2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58 #include <linux/sunrpc/clnt.h>
60 #define NFSDDBG_FACILITY NFSDDBG_PROC
63 static time_t lease_time = 90; /* default lease time */
64 static time_t user_lease_time = 90;
65 static time_t boot_time;
66 static u32 current_ownerid = 1;
67 static u32 current_fileid = 1;
68 static u32 current_delegid = 1;
70 static stateid_t zerostateid; /* bits all 0 */
71 static stateid_t onestateid; /* bits all 1 */
72 static u64 current_sessionid = 1;
74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
75 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
77 /* forward declarations */
78 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
79 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
80 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
81 static void nfs4_set_recdir(char *recdir);
85 /* Currently used for almost all code touching nfsv4 state: */
86 static DEFINE_MUTEX(client_mutex);
89 * Currently used for the del_recall_lru and file hash table. In an
90 * effort to decrease the scope of the client_mutex, this spinlock may
91 * eventually cover more:
93 static DEFINE_SPINLOCK(recall_lock);
95 static struct kmem_cache *stateowner_slab = NULL;
96 static struct kmem_cache *file_slab = NULL;
97 static struct kmem_cache *stateid_slab = NULL;
98 static struct kmem_cache *deleg_slab = NULL;
101 nfs4_lock_state(void)
103 mutex_lock(&client_mutex);
107 nfs4_unlock_state(void)
109 mutex_unlock(&client_mutex);
113 opaque_hashval(const void *ptr, int nbytes)
115 unsigned char *cptr = (unsigned char *) ptr;
125 static struct list_head del_recall_lru;
128 put_nfs4_file(struct nfs4_file *fi)
130 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
131 list_del(&fi->fi_hash);
132 spin_unlock(&recall_lock);
134 kmem_cache_free(file_slab, fi);
139 get_nfs4_file(struct nfs4_file *fi)
141 atomic_inc(&fi->fi_ref);
144 static int num_delegations;
145 unsigned int max_delegations;
148 * Open owner state (share locks)
151 /* hash tables for nfs4_stateowner */
152 #define OWNER_HASH_BITS 8
153 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
154 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
156 #define ownerid_hashval(id) \
157 ((id) & OWNER_HASH_MASK)
158 #define ownerstr_hashval(clientid, ownername) \
159 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
161 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
162 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
164 /* hash table for nfs4_file */
165 #define FILE_HASH_BITS 8
166 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
167 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
168 /* hash table for (open)nfs4_stateid */
169 #define STATEID_HASH_BITS 10
170 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
171 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
173 #define file_hashval(x) \
174 hash_ptr(x, FILE_HASH_BITS)
175 #define stateid_hashval(owner_id, file_id) \
176 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
178 static struct list_head file_hashtbl[FILE_HASH_SIZE];
179 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
181 static struct nfs4_delegation *
182 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
184 struct nfs4_delegation *dp;
185 struct nfs4_file *fp = stp->st_file;
186 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
188 dprintk("NFSD alloc_init_deleg\n");
189 if (fp->fi_had_conflict)
191 if (num_delegations > max_delegations)
193 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
197 INIT_LIST_HEAD(&dp->dl_perfile);
198 INIT_LIST_HEAD(&dp->dl_perclnt);
199 INIT_LIST_HEAD(&dp->dl_recall_lru);
204 get_file(stp->st_vfs_file);
205 dp->dl_vfs_file = stp->st_vfs_file;
207 dp->dl_ident = cb->cb_ident;
208 dp->dl_stateid.si_boot = get_seconds();
209 dp->dl_stateid.si_stateownerid = current_delegid++;
210 dp->dl_stateid.si_fileid = 0;
211 dp->dl_stateid.si_generation = 0;
212 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
214 atomic_set(&dp->dl_count, 1);
215 list_add(&dp->dl_perfile, &fp->fi_delegations);
216 list_add(&dp->dl_perclnt, &clp->cl_delegations);
221 nfs4_put_delegation(struct nfs4_delegation *dp)
223 if (atomic_dec_and_test(&dp->dl_count)) {
224 dprintk("NFSD: freeing dp %p\n",dp);
225 put_nfs4_file(dp->dl_file);
226 kmem_cache_free(deleg_slab, dp);
231 /* Remove the associated file_lock first, then remove the delegation.
232 * lease_modify() is called to remove the FS_LEASE file_lock from
233 * the i_flock list, eventually calling nfsd's lock_manager
234 * fl_release_callback.
237 nfs4_close_delegation(struct nfs4_delegation *dp)
239 struct file *filp = dp->dl_vfs_file;
241 dprintk("NFSD: close_delegation dp %p\n",dp);
242 dp->dl_vfs_file = NULL;
243 /* The following nfsd_close may not actually close the file,
244 * but we want to remove the lease in any case. */
246 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
250 /* Called under the state lock. */
252 unhash_delegation(struct nfs4_delegation *dp)
254 list_del_init(&dp->dl_perfile);
255 list_del_init(&dp->dl_perclnt);
256 spin_lock(&recall_lock);
257 list_del_init(&dp->dl_recall_lru);
258 spin_unlock(&recall_lock);
259 nfs4_close_delegation(dp);
260 nfs4_put_delegation(dp);
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS 4
269 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
272 #define clientid_hashval(id) \
273 ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
277 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278 * used in reboot/reset lease grace period processing
280 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281 * setclientid_confirmed info.
283 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
286 * client_lru holds client queue ordered by nfs4_client.cl_time
289 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290 * for last close replay.
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
303 list_del(&stp->st_hash);
304 list_del(&stp->st_perfile);
305 list_del(&stp->st_perstateowner);
308 static void free_generic_stateid(struct nfs4_stateid *stp)
310 put_nfs4_file(stp->st_file);
311 kmem_cache_free(stateid_slab, stp);
314 static void release_lock_stateid(struct nfs4_stateid *stp)
316 unhash_generic_stateid(stp);
317 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318 free_generic_stateid(stp);
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
323 struct nfs4_stateid *stp;
325 list_del(&sop->so_idhash);
326 list_del(&sop->so_strhash);
327 list_del(&sop->so_perstateid);
328 while (!list_empty(&sop->so_stateids)) {
329 stp = list_first_entry(&sop->so_stateids,
330 struct nfs4_stateid, st_perstateowner);
331 release_lock_stateid(stp);
335 static void release_lockowner(struct nfs4_stateowner *sop)
337 unhash_lockowner(sop);
338 nfs4_put_stateowner(sop);
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
344 struct nfs4_stateowner *lock_sop;
346 while (!list_empty(&open_stp->st_lockowners)) {
347 lock_sop = list_entry(open_stp->st_lockowners.next,
348 struct nfs4_stateowner, so_perstateid);
349 /* list_del(&open_stp->st_lockowners); */
350 BUG_ON(lock_sop->so_is_open_owner);
351 release_lockowner(lock_sop);
355 static void release_open_stateid(struct nfs4_stateid *stp)
357 unhash_generic_stateid(stp);
358 release_stateid_lockowners(stp);
359 nfsd_close(stp->st_vfs_file);
360 free_generic_stateid(stp);
363 static void unhash_openowner(struct nfs4_stateowner *sop)
365 struct nfs4_stateid *stp;
367 list_del(&sop->so_idhash);
368 list_del(&sop->so_strhash);
369 list_del(&sop->so_perclient);
370 list_del(&sop->so_perstateid); /* XXX: necessary? */
371 while (!list_empty(&sop->so_stateids)) {
372 stp = list_first_entry(&sop->so_stateids,
373 struct nfs4_stateid, st_perstateowner);
374 release_open_stateid(stp);
378 static void release_openowner(struct nfs4_stateowner *sop)
380 unhash_openowner(sop);
381 list_del(&sop->so_close_lru);
382 nfs4_put_stateowner(sop);
385 static DEFINE_SPINLOCK(sessionid_lock);
386 #define SESSION_HASH_SIZE 512
387 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
390 hash_sessionid(struct nfs4_sessionid *sessionid)
392 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
394 return sid->sequence % SESSION_HASH_SIZE;
398 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
400 u32 *ptr = (u32 *)(&sessionid->data[0]);
401 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
405 gen_sessionid(struct nfsd4_session *ses)
407 struct nfs4_client *clp = ses->se_client;
408 struct nfsd4_sessionid *sid;
410 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
411 sid->clientid = clp->cl_clientid;
412 sid->sequence = current_sessionid++;
417 * The protocol defines ca_maxresponssize_cached to include the size of
418 * the rpc header, but all we need to cache is the data starting after
419 * the end of the initial SEQUENCE operation--the rest we regenerate
420 * each time. Therefore we can advertise a ca_maxresponssize_cached
421 * value that is the number of bytes in our cache plus a few additional
422 * bytes. In order to stay on the safe side, and not promise more than
423 * we can cache, those additional bytes must be the minimum possible: 24
424 * bytes of rpc header (xid through accept state, with AUTH_NULL
425 * verifier), 12 for the compound header (with zero-length tag), and 44
426 * for the SEQUENCE op response:
428 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
431 * Give the client the number of ca_maxresponsesize_cached slots it
432 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
433 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
434 * than NFSD_MAX_SLOTS_PER_SESSION.
436 * If we run out of reserved DRC memory we should (up to a point)
437 * re-negotiate active sessions and reduce their slot usage to make
438 * rooom for new connections. For now we just fail the create session.
440 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
442 int mem, size = fchan->maxresp_cached;
444 if (fchan->maxreqs < 1)
447 if (size < NFSD_MIN_HDR_SEQ_SZ)
448 size = NFSD_MIN_HDR_SEQ_SZ;
449 size -= NFSD_MIN_HDR_SEQ_SZ;
450 if (size > NFSD_SLOT_CACHE_SIZE)
451 size = NFSD_SLOT_CACHE_SIZE;
453 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
454 mem = fchan->maxreqs * size;
455 if (mem > NFSD_MAX_MEM_PER_SESSION) {
456 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
457 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
458 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
459 mem = fchan->maxreqs * size;
462 spin_lock(&nfsd_drc_lock);
463 /* bound the total session drc memory ussage */
464 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
465 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
466 mem = fchan->maxreqs * size;
468 nfsd_drc_mem_used += mem;
469 spin_unlock(&nfsd_drc_lock);
471 if (fchan->maxreqs == 0)
472 return nfserr_serverfault;
474 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
479 * fchan holds the client values on input, and the server values on output
480 * sv_max_mesg is the maximum payload plus one page for overhead.
482 static int init_forechannel_attrs(struct svc_rqst *rqstp,
483 struct nfsd4_channel_attrs *session_fchan,
484 struct nfsd4_channel_attrs *fchan)
487 __u32 maxcount = nfsd_serv->sv_max_mesg;
489 /* headerpadsz set to zero in encode routine */
491 /* Use the client's max request and max response size if possible */
492 if (fchan->maxreq_sz > maxcount)
493 fchan->maxreq_sz = maxcount;
494 session_fchan->maxreq_sz = fchan->maxreq_sz;
496 if (fchan->maxresp_sz > maxcount)
497 fchan->maxresp_sz = maxcount;
498 session_fchan->maxresp_sz = fchan->maxresp_sz;
500 /* Use the client's maxops if possible */
501 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
502 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
503 session_fchan->maxops = fchan->maxops;
505 /* FIXME: Error means no more DRC pages so the server should
506 * recover pages from existing sessions. For now fail session
509 status = set_forechannel_drc_size(fchan);
511 session_fchan->maxresp_cached = fchan->maxresp_cached;
512 session_fchan->maxreqs = fchan->maxreqs;
514 dprintk("%s status %d\n", __func__, status);
519 free_session_slots(struct nfsd4_session *ses)
523 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
524 kfree(ses->se_slots[i]);
528 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
529 struct nfsd4_create_session *cses)
531 struct nfsd4_session *new, tmp;
532 struct nfsd4_slot *sp;
533 int idx, slotsize, cachesize, i;
536 memset(&tmp, 0, sizeof(tmp));
538 /* FIXME: For now, we just accept the client back channel attributes. */
539 tmp.se_bchannel = cses->back_channel;
540 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
541 &cses->fore_channel);
545 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
546 + sizeof(struct nfsd4_session) > PAGE_SIZE);
548 status = nfserr_serverfault;
549 /* allocate struct nfsd4_session and slot table pointers in one piece */
550 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
551 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
555 memcpy(new, &tmp, sizeof(*new));
557 /* allocate each struct nfsd4_slot and data cache in one piece */
558 cachesize = new->se_fchannel.maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
559 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
560 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
563 new->se_slots[i] = sp;
566 new->se_client = clp;
568 idx = hash_sessionid(&new->se_sessionid);
569 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
570 NFS4_MAX_SESSIONID_LEN);
572 new->se_flags = cses->flags;
573 kref_init(&new->se_ref);
574 spin_lock(&sessionid_lock);
575 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
576 list_add(&new->se_perclnt, &clp->cl_sessions);
577 spin_unlock(&sessionid_lock);
583 free_session_slots(new);
588 /* caller must hold sessionid_lock */
589 static struct nfsd4_session *
590 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
592 struct nfsd4_session *elem;
595 dump_sessionid(__func__, sessionid);
596 idx = hash_sessionid(sessionid);
597 dprintk("%s: idx is %d\n", __func__, idx);
598 /* Search in the appropriate list */
599 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
600 dump_sessionid("list traversal", &elem->se_sessionid);
601 if (!memcmp(elem->se_sessionid.data, sessionid->data,
602 NFS4_MAX_SESSIONID_LEN)) {
607 dprintk("%s: session not found\n", __func__);
611 /* caller must hold sessionid_lock */
613 unhash_session(struct nfsd4_session *ses)
615 list_del(&ses->se_hash);
616 list_del(&ses->se_perclnt);
620 release_session(struct nfsd4_session *ses)
622 spin_lock(&sessionid_lock);
624 spin_unlock(&sessionid_lock);
625 nfsd4_put_session(ses);
629 free_session(struct kref *kref)
631 struct nfsd4_session *ses;
634 ses = container_of(kref, struct nfsd4_session, se_ref);
635 spin_lock(&nfsd_drc_lock);
636 mem = ses->se_fchannel.maxreqs
637 * (ses->se_fchannel.maxresp_cached - NFSD_MIN_HDR_SEQ_SZ);
638 nfsd_drc_mem_used -= mem;
639 spin_unlock(&nfsd_drc_lock);
640 free_session_slots(ses);
645 renew_client(struct nfs4_client *clp)
648 * Move client to the end to the LRU list.
650 dprintk("renewing client (clientid %08x/%08x)\n",
651 clp->cl_clientid.cl_boot,
652 clp->cl_clientid.cl_id);
653 list_move_tail(&clp->cl_lru, &client_lru);
654 clp->cl_time = get_seconds();
657 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
659 STALE_CLIENTID(clientid_t *clid)
661 if (clid->cl_boot == boot_time)
663 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
664 clid->cl_boot, clid->cl_id, boot_time);
669 * XXX Should we use a slab cache ?
670 * This type of memory management is somewhat inefficient, but we use it
671 * anyway since SETCLIENTID is not a common operation.
673 static struct nfs4_client *alloc_client(struct xdr_netobj name)
675 struct nfs4_client *clp;
677 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
680 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
681 if (clp->cl_name.data == NULL) {
685 memcpy(clp->cl_name.data, name.data, name.len);
686 clp->cl_name.len = name.len;
691 shutdown_callback_client(struct nfs4_client *clp)
693 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
697 * Callback threads take a reference on the client, so there
698 * should be no outstanding callbacks at this point.
700 clp->cl_cb_conn.cb_client = NULL;
701 rpc_shutdown_client(clnt);
706 free_client(struct nfs4_client *clp)
708 shutdown_callback_client(clp);
710 svc_xprt_put(clp->cl_cb_xprt);
711 if (clp->cl_cred.cr_group_info)
712 put_group_info(clp->cl_cred.cr_group_info);
713 kfree(clp->cl_principal);
714 kfree(clp->cl_name.data);
719 put_nfs4_client(struct nfs4_client *clp)
721 if (atomic_dec_and_test(&clp->cl_count))
726 expire_client(struct nfs4_client *clp)
728 struct nfs4_stateowner *sop;
729 struct nfs4_delegation *dp;
730 struct list_head reaplist;
732 dprintk("NFSD: expire_client cl_count %d\n",
733 atomic_read(&clp->cl_count));
735 INIT_LIST_HEAD(&reaplist);
736 spin_lock(&recall_lock);
737 while (!list_empty(&clp->cl_delegations)) {
738 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
739 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
741 list_del_init(&dp->dl_perclnt);
742 list_move(&dp->dl_recall_lru, &reaplist);
744 spin_unlock(&recall_lock);
745 while (!list_empty(&reaplist)) {
746 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
747 list_del_init(&dp->dl_recall_lru);
748 unhash_delegation(dp);
750 list_del(&clp->cl_idhash);
751 list_del(&clp->cl_strhash);
752 list_del(&clp->cl_lru);
753 while (!list_empty(&clp->cl_openowners)) {
754 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
755 release_openowner(sop);
757 while (!list_empty(&clp->cl_sessions)) {
758 struct nfsd4_session *ses;
759 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
761 release_session(ses);
763 put_nfs4_client(clp);
766 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
768 memcpy(target->cl_verifier.data, source->data,
769 sizeof(target->cl_verifier.data));
772 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
774 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
775 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
778 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
780 target->cr_uid = source->cr_uid;
781 target->cr_gid = source->cr_gid;
782 target->cr_group_info = source->cr_group_info;
783 get_group_info(target->cr_group_info);
786 static int same_name(const char *n1, const char *n2)
788 return 0 == memcmp(n1, n2, HEXDIR_LEN);
792 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
794 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
798 same_clid(clientid_t *cl1, clientid_t *cl2)
800 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
803 /* XXX what about NGROUP */
805 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
807 return cr1->cr_uid == cr2->cr_uid;
810 static void gen_clid(struct nfs4_client *clp)
812 static u32 current_clientid = 1;
814 clp->cl_clientid.cl_boot = boot_time;
815 clp->cl_clientid.cl_id = current_clientid++;
818 static void gen_confirm(struct nfs4_client *clp)
823 p = (u32 *)clp->cl_confirm.data;
824 *p++ = get_seconds();
828 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
829 struct svc_rqst *rqstp, nfs4_verifier *verf)
831 struct nfs4_client *clp;
832 struct sockaddr *sa = svc_addr(rqstp);
835 clp = alloc_client(name);
839 princ = svc_gss_principal(rqstp);
841 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
842 if (clp->cl_principal == NULL) {
848 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
849 atomic_set(&clp->cl_count, 1);
850 atomic_set(&clp->cl_cb_conn.cb_set, 0);
851 INIT_LIST_HEAD(&clp->cl_idhash);
852 INIT_LIST_HEAD(&clp->cl_strhash);
853 INIT_LIST_HEAD(&clp->cl_openowners);
854 INIT_LIST_HEAD(&clp->cl_delegations);
855 INIT_LIST_HEAD(&clp->cl_sessions);
856 INIT_LIST_HEAD(&clp->cl_lru);
857 clear_bit(0, &clp->cl_cb_slot_busy);
858 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
859 copy_verf(clp, verf);
860 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
861 clp->cl_flavor = rqstp->rq_flavor;
862 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
868 static int check_name(struct xdr_netobj name)
872 if (name.len > NFS4_OPAQUE_LIMIT) {
873 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
880 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
882 unsigned int idhashval;
884 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
885 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
886 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
887 list_add_tail(&clp->cl_lru, &client_lru);
888 clp->cl_time = get_seconds();
892 move_to_confirmed(struct nfs4_client *clp)
894 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
895 unsigned int strhashval;
897 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
898 list_del_init(&clp->cl_strhash);
899 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
900 strhashval = clientstr_hashval(clp->cl_recdir);
901 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
905 static struct nfs4_client *
906 find_confirmed_client(clientid_t *clid)
908 struct nfs4_client *clp;
909 unsigned int idhashval = clientid_hashval(clid->cl_id);
911 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
912 if (same_clid(&clp->cl_clientid, clid))
918 static struct nfs4_client *
919 find_unconfirmed_client(clientid_t *clid)
921 struct nfs4_client *clp;
922 unsigned int idhashval = clientid_hashval(clid->cl_id);
924 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
925 if (same_clid(&clp->cl_clientid, clid))
932 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
933 * parameter. Matching is based on the fact the at least one of the
934 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
936 * FIXME: we need to unify the clientid namespaces for nfsv4.x
937 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
938 * and SET_CLIENTID{,_CONFIRM}
941 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
943 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
944 return use_exchange_id == has_exchange_flags;
947 static struct nfs4_client *
948 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
949 bool use_exchange_id)
951 struct nfs4_client *clp;
953 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
954 if (same_name(clp->cl_recdir, dname) &&
955 match_clientid_establishment(clp, use_exchange_id))
961 static struct nfs4_client *
962 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
963 bool use_exchange_id)
965 struct nfs4_client *clp;
967 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
968 if (same_name(clp->cl_recdir, dname) &&
969 match_clientid_establishment(clp, use_exchange_id))
976 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
978 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
979 unsigned short expected_family;
981 /* Currently, we only support tcp and tcp6 for the callback channel */
982 if (se->se_callback_netid_len == 3 &&
983 !memcmp(se->se_callback_netid_val, "tcp", 3))
984 expected_family = AF_INET;
985 else if (se->se_callback_netid_len == 4 &&
986 !memcmp(se->se_callback_netid_val, "tcp6", 4))
987 expected_family = AF_INET6;
991 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
992 se->se_callback_addr_len,
993 (struct sockaddr *) &cb->cb_addr,
994 sizeof(cb->cb_addr));
996 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
999 if (cb->cb_addr.ss_family == AF_INET6)
1000 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
1002 cb->cb_minorversion = 0;
1003 cb->cb_prog = se->se_callback_prog;
1004 cb->cb_ident = se->se_callback_ident;
1007 cb->cb_addr.ss_family = AF_UNSPEC;
1009 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1010 "will not receive delegations\n",
1011 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1017 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1020 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1022 struct nfsd4_slot *slot = resp->cstate.slot;
1025 dprintk("--> %s slot %p\n", __func__, slot);
1027 slot->sl_opcnt = resp->opcnt;
1028 slot->sl_status = resp->cstate.status;
1030 if (nfsd4_not_cached(resp)) {
1031 slot->sl_datalen = 0;
1034 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1035 base = (char *)resp->cstate.datap -
1036 (char *)resp->xbuf->head[0].iov_base;
1037 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1039 WARN("%s: sessions DRC could not cache compound\n", __func__);
1044 * Encode the replay sequence operation from the slot values.
1045 * If cachethis is FALSE encode the uncached rep error on the next
1046 * operation which sets resp->p and increments resp->opcnt for
1047 * nfs4svc_encode_compoundres.
1051 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1052 struct nfsd4_compoundres *resp)
1054 struct nfsd4_op *op;
1055 struct nfsd4_slot *slot = resp->cstate.slot;
1057 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1058 resp->opcnt, resp->cstate.slot->sl_cachethis);
1060 /* Encode the replayed sequence operation */
1061 op = &args->ops[resp->opcnt - 1];
1062 nfsd4_encode_operation(resp, op);
1064 /* Return nfserr_retry_uncached_rep in next operation. */
1065 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1066 op = &args->ops[resp->opcnt++];
1067 op->status = nfserr_retry_uncached_rep;
1068 nfsd4_encode_operation(resp, op);
1074 * The sequence operation is not cached because we can use the slot and
1078 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1079 struct nfsd4_sequence *seq)
1081 struct nfsd4_slot *slot = resp->cstate.slot;
1084 dprintk("--> %s slot %p\n", __func__, slot);
1086 /* Either returns 0 or nfserr_retry_uncached */
1087 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1088 if (status == nfserr_retry_uncached_rep)
1091 /* The sequence operation has been encoded, cstate->datap set. */
1092 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1094 resp->opcnt = slot->sl_opcnt;
1095 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1096 status = slot->sl_status;
1102 * Set the exchange_id flags returned by the server.
1105 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1107 /* pNFS is not supported */
1108 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1110 /* Referrals are supported, Migration is not. */
1111 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1113 /* set the wire flags to return to client. */
1114 clid->flags = new->cl_exchange_flags;
1118 nfsd4_exchange_id(struct svc_rqst *rqstp,
1119 struct nfsd4_compound_state *cstate,
1120 struct nfsd4_exchange_id *exid)
1122 struct nfs4_client *unconf, *conf, *new;
1124 unsigned int strhashval;
1125 char dname[HEXDIR_LEN];
1126 char addr_str[INET6_ADDRSTRLEN];
1127 nfs4_verifier verf = exid->verifier;
1128 struct sockaddr *sa = svc_addr(rqstp);
1130 rpc_ntop(sa, addr_str, sizeof(addr_str));
1131 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1132 "ip_addr=%s flags %x, spa_how %d\n",
1133 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1134 addr_str, exid->flags, exid->spa_how);
1136 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1137 return nfserr_inval;
1139 /* Currently only support SP4_NONE */
1140 switch (exid->spa_how) {
1144 return nfserr_encr_alg_unsupp;
1146 BUG(); /* checked by xdr code */
1148 return nfserr_serverfault; /* no excuse :-/ */
1151 status = nfs4_make_rec_clidname(dname, &exid->clname);
1156 strhashval = clientstr_hashval(dname);
1161 conf = find_confirmed_client_by_str(dname, strhashval, true);
1163 if (!same_verf(&verf, &conf->cl_verifier)) {
1164 /* 18.35.4 case 8 */
1165 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1166 status = nfserr_not_same;
1169 /* Client reboot: destroy old state */
1170 expire_client(conf);
1173 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1174 /* 18.35.4 case 9 */
1175 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1176 status = nfserr_perm;
1179 expire_client(conf);
1183 * Set bit when the owner id and verifier map to an already
1184 * confirmed client id (18.35.3).
1186 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1189 * Falling into 18.35.4 case 2, possible router replay.
1190 * Leave confirmed record intact and return same result.
1192 copy_verf(conf, &verf);
1197 /* 18.35.4 case 7 */
1198 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1199 status = nfserr_noent;
1203 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1206 * Possible retry or client restart. Per 18.35.4 case 4,
1207 * a new unconfirmed record should be generated regardless
1208 * of whether any properties have changed.
1210 expire_client(unconf);
1215 new = create_client(exid->clname, dname, rqstp, &verf);
1217 status = nfserr_serverfault;
1222 add_to_unconfirmed(new, strhashval);
1224 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1225 exid->clientid.cl_id = new->cl_clientid.cl_id;
1228 nfsd4_set_ex_flags(new, exid);
1230 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1231 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1235 nfs4_unlock_state();
1237 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1242 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1244 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1247 /* The slot is in use, and no response has been sent. */
1249 if (seqid == slot_seqid)
1250 return nfserr_jukebox;
1252 return nfserr_seq_misordered;
1255 if (likely(seqid == slot_seqid + 1))
1258 if (seqid == slot_seqid)
1259 return nfserr_replay_cache;
1261 if (seqid == 1 && (slot_seqid + 1) == 0)
1263 /* Misordered replay or misordered new request */
1264 return nfserr_seq_misordered;
1268 * Cache the create session result into the create session single DRC
1269 * slot cache by saving the xdr structure. sl_seqid has been set.
1270 * Do this for solo or embedded create session operations.
1273 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1274 struct nfsd4_clid_slot *slot, int nfserr)
1276 slot->sl_status = nfserr;
1277 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1281 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1282 struct nfsd4_clid_slot *slot)
1284 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1285 return slot->sl_status;
1289 nfsd4_create_session(struct svc_rqst *rqstp,
1290 struct nfsd4_compound_state *cstate,
1291 struct nfsd4_create_session *cr_ses)
1293 struct sockaddr *sa = svc_addr(rqstp);
1294 struct nfs4_client *conf, *unconf;
1295 struct nfsd4_clid_slot *cs_slot = NULL;
1299 unconf = find_unconfirmed_client(&cr_ses->clientid);
1300 conf = find_confirmed_client(&cr_ses->clientid);
1303 cs_slot = &conf->cl_cs_slot;
1304 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1305 if (status == nfserr_replay_cache) {
1306 dprintk("Got a create_session replay! seqid= %d\n",
1308 /* Return the cached reply status */
1309 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1311 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1312 status = nfserr_seq_misordered;
1313 dprintk("Sequence misordered!\n");
1314 dprintk("Expected seqid= %d but got seqid= %d\n",
1315 cs_slot->sl_seqid, cr_ses->seqid);
1318 cs_slot->sl_seqid++;
1319 } else if (unconf) {
1320 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1321 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1322 status = nfserr_clid_inuse;
1326 cs_slot = &unconf->cl_cs_slot;
1327 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1329 /* an unconfirmed replay returns misordered */
1330 status = nfserr_seq_misordered;
1334 cs_slot->sl_seqid++; /* from 0 to 1 */
1335 move_to_confirmed(unconf);
1338 * We do not support RDMA or persistent sessions
1340 cr_ses->flags &= ~SESSION4_PERSIST;
1341 cr_ses->flags &= ~SESSION4_RDMA;
1343 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1344 unconf->cl_cb_xprt = rqstp->rq_xprt;
1345 svc_xprt_get(unconf->cl_cb_xprt);
1347 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1349 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1350 unconf->cl_cb_conn.cb_minorversion =
1351 cstate->minorversion;
1352 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1353 unconf->cl_cb_seq_nr = 1;
1354 nfsd4_probe_callback(unconf);
1358 status = nfserr_stale_clientid;
1362 status = alloc_init_session(rqstp, conf, cr_ses);
1366 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1367 NFS4_MAX_SESSIONID_LEN);
1368 cr_ses->seqid = cs_slot->sl_seqid;
1371 /* cache solo and embedded create sessions under the state lock */
1372 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1374 nfs4_unlock_state();
1375 dprintk("%s returns %d\n", __func__, ntohl(status));
1380 nfsd4_destroy_session(struct svc_rqst *r,
1381 struct nfsd4_compound_state *cstate,
1382 struct nfsd4_destroy_session *sessionid)
1384 struct nfsd4_session *ses;
1385 u32 status = nfserr_badsession;
1388 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1389 * - Should we return nfserr_back_chan_busy if waiting for
1390 * callbacks on to-be-destroyed session?
1391 * - Do we need to clear any callback info from previous session?
1394 dump_sessionid(__func__, &sessionid->sessionid);
1395 spin_lock(&sessionid_lock);
1396 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1398 spin_unlock(&sessionid_lock);
1402 unhash_session(ses);
1403 spin_unlock(&sessionid_lock);
1405 /* wait for callbacks */
1406 shutdown_callback_client(ses->se_client);
1407 nfsd4_put_session(ses);
1410 dprintk("%s returns %d\n", __func__, ntohl(status));
1415 nfsd4_sequence(struct svc_rqst *rqstp,
1416 struct nfsd4_compound_state *cstate,
1417 struct nfsd4_sequence *seq)
1419 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1420 struct nfsd4_session *session;
1421 struct nfsd4_slot *slot;
1424 if (resp->opcnt != 1)
1425 return nfserr_sequence_pos;
1427 spin_lock(&sessionid_lock);
1428 status = nfserr_badsession;
1429 session = find_in_sessionid_hashtbl(&seq->sessionid);
1433 status = nfserr_badslot;
1434 if (seq->slotid >= session->se_fchannel.maxreqs)
1437 slot = session->se_slots[seq->slotid];
1438 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1440 /* We do not negotiate the number of slots yet, so set the
1441 * maxslots to the session maxreqs which is used to encode
1442 * sr_highest_slotid and the sr_target_slot id to maxslots */
1443 seq->maxslots = session->se_fchannel.maxreqs;
1445 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1446 if (status == nfserr_replay_cache) {
1447 cstate->slot = slot;
1448 cstate->session = session;
1449 /* Return the cached reply status and set cstate->status
1450 * for nfsd4_proc_compound processing */
1451 status = nfsd4_replay_cache_entry(resp, seq);
1452 cstate->status = nfserr_replay_cache;
1458 /* Success! bump slot seqid */
1459 slot->sl_inuse = true;
1460 slot->sl_seqid = seq->seqid;
1461 slot->sl_cachethis = seq->cachethis;
1463 cstate->slot = slot;
1464 cstate->session = session;
1466 /* Hold a session reference until done processing the compound:
1467 * nfsd4_put_session called only if the cstate slot is set.
1469 nfsd4_get_session(session);
1471 spin_unlock(&sessionid_lock);
1472 /* Renew the clientid on success and on replay */
1473 if (cstate->session) {
1475 renew_client(session->se_client);
1476 nfs4_unlock_state();
1478 dprintk("%s: return %d\n", __func__, ntohl(status));
1483 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1484 struct nfsd4_setclientid *setclid)
1486 struct sockaddr *sa = svc_addr(rqstp);
1487 struct xdr_netobj clname = {
1488 .len = setclid->se_namelen,
1489 .data = setclid->se_name,
1491 nfs4_verifier clverifier = setclid->se_verf;
1492 unsigned int strhashval;
1493 struct nfs4_client *conf, *unconf, *new;
1495 char dname[HEXDIR_LEN];
1497 if (!check_name(clname))
1498 return nfserr_inval;
1500 status = nfs4_make_rec_clidname(dname, &clname);
1505 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1506 * We get here on a DRC miss.
1509 strhashval = clientstr_hashval(dname);
1512 conf = find_confirmed_client_by_str(dname, strhashval, false);
1514 /* RFC 3530 14.2.33 CASE 0: */
1515 status = nfserr_clid_inuse;
1516 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1517 char addr_str[INET6_ADDRSTRLEN];
1518 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1520 dprintk("NFSD: setclientid: string in use by client "
1521 "at %s\n", addr_str);
1526 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1527 * has a description of SETCLIENTID request processing consisting
1528 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1530 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1531 status = nfserr_resource;
1534 * RFC 3530 14.2.33 CASE 4:
1535 * placed first, because it is the normal case
1538 expire_client(unconf);
1539 new = create_client(clname, dname, rqstp, &clverifier);
1543 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1545 * RFC 3530 14.2.33 CASE 1:
1546 * probable callback update
1549 /* Note this is removing unconfirmed {*x***},
1550 * which is stronger than RFC recommended {vxc**}.
1551 * This has the advantage that there is at most
1552 * one {*x***} in either list at any time.
1554 expire_client(unconf);
1556 new = create_client(clname, dname, rqstp, &clverifier);
1559 copy_clid(new, conf);
1560 } else if (!unconf) {
1562 * RFC 3530 14.2.33 CASE 2:
1563 * probable client reboot; state will be removed if
1566 new = create_client(clname, dname, rqstp, &clverifier);
1572 * RFC 3530 14.2.33 CASE 3:
1573 * probable client reboot; state will be removed if
1576 expire_client(unconf);
1577 new = create_client(clname, dname, rqstp, &clverifier);
1582 gen_callback(new, setclid, rpc_get_scope_id(sa));
1583 add_to_unconfirmed(new, strhashval);
1584 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1585 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1586 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1589 nfs4_unlock_state();
1595 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1596 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1597 * bullets, labeled as CASE1 - CASE4 below.
1600 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1601 struct nfsd4_compound_state *cstate,
1602 struct nfsd4_setclientid_confirm *setclientid_confirm)
1604 struct sockaddr *sa = svc_addr(rqstp);
1605 struct nfs4_client *conf, *unconf;
1606 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1607 clientid_t * clid = &setclientid_confirm->sc_clientid;
1610 if (STALE_CLIENTID(clid))
1611 return nfserr_stale_clientid;
1613 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1614 * We get here on a DRC miss.
1619 conf = find_confirmed_client(clid);
1620 unconf = find_unconfirmed_client(clid);
1622 status = nfserr_clid_inuse;
1623 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1625 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1629 * section 14.2.34 of RFC 3530 has a description of
1630 * SETCLIENTID_CONFIRM request processing consisting
1631 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1633 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1635 * RFC 3530 14.2.34 CASE 1:
1638 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1639 status = nfserr_clid_inuse;
1641 /* XXX: We just turn off callbacks until we can handle
1642 * change request correctly. */
1643 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1644 expire_client(unconf);
1648 } else if (conf && !unconf) {
1650 * RFC 3530 14.2.34 CASE 2:
1651 * probable retransmitted request; play it safe and
1654 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1655 status = nfserr_clid_inuse;
1658 } else if (!conf && unconf
1659 && same_verf(&unconf->cl_confirm, &confirm)) {
1661 * RFC 3530 14.2.34 CASE 3:
1662 * Normal case; new or rebooted client:
1664 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1665 status = nfserr_clid_inuse;
1668 clientstr_hashval(unconf->cl_recdir);
1669 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1672 nfsd4_remove_clid_dir(conf);
1673 expire_client(conf);
1675 move_to_confirmed(unconf);
1677 nfsd4_probe_callback(conf);
1680 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1681 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1684 * RFC 3530 14.2.34 CASE 4:
1685 * Client probably hasn't noticed that we rebooted yet.
1687 status = nfserr_stale_clientid;
1689 /* check that we have hit one of the cases...*/
1690 status = nfserr_clid_inuse;
1693 nfs4_unlock_state();
1697 /* OPEN Share state helper functions */
1698 static inline struct nfs4_file *
1699 alloc_init_file(struct inode *ino)
1701 struct nfs4_file *fp;
1702 unsigned int hashval = file_hashval(ino);
1704 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1706 atomic_set(&fp->fi_ref, 1);
1707 INIT_LIST_HEAD(&fp->fi_hash);
1708 INIT_LIST_HEAD(&fp->fi_stateids);
1709 INIT_LIST_HEAD(&fp->fi_delegations);
1710 spin_lock(&recall_lock);
1711 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1712 spin_unlock(&recall_lock);
1713 fp->fi_inode = igrab(ino);
1714 fp->fi_id = current_fileid++;
1715 fp->fi_had_conflict = false;
1722 nfsd4_free_slab(struct kmem_cache **slab)
1726 kmem_cache_destroy(*slab);
1731 nfsd4_free_slabs(void)
1733 nfsd4_free_slab(&stateowner_slab);
1734 nfsd4_free_slab(&file_slab);
1735 nfsd4_free_slab(&stateid_slab);
1736 nfsd4_free_slab(&deleg_slab);
1740 nfsd4_init_slabs(void)
1742 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1743 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1744 if (stateowner_slab == NULL)
1746 file_slab = kmem_cache_create("nfsd4_files",
1747 sizeof(struct nfs4_file), 0, 0, NULL);
1748 if (file_slab == NULL)
1750 stateid_slab = kmem_cache_create("nfsd4_stateids",
1751 sizeof(struct nfs4_stateid), 0, 0, NULL);
1752 if (stateid_slab == NULL)
1754 deleg_slab = kmem_cache_create("nfsd4_delegations",
1755 sizeof(struct nfs4_delegation), 0, 0, NULL);
1756 if (deleg_slab == NULL)
1761 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1766 nfs4_free_stateowner(struct kref *kref)
1768 struct nfs4_stateowner *sop =
1769 container_of(kref, struct nfs4_stateowner, so_ref);
1770 kfree(sop->so_owner.data);
1771 kmem_cache_free(stateowner_slab, sop);
1774 static inline struct nfs4_stateowner *
1775 alloc_stateowner(struct xdr_netobj *owner)
1777 struct nfs4_stateowner *sop;
1779 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1780 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1781 memcpy(sop->so_owner.data, owner->data, owner->len);
1782 sop->so_owner.len = owner->len;
1783 kref_init(&sop->so_ref);
1786 kmem_cache_free(stateowner_slab, sop);
1791 static struct nfs4_stateowner *
1792 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1793 struct nfs4_stateowner *sop;
1794 struct nfs4_replay *rp;
1795 unsigned int idhashval;
1797 if (!(sop = alloc_stateowner(&open->op_owner)))
1799 idhashval = ownerid_hashval(current_ownerid);
1800 INIT_LIST_HEAD(&sop->so_idhash);
1801 INIT_LIST_HEAD(&sop->so_strhash);
1802 INIT_LIST_HEAD(&sop->so_perclient);
1803 INIT_LIST_HEAD(&sop->so_stateids);
1804 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1805 INIT_LIST_HEAD(&sop->so_close_lru);
1807 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1808 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1809 list_add(&sop->so_perclient, &clp->cl_openowners);
1810 sop->so_is_open_owner = 1;
1811 sop->so_id = current_ownerid++;
1812 sop->so_client = clp;
1813 sop->so_seqid = open->op_seqid;
1814 sop->so_confirmed = 0;
1815 rp = &sop->so_replay;
1816 rp->rp_status = nfserr_serverfault;
1818 rp->rp_buf = rp->rp_ibuf;
1823 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1824 struct nfs4_stateowner *sop = open->op_stateowner;
1825 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1827 INIT_LIST_HEAD(&stp->st_hash);
1828 INIT_LIST_HEAD(&stp->st_perstateowner);
1829 INIT_LIST_HEAD(&stp->st_lockowners);
1830 INIT_LIST_HEAD(&stp->st_perfile);
1831 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1832 list_add(&stp->st_perstateowner, &sop->so_stateids);
1833 list_add(&stp->st_perfile, &fp->fi_stateids);
1834 stp->st_stateowner = sop;
1837 stp->st_stateid.si_boot = get_seconds();
1838 stp->st_stateid.si_stateownerid = sop->so_id;
1839 stp->st_stateid.si_fileid = fp->fi_id;
1840 stp->st_stateid.si_generation = 0;
1841 stp->st_access_bmap = 0;
1842 stp->st_deny_bmap = 0;
1843 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1844 &stp->st_access_bmap);
1845 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1846 stp->st_openstp = NULL;
1850 move_to_close_lru(struct nfs4_stateowner *sop)
1852 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1854 list_move_tail(&sop->so_close_lru, &close_lru);
1855 sop->so_time = get_seconds();
1859 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1862 return (sop->so_owner.len == owner->len) &&
1863 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1864 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1867 static struct nfs4_stateowner *
1868 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1870 struct nfs4_stateowner *so = NULL;
1872 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1873 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1879 /* search file_hashtbl[] for file */
1880 static struct nfs4_file *
1881 find_file(struct inode *ino)
1883 unsigned int hashval = file_hashval(ino);
1884 struct nfs4_file *fp;
1886 spin_lock(&recall_lock);
1887 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1888 if (fp->fi_inode == ino) {
1890 spin_unlock(&recall_lock);
1894 spin_unlock(&recall_lock);
1898 static inline int access_valid(u32 x, u32 minorversion)
1900 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1902 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1904 x &= ~NFS4_SHARE_ACCESS_MASK;
1905 if (minorversion && x) {
1906 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1908 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1910 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1917 static inline int deny_valid(u32 x)
1919 /* Note: unlike access bits, deny bits may be zero. */
1920 return x <= NFS4_SHARE_DENY_BOTH;
1924 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1925 * st_{access,deny}_bmap field of the stateid, in order to track not
1926 * only what share bits are currently in force, but also what
1927 * combinations of share bits previous opens have used. This allows us
1928 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1929 * return an error if the client attempt to downgrade to a combination
1930 * of share bits not explicable by closing some of its previous opens.
1932 * XXX: This enforcement is actually incomplete, since we don't keep
1933 * track of access/deny bit combinations; so, e.g., we allow:
1935 * OPEN allow read, deny write
1936 * OPEN allow both, deny none
1937 * DOWNGRADE allow read, deny none
1939 * which we should reject.
1942 set_access(unsigned int *access, unsigned long bmap) {
1946 for (i = 1; i < 4; i++) {
1947 if (test_bit(i, &bmap))
1953 set_deny(unsigned int *deny, unsigned long bmap) {
1957 for (i = 0; i < 4; i++) {
1958 if (test_bit(i, &bmap))
1964 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1965 unsigned int access, deny;
1967 set_access(&access, stp->st_access_bmap);
1968 set_deny(&deny, stp->st_deny_bmap);
1969 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1975 * Called to check deny when READ with all zero stateid or
1976 * WRITE with all zero or all one stateid
1979 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1981 struct inode *ino = current_fh->fh_dentry->d_inode;
1982 struct nfs4_file *fp;
1983 struct nfs4_stateid *stp;
1986 dprintk("NFSD: nfs4_share_conflict\n");
1988 fp = find_file(ino);
1991 ret = nfserr_locked;
1992 /* Search for conflicting share reservations */
1993 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1994 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1995 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2005 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2007 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2008 drop_file_write_access(filp);
2009 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2014 * Spawn a thread to perform a recall on the delegation represented
2015 * by the lease (file_lock)
2017 * Called from break_lease() with lock_kernel() held.
2018 * Note: we assume break_lease will only call this *once* for any given
2022 void nfsd_break_deleg_cb(struct file_lock *fl)
2024 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2026 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2030 /* We're assuming the state code never drops its reference
2031 * without first removing the lease. Since we're in this lease
2032 * callback (and since the lease code is serialized by the kernel
2033 * lock) we know the server hasn't removed the lease yet, we know
2034 * it's safe to take a reference: */
2035 atomic_inc(&dp->dl_count);
2036 atomic_inc(&dp->dl_client->cl_count);
2038 spin_lock(&recall_lock);
2039 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2040 spin_unlock(&recall_lock);
2042 /* only place dl_time is set. protected by lock_kernel*/
2043 dp->dl_time = get_seconds();
2046 * We don't want the locks code to timeout the lease for us;
2047 * we'll remove it ourself if the delegation isn't returned
2050 fl->fl_break_time = 0;
2052 dp->dl_file->fi_had_conflict = true;
2053 nfsd4_cb_recall(dp);
2057 * The file_lock is being reapd.
2059 * Called by locks_free_lock() with lock_kernel() held.
2062 void nfsd_release_deleg_cb(struct file_lock *fl)
2064 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2066 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2068 if (!(fl->fl_flags & FL_LEASE) || !dp)
2070 dp->dl_flock = NULL;
2074 * Set the delegation file_lock back pointer.
2076 * Called from setlease() with lock_kernel() held.
2079 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2081 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2083 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2090 * Called from setlease() with lock_kernel() held
2093 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2095 struct nfs4_delegation *onlistd =
2096 (struct nfs4_delegation *)onlist->fl_owner;
2097 struct nfs4_delegation *tryd =
2098 (struct nfs4_delegation *)try->fl_owner;
2100 if (onlist->fl_lmops != try->fl_lmops)
2103 return onlistd->dl_client == tryd->dl_client;
2108 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2111 return lease_modify(onlist, arg);
2116 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2117 .fl_break = nfsd_break_deleg_cb,
2118 .fl_release_private = nfsd_release_deleg_cb,
2119 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2120 .fl_mylease = nfsd_same_client_deleg_cb,
2121 .fl_change = nfsd_change_deleg_cb,
2126 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2127 struct nfsd4_open *open)
2129 clientid_t *clientid = &open->op_clientid;
2130 struct nfs4_client *clp = NULL;
2131 unsigned int strhashval;
2132 struct nfs4_stateowner *sop = NULL;
2134 if (!check_name(open->op_owner))
2135 return nfserr_inval;
2137 if (STALE_CLIENTID(&open->op_clientid))
2138 return nfserr_stale_clientid;
2140 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2141 sop = find_openstateowner_str(strhashval, open);
2142 open->op_stateowner = sop;
2144 /* Make sure the client's lease hasn't expired. */
2145 clp = find_confirmed_client(clientid);
2147 return nfserr_expired;
2150 /* When sessions are used, skip open sequenceid processing */
2151 if (nfsd4_has_session(cstate))
2153 if (!sop->so_confirmed) {
2154 /* Replace unconfirmed owners without checking for replay. */
2155 clp = sop->so_client;
2156 release_openowner(sop);
2157 open->op_stateowner = NULL;
2160 if (open->op_seqid == sop->so_seqid - 1) {
2161 if (sop->so_replay.rp_buflen)
2162 return nfserr_replay_me;
2163 /* The original OPEN failed so spectacularly
2164 * that we don't even have replay data saved!
2165 * Therefore, we have no choice but to continue
2166 * processing this OPEN; presumably, we'll
2167 * fail again for the same reason.
2169 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2172 if (open->op_seqid != sop->so_seqid)
2173 return nfserr_bad_seqid;
2175 if (open->op_stateowner == NULL) {
2176 sop = alloc_init_open_stateowner(strhashval, clp, open);
2178 return nfserr_resource;
2179 open->op_stateowner = sop;
2181 list_del_init(&sop->so_close_lru);
2182 renew_client(sop->so_client);
2186 static inline __be32
2187 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2189 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2190 return nfserr_openmode;
2195 static struct nfs4_delegation *
2196 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2198 struct nfs4_delegation *dp;
2200 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2201 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2208 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2209 struct nfs4_delegation **dp)
2212 __be32 status = nfserr_bad_stateid;
2214 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2217 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2218 RD_STATE : WR_STATE;
2219 status = nfs4_check_delegmode(*dp, flags);
2223 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2227 open->op_stateowner->so_confirmed = 1;
2232 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2234 struct nfs4_stateid *local;
2235 __be32 status = nfserr_share_denied;
2236 struct nfs4_stateowner *sop = open->op_stateowner;
2238 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2239 /* ignore lock owners */
2240 if (local->st_stateowner->so_is_open_owner == 0)
2242 /* remember if we have seen this open owner */
2243 if (local->st_stateowner == sop)
2245 /* check for conflicting share reservations */
2246 if (!test_share(local, open))
2254 static inline struct nfs4_stateid *
2255 nfs4_alloc_stateid(void)
2257 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2261 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2262 struct nfs4_delegation *dp,
2263 struct svc_fh *cur_fh, int flags)
2265 struct nfs4_stateid *stp;
2267 stp = nfs4_alloc_stateid();
2269 return nfserr_resource;
2272 get_file(dp->dl_vfs_file);
2273 stp->st_vfs_file = dp->dl_vfs_file;
2276 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2279 if (status == nfserr_dropit)
2280 status = nfserr_jukebox;
2281 kmem_cache_free(stateid_slab, stp);
2289 static inline __be32
2290 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2291 struct nfsd4_open *open)
2293 struct iattr iattr = {
2294 .ia_valid = ATTR_SIZE,
2297 if (!open->op_truncate)
2299 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2300 return nfserr_inval;
2301 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2305 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2307 struct file *filp = stp->st_vfs_file;
2308 struct inode *inode = filp->f_path.dentry->d_inode;
2309 unsigned int share_access, new_writer;
2312 set_access(&share_access, stp->st_access_bmap);
2313 new_writer = (~share_access) & open->op_share_access
2314 & NFS4_SHARE_ACCESS_WRITE;
2317 int err = get_write_access(inode);
2319 return nfserrno(err);
2320 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2322 return nfserrno(err);
2323 file_take_write(filp);
2325 status = nfsd4_truncate(rqstp, cur_fh, open);
2328 put_write_access(inode);
2331 /* remember the open */
2332 filp->f_mode |= open->op_share_access;
2333 __set_bit(open->op_share_access, &stp->st_access_bmap);
2334 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2341 nfs4_set_claim_prev(struct nfsd4_open *open)
2343 open->op_stateowner->so_confirmed = 1;
2344 open->op_stateowner->so_client->cl_firststate = 1;
2348 * Attempt to hand out a delegation.
2351 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2353 struct nfs4_delegation *dp;
2354 struct nfs4_stateowner *sop = stp->st_stateowner;
2355 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2356 struct file_lock fl, *flp = &fl;
2357 int status, flag = 0;
2359 flag = NFS4_OPEN_DELEGATE_NONE;
2360 open->op_recall = 0;
2361 switch (open->op_claim_type) {
2362 case NFS4_OPEN_CLAIM_PREVIOUS:
2363 if (!atomic_read(&cb->cb_set))
2364 open->op_recall = 1;
2365 flag = open->op_delegate_type;
2366 if (flag == NFS4_OPEN_DELEGATE_NONE)
2369 case NFS4_OPEN_CLAIM_NULL:
2370 /* Let's not give out any delegations till everyone's
2371 * had the chance to reclaim theirs.... */
2372 if (locks_in_grace())
2374 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2376 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2377 flag = NFS4_OPEN_DELEGATE_WRITE;
2379 flag = NFS4_OPEN_DELEGATE_READ;
2385 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2387 flag = NFS4_OPEN_DELEGATE_NONE;
2390 locks_init_lock(&fl);
2391 fl.fl_lmops = &nfsd_lease_mng_ops;
2392 fl.fl_flags = FL_LEASE;
2393 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2394 fl.fl_end = OFFSET_MAX;
2395 fl.fl_owner = (fl_owner_t)dp;
2396 fl.fl_file = stp->st_vfs_file;
2397 fl.fl_pid = current->tgid;
2399 /* vfs_setlease checks to see if delegation should be handed out.
2400 * the lock_manager callbacks fl_mylease and fl_change are used
2402 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2403 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2404 unhash_delegation(dp);
2405 flag = NFS4_OPEN_DELEGATE_NONE;
2409 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2411 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2412 dp->dl_stateid.si_boot,
2413 dp->dl_stateid.si_stateownerid,
2414 dp->dl_stateid.si_fileid,
2415 dp->dl_stateid.si_generation);
2417 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2418 && flag == NFS4_OPEN_DELEGATE_NONE
2419 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2420 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2421 open->op_delegate_type = flag;
2425 * called with nfs4_lock_state() held.
2428 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2430 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2431 struct nfs4_file *fp = NULL;
2432 struct inode *ino = current_fh->fh_dentry->d_inode;
2433 struct nfs4_stateid *stp = NULL;
2434 struct nfs4_delegation *dp = NULL;
2437 status = nfserr_inval;
2438 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2439 || !deny_valid(open->op_share_deny))
2442 * Lookup file; if found, lookup stateid and check open request,
2443 * and check for delegations in the process of being recalled.
2444 * If not found, create the nfs4_file struct
2446 fp = find_file(ino);
2448 if ((status = nfs4_check_open(fp, open, &stp)))
2450 status = nfs4_check_deleg(fp, open, &dp);
2454 status = nfserr_bad_stateid;
2455 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2457 status = nfserr_resource;
2458 fp = alloc_init_file(ino);
2464 * OPEN the file, or upgrade an existing OPEN.
2465 * If truncate fails, the OPEN fails.
2468 /* Stateid was found, this is an OPEN upgrade */
2469 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2472 update_stateid(&stp->st_stateid);
2474 /* Stateid was not found, this is a new OPEN */
2476 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2477 flags |= NFSD_MAY_READ;
2478 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2479 flags |= NFSD_MAY_WRITE;
2480 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2483 init_stateid(stp, fp, open);
2484 status = nfsd4_truncate(rqstp, current_fh, open);
2486 release_open_stateid(stp);
2489 if (nfsd4_has_session(&resp->cstate))
2490 update_stateid(&stp->st_stateid);
2492 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2494 if (nfsd4_has_session(&resp->cstate))
2495 open->op_stateowner->so_confirmed = 1;
2498 * Attempt to hand out a delegation. No error return, because the
2499 * OPEN succeeds even if we fail.
2501 nfs4_open_delegation(current_fh, open, stp);
2505 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2506 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2507 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2511 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2512 nfs4_set_claim_prev(open);
2514 * To finish the open response, we just need to set the rflags.
2516 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2517 if (!open->op_stateowner->so_confirmed &&
2518 !nfsd4_has_session(&resp->cstate))
2519 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2525 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2528 struct nfs4_client *clp;
2532 dprintk("process_renew(%08x/%08x): starting\n",
2533 clid->cl_boot, clid->cl_id);
2534 status = nfserr_stale_clientid;
2535 if (STALE_CLIENTID(clid))
2537 clp = find_confirmed_client(clid);
2538 status = nfserr_expired;
2540 /* We assume the client took too long to RENEW. */
2541 dprintk("nfsd4_renew: clientid not found!\n");
2545 status = nfserr_cb_path_down;
2546 if (!list_empty(&clp->cl_delegations)
2547 && !atomic_read(&clp->cl_cb_conn.cb_set))
2551 nfs4_unlock_state();
2555 struct lock_manager nfsd4_manager = {
2559 nfsd4_end_grace(void)
2561 dprintk("NFSD: end of grace period\n");
2562 nfsd4_recdir_purge_old();
2563 locks_end_grace(&nfsd4_manager);
2567 nfs4_laundromat(void)
2569 struct nfs4_client *clp;
2570 struct nfs4_stateowner *sop;
2571 struct nfs4_delegation *dp;
2572 struct list_head *pos, *next, reaplist;
2573 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2574 time_t t, clientid_val = NFSD_LEASE_TIME;
2575 time_t u, test_val = NFSD_LEASE_TIME;
2579 dprintk("NFSD: laundromat service - starting\n");
2580 if (locks_in_grace())
2582 list_for_each_safe(pos, next, &client_lru) {
2583 clp = list_entry(pos, struct nfs4_client, cl_lru);
2584 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2585 t = clp->cl_time - cutoff;
2586 if (clientid_val > t)
2590 dprintk("NFSD: purging unused client (clientid %08x)\n",
2591 clp->cl_clientid.cl_id);
2592 nfsd4_remove_clid_dir(clp);
2595 INIT_LIST_HEAD(&reaplist);
2596 spin_lock(&recall_lock);
2597 list_for_each_safe(pos, next, &del_recall_lru) {
2598 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2599 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2600 u = dp->dl_time - cutoff;
2605 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2607 list_move(&dp->dl_recall_lru, &reaplist);
2609 spin_unlock(&recall_lock);
2610 list_for_each_safe(pos, next, &reaplist) {
2611 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2612 list_del_init(&dp->dl_recall_lru);
2613 unhash_delegation(dp);
2615 test_val = NFSD_LEASE_TIME;
2616 list_for_each_safe(pos, next, &close_lru) {
2617 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2618 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2619 u = sop->so_time - cutoff;
2624 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2626 release_openowner(sop);
2628 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2629 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2630 nfs4_unlock_state();
2631 return clientid_val;
2634 static struct workqueue_struct *laundry_wq;
2635 static void laundromat_main(struct work_struct *);
2636 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2639 laundromat_main(struct work_struct *not_used)
2643 t = nfs4_laundromat();
2644 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2645 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2648 static struct nfs4_stateowner *
2649 search_close_lru(u32 st_id, int flags)
2651 struct nfs4_stateowner *local = NULL;
2653 if (flags & CLOSE_STATE) {
2654 list_for_each_entry(local, &close_lru, so_close_lru) {
2655 if (local->so_id == st_id)
2663 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2665 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2669 STALE_STATEID(stateid_t *stateid)
2671 if (time_after((unsigned long)boot_time,
2672 (unsigned long)stateid->si_boot)) {
2673 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2674 stateid->si_boot, stateid->si_stateownerid,
2675 stateid->si_fileid, stateid->si_generation);
2682 EXPIRED_STATEID(stateid_t *stateid)
2684 if (time_before((unsigned long)boot_time,
2685 ((unsigned long)stateid->si_boot)) &&
2686 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2687 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2688 stateid->si_boot, stateid->si_stateownerid,
2689 stateid->si_fileid, stateid->si_generation);
2696 stateid_error_map(stateid_t *stateid)
2698 if (STALE_STATEID(stateid))
2699 return nfserr_stale_stateid;
2700 if (EXPIRED_STATEID(stateid))
2701 return nfserr_expired;
2703 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2704 stateid->si_boot, stateid->si_stateownerid,
2705 stateid->si_fileid, stateid->si_generation);
2706 return nfserr_bad_stateid;
2710 access_permit_read(unsigned long access_bmap)
2712 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2713 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2714 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2718 access_permit_write(unsigned long access_bmap)
2720 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2721 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2725 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2727 __be32 status = nfserr_openmode;
2729 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2731 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2738 static inline __be32
2739 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2741 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2743 else if (locks_in_grace()) {
2744 /* Answer in remaining cases depends on existance of
2745 * conflicting state; so we must wait out the grace period. */
2746 return nfserr_grace;
2747 } else if (flags & WR_STATE)
2748 return nfs4_share_conflict(current_fh,
2749 NFS4_SHARE_DENY_WRITE);
2750 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2751 return nfs4_share_conflict(current_fh,
2752 NFS4_SHARE_DENY_READ);
2756 * Allow READ/WRITE during grace period on recovered state only for files
2757 * that are not able to provide mandatory locking.
2760 grace_disallows_io(struct inode *inode)
2762 return locks_in_grace() && mandatory_lock(inode);
2765 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2768 * When sessions are used the stateid generation number is ignored
2771 if ((flags & HAS_SESSION) && in->si_generation == 0)
2774 /* If the client sends us a stateid from the future, it's buggy: */
2775 if (in->si_generation > ref->si_generation)
2776 return nfserr_bad_stateid;
2778 * The following, however, can happen. For example, if the
2779 * client sends an open and some IO at the same time, the open
2780 * may bump si_generation while the IO is still in flight.
2781 * Thanks to hard links and renames, the client never knows what
2782 * file an open will affect. So it could avoid that situation
2783 * only by serializing all opens and IO from the same open
2784 * owner. To recover from the old_stateid error, the client
2785 * will just have to retry the IO:
2787 if (in->si_generation < ref->si_generation)
2788 return nfserr_old_stateid;
2793 static int is_delegation_stateid(stateid_t *stateid)
2795 return stateid->si_fileid == 0;
2799 * Checks for stateid operations
2802 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2803 stateid_t *stateid, int flags, struct file **filpp)
2805 struct nfs4_stateid *stp = NULL;
2806 struct nfs4_delegation *dp = NULL;
2807 struct svc_fh *current_fh = &cstate->current_fh;
2808 struct inode *ino = current_fh->fh_dentry->d_inode;
2814 if (grace_disallows_io(ino))
2815 return nfserr_grace;
2817 if (nfsd4_has_session(cstate))
2818 flags |= HAS_SESSION;
2820 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2821 return check_special_stateids(current_fh, stateid, flags);
2823 status = nfserr_stale_stateid;
2824 if (STALE_STATEID(stateid))
2827 status = nfserr_bad_stateid;
2828 if (is_delegation_stateid(stateid)) {
2829 dp = find_delegation_stateid(ino, stateid);
2831 status = stateid_error_map(stateid);
2834 status = check_stateid_generation(stateid, &dp->dl_stateid,
2838 status = nfs4_check_delegmode(dp, flags);
2841 renew_client(dp->dl_client);
2843 *filpp = dp->dl_vfs_file;
2844 } else { /* open or lock stateid */
2845 stp = find_stateid(stateid, flags);
2847 status = stateid_error_map(stateid);
2850 if (nfs4_check_fh(current_fh, stp))
2852 if (!stp->st_stateowner->so_confirmed)
2854 status = check_stateid_generation(stateid, &stp->st_stateid,
2858 status = nfs4_check_openmode(stp, flags);
2861 renew_client(stp->st_stateowner->so_client);
2863 *filpp = stp->st_vfs_file;
2873 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2874 RD_STATE : WR_STATE;
2878 * Checks for sequence id mutating operations.
2881 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2882 stateid_t *stateid, int flags,
2883 struct nfs4_stateowner **sopp,
2884 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2886 struct nfs4_stateid *stp;
2887 struct nfs4_stateowner *sop;
2888 struct svc_fh *current_fh = &cstate->current_fh;
2891 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2892 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2893 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2894 stateid->si_generation);
2899 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2900 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2901 return nfserr_bad_stateid;
2904 if (STALE_STATEID(stateid))
2905 return nfserr_stale_stateid;
2907 if (nfsd4_has_session(cstate))
2908 flags |= HAS_SESSION;
2911 * We return BAD_STATEID if filehandle doesn't match stateid,
2912 * the confirmed flag is incorrecly set, or the generation
2913 * number is incorrect.
2915 stp = find_stateid(stateid, flags);
2918 * Also, we should make sure this isn't just the result of
2921 sop = search_close_lru(stateid->si_stateownerid, flags);
2923 return stateid_error_map(stateid);
2929 *sopp = sop = stp->st_stateowner;
2932 clientid_t *lockclid = &lock->v.new.clientid;
2933 struct nfs4_client *clp = sop->so_client;
2937 lkflg = setlkflg(lock->lk_type);
2939 if (lock->lk_is_new) {
2940 if (!sop->so_is_open_owner)
2941 return nfserr_bad_stateid;
2942 if (!(flags & HAS_SESSION) &&
2943 !same_clid(&clp->cl_clientid, lockclid))
2944 return nfserr_bad_stateid;
2945 /* stp is the open stateid */
2946 status = nfs4_check_openmode(stp, lkflg);
2950 /* stp is the lock stateid */
2951 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2957 if (nfs4_check_fh(current_fh, stp)) {
2958 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2959 return nfserr_bad_stateid;
2963 * We now validate the seqid and stateid generation numbers.
2964 * For the moment, we ignore the possibility of
2965 * generation number wraparound.
2967 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2970 if (sop->so_confirmed && flags & CONFIRM) {
2971 dprintk("NFSD: preprocess_seqid_op: expected"
2972 " unconfirmed stateowner!\n");
2973 return nfserr_bad_stateid;
2975 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2976 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2977 " confirmed yet!\n");
2978 return nfserr_bad_stateid;
2980 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2983 renew_client(sop->so_client);
2987 if (seqid == sop->so_seqid - 1) {
2988 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2989 /* indicate replay to calling function */
2990 return nfserr_replay_me;
2992 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2993 sop->so_seqid, seqid);
2995 return nfserr_bad_seqid;
2999 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3000 struct nfsd4_open_confirm *oc)
3003 struct nfs4_stateowner *sop;
3004 struct nfs4_stateid *stp;
3006 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3007 (int)cstate->current_fh.fh_dentry->d_name.len,
3008 cstate->current_fh.fh_dentry->d_name.name);
3010 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3016 if ((status = nfs4_preprocess_seqid_op(cstate,
3017 oc->oc_seqid, &oc->oc_req_stateid,
3018 CONFIRM | OPEN_STATE,
3019 &oc->oc_stateowner, &stp, NULL)))
3022 sop = oc->oc_stateowner;
3023 sop->so_confirmed = 1;
3024 update_stateid(&stp->st_stateid);
3025 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3026 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3027 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3028 stp->st_stateid.si_boot,
3029 stp->st_stateid.si_stateownerid,
3030 stp->st_stateid.si_fileid,
3031 stp->st_stateid.si_generation);
3033 nfsd4_create_clid_dir(sop->so_client);
3035 if (oc->oc_stateowner) {
3036 nfs4_get_stateowner(oc->oc_stateowner);
3037 cstate->replay_owner = oc->oc_stateowner;
3039 nfs4_unlock_state();
3045 * unset all bits in union bitmap (bmap) that
3046 * do not exist in share (from successful OPEN_DOWNGRADE)
3049 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3052 for (i = 1; i < 4; i++) {
3053 if ((i & access) != i)
3054 __clear_bit(i, bmap);
3059 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3062 for (i = 0; i < 4; i++) {
3063 if ((i & deny) != i)
3064 __clear_bit(i, bmap);
3069 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3070 struct nfsd4_compound_state *cstate,
3071 struct nfsd4_open_downgrade *od)
3074 struct nfs4_stateid *stp;
3075 unsigned int share_access;
3077 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3078 (int)cstate->current_fh.fh_dentry->d_name.len,
3079 cstate->current_fh.fh_dentry->d_name.name);
3081 if (!access_valid(od->od_share_access, cstate->minorversion)
3082 || !deny_valid(od->od_share_deny))
3083 return nfserr_inval;
3086 if ((status = nfs4_preprocess_seqid_op(cstate,
3090 &od->od_stateowner, &stp, NULL)))
3093 status = nfserr_inval;
3094 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3095 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3096 stp->st_access_bmap, od->od_share_access);
3099 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3100 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3101 stp->st_deny_bmap, od->od_share_deny);
3104 set_access(&share_access, stp->st_access_bmap);
3105 nfs4_file_downgrade(stp->st_vfs_file,
3106 share_access & ~od->od_share_access);
3108 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3109 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3111 update_stateid(&stp->st_stateid);
3112 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3115 if (od->od_stateowner) {
3116 nfs4_get_stateowner(od->od_stateowner);
3117 cstate->replay_owner = od->od_stateowner;
3119 nfs4_unlock_state();
3124 * nfs4_unlock_state() called after encode
3127 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3128 struct nfsd4_close *close)
3131 struct nfs4_stateid *stp;
3133 dprintk("NFSD: nfsd4_close on file %.*s\n",
3134 (int)cstate->current_fh.fh_dentry->d_name.len,
3135 cstate->current_fh.fh_dentry->d_name.name);
3138 /* check close_lru for replay */
3139 if ((status = nfs4_preprocess_seqid_op(cstate,
3142 OPEN_STATE | CLOSE_STATE,
3143 &close->cl_stateowner, &stp, NULL)))
3146 update_stateid(&stp->st_stateid);
3147 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3149 /* release_stateid() calls nfsd_close() if needed */
3150 release_open_stateid(stp);
3152 /* place unused nfs4_stateowners on so_close_lru list to be
3153 * released by the laundromat service after the lease period
3154 * to enable us to handle CLOSE replay
3156 if (list_empty(&close->cl_stateowner->so_stateids))
3157 move_to_close_lru(close->cl_stateowner);
3159 if (close->cl_stateowner) {
3160 nfs4_get_stateowner(close->cl_stateowner);
3161 cstate->replay_owner = close->cl_stateowner;
3163 nfs4_unlock_state();
3168 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3169 struct nfsd4_delegreturn *dr)
3171 struct nfs4_delegation *dp;
3172 stateid_t *stateid = &dr->dr_stateid;
3173 struct inode *inode;
3177 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3179 inode = cstate->current_fh.fh_dentry->d_inode;
3181 if (nfsd4_has_session(cstate))
3182 flags |= HAS_SESSION;
3184 status = nfserr_bad_stateid;
3185 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3187 status = nfserr_stale_stateid;
3188 if (STALE_STATEID(stateid))
3190 status = nfserr_bad_stateid;
3191 if (!is_delegation_stateid(stateid))
3193 dp = find_delegation_stateid(inode, stateid);
3195 status = stateid_error_map(stateid);
3198 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3201 renew_client(dp->dl_client);
3203 unhash_delegation(dp);
3205 nfs4_unlock_state();
3212 * Lock owner state (byte-range locks)
3214 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3215 #define LOCK_HASH_BITS 8
3216 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3217 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3220 end_offset(u64 start, u64 len)
3225 return end >= start ? end: NFS4_MAX_UINT64;
3228 /* last octet in a range */
3230 last_byte_offset(u64 start, u64 len)
3236 return end > start ? end - 1: NFS4_MAX_UINT64;
3239 #define lockownerid_hashval(id) \
3240 ((id) & LOCK_HASH_MASK)
3242 static inline unsigned int
3243 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3244 struct xdr_netobj *ownername)
3246 return (file_hashval(inode) + cl_id
3247 + opaque_hashval(ownername->data, ownername->len))
3251 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3252 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3253 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3255 static struct nfs4_stateid *
3256 find_stateid(stateid_t *stid, int flags)
3258 struct nfs4_stateid *local;
3259 u32 st_id = stid->si_stateownerid;
3260 u32 f_id = stid->si_fileid;
3261 unsigned int hashval;
3263 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3264 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3265 hashval = stateid_hashval(st_id, f_id);
3266 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3267 if ((local->st_stateid.si_stateownerid == st_id) &&
3268 (local->st_stateid.si_fileid == f_id))
3273 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3274 hashval = stateid_hashval(st_id, f_id);
3275 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3276 if ((local->st_stateid.si_stateownerid == st_id) &&
3277 (local->st_stateid.si_fileid == f_id))
3284 static struct nfs4_delegation *
3285 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3287 struct nfs4_file *fp;
3288 struct nfs4_delegation *dl;
3290 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3291 stid->si_boot, stid->si_stateownerid,
3292 stid->si_fileid, stid->si_generation);
3294 fp = find_file(ino);
3297 dl = find_delegation_file(fp, stid);
3303 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3304 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3305 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3306 * locking, this prevents us from being completely protocol-compliant. The
3307 * real solution to this problem is to start using unsigned file offsets in
3308 * the VFS, but this is a very deep change!
3311 nfs4_transform_lock_offset(struct file_lock *lock)
3313 if (lock->fl_start < 0)
3314 lock->fl_start = OFFSET_MAX;
3315 if (lock->fl_end < 0)
3316 lock->fl_end = OFFSET_MAX;
3319 /* Hack!: For now, we're defining this just so we can use a pointer to it
3320 * as a unique cookie to identify our (NFSv4's) posix locks. */
3321 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3325 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3327 struct nfs4_stateowner *sop;
3330 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3331 sop = (struct nfs4_stateowner *) fl->fl_owner;
3332 hval = lockownerid_hashval(sop->so_id);
3333 kref_get(&sop->so_ref);
3335 deny->ld_clientid = sop->so_client->cl_clientid;
3337 deny->ld_sop = NULL;
3338 deny->ld_clientid.cl_boot = 0;
3339 deny->ld_clientid.cl_id = 0;
3341 deny->ld_start = fl->fl_start;
3342 deny->ld_length = NFS4_MAX_UINT64;
3343 if (fl->fl_end != NFS4_MAX_UINT64)
3344 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3345 deny->ld_type = NFS4_READ_LT;
3346 if (fl->fl_type != F_RDLCK)
3347 deny->ld_type = NFS4_WRITE_LT;
3350 static struct nfs4_stateowner *
3351 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3352 struct xdr_netobj *owner)
3354 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3355 struct nfs4_stateowner *op;
3357 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3358 if (same_owner_str(op, owner, clid))
3365 * Alloc a lock owner structure.
3366 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3369 * strhashval = lock_ownerstr_hashval
3372 static struct nfs4_stateowner *
3373 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3374 struct nfs4_stateowner *sop;
3375 struct nfs4_replay *rp;
3376 unsigned int idhashval;
3378 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3380 idhashval = lockownerid_hashval(current_ownerid);
3381 INIT_LIST_HEAD(&sop->so_idhash);
3382 INIT_LIST_HEAD(&sop->so_strhash);
3383 INIT_LIST_HEAD(&sop->so_perclient);
3384 INIT_LIST_HEAD(&sop->so_stateids);
3385 INIT_LIST_HEAD(&sop->so_perstateid);
3386 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3388 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3389 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3390 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3391 sop->so_is_open_owner = 0;
3392 sop->so_id = current_ownerid++;
3393 sop->so_client = clp;
3394 /* It is the openowner seqid that will be incremented in encode in the
3395 * case of new lockowners; so increment the lock seqid manually: */
3396 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3397 sop->so_confirmed = 1;
3398 rp = &sop->so_replay;
3399 rp->rp_status = nfserr_serverfault;
3401 rp->rp_buf = rp->rp_ibuf;
3405 static struct nfs4_stateid *
3406 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3408 struct nfs4_stateid *stp;
3409 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3411 stp = nfs4_alloc_stateid();
3414 INIT_LIST_HEAD(&stp->st_hash);
3415 INIT_LIST_HEAD(&stp->st_perfile);
3416 INIT_LIST_HEAD(&stp->st_perstateowner);
3417 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3418 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3419 list_add(&stp->st_perfile, &fp->fi_stateids);
3420 list_add(&stp->st_perstateowner, &sop->so_stateids);
3421 stp->st_stateowner = sop;
3424 stp->st_stateid.si_boot = get_seconds();
3425 stp->st_stateid.si_stateownerid = sop->so_id;
3426 stp->st_stateid.si_fileid = fp->fi_id;
3427 stp->st_stateid.si_generation = 0;
3428 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3429 stp->st_access_bmap = open_stp->st_access_bmap;
3430 stp->st_deny_bmap = open_stp->st_deny_bmap;
3431 stp->st_openstp = open_stp;
3438 check_lock_length(u64 offset, u64 length)
3440 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3441 LOFF_OVERFLOW(offset, length)));
3448 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3449 struct nfsd4_lock *lock)
3451 struct nfs4_stateowner *open_sop = NULL;
3452 struct nfs4_stateowner *lock_sop = NULL;
3453 struct nfs4_stateid *lock_stp;
3455 struct file_lock file_lock;
3456 struct file_lock conflock;
3458 unsigned int strhashval;
3462 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3463 (long long) lock->lk_offset,
3464 (long long) lock->lk_length);
3466 if (check_lock_length(lock->lk_offset, lock->lk_length))
3467 return nfserr_inval;
3469 if ((status = fh_verify(rqstp, &cstate->current_fh,
3470 S_IFREG, NFSD_MAY_LOCK))) {
3471 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3477 if (lock->lk_is_new) {
3479 * Client indicates that this is a new lockowner.
3480 * Use open owner and open stateid to create lock owner and
3483 struct nfs4_stateid *open_stp = NULL;
3484 struct nfs4_file *fp;
3486 status = nfserr_stale_clientid;
3487 if (!nfsd4_has_session(cstate) &&
3488 STALE_CLIENTID(&lock->lk_new_clientid))
3491 /* validate and update open stateid and open seqid */
3492 status = nfs4_preprocess_seqid_op(cstate,
3493 lock->lk_new_open_seqid,
3494 &lock->lk_new_open_stateid,
3496 &lock->lk_replay_owner, &open_stp,
3500 open_sop = lock->lk_replay_owner;
3501 /* create lockowner and lock stateid */
3502 fp = open_stp->st_file;
3503 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3504 open_sop->so_client->cl_clientid.cl_id,
3505 &lock->v.new.owner);
3506 /* XXX: Do we need to check for duplicate stateowners on
3507 * the same file, or should they just be allowed (and
3508 * create new stateids)? */
3509 status = nfserr_resource;
3510 lock_sop = alloc_init_lock_stateowner(strhashval,
3511 open_sop->so_client, open_stp, lock);
3512 if (lock_sop == NULL)
3514 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3515 if (lock_stp == NULL)
3518 /* lock (lock owner + lock stateid) already exists */
3519 status = nfs4_preprocess_seqid_op(cstate,
3520 lock->lk_old_lock_seqid,
3521 &lock->lk_old_lock_stateid,
3523 &lock->lk_replay_owner, &lock_stp, lock);
3526 lock_sop = lock->lk_replay_owner;
3528 /* lock->lk_replay_owner and lock_stp have been created or found */
3529 filp = lock_stp->st_vfs_file;
3531 status = nfserr_grace;
3532 if (locks_in_grace() && !lock->lk_reclaim)
3534 status = nfserr_no_grace;
3535 if (!locks_in_grace() && lock->lk_reclaim)
3538 locks_init_lock(&file_lock);
3539 switch (lock->lk_type) {
3542 file_lock.fl_type = F_RDLCK;
3546 case NFS4_WRITEW_LT:
3547 file_lock.fl_type = F_WRLCK;
3551 status = nfserr_inval;
3554 file_lock.fl_owner = (fl_owner_t)lock_sop;
3555 file_lock.fl_pid = current->tgid;
3556 file_lock.fl_file = filp;
3557 file_lock.fl_flags = FL_POSIX;
3558 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3560 file_lock.fl_start = lock->lk_offset;
3561 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3562 nfs4_transform_lock_offset(&file_lock);
3565 * Try to lock the file in the VFS.
3566 * Note: locks.c uses the BKL to protect the inode's lock list.
3569 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3571 case 0: /* success! */
3572 update_stateid(&lock_stp->st_stateid);
3573 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3577 case (EAGAIN): /* conflock holds conflicting lock */
3578 status = nfserr_denied;
3579 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3580 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3583 status = nfserr_deadlock;
3586 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3587 status = nfserr_resource;
3591 if (status && lock->lk_is_new && lock_sop)
3592 release_lockowner(lock_sop);
3593 if (lock->lk_replay_owner) {
3594 nfs4_get_stateowner(lock->lk_replay_owner);
3595 cstate->replay_owner = lock->lk_replay_owner;
3597 nfs4_unlock_state();
3602 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3603 * so we do a temporary open here just to get an open file to pass to
3604 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3607 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3612 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3615 err = vfs_test_lock(file, lock);
3624 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3625 struct nfsd4_lockt *lockt)
3627 struct inode *inode;
3628 struct file_lock file_lock;
3632 if (locks_in_grace())
3633 return nfserr_grace;
3635 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3636 return nfserr_inval;
3638 lockt->lt_stateowner = NULL;
3641 status = nfserr_stale_clientid;
3642 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3645 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3646 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3647 if (status == nfserr_symlink)
3648 status = nfserr_inval;
3652 inode = cstate->current_fh.fh_dentry->d_inode;
3653 locks_init_lock(&file_lock);
3654 switch (lockt->lt_type) {
3657 file_lock.fl_type = F_RDLCK;
3660 case NFS4_WRITEW_LT:
3661 file_lock.fl_type = F_WRLCK;
3664 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3665 status = nfserr_inval;
3669 lockt->lt_stateowner = find_lockstateowner_str(inode,
3670 &lockt->lt_clientid, &lockt->lt_owner);
3671 if (lockt->lt_stateowner)
3672 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3673 file_lock.fl_pid = current->tgid;
3674 file_lock.fl_flags = FL_POSIX;
3676 file_lock.fl_start = lockt->lt_offset;
3677 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3679 nfs4_transform_lock_offset(&file_lock);
3682 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3684 status = nfserrno(error);
3687 if (file_lock.fl_type != F_UNLCK) {
3688 status = nfserr_denied;
3689 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3692 nfs4_unlock_state();
3697 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3698 struct nfsd4_locku *locku)
3700 struct nfs4_stateid *stp;
3701 struct file *filp = NULL;
3702 struct file_lock file_lock;
3706 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3707 (long long) locku->lu_offset,
3708 (long long) locku->lu_length);
3710 if (check_lock_length(locku->lu_offset, locku->lu_length))
3711 return nfserr_inval;
3715 if ((status = nfs4_preprocess_seqid_op(cstate,
3719 &locku->lu_stateowner, &stp, NULL)))
3722 filp = stp->st_vfs_file;
3724 locks_init_lock(&file_lock);
3725 file_lock.fl_type = F_UNLCK;
3726 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3727 file_lock.fl_pid = current->tgid;
3728 file_lock.fl_file = filp;
3729 file_lock.fl_flags = FL_POSIX;
3730 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3731 file_lock.fl_start = locku->lu_offset;
3733 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3734 nfs4_transform_lock_offset(&file_lock);
3737 * Try to unlock the file in the VFS.
3739 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3741 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3745 * OK, unlock succeeded; the only thing left to do is update the stateid.
3747 update_stateid(&stp->st_stateid);
3748 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3751 if (locku->lu_stateowner) {
3752 nfs4_get_stateowner(locku->lu_stateowner);
3753 cstate->replay_owner = locku->lu_stateowner;
3755 nfs4_unlock_state();
3759 status = nfserrno(err);
3765 * 1: locks held by lockowner
3766 * 0: no locks held by lockowner
3769 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3771 struct file_lock **flpp;
3772 struct inode *inode = filp->f_path.dentry->d_inode;
3776 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3777 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3788 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3789 struct nfsd4_compound_state *cstate,
3790 struct nfsd4_release_lockowner *rlockowner)
3792 clientid_t *clid = &rlockowner->rl_clientid;
3793 struct nfs4_stateowner *sop;
3794 struct nfs4_stateid *stp;
3795 struct xdr_netobj *owner = &rlockowner->rl_owner;
3796 struct list_head matches;
3800 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3801 clid->cl_boot, clid->cl_id);
3803 /* XXX check for lease expiration */
3805 status = nfserr_stale_clientid;
3806 if (STALE_CLIENTID(clid))
3811 status = nfserr_locks_held;
3812 /* XXX: we're doing a linear search through all the lockowners.
3813 * Yipes! For now we'll just hope clients aren't really using
3814 * release_lockowner much, but eventually we have to fix these
3815 * data structures. */
3816 INIT_LIST_HEAD(&matches);
3817 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3818 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3819 if (!same_owner_str(sop, owner, clid))
3821 list_for_each_entry(stp, &sop->so_stateids,
3823 if (check_for_locks(stp->st_vfs_file, sop))
3825 /* Note: so_perclient unused for lockowners,
3826 * so it's OK to fool with here. */
3827 list_add(&sop->so_perclient, &matches);
3831 /* Clients probably won't expect us to return with some (but not all)
3832 * of the lockowner state released; so don't release any until all
3833 * have been checked. */
3835 while (!list_empty(&matches)) {
3836 sop = list_entry(matches.next, struct nfs4_stateowner,
3838 /* unhash_stateowner deletes so_perclient only
3839 * for openowners. */
3840 list_del(&sop->so_perclient);
3841 release_lockowner(sop);
3844 nfs4_unlock_state();
3848 static inline struct nfs4_client_reclaim *
3851 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3855 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3857 unsigned int strhashval = clientstr_hashval(name);
3858 struct nfs4_client *clp;
3860 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3865 * failure => all reset bets are off, nfserr_no_grace...
3868 nfs4_client_to_reclaim(const char *name)
3870 unsigned int strhashval;
3871 struct nfs4_client_reclaim *crp = NULL;
3873 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3874 crp = alloc_reclaim();
3877 strhashval = clientstr_hashval(name);
3878 INIT_LIST_HEAD(&crp->cr_strhash);
3879 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3880 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3881 reclaim_str_hashtbl_size++;
3886 nfs4_release_reclaim(void)
3888 struct nfs4_client_reclaim *crp = NULL;
3891 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3892 while (!list_empty(&reclaim_str_hashtbl[i])) {
3893 crp = list_entry(reclaim_str_hashtbl[i].next,
3894 struct nfs4_client_reclaim, cr_strhash);
3895 list_del(&crp->cr_strhash);
3897 reclaim_str_hashtbl_size--;
3900 BUG_ON(reclaim_str_hashtbl_size);
3904 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3905 static struct nfs4_client_reclaim *
3906 nfs4_find_reclaim_client(clientid_t *clid)
3908 unsigned int strhashval;
3909 struct nfs4_client *clp;
3910 struct nfs4_client_reclaim *crp = NULL;
3913 /* find clientid in conf_id_hashtbl */
3914 clp = find_confirmed_client(clid);
3918 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3919 clp->cl_name.len, clp->cl_name.data,
3922 /* find clp->cl_name in reclaim_str_hashtbl */
3923 strhashval = clientstr_hashval(clp->cl_recdir);
3924 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3925 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3933 * Called from OPEN. Look for clientid in reclaim list.
3936 nfs4_check_open_reclaim(clientid_t *clid)
3938 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3941 /* initialization to perform at module load time: */
3944 nfs4_state_init(void)
3948 status = nfsd4_init_slabs();
3951 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3952 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3953 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3954 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3955 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3956 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3958 for (i = 0; i < SESSION_HASH_SIZE; i++)
3959 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3960 for (i = 0; i < FILE_HASH_SIZE; i++) {
3961 INIT_LIST_HEAD(&file_hashtbl[i]);
3963 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3964 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3965 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3967 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3968 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3969 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3971 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3972 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3973 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3975 memset(&onestateid, ~0, sizeof(stateid_t));
3976 INIT_LIST_HEAD(&close_lru);
3977 INIT_LIST_HEAD(&client_lru);
3978 INIT_LIST_HEAD(&del_recall_lru);
3979 reclaim_str_hashtbl_size = 0;
3984 nfsd4_load_reboot_recovery_data(void)
3989 nfsd4_init_recdir(user_recovery_dirname);
3990 status = nfsd4_recdir_load();
3991 nfs4_unlock_state();
3993 printk("NFSD: Failure reading reboot recovery data\n");
3997 get_nfs4_grace_period(void)
3999 return max(user_lease_time, lease_time) * HZ;
4003 * Since the lifetime of a delegation isn't limited to that of an open, a
4004 * client may quite reasonably hang on to a delegation as long as it has
4005 * the inode cached. This becomes an obvious problem the first time a
4006 * client's inode cache approaches the size of the server's total memory.
4008 * For now we avoid this problem by imposing a hard limit on the number
4009 * of delegations, which varies according to the server's memory size.
4012 set_max_delegations(void)
4015 * Allow at most 4 delegations per megabyte of RAM. Quick
4016 * estimates suggest that in the worst case (where every delegation
4017 * is for a different inode), a delegation could take about 1.5K,
4018 * giving a worst case usage of about 6% of memory.
4020 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4023 /* initialization to perform when the nfsd service is started: */
4026 __nfs4_state_start(void)
4028 unsigned long grace_time;
4030 boot_time = get_seconds();
4031 grace_time = get_nfs4_grace_period();
4032 lease_time = user_lease_time;
4033 locks_start_grace(&nfsd4_manager);
4034 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4036 laundry_wq = create_singlethread_workqueue("nfsd4");
4037 if (laundry_wq == NULL)
4039 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4040 set_max_delegations();
4041 return set_callback_cred();
4045 nfs4_state_start(void)
4051 nfsd4_load_reboot_recovery_data();
4052 ret = __nfs4_state_start();
4060 nfs4_lease_time(void)
4066 __nfs4_state_shutdown(void)
4069 struct nfs4_client *clp = NULL;
4070 struct nfs4_delegation *dp = NULL;
4071 struct list_head *pos, *next, reaplist;
4073 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4074 while (!list_empty(&conf_id_hashtbl[i])) {
4075 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4078 while (!list_empty(&unconf_str_hashtbl[i])) {
4079 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4083 INIT_LIST_HEAD(&reaplist);
4084 spin_lock(&recall_lock);
4085 list_for_each_safe(pos, next, &del_recall_lru) {
4086 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4087 list_move(&dp->dl_recall_lru, &reaplist);
4089 spin_unlock(&recall_lock);
4090 list_for_each_safe(pos, next, &reaplist) {
4091 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4092 list_del_init(&dp->dl_recall_lru);
4093 unhash_delegation(dp);
4096 nfsd4_shutdown_recdir();
4101 nfs4_state_shutdown(void)
4103 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4104 destroy_workqueue(laundry_wq);
4105 locks_end_grace(&nfsd4_manager);
4107 nfs4_release_reclaim();
4108 __nfs4_state_shutdown();
4109 nfs4_unlock_state();
4113 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4114 * accessed when nfsd is starting.
4117 nfs4_set_recdir(char *recdir)
4119 strcpy(user_recovery_dirname, recdir);
4123 * Change the NFSv4 recovery directory to recdir.
4126 nfs4_reset_recoverydir(char *recdir)
4131 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4135 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4136 nfs4_set_recdir(recdir);
4144 nfs4_recoverydir(void)
4146 return user_recovery_dirname;
4150 * Called when leasetime is changed.
4152 * The only way the protocol gives us to handle on-the-fly lease changes is to
4153 * simulate a reboot. Instead of doing that, we just wait till the next time
4154 * we start to register any changes in lease time. If the administrator
4155 * really wants to change the lease time *now*, they can go ahead and bring
4156 * nfsd down and then back up again after changing the lease time.
4158 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4159 * when nfsd is starting
4162 nfs4_reset_lease(time_t leasetime)
4164 user_lease_time = leasetime;