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 * We don't actually need to cache the rpc and session headers, so we
529 * can allocate a little less for each slot:
531 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
533 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
537 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
538 struct nfsd4_create_session *cses)
540 struct nfsd4_session *new, tmp;
541 struct nfsd4_slot *sp;
542 int idx, slotsize, cachesize, i;
545 memset(&tmp, 0, sizeof(tmp));
547 /* FIXME: For now, we just accept the client back channel attributes. */
548 tmp.se_bchannel = cses->back_channel;
549 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
550 &cses->fore_channel);
554 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
555 + sizeof(struct nfsd4_session) > PAGE_SIZE);
557 status = nfserr_serverfault;
558 /* allocate struct nfsd4_session and slot table pointers in one piece */
559 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
560 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
564 memcpy(new, &tmp, sizeof(*new));
566 /* allocate each struct nfsd4_slot and data cache in one piece */
567 cachesize = slot_bytes(&new->se_fchannel);
568 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
569 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
572 new->se_slots[i] = sp;
575 new->se_client = clp;
577 idx = hash_sessionid(&new->se_sessionid);
578 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
579 NFS4_MAX_SESSIONID_LEN);
581 new->se_flags = cses->flags;
582 kref_init(&new->se_ref);
583 spin_lock(&sessionid_lock);
584 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
585 list_add(&new->se_perclnt, &clp->cl_sessions);
586 spin_unlock(&sessionid_lock);
592 free_session_slots(new);
597 /* caller must hold sessionid_lock */
598 static struct nfsd4_session *
599 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
601 struct nfsd4_session *elem;
604 dump_sessionid(__func__, sessionid);
605 idx = hash_sessionid(sessionid);
606 dprintk("%s: idx is %d\n", __func__, idx);
607 /* Search in the appropriate list */
608 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
609 dump_sessionid("list traversal", &elem->se_sessionid);
610 if (!memcmp(elem->se_sessionid.data, sessionid->data,
611 NFS4_MAX_SESSIONID_LEN)) {
616 dprintk("%s: session not found\n", __func__);
620 /* caller must hold sessionid_lock */
622 unhash_session(struct nfsd4_session *ses)
624 list_del(&ses->se_hash);
625 list_del(&ses->se_perclnt);
629 release_session(struct nfsd4_session *ses)
631 spin_lock(&sessionid_lock);
633 spin_unlock(&sessionid_lock);
634 nfsd4_put_session(ses);
638 free_session(struct kref *kref)
640 struct nfsd4_session *ses;
643 ses = container_of(kref, struct nfsd4_session, se_ref);
644 spin_lock(&nfsd_drc_lock);
645 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
646 nfsd_drc_mem_used -= mem;
647 spin_unlock(&nfsd_drc_lock);
648 free_session_slots(ses);
653 renew_client(struct nfs4_client *clp)
656 * Move client to the end to the LRU list.
658 dprintk("renewing client (clientid %08x/%08x)\n",
659 clp->cl_clientid.cl_boot,
660 clp->cl_clientid.cl_id);
661 list_move_tail(&clp->cl_lru, &client_lru);
662 clp->cl_time = get_seconds();
665 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
667 STALE_CLIENTID(clientid_t *clid)
669 if (clid->cl_boot == boot_time)
671 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
672 clid->cl_boot, clid->cl_id, boot_time);
677 * XXX Should we use a slab cache ?
678 * This type of memory management is somewhat inefficient, but we use it
679 * anyway since SETCLIENTID is not a common operation.
681 static struct nfs4_client *alloc_client(struct xdr_netobj name)
683 struct nfs4_client *clp;
685 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
688 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
689 if (clp->cl_name.data == NULL) {
693 memcpy(clp->cl_name.data, name.data, name.len);
694 clp->cl_name.len = name.len;
699 shutdown_callback_client(struct nfs4_client *clp)
701 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
705 * Callback threads take a reference on the client, so there
706 * should be no outstanding callbacks at this point.
708 clp->cl_cb_conn.cb_client = NULL;
709 rpc_shutdown_client(clnt);
714 free_client(struct nfs4_client *clp)
716 shutdown_callback_client(clp);
718 svc_xprt_put(clp->cl_cb_xprt);
719 if (clp->cl_cred.cr_group_info)
720 put_group_info(clp->cl_cred.cr_group_info);
721 kfree(clp->cl_principal);
722 kfree(clp->cl_name.data);
727 put_nfs4_client(struct nfs4_client *clp)
729 if (atomic_dec_and_test(&clp->cl_count))
734 expire_client(struct nfs4_client *clp)
736 struct nfs4_stateowner *sop;
737 struct nfs4_delegation *dp;
738 struct list_head reaplist;
740 dprintk("NFSD: expire_client cl_count %d\n",
741 atomic_read(&clp->cl_count));
743 INIT_LIST_HEAD(&reaplist);
744 spin_lock(&recall_lock);
745 while (!list_empty(&clp->cl_delegations)) {
746 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
747 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
749 list_del_init(&dp->dl_perclnt);
750 list_move(&dp->dl_recall_lru, &reaplist);
752 spin_unlock(&recall_lock);
753 while (!list_empty(&reaplist)) {
754 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
755 list_del_init(&dp->dl_recall_lru);
756 unhash_delegation(dp);
758 list_del(&clp->cl_idhash);
759 list_del(&clp->cl_strhash);
760 list_del(&clp->cl_lru);
761 while (!list_empty(&clp->cl_openowners)) {
762 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
763 release_openowner(sop);
765 while (!list_empty(&clp->cl_sessions)) {
766 struct nfsd4_session *ses;
767 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
769 release_session(ses);
771 put_nfs4_client(clp);
774 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
776 memcpy(target->cl_verifier.data, source->data,
777 sizeof(target->cl_verifier.data));
780 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
782 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
783 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
786 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
788 target->cr_uid = source->cr_uid;
789 target->cr_gid = source->cr_gid;
790 target->cr_group_info = source->cr_group_info;
791 get_group_info(target->cr_group_info);
794 static int same_name(const char *n1, const char *n2)
796 return 0 == memcmp(n1, n2, HEXDIR_LEN);
800 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
802 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
806 same_clid(clientid_t *cl1, clientid_t *cl2)
808 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
811 /* XXX what about NGROUP */
813 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
815 return cr1->cr_uid == cr2->cr_uid;
818 static void gen_clid(struct nfs4_client *clp)
820 static u32 current_clientid = 1;
822 clp->cl_clientid.cl_boot = boot_time;
823 clp->cl_clientid.cl_id = current_clientid++;
826 static void gen_confirm(struct nfs4_client *clp)
831 p = (u32 *)clp->cl_confirm.data;
832 *p++ = get_seconds();
836 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
837 struct svc_rqst *rqstp, nfs4_verifier *verf)
839 struct nfs4_client *clp;
840 struct sockaddr *sa = svc_addr(rqstp);
843 clp = alloc_client(name);
847 princ = svc_gss_principal(rqstp);
849 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
850 if (clp->cl_principal == NULL) {
856 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
857 atomic_set(&clp->cl_count, 1);
858 atomic_set(&clp->cl_cb_conn.cb_set, 0);
859 INIT_LIST_HEAD(&clp->cl_idhash);
860 INIT_LIST_HEAD(&clp->cl_strhash);
861 INIT_LIST_HEAD(&clp->cl_openowners);
862 INIT_LIST_HEAD(&clp->cl_delegations);
863 INIT_LIST_HEAD(&clp->cl_sessions);
864 INIT_LIST_HEAD(&clp->cl_lru);
865 clear_bit(0, &clp->cl_cb_slot_busy);
866 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
867 copy_verf(clp, verf);
868 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
869 clp->cl_flavor = rqstp->rq_flavor;
870 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
876 static int check_name(struct xdr_netobj name)
880 if (name.len > NFS4_OPAQUE_LIMIT) {
881 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
888 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
890 unsigned int idhashval;
892 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
893 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
894 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
895 list_add_tail(&clp->cl_lru, &client_lru);
896 clp->cl_time = get_seconds();
900 move_to_confirmed(struct nfs4_client *clp)
902 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
903 unsigned int strhashval;
905 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
906 list_del_init(&clp->cl_strhash);
907 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
908 strhashval = clientstr_hashval(clp->cl_recdir);
909 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
913 static struct nfs4_client *
914 find_confirmed_client(clientid_t *clid)
916 struct nfs4_client *clp;
917 unsigned int idhashval = clientid_hashval(clid->cl_id);
919 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
920 if (same_clid(&clp->cl_clientid, clid))
926 static struct nfs4_client *
927 find_unconfirmed_client(clientid_t *clid)
929 struct nfs4_client *clp;
930 unsigned int idhashval = clientid_hashval(clid->cl_id);
932 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
933 if (same_clid(&clp->cl_clientid, clid))
940 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
941 * parameter. Matching is based on the fact the at least one of the
942 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
944 * FIXME: we need to unify the clientid namespaces for nfsv4.x
945 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
946 * and SET_CLIENTID{,_CONFIRM}
949 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
951 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
952 return use_exchange_id == has_exchange_flags;
955 static struct nfs4_client *
956 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
957 bool use_exchange_id)
959 struct nfs4_client *clp;
961 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
962 if (same_name(clp->cl_recdir, dname) &&
963 match_clientid_establishment(clp, use_exchange_id))
969 static struct nfs4_client *
970 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
971 bool use_exchange_id)
973 struct nfs4_client *clp;
975 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
976 if (same_name(clp->cl_recdir, dname) &&
977 match_clientid_establishment(clp, use_exchange_id))
984 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
986 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
987 unsigned short expected_family;
989 /* Currently, we only support tcp and tcp6 for the callback channel */
990 if (se->se_callback_netid_len == 3 &&
991 !memcmp(se->se_callback_netid_val, "tcp", 3))
992 expected_family = AF_INET;
993 else if (se->se_callback_netid_len == 4 &&
994 !memcmp(se->se_callback_netid_val, "tcp6", 4))
995 expected_family = AF_INET6;
999 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1000 se->se_callback_addr_len,
1001 (struct sockaddr *) &cb->cb_addr,
1002 sizeof(cb->cb_addr));
1004 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
1007 if (cb->cb_addr.ss_family == AF_INET6)
1008 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
1010 cb->cb_minorversion = 0;
1011 cb->cb_prog = se->se_callback_prog;
1012 cb->cb_ident = se->se_callback_ident;
1015 cb->cb_addr.ss_family = AF_UNSPEC;
1017 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1018 "will not receive delegations\n",
1019 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1025 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1028 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1030 struct nfsd4_slot *slot = resp->cstate.slot;
1033 dprintk("--> %s slot %p\n", __func__, slot);
1035 slot->sl_opcnt = resp->opcnt;
1036 slot->sl_status = resp->cstate.status;
1038 if (nfsd4_not_cached(resp)) {
1039 slot->sl_datalen = 0;
1042 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1043 base = (char *)resp->cstate.datap -
1044 (char *)resp->xbuf->head[0].iov_base;
1045 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1047 WARN("%s: sessions DRC could not cache compound\n", __func__);
1052 * Encode the replay sequence operation from the slot values.
1053 * If cachethis is FALSE encode the uncached rep error on the next
1054 * operation which sets resp->p and increments resp->opcnt for
1055 * nfs4svc_encode_compoundres.
1059 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1060 struct nfsd4_compoundres *resp)
1062 struct nfsd4_op *op;
1063 struct nfsd4_slot *slot = resp->cstate.slot;
1065 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1066 resp->opcnt, resp->cstate.slot->sl_cachethis);
1068 /* Encode the replayed sequence operation */
1069 op = &args->ops[resp->opcnt - 1];
1070 nfsd4_encode_operation(resp, op);
1072 /* Return nfserr_retry_uncached_rep in next operation. */
1073 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1074 op = &args->ops[resp->opcnt++];
1075 op->status = nfserr_retry_uncached_rep;
1076 nfsd4_encode_operation(resp, op);
1082 * The sequence operation is not cached because we can use the slot and
1086 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1087 struct nfsd4_sequence *seq)
1089 struct nfsd4_slot *slot = resp->cstate.slot;
1092 dprintk("--> %s slot %p\n", __func__, slot);
1094 /* Either returns 0 or nfserr_retry_uncached */
1095 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1096 if (status == nfserr_retry_uncached_rep)
1099 /* The sequence operation has been encoded, cstate->datap set. */
1100 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1102 resp->opcnt = slot->sl_opcnt;
1103 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1104 status = slot->sl_status;
1110 * Set the exchange_id flags returned by the server.
1113 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1115 /* pNFS is not supported */
1116 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1118 /* Referrals are supported, Migration is not. */
1119 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1121 /* set the wire flags to return to client. */
1122 clid->flags = new->cl_exchange_flags;
1126 nfsd4_exchange_id(struct svc_rqst *rqstp,
1127 struct nfsd4_compound_state *cstate,
1128 struct nfsd4_exchange_id *exid)
1130 struct nfs4_client *unconf, *conf, *new;
1132 unsigned int strhashval;
1133 char dname[HEXDIR_LEN];
1134 char addr_str[INET6_ADDRSTRLEN];
1135 nfs4_verifier verf = exid->verifier;
1136 struct sockaddr *sa = svc_addr(rqstp);
1138 rpc_ntop(sa, addr_str, sizeof(addr_str));
1139 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1140 "ip_addr=%s flags %x, spa_how %d\n",
1141 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1142 addr_str, exid->flags, exid->spa_how);
1144 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1145 return nfserr_inval;
1147 /* Currently only support SP4_NONE */
1148 switch (exid->spa_how) {
1152 return nfserr_encr_alg_unsupp;
1154 BUG(); /* checked by xdr code */
1156 return nfserr_serverfault; /* no excuse :-/ */
1159 status = nfs4_make_rec_clidname(dname, &exid->clname);
1164 strhashval = clientstr_hashval(dname);
1169 conf = find_confirmed_client_by_str(dname, strhashval, true);
1171 if (!same_verf(&verf, &conf->cl_verifier)) {
1172 /* 18.35.4 case 8 */
1173 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1174 status = nfserr_not_same;
1177 /* Client reboot: destroy old state */
1178 expire_client(conf);
1181 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1182 /* 18.35.4 case 9 */
1183 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1184 status = nfserr_perm;
1187 expire_client(conf);
1191 * Set bit when the owner id and verifier map to an already
1192 * confirmed client id (18.35.3).
1194 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1197 * Falling into 18.35.4 case 2, possible router replay.
1198 * Leave confirmed record intact and return same result.
1200 copy_verf(conf, &verf);
1205 /* 18.35.4 case 7 */
1206 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1207 status = nfserr_noent;
1211 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1214 * Possible retry or client restart. Per 18.35.4 case 4,
1215 * a new unconfirmed record should be generated regardless
1216 * of whether any properties have changed.
1218 expire_client(unconf);
1223 new = create_client(exid->clname, dname, rqstp, &verf);
1225 status = nfserr_serverfault;
1230 add_to_unconfirmed(new, strhashval);
1232 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1233 exid->clientid.cl_id = new->cl_clientid.cl_id;
1236 nfsd4_set_ex_flags(new, exid);
1238 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1239 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1243 nfs4_unlock_state();
1245 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1250 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1252 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1255 /* The slot is in use, and no response has been sent. */
1257 if (seqid == slot_seqid)
1258 return nfserr_jukebox;
1260 return nfserr_seq_misordered;
1263 if (likely(seqid == slot_seqid + 1))
1266 if (seqid == slot_seqid)
1267 return nfserr_replay_cache;
1269 if (seqid == 1 && (slot_seqid + 1) == 0)
1271 /* Misordered replay or misordered new request */
1272 return nfserr_seq_misordered;
1276 * Cache the create session result into the create session single DRC
1277 * slot cache by saving the xdr structure. sl_seqid has been set.
1278 * Do this for solo or embedded create session operations.
1281 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1282 struct nfsd4_clid_slot *slot, int nfserr)
1284 slot->sl_status = nfserr;
1285 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1289 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1290 struct nfsd4_clid_slot *slot)
1292 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1293 return slot->sl_status;
1297 nfsd4_create_session(struct svc_rqst *rqstp,
1298 struct nfsd4_compound_state *cstate,
1299 struct nfsd4_create_session *cr_ses)
1301 struct sockaddr *sa = svc_addr(rqstp);
1302 struct nfs4_client *conf, *unconf;
1303 struct nfsd4_clid_slot *cs_slot = NULL;
1307 unconf = find_unconfirmed_client(&cr_ses->clientid);
1308 conf = find_confirmed_client(&cr_ses->clientid);
1311 cs_slot = &conf->cl_cs_slot;
1312 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1313 if (status == nfserr_replay_cache) {
1314 dprintk("Got a create_session replay! seqid= %d\n",
1316 /* Return the cached reply status */
1317 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1319 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1320 status = nfserr_seq_misordered;
1321 dprintk("Sequence misordered!\n");
1322 dprintk("Expected seqid= %d but got seqid= %d\n",
1323 cs_slot->sl_seqid, cr_ses->seqid);
1326 cs_slot->sl_seqid++;
1327 } else if (unconf) {
1328 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1329 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1330 status = nfserr_clid_inuse;
1334 cs_slot = &unconf->cl_cs_slot;
1335 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1337 /* an unconfirmed replay returns misordered */
1338 status = nfserr_seq_misordered;
1342 cs_slot->sl_seqid++; /* from 0 to 1 */
1343 move_to_confirmed(unconf);
1346 * We do not support RDMA or persistent sessions
1348 cr_ses->flags &= ~SESSION4_PERSIST;
1349 cr_ses->flags &= ~SESSION4_RDMA;
1351 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1352 unconf->cl_cb_xprt = rqstp->rq_xprt;
1353 svc_xprt_get(unconf->cl_cb_xprt);
1355 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1357 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1358 unconf->cl_cb_conn.cb_minorversion =
1359 cstate->minorversion;
1360 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1361 unconf->cl_cb_seq_nr = 1;
1362 nfsd4_probe_callback(unconf);
1366 status = nfserr_stale_clientid;
1370 status = alloc_init_session(rqstp, conf, cr_ses);
1374 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1375 NFS4_MAX_SESSIONID_LEN);
1376 cr_ses->seqid = cs_slot->sl_seqid;
1379 /* cache solo and embedded create sessions under the state lock */
1380 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1382 nfs4_unlock_state();
1383 dprintk("%s returns %d\n", __func__, ntohl(status));
1388 nfsd4_destroy_session(struct svc_rqst *r,
1389 struct nfsd4_compound_state *cstate,
1390 struct nfsd4_destroy_session *sessionid)
1392 struct nfsd4_session *ses;
1393 u32 status = nfserr_badsession;
1396 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1397 * - Should we return nfserr_back_chan_busy if waiting for
1398 * callbacks on to-be-destroyed session?
1399 * - Do we need to clear any callback info from previous session?
1402 dump_sessionid(__func__, &sessionid->sessionid);
1403 spin_lock(&sessionid_lock);
1404 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1406 spin_unlock(&sessionid_lock);
1410 unhash_session(ses);
1411 spin_unlock(&sessionid_lock);
1413 /* wait for callbacks */
1414 shutdown_callback_client(ses->se_client);
1415 nfsd4_put_session(ses);
1418 dprintk("%s returns %d\n", __func__, ntohl(status));
1423 nfsd4_sequence(struct svc_rqst *rqstp,
1424 struct nfsd4_compound_state *cstate,
1425 struct nfsd4_sequence *seq)
1427 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1428 struct nfsd4_session *session;
1429 struct nfsd4_slot *slot;
1432 if (resp->opcnt != 1)
1433 return nfserr_sequence_pos;
1435 spin_lock(&sessionid_lock);
1436 status = nfserr_badsession;
1437 session = find_in_sessionid_hashtbl(&seq->sessionid);
1441 status = nfserr_badslot;
1442 if (seq->slotid >= session->se_fchannel.maxreqs)
1445 slot = session->se_slots[seq->slotid];
1446 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1448 /* We do not negotiate the number of slots yet, so set the
1449 * maxslots to the session maxreqs which is used to encode
1450 * sr_highest_slotid and the sr_target_slot id to maxslots */
1451 seq->maxslots = session->se_fchannel.maxreqs;
1453 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1454 if (status == nfserr_replay_cache) {
1455 cstate->slot = slot;
1456 cstate->session = session;
1457 /* Return the cached reply status and set cstate->status
1458 * for nfsd4_proc_compound processing */
1459 status = nfsd4_replay_cache_entry(resp, seq);
1460 cstate->status = nfserr_replay_cache;
1466 /* Success! bump slot seqid */
1467 slot->sl_inuse = true;
1468 slot->sl_seqid = seq->seqid;
1469 slot->sl_cachethis = seq->cachethis;
1471 cstate->slot = slot;
1472 cstate->session = session;
1474 /* Hold a session reference until done processing the compound:
1475 * nfsd4_put_session called only if the cstate slot is set.
1477 nfsd4_get_session(session);
1479 spin_unlock(&sessionid_lock);
1480 /* Renew the clientid on success and on replay */
1481 if (cstate->session) {
1483 renew_client(session->se_client);
1484 nfs4_unlock_state();
1486 dprintk("%s: return %d\n", __func__, ntohl(status));
1491 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1492 struct nfsd4_setclientid *setclid)
1494 struct sockaddr *sa = svc_addr(rqstp);
1495 struct xdr_netobj clname = {
1496 .len = setclid->se_namelen,
1497 .data = setclid->se_name,
1499 nfs4_verifier clverifier = setclid->se_verf;
1500 unsigned int strhashval;
1501 struct nfs4_client *conf, *unconf, *new;
1503 char dname[HEXDIR_LEN];
1505 if (!check_name(clname))
1506 return nfserr_inval;
1508 status = nfs4_make_rec_clidname(dname, &clname);
1513 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1514 * We get here on a DRC miss.
1517 strhashval = clientstr_hashval(dname);
1520 conf = find_confirmed_client_by_str(dname, strhashval, false);
1522 /* RFC 3530 14.2.33 CASE 0: */
1523 status = nfserr_clid_inuse;
1524 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1525 char addr_str[INET6_ADDRSTRLEN];
1526 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1528 dprintk("NFSD: setclientid: string in use by client "
1529 "at %s\n", addr_str);
1534 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1535 * has a description of SETCLIENTID request processing consisting
1536 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1538 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1539 status = nfserr_resource;
1542 * RFC 3530 14.2.33 CASE 4:
1543 * placed first, because it is the normal case
1546 expire_client(unconf);
1547 new = create_client(clname, dname, rqstp, &clverifier);
1551 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1553 * RFC 3530 14.2.33 CASE 1:
1554 * probable callback update
1557 /* Note this is removing unconfirmed {*x***},
1558 * which is stronger than RFC recommended {vxc**}.
1559 * This has the advantage that there is at most
1560 * one {*x***} in either list at any time.
1562 expire_client(unconf);
1564 new = create_client(clname, dname, rqstp, &clverifier);
1567 copy_clid(new, conf);
1568 } else if (!unconf) {
1570 * RFC 3530 14.2.33 CASE 2:
1571 * probable client reboot; state will be removed if
1574 new = create_client(clname, dname, rqstp, &clverifier);
1580 * RFC 3530 14.2.33 CASE 3:
1581 * probable client reboot; state will be removed if
1584 expire_client(unconf);
1585 new = create_client(clname, dname, rqstp, &clverifier);
1590 gen_callback(new, setclid, rpc_get_scope_id(sa));
1591 add_to_unconfirmed(new, strhashval);
1592 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1593 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1594 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1597 nfs4_unlock_state();
1603 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1604 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1605 * bullets, labeled as CASE1 - CASE4 below.
1608 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1609 struct nfsd4_compound_state *cstate,
1610 struct nfsd4_setclientid_confirm *setclientid_confirm)
1612 struct sockaddr *sa = svc_addr(rqstp);
1613 struct nfs4_client *conf, *unconf;
1614 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1615 clientid_t * clid = &setclientid_confirm->sc_clientid;
1618 if (STALE_CLIENTID(clid))
1619 return nfserr_stale_clientid;
1621 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1622 * We get here on a DRC miss.
1627 conf = find_confirmed_client(clid);
1628 unconf = find_unconfirmed_client(clid);
1630 status = nfserr_clid_inuse;
1631 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1633 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1637 * section 14.2.34 of RFC 3530 has a description of
1638 * SETCLIENTID_CONFIRM request processing consisting
1639 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1641 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1643 * RFC 3530 14.2.34 CASE 1:
1646 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1647 status = nfserr_clid_inuse;
1649 /* XXX: We just turn off callbacks until we can handle
1650 * change request correctly. */
1651 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1652 expire_client(unconf);
1656 } else if (conf && !unconf) {
1658 * RFC 3530 14.2.34 CASE 2:
1659 * probable retransmitted request; play it safe and
1662 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1663 status = nfserr_clid_inuse;
1666 } else if (!conf && unconf
1667 && same_verf(&unconf->cl_confirm, &confirm)) {
1669 * RFC 3530 14.2.34 CASE 3:
1670 * Normal case; new or rebooted client:
1672 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1673 status = nfserr_clid_inuse;
1676 clientstr_hashval(unconf->cl_recdir);
1677 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1680 nfsd4_remove_clid_dir(conf);
1681 expire_client(conf);
1683 move_to_confirmed(unconf);
1685 nfsd4_probe_callback(conf);
1688 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1689 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1692 * RFC 3530 14.2.34 CASE 4:
1693 * Client probably hasn't noticed that we rebooted yet.
1695 status = nfserr_stale_clientid;
1697 /* check that we have hit one of the cases...*/
1698 status = nfserr_clid_inuse;
1701 nfs4_unlock_state();
1705 /* OPEN Share state helper functions */
1706 static inline struct nfs4_file *
1707 alloc_init_file(struct inode *ino)
1709 struct nfs4_file *fp;
1710 unsigned int hashval = file_hashval(ino);
1712 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1714 atomic_set(&fp->fi_ref, 1);
1715 INIT_LIST_HEAD(&fp->fi_hash);
1716 INIT_LIST_HEAD(&fp->fi_stateids);
1717 INIT_LIST_HEAD(&fp->fi_delegations);
1718 spin_lock(&recall_lock);
1719 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1720 spin_unlock(&recall_lock);
1721 fp->fi_inode = igrab(ino);
1722 fp->fi_id = current_fileid++;
1723 fp->fi_had_conflict = false;
1730 nfsd4_free_slab(struct kmem_cache **slab)
1734 kmem_cache_destroy(*slab);
1739 nfsd4_free_slabs(void)
1741 nfsd4_free_slab(&stateowner_slab);
1742 nfsd4_free_slab(&file_slab);
1743 nfsd4_free_slab(&stateid_slab);
1744 nfsd4_free_slab(&deleg_slab);
1748 nfsd4_init_slabs(void)
1750 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1751 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1752 if (stateowner_slab == NULL)
1754 file_slab = kmem_cache_create("nfsd4_files",
1755 sizeof(struct nfs4_file), 0, 0, NULL);
1756 if (file_slab == NULL)
1758 stateid_slab = kmem_cache_create("nfsd4_stateids",
1759 sizeof(struct nfs4_stateid), 0, 0, NULL);
1760 if (stateid_slab == NULL)
1762 deleg_slab = kmem_cache_create("nfsd4_delegations",
1763 sizeof(struct nfs4_delegation), 0, 0, NULL);
1764 if (deleg_slab == NULL)
1769 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1774 nfs4_free_stateowner(struct kref *kref)
1776 struct nfs4_stateowner *sop =
1777 container_of(kref, struct nfs4_stateowner, so_ref);
1778 kfree(sop->so_owner.data);
1779 kmem_cache_free(stateowner_slab, sop);
1782 static inline struct nfs4_stateowner *
1783 alloc_stateowner(struct xdr_netobj *owner)
1785 struct nfs4_stateowner *sop;
1787 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1788 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1789 memcpy(sop->so_owner.data, owner->data, owner->len);
1790 sop->so_owner.len = owner->len;
1791 kref_init(&sop->so_ref);
1794 kmem_cache_free(stateowner_slab, sop);
1799 static struct nfs4_stateowner *
1800 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1801 struct nfs4_stateowner *sop;
1802 struct nfs4_replay *rp;
1803 unsigned int idhashval;
1805 if (!(sop = alloc_stateowner(&open->op_owner)))
1807 idhashval = ownerid_hashval(current_ownerid);
1808 INIT_LIST_HEAD(&sop->so_idhash);
1809 INIT_LIST_HEAD(&sop->so_strhash);
1810 INIT_LIST_HEAD(&sop->so_perclient);
1811 INIT_LIST_HEAD(&sop->so_stateids);
1812 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1813 INIT_LIST_HEAD(&sop->so_close_lru);
1815 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1816 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1817 list_add(&sop->so_perclient, &clp->cl_openowners);
1818 sop->so_is_open_owner = 1;
1819 sop->so_id = current_ownerid++;
1820 sop->so_client = clp;
1821 sop->so_seqid = open->op_seqid;
1822 sop->so_confirmed = 0;
1823 rp = &sop->so_replay;
1824 rp->rp_status = nfserr_serverfault;
1826 rp->rp_buf = rp->rp_ibuf;
1831 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1832 struct nfs4_stateowner *sop = open->op_stateowner;
1833 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1835 INIT_LIST_HEAD(&stp->st_hash);
1836 INIT_LIST_HEAD(&stp->st_perstateowner);
1837 INIT_LIST_HEAD(&stp->st_lockowners);
1838 INIT_LIST_HEAD(&stp->st_perfile);
1839 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1840 list_add(&stp->st_perstateowner, &sop->so_stateids);
1841 list_add(&stp->st_perfile, &fp->fi_stateids);
1842 stp->st_stateowner = sop;
1845 stp->st_stateid.si_boot = get_seconds();
1846 stp->st_stateid.si_stateownerid = sop->so_id;
1847 stp->st_stateid.si_fileid = fp->fi_id;
1848 stp->st_stateid.si_generation = 0;
1849 stp->st_access_bmap = 0;
1850 stp->st_deny_bmap = 0;
1851 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1852 &stp->st_access_bmap);
1853 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1854 stp->st_openstp = NULL;
1858 move_to_close_lru(struct nfs4_stateowner *sop)
1860 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1862 list_move_tail(&sop->so_close_lru, &close_lru);
1863 sop->so_time = get_seconds();
1867 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1870 return (sop->so_owner.len == owner->len) &&
1871 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1872 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1875 static struct nfs4_stateowner *
1876 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1878 struct nfs4_stateowner *so = NULL;
1880 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1881 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1887 /* search file_hashtbl[] for file */
1888 static struct nfs4_file *
1889 find_file(struct inode *ino)
1891 unsigned int hashval = file_hashval(ino);
1892 struct nfs4_file *fp;
1894 spin_lock(&recall_lock);
1895 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1896 if (fp->fi_inode == ino) {
1898 spin_unlock(&recall_lock);
1902 spin_unlock(&recall_lock);
1906 static inline int access_valid(u32 x, u32 minorversion)
1908 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1910 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1912 x &= ~NFS4_SHARE_ACCESS_MASK;
1913 if (minorversion && x) {
1914 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1916 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1918 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1925 static inline int deny_valid(u32 x)
1927 /* Note: unlike access bits, deny bits may be zero. */
1928 return x <= NFS4_SHARE_DENY_BOTH;
1932 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1933 * st_{access,deny}_bmap field of the stateid, in order to track not
1934 * only what share bits are currently in force, but also what
1935 * combinations of share bits previous opens have used. This allows us
1936 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1937 * return an error if the client attempt to downgrade to a combination
1938 * of share bits not explicable by closing some of its previous opens.
1940 * XXX: This enforcement is actually incomplete, since we don't keep
1941 * track of access/deny bit combinations; so, e.g., we allow:
1943 * OPEN allow read, deny write
1944 * OPEN allow both, deny none
1945 * DOWNGRADE allow read, deny none
1947 * which we should reject.
1950 set_access(unsigned int *access, unsigned long bmap) {
1954 for (i = 1; i < 4; i++) {
1955 if (test_bit(i, &bmap))
1961 set_deny(unsigned int *deny, unsigned long bmap) {
1965 for (i = 0; i < 4; i++) {
1966 if (test_bit(i, &bmap))
1972 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1973 unsigned int access, deny;
1975 set_access(&access, stp->st_access_bmap);
1976 set_deny(&deny, stp->st_deny_bmap);
1977 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1983 * Called to check deny when READ with all zero stateid or
1984 * WRITE with all zero or all one stateid
1987 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1989 struct inode *ino = current_fh->fh_dentry->d_inode;
1990 struct nfs4_file *fp;
1991 struct nfs4_stateid *stp;
1994 dprintk("NFSD: nfs4_share_conflict\n");
1996 fp = find_file(ino);
1999 ret = nfserr_locked;
2000 /* Search for conflicting share reservations */
2001 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2002 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2003 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2013 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2015 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2016 drop_file_write_access(filp);
2017 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2022 * Spawn a thread to perform a recall on the delegation represented
2023 * by the lease (file_lock)
2025 * Called from break_lease() with lock_kernel() held.
2026 * Note: we assume break_lease will only call this *once* for any given
2030 void nfsd_break_deleg_cb(struct file_lock *fl)
2032 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2034 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2038 /* We're assuming the state code never drops its reference
2039 * without first removing the lease. Since we're in this lease
2040 * callback (and since the lease code is serialized by the kernel
2041 * lock) we know the server hasn't removed the lease yet, we know
2042 * it's safe to take a reference: */
2043 atomic_inc(&dp->dl_count);
2044 atomic_inc(&dp->dl_client->cl_count);
2046 spin_lock(&recall_lock);
2047 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2048 spin_unlock(&recall_lock);
2050 /* only place dl_time is set. protected by lock_kernel*/
2051 dp->dl_time = get_seconds();
2054 * We don't want the locks code to timeout the lease for us;
2055 * we'll remove it ourself if the delegation isn't returned
2058 fl->fl_break_time = 0;
2060 dp->dl_file->fi_had_conflict = true;
2061 nfsd4_cb_recall(dp);
2065 * The file_lock is being reapd.
2067 * Called by locks_free_lock() with lock_kernel() held.
2070 void nfsd_release_deleg_cb(struct file_lock *fl)
2072 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2074 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2076 if (!(fl->fl_flags & FL_LEASE) || !dp)
2078 dp->dl_flock = NULL;
2082 * Set the delegation file_lock back pointer.
2084 * Called from setlease() with lock_kernel() held.
2087 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2089 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2091 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2098 * Called from setlease() with lock_kernel() held
2101 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2103 struct nfs4_delegation *onlistd =
2104 (struct nfs4_delegation *)onlist->fl_owner;
2105 struct nfs4_delegation *tryd =
2106 (struct nfs4_delegation *)try->fl_owner;
2108 if (onlist->fl_lmops != try->fl_lmops)
2111 return onlistd->dl_client == tryd->dl_client;
2116 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2119 return lease_modify(onlist, arg);
2124 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2125 .fl_break = nfsd_break_deleg_cb,
2126 .fl_release_private = nfsd_release_deleg_cb,
2127 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2128 .fl_mylease = nfsd_same_client_deleg_cb,
2129 .fl_change = nfsd_change_deleg_cb,
2134 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2135 struct nfsd4_open *open)
2137 clientid_t *clientid = &open->op_clientid;
2138 struct nfs4_client *clp = NULL;
2139 unsigned int strhashval;
2140 struct nfs4_stateowner *sop = NULL;
2142 if (!check_name(open->op_owner))
2143 return nfserr_inval;
2145 if (STALE_CLIENTID(&open->op_clientid))
2146 return nfserr_stale_clientid;
2148 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2149 sop = find_openstateowner_str(strhashval, open);
2150 open->op_stateowner = sop;
2152 /* Make sure the client's lease hasn't expired. */
2153 clp = find_confirmed_client(clientid);
2155 return nfserr_expired;
2158 /* When sessions are used, skip open sequenceid processing */
2159 if (nfsd4_has_session(cstate))
2161 if (!sop->so_confirmed) {
2162 /* Replace unconfirmed owners without checking for replay. */
2163 clp = sop->so_client;
2164 release_openowner(sop);
2165 open->op_stateowner = NULL;
2168 if (open->op_seqid == sop->so_seqid - 1) {
2169 if (sop->so_replay.rp_buflen)
2170 return nfserr_replay_me;
2171 /* The original OPEN failed so spectacularly
2172 * that we don't even have replay data saved!
2173 * Therefore, we have no choice but to continue
2174 * processing this OPEN; presumably, we'll
2175 * fail again for the same reason.
2177 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2180 if (open->op_seqid != sop->so_seqid)
2181 return nfserr_bad_seqid;
2183 if (open->op_stateowner == NULL) {
2184 sop = alloc_init_open_stateowner(strhashval, clp, open);
2186 return nfserr_resource;
2187 open->op_stateowner = sop;
2189 list_del_init(&sop->so_close_lru);
2190 renew_client(sop->so_client);
2194 static inline __be32
2195 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2197 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2198 return nfserr_openmode;
2203 static struct nfs4_delegation *
2204 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2206 struct nfs4_delegation *dp;
2208 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2209 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2216 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2217 struct nfs4_delegation **dp)
2220 __be32 status = nfserr_bad_stateid;
2222 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2225 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2226 RD_STATE : WR_STATE;
2227 status = nfs4_check_delegmode(*dp, flags);
2231 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2235 open->op_stateowner->so_confirmed = 1;
2240 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2242 struct nfs4_stateid *local;
2243 __be32 status = nfserr_share_denied;
2244 struct nfs4_stateowner *sop = open->op_stateowner;
2246 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2247 /* ignore lock owners */
2248 if (local->st_stateowner->so_is_open_owner == 0)
2250 /* remember if we have seen this open owner */
2251 if (local->st_stateowner == sop)
2253 /* check for conflicting share reservations */
2254 if (!test_share(local, open))
2262 static inline struct nfs4_stateid *
2263 nfs4_alloc_stateid(void)
2265 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2269 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2270 struct nfs4_delegation *dp,
2271 struct svc_fh *cur_fh, int flags)
2273 struct nfs4_stateid *stp;
2275 stp = nfs4_alloc_stateid();
2277 return nfserr_resource;
2280 get_file(dp->dl_vfs_file);
2281 stp->st_vfs_file = dp->dl_vfs_file;
2284 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2287 if (status == nfserr_dropit)
2288 status = nfserr_jukebox;
2289 kmem_cache_free(stateid_slab, stp);
2297 static inline __be32
2298 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2299 struct nfsd4_open *open)
2301 struct iattr iattr = {
2302 .ia_valid = ATTR_SIZE,
2305 if (!open->op_truncate)
2307 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2308 return nfserr_inval;
2309 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2313 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2315 struct file *filp = stp->st_vfs_file;
2316 struct inode *inode = filp->f_path.dentry->d_inode;
2317 unsigned int share_access, new_writer;
2320 set_access(&share_access, stp->st_access_bmap);
2321 new_writer = (~share_access) & open->op_share_access
2322 & NFS4_SHARE_ACCESS_WRITE;
2325 int err = get_write_access(inode);
2327 return nfserrno(err);
2328 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2330 return nfserrno(err);
2331 file_take_write(filp);
2333 status = nfsd4_truncate(rqstp, cur_fh, open);
2336 put_write_access(inode);
2339 /* remember the open */
2340 filp->f_mode |= open->op_share_access;
2341 __set_bit(open->op_share_access, &stp->st_access_bmap);
2342 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2349 nfs4_set_claim_prev(struct nfsd4_open *open)
2351 open->op_stateowner->so_confirmed = 1;
2352 open->op_stateowner->so_client->cl_firststate = 1;
2356 * Attempt to hand out a delegation.
2359 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2361 struct nfs4_delegation *dp;
2362 struct nfs4_stateowner *sop = stp->st_stateowner;
2363 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2364 struct file_lock fl, *flp = &fl;
2365 int status, flag = 0;
2367 flag = NFS4_OPEN_DELEGATE_NONE;
2368 open->op_recall = 0;
2369 switch (open->op_claim_type) {
2370 case NFS4_OPEN_CLAIM_PREVIOUS:
2371 if (!atomic_read(&cb->cb_set))
2372 open->op_recall = 1;
2373 flag = open->op_delegate_type;
2374 if (flag == NFS4_OPEN_DELEGATE_NONE)
2377 case NFS4_OPEN_CLAIM_NULL:
2378 /* Let's not give out any delegations till everyone's
2379 * had the chance to reclaim theirs.... */
2380 if (locks_in_grace())
2382 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2384 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2385 flag = NFS4_OPEN_DELEGATE_WRITE;
2387 flag = NFS4_OPEN_DELEGATE_READ;
2393 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2395 flag = NFS4_OPEN_DELEGATE_NONE;
2398 locks_init_lock(&fl);
2399 fl.fl_lmops = &nfsd_lease_mng_ops;
2400 fl.fl_flags = FL_LEASE;
2401 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2402 fl.fl_end = OFFSET_MAX;
2403 fl.fl_owner = (fl_owner_t)dp;
2404 fl.fl_file = stp->st_vfs_file;
2405 fl.fl_pid = current->tgid;
2407 /* vfs_setlease checks to see if delegation should be handed out.
2408 * the lock_manager callbacks fl_mylease and fl_change are used
2410 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2411 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2412 unhash_delegation(dp);
2413 flag = NFS4_OPEN_DELEGATE_NONE;
2417 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2419 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2420 dp->dl_stateid.si_boot,
2421 dp->dl_stateid.si_stateownerid,
2422 dp->dl_stateid.si_fileid,
2423 dp->dl_stateid.si_generation);
2425 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2426 && flag == NFS4_OPEN_DELEGATE_NONE
2427 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2428 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2429 open->op_delegate_type = flag;
2433 * called with nfs4_lock_state() held.
2436 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2438 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2439 struct nfs4_file *fp = NULL;
2440 struct inode *ino = current_fh->fh_dentry->d_inode;
2441 struct nfs4_stateid *stp = NULL;
2442 struct nfs4_delegation *dp = NULL;
2445 status = nfserr_inval;
2446 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2447 || !deny_valid(open->op_share_deny))
2450 * Lookup file; if found, lookup stateid and check open request,
2451 * and check for delegations in the process of being recalled.
2452 * If not found, create the nfs4_file struct
2454 fp = find_file(ino);
2456 if ((status = nfs4_check_open(fp, open, &stp)))
2458 status = nfs4_check_deleg(fp, open, &dp);
2462 status = nfserr_bad_stateid;
2463 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2465 status = nfserr_resource;
2466 fp = alloc_init_file(ino);
2472 * OPEN the file, or upgrade an existing OPEN.
2473 * If truncate fails, the OPEN fails.
2476 /* Stateid was found, this is an OPEN upgrade */
2477 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2480 update_stateid(&stp->st_stateid);
2482 /* Stateid was not found, this is a new OPEN */
2484 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2485 flags |= NFSD_MAY_READ;
2486 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2487 flags |= NFSD_MAY_WRITE;
2488 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2491 init_stateid(stp, fp, open);
2492 status = nfsd4_truncate(rqstp, current_fh, open);
2494 release_open_stateid(stp);
2497 if (nfsd4_has_session(&resp->cstate))
2498 update_stateid(&stp->st_stateid);
2500 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2502 if (nfsd4_has_session(&resp->cstate))
2503 open->op_stateowner->so_confirmed = 1;
2506 * Attempt to hand out a delegation. No error return, because the
2507 * OPEN succeeds even if we fail.
2509 nfs4_open_delegation(current_fh, open, stp);
2513 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2514 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2515 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2519 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2520 nfs4_set_claim_prev(open);
2522 * To finish the open response, we just need to set the rflags.
2524 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2525 if (!open->op_stateowner->so_confirmed &&
2526 !nfsd4_has_session(&resp->cstate))
2527 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2533 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2536 struct nfs4_client *clp;
2540 dprintk("process_renew(%08x/%08x): starting\n",
2541 clid->cl_boot, clid->cl_id);
2542 status = nfserr_stale_clientid;
2543 if (STALE_CLIENTID(clid))
2545 clp = find_confirmed_client(clid);
2546 status = nfserr_expired;
2548 /* We assume the client took too long to RENEW. */
2549 dprintk("nfsd4_renew: clientid not found!\n");
2553 status = nfserr_cb_path_down;
2554 if (!list_empty(&clp->cl_delegations)
2555 && !atomic_read(&clp->cl_cb_conn.cb_set))
2559 nfs4_unlock_state();
2563 struct lock_manager nfsd4_manager = {
2567 nfsd4_end_grace(void)
2569 dprintk("NFSD: end of grace period\n");
2570 nfsd4_recdir_purge_old();
2571 locks_end_grace(&nfsd4_manager);
2575 nfs4_laundromat(void)
2577 struct nfs4_client *clp;
2578 struct nfs4_stateowner *sop;
2579 struct nfs4_delegation *dp;
2580 struct list_head *pos, *next, reaplist;
2581 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2582 time_t t, clientid_val = NFSD_LEASE_TIME;
2583 time_t u, test_val = NFSD_LEASE_TIME;
2587 dprintk("NFSD: laundromat service - starting\n");
2588 if (locks_in_grace())
2590 list_for_each_safe(pos, next, &client_lru) {
2591 clp = list_entry(pos, struct nfs4_client, cl_lru);
2592 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2593 t = clp->cl_time - cutoff;
2594 if (clientid_val > t)
2598 dprintk("NFSD: purging unused client (clientid %08x)\n",
2599 clp->cl_clientid.cl_id);
2600 nfsd4_remove_clid_dir(clp);
2603 INIT_LIST_HEAD(&reaplist);
2604 spin_lock(&recall_lock);
2605 list_for_each_safe(pos, next, &del_recall_lru) {
2606 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2607 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2608 u = dp->dl_time - cutoff;
2613 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2615 list_move(&dp->dl_recall_lru, &reaplist);
2617 spin_unlock(&recall_lock);
2618 list_for_each_safe(pos, next, &reaplist) {
2619 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2620 list_del_init(&dp->dl_recall_lru);
2621 unhash_delegation(dp);
2623 test_val = NFSD_LEASE_TIME;
2624 list_for_each_safe(pos, next, &close_lru) {
2625 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2626 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2627 u = sop->so_time - cutoff;
2632 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2634 release_openowner(sop);
2636 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2637 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2638 nfs4_unlock_state();
2639 return clientid_val;
2642 static struct workqueue_struct *laundry_wq;
2643 static void laundromat_main(struct work_struct *);
2644 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2647 laundromat_main(struct work_struct *not_used)
2651 t = nfs4_laundromat();
2652 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2653 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2656 static struct nfs4_stateowner *
2657 search_close_lru(u32 st_id, int flags)
2659 struct nfs4_stateowner *local = NULL;
2661 if (flags & CLOSE_STATE) {
2662 list_for_each_entry(local, &close_lru, so_close_lru) {
2663 if (local->so_id == st_id)
2671 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2673 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2677 STALE_STATEID(stateid_t *stateid)
2679 if (time_after((unsigned long)boot_time,
2680 (unsigned long)stateid->si_boot)) {
2681 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2682 stateid->si_boot, stateid->si_stateownerid,
2683 stateid->si_fileid, stateid->si_generation);
2690 EXPIRED_STATEID(stateid_t *stateid)
2692 if (time_before((unsigned long)boot_time,
2693 ((unsigned long)stateid->si_boot)) &&
2694 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2695 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2696 stateid->si_boot, stateid->si_stateownerid,
2697 stateid->si_fileid, stateid->si_generation);
2704 stateid_error_map(stateid_t *stateid)
2706 if (STALE_STATEID(stateid))
2707 return nfserr_stale_stateid;
2708 if (EXPIRED_STATEID(stateid))
2709 return nfserr_expired;
2711 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2712 stateid->si_boot, stateid->si_stateownerid,
2713 stateid->si_fileid, stateid->si_generation);
2714 return nfserr_bad_stateid;
2718 access_permit_read(unsigned long access_bmap)
2720 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2721 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2722 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2726 access_permit_write(unsigned long access_bmap)
2728 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2729 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2733 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2735 __be32 status = nfserr_openmode;
2737 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2739 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2746 static inline __be32
2747 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2749 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2751 else if (locks_in_grace()) {
2752 /* Answer in remaining cases depends on existance of
2753 * conflicting state; so we must wait out the grace period. */
2754 return nfserr_grace;
2755 } else if (flags & WR_STATE)
2756 return nfs4_share_conflict(current_fh,
2757 NFS4_SHARE_DENY_WRITE);
2758 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2759 return nfs4_share_conflict(current_fh,
2760 NFS4_SHARE_DENY_READ);
2764 * Allow READ/WRITE during grace period on recovered state only for files
2765 * that are not able to provide mandatory locking.
2768 grace_disallows_io(struct inode *inode)
2770 return locks_in_grace() && mandatory_lock(inode);
2773 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2776 * When sessions are used the stateid generation number is ignored
2779 if ((flags & HAS_SESSION) && in->si_generation == 0)
2782 /* If the client sends us a stateid from the future, it's buggy: */
2783 if (in->si_generation > ref->si_generation)
2784 return nfserr_bad_stateid;
2786 * The following, however, can happen. For example, if the
2787 * client sends an open and some IO at the same time, the open
2788 * may bump si_generation while the IO is still in flight.
2789 * Thanks to hard links and renames, the client never knows what
2790 * file an open will affect. So it could avoid that situation
2791 * only by serializing all opens and IO from the same open
2792 * owner. To recover from the old_stateid error, the client
2793 * will just have to retry the IO:
2795 if (in->si_generation < ref->si_generation)
2796 return nfserr_old_stateid;
2801 static int is_delegation_stateid(stateid_t *stateid)
2803 return stateid->si_fileid == 0;
2807 * Checks for stateid operations
2810 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2811 stateid_t *stateid, int flags, struct file **filpp)
2813 struct nfs4_stateid *stp = NULL;
2814 struct nfs4_delegation *dp = NULL;
2815 struct svc_fh *current_fh = &cstate->current_fh;
2816 struct inode *ino = current_fh->fh_dentry->d_inode;
2822 if (grace_disallows_io(ino))
2823 return nfserr_grace;
2825 if (nfsd4_has_session(cstate))
2826 flags |= HAS_SESSION;
2828 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2829 return check_special_stateids(current_fh, stateid, flags);
2831 status = nfserr_stale_stateid;
2832 if (STALE_STATEID(stateid))
2835 status = nfserr_bad_stateid;
2836 if (is_delegation_stateid(stateid)) {
2837 dp = find_delegation_stateid(ino, stateid);
2839 status = stateid_error_map(stateid);
2842 status = check_stateid_generation(stateid, &dp->dl_stateid,
2846 status = nfs4_check_delegmode(dp, flags);
2849 renew_client(dp->dl_client);
2851 *filpp = dp->dl_vfs_file;
2852 } else { /* open or lock stateid */
2853 stp = find_stateid(stateid, flags);
2855 status = stateid_error_map(stateid);
2858 if (nfs4_check_fh(current_fh, stp))
2860 if (!stp->st_stateowner->so_confirmed)
2862 status = check_stateid_generation(stateid, &stp->st_stateid,
2866 status = nfs4_check_openmode(stp, flags);
2869 renew_client(stp->st_stateowner->so_client);
2871 *filpp = stp->st_vfs_file;
2881 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2882 RD_STATE : WR_STATE;
2886 * Checks for sequence id mutating operations.
2889 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2890 stateid_t *stateid, int flags,
2891 struct nfs4_stateowner **sopp,
2892 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2894 struct nfs4_stateid *stp;
2895 struct nfs4_stateowner *sop;
2896 struct svc_fh *current_fh = &cstate->current_fh;
2899 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2900 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2901 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2902 stateid->si_generation);
2907 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2908 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2909 return nfserr_bad_stateid;
2912 if (STALE_STATEID(stateid))
2913 return nfserr_stale_stateid;
2915 if (nfsd4_has_session(cstate))
2916 flags |= HAS_SESSION;
2919 * We return BAD_STATEID if filehandle doesn't match stateid,
2920 * the confirmed flag is incorrecly set, or the generation
2921 * number is incorrect.
2923 stp = find_stateid(stateid, flags);
2926 * Also, we should make sure this isn't just the result of
2929 sop = search_close_lru(stateid->si_stateownerid, flags);
2931 return stateid_error_map(stateid);
2937 *sopp = sop = stp->st_stateowner;
2940 clientid_t *lockclid = &lock->v.new.clientid;
2941 struct nfs4_client *clp = sop->so_client;
2945 lkflg = setlkflg(lock->lk_type);
2947 if (lock->lk_is_new) {
2948 if (!sop->so_is_open_owner)
2949 return nfserr_bad_stateid;
2950 if (!(flags & HAS_SESSION) &&
2951 !same_clid(&clp->cl_clientid, lockclid))
2952 return nfserr_bad_stateid;
2953 /* stp is the open stateid */
2954 status = nfs4_check_openmode(stp, lkflg);
2958 /* stp is the lock stateid */
2959 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2965 if (nfs4_check_fh(current_fh, stp)) {
2966 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2967 return nfserr_bad_stateid;
2971 * We now validate the seqid and stateid generation numbers.
2972 * For the moment, we ignore the possibility of
2973 * generation number wraparound.
2975 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2978 if (sop->so_confirmed && flags & CONFIRM) {
2979 dprintk("NFSD: preprocess_seqid_op: expected"
2980 " unconfirmed stateowner!\n");
2981 return nfserr_bad_stateid;
2983 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2984 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2985 " confirmed yet!\n");
2986 return nfserr_bad_stateid;
2988 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2991 renew_client(sop->so_client);
2995 if (seqid == sop->so_seqid - 1) {
2996 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2997 /* indicate replay to calling function */
2998 return nfserr_replay_me;
3000 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3001 sop->so_seqid, seqid);
3003 return nfserr_bad_seqid;
3007 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3008 struct nfsd4_open_confirm *oc)
3011 struct nfs4_stateowner *sop;
3012 struct nfs4_stateid *stp;
3014 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3015 (int)cstate->current_fh.fh_dentry->d_name.len,
3016 cstate->current_fh.fh_dentry->d_name.name);
3018 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3024 if ((status = nfs4_preprocess_seqid_op(cstate,
3025 oc->oc_seqid, &oc->oc_req_stateid,
3026 CONFIRM | OPEN_STATE,
3027 &oc->oc_stateowner, &stp, NULL)))
3030 sop = oc->oc_stateowner;
3031 sop->so_confirmed = 1;
3032 update_stateid(&stp->st_stateid);
3033 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3034 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3035 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3036 stp->st_stateid.si_boot,
3037 stp->st_stateid.si_stateownerid,
3038 stp->st_stateid.si_fileid,
3039 stp->st_stateid.si_generation);
3041 nfsd4_create_clid_dir(sop->so_client);
3043 if (oc->oc_stateowner) {
3044 nfs4_get_stateowner(oc->oc_stateowner);
3045 cstate->replay_owner = oc->oc_stateowner;
3047 nfs4_unlock_state();
3053 * unset all bits in union bitmap (bmap) that
3054 * do not exist in share (from successful OPEN_DOWNGRADE)
3057 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3060 for (i = 1; i < 4; i++) {
3061 if ((i & access) != i)
3062 __clear_bit(i, bmap);
3067 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3070 for (i = 0; i < 4; i++) {
3071 if ((i & deny) != i)
3072 __clear_bit(i, bmap);
3077 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3078 struct nfsd4_compound_state *cstate,
3079 struct nfsd4_open_downgrade *od)
3082 struct nfs4_stateid *stp;
3083 unsigned int share_access;
3085 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3086 (int)cstate->current_fh.fh_dentry->d_name.len,
3087 cstate->current_fh.fh_dentry->d_name.name);
3089 if (!access_valid(od->od_share_access, cstate->minorversion)
3090 || !deny_valid(od->od_share_deny))
3091 return nfserr_inval;
3094 if ((status = nfs4_preprocess_seqid_op(cstate,
3098 &od->od_stateowner, &stp, NULL)))
3101 status = nfserr_inval;
3102 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3103 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3104 stp->st_access_bmap, od->od_share_access);
3107 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3108 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3109 stp->st_deny_bmap, od->od_share_deny);
3112 set_access(&share_access, stp->st_access_bmap);
3113 nfs4_file_downgrade(stp->st_vfs_file,
3114 share_access & ~od->od_share_access);
3116 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3117 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3119 update_stateid(&stp->st_stateid);
3120 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3123 if (od->od_stateowner) {
3124 nfs4_get_stateowner(od->od_stateowner);
3125 cstate->replay_owner = od->od_stateowner;
3127 nfs4_unlock_state();
3132 * nfs4_unlock_state() called after encode
3135 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3136 struct nfsd4_close *close)
3139 struct nfs4_stateid *stp;
3141 dprintk("NFSD: nfsd4_close on file %.*s\n",
3142 (int)cstate->current_fh.fh_dentry->d_name.len,
3143 cstate->current_fh.fh_dentry->d_name.name);
3146 /* check close_lru for replay */
3147 if ((status = nfs4_preprocess_seqid_op(cstate,
3150 OPEN_STATE | CLOSE_STATE,
3151 &close->cl_stateowner, &stp, NULL)))
3154 update_stateid(&stp->st_stateid);
3155 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3157 /* release_stateid() calls nfsd_close() if needed */
3158 release_open_stateid(stp);
3160 /* place unused nfs4_stateowners on so_close_lru list to be
3161 * released by the laundromat service after the lease period
3162 * to enable us to handle CLOSE replay
3164 if (list_empty(&close->cl_stateowner->so_stateids))
3165 move_to_close_lru(close->cl_stateowner);
3167 if (close->cl_stateowner) {
3168 nfs4_get_stateowner(close->cl_stateowner);
3169 cstate->replay_owner = close->cl_stateowner;
3171 nfs4_unlock_state();
3176 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3177 struct nfsd4_delegreturn *dr)
3179 struct nfs4_delegation *dp;
3180 stateid_t *stateid = &dr->dr_stateid;
3181 struct inode *inode;
3185 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3187 inode = cstate->current_fh.fh_dentry->d_inode;
3189 if (nfsd4_has_session(cstate))
3190 flags |= HAS_SESSION;
3192 status = nfserr_bad_stateid;
3193 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3195 status = nfserr_stale_stateid;
3196 if (STALE_STATEID(stateid))
3198 status = nfserr_bad_stateid;
3199 if (!is_delegation_stateid(stateid))
3201 dp = find_delegation_stateid(inode, stateid);
3203 status = stateid_error_map(stateid);
3206 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3209 renew_client(dp->dl_client);
3211 unhash_delegation(dp);
3213 nfs4_unlock_state();
3220 * Lock owner state (byte-range locks)
3222 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3223 #define LOCK_HASH_BITS 8
3224 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3225 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3228 end_offset(u64 start, u64 len)
3233 return end >= start ? end: NFS4_MAX_UINT64;
3236 /* last octet in a range */
3238 last_byte_offset(u64 start, u64 len)
3244 return end > start ? end - 1: NFS4_MAX_UINT64;
3247 #define lockownerid_hashval(id) \
3248 ((id) & LOCK_HASH_MASK)
3250 static inline unsigned int
3251 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3252 struct xdr_netobj *ownername)
3254 return (file_hashval(inode) + cl_id
3255 + opaque_hashval(ownername->data, ownername->len))
3259 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3260 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3261 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3263 static struct nfs4_stateid *
3264 find_stateid(stateid_t *stid, int flags)
3266 struct nfs4_stateid *local;
3267 u32 st_id = stid->si_stateownerid;
3268 u32 f_id = stid->si_fileid;
3269 unsigned int hashval;
3271 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3272 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3273 hashval = stateid_hashval(st_id, f_id);
3274 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3275 if ((local->st_stateid.si_stateownerid == st_id) &&
3276 (local->st_stateid.si_fileid == f_id))
3281 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3282 hashval = stateid_hashval(st_id, f_id);
3283 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3284 if ((local->st_stateid.si_stateownerid == st_id) &&
3285 (local->st_stateid.si_fileid == f_id))
3292 static struct nfs4_delegation *
3293 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3295 struct nfs4_file *fp;
3296 struct nfs4_delegation *dl;
3298 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3299 stid->si_boot, stid->si_stateownerid,
3300 stid->si_fileid, stid->si_generation);
3302 fp = find_file(ino);
3305 dl = find_delegation_file(fp, stid);
3311 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3312 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3313 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3314 * locking, this prevents us from being completely protocol-compliant. The
3315 * real solution to this problem is to start using unsigned file offsets in
3316 * the VFS, but this is a very deep change!
3319 nfs4_transform_lock_offset(struct file_lock *lock)
3321 if (lock->fl_start < 0)
3322 lock->fl_start = OFFSET_MAX;
3323 if (lock->fl_end < 0)
3324 lock->fl_end = OFFSET_MAX;
3327 /* Hack!: For now, we're defining this just so we can use a pointer to it
3328 * as a unique cookie to identify our (NFSv4's) posix locks. */
3329 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3333 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3335 struct nfs4_stateowner *sop;
3338 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3339 sop = (struct nfs4_stateowner *) fl->fl_owner;
3340 hval = lockownerid_hashval(sop->so_id);
3341 kref_get(&sop->so_ref);
3343 deny->ld_clientid = sop->so_client->cl_clientid;
3345 deny->ld_sop = NULL;
3346 deny->ld_clientid.cl_boot = 0;
3347 deny->ld_clientid.cl_id = 0;
3349 deny->ld_start = fl->fl_start;
3350 deny->ld_length = NFS4_MAX_UINT64;
3351 if (fl->fl_end != NFS4_MAX_UINT64)
3352 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3353 deny->ld_type = NFS4_READ_LT;
3354 if (fl->fl_type != F_RDLCK)
3355 deny->ld_type = NFS4_WRITE_LT;
3358 static struct nfs4_stateowner *
3359 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3360 struct xdr_netobj *owner)
3362 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3363 struct nfs4_stateowner *op;
3365 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3366 if (same_owner_str(op, owner, clid))
3373 * Alloc a lock owner structure.
3374 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3377 * strhashval = lock_ownerstr_hashval
3380 static struct nfs4_stateowner *
3381 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3382 struct nfs4_stateowner *sop;
3383 struct nfs4_replay *rp;
3384 unsigned int idhashval;
3386 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3388 idhashval = lockownerid_hashval(current_ownerid);
3389 INIT_LIST_HEAD(&sop->so_idhash);
3390 INIT_LIST_HEAD(&sop->so_strhash);
3391 INIT_LIST_HEAD(&sop->so_perclient);
3392 INIT_LIST_HEAD(&sop->so_stateids);
3393 INIT_LIST_HEAD(&sop->so_perstateid);
3394 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3396 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3397 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3398 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3399 sop->so_is_open_owner = 0;
3400 sop->so_id = current_ownerid++;
3401 sop->so_client = clp;
3402 /* It is the openowner seqid that will be incremented in encode in the
3403 * case of new lockowners; so increment the lock seqid manually: */
3404 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3405 sop->so_confirmed = 1;
3406 rp = &sop->so_replay;
3407 rp->rp_status = nfserr_serverfault;
3409 rp->rp_buf = rp->rp_ibuf;
3413 static struct nfs4_stateid *
3414 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3416 struct nfs4_stateid *stp;
3417 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3419 stp = nfs4_alloc_stateid();
3422 INIT_LIST_HEAD(&stp->st_hash);
3423 INIT_LIST_HEAD(&stp->st_perfile);
3424 INIT_LIST_HEAD(&stp->st_perstateowner);
3425 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3426 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3427 list_add(&stp->st_perfile, &fp->fi_stateids);
3428 list_add(&stp->st_perstateowner, &sop->so_stateids);
3429 stp->st_stateowner = sop;
3432 stp->st_stateid.si_boot = get_seconds();
3433 stp->st_stateid.si_stateownerid = sop->so_id;
3434 stp->st_stateid.si_fileid = fp->fi_id;
3435 stp->st_stateid.si_generation = 0;
3436 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3437 stp->st_access_bmap = open_stp->st_access_bmap;
3438 stp->st_deny_bmap = open_stp->st_deny_bmap;
3439 stp->st_openstp = open_stp;
3446 check_lock_length(u64 offset, u64 length)
3448 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3449 LOFF_OVERFLOW(offset, length)));
3456 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3457 struct nfsd4_lock *lock)
3459 struct nfs4_stateowner *open_sop = NULL;
3460 struct nfs4_stateowner *lock_sop = NULL;
3461 struct nfs4_stateid *lock_stp;
3463 struct file_lock file_lock;
3464 struct file_lock conflock;
3466 unsigned int strhashval;
3470 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3471 (long long) lock->lk_offset,
3472 (long long) lock->lk_length);
3474 if (check_lock_length(lock->lk_offset, lock->lk_length))
3475 return nfserr_inval;
3477 if ((status = fh_verify(rqstp, &cstate->current_fh,
3478 S_IFREG, NFSD_MAY_LOCK))) {
3479 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3485 if (lock->lk_is_new) {
3487 * Client indicates that this is a new lockowner.
3488 * Use open owner and open stateid to create lock owner and
3491 struct nfs4_stateid *open_stp = NULL;
3492 struct nfs4_file *fp;
3494 status = nfserr_stale_clientid;
3495 if (!nfsd4_has_session(cstate) &&
3496 STALE_CLIENTID(&lock->lk_new_clientid))
3499 /* validate and update open stateid and open seqid */
3500 status = nfs4_preprocess_seqid_op(cstate,
3501 lock->lk_new_open_seqid,
3502 &lock->lk_new_open_stateid,
3504 &lock->lk_replay_owner, &open_stp,
3508 open_sop = lock->lk_replay_owner;
3509 /* create lockowner and lock stateid */
3510 fp = open_stp->st_file;
3511 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3512 open_sop->so_client->cl_clientid.cl_id,
3513 &lock->v.new.owner);
3514 /* XXX: Do we need to check for duplicate stateowners on
3515 * the same file, or should they just be allowed (and
3516 * create new stateids)? */
3517 status = nfserr_resource;
3518 lock_sop = alloc_init_lock_stateowner(strhashval,
3519 open_sop->so_client, open_stp, lock);
3520 if (lock_sop == NULL)
3522 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3523 if (lock_stp == NULL)
3526 /* lock (lock owner + lock stateid) already exists */
3527 status = nfs4_preprocess_seqid_op(cstate,
3528 lock->lk_old_lock_seqid,
3529 &lock->lk_old_lock_stateid,
3531 &lock->lk_replay_owner, &lock_stp, lock);
3534 lock_sop = lock->lk_replay_owner;
3536 /* lock->lk_replay_owner and lock_stp have been created or found */
3537 filp = lock_stp->st_vfs_file;
3539 status = nfserr_grace;
3540 if (locks_in_grace() && !lock->lk_reclaim)
3542 status = nfserr_no_grace;
3543 if (!locks_in_grace() && lock->lk_reclaim)
3546 locks_init_lock(&file_lock);
3547 switch (lock->lk_type) {
3550 file_lock.fl_type = F_RDLCK;
3554 case NFS4_WRITEW_LT:
3555 file_lock.fl_type = F_WRLCK;
3559 status = nfserr_inval;
3562 file_lock.fl_owner = (fl_owner_t)lock_sop;
3563 file_lock.fl_pid = current->tgid;
3564 file_lock.fl_file = filp;
3565 file_lock.fl_flags = FL_POSIX;
3566 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3568 file_lock.fl_start = lock->lk_offset;
3569 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3570 nfs4_transform_lock_offset(&file_lock);
3573 * Try to lock the file in the VFS.
3574 * Note: locks.c uses the BKL to protect the inode's lock list.
3577 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3579 case 0: /* success! */
3580 update_stateid(&lock_stp->st_stateid);
3581 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3585 case (EAGAIN): /* conflock holds conflicting lock */
3586 status = nfserr_denied;
3587 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3588 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3591 status = nfserr_deadlock;
3594 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3595 status = nfserr_resource;
3599 if (status && lock->lk_is_new && lock_sop)
3600 release_lockowner(lock_sop);
3601 if (lock->lk_replay_owner) {
3602 nfs4_get_stateowner(lock->lk_replay_owner);
3603 cstate->replay_owner = lock->lk_replay_owner;
3605 nfs4_unlock_state();
3610 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3611 * so we do a temporary open here just to get an open file to pass to
3612 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3615 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3620 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3623 err = vfs_test_lock(file, lock);
3632 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3633 struct nfsd4_lockt *lockt)
3635 struct inode *inode;
3636 struct file_lock file_lock;
3640 if (locks_in_grace())
3641 return nfserr_grace;
3643 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3644 return nfserr_inval;
3646 lockt->lt_stateowner = NULL;
3649 status = nfserr_stale_clientid;
3650 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3653 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3654 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3655 if (status == nfserr_symlink)
3656 status = nfserr_inval;
3660 inode = cstate->current_fh.fh_dentry->d_inode;
3661 locks_init_lock(&file_lock);
3662 switch (lockt->lt_type) {
3665 file_lock.fl_type = F_RDLCK;
3668 case NFS4_WRITEW_LT:
3669 file_lock.fl_type = F_WRLCK;
3672 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3673 status = nfserr_inval;
3677 lockt->lt_stateowner = find_lockstateowner_str(inode,
3678 &lockt->lt_clientid, &lockt->lt_owner);
3679 if (lockt->lt_stateowner)
3680 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3681 file_lock.fl_pid = current->tgid;
3682 file_lock.fl_flags = FL_POSIX;
3684 file_lock.fl_start = lockt->lt_offset;
3685 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3687 nfs4_transform_lock_offset(&file_lock);
3690 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3692 status = nfserrno(error);
3695 if (file_lock.fl_type != F_UNLCK) {
3696 status = nfserr_denied;
3697 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3700 nfs4_unlock_state();
3705 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3706 struct nfsd4_locku *locku)
3708 struct nfs4_stateid *stp;
3709 struct file *filp = NULL;
3710 struct file_lock file_lock;
3714 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3715 (long long) locku->lu_offset,
3716 (long long) locku->lu_length);
3718 if (check_lock_length(locku->lu_offset, locku->lu_length))
3719 return nfserr_inval;
3723 if ((status = nfs4_preprocess_seqid_op(cstate,
3727 &locku->lu_stateowner, &stp, NULL)))
3730 filp = stp->st_vfs_file;
3732 locks_init_lock(&file_lock);
3733 file_lock.fl_type = F_UNLCK;
3734 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3735 file_lock.fl_pid = current->tgid;
3736 file_lock.fl_file = filp;
3737 file_lock.fl_flags = FL_POSIX;
3738 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3739 file_lock.fl_start = locku->lu_offset;
3741 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3742 nfs4_transform_lock_offset(&file_lock);
3745 * Try to unlock the file in the VFS.
3747 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3749 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3753 * OK, unlock succeeded; the only thing left to do is update the stateid.
3755 update_stateid(&stp->st_stateid);
3756 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3759 if (locku->lu_stateowner) {
3760 nfs4_get_stateowner(locku->lu_stateowner);
3761 cstate->replay_owner = locku->lu_stateowner;
3763 nfs4_unlock_state();
3767 status = nfserrno(err);
3773 * 1: locks held by lockowner
3774 * 0: no locks held by lockowner
3777 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3779 struct file_lock **flpp;
3780 struct inode *inode = filp->f_path.dentry->d_inode;
3784 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3785 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3796 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3797 struct nfsd4_compound_state *cstate,
3798 struct nfsd4_release_lockowner *rlockowner)
3800 clientid_t *clid = &rlockowner->rl_clientid;
3801 struct nfs4_stateowner *sop;
3802 struct nfs4_stateid *stp;
3803 struct xdr_netobj *owner = &rlockowner->rl_owner;
3804 struct list_head matches;
3808 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3809 clid->cl_boot, clid->cl_id);
3811 /* XXX check for lease expiration */
3813 status = nfserr_stale_clientid;
3814 if (STALE_CLIENTID(clid))
3819 status = nfserr_locks_held;
3820 /* XXX: we're doing a linear search through all the lockowners.
3821 * Yipes! For now we'll just hope clients aren't really using
3822 * release_lockowner much, but eventually we have to fix these
3823 * data structures. */
3824 INIT_LIST_HEAD(&matches);
3825 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3826 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3827 if (!same_owner_str(sop, owner, clid))
3829 list_for_each_entry(stp, &sop->so_stateids,
3831 if (check_for_locks(stp->st_vfs_file, sop))
3833 /* Note: so_perclient unused for lockowners,
3834 * so it's OK to fool with here. */
3835 list_add(&sop->so_perclient, &matches);
3839 /* Clients probably won't expect us to return with some (but not all)
3840 * of the lockowner state released; so don't release any until all
3841 * have been checked. */
3843 while (!list_empty(&matches)) {
3844 sop = list_entry(matches.next, struct nfs4_stateowner,
3846 /* unhash_stateowner deletes so_perclient only
3847 * for openowners. */
3848 list_del(&sop->so_perclient);
3849 release_lockowner(sop);
3852 nfs4_unlock_state();
3856 static inline struct nfs4_client_reclaim *
3859 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3863 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3865 unsigned int strhashval = clientstr_hashval(name);
3866 struct nfs4_client *clp;
3868 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3873 * failure => all reset bets are off, nfserr_no_grace...
3876 nfs4_client_to_reclaim(const char *name)
3878 unsigned int strhashval;
3879 struct nfs4_client_reclaim *crp = NULL;
3881 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3882 crp = alloc_reclaim();
3885 strhashval = clientstr_hashval(name);
3886 INIT_LIST_HEAD(&crp->cr_strhash);
3887 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3888 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3889 reclaim_str_hashtbl_size++;
3894 nfs4_release_reclaim(void)
3896 struct nfs4_client_reclaim *crp = NULL;
3899 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3900 while (!list_empty(&reclaim_str_hashtbl[i])) {
3901 crp = list_entry(reclaim_str_hashtbl[i].next,
3902 struct nfs4_client_reclaim, cr_strhash);
3903 list_del(&crp->cr_strhash);
3905 reclaim_str_hashtbl_size--;
3908 BUG_ON(reclaim_str_hashtbl_size);
3912 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3913 static struct nfs4_client_reclaim *
3914 nfs4_find_reclaim_client(clientid_t *clid)
3916 unsigned int strhashval;
3917 struct nfs4_client *clp;
3918 struct nfs4_client_reclaim *crp = NULL;
3921 /* find clientid in conf_id_hashtbl */
3922 clp = find_confirmed_client(clid);
3926 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3927 clp->cl_name.len, clp->cl_name.data,
3930 /* find clp->cl_name in reclaim_str_hashtbl */
3931 strhashval = clientstr_hashval(clp->cl_recdir);
3932 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3933 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3941 * Called from OPEN. Look for clientid in reclaim list.
3944 nfs4_check_open_reclaim(clientid_t *clid)
3946 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3949 /* initialization to perform at module load time: */
3952 nfs4_state_init(void)
3956 status = nfsd4_init_slabs();
3959 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3960 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3961 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3962 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3963 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3964 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3966 for (i = 0; i < SESSION_HASH_SIZE; i++)
3967 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3968 for (i = 0; i < FILE_HASH_SIZE; i++) {
3969 INIT_LIST_HEAD(&file_hashtbl[i]);
3971 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3972 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3973 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3975 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3976 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3977 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3979 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3980 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3981 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3983 memset(&onestateid, ~0, sizeof(stateid_t));
3984 INIT_LIST_HEAD(&close_lru);
3985 INIT_LIST_HEAD(&client_lru);
3986 INIT_LIST_HEAD(&del_recall_lru);
3987 reclaim_str_hashtbl_size = 0;
3992 nfsd4_load_reboot_recovery_data(void)
3997 nfsd4_init_recdir(user_recovery_dirname);
3998 status = nfsd4_recdir_load();
3999 nfs4_unlock_state();
4001 printk("NFSD: Failure reading reboot recovery data\n");
4005 get_nfs4_grace_period(void)
4007 return max(user_lease_time, lease_time) * HZ;
4011 * Since the lifetime of a delegation isn't limited to that of an open, a
4012 * client may quite reasonably hang on to a delegation as long as it has
4013 * the inode cached. This becomes an obvious problem the first time a
4014 * client's inode cache approaches the size of the server's total memory.
4016 * For now we avoid this problem by imposing a hard limit on the number
4017 * of delegations, which varies according to the server's memory size.
4020 set_max_delegations(void)
4023 * Allow at most 4 delegations per megabyte of RAM. Quick
4024 * estimates suggest that in the worst case (where every delegation
4025 * is for a different inode), a delegation could take about 1.5K,
4026 * giving a worst case usage of about 6% of memory.
4028 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4031 /* initialization to perform when the nfsd service is started: */
4034 __nfs4_state_start(void)
4036 unsigned long grace_time;
4038 boot_time = get_seconds();
4039 grace_time = get_nfs4_grace_period();
4040 lease_time = user_lease_time;
4041 locks_start_grace(&nfsd4_manager);
4042 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4044 laundry_wq = create_singlethread_workqueue("nfsd4");
4045 if (laundry_wq == NULL)
4047 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4048 set_max_delegations();
4049 return set_callback_cred();
4053 nfs4_state_start(void)
4059 nfsd4_load_reboot_recovery_data();
4060 ret = __nfs4_state_start();
4068 nfs4_lease_time(void)
4074 __nfs4_state_shutdown(void)
4077 struct nfs4_client *clp = NULL;
4078 struct nfs4_delegation *dp = NULL;
4079 struct list_head *pos, *next, reaplist;
4081 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4082 while (!list_empty(&conf_id_hashtbl[i])) {
4083 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4086 while (!list_empty(&unconf_str_hashtbl[i])) {
4087 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4091 INIT_LIST_HEAD(&reaplist);
4092 spin_lock(&recall_lock);
4093 list_for_each_safe(pos, next, &del_recall_lru) {
4094 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4095 list_move(&dp->dl_recall_lru, &reaplist);
4097 spin_unlock(&recall_lock);
4098 list_for_each_safe(pos, next, &reaplist) {
4099 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4100 list_del_init(&dp->dl_recall_lru);
4101 unhash_delegation(dp);
4104 nfsd4_shutdown_recdir();
4109 nfs4_state_shutdown(void)
4111 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4112 destroy_workqueue(laundry_wq);
4113 locks_end_grace(&nfsd4_manager);
4115 nfs4_release_reclaim();
4116 __nfs4_state_shutdown();
4117 nfs4_unlock_state();
4121 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4122 * accessed when nfsd is starting.
4125 nfs4_set_recdir(char *recdir)
4127 strcpy(user_recovery_dirname, recdir);
4131 * Change the NFSv4 recovery directory to recdir.
4134 nfs4_reset_recoverydir(char *recdir)
4139 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4143 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4144 nfs4_set_recdir(recdir);
4152 nfs4_recoverydir(void)
4154 return user_recovery_dirname;
4158 * Called when leasetime is changed.
4160 * The only way the protocol gives us to handle on-the-fly lease changes is to
4161 * simulate a reboot. Instead of doing that, we just wait till the next time
4162 * we start to register any changes in lease time. If the administrator
4163 * really wants to change the lease time *now*, they can go ahead and bring
4164 * nfsd down and then back up again after changing the lease time.
4166 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4167 * when nfsd is starting
4170 nfs4_reset_lease(time_t leasetime)
4172 user_lease_time = leasetime;