2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58 #include <linux/sunrpc/clnt.h>
60 #define NFSDDBG_FACILITY NFSDDBG_PROC
63 static time_t lease_time = 90; /* default lease time */
64 static time_t user_lease_time = 90;
65 static time_t boot_time;
66 static u32 current_ownerid = 1;
67 static u32 current_fileid = 1;
68 static u32 current_delegid = 1;
70 static stateid_t zerostateid; /* bits all 0 */
71 static stateid_t onestateid; /* bits all 1 */
72 static u64 current_sessionid = 1;
74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
75 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
77 /* forward declarations */
78 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
79 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
80 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
81 static void nfs4_set_recdir(char *recdir);
85 /* Currently used for almost all code touching nfsv4 state: */
86 static DEFINE_MUTEX(client_mutex);
89 * Currently used for the del_recall_lru and file hash table. In an
90 * effort to decrease the scope of the client_mutex, this spinlock may
91 * eventually cover more:
93 static DEFINE_SPINLOCK(recall_lock);
95 static struct kmem_cache *stateowner_slab = NULL;
96 static struct kmem_cache *file_slab = NULL;
97 static struct kmem_cache *stateid_slab = NULL;
98 static struct kmem_cache *deleg_slab = NULL;
101 nfs4_lock_state(void)
103 mutex_lock(&client_mutex);
107 nfs4_unlock_state(void)
109 mutex_unlock(&client_mutex);
113 opaque_hashval(const void *ptr, int nbytes)
115 unsigned char *cptr = (unsigned char *) ptr;
125 static struct list_head del_recall_lru;
128 put_nfs4_file(struct nfs4_file *fi)
130 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
131 list_del(&fi->fi_hash);
132 spin_unlock(&recall_lock);
134 kmem_cache_free(file_slab, fi);
139 get_nfs4_file(struct nfs4_file *fi)
141 atomic_inc(&fi->fi_ref);
144 static int num_delegations;
145 unsigned int max_delegations;
148 * Open owner state (share locks)
151 /* hash tables for nfs4_stateowner */
152 #define OWNER_HASH_BITS 8
153 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
154 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
156 #define ownerid_hashval(id) \
157 ((id) & OWNER_HASH_MASK)
158 #define ownerstr_hashval(clientid, ownername) \
159 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
161 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
162 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
164 /* hash table for nfs4_file */
165 #define FILE_HASH_BITS 8
166 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
167 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
168 /* hash table for (open)nfs4_stateid */
169 #define STATEID_HASH_BITS 10
170 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
171 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
173 #define file_hashval(x) \
174 hash_ptr(x, FILE_HASH_BITS)
175 #define stateid_hashval(owner_id, file_id) \
176 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
178 static struct list_head file_hashtbl[FILE_HASH_SIZE];
179 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
181 static struct nfs4_delegation *
182 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
184 struct nfs4_delegation *dp;
185 struct nfs4_file *fp = stp->st_file;
186 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
188 dprintk("NFSD alloc_init_deleg\n");
189 if (fp->fi_had_conflict)
191 if (num_delegations > max_delegations)
193 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
197 INIT_LIST_HEAD(&dp->dl_perfile);
198 INIT_LIST_HEAD(&dp->dl_perclnt);
199 INIT_LIST_HEAD(&dp->dl_recall_lru);
204 get_file(stp->st_vfs_file);
205 dp->dl_vfs_file = stp->st_vfs_file;
207 dp->dl_ident = cb->cb_ident;
208 dp->dl_stateid.si_boot = get_seconds();
209 dp->dl_stateid.si_stateownerid = current_delegid++;
210 dp->dl_stateid.si_fileid = 0;
211 dp->dl_stateid.si_generation = 0;
212 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
214 atomic_set(&dp->dl_count, 1);
215 list_add(&dp->dl_perfile, &fp->fi_delegations);
216 list_add(&dp->dl_perclnt, &clp->cl_delegations);
221 nfs4_put_delegation(struct nfs4_delegation *dp)
223 if (atomic_dec_and_test(&dp->dl_count)) {
224 dprintk("NFSD: freeing dp %p\n",dp);
225 put_nfs4_file(dp->dl_file);
226 kmem_cache_free(deleg_slab, dp);
231 /* Remove the associated file_lock first, then remove the delegation.
232 * lease_modify() is called to remove the FS_LEASE file_lock from
233 * the i_flock list, eventually calling nfsd's lock_manager
234 * fl_release_callback.
237 nfs4_close_delegation(struct nfs4_delegation *dp)
239 struct file *filp = dp->dl_vfs_file;
241 dprintk("NFSD: close_delegation dp %p\n",dp);
242 dp->dl_vfs_file = NULL;
243 /* The following nfsd_close may not actually close the file,
244 * but we want to remove the lease in any case. */
246 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
250 /* Called under the state lock. */
252 unhash_delegation(struct nfs4_delegation *dp)
254 list_del_init(&dp->dl_perfile);
255 list_del_init(&dp->dl_perclnt);
256 spin_lock(&recall_lock);
257 list_del_init(&dp->dl_recall_lru);
258 spin_unlock(&recall_lock);
259 nfs4_close_delegation(dp);
260 nfs4_put_delegation(dp);
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS 4
269 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
272 #define clientid_hashval(id) \
273 ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
277 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278 * used in reboot/reset lease grace period processing
280 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281 * setclientid_confirmed info.
283 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
286 * client_lru holds client queue ordered by nfs4_client.cl_time
289 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290 * for last close replay.
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
303 list_del(&stp->st_hash);
304 list_del(&stp->st_perfile);
305 list_del(&stp->st_perstateowner);
308 static void free_generic_stateid(struct nfs4_stateid *stp)
310 put_nfs4_file(stp->st_file);
311 kmem_cache_free(stateid_slab, stp);
314 static void release_lock_stateid(struct nfs4_stateid *stp)
316 unhash_generic_stateid(stp);
317 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318 free_generic_stateid(stp);
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
323 struct nfs4_stateid *stp;
325 list_del(&sop->so_idhash);
326 list_del(&sop->so_strhash);
327 list_del(&sop->so_perstateid);
328 while (!list_empty(&sop->so_stateids)) {
329 stp = list_first_entry(&sop->so_stateids,
330 struct nfs4_stateid, st_perstateowner);
331 release_lock_stateid(stp);
335 static void release_lockowner(struct nfs4_stateowner *sop)
337 unhash_lockowner(sop);
338 nfs4_put_stateowner(sop);
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
344 struct nfs4_stateowner *lock_sop;
346 while (!list_empty(&open_stp->st_lockowners)) {
347 lock_sop = list_entry(open_stp->st_lockowners.next,
348 struct nfs4_stateowner, so_perstateid);
349 /* list_del(&open_stp->st_lockowners); */
350 BUG_ON(lock_sop->so_is_open_owner);
351 release_lockowner(lock_sop);
355 static void release_open_stateid(struct nfs4_stateid *stp)
357 unhash_generic_stateid(stp);
358 release_stateid_lockowners(stp);
359 nfsd_close(stp->st_vfs_file);
360 free_generic_stateid(stp);
363 static void unhash_openowner(struct nfs4_stateowner *sop)
365 struct nfs4_stateid *stp;
367 list_del(&sop->so_idhash);
368 list_del(&sop->so_strhash);
369 list_del(&sop->so_perclient);
370 list_del(&sop->so_perstateid); /* XXX: necessary? */
371 while (!list_empty(&sop->so_stateids)) {
372 stp = list_first_entry(&sop->so_stateids,
373 struct nfs4_stateid, st_perstateowner);
374 release_open_stateid(stp);
378 static void release_openowner(struct nfs4_stateowner *sop)
380 unhash_openowner(sop);
381 list_del(&sop->so_close_lru);
382 nfs4_put_stateowner(sop);
385 static DEFINE_SPINLOCK(sessionid_lock);
386 #define SESSION_HASH_SIZE 512
387 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
390 hash_sessionid(struct nfs4_sessionid *sessionid)
392 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
394 return sid->sequence % SESSION_HASH_SIZE;
398 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
400 u32 *ptr = (u32 *)(&sessionid->data[0]);
401 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
405 gen_sessionid(struct nfsd4_session *ses)
407 struct nfs4_client *clp = ses->se_client;
408 struct nfsd4_sessionid *sid;
410 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
411 sid->clientid = clp->cl_clientid;
412 sid->sequence = current_sessionid++;
417 * Give the client the number of slots it requests bound by
418 * NFSD_MAX_SLOTS_PER_SESSION and by nfsd_drc_max_mem.
420 * If we run out of reserved DRC memory we should (up to a point) re-negotiate
421 * active sessions and reduce their slot usage to make rooom for new
422 * connections. For now we just fail the create session.
424 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
428 if (fchan->maxreqs < 1)
430 else if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
431 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
433 mem = fchan->maxreqs * NFSD_SLOT_CACHE_SIZE;
435 spin_lock(&nfsd_drc_lock);
436 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem)
437 mem = ((nfsd_drc_max_mem - nfsd_drc_mem_used) /
438 NFSD_SLOT_CACHE_SIZE) * NFSD_SLOT_CACHE_SIZE;
439 nfsd_drc_mem_used += mem;
440 spin_unlock(&nfsd_drc_lock);
442 fchan->maxreqs = mem / NFSD_SLOT_CACHE_SIZE;
443 if (fchan->maxreqs == 0)
444 return nfserr_resource;
449 * fchan holds the client values on input, and the server values on output
451 static int init_forechannel_attrs(struct svc_rqst *rqstp,
452 struct nfsd4_channel_attrs *session_fchan,
453 struct nfsd4_channel_attrs *fchan)
456 __u32 maxcount = svc_max_payload(rqstp);
458 /* headerpadsz set to zero in encode routine */
460 /* Use the client's max request and max response size if possible */
461 if (fchan->maxreq_sz > maxcount)
462 fchan->maxreq_sz = maxcount;
463 session_fchan->maxreq_sz = fchan->maxreq_sz;
465 if (fchan->maxresp_sz > maxcount)
466 fchan->maxresp_sz = maxcount;
467 session_fchan->maxresp_sz = fchan->maxresp_sz;
469 session_fchan->maxresp_cached = NFSD_SLOT_CACHE_SIZE;
470 fchan->maxresp_cached = session_fchan->maxresp_cached;
472 /* Use the client's maxops if possible */
473 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
474 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
475 session_fchan->maxops = fchan->maxops;
477 /* FIXME: Error means no more DRC pages so the server should
478 * recover pages from existing sessions. For now fail session
481 status = set_forechannel_maxreqs(fchan);
483 session_fchan->maxreqs = fchan->maxreqs;
488 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
489 struct nfsd4_create_session *cses)
491 struct nfsd4_session *new, tmp;
492 int idx, status = nfserr_resource, slotsize;
494 memset(&tmp, 0, sizeof(tmp));
496 /* FIXME: For now, we just accept the client back channel attributes. */
497 tmp.se_bchannel = cses->back_channel;
498 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
499 &cses->fore_channel);
503 /* allocate struct nfsd4_session and slot table in one piece */
504 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot);
505 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
509 memcpy(new, &tmp, sizeof(*new));
511 new->se_client = clp;
513 idx = hash_sessionid(&new->se_sessionid);
514 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
515 NFS4_MAX_SESSIONID_LEN);
517 new->se_flags = cses->flags;
518 kref_init(&new->se_ref);
519 spin_lock(&sessionid_lock);
520 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
521 list_add(&new->se_perclnt, &clp->cl_sessions);
522 spin_unlock(&sessionid_lock);
529 /* caller must hold sessionid_lock */
530 static struct nfsd4_session *
531 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
533 struct nfsd4_session *elem;
536 dump_sessionid(__func__, sessionid);
537 idx = hash_sessionid(sessionid);
538 dprintk("%s: idx is %d\n", __func__, idx);
539 /* Search in the appropriate list */
540 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
541 dump_sessionid("list traversal", &elem->se_sessionid);
542 if (!memcmp(elem->se_sessionid.data, sessionid->data,
543 NFS4_MAX_SESSIONID_LEN)) {
548 dprintk("%s: session not found\n", __func__);
552 /* caller must hold sessionid_lock */
554 unhash_session(struct nfsd4_session *ses)
556 list_del(&ses->se_hash);
557 list_del(&ses->se_perclnt);
561 release_session(struct nfsd4_session *ses)
563 spin_lock(&sessionid_lock);
565 spin_unlock(&sessionid_lock);
566 nfsd4_put_session(ses);
569 static void nfsd4_release_respages(struct page **respages, short resused);
572 free_session(struct kref *kref)
574 struct nfsd4_session *ses;
577 ses = container_of(kref, struct nfsd4_session, se_ref);
578 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
579 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
580 nfsd4_release_respages(e->ce_respages, e->ce_resused);
582 spin_lock(&nfsd_drc_lock);
583 nfsd_drc_mem_used -= ses->se_fchannel.maxreqs * NFSD_SLOT_CACHE_SIZE;
584 spin_unlock(&nfsd_drc_lock);
589 renew_client(struct nfs4_client *clp)
592 * Move client to the end to the LRU list.
594 dprintk("renewing client (clientid %08x/%08x)\n",
595 clp->cl_clientid.cl_boot,
596 clp->cl_clientid.cl_id);
597 list_move_tail(&clp->cl_lru, &client_lru);
598 clp->cl_time = get_seconds();
601 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
603 STALE_CLIENTID(clientid_t *clid)
605 if (clid->cl_boot == boot_time)
607 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
608 clid->cl_boot, clid->cl_id, boot_time);
613 * XXX Should we use a slab cache ?
614 * This type of memory management is somewhat inefficient, but we use it
615 * anyway since SETCLIENTID is not a common operation.
617 static struct nfs4_client *alloc_client(struct xdr_netobj name)
619 struct nfs4_client *clp;
621 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
624 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
625 if (clp->cl_name.data == NULL) {
629 memcpy(clp->cl_name.data, name.data, name.len);
630 clp->cl_name.len = name.len;
635 shutdown_callback_client(struct nfs4_client *clp)
637 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
641 * Callback threads take a reference on the client, so there
642 * should be no outstanding callbacks at this point.
644 clp->cl_cb_conn.cb_client = NULL;
645 rpc_shutdown_client(clnt);
647 if (clp->cl_cb_conn.cb_cred) {
648 put_rpccred(clp->cl_cb_conn.cb_cred);
649 clp->cl_cb_conn.cb_cred = NULL;
654 free_client(struct nfs4_client *clp)
656 shutdown_callback_client(clp);
657 if (clp->cl_cred.cr_group_info)
658 put_group_info(clp->cl_cred.cr_group_info);
659 kfree(clp->cl_principal);
660 kfree(clp->cl_name.data);
665 put_nfs4_client(struct nfs4_client *clp)
667 if (atomic_dec_and_test(&clp->cl_count))
672 expire_client(struct nfs4_client *clp)
674 struct nfs4_stateowner *sop;
675 struct nfs4_delegation *dp;
676 struct list_head reaplist;
678 dprintk("NFSD: expire_client cl_count %d\n",
679 atomic_read(&clp->cl_count));
681 INIT_LIST_HEAD(&reaplist);
682 spin_lock(&recall_lock);
683 while (!list_empty(&clp->cl_delegations)) {
684 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
685 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
687 list_del_init(&dp->dl_perclnt);
688 list_move(&dp->dl_recall_lru, &reaplist);
690 spin_unlock(&recall_lock);
691 while (!list_empty(&reaplist)) {
692 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
693 list_del_init(&dp->dl_recall_lru);
694 unhash_delegation(dp);
696 list_del(&clp->cl_idhash);
697 list_del(&clp->cl_strhash);
698 list_del(&clp->cl_lru);
699 while (!list_empty(&clp->cl_openowners)) {
700 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
701 release_openowner(sop);
703 while (!list_empty(&clp->cl_sessions)) {
704 struct nfsd4_session *ses;
705 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
707 release_session(ses);
709 put_nfs4_client(clp);
712 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
714 struct nfs4_client *clp;
716 clp = alloc_client(name);
719 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
720 atomic_set(&clp->cl_count, 1);
721 atomic_set(&clp->cl_cb_conn.cb_set, 0);
722 INIT_LIST_HEAD(&clp->cl_idhash);
723 INIT_LIST_HEAD(&clp->cl_strhash);
724 INIT_LIST_HEAD(&clp->cl_openowners);
725 INIT_LIST_HEAD(&clp->cl_delegations);
726 INIT_LIST_HEAD(&clp->cl_sessions);
727 INIT_LIST_HEAD(&clp->cl_lru);
731 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
733 memcpy(target->cl_verifier.data, source->data,
734 sizeof(target->cl_verifier.data));
737 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
739 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
740 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
743 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
745 target->cr_uid = source->cr_uid;
746 target->cr_gid = source->cr_gid;
747 target->cr_group_info = source->cr_group_info;
748 get_group_info(target->cr_group_info);
751 static int same_name(const char *n1, const char *n2)
753 return 0 == memcmp(n1, n2, HEXDIR_LEN);
757 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
759 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
763 same_clid(clientid_t *cl1, clientid_t *cl2)
765 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
768 /* XXX what about NGROUP */
770 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
772 return cr1->cr_uid == cr2->cr_uid;
775 static void gen_clid(struct nfs4_client *clp)
777 static u32 current_clientid = 1;
779 clp->cl_clientid.cl_boot = boot_time;
780 clp->cl_clientid.cl_id = current_clientid++;
783 static void gen_confirm(struct nfs4_client *clp)
788 p = (u32 *)clp->cl_confirm.data;
789 *p++ = get_seconds();
793 static int check_name(struct xdr_netobj name)
797 if (name.len > NFS4_OPAQUE_LIMIT) {
798 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
805 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
807 unsigned int idhashval;
809 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
810 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
811 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
812 list_add_tail(&clp->cl_lru, &client_lru);
813 clp->cl_time = get_seconds();
817 move_to_confirmed(struct nfs4_client *clp)
819 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
820 unsigned int strhashval;
822 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
823 list_del_init(&clp->cl_strhash);
824 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
825 strhashval = clientstr_hashval(clp->cl_recdir);
826 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
830 static struct nfs4_client *
831 find_confirmed_client(clientid_t *clid)
833 struct nfs4_client *clp;
834 unsigned int idhashval = clientid_hashval(clid->cl_id);
836 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
837 if (same_clid(&clp->cl_clientid, clid))
843 static struct nfs4_client *
844 find_unconfirmed_client(clientid_t *clid)
846 struct nfs4_client *clp;
847 unsigned int idhashval = clientid_hashval(clid->cl_id);
849 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
850 if (same_clid(&clp->cl_clientid, clid))
857 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
858 * parameter. Matching is based on the fact the at least one of the
859 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
861 * FIXME: we need to unify the clientid namespaces for nfsv4.x
862 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
863 * and SET_CLIENTID{,_CONFIRM}
866 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
868 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
869 return use_exchange_id == has_exchange_flags;
872 static struct nfs4_client *
873 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
874 bool use_exchange_id)
876 struct nfs4_client *clp;
878 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
879 if (same_name(clp->cl_recdir, dname) &&
880 match_clientid_establishment(clp, use_exchange_id))
886 static struct nfs4_client *
887 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
888 bool use_exchange_id)
890 struct nfs4_client *clp;
892 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
893 if (same_name(clp->cl_recdir, dname) &&
894 match_clientid_establishment(clp, use_exchange_id))
901 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
903 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
904 unsigned short expected_family;
906 /* Currently, we only support tcp and tcp6 for the callback channel */
907 if (se->se_callback_netid_len == 3 &&
908 !memcmp(se->se_callback_netid_val, "tcp", 3))
909 expected_family = AF_INET;
910 else if (se->se_callback_netid_len == 4 &&
911 !memcmp(se->se_callback_netid_val, "tcp6", 4))
912 expected_family = AF_INET6;
916 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
917 se->se_callback_addr_len,
918 (struct sockaddr *) &cb->cb_addr,
919 sizeof(cb->cb_addr));
921 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
924 if (cb->cb_addr.ss_family == AF_INET6)
925 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
927 cb->cb_minorversion = 0;
928 cb->cb_prog = se->se_callback_prog;
929 cb->cb_ident = se->se_callback_ident;
932 cb->cb_addr.ss_family = AF_UNSPEC;
934 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
935 "will not receive delegations\n",
936 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
942 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
944 struct nfsd4_compoundres *resp = rqstp->rq_resp;
946 resp->cstate.statp = statp;
950 * Dereference the result pages.
953 nfsd4_release_respages(struct page **respages, short resused)
957 dprintk("--> %s\n", __func__);
958 for (i = 0; i < resused; i++) {
961 put_page(respages[i]);
967 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
971 for (i = 0; i < count; i++) {
972 topages[i] = frompages[i];
975 get_page(topages[i]);
980 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
981 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
982 * length of the XDR response is less than se_fmaxresp_cached
983 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
984 * of the reply (e.g. readdir).
986 * Store the base and length of the rq_req.head[0] page
987 * of the NFSv4.1 data, just past the rpc header.
990 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
992 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
993 struct svc_rqst *rqstp = resp->rqstp;
994 struct kvec *resv = &rqstp->rq_res.head[0];
996 dprintk("--> %s entry %p\n", __func__, entry);
998 nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
999 entry->ce_opcnt = resp->opcnt;
1000 entry->ce_status = resp->cstate.status;
1003 * Don't need a page to cache just the sequence operation - the slot
1007 if (nfsd4_not_cached(resp)) {
1008 entry->ce_resused = 0;
1009 entry->ce_rpchdrlen = 0;
1010 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1011 resp->cstate.slot->sl_cache_entry.ce_cachethis);
1014 entry->ce_resused = rqstp->rq_resused;
1015 if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1016 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1017 nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1019 entry->ce_datav.iov_base = resp->cstate.statp;
1020 entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1021 (char *)page_address(rqstp->rq_respages[0]));
1022 /* Current request rpc header length*/
1023 entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1024 (char *)page_address(rqstp->rq_respages[0]);
1028 * We keep the rpc header, but take the nfs reply from the replycache.
1031 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1032 struct nfsd4_cache_entry *entry)
1034 struct svc_rqst *rqstp = resp->rqstp;
1035 struct kvec *resv = &resp->rqstp->rq_res.head[0];
1038 /* Current request rpc header length*/
1039 len = (char *)resp->cstate.statp -
1040 (char *)page_address(rqstp->rq_respages[0]);
1041 if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1042 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1043 entry->ce_datav.iov_len);
1046 /* copy the cached reply nfsd data past the current rpc header */
1047 memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1048 entry->ce_datav.iov_len);
1049 resv->iov_len = len + entry->ce_datav.iov_len;
1054 * Encode the replay sequence operation from the slot values.
1055 * If cachethis is FALSE encode the uncached rep error on the next
1056 * operation which sets resp->p and increments resp->opcnt for
1057 * nfs4svc_encode_compoundres.
1061 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1062 struct nfsd4_compoundres *resp)
1064 struct nfsd4_op *op;
1065 struct nfsd4_slot *slot = resp->cstate.slot;
1067 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1068 resp->opcnt, resp->cstate.slot->sl_cache_entry.ce_cachethis);
1070 /* Encode the replayed sequence operation */
1071 op = &args->ops[resp->opcnt - 1];
1072 nfsd4_encode_operation(resp, op);
1074 /* Return nfserr_retry_uncached_rep in next operation. */
1075 if (args->opcnt > 1 && slot->sl_cache_entry.ce_cachethis == 0) {
1076 op = &args->ops[resp->opcnt++];
1077 op->status = nfserr_retry_uncached_rep;
1078 nfsd4_encode_operation(resp, op);
1084 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1085 * cached page. Replace any futher replay pages from the cache.
1088 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1089 struct nfsd4_sequence *seq)
1091 struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1094 dprintk("--> %s entry %p\n", __func__, entry);
1097 * If this is just the sequence operation, we did not keep
1098 * a page in the cache entry because we can just use the
1099 * slot info stored in struct nfsd4_sequence that was checked
1100 * against the slot in nfsd4_sequence().
1102 * This occurs when seq->cachethis is FALSE, or when the client
1103 * session inactivity timer fires and a solo sequence operation
1104 * is sent (lease renewal).
1106 seq->maxslots = resp->cstate.session->se_fchannel.maxreqs;
1108 /* Either returns 0 or nfserr_retry_uncached */
1109 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1110 if (status == nfserr_retry_uncached_rep)
1113 if (!nfsd41_copy_replay_data(resp, entry)) {
1115 * Not enough room to use the replay rpc header, send the
1116 * cached header. Release all the allocated result pages.
1118 svc_free_res_pages(resp->rqstp);
1119 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1122 /* Release all but the first allocated result page */
1124 resp->rqstp->rq_resused--;
1125 svc_free_res_pages(resp->rqstp);
1127 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1128 &entry->ce_respages[1],
1129 entry->ce_resused - 1);
1132 resp->rqstp->rq_resused = entry->ce_resused;
1133 resp->opcnt = entry->ce_opcnt;
1134 resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1135 status = entry->ce_status;
1141 * Set the exchange_id flags returned by the server.
1144 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1146 /* pNFS is not supported */
1147 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1149 /* Referrals are supported, Migration is not. */
1150 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1152 /* set the wire flags to return to client. */
1153 clid->flags = new->cl_exchange_flags;
1157 nfsd4_exchange_id(struct svc_rqst *rqstp,
1158 struct nfsd4_compound_state *cstate,
1159 struct nfsd4_exchange_id *exid)
1161 struct nfs4_client *unconf, *conf, *new;
1163 unsigned int strhashval;
1164 char dname[HEXDIR_LEN];
1165 char addr_str[INET6_ADDRSTRLEN];
1166 nfs4_verifier verf = exid->verifier;
1167 struct sockaddr *sa = svc_addr(rqstp);
1169 rpc_ntop(sa, addr_str, sizeof(addr_str));
1170 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1171 "ip_addr=%s flags %x, spa_how %d\n",
1172 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1173 addr_str, exid->flags, exid->spa_how);
1175 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1176 return nfserr_inval;
1178 /* Currently only support SP4_NONE */
1179 switch (exid->spa_how) {
1183 return nfserr_encr_alg_unsupp;
1185 BUG(); /* checked by xdr code */
1187 return nfserr_serverfault; /* no excuse :-/ */
1190 status = nfs4_make_rec_clidname(dname, &exid->clname);
1195 strhashval = clientstr_hashval(dname);
1200 conf = find_confirmed_client_by_str(dname, strhashval, true);
1202 if (!same_verf(&verf, &conf->cl_verifier)) {
1203 /* 18.35.4 case 8 */
1204 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1205 status = nfserr_not_same;
1208 /* Client reboot: destroy old state */
1209 expire_client(conf);
1212 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1213 /* 18.35.4 case 9 */
1214 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1215 status = nfserr_perm;
1218 expire_client(conf);
1222 * Set bit when the owner id and verifier map to an already
1223 * confirmed client id (18.35.3).
1225 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1228 * Falling into 18.35.4 case 2, possible router replay.
1229 * Leave confirmed record intact and return same result.
1231 copy_verf(conf, &verf);
1236 /* 18.35.4 case 7 */
1237 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1238 status = nfserr_noent;
1242 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1245 * Possible retry or client restart. Per 18.35.4 case 4,
1246 * a new unconfirmed record should be generated regardless
1247 * of whether any properties have changed.
1249 expire_client(unconf);
1254 new = create_client(exid->clname, dname);
1256 status = nfserr_resource;
1260 copy_verf(new, &verf);
1261 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1262 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1265 add_to_unconfirmed(new, strhashval);
1267 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1268 exid->clientid.cl_id = new->cl_clientid.cl_id;
1271 nfsd4_set_ex_flags(new, exid);
1273 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1274 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1278 nfs4_unlock_state();
1280 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1285 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1287 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1290 /* The slot is in use, and no response has been sent. */
1292 if (seqid == slot_seqid)
1293 return nfserr_jukebox;
1295 return nfserr_seq_misordered;
1298 if (likely(seqid == slot_seqid + 1))
1301 if (seqid == slot_seqid)
1302 return nfserr_replay_cache;
1304 if (seqid == 1 && (slot_seqid + 1) == 0)
1306 /* Misordered replay or misordered new request */
1307 return nfserr_seq_misordered;
1311 * Cache the create session result into the create session single DRC
1312 * slot cache by saving the xdr structure. sl_seqid has been set.
1313 * Do this for solo or embedded create session operations.
1316 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1317 struct nfsd4_clid_slot *slot, int nfserr)
1319 slot->sl_status = nfserr;
1320 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1324 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1325 struct nfsd4_clid_slot *slot)
1327 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1328 return slot->sl_status;
1332 nfsd4_create_session(struct svc_rqst *rqstp,
1333 struct nfsd4_compound_state *cstate,
1334 struct nfsd4_create_session *cr_ses)
1336 struct sockaddr *sa = svc_addr(rqstp);
1337 struct nfs4_client *conf, *unconf;
1338 struct nfsd4_clid_slot *cs_slot = NULL;
1342 unconf = find_unconfirmed_client(&cr_ses->clientid);
1343 conf = find_confirmed_client(&cr_ses->clientid);
1346 cs_slot = &conf->cl_cs_slot;
1347 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1348 if (status == nfserr_replay_cache) {
1349 dprintk("Got a create_session replay! seqid= %d\n",
1351 /* Return the cached reply status */
1352 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1354 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1355 status = nfserr_seq_misordered;
1356 dprintk("Sequence misordered!\n");
1357 dprintk("Expected seqid= %d but got seqid= %d\n",
1358 cs_slot->sl_seqid, cr_ses->seqid);
1361 cs_slot->sl_seqid++;
1362 } else if (unconf) {
1363 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1364 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1365 status = nfserr_clid_inuse;
1369 cs_slot = &unconf->cl_cs_slot;
1370 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1372 /* an unconfirmed replay returns misordered */
1373 status = nfserr_seq_misordered;
1377 cs_slot->sl_seqid++; /* from 0 to 1 */
1378 move_to_confirmed(unconf);
1381 * We do not support RDMA or persistent sessions
1383 cr_ses->flags &= ~SESSION4_PERSIST;
1384 cr_ses->flags &= ~SESSION4_RDMA;
1388 status = nfserr_stale_clientid;
1392 status = alloc_init_session(rqstp, conf, cr_ses);
1396 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1397 NFS4_MAX_SESSIONID_LEN);
1398 cr_ses->seqid = cs_slot->sl_seqid;
1401 /* cache solo and embedded create sessions under the state lock */
1402 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1404 nfs4_unlock_state();
1405 dprintk("%s returns %d\n", __func__, ntohl(status));
1410 nfsd4_destroy_session(struct svc_rqst *r,
1411 struct nfsd4_compound_state *cstate,
1412 struct nfsd4_destroy_session *sessionid)
1414 struct nfsd4_session *ses;
1415 u32 status = nfserr_badsession;
1418 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1419 * - Should we return nfserr_back_chan_busy if waiting for
1420 * callbacks on to-be-destroyed session?
1421 * - Do we need to clear any callback info from previous session?
1424 dump_sessionid(__func__, &sessionid->sessionid);
1425 spin_lock(&sessionid_lock);
1426 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1428 spin_unlock(&sessionid_lock);
1432 unhash_session(ses);
1433 spin_unlock(&sessionid_lock);
1435 /* wait for callbacks */
1436 shutdown_callback_client(ses->se_client);
1437 nfsd4_put_session(ses);
1440 dprintk("%s returns %d\n", __func__, ntohl(status));
1445 nfsd4_sequence(struct svc_rqst *rqstp,
1446 struct nfsd4_compound_state *cstate,
1447 struct nfsd4_sequence *seq)
1449 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1450 struct nfsd4_session *session;
1451 struct nfsd4_slot *slot;
1454 if (resp->opcnt != 1)
1455 return nfserr_sequence_pos;
1457 spin_lock(&sessionid_lock);
1458 status = nfserr_badsession;
1459 session = find_in_sessionid_hashtbl(&seq->sessionid);
1463 status = nfserr_badslot;
1464 if (seq->slotid >= session->se_fchannel.maxreqs)
1467 slot = &session->se_slots[seq->slotid];
1468 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1470 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1471 if (status == nfserr_replay_cache) {
1472 cstate->slot = slot;
1473 cstate->session = session;
1474 /* Return the cached reply status and set cstate->status
1475 * for nfsd4_svc_encode_compoundres processing */
1476 status = nfsd4_replay_cache_entry(resp, seq);
1477 cstate->status = nfserr_replay_cache;
1483 /* Success! bump slot seqid */
1484 slot->sl_inuse = true;
1485 slot->sl_seqid = seq->seqid;
1486 slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1488 cstate->slot = slot;
1489 cstate->session = session;
1491 /* Hold a session reference until done processing the compound:
1492 * nfsd4_put_session called only if the cstate slot is set.
1494 nfsd4_get_session(session);
1496 spin_unlock(&sessionid_lock);
1497 /* Renew the clientid on success and on replay */
1498 if (cstate->session) {
1500 renew_client(session->se_client);
1501 nfs4_unlock_state();
1503 dprintk("%s: return %d\n", __func__, ntohl(status));
1508 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1509 struct nfsd4_setclientid *setclid)
1511 struct sockaddr *sa = svc_addr(rqstp);
1512 struct xdr_netobj clname = {
1513 .len = setclid->se_namelen,
1514 .data = setclid->se_name,
1516 nfs4_verifier clverifier = setclid->se_verf;
1517 unsigned int strhashval;
1518 struct nfs4_client *conf, *unconf, *new;
1521 char dname[HEXDIR_LEN];
1523 if (!check_name(clname))
1524 return nfserr_inval;
1526 status = nfs4_make_rec_clidname(dname, &clname);
1531 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1532 * We get here on a DRC miss.
1535 strhashval = clientstr_hashval(dname);
1538 conf = find_confirmed_client_by_str(dname, strhashval, false);
1540 /* RFC 3530 14.2.33 CASE 0: */
1541 status = nfserr_clid_inuse;
1542 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1543 char addr_str[INET6_ADDRSTRLEN];
1544 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1546 dprintk("NFSD: setclientid: string in use by client "
1547 "at %s\n", addr_str);
1552 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1553 * has a description of SETCLIENTID request processing consisting
1554 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1556 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1557 status = nfserr_resource;
1560 * RFC 3530 14.2.33 CASE 4:
1561 * placed first, because it is the normal case
1564 expire_client(unconf);
1565 new = create_client(clname, dname);
1569 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1571 * RFC 3530 14.2.33 CASE 1:
1572 * probable callback update
1575 /* Note this is removing unconfirmed {*x***},
1576 * which is stronger than RFC recommended {vxc**}.
1577 * This has the advantage that there is at most
1578 * one {*x***} in either list at any time.
1580 expire_client(unconf);
1582 new = create_client(clname, dname);
1585 copy_clid(new, conf);
1586 } else if (!unconf) {
1588 * RFC 3530 14.2.33 CASE 2:
1589 * probable client reboot; state will be removed if
1592 new = create_client(clname, dname);
1598 * RFC 3530 14.2.33 CASE 3:
1599 * probable client reboot; state will be removed if
1602 expire_client(unconf);
1603 new = create_client(clname, dname);
1608 copy_verf(new, &clverifier);
1609 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1610 new->cl_flavor = rqstp->rq_flavor;
1611 princ = svc_gss_principal(rqstp);
1613 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1614 if (new->cl_principal == NULL) {
1619 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1621 gen_callback(new, setclid, rpc_get_scope_id(sa));
1622 add_to_unconfirmed(new, strhashval);
1623 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1624 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1625 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1628 nfs4_unlock_state();
1634 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1635 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1636 * bullets, labeled as CASE1 - CASE4 below.
1639 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1640 struct nfsd4_compound_state *cstate,
1641 struct nfsd4_setclientid_confirm *setclientid_confirm)
1643 struct sockaddr *sa = svc_addr(rqstp);
1644 struct nfs4_client *conf, *unconf;
1645 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1646 clientid_t * clid = &setclientid_confirm->sc_clientid;
1649 if (STALE_CLIENTID(clid))
1650 return nfserr_stale_clientid;
1652 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1653 * We get here on a DRC miss.
1658 conf = find_confirmed_client(clid);
1659 unconf = find_unconfirmed_client(clid);
1661 status = nfserr_clid_inuse;
1662 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1664 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1668 * section 14.2.34 of RFC 3530 has a description of
1669 * SETCLIENTID_CONFIRM request processing consisting
1670 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1672 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1674 * RFC 3530 14.2.34 CASE 1:
1677 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1678 status = nfserr_clid_inuse;
1680 /* XXX: We just turn off callbacks until we can handle
1681 * change request correctly. */
1682 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1683 expire_client(unconf);
1687 } else if (conf && !unconf) {
1689 * RFC 3530 14.2.34 CASE 2:
1690 * probable retransmitted request; play it safe and
1693 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1694 status = nfserr_clid_inuse;
1697 } else if (!conf && unconf
1698 && same_verf(&unconf->cl_confirm, &confirm)) {
1700 * RFC 3530 14.2.34 CASE 3:
1701 * Normal case; new or rebooted client:
1703 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1704 status = nfserr_clid_inuse;
1707 clientstr_hashval(unconf->cl_recdir);
1708 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1711 nfsd4_remove_clid_dir(conf);
1712 expire_client(conf);
1714 move_to_confirmed(unconf);
1716 nfsd4_probe_callback(conf);
1719 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1720 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1723 * RFC 3530 14.2.34 CASE 4:
1724 * Client probably hasn't noticed that we rebooted yet.
1726 status = nfserr_stale_clientid;
1728 /* check that we have hit one of the cases...*/
1729 status = nfserr_clid_inuse;
1732 nfs4_unlock_state();
1736 /* OPEN Share state helper functions */
1737 static inline struct nfs4_file *
1738 alloc_init_file(struct inode *ino)
1740 struct nfs4_file *fp;
1741 unsigned int hashval = file_hashval(ino);
1743 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1745 atomic_set(&fp->fi_ref, 1);
1746 INIT_LIST_HEAD(&fp->fi_hash);
1747 INIT_LIST_HEAD(&fp->fi_stateids);
1748 INIT_LIST_HEAD(&fp->fi_delegations);
1749 spin_lock(&recall_lock);
1750 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1751 spin_unlock(&recall_lock);
1752 fp->fi_inode = igrab(ino);
1753 fp->fi_id = current_fileid++;
1754 fp->fi_had_conflict = false;
1761 nfsd4_free_slab(struct kmem_cache **slab)
1765 kmem_cache_destroy(*slab);
1770 nfsd4_free_slabs(void)
1772 nfsd4_free_slab(&stateowner_slab);
1773 nfsd4_free_slab(&file_slab);
1774 nfsd4_free_slab(&stateid_slab);
1775 nfsd4_free_slab(&deleg_slab);
1779 nfsd4_init_slabs(void)
1781 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1782 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1783 if (stateowner_slab == NULL)
1785 file_slab = kmem_cache_create("nfsd4_files",
1786 sizeof(struct nfs4_file), 0, 0, NULL);
1787 if (file_slab == NULL)
1789 stateid_slab = kmem_cache_create("nfsd4_stateids",
1790 sizeof(struct nfs4_stateid), 0, 0, NULL);
1791 if (stateid_slab == NULL)
1793 deleg_slab = kmem_cache_create("nfsd4_delegations",
1794 sizeof(struct nfs4_delegation), 0, 0, NULL);
1795 if (deleg_slab == NULL)
1800 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1805 nfs4_free_stateowner(struct kref *kref)
1807 struct nfs4_stateowner *sop =
1808 container_of(kref, struct nfs4_stateowner, so_ref);
1809 kfree(sop->so_owner.data);
1810 kmem_cache_free(stateowner_slab, sop);
1813 static inline struct nfs4_stateowner *
1814 alloc_stateowner(struct xdr_netobj *owner)
1816 struct nfs4_stateowner *sop;
1818 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1819 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1820 memcpy(sop->so_owner.data, owner->data, owner->len);
1821 sop->so_owner.len = owner->len;
1822 kref_init(&sop->so_ref);
1825 kmem_cache_free(stateowner_slab, sop);
1830 static struct nfs4_stateowner *
1831 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1832 struct nfs4_stateowner *sop;
1833 struct nfs4_replay *rp;
1834 unsigned int idhashval;
1836 if (!(sop = alloc_stateowner(&open->op_owner)))
1838 idhashval = ownerid_hashval(current_ownerid);
1839 INIT_LIST_HEAD(&sop->so_idhash);
1840 INIT_LIST_HEAD(&sop->so_strhash);
1841 INIT_LIST_HEAD(&sop->so_perclient);
1842 INIT_LIST_HEAD(&sop->so_stateids);
1843 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1844 INIT_LIST_HEAD(&sop->so_close_lru);
1846 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1847 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1848 list_add(&sop->so_perclient, &clp->cl_openowners);
1849 sop->so_is_open_owner = 1;
1850 sop->so_id = current_ownerid++;
1851 sop->so_client = clp;
1852 sop->so_seqid = open->op_seqid;
1853 sop->so_confirmed = 0;
1854 rp = &sop->so_replay;
1855 rp->rp_status = nfserr_serverfault;
1857 rp->rp_buf = rp->rp_ibuf;
1862 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1863 struct nfs4_stateowner *sop = open->op_stateowner;
1864 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1866 INIT_LIST_HEAD(&stp->st_hash);
1867 INIT_LIST_HEAD(&stp->st_perstateowner);
1868 INIT_LIST_HEAD(&stp->st_lockowners);
1869 INIT_LIST_HEAD(&stp->st_perfile);
1870 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1871 list_add(&stp->st_perstateowner, &sop->so_stateids);
1872 list_add(&stp->st_perfile, &fp->fi_stateids);
1873 stp->st_stateowner = sop;
1876 stp->st_stateid.si_boot = get_seconds();
1877 stp->st_stateid.si_stateownerid = sop->so_id;
1878 stp->st_stateid.si_fileid = fp->fi_id;
1879 stp->st_stateid.si_generation = 0;
1880 stp->st_access_bmap = 0;
1881 stp->st_deny_bmap = 0;
1882 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1883 &stp->st_access_bmap);
1884 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1885 stp->st_openstp = NULL;
1889 move_to_close_lru(struct nfs4_stateowner *sop)
1891 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1893 list_move_tail(&sop->so_close_lru, &close_lru);
1894 sop->so_time = get_seconds();
1898 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1901 return (sop->so_owner.len == owner->len) &&
1902 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1903 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1906 static struct nfs4_stateowner *
1907 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1909 struct nfs4_stateowner *so = NULL;
1911 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1912 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1918 /* search file_hashtbl[] for file */
1919 static struct nfs4_file *
1920 find_file(struct inode *ino)
1922 unsigned int hashval = file_hashval(ino);
1923 struct nfs4_file *fp;
1925 spin_lock(&recall_lock);
1926 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1927 if (fp->fi_inode == ino) {
1929 spin_unlock(&recall_lock);
1933 spin_unlock(&recall_lock);
1937 static inline int access_valid(u32 x, u32 minorversion)
1939 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1941 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1943 x &= ~NFS4_SHARE_ACCESS_MASK;
1944 if (minorversion && x) {
1945 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1947 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1949 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1956 static inline int deny_valid(u32 x)
1958 /* Note: unlike access bits, deny bits may be zero. */
1959 return x <= NFS4_SHARE_DENY_BOTH;
1963 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1964 * st_{access,deny}_bmap field of the stateid, in order to track not
1965 * only what share bits are currently in force, but also what
1966 * combinations of share bits previous opens have used. This allows us
1967 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1968 * return an error if the client attempt to downgrade to a combination
1969 * of share bits not explicable by closing some of its previous opens.
1971 * XXX: This enforcement is actually incomplete, since we don't keep
1972 * track of access/deny bit combinations; so, e.g., we allow:
1974 * OPEN allow read, deny write
1975 * OPEN allow both, deny none
1976 * DOWNGRADE allow read, deny none
1978 * which we should reject.
1981 set_access(unsigned int *access, unsigned long bmap) {
1985 for (i = 1; i < 4; i++) {
1986 if (test_bit(i, &bmap))
1992 set_deny(unsigned int *deny, unsigned long bmap) {
1996 for (i = 0; i < 4; i++) {
1997 if (test_bit(i, &bmap))
2003 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2004 unsigned int access, deny;
2006 set_access(&access, stp->st_access_bmap);
2007 set_deny(&deny, stp->st_deny_bmap);
2008 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2014 * Called to check deny when READ with all zero stateid or
2015 * WRITE with all zero or all one stateid
2018 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2020 struct inode *ino = current_fh->fh_dentry->d_inode;
2021 struct nfs4_file *fp;
2022 struct nfs4_stateid *stp;
2025 dprintk("NFSD: nfs4_share_conflict\n");
2027 fp = find_file(ino);
2030 ret = nfserr_locked;
2031 /* Search for conflicting share reservations */
2032 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2033 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2034 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2044 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2046 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2047 drop_file_write_access(filp);
2048 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2053 * Spawn a thread to perform a recall on the delegation represented
2054 * by the lease (file_lock)
2056 * Called from break_lease() with lock_kernel() held.
2057 * Note: we assume break_lease will only call this *once* for any given
2061 void nfsd_break_deleg_cb(struct file_lock *fl)
2063 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2065 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2069 /* We're assuming the state code never drops its reference
2070 * without first removing the lease. Since we're in this lease
2071 * callback (and since the lease code is serialized by the kernel
2072 * lock) we know the server hasn't removed the lease yet, we know
2073 * it's safe to take a reference: */
2074 atomic_inc(&dp->dl_count);
2075 atomic_inc(&dp->dl_client->cl_count);
2077 spin_lock(&recall_lock);
2078 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2079 spin_unlock(&recall_lock);
2081 /* only place dl_time is set. protected by lock_kernel*/
2082 dp->dl_time = get_seconds();
2085 * We don't want the locks code to timeout the lease for us;
2086 * we'll remove it ourself if the delegation isn't returned
2089 fl->fl_break_time = 0;
2091 dp->dl_file->fi_had_conflict = true;
2092 nfsd4_cb_recall(dp);
2096 * The file_lock is being reapd.
2098 * Called by locks_free_lock() with lock_kernel() held.
2101 void nfsd_release_deleg_cb(struct file_lock *fl)
2103 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2105 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2107 if (!(fl->fl_flags & FL_LEASE) || !dp)
2109 dp->dl_flock = NULL;
2113 * Set the delegation file_lock back pointer.
2115 * Called from setlease() with lock_kernel() held.
2118 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2120 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2122 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2129 * Called from setlease() with lock_kernel() held
2132 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2134 struct nfs4_delegation *onlistd =
2135 (struct nfs4_delegation *)onlist->fl_owner;
2136 struct nfs4_delegation *tryd =
2137 (struct nfs4_delegation *)try->fl_owner;
2139 if (onlist->fl_lmops != try->fl_lmops)
2142 return onlistd->dl_client == tryd->dl_client;
2147 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2150 return lease_modify(onlist, arg);
2155 static struct lock_manager_operations nfsd_lease_mng_ops = {
2156 .fl_break = nfsd_break_deleg_cb,
2157 .fl_release_private = nfsd_release_deleg_cb,
2158 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2159 .fl_mylease = nfsd_same_client_deleg_cb,
2160 .fl_change = nfsd_change_deleg_cb,
2165 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2166 struct nfsd4_open *open)
2168 clientid_t *clientid = &open->op_clientid;
2169 struct nfs4_client *clp = NULL;
2170 unsigned int strhashval;
2171 struct nfs4_stateowner *sop = NULL;
2173 if (!check_name(open->op_owner))
2174 return nfserr_inval;
2176 if (STALE_CLIENTID(&open->op_clientid))
2177 return nfserr_stale_clientid;
2179 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2180 sop = find_openstateowner_str(strhashval, open);
2181 open->op_stateowner = sop;
2183 /* Make sure the client's lease hasn't expired. */
2184 clp = find_confirmed_client(clientid);
2186 return nfserr_expired;
2189 /* When sessions are used, skip open sequenceid processing */
2190 if (nfsd4_has_session(cstate))
2192 if (!sop->so_confirmed) {
2193 /* Replace unconfirmed owners without checking for replay. */
2194 clp = sop->so_client;
2195 release_openowner(sop);
2196 open->op_stateowner = NULL;
2199 if (open->op_seqid == sop->so_seqid - 1) {
2200 if (sop->so_replay.rp_buflen)
2201 return nfserr_replay_me;
2202 /* The original OPEN failed so spectacularly
2203 * that we don't even have replay data saved!
2204 * Therefore, we have no choice but to continue
2205 * processing this OPEN; presumably, we'll
2206 * fail again for the same reason.
2208 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2211 if (open->op_seqid != sop->so_seqid)
2212 return nfserr_bad_seqid;
2214 if (open->op_stateowner == NULL) {
2215 sop = alloc_init_open_stateowner(strhashval, clp, open);
2217 return nfserr_resource;
2218 open->op_stateowner = sop;
2220 list_del_init(&sop->so_close_lru);
2221 renew_client(sop->so_client);
2225 static inline __be32
2226 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2228 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2229 return nfserr_openmode;
2234 static struct nfs4_delegation *
2235 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2237 struct nfs4_delegation *dp;
2239 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2240 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2247 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2248 struct nfs4_delegation **dp)
2251 __be32 status = nfserr_bad_stateid;
2253 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2256 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2257 RD_STATE : WR_STATE;
2258 status = nfs4_check_delegmode(*dp, flags);
2262 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2266 open->op_stateowner->so_confirmed = 1;
2271 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2273 struct nfs4_stateid *local;
2274 __be32 status = nfserr_share_denied;
2275 struct nfs4_stateowner *sop = open->op_stateowner;
2277 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2278 /* ignore lock owners */
2279 if (local->st_stateowner->so_is_open_owner == 0)
2281 /* remember if we have seen this open owner */
2282 if (local->st_stateowner == sop)
2284 /* check for conflicting share reservations */
2285 if (!test_share(local, open))
2293 static inline struct nfs4_stateid *
2294 nfs4_alloc_stateid(void)
2296 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2300 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2301 struct nfs4_delegation *dp,
2302 struct svc_fh *cur_fh, int flags)
2304 struct nfs4_stateid *stp;
2306 stp = nfs4_alloc_stateid();
2308 return nfserr_resource;
2311 get_file(dp->dl_vfs_file);
2312 stp->st_vfs_file = dp->dl_vfs_file;
2315 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2318 if (status == nfserr_dropit)
2319 status = nfserr_jukebox;
2320 kmem_cache_free(stateid_slab, stp);
2328 static inline __be32
2329 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2330 struct nfsd4_open *open)
2332 struct iattr iattr = {
2333 .ia_valid = ATTR_SIZE,
2336 if (!open->op_truncate)
2338 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2339 return nfserr_inval;
2340 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2344 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2346 struct file *filp = stp->st_vfs_file;
2347 struct inode *inode = filp->f_path.dentry->d_inode;
2348 unsigned int share_access, new_writer;
2351 set_access(&share_access, stp->st_access_bmap);
2352 new_writer = (~share_access) & open->op_share_access
2353 & NFS4_SHARE_ACCESS_WRITE;
2356 int err = get_write_access(inode);
2358 return nfserrno(err);
2359 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2361 return nfserrno(err);
2362 file_take_write(filp);
2364 status = nfsd4_truncate(rqstp, cur_fh, open);
2367 put_write_access(inode);
2370 /* remember the open */
2371 filp->f_mode |= open->op_share_access;
2372 __set_bit(open->op_share_access, &stp->st_access_bmap);
2373 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2380 nfs4_set_claim_prev(struct nfsd4_open *open)
2382 open->op_stateowner->so_confirmed = 1;
2383 open->op_stateowner->so_client->cl_firststate = 1;
2387 * Attempt to hand out a delegation.
2390 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2392 struct nfs4_delegation *dp;
2393 struct nfs4_stateowner *sop = stp->st_stateowner;
2394 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2395 struct file_lock fl, *flp = &fl;
2396 int status, flag = 0;
2398 flag = NFS4_OPEN_DELEGATE_NONE;
2399 open->op_recall = 0;
2400 switch (open->op_claim_type) {
2401 case NFS4_OPEN_CLAIM_PREVIOUS:
2402 if (!atomic_read(&cb->cb_set))
2403 open->op_recall = 1;
2404 flag = open->op_delegate_type;
2405 if (flag == NFS4_OPEN_DELEGATE_NONE)
2408 case NFS4_OPEN_CLAIM_NULL:
2409 /* Let's not give out any delegations till everyone's
2410 * had the chance to reclaim theirs.... */
2411 if (locks_in_grace())
2413 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2415 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2416 flag = NFS4_OPEN_DELEGATE_WRITE;
2418 flag = NFS4_OPEN_DELEGATE_READ;
2424 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2426 flag = NFS4_OPEN_DELEGATE_NONE;
2429 locks_init_lock(&fl);
2430 fl.fl_lmops = &nfsd_lease_mng_ops;
2431 fl.fl_flags = FL_LEASE;
2432 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2433 fl.fl_end = OFFSET_MAX;
2434 fl.fl_owner = (fl_owner_t)dp;
2435 fl.fl_file = stp->st_vfs_file;
2436 fl.fl_pid = current->tgid;
2438 /* vfs_setlease checks to see if delegation should be handed out.
2439 * the lock_manager callbacks fl_mylease and fl_change are used
2441 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2442 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2443 unhash_delegation(dp);
2444 flag = NFS4_OPEN_DELEGATE_NONE;
2448 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2450 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2451 dp->dl_stateid.si_boot,
2452 dp->dl_stateid.si_stateownerid,
2453 dp->dl_stateid.si_fileid,
2454 dp->dl_stateid.si_generation);
2456 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2457 && flag == NFS4_OPEN_DELEGATE_NONE
2458 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2459 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2460 open->op_delegate_type = flag;
2464 * called with nfs4_lock_state() held.
2467 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2469 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2470 struct nfs4_file *fp = NULL;
2471 struct inode *ino = current_fh->fh_dentry->d_inode;
2472 struct nfs4_stateid *stp = NULL;
2473 struct nfs4_delegation *dp = NULL;
2476 status = nfserr_inval;
2477 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2478 || !deny_valid(open->op_share_deny))
2481 * Lookup file; if found, lookup stateid and check open request,
2482 * and check for delegations in the process of being recalled.
2483 * If not found, create the nfs4_file struct
2485 fp = find_file(ino);
2487 if ((status = nfs4_check_open(fp, open, &stp)))
2489 status = nfs4_check_deleg(fp, open, &dp);
2493 status = nfserr_bad_stateid;
2494 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2496 status = nfserr_resource;
2497 fp = alloc_init_file(ino);
2503 * OPEN the file, or upgrade an existing OPEN.
2504 * If truncate fails, the OPEN fails.
2507 /* Stateid was found, this is an OPEN upgrade */
2508 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2511 update_stateid(&stp->st_stateid);
2513 /* Stateid was not found, this is a new OPEN */
2515 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2516 flags |= NFSD_MAY_READ;
2517 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2518 flags |= NFSD_MAY_WRITE;
2519 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2522 init_stateid(stp, fp, open);
2523 status = nfsd4_truncate(rqstp, current_fh, open);
2525 release_open_stateid(stp);
2528 if (nfsd4_has_session(&resp->cstate))
2529 update_stateid(&stp->st_stateid);
2531 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2533 if (nfsd4_has_session(&resp->cstate))
2534 open->op_stateowner->so_confirmed = 1;
2537 * Attempt to hand out a delegation. No error return, because the
2538 * OPEN succeeds even if we fail.
2540 nfs4_open_delegation(current_fh, open, stp);
2544 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2545 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2546 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2550 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2551 nfs4_set_claim_prev(open);
2553 * To finish the open response, we just need to set the rflags.
2555 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2556 if (!open->op_stateowner->so_confirmed &&
2557 !nfsd4_has_session(&resp->cstate))
2558 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2564 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2567 struct nfs4_client *clp;
2571 dprintk("process_renew(%08x/%08x): starting\n",
2572 clid->cl_boot, clid->cl_id);
2573 status = nfserr_stale_clientid;
2574 if (STALE_CLIENTID(clid))
2576 clp = find_confirmed_client(clid);
2577 status = nfserr_expired;
2579 /* We assume the client took too long to RENEW. */
2580 dprintk("nfsd4_renew: clientid not found!\n");
2584 status = nfserr_cb_path_down;
2585 if (!list_empty(&clp->cl_delegations)
2586 && !atomic_read(&clp->cl_cb_conn.cb_set))
2590 nfs4_unlock_state();
2594 struct lock_manager nfsd4_manager = {
2598 nfsd4_end_grace(void)
2600 dprintk("NFSD: end of grace period\n");
2601 nfsd4_recdir_purge_old();
2602 locks_end_grace(&nfsd4_manager);
2606 nfs4_laundromat(void)
2608 struct nfs4_client *clp;
2609 struct nfs4_stateowner *sop;
2610 struct nfs4_delegation *dp;
2611 struct list_head *pos, *next, reaplist;
2612 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2613 time_t t, clientid_val = NFSD_LEASE_TIME;
2614 time_t u, test_val = NFSD_LEASE_TIME;
2618 dprintk("NFSD: laundromat service - starting\n");
2619 if (locks_in_grace())
2621 list_for_each_safe(pos, next, &client_lru) {
2622 clp = list_entry(pos, struct nfs4_client, cl_lru);
2623 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2624 t = clp->cl_time - cutoff;
2625 if (clientid_val > t)
2629 dprintk("NFSD: purging unused client (clientid %08x)\n",
2630 clp->cl_clientid.cl_id);
2631 nfsd4_remove_clid_dir(clp);
2634 INIT_LIST_HEAD(&reaplist);
2635 spin_lock(&recall_lock);
2636 list_for_each_safe(pos, next, &del_recall_lru) {
2637 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2638 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2639 u = dp->dl_time - cutoff;
2644 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2646 list_move(&dp->dl_recall_lru, &reaplist);
2648 spin_unlock(&recall_lock);
2649 list_for_each_safe(pos, next, &reaplist) {
2650 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2651 list_del_init(&dp->dl_recall_lru);
2652 unhash_delegation(dp);
2654 test_val = NFSD_LEASE_TIME;
2655 list_for_each_safe(pos, next, &close_lru) {
2656 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2657 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2658 u = sop->so_time - cutoff;
2663 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2665 release_openowner(sop);
2667 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2668 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2669 nfs4_unlock_state();
2670 return clientid_val;
2673 static struct workqueue_struct *laundry_wq;
2674 static void laundromat_main(struct work_struct *);
2675 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2678 laundromat_main(struct work_struct *not_used)
2682 t = nfs4_laundromat();
2683 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2684 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2687 static struct nfs4_stateowner *
2688 search_close_lru(u32 st_id, int flags)
2690 struct nfs4_stateowner *local = NULL;
2692 if (flags & CLOSE_STATE) {
2693 list_for_each_entry(local, &close_lru, so_close_lru) {
2694 if (local->so_id == st_id)
2702 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2704 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2708 STALE_STATEID(stateid_t *stateid)
2710 if (time_after((unsigned long)boot_time,
2711 (unsigned long)stateid->si_boot)) {
2712 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2713 stateid->si_boot, stateid->si_stateownerid,
2714 stateid->si_fileid, stateid->si_generation);
2721 EXPIRED_STATEID(stateid_t *stateid)
2723 if (time_before((unsigned long)boot_time,
2724 ((unsigned long)stateid->si_boot)) &&
2725 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2726 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2727 stateid->si_boot, stateid->si_stateownerid,
2728 stateid->si_fileid, stateid->si_generation);
2735 stateid_error_map(stateid_t *stateid)
2737 if (STALE_STATEID(stateid))
2738 return nfserr_stale_stateid;
2739 if (EXPIRED_STATEID(stateid))
2740 return nfserr_expired;
2742 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2743 stateid->si_boot, stateid->si_stateownerid,
2744 stateid->si_fileid, stateid->si_generation);
2745 return nfserr_bad_stateid;
2749 access_permit_read(unsigned long access_bmap)
2751 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2752 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2753 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2757 access_permit_write(unsigned long access_bmap)
2759 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2760 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2764 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2766 __be32 status = nfserr_openmode;
2768 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2770 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2777 static inline __be32
2778 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2780 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2782 else if (locks_in_grace()) {
2783 /* Answer in remaining cases depends on existance of
2784 * conflicting state; so we must wait out the grace period. */
2785 return nfserr_grace;
2786 } else if (flags & WR_STATE)
2787 return nfs4_share_conflict(current_fh,
2788 NFS4_SHARE_DENY_WRITE);
2789 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2790 return nfs4_share_conflict(current_fh,
2791 NFS4_SHARE_DENY_READ);
2795 * Allow READ/WRITE during grace period on recovered state only for files
2796 * that are not able to provide mandatory locking.
2799 grace_disallows_io(struct inode *inode)
2801 return locks_in_grace() && mandatory_lock(inode);
2804 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2807 * When sessions are used the stateid generation number is ignored
2810 if ((flags & HAS_SESSION) && in->si_generation == 0)
2813 /* If the client sends us a stateid from the future, it's buggy: */
2814 if (in->si_generation > ref->si_generation)
2815 return nfserr_bad_stateid;
2817 * The following, however, can happen. For example, if the
2818 * client sends an open and some IO at the same time, the open
2819 * may bump si_generation while the IO is still in flight.
2820 * Thanks to hard links and renames, the client never knows what
2821 * file an open will affect. So it could avoid that situation
2822 * only by serializing all opens and IO from the same open
2823 * owner. To recover from the old_stateid error, the client
2824 * will just have to retry the IO:
2826 if (in->si_generation < ref->si_generation)
2827 return nfserr_old_stateid;
2832 static int is_delegation_stateid(stateid_t *stateid)
2834 return stateid->si_fileid == 0;
2838 * Checks for stateid operations
2841 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2842 stateid_t *stateid, int flags, struct file **filpp)
2844 struct nfs4_stateid *stp = NULL;
2845 struct nfs4_delegation *dp = NULL;
2846 struct svc_fh *current_fh = &cstate->current_fh;
2847 struct inode *ino = current_fh->fh_dentry->d_inode;
2853 if (grace_disallows_io(ino))
2854 return nfserr_grace;
2856 if (nfsd4_has_session(cstate))
2857 flags |= HAS_SESSION;
2859 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2860 return check_special_stateids(current_fh, stateid, flags);
2862 status = nfserr_stale_stateid;
2863 if (STALE_STATEID(stateid))
2866 status = nfserr_bad_stateid;
2867 if (is_delegation_stateid(stateid)) {
2868 dp = find_delegation_stateid(ino, stateid);
2870 status = stateid_error_map(stateid);
2873 status = check_stateid_generation(stateid, &dp->dl_stateid,
2877 status = nfs4_check_delegmode(dp, flags);
2880 renew_client(dp->dl_client);
2882 *filpp = dp->dl_vfs_file;
2883 } else { /* open or lock stateid */
2884 stp = find_stateid(stateid, flags);
2886 status = stateid_error_map(stateid);
2889 if (nfs4_check_fh(current_fh, stp))
2891 if (!stp->st_stateowner->so_confirmed)
2893 status = check_stateid_generation(stateid, &stp->st_stateid,
2897 status = nfs4_check_openmode(stp, flags);
2900 renew_client(stp->st_stateowner->so_client);
2902 *filpp = stp->st_vfs_file;
2912 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2913 RD_STATE : WR_STATE;
2917 * Checks for sequence id mutating operations.
2920 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2921 stateid_t *stateid, int flags,
2922 struct nfs4_stateowner **sopp,
2923 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2925 struct nfs4_stateid *stp;
2926 struct nfs4_stateowner *sop;
2927 struct svc_fh *current_fh = &cstate->current_fh;
2930 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2931 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2932 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2933 stateid->si_generation);
2938 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2939 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2940 return nfserr_bad_stateid;
2943 if (STALE_STATEID(stateid))
2944 return nfserr_stale_stateid;
2946 if (nfsd4_has_session(cstate))
2947 flags |= HAS_SESSION;
2950 * We return BAD_STATEID if filehandle doesn't match stateid,
2951 * the confirmed flag is incorrecly set, or the generation
2952 * number is incorrect.
2954 stp = find_stateid(stateid, flags);
2957 * Also, we should make sure this isn't just the result of
2960 sop = search_close_lru(stateid->si_stateownerid, flags);
2962 return stateid_error_map(stateid);
2968 *sopp = sop = stp->st_stateowner;
2971 clientid_t *lockclid = &lock->v.new.clientid;
2972 struct nfs4_client *clp = sop->so_client;
2976 lkflg = setlkflg(lock->lk_type);
2978 if (lock->lk_is_new) {
2979 if (!sop->so_is_open_owner)
2980 return nfserr_bad_stateid;
2981 if (!(flags & HAS_SESSION) &&
2982 !same_clid(&clp->cl_clientid, lockclid))
2983 return nfserr_bad_stateid;
2984 /* stp is the open stateid */
2985 status = nfs4_check_openmode(stp, lkflg);
2989 /* stp is the lock stateid */
2990 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2996 if (nfs4_check_fh(current_fh, stp)) {
2997 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2998 return nfserr_bad_stateid;
3002 * We now validate the seqid and stateid generation numbers.
3003 * For the moment, we ignore the possibility of
3004 * generation number wraparound.
3006 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3009 if (sop->so_confirmed && flags & CONFIRM) {
3010 dprintk("NFSD: preprocess_seqid_op: expected"
3011 " unconfirmed stateowner!\n");
3012 return nfserr_bad_stateid;
3014 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3015 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3016 " confirmed yet!\n");
3017 return nfserr_bad_stateid;
3019 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3022 renew_client(sop->so_client);
3026 if (seqid == sop->so_seqid - 1) {
3027 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3028 /* indicate replay to calling function */
3029 return nfserr_replay_me;
3031 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3032 sop->so_seqid, seqid);
3034 return nfserr_bad_seqid;
3038 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3039 struct nfsd4_open_confirm *oc)
3042 struct nfs4_stateowner *sop;
3043 struct nfs4_stateid *stp;
3045 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3046 (int)cstate->current_fh.fh_dentry->d_name.len,
3047 cstate->current_fh.fh_dentry->d_name.name);
3049 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3055 if ((status = nfs4_preprocess_seqid_op(cstate,
3056 oc->oc_seqid, &oc->oc_req_stateid,
3057 CONFIRM | OPEN_STATE,
3058 &oc->oc_stateowner, &stp, NULL)))
3061 sop = oc->oc_stateowner;
3062 sop->so_confirmed = 1;
3063 update_stateid(&stp->st_stateid);
3064 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3065 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3066 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3067 stp->st_stateid.si_boot,
3068 stp->st_stateid.si_stateownerid,
3069 stp->st_stateid.si_fileid,
3070 stp->st_stateid.si_generation);
3072 nfsd4_create_clid_dir(sop->so_client);
3074 if (oc->oc_stateowner) {
3075 nfs4_get_stateowner(oc->oc_stateowner);
3076 cstate->replay_owner = oc->oc_stateowner;
3078 nfs4_unlock_state();
3084 * unset all bits in union bitmap (bmap) that
3085 * do not exist in share (from successful OPEN_DOWNGRADE)
3088 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3091 for (i = 1; i < 4; i++) {
3092 if ((i & access) != i)
3093 __clear_bit(i, bmap);
3098 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3101 for (i = 0; i < 4; i++) {
3102 if ((i & deny) != i)
3103 __clear_bit(i, bmap);
3108 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3109 struct nfsd4_compound_state *cstate,
3110 struct nfsd4_open_downgrade *od)
3113 struct nfs4_stateid *stp;
3114 unsigned int share_access;
3116 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3117 (int)cstate->current_fh.fh_dentry->d_name.len,
3118 cstate->current_fh.fh_dentry->d_name.name);
3120 if (!access_valid(od->od_share_access, cstate->minorversion)
3121 || !deny_valid(od->od_share_deny))
3122 return nfserr_inval;
3125 if ((status = nfs4_preprocess_seqid_op(cstate,
3129 &od->od_stateowner, &stp, NULL)))
3132 status = nfserr_inval;
3133 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3134 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3135 stp->st_access_bmap, od->od_share_access);
3138 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3139 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3140 stp->st_deny_bmap, od->od_share_deny);
3143 set_access(&share_access, stp->st_access_bmap);
3144 nfs4_file_downgrade(stp->st_vfs_file,
3145 share_access & ~od->od_share_access);
3147 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3148 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3150 update_stateid(&stp->st_stateid);
3151 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3154 if (od->od_stateowner) {
3155 nfs4_get_stateowner(od->od_stateowner);
3156 cstate->replay_owner = od->od_stateowner;
3158 nfs4_unlock_state();
3163 * nfs4_unlock_state() called after encode
3166 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3167 struct nfsd4_close *close)
3170 struct nfs4_stateid *stp;
3172 dprintk("NFSD: nfsd4_close on file %.*s\n",
3173 (int)cstate->current_fh.fh_dentry->d_name.len,
3174 cstate->current_fh.fh_dentry->d_name.name);
3177 /* check close_lru for replay */
3178 if ((status = nfs4_preprocess_seqid_op(cstate,
3181 OPEN_STATE | CLOSE_STATE,
3182 &close->cl_stateowner, &stp, NULL)))
3185 update_stateid(&stp->st_stateid);
3186 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3188 /* release_stateid() calls nfsd_close() if needed */
3189 release_open_stateid(stp);
3191 /* place unused nfs4_stateowners on so_close_lru list to be
3192 * released by the laundromat service after the lease period
3193 * to enable us to handle CLOSE replay
3195 if (list_empty(&close->cl_stateowner->so_stateids))
3196 move_to_close_lru(close->cl_stateowner);
3198 if (close->cl_stateowner) {
3199 nfs4_get_stateowner(close->cl_stateowner);
3200 cstate->replay_owner = close->cl_stateowner;
3202 nfs4_unlock_state();
3207 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3208 struct nfsd4_delegreturn *dr)
3210 struct nfs4_delegation *dp;
3211 stateid_t *stateid = &dr->dr_stateid;
3212 struct inode *inode;
3216 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3218 inode = cstate->current_fh.fh_dentry->d_inode;
3220 if (nfsd4_has_session(cstate))
3221 flags |= HAS_SESSION;
3223 status = nfserr_bad_stateid;
3224 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3226 status = nfserr_stale_stateid;
3227 if (STALE_STATEID(stateid))
3229 status = nfserr_bad_stateid;
3230 if (!is_delegation_stateid(stateid))
3232 dp = find_delegation_stateid(inode, stateid);
3234 status = stateid_error_map(stateid);
3237 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3240 renew_client(dp->dl_client);
3242 unhash_delegation(dp);
3244 nfs4_unlock_state();
3251 * Lock owner state (byte-range locks)
3253 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3254 #define LOCK_HASH_BITS 8
3255 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3256 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3259 end_offset(u64 start, u64 len)
3264 return end >= start ? end: NFS4_MAX_UINT64;
3267 /* last octet in a range */
3269 last_byte_offset(u64 start, u64 len)
3275 return end > start ? end - 1: NFS4_MAX_UINT64;
3278 #define lockownerid_hashval(id) \
3279 ((id) & LOCK_HASH_MASK)
3281 static inline unsigned int
3282 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3283 struct xdr_netobj *ownername)
3285 return (file_hashval(inode) + cl_id
3286 + opaque_hashval(ownername->data, ownername->len))
3290 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3291 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3292 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3294 static struct nfs4_stateid *
3295 find_stateid(stateid_t *stid, int flags)
3297 struct nfs4_stateid *local;
3298 u32 st_id = stid->si_stateownerid;
3299 u32 f_id = stid->si_fileid;
3300 unsigned int hashval;
3302 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3303 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3304 hashval = stateid_hashval(st_id, f_id);
3305 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3306 if ((local->st_stateid.si_stateownerid == st_id) &&
3307 (local->st_stateid.si_fileid == f_id))
3312 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3313 hashval = stateid_hashval(st_id, f_id);
3314 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3315 if ((local->st_stateid.si_stateownerid == st_id) &&
3316 (local->st_stateid.si_fileid == f_id))
3323 static struct nfs4_delegation *
3324 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3326 struct nfs4_file *fp;
3327 struct nfs4_delegation *dl;
3329 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3330 stid->si_boot, stid->si_stateownerid,
3331 stid->si_fileid, stid->si_generation);
3333 fp = find_file(ino);
3336 dl = find_delegation_file(fp, stid);
3342 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3343 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3344 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3345 * locking, this prevents us from being completely protocol-compliant. The
3346 * real solution to this problem is to start using unsigned file offsets in
3347 * the VFS, but this is a very deep change!
3350 nfs4_transform_lock_offset(struct file_lock *lock)
3352 if (lock->fl_start < 0)
3353 lock->fl_start = OFFSET_MAX;
3354 if (lock->fl_end < 0)
3355 lock->fl_end = OFFSET_MAX;
3358 /* Hack!: For now, we're defining this just so we can use a pointer to it
3359 * as a unique cookie to identify our (NFSv4's) posix locks. */
3360 static struct lock_manager_operations nfsd_posix_mng_ops = {
3364 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3366 struct nfs4_stateowner *sop;
3369 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3370 sop = (struct nfs4_stateowner *) fl->fl_owner;
3371 hval = lockownerid_hashval(sop->so_id);
3372 kref_get(&sop->so_ref);
3374 deny->ld_clientid = sop->so_client->cl_clientid;
3376 deny->ld_sop = NULL;
3377 deny->ld_clientid.cl_boot = 0;
3378 deny->ld_clientid.cl_id = 0;
3380 deny->ld_start = fl->fl_start;
3381 deny->ld_length = NFS4_MAX_UINT64;
3382 if (fl->fl_end != NFS4_MAX_UINT64)
3383 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3384 deny->ld_type = NFS4_READ_LT;
3385 if (fl->fl_type != F_RDLCK)
3386 deny->ld_type = NFS4_WRITE_LT;
3389 static struct nfs4_stateowner *
3390 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3391 struct xdr_netobj *owner)
3393 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3394 struct nfs4_stateowner *op;
3396 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3397 if (same_owner_str(op, owner, clid))
3404 * Alloc a lock owner structure.
3405 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3408 * strhashval = lock_ownerstr_hashval
3411 static struct nfs4_stateowner *
3412 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3413 struct nfs4_stateowner *sop;
3414 struct nfs4_replay *rp;
3415 unsigned int idhashval;
3417 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3419 idhashval = lockownerid_hashval(current_ownerid);
3420 INIT_LIST_HEAD(&sop->so_idhash);
3421 INIT_LIST_HEAD(&sop->so_strhash);
3422 INIT_LIST_HEAD(&sop->so_perclient);
3423 INIT_LIST_HEAD(&sop->so_stateids);
3424 INIT_LIST_HEAD(&sop->so_perstateid);
3425 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3427 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3428 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3429 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3430 sop->so_is_open_owner = 0;
3431 sop->so_id = current_ownerid++;
3432 sop->so_client = clp;
3433 /* It is the openowner seqid that will be incremented in encode in the
3434 * case of new lockowners; so increment the lock seqid manually: */
3435 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3436 sop->so_confirmed = 1;
3437 rp = &sop->so_replay;
3438 rp->rp_status = nfserr_serverfault;
3440 rp->rp_buf = rp->rp_ibuf;
3444 static struct nfs4_stateid *
3445 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3447 struct nfs4_stateid *stp;
3448 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3450 stp = nfs4_alloc_stateid();
3453 INIT_LIST_HEAD(&stp->st_hash);
3454 INIT_LIST_HEAD(&stp->st_perfile);
3455 INIT_LIST_HEAD(&stp->st_perstateowner);
3456 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3457 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3458 list_add(&stp->st_perfile, &fp->fi_stateids);
3459 list_add(&stp->st_perstateowner, &sop->so_stateids);
3460 stp->st_stateowner = sop;
3463 stp->st_stateid.si_boot = get_seconds();
3464 stp->st_stateid.si_stateownerid = sop->so_id;
3465 stp->st_stateid.si_fileid = fp->fi_id;
3466 stp->st_stateid.si_generation = 0;
3467 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3468 stp->st_access_bmap = open_stp->st_access_bmap;
3469 stp->st_deny_bmap = open_stp->st_deny_bmap;
3470 stp->st_openstp = open_stp;
3477 check_lock_length(u64 offset, u64 length)
3479 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3480 LOFF_OVERFLOW(offset, length)));
3487 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3488 struct nfsd4_lock *lock)
3490 struct nfs4_stateowner *open_sop = NULL;
3491 struct nfs4_stateowner *lock_sop = NULL;
3492 struct nfs4_stateid *lock_stp;
3494 struct file_lock file_lock;
3495 struct file_lock conflock;
3497 unsigned int strhashval;
3501 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3502 (long long) lock->lk_offset,
3503 (long long) lock->lk_length);
3505 if (check_lock_length(lock->lk_offset, lock->lk_length))
3506 return nfserr_inval;
3508 if ((status = fh_verify(rqstp, &cstate->current_fh,
3509 S_IFREG, NFSD_MAY_LOCK))) {
3510 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3516 if (lock->lk_is_new) {
3518 * Client indicates that this is a new lockowner.
3519 * Use open owner and open stateid to create lock owner and
3522 struct nfs4_stateid *open_stp = NULL;
3523 struct nfs4_file *fp;
3525 status = nfserr_stale_clientid;
3526 if (!nfsd4_has_session(cstate) &&
3527 STALE_CLIENTID(&lock->lk_new_clientid))
3530 /* validate and update open stateid and open seqid */
3531 status = nfs4_preprocess_seqid_op(cstate,
3532 lock->lk_new_open_seqid,
3533 &lock->lk_new_open_stateid,
3535 &lock->lk_replay_owner, &open_stp,
3539 open_sop = lock->lk_replay_owner;
3540 /* create lockowner and lock stateid */
3541 fp = open_stp->st_file;
3542 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3543 open_sop->so_client->cl_clientid.cl_id,
3544 &lock->v.new.owner);
3545 /* XXX: Do we need to check for duplicate stateowners on
3546 * the same file, or should they just be allowed (and
3547 * create new stateids)? */
3548 status = nfserr_resource;
3549 lock_sop = alloc_init_lock_stateowner(strhashval,
3550 open_sop->so_client, open_stp, lock);
3551 if (lock_sop == NULL)
3553 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3554 if (lock_stp == NULL)
3557 /* lock (lock owner + lock stateid) already exists */
3558 status = nfs4_preprocess_seqid_op(cstate,
3559 lock->lk_old_lock_seqid,
3560 &lock->lk_old_lock_stateid,
3562 &lock->lk_replay_owner, &lock_stp, lock);
3565 lock_sop = lock->lk_replay_owner;
3567 /* lock->lk_replay_owner and lock_stp have been created or found */
3568 filp = lock_stp->st_vfs_file;
3570 status = nfserr_grace;
3571 if (locks_in_grace() && !lock->lk_reclaim)
3573 status = nfserr_no_grace;
3574 if (!locks_in_grace() && lock->lk_reclaim)
3577 locks_init_lock(&file_lock);
3578 switch (lock->lk_type) {
3581 file_lock.fl_type = F_RDLCK;
3585 case NFS4_WRITEW_LT:
3586 file_lock.fl_type = F_WRLCK;
3590 status = nfserr_inval;
3593 file_lock.fl_owner = (fl_owner_t)lock_sop;
3594 file_lock.fl_pid = current->tgid;
3595 file_lock.fl_file = filp;
3596 file_lock.fl_flags = FL_POSIX;
3597 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3599 file_lock.fl_start = lock->lk_offset;
3600 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3601 nfs4_transform_lock_offset(&file_lock);
3604 * Try to lock the file in the VFS.
3605 * Note: locks.c uses the BKL to protect the inode's lock list.
3608 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3610 case 0: /* success! */
3611 update_stateid(&lock_stp->st_stateid);
3612 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3616 case (EAGAIN): /* conflock holds conflicting lock */
3617 status = nfserr_denied;
3618 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3619 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3622 status = nfserr_deadlock;
3625 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3626 status = nfserr_resource;
3630 if (status && lock->lk_is_new && lock_sop)
3631 release_lockowner(lock_sop);
3632 if (lock->lk_replay_owner) {
3633 nfs4_get_stateowner(lock->lk_replay_owner);
3634 cstate->replay_owner = lock->lk_replay_owner;
3636 nfs4_unlock_state();
3641 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3642 * so we do a temporary open here just to get an open file to pass to
3643 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3646 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3651 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3654 err = vfs_test_lock(file, lock);
3663 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3664 struct nfsd4_lockt *lockt)
3666 struct inode *inode;
3667 struct file_lock file_lock;
3671 if (locks_in_grace())
3672 return nfserr_grace;
3674 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3675 return nfserr_inval;
3677 lockt->lt_stateowner = NULL;
3680 status = nfserr_stale_clientid;
3681 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3684 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3685 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3686 if (status == nfserr_symlink)
3687 status = nfserr_inval;
3691 inode = cstate->current_fh.fh_dentry->d_inode;
3692 locks_init_lock(&file_lock);
3693 switch (lockt->lt_type) {
3696 file_lock.fl_type = F_RDLCK;
3699 case NFS4_WRITEW_LT:
3700 file_lock.fl_type = F_WRLCK;
3703 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3704 status = nfserr_inval;
3708 lockt->lt_stateowner = find_lockstateowner_str(inode,
3709 &lockt->lt_clientid, &lockt->lt_owner);
3710 if (lockt->lt_stateowner)
3711 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3712 file_lock.fl_pid = current->tgid;
3713 file_lock.fl_flags = FL_POSIX;
3715 file_lock.fl_start = lockt->lt_offset;
3716 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3718 nfs4_transform_lock_offset(&file_lock);
3721 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3723 status = nfserrno(error);
3726 if (file_lock.fl_type != F_UNLCK) {
3727 status = nfserr_denied;
3728 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3731 nfs4_unlock_state();
3736 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3737 struct nfsd4_locku *locku)
3739 struct nfs4_stateid *stp;
3740 struct file *filp = NULL;
3741 struct file_lock file_lock;
3745 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3746 (long long) locku->lu_offset,
3747 (long long) locku->lu_length);
3749 if (check_lock_length(locku->lu_offset, locku->lu_length))
3750 return nfserr_inval;
3754 if ((status = nfs4_preprocess_seqid_op(cstate,
3758 &locku->lu_stateowner, &stp, NULL)))
3761 filp = stp->st_vfs_file;
3763 locks_init_lock(&file_lock);
3764 file_lock.fl_type = F_UNLCK;
3765 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3766 file_lock.fl_pid = current->tgid;
3767 file_lock.fl_file = filp;
3768 file_lock.fl_flags = FL_POSIX;
3769 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3770 file_lock.fl_start = locku->lu_offset;
3772 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3773 nfs4_transform_lock_offset(&file_lock);
3776 * Try to unlock the file in the VFS.
3778 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3780 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3784 * OK, unlock succeeded; the only thing left to do is update the stateid.
3786 update_stateid(&stp->st_stateid);
3787 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3790 if (locku->lu_stateowner) {
3791 nfs4_get_stateowner(locku->lu_stateowner);
3792 cstate->replay_owner = locku->lu_stateowner;
3794 nfs4_unlock_state();
3798 status = nfserrno(err);
3804 * 1: locks held by lockowner
3805 * 0: no locks held by lockowner
3808 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3810 struct file_lock **flpp;
3811 struct inode *inode = filp->f_path.dentry->d_inode;
3815 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3816 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3827 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3828 struct nfsd4_compound_state *cstate,
3829 struct nfsd4_release_lockowner *rlockowner)
3831 clientid_t *clid = &rlockowner->rl_clientid;
3832 struct nfs4_stateowner *sop;
3833 struct nfs4_stateid *stp;
3834 struct xdr_netobj *owner = &rlockowner->rl_owner;
3835 struct list_head matches;
3839 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3840 clid->cl_boot, clid->cl_id);
3842 /* XXX check for lease expiration */
3844 status = nfserr_stale_clientid;
3845 if (STALE_CLIENTID(clid))
3850 status = nfserr_locks_held;
3851 /* XXX: we're doing a linear search through all the lockowners.
3852 * Yipes! For now we'll just hope clients aren't really using
3853 * release_lockowner much, but eventually we have to fix these
3854 * data structures. */
3855 INIT_LIST_HEAD(&matches);
3856 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3857 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3858 if (!same_owner_str(sop, owner, clid))
3860 list_for_each_entry(stp, &sop->so_stateids,
3862 if (check_for_locks(stp->st_vfs_file, sop))
3864 /* Note: so_perclient unused for lockowners,
3865 * so it's OK to fool with here. */
3866 list_add(&sop->so_perclient, &matches);
3870 /* Clients probably won't expect us to return with some (but not all)
3871 * of the lockowner state released; so don't release any until all
3872 * have been checked. */
3874 while (!list_empty(&matches)) {
3875 sop = list_entry(matches.next, struct nfs4_stateowner,
3877 /* unhash_stateowner deletes so_perclient only
3878 * for openowners. */
3879 list_del(&sop->so_perclient);
3880 release_lockowner(sop);
3883 nfs4_unlock_state();
3887 static inline struct nfs4_client_reclaim *
3890 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3894 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3896 unsigned int strhashval = clientstr_hashval(name);
3897 struct nfs4_client *clp;
3899 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3904 * failure => all reset bets are off, nfserr_no_grace...
3907 nfs4_client_to_reclaim(const char *name)
3909 unsigned int strhashval;
3910 struct nfs4_client_reclaim *crp = NULL;
3912 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3913 crp = alloc_reclaim();
3916 strhashval = clientstr_hashval(name);
3917 INIT_LIST_HEAD(&crp->cr_strhash);
3918 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3919 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3920 reclaim_str_hashtbl_size++;
3925 nfs4_release_reclaim(void)
3927 struct nfs4_client_reclaim *crp = NULL;
3930 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3931 while (!list_empty(&reclaim_str_hashtbl[i])) {
3932 crp = list_entry(reclaim_str_hashtbl[i].next,
3933 struct nfs4_client_reclaim, cr_strhash);
3934 list_del(&crp->cr_strhash);
3936 reclaim_str_hashtbl_size--;
3939 BUG_ON(reclaim_str_hashtbl_size);
3943 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3944 static struct nfs4_client_reclaim *
3945 nfs4_find_reclaim_client(clientid_t *clid)
3947 unsigned int strhashval;
3948 struct nfs4_client *clp;
3949 struct nfs4_client_reclaim *crp = NULL;
3952 /* find clientid in conf_id_hashtbl */
3953 clp = find_confirmed_client(clid);
3957 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3958 clp->cl_name.len, clp->cl_name.data,
3961 /* find clp->cl_name in reclaim_str_hashtbl */
3962 strhashval = clientstr_hashval(clp->cl_recdir);
3963 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3964 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3972 * Called from OPEN. Look for clientid in reclaim list.
3975 nfs4_check_open_reclaim(clientid_t *clid)
3977 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3980 /* initialization to perform at module load time: */
3983 nfs4_state_init(void)
3987 status = nfsd4_init_slabs();
3990 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3991 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3992 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3993 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3994 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3995 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3997 for (i = 0; i < SESSION_HASH_SIZE; i++)
3998 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3999 for (i = 0; i < FILE_HASH_SIZE; i++) {
4000 INIT_LIST_HEAD(&file_hashtbl[i]);
4002 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4003 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4004 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4006 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4007 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4008 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4010 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4011 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4012 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4014 memset(&onestateid, ~0, sizeof(stateid_t));
4015 INIT_LIST_HEAD(&close_lru);
4016 INIT_LIST_HEAD(&client_lru);
4017 INIT_LIST_HEAD(&del_recall_lru);
4018 reclaim_str_hashtbl_size = 0;
4023 nfsd4_load_reboot_recovery_data(void)
4028 nfsd4_init_recdir(user_recovery_dirname);
4029 status = nfsd4_recdir_load();
4030 nfs4_unlock_state();
4032 printk("NFSD: Failure reading reboot recovery data\n");
4036 get_nfs4_grace_period(void)
4038 return max(user_lease_time, lease_time) * HZ;
4042 * Since the lifetime of a delegation isn't limited to that of an open, a
4043 * client may quite reasonably hang on to a delegation as long as it has
4044 * the inode cached. This becomes an obvious problem the first time a
4045 * client's inode cache approaches the size of the server's total memory.
4047 * For now we avoid this problem by imposing a hard limit on the number
4048 * of delegations, which varies according to the server's memory size.
4051 set_max_delegations(void)
4054 * Allow at most 4 delegations per megabyte of RAM. Quick
4055 * estimates suggest that in the worst case (where every delegation
4056 * is for a different inode), a delegation could take about 1.5K,
4057 * giving a worst case usage of about 6% of memory.
4059 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4062 /* initialization to perform when the nfsd service is started: */
4065 __nfs4_state_start(void)
4067 unsigned long grace_time;
4069 boot_time = get_seconds();
4070 grace_time = get_nfs4_grace_period();
4071 lease_time = user_lease_time;
4072 locks_start_grace(&nfsd4_manager);
4073 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4075 laundry_wq = create_singlethread_workqueue("nfsd4");
4076 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4077 set_max_delegations();
4081 nfs4_state_start(void)
4085 nfsd4_load_reboot_recovery_data();
4086 __nfs4_state_start();
4092 nfs4_lease_time(void)
4098 __nfs4_state_shutdown(void)
4101 struct nfs4_client *clp = NULL;
4102 struct nfs4_delegation *dp = NULL;
4103 struct list_head *pos, *next, reaplist;
4105 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4106 while (!list_empty(&conf_id_hashtbl[i])) {
4107 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4110 while (!list_empty(&unconf_str_hashtbl[i])) {
4111 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4115 INIT_LIST_HEAD(&reaplist);
4116 spin_lock(&recall_lock);
4117 list_for_each_safe(pos, next, &del_recall_lru) {
4118 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4119 list_move(&dp->dl_recall_lru, &reaplist);
4121 spin_unlock(&recall_lock);
4122 list_for_each_safe(pos, next, &reaplist) {
4123 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4124 list_del_init(&dp->dl_recall_lru);
4125 unhash_delegation(dp);
4128 nfsd4_shutdown_recdir();
4133 nfs4_state_shutdown(void)
4135 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4136 destroy_workqueue(laundry_wq);
4137 locks_end_grace(&nfsd4_manager);
4139 nfs4_release_reclaim();
4140 __nfs4_state_shutdown();
4141 nfs4_unlock_state();
4145 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4146 * accessed when nfsd is starting.
4149 nfs4_set_recdir(char *recdir)
4151 strcpy(user_recovery_dirname, recdir);
4155 * Change the NFSv4 recovery directory to recdir.
4158 nfs4_reset_recoverydir(char *recdir)
4163 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4167 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4168 nfs4_set_recdir(recdir);
4176 nfs4_recoverydir(void)
4178 return user_recovery_dirname;
4182 * Called when leasetime is changed.
4184 * The only way the protocol gives us to handle on-the-fly lease changes is to
4185 * simulate a reboot. Instead of doing that, we just wait till the next time
4186 * we start to register any changes in lease time. If the administrator
4187 * really wants to change the lease time *now*, they can go ahead and bring
4188 * nfsd down and then back up again after changing the lease time.
4190 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4191 * when nfsd is starting
4194 nfs4_reset_lease(time_t leasetime)
4196 user_lease_time = leasetime;