = dp->dl_vfs_file;
2886 } else { /* open or lock stateid */
2887 stp = find_stateid(stateid, flags);
2889 status = stateid_error_map(stateid);
2892 if (nfs4_check_fh(current_fh, stp))
2894 if (!stp->st_stateowner->so_confirmed)
2896 status = check_stateid_generation(stateid, &stp->st_stateid,
2900 status = nfs4_check_openmode(stp, flags);
2903 renew_client(stp->st_stateowner->so_client);
2905 *filpp = stp->st_vfs_file;
2915 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2916 RD_STATE : WR_STATE;
2920 * Checks for sequence id mutating operations.
2923 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2924 stateid_t *stateid, int flags,
2925 struct nfs4_stateowner **sopp,
2926 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2928 struct nfs4_stateid *stp;
2929 struct nfs4_stateowner *sop;
2930 struct svc_fh *current_fh = &cstate->current_fh;
2933 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2934 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2935 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2936 stateid->si_generation);
2941 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2942 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2943 return nfserr_bad_stateid;
2946 if (STALE_STATEID(stateid))
2947 return nfserr_stale_stateid;
2949 if (nfsd4_has_session(cstate))
2950 flags |= HAS_SESSION;
2953 * We return BAD_STATEID if filehandle doesn't match stateid,
2954 * the confirmed flag is incorrecly set, or the generation
2955 * number is incorrect.
2957 stp = find_stateid(stateid, flags);
2960 * Also, we should make sure this isn't just the result of
2963 sop = search_close_lru(stateid->si_stateownerid, flags);
2965 return stateid_error_map(stateid);
2971 *sopp = sop = stp->st_stateowner;
2974 clientid_t *lockclid = &lock->v.new.clientid;
2975 struct nfs4_client *clp = sop->so_client;
2979 lkflg = setlkflg(lock->lk_type);
2981 if (lock->lk_is_new) {
2982 if (!sop->so_is_open_owner)
2983 return nfserr_bad_stateid;
2984 if (!(flags & HAS_SESSION) &&
2985 !same_clid(&clp->cl_clientid, lockclid))
2986 return nfserr_bad_stateid;
2987 /* stp is the open stateid */
2988 status = nfs4_check_openmode(stp, lkflg);
2992 /* stp is the lock stateid */
2993 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2999 if (nfs4_check_fh(current_fh, stp)) {
3000 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3001 return nfserr_bad_stateid;
3005 * We now validate the seqid and stateid generation numbers.
3006 * For the moment, we ignore the possibility of
3007 * generation number wraparound.
3009 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3012 if (sop->so_confirmed && flags & CONFIRM) {
3013 dprintk("NFSD: preprocess_seqid_op: expected"
3014 " unconfirmed stateowner!\n");
3015 return nfserr_bad_stateid;
3017 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3018 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3019 " confirmed yet!\n");
3020 return nfserr_bad_stateid;
3022 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3025 renew_client(sop->so_client);
3029 if (seqid == sop->so_seqid - 1) {
3030 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3031 /* indicate replay to calling function */
3032 return nfserr_replay_me;
3034 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3035 sop->so_seqid, seqid);
3037 return nfserr_bad_seqid;
3041 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3042 struct nfsd4_open_confirm *oc)
3045 struct nfs4_stateowner *sop;
3046 struct nfs4_stateid *stp;
3048 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3049 (int)cstate->current_fh.fh_dentry->d_name.len,
3050 cstate->current_fh.fh_dentry->d_name.name);
3052 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3058 if ((status = nfs4_preprocess_seqid_op(cstate,
3059 oc->oc_seqid, &oc->oc_req_stateid,
3060 CONFIRM | OPEN_STATE,
3061 &oc->oc_stateowner, &stp, NULL)))
3064 sop = oc->oc_stateowner;
3065 sop->so_confirmed = 1;
3066 update_stateid(&stp->st_stateid);
3067 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3068 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3069 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3070 stp->st_stateid.si_boot,
3071 stp->st_stateid.si_stateownerid,
3072 stp->st_stateid.si_fileid,
3073 stp->st_stateid.si_generation);
3075 nfsd4_create_clid_dir(sop->so_client);
3077 if (oc->oc_stateowner) {
3078 nfs4_get_stateowner(oc->oc_stateowner);
3079 cstate->replay_owner = oc->oc_stateowner;
3081 nfs4_unlock_state();
3087 * unset all bits in union bitmap (bmap) that
3088 * do not exist in share (from successful OPEN_DOWNGRADE)
3091 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3094 for (i = 1; i < 4; i++) {
3095 if ((i & access) != i)
3096 __clear_bit(i, bmap);
3101 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3104 for (i = 0; i < 4; i++) {
3105 if ((i & deny) != i)
3106 __clear_bit(i, bmap);
3111 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3112 struct nfsd4_compound_state *cstate,
3113 struct nfsd4_open_downgrade *od)
3116 struct nfs4_stateid *stp;
3117 unsigned int share_access;
3119 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3120 (int)cstate->current_fh.fh_dentry->d_name.len,
3121 cstate->current_fh.fh_dentry->d_name.name);
3123 if (!access_valid(od->od_share_access, cstate->minorversion)
3124 || !deny_valid(od->od_share_deny))
3125 return nfserr_inval;
3128 if ((status = nfs4_preprocess_seqid_op(cstate,
3132 &od->od_stateowner, &stp, NULL)))
3135 status = nfserr_inval;
3136 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3137 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3138 stp->st_access_bmap, od->od_share_access);
3141 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3142 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3143 stp->st_deny_bmap, od->od_share_deny);
3146 set_access(&share_access, stp->st_access_bmap);
3147 nfs4_file_downgrade(stp->st_vfs_file,
3148 share_access & ~od->od_share_access);
3150 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3151 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3153 update_stateid(&stp->st_stateid);
3154 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3157 if (od->od_stateowner) {
3158 nfs4_get_stateowner(od->od_stateowner);
3159 cstate->replay_owner = od->od_stateowner;
3161 nfs4_unlock_state();
3166 * nfs4_unlock_state() called after encode
3169 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3170 struct nfsd4_close *close)
3173 struct nfs4_stateid *stp;
3175 dprintk("NFSD: nfsd4_close on file %.*s\n",
3176 (int)cstate->current_fh.fh_dentry->d_name.len,
3177 cstate->current_fh.fh_dentry->d_name.name);
3180 /* check close_lru for replay */
3181 if ((status = nfs4_preprocess_seqid_op(cstate,
3184 OPEN_STATE | CLOSE_STATE,
3185 &close->cl_stateowner, &stp, NULL)))
3188 update_stateid(&stp->st_stateid);
3189 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3191 /* release_stateid() calls nfsd_close() if needed */
3192 release_open_stateid(stp);
3194 /* place unused nfs4_stateowners on so_close_lru list to be
3195 * released by the laundromat service after the lease period
3196 * to enable us to handle CLOSE replay
3198 if (list_empty(&close->cl_stateowner->so_stateids))
3199 move_to_close_lru(close->cl_stateowner);
3201 if (close->cl_stateowner) {
3202 nfs4_get_stateowner(close->cl_stateowner);
3203 cstate->replay_owner = close->cl_stateowner;
3205 nfs4_unlock_state();
3210 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3211 struct nfsd4_delegreturn *dr)
3213 struct nfs4_delegation *dp;
3214 stateid_t *stateid = &dr->dr_stateid;
3215 struct inode *inode;
3219 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3221 inode = cstate->current_fh.fh_dentry->d_inode;
3223 if (nfsd4_has_session(cstate))
3224 flags |= HAS_SESSION;
3226 status = nfserr_bad_stateid;
3227 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3229 status = nfserr_stale_stateid;
3230 if (STALE_STATEID(stateid))
3232 status = nfserr_bad_stateid;
3233 if (!is_delegation_stateid(stateid))
3235 dp = find_delegation_stateid(inode, stateid);
3237 status = stateid_error_map(stateid);
3240 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3243 renew_client(dp->dl_client);
3245 unhash_delegation(dp);
3247 nfs4_unlock_state();
3254 * Lock owner state (byte-range locks)
3256 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3257 #define LOCK_HASH_BITS 8
3258 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3259 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3262 end_offset(u64 start, u64 len)
3267 return end >= start ? end: NFS4_MAX_UINT64;
3270 /* last octet in a range */
3272 last_byte_offset(u64 start, u64 len)
3278 return end > start ? end - 1: NFS4_MAX_UINT64;
3281 #define lockownerid_hashval(id) \
3282 ((id) & LOCK_HASH_MASK)
3284 static inline unsigned int
3285 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3286 struct xdr_netobj *ownername)
3288 return (file_hashval(inode) + cl_id
3289 + opaque_hashval(ownername->data, ownername->len))
3293 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3294 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3295 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3297 static struct nfs4_stateid *
3298 find_stateid(stateid_t *stid, int flags)
3300 struct nfs4_stateid *local;
3301 u32 st_id = stid->si_stateownerid;
3302 u32 f_id = stid->si_fileid;
3303 unsigned int hashval;
3305 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3306 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3307 hashval = stateid_hashval(st_id, f_id);
3308 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3309 if ((local->st_stateid.si_stateownerid == st_id) &&
3310 (local->st_stateid.si_fileid == f_id))
3315 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3316 hashval = stateid_hashval(st_id, f_id);
3317 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3318 if ((local->st_stateid.si_stateownerid == st_id) &&
3319 (local->st_stateid.si_fileid == f_id))
3326 static struct nfs4_delegation *
3327 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3329 struct nfs4_file *fp;
3330 struct nfs4_delegation *dl;
3332 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3333 stid->si_boot, stid->si_stateownerid,
3334 stid->si_fileid, stid->si_generation);
3336 fp = find_file(ino);
3339 dl = find_delegation_file(fp, stid);
3345 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3346 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3347 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3348 * locking, this prevents us from being completely protocol-compliant. The
3349 * real solution to this problem is to start using unsigned file offsets in
3350 * the VFS, but this is a very deep change!
3353 nfs4_transform_lock_offset(struct file_lock *lock)
3355 if (lock->fl_start < 0)
3356 lock->fl_start = OFFSET_MAX;
3357 if (lock->fl_end < 0)
3358 lock->fl_end = OFFSET_MAX;
3361 /* Hack!: For now, we're defining this just so we can use a pointer to it
3362 * as a unique cookie to identify our (NFSv4's) posix locks. */
3363 static struct lock_manager_operations nfsd_posix_mng_ops = {
3367 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3369 struct nfs4_stateowner *sop;
3372 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3373 sop = (struct nfs4_stateowner *) fl->fl_owner;
3374 hval = lockownerid_hashval(sop->so_id);
3375 kref_get(&sop->so_ref);
3377 deny->ld_clientid = sop->so_client->cl_clientid;
3379 deny->ld_sop = NULL;
3380 deny->ld_clientid.cl_boot = 0;
3381 deny->ld_clientid.cl_id = 0;
3383 deny->ld_start = fl->fl_start;
3384 deny->ld_length = NFS4_MAX_UINT64;
3385 if (fl->fl_end != NFS4_MAX_UINT64)
3386 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3387 deny->ld_type = NFS4_READ_LT;
3388 if (fl->fl_type != F_RDLCK)
3389 deny->ld_type = NFS4_WRITE_LT;
3392 static struct nfs4_stateowner *
3393 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3394 struct xdr_netobj *owner)
3396 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3397 struct nfs4_stateowner *op;
3399 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3400 if (same_owner_str(op, owner, clid))
3407 * Alloc a lock owner structure.
3408 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3411 * strhashval = lock_ownerstr_hashval
3414 static struct nfs4_stateowner *
3415 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3416 struct nfs4_stateowner *sop;
3417 struct nfs4_replay *rp;
3418 unsigned int idhashval;
3420 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3422 idhashval = lockownerid_hashval(current_ownerid);
3423 INIT_LIST_HEAD(&sop->so_idhash);
3424 INIT_LIST_HEAD(&sop->so_strhash);
3425 INIT_LIST_HEAD(&sop->so_perclient);
3426 INIT_LIST_HEAD(&sop->so_stateids);
3427 INIT_LIST_HEAD(&sop->so_perstateid);
3428 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3430 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3431 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3432 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3433 sop->so_is_open_owner = 0;
3434 sop->so_id = current_ownerid++;
3435 sop->so_client = clp;
3436 /* It is the openowner seqid that will be incremented in encode in the
3437 * case of new lockowners; so increment the lock seqid manually: */
3438 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3439 sop->so_confirmed = 1;
3440 rp = &sop->so_replay;
3441 rp->rp_status = nfserr_serverfault;
3443 rp->rp_buf = rp->rp_ibuf;
3447 static struct nfs4_stateid *
3448 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3450 struct nfs4_stateid *stp;
3451 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3453 stp = nfs4_alloc_stateid();
3456 INIT_LIST_HEAD(&stp->st_hash);
3457 INIT_LIST_HEAD(&stp->st_perfile);
3458 INIT_LIST_HEAD(&stp->st_perstateowner);
3459 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3460 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3461 list_add(&stp->st_perfile, &fp->fi_stateids);
3462 list_add(&stp->st_perstateowner, &sop->so_stateids);
3463 stp->st_stateowner = sop;
3466 stp->st_stateid.si_boot = get_seconds();
3467 stp->st_stateid.si_stateownerid = sop->so_id;
3468 stp->st_stateid.si_fileid = fp->fi_id;
3469 stp->st_stateid.si_generation = 0;
3470 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3471 stp->st_access_bmap = open_stp->st_access_bmap;
3472 stp->st_deny_bmap = open_stp->st_deny_bmap;
3473 stp->st_openstp = open_stp;
3480 check_lock_length(u64 offset, u64 length)
3482 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3483 LOFF_OVERFLOW(offset, length)));
3490 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3491 struct nfsd4_lock *lock)
3493 struct nfs4_stateowner *open_sop = NULL;
3494 struct nfs4_stateowner *lock_sop = NULL;
3495 struct nfs4_stateid *lock_stp;
3497 struct file_lock file_lock;
3498 struct file_lock conflock;
3500 unsigned int strhashval;
3504 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3505 (long long) lock->lk_offset,
3506 (long long) lock->lk_length);
3508 if (check_lock_length(lock->lk_offset, lock->lk_length))
3509 return nfserr_inval;
3511 if ((status = fh_verify(rqstp, &cstate->current_fh,
3512 S_IFREG, NFSD_MAY_LOCK))) {
3513 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3519 if (lock->lk_is_new) {
3521 * Client indicates that this is a new lockowner.
3522 * Use open owner and open stateid to create lock owner and
3525 struct nfs4_stateid *open_stp = NULL;
3526 struct nfs4_file *fp;
3528 status = nfserr_stale_clientid;
3529 if (!nfsd4_has_session(cstate) &&
3530 STALE_CLIENTID(&lock->lk_new_clientid))
3533 /* validate and update open stateid and open seqid */
3534 status = nfs4_preprocess_seqid_op(cstate,
3535 lock->lk_new_open_seqid,
3536 &lock->lk_new_open_stateid,
3538 &lock->lk_replay_owner, &open_stp,
3542 open_sop = lock->lk_replay_owner;
3543 /* create lockowner and lock stateid */
3544 fp = open_stp->st_file;
3545 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3546 open_sop->so_client->cl_clientid.cl_id,
3547 &lock->v.new.owner);
3548 /* XXX: Do we need to check for duplicate stateowners on
3549 * the same file, or should they just be allowed (and
3550 * create new stateids)? */
3551 status = nfserr_resource;
3552 lock_sop = alloc_init_lock_stateowner(strhashval,
3553 open_sop->so_client, open_stp, lock);
3554 if (lock_sop == NULL)
3556 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3557 if (lock_stp == NULL)
3560 /* lock (lock owner + lock stateid) already exists */
3561 status = nfs4_preprocess_seqid_op(cstate,
3562 lock->lk_old_lock_seqid,
3563 &lock->lk_old_lock_stateid,
3565 &lock->lk_replay_owner, &lock_stp, lock);
3568 lock_sop = lock->lk_replay_owner;
3570 /* lock->lk_replay_owner and lock_stp have been created or found */
3571 filp = lock_stp->st_vfs_file;
3573 status = nfserr_grace;
3574 if (locks_in_grace() && !lock->lk_reclaim)
3576 status = nfserr_no_grace;
3577 if (!locks_in_grace() && lock->lk_reclaim)
3580 locks_init_lock(&file_lock);
3581 switch (lock->lk_type) {
3584 file_lock.fl_type = F_RDLCK;
3588 case NFS4_WRITEW_LT:
3589 file_lock.fl_type = F_WRLCK;
3593 status = nfserr_inval;
3596 file_lock.fl_owner = (fl_owner_t)lock_sop;
3597 file_lock.fl_pid = current->tgid;
3598 file_lock.fl_file = filp;
3599 file_lock.fl_flags = FL_POSIX;
3600 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3602 file_lock.fl_start = lock->lk_offset;
3603 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3604 nfs4_transform_lock_offset(&file_lock);
3607 * Try to lock the file in the VFS.
3608 * Note: locks.c uses the BKL to protect the inode's lock list.
3611 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3613 case 0: /* success! */
3614 update_stateid(&lock_stp->st_stateid);
3615 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3619 case (EAGAIN): /* conflock holds conflicting lock */
3620 status = nfserr_denied;
3621 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3622 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3625 status = nfserr_deadlock;
3628 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3629 status = nfserr_resource;
3633 if (status && lock->lk_is_new && lock_sop)
3634 release_lockowner(lock_sop);
3635 if (lock->lk_replay_owner) {
3636 nfs4_get_stateowner(lock->lk_replay_owner);
3637 cstate->replay_owner = lock->lk_replay_owner;
3639 nfs4_unlock_state();
3644 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3645 * so we do a temporary open here just to get an open file to pass to
3646 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3649 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3654 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3657 err = vfs_test_lock(file, lock);
3666 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3667 struct nfsd4_lockt *lockt)
3669 struct inode *inode;
3670 struct file_lock file_lock;
3674 if (locks_in_grace())
3675 return nfserr_grace;
3677 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3678 return nfserr_inval;
3680 lockt->lt_stateowner = NULL;
3683 status = nfserr_stale_clientid;
3684 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3687 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3688 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3689 if (status == nfserr_symlink)
3690 status = nfserr_inval;
3694 inode = cstate->current_fh.fh_dentry->d_inode;
3695 locks_init_lock(&file_lock);
3696 switch (lockt->lt_type) {
3699 file_lock.fl_type = F_RDLCK;
3702 case NFS4_WRITEW_LT:
3703 file_lock.fl_type = F_WRLCK;
3706 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3707 status = nfserr_inval;
3711 lockt->lt_stateowner = find_lockstateowner_str(inode,
3712 &lockt->lt_clientid, &lockt->lt_owner);
3713 if (lockt->lt_stateowner)
3714 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3715 file_lock.fl_pid = current->tgid;
3716 file_lock.fl_flags = FL_POSIX;
3718 file_lock.fl_start = lockt->lt_offset;
3719 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3721 nfs4_transform_lock_offset(&file_lock);
3724 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3726 status = nfserrno(error);
3729 if (file_lock.fl_type != F_UNLCK) {
3730 status = nfserr_denied;
3731 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3734 nfs4_unlock_state();
3739 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3740 struct nfsd4_locku *locku)
3742 struct nfs4_stateid *stp;
3743 struct file *filp = NULL;
3744 struct file_lock file_lock;
3748 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3749 (long long) locku->lu_offset,
3750 (long long) locku->lu_length);
3752 if (check_lock_length(locku->lu_offset, locku->lu_length))
3753 return nfserr_inval;
3757 if ((status = nfs4_preprocess_seqid_op(cstate,
3761 &locku->lu_stateowner, &stp, NULL)))
3764 filp = stp->st_vfs_file;
3766 locks_init_lock(&file_lock);
3767 file_lock.fl_type = F_UNLCK;
3768 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3769 file_lock.fl_pid = current->tgid;
3770 file_lock.fl_file = filp;
3771 file_lock.fl_flags = FL_POSIX;
3772 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3773 file_lock.fl_start = locku->lu_offset;
3775 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3776 nfs4_transform_lock_offset(&file_lock);
3779 * Try to unlock the file in the VFS.
3781 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3783 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3787 * OK, unlock succeeded; the only thing left to do is update the stateid.
3789 update_stateid(&stp->st_stateid);
3790 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3793 if (locku->lu_stateowner) {
3794 nfs4_get_stateowner(locku->lu_stateowner);
3795 cstate->replay_owner = locku->lu_stateowner;
3797 nfs4_unlock_state();
3801 status = nfserrno(err);
3807 * 1: locks held by lockowner
3808 * 0: no locks held by lockowner
3811 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3813 struct file_lock **flpp;
3814 struct inode *inode = filp->f_path.dentry->d_inode;
3818 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3819 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3830 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3831 struct nfsd4_compound_state *cstate,
3832 struct nfsd4_release_lockowner *rlockowner)
3834 clientid_t *clid = &rlockowner->rl_clientid;
3835 struct nfs4_stateowner *sop;
3836 struct nfs4_stateid *stp;
3837 struct xdr_netobj *owner = &rlockowner->rl_owner;
3838 struct list_head matches;
3842 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3843 clid->cl_boot, clid->cl_id);
3845 /* XXX check for lease expiration */
3847 status = nfserr_stale_clientid;
3848 if (STALE_CLIENTID(clid))
3853 status = nfserr_locks_held;
3854 /* XXX: we're doing a linear search through all the lockowners.
3855 * Yipes! For now we'll just hope clients aren't really using
3856 * release_lockowner much, but eventually we have to fix these
3857 * data structures. */
3858 INIT_LIST_HEAD(&matches);
3859 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3860 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3861 if (!same_owner_str(sop, owner, clid))
3863 list_for_each_entry(stp, &sop->so_stateids,
3865 if (check_for_locks(stp->st_vfs_file, sop))
3867 /* Note: so_perclient unused for lockowners,
3868 * so it's OK to fool with here. */
3869 list_add(&sop->so_perclient, &matches);
3873 /* Clients probably won't expect us to return with some (but not all)
3874 * of the lockowner state released; so don't release any until all
3875 * have been checked. */
3877 while (!list_empty(&matches)) {
3878 sop = list_entry(matches.next, struct nfs4_stateowner,
3880 /* unhash_stateowner deletes so_perclient only
3881 * for openowners. */
3882 list_del(&sop->so_perclient);
3883 release_lockowner(sop);
3886 nfs4_unlock_state();
3890 static inline struct nfs4_client_reclaim *
3893 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3897 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3899 unsigned int strhashval = clientstr_hashval(name);
3900 struct nfs4_client *clp;
3902 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3907 * failure => all reset bets are off, nfserr_no_grace...
3910 nfs4_client_to_reclaim(const char *name)
3912 unsigned int strhashval;
3913 struct nfs4_client_reclaim *crp = NULL;
3915 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3916 crp = alloc_reclaim();
3919 strhashval = clientstr_hashval(name);
3920 INIT_LIST_HEAD(&crp->cr_strhash);
3921 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3922 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3923 reclaim_str_hashtbl_size++;
3928 nfs4_release_reclaim(void)
3930 struct nfs4_client_reclaim *crp = NULL;
3933 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3934 while (!list_empty(&reclaim_str_hashtbl[i])) {
3935 crp = list_entry(reclaim_str_hashtbl[i].next,
3936 struct nfs4_client_reclaim, cr_strhash);
3937 list_del(&crp->cr_strhash);
3939 reclaim_str_hashtbl_size--;
3942 BUG_ON(reclaim_str_hashtbl_size);
3946 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3947 static struct nfs4_client_reclaim *
3948 nfs4_find_reclaim_client(clientid_t *clid)
3950 unsigned int strhashval;
3951 struct nfs4_client *clp;
3952 struct nfs4_client_reclaim *crp = NULL;
3955 /* find clientid in conf_id_hashtbl */
3956 clp = find_confirmed_client(clid);
3960 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3961 clp->cl_name.len, clp->cl_name.data,
3964 /* find clp->cl_name in reclaim_str_hashtbl */
3965 strhashval = clientstr_hashval(clp->cl_recdir);
3966 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3967 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3975 * Called from OPEN. Look for clientid in reclaim list.
3978 nfs4_check_open_reclaim(clientid_t *clid)
3980 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3983 /* initialization to perform at module load time: */
3986 nfs4_state_init(void)
3990 status = nfsd4_init_slabs();
3993 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3994 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3995 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3996 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3997 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3998 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4000 for (i = 0; i < SESSION_HASH_SIZE; i++)
4001 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4002 for (i = 0; i < FILE_HASH_SIZE; i++) {
4003 INIT_LIST_HEAD(&file_hashtbl[i]);
4005 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4006 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4007 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4009 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4010 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4011 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4013 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4014 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4015 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4017 memset(&onestateid, ~0, sizeof(stateid_t));
4018 INIT_LIST_HEAD(&close_lru);
4019 INIT_LIST_HEAD(&client_lru);
4020 INIT_LIST_HEAD(&del_recall_lru);
4021 reclaim_str_hashtbl_size = 0;
4026 nfsd4_load_reboot_recovery_data(void)
4031 nfsd4_init_recdir(user_recovery_dirname);
4032 status = nfsd4_recdir_load();
4033 nfs4_unlock_state();
4035 printk("NFSD: Failure reading reboot recovery data\n");
4039 get_nfs4_grace_period(void)
4041 return max(user_lease_time, lease_time) * HZ;
4045 * Since the lifetime of a delegation isn't limited to that of an open, a
4046 * client may quite reasonably hang on to a delegation as long as it has
4047 * the inode cached. This becomes an obvious problem the first time a
4048 * client's inode cache approaches the size of the server's total memory.
4050 * For now we avoid this problem by imposing a hard limit on the number
4051 * of delegations, which varies according to the server's memory size.
4054 set_max_delegations(void)
4057 * Allow at most 4 delegations per megabyte of RAM. Quick
4058 * estimates suggest that in the worst case (where every delegation
4059 * is for a different inode), a delegation could take about 1.5K,
4060 * giving a worst case usage of about 6% of memory.
4062 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4065 /* initialization to perform when the nfsd service is started: */
4068 __nfs4_state_start(void)
4070 unsigned long grace_time;
4072 boot_time = get_seconds();
4073 grace_time = get_nfs4_grace_period();
4074 lease_time = user_lease_time;
4075 locks_start_grace(&nfsd4_manager);
4076 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4078 laundry_wq = create_singlethread_workqueue("nfsd4");
4079 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4080 set_max_delegations();
4084 nfs4_state_start(void)
4088 nfsd4_load_reboot_recovery_data();
4089 __nfs4_state_start();
4095 nfs4_lease_time(void)
4101 __nfs4_state_shutdown(void)
4104 struct nfs4_client *clp = NULL;
4105 struct nfs4_delegation *dp = NULL;
4106 struct list_head *pos, *next, reaplist;
4108 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4109 while (!list_empty(&conf_id_hashtbl[i])) {
4110 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4113 while (!list_empty(&unconf_str_hashtbl[i])) {
4114 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4118 INIT_LIST_HEAD(&reaplist);
4119 spin_lock(&recall_lock);
4120 list_for_each_safe(pos, next, &del_recall_lru) {
4121 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4122 list_move(&dp->dl_recall_lru, &reaplist);
4124 spin_unlock(&recall_lock);
4125 list_for_each_safe(pos, next, &reaplist) {
4126 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4127 list_del_init(&dp->dl_recall_lru);
4128 unhash_delegation(dp);
4131 nfsd4_shutdown_recdir();
4136 nfs4_state_shutdown(void)
4138 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4139 destroy_workqueue(laundry_wq);
4140 locks_end_grace(&nfsd4_manager);
4142 nfs4_release_reclaim();
4143 __nfs4_state_shutdown();
4144 nfs4_unlock_state();
4148 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4149 * accessed when nfsd is starting.
4152 nfs4_set_recdir(char *recdir)
4154 strcpy(user_recovery_dirname, recdir);
4158 * Change the NFSv4 recovery directory to recdir.
4161 nfs4_reset_recoverydir(char *recdir)
4166 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4170 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4171 nfs4_set_recdir(recdir);
4179 nfs4_recoverydir(void)
4181 return user_recovery_dirname;
4185 * Called when leasetime is changed.
4187 * The only way the protocol gives us to handle on-the-fly lease changes is to
4188 * simulate a reboot. Instead of doing that, we just wait till the next time
4189 * we start to register any changes in lease time. If the administrator
4190 * really wants to change the lease time *now*, they can go ahead and bring
4191 * nfsd down and then back up again after changing the lease time.
4193 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4194 * when nfsd is starting
4197 nfs4_reset_lease(time_t leasetime)
4199 user_lease_time = leasetime;