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/file.h>
38 #include <linux/smp_lock.h>
39 #include <linux/namei.h>
40 #include <linux/swap.h>
41 #include <linux/sunrpc/svcauth_gss.h>
42 #include <linux/sunrpc/clnt.h>
46 #define NFSDDBG_FACILITY NFSDDBG_PROC
49 static time_t lease_time = 90; /* default lease time */
50 static time_t user_lease_time = 90;
51 static time_t boot_time;
52 static u32 current_ownerid = 1;
53 static u32 current_fileid = 1;
54 static u32 current_delegid = 1;
56 static stateid_t zerostateid; /* bits all 0 */
57 static stateid_t onestateid; /* bits all 1 */
58 static u64 current_sessionid = 1;
60 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
61 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
63 /* forward declarations */
64 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
65 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
66 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
67 static void nfs4_set_recdir(char *recdir);
71 /* Currently used for almost all code touching nfsv4 state: */
72 static DEFINE_MUTEX(client_mutex);
75 * Currently used for the del_recall_lru and file hash table. In an
76 * effort to decrease the scope of the client_mutex, this spinlock may
77 * eventually cover more:
79 static DEFINE_SPINLOCK(recall_lock);
81 static struct kmem_cache *stateowner_slab = NULL;
82 static struct kmem_cache *file_slab = NULL;
83 static struct kmem_cache *stateid_slab = NULL;
84 static struct kmem_cache *deleg_slab = NULL;
89 mutex_lock(&client_mutex);
93 nfs4_unlock_state(void)
95 mutex_unlock(&client_mutex);
99 opaque_hashval(const void *ptr, int nbytes)
101 unsigned char *cptr = (unsigned char *) ptr;
111 static struct list_head del_recall_lru;
114 put_nfs4_file(struct nfs4_file *fi)
116 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
117 list_del(&fi->fi_hash);
118 spin_unlock(&recall_lock);
120 kmem_cache_free(file_slab, fi);
125 get_nfs4_file(struct nfs4_file *fi)
127 atomic_inc(&fi->fi_ref);
130 static int num_delegations;
131 unsigned int max_delegations;
134 * Open owner state (share locks)
137 /* hash tables for nfs4_stateowner */
138 #define OWNER_HASH_BITS 8
139 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
140 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
142 #define ownerid_hashval(id) \
143 ((id) & OWNER_HASH_MASK)
144 #define ownerstr_hashval(clientid, ownername) \
145 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
147 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
148 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
150 /* hash table for nfs4_file */
151 #define FILE_HASH_BITS 8
152 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
153 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
154 /* hash table for (open)nfs4_stateid */
155 #define STATEID_HASH_BITS 10
156 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
157 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
159 #define file_hashval(x) \
160 hash_ptr(x, FILE_HASH_BITS)
161 #define stateid_hashval(owner_id, file_id) \
162 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
164 static struct list_head file_hashtbl[FILE_HASH_SIZE];
165 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
167 static struct nfs4_delegation *
168 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
170 struct nfs4_delegation *dp;
171 struct nfs4_file *fp = stp->st_file;
172 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
174 dprintk("NFSD alloc_init_deleg\n");
175 if (fp->fi_had_conflict)
177 if (num_delegations > max_delegations)
179 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
183 INIT_LIST_HEAD(&dp->dl_perfile);
184 INIT_LIST_HEAD(&dp->dl_perclnt);
185 INIT_LIST_HEAD(&dp->dl_recall_lru);
190 get_file(stp->st_vfs_file);
191 dp->dl_vfs_file = stp->st_vfs_file;
193 dp->dl_ident = cb->cb_ident;
194 dp->dl_stateid.si_boot = get_seconds();
195 dp->dl_stateid.si_stateownerid = current_delegid++;
196 dp->dl_stateid.si_fileid = 0;
197 dp->dl_stateid.si_generation = 0;
198 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
200 atomic_set(&dp->dl_count, 1);
201 list_add(&dp->dl_perfile, &fp->fi_delegations);
202 list_add(&dp->dl_perclnt, &clp->cl_delegations);
207 nfs4_put_delegation(struct nfs4_delegation *dp)
209 if (atomic_dec_and_test(&dp->dl_count)) {
210 dprintk("NFSD: freeing dp %p\n",dp);
211 put_nfs4_file(dp->dl_file);
212 kmem_cache_free(deleg_slab, dp);
217 /* Remove the associated file_lock first, then remove the delegation.
218 * lease_modify() is called to remove the FS_LEASE file_lock from
219 * the i_flock list, eventually calling nfsd's lock_manager
220 * fl_release_callback.
223 nfs4_close_delegation(struct nfs4_delegation *dp)
225 struct file *filp = dp->dl_vfs_file;
227 dprintk("NFSD: close_delegation dp %p\n",dp);
228 dp->dl_vfs_file = NULL;
229 /* The following nfsd_close may not actually close the file,
230 * but we want to remove the lease in any case. */
232 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
236 /* Called under the state lock. */
238 unhash_delegation(struct nfs4_delegation *dp)
240 list_del_init(&dp->dl_perfile);
241 list_del_init(&dp->dl_perclnt);
242 spin_lock(&recall_lock);
243 list_del_init(&dp->dl_recall_lru);
244 spin_unlock(&recall_lock);
245 nfs4_close_delegation(dp);
246 nfs4_put_delegation(dp);
253 /* Hash tables for nfs4_clientid state */
254 #define CLIENT_HASH_BITS 4
255 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
256 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
258 #define clientid_hashval(id) \
259 ((id) & CLIENT_HASH_MASK)
260 #define clientstr_hashval(name) \
261 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
263 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
264 * used in reboot/reset lease grace period processing
266 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
267 * setclientid_confirmed info.
269 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
272 * client_lru holds client queue ordered by nfs4_client.cl_time
275 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
276 * for last close replay.
278 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
279 static int reclaim_str_hashtbl_size = 0;
280 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
281 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
282 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
283 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
284 static struct list_head client_lru;
285 static struct list_head close_lru;
287 static void unhash_generic_stateid(struct nfs4_stateid *stp)
289 list_del(&stp->st_hash);
290 list_del(&stp->st_perfile);
291 list_del(&stp->st_perstateowner);
294 static void free_generic_stateid(struct nfs4_stateid *stp)
296 put_nfs4_file(stp->st_file);
297 kmem_cache_free(stateid_slab, stp);
300 static void release_lock_stateid(struct nfs4_stateid *stp)
302 unhash_generic_stateid(stp);
303 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
304 free_generic_stateid(stp);
307 static void unhash_lockowner(struct nfs4_stateowner *sop)
309 struct nfs4_stateid *stp;
311 list_del(&sop->so_idhash);
312 list_del(&sop->so_strhash);
313 list_del(&sop->so_perstateid);
314 while (!list_empty(&sop->so_stateids)) {
315 stp = list_first_entry(&sop->so_stateids,
316 struct nfs4_stateid, st_perstateowner);
317 release_lock_stateid(stp);
321 static void release_lockowner(struct nfs4_stateowner *sop)
323 unhash_lockowner(sop);
324 nfs4_put_stateowner(sop);
328 release_stateid_lockowners(struct nfs4_stateid *open_stp)
330 struct nfs4_stateowner *lock_sop;
332 while (!list_empty(&open_stp->st_lockowners)) {
333 lock_sop = list_entry(open_stp->st_lockowners.next,
334 struct nfs4_stateowner, so_perstateid);
335 /* list_del(&open_stp->st_lockowners); */
336 BUG_ON(lock_sop->so_is_open_owner);
337 release_lockowner(lock_sop);
341 static void release_open_stateid(struct nfs4_stateid *stp)
343 unhash_generic_stateid(stp);
344 release_stateid_lockowners(stp);
345 nfsd_close(stp->st_vfs_file);
346 free_generic_stateid(stp);
349 static void unhash_openowner(struct nfs4_stateowner *sop)
351 struct nfs4_stateid *stp;
353 list_del(&sop->so_idhash);
354 list_del(&sop->so_strhash);
355 list_del(&sop->so_perclient);
356 list_del(&sop->so_perstateid); /* XXX: necessary? */
357 while (!list_empty(&sop->so_stateids)) {
358 stp = list_first_entry(&sop->so_stateids,
359 struct nfs4_stateid, st_perstateowner);
360 release_open_stateid(stp);
364 static void release_openowner(struct nfs4_stateowner *sop)
366 unhash_openowner(sop);
367 list_del(&sop->so_close_lru);
368 nfs4_put_stateowner(sop);
371 static DEFINE_SPINLOCK(sessionid_lock);
372 #define SESSION_HASH_SIZE 512
373 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
376 hash_sessionid(struct nfs4_sessionid *sessionid)
378 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
380 return sid->sequence % SESSION_HASH_SIZE;
384 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
386 u32 *ptr = (u32 *)(&sessionid->data[0]);
387 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
391 gen_sessionid(struct nfsd4_session *ses)
393 struct nfs4_client *clp = ses->se_client;
394 struct nfsd4_sessionid *sid;
396 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
397 sid->clientid = clp->cl_clientid;
398 sid->sequence = current_sessionid++;
403 * The protocol defines ca_maxresponssize_cached to include the size of
404 * the rpc header, but all we need to cache is the data starting after
405 * the end of the initial SEQUENCE operation--the rest we regenerate
406 * each time. Therefore we can advertise a ca_maxresponssize_cached
407 * value that is the number of bytes in our cache plus a few additional
408 * bytes. In order to stay on the safe side, and not promise more than
409 * we can cache, those additional bytes must be the minimum possible: 24
410 * bytes of rpc header (xid through accept state, with AUTH_NULL
411 * verifier), 12 for the compound header (with zero-length tag), and 44
412 * for the SEQUENCE op response:
414 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
417 * Give the client the number of ca_maxresponsesize_cached slots it
418 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
419 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
420 * than NFSD_MAX_SLOTS_PER_SESSION.
422 * If we run out of reserved DRC memory we should (up to a point)
423 * re-negotiate active sessions and reduce their slot usage to make
424 * rooom for new connections. For now we just fail the create session.
426 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
428 int mem, size = fchan->maxresp_cached;
430 if (fchan->maxreqs < 1)
433 if (size < NFSD_MIN_HDR_SEQ_SZ)
434 size = NFSD_MIN_HDR_SEQ_SZ;
435 size -= NFSD_MIN_HDR_SEQ_SZ;
436 if (size > NFSD_SLOT_CACHE_SIZE)
437 size = NFSD_SLOT_CACHE_SIZE;
439 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
440 mem = fchan->maxreqs * size;
441 if (mem > NFSD_MAX_MEM_PER_SESSION) {
442 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
443 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
444 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
445 mem = fchan->maxreqs * size;
448 spin_lock(&nfsd_drc_lock);
449 /* bound the total session drc memory ussage */
450 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
451 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
452 mem = fchan->maxreqs * size;
454 nfsd_drc_mem_used += mem;
455 spin_unlock(&nfsd_drc_lock);
457 if (fchan->maxreqs == 0)
458 return nfserr_serverfault;
460 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
465 * fchan holds the client values on input, and the server values on output
466 * sv_max_mesg is the maximum payload plus one page for overhead.
468 static int init_forechannel_attrs(struct svc_rqst *rqstp,
469 struct nfsd4_channel_attrs *session_fchan,
470 struct nfsd4_channel_attrs *fchan)
473 __u32 maxcount = nfsd_serv->sv_max_mesg;
475 /* headerpadsz set to zero in encode routine */
477 /* Use the client's max request and max response size if possible */
478 if (fchan->maxreq_sz > maxcount)
479 fchan->maxreq_sz = maxcount;
480 session_fchan->maxreq_sz = fchan->maxreq_sz;
482 if (fchan->maxresp_sz > maxcount)
483 fchan->maxresp_sz = maxcount;
484 session_fchan->maxresp_sz = fchan->maxresp_sz;
486 /* Use the client's maxops if possible */
487 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
488 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
489 session_fchan->maxops = fchan->maxops;
491 /* FIXME: Error means no more DRC pages so the server should
492 * recover pages from existing sessions. For now fail session
495 status = set_forechannel_drc_size(fchan);
497 session_fchan->maxresp_cached = fchan->maxresp_cached;
498 session_fchan->maxreqs = fchan->maxreqs;
500 dprintk("%s status %d\n", __func__, status);
505 free_session_slots(struct nfsd4_session *ses)
509 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
510 kfree(ses->se_slots[i]);
514 * We don't actually need to cache the rpc and session headers, so we
515 * can allocate a little less for each slot:
517 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
519 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
523 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
524 struct nfsd4_create_session *cses)
526 struct nfsd4_session *new, tmp;
527 struct nfsd4_slot *sp;
528 int idx, slotsize, cachesize, i;
531 memset(&tmp, 0, sizeof(tmp));
533 /* FIXME: For now, we just accept the client back channel attributes. */
534 tmp.se_bchannel = cses->back_channel;
535 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
536 &cses->fore_channel);
540 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
541 + sizeof(struct nfsd4_session) > PAGE_SIZE);
543 status = nfserr_serverfault;
544 /* allocate struct nfsd4_session and slot table pointers in one piece */
545 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
546 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
550 memcpy(new, &tmp, sizeof(*new));
552 /* allocate each struct nfsd4_slot and data cache in one piece */
553 cachesize = slot_bytes(&new->se_fchannel);
554 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
555 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
558 new->se_slots[i] = sp;
561 new->se_client = clp;
563 idx = hash_sessionid(&new->se_sessionid);
564 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
565 NFS4_MAX_SESSIONID_LEN);
567 new->se_flags = cses->flags;
568 kref_init(&new->se_ref);
569 spin_lock(&sessionid_lock);
570 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
571 list_add(&new->se_perclnt, &clp->cl_sessions);
572 spin_unlock(&sessionid_lock);
578 free_session_slots(new);
583 /* caller must hold sessionid_lock */
584 static struct nfsd4_session *
585 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
587 struct nfsd4_session *elem;
590 dump_sessionid(__func__, sessionid);
591 idx = hash_sessionid(sessionid);
592 dprintk("%s: idx is %d\n", __func__, idx);
593 /* Search in the appropriate list */
594 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
595 dump_sessionid("list traversal", &elem->se_sessionid);
596 if (!memcmp(elem->se_sessionid.data, sessionid->data,
597 NFS4_MAX_SESSIONID_LEN)) {
602 dprintk("%s: session not found\n", __func__);
606 /* caller must hold sessionid_lock */
608 unhash_session(struct nfsd4_session *ses)
610 list_del(&ses->se_hash);
611 list_del(&ses->se_perclnt);
615 release_session(struct nfsd4_session *ses)
617 spin_lock(&sessionid_lock);
619 spin_unlock(&sessionid_lock);
620 nfsd4_put_session(ses);
624 free_session(struct kref *kref)
626 struct nfsd4_session *ses;
629 ses = container_of(kref, struct nfsd4_session, se_ref);
630 spin_lock(&nfsd_drc_lock);
631 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
632 nfsd_drc_mem_used -= mem;
633 spin_unlock(&nfsd_drc_lock);
634 free_session_slots(ses);
639 renew_client(struct nfs4_client *clp)
642 * Move client to the end to the LRU list.
644 dprintk("renewing client (clientid %08x/%08x)\n",
645 clp->cl_clientid.cl_boot,
646 clp->cl_clientid.cl_id);
647 list_move_tail(&clp->cl_lru, &client_lru);
648 clp->cl_time = get_seconds();
651 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
653 STALE_CLIENTID(clientid_t *clid)
655 if (clid->cl_boot == boot_time)
657 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
658 clid->cl_boot, clid->cl_id, boot_time);
663 * XXX Should we use a slab cache ?
664 * This type of memory management is somewhat inefficient, but we use it
665 * anyway since SETCLIENTID is not a common operation.
667 static struct nfs4_client *alloc_client(struct xdr_netobj name)
669 struct nfs4_client *clp;
671 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
674 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
675 if (clp->cl_name.data == NULL) {
679 memcpy(clp->cl_name.data, name.data, name.len);
680 clp->cl_name.len = name.len;
685 shutdown_callback_client(struct nfs4_client *clp)
687 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
691 * Callback threads take a reference on the client, so there
692 * should be no outstanding callbacks at this point.
694 clp->cl_cb_conn.cb_client = NULL;
695 rpc_shutdown_client(clnt);
700 free_client(struct nfs4_client *clp)
702 shutdown_callback_client(clp);
704 svc_xprt_put(clp->cl_cb_xprt);
705 if (clp->cl_cred.cr_group_info)
706 put_group_info(clp->cl_cred.cr_group_info);
707 kfree(clp->cl_principal);
708 kfree(clp->cl_name.data);
713 put_nfs4_client(struct nfs4_client *clp)
715 if (atomic_dec_and_test(&clp->cl_count))
720 expire_client(struct nfs4_client *clp)
722 struct nfs4_stateowner *sop;
723 struct nfs4_delegation *dp;
724 struct list_head reaplist;
726 dprintk("NFSD: expire_client cl_count %d\n",
727 atomic_read(&clp->cl_count));
729 INIT_LIST_HEAD(&reaplist);
730 spin_lock(&recall_lock);
731 while (!list_empty(&clp->cl_delegations)) {
732 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
733 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
735 list_del_init(&dp->dl_perclnt);
736 list_move(&dp->dl_recall_lru, &reaplist);
738 spin_unlock(&recall_lock);
739 while (!list_empty(&reaplist)) {
740 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
741 list_del_init(&dp->dl_recall_lru);
742 unhash_delegation(dp);
744 list_del(&clp->cl_idhash);
745 list_del(&clp->cl_strhash);
746 list_del(&clp->cl_lru);
747 while (!list_empty(&clp->cl_openowners)) {
748 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
749 release_openowner(sop);
751 while (!list_empty(&clp->cl_sessions)) {
752 struct nfsd4_session *ses;
753 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
755 release_session(ses);
757 put_nfs4_client(clp);
760 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
762 memcpy(target->cl_verifier.data, source->data,
763 sizeof(target->cl_verifier.data));
766 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
768 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
769 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
772 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
774 target->cr_uid = source->cr_uid;
775 target->cr_gid = source->cr_gid;
776 target->cr_group_info = source->cr_group_info;
777 get_group_info(target->cr_group_info);
780 static int same_name(const char *n1, const char *n2)
782 return 0 == memcmp(n1, n2, HEXDIR_LEN);
786 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
788 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
792 same_clid(clientid_t *cl1, clientid_t *cl2)
794 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
797 /* XXX what about NGROUP */
799 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
801 return cr1->cr_uid == cr2->cr_uid;
804 static void gen_clid(struct nfs4_client *clp)
806 static u32 current_clientid = 1;
808 clp->cl_clientid.cl_boot = boot_time;
809 clp->cl_clientid.cl_id = current_clientid++;
812 static void gen_confirm(struct nfs4_client *clp)
817 p = (u32 *)clp->cl_confirm.data;
818 *p++ = get_seconds();
822 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
823 struct svc_rqst *rqstp, nfs4_verifier *verf)
825 struct nfs4_client *clp;
826 struct sockaddr *sa = svc_addr(rqstp);
829 clp = alloc_client(name);
833 princ = svc_gss_principal(rqstp);
835 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
836 if (clp->cl_principal == NULL) {
842 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
843 atomic_set(&clp->cl_count, 1);
844 atomic_set(&clp->cl_cb_conn.cb_set, 0);
845 INIT_LIST_HEAD(&clp->cl_idhash);
846 INIT_LIST_HEAD(&clp->cl_strhash);
847 INIT_LIST_HEAD(&clp->cl_openowners);
848 INIT_LIST_HEAD(&clp->cl_delegations);
849 INIT_LIST_HEAD(&clp->cl_sessions);
850 INIT_LIST_HEAD(&clp->cl_lru);
851 clear_bit(0, &clp->cl_cb_slot_busy);
852 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
853 copy_verf(clp, verf);
854 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
855 clp->cl_flavor = rqstp->rq_flavor;
856 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
862 static int check_name(struct xdr_netobj name)
866 if (name.len > NFS4_OPAQUE_LIMIT) {
867 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
874 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
876 unsigned int idhashval;
878 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
879 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
880 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
881 list_add_tail(&clp->cl_lru, &client_lru);
882 clp->cl_time = get_seconds();
886 move_to_confirmed(struct nfs4_client *clp)
888 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
889 unsigned int strhashval;
891 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
892 list_del_init(&clp->cl_strhash);
893 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
894 strhashval = clientstr_hashval(clp->cl_recdir);
895 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
899 static struct nfs4_client *
900 find_confirmed_client(clientid_t *clid)
902 struct nfs4_client *clp;
903 unsigned int idhashval = clientid_hashval(clid->cl_id);
905 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
906 if (same_clid(&clp->cl_clientid, clid))
912 static struct nfs4_client *
913 find_unconfirmed_client(clientid_t *clid)
915 struct nfs4_client *clp;
916 unsigned int idhashval = clientid_hashval(clid->cl_id);
918 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
919 if (same_clid(&clp->cl_clientid, clid))
926 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
927 * parameter. Matching is based on the fact the at least one of the
928 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
930 * FIXME: we need to unify the clientid namespaces for nfsv4.x
931 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
932 * and SET_CLIENTID{,_CONFIRM}
935 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
937 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
938 return use_exchange_id == has_exchange_flags;
941 static struct nfs4_client *
942 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
943 bool use_exchange_id)
945 struct nfs4_client *clp;
947 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
948 if (same_name(clp->cl_recdir, dname) &&
949 match_clientid_establishment(clp, use_exchange_id))
955 static struct nfs4_client *
956 find_unconfirmed_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, &unconf_str_hashtbl[hashval], cl_strhash) {
962 if (same_name(clp->cl_recdir, dname) &&
963 match_clientid_establishment(clp, use_exchange_id))
970 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
972 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
973 unsigned short expected_family;
975 /* Currently, we only support tcp and tcp6 for the callback channel */
976 if (se->se_callback_netid_len == 3 &&
977 !memcmp(se->se_callback_netid_val, "tcp", 3))
978 expected_family = AF_INET;
979 else if (se->se_callback_netid_len == 4 &&
980 !memcmp(se->se_callback_netid_val, "tcp6", 4))
981 expected_family = AF_INET6;
985 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
986 se->se_callback_addr_len,
987 (struct sockaddr *) &cb->cb_addr,
988 sizeof(cb->cb_addr));
990 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
993 if (cb->cb_addr.ss_family == AF_INET6)
994 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
996 cb->cb_minorversion = 0;
997 cb->cb_prog = se->se_callback_prog;
998 cb->cb_ident = se->se_callback_ident;
1001 cb->cb_addr.ss_family = AF_UNSPEC;
1003 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1004 "will not receive delegations\n",
1005 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1011 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1014 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1016 struct nfsd4_slot *slot = resp->cstate.slot;
1019 dprintk("--> %s slot %p\n", __func__, slot);
1021 slot->sl_opcnt = resp->opcnt;
1022 slot->sl_status = resp->cstate.status;
1024 if (nfsd4_not_cached(resp)) {
1025 slot->sl_datalen = 0;
1028 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1029 base = (char *)resp->cstate.datap -
1030 (char *)resp->xbuf->head[0].iov_base;
1031 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1033 WARN("%s: sessions DRC could not cache compound\n", __func__);
1038 * Encode the replay sequence operation from the slot values.
1039 * If cachethis is FALSE encode the uncached rep error on the next
1040 * operation which sets resp->p and increments resp->opcnt for
1041 * nfs4svc_encode_compoundres.
1045 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1046 struct nfsd4_compoundres *resp)
1048 struct nfsd4_op *op;
1049 struct nfsd4_slot *slot = resp->cstate.slot;
1051 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1052 resp->opcnt, resp->cstate.slot->sl_cachethis);
1054 /* Encode the replayed sequence operation */
1055 op = &args->ops[resp->opcnt - 1];
1056 nfsd4_encode_operation(resp, op);
1058 /* Return nfserr_retry_uncached_rep in next operation. */
1059 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1060 op = &args->ops[resp->opcnt++];
1061 op->status = nfserr_retry_uncached_rep;
1062 nfsd4_encode_operation(resp, op);
1068 * The sequence operation is not cached because we can use the slot and
1072 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1073 struct nfsd4_sequence *seq)
1075 struct nfsd4_slot *slot = resp->cstate.slot;
1078 dprintk("--> %s slot %p\n", __func__, slot);
1080 /* Either returns 0 or nfserr_retry_uncached */
1081 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1082 if (status == nfserr_retry_uncached_rep)
1085 /* The sequence operation has been encoded, cstate->datap set. */
1086 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1088 resp->opcnt = slot->sl_opcnt;
1089 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1090 status = slot->sl_status;
1096 * Set the exchange_id flags returned by the server.
1099 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1101 /* pNFS is not supported */
1102 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1104 /* Referrals are supported, Migration is not. */
1105 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1107 /* set the wire flags to return to client. */
1108 clid->flags = new->cl_exchange_flags;
1112 nfsd4_exchange_id(struct svc_rqst *rqstp,
1113 struct nfsd4_compound_state *cstate,
1114 struct nfsd4_exchange_id *exid)
1116 struct nfs4_client *unconf, *conf, *new;
1118 unsigned int strhashval;
1119 char dname[HEXDIR_LEN];
1120 char addr_str[INET6_ADDRSTRLEN];
1121 nfs4_verifier verf = exid->verifier;
1122 struct sockaddr *sa = svc_addr(rqstp);
1124 rpc_ntop(sa, addr_str, sizeof(addr_str));
1125 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1126 "ip_addr=%s flags %x, spa_how %d\n",
1127 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1128 addr_str, exid->flags, exid->spa_how);
1130 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1131 return nfserr_inval;
1133 /* Currently only support SP4_NONE */
1134 switch (exid->spa_how) {
1138 return nfserr_encr_alg_unsupp;
1140 BUG(); /* checked by xdr code */
1142 return nfserr_serverfault; /* no excuse :-/ */
1145 status = nfs4_make_rec_clidname(dname, &exid->clname);
1150 strhashval = clientstr_hashval(dname);
1155 conf = find_confirmed_client_by_str(dname, strhashval, true);
1157 if (!same_verf(&verf, &conf->cl_verifier)) {
1158 /* 18.35.4 case 8 */
1159 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1160 status = nfserr_not_same;
1163 /* Client reboot: destroy old state */
1164 expire_client(conf);
1167 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1168 /* 18.35.4 case 9 */
1169 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1170 status = nfserr_perm;
1173 expire_client(conf);
1177 * Set bit when the owner id and verifier map to an already
1178 * confirmed client id (18.35.3).
1180 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1183 * Falling into 18.35.4 case 2, possible router replay.
1184 * Leave confirmed record intact and return same result.
1186 copy_verf(conf, &verf);
1191 /* 18.35.4 case 7 */
1192 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1193 status = nfserr_noent;
1197 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1200 * Possible retry or client restart. Per 18.35.4 case 4,
1201 * a new unconfirmed record should be generated regardless
1202 * of whether any properties have changed.
1204 expire_client(unconf);
1209 new = create_client(exid->clname, dname, rqstp, &verf);
1211 status = nfserr_serverfault;
1216 add_to_unconfirmed(new, strhashval);
1218 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1219 exid->clientid.cl_id = new->cl_clientid.cl_id;
1222 nfsd4_set_ex_flags(new, exid);
1224 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1225 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1229 nfs4_unlock_state();
1231 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1236 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1238 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1241 /* The slot is in use, and no response has been sent. */
1243 if (seqid == slot_seqid)
1244 return nfserr_jukebox;
1246 return nfserr_seq_misordered;
1249 if (likely(seqid == slot_seqid + 1))
1252 if (seqid == slot_seqid)
1253 return nfserr_replay_cache;
1255 if (seqid == 1 && (slot_seqid + 1) == 0)
1257 /* Misordered replay or misordered new request */
1258 return nfserr_seq_misordered;
1262 * Cache the create session result into the create session single DRC
1263 * slot cache by saving the xdr structure. sl_seqid has been set.
1264 * Do this for solo or embedded create session operations.
1267 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1268 struct nfsd4_clid_slot *slot, int nfserr)
1270 slot->sl_status = nfserr;
1271 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1275 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1276 struct nfsd4_clid_slot *slot)
1278 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1279 return slot->sl_status;
1283 nfsd4_create_session(struct svc_rqst *rqstp,
1284 struct nfsd4_compound_state *cstate,
1285 struct nfsd4_create_session *cr_ses)
1287 struct sockaddr *sa = svc_addr(rqstp);
1288 struct nfs4_client *conf, *unconf;
1289 struct nfsd4_clid_slot *cs_slot = NULL;
1293 unconf = find_unconfirmed_client(&cr_ses->clientid);
1294 conf = find_confirmed_client(&cr_ses->clientid);
1297 cs_slot = &conf->cl_cs_slot;
1298 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1299 if (status == nfserr_replay_cache) {
1300 dprintk("Got a create_session replay! seqid= %d\n",
1302 /* Return the cached reply status */
1303 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1305 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1306 status = nfserr_seq_misordered;
1307 dprintk("Sequence misordered!\n");
1308 dprintk("Expected seqid= %d but got seqid= %d\n",
1309 cs_slot->sl_seqid, cr_ses->seqid);
1312 cs_slot->sl_seqid++;
1313 } else if (unconf) {
1314 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1315 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1316 status = nfserr_clid_inuse;
1320 cs_slot = &unconf->cl_cs_slot;
1321 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1323 /* an unconfirmed replay returns misordered */
1324 status = nfserr_seq_misordered;
1328 cs_slot->sl_seqid++; /* from 0 to 1 */
1329 move_to_confirmed(unconf);
1332 * We do not support RDMA or persistent sessions
1334 cr_ses->flags &= ~SESSION4_PERSIST;
1335 cr_ses->flags &= ~SESSION4_RDMA;
1337 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1338 unconf->cl_cb_xprt = rqstp->rq_xprt;
1339 svc_xprt_get(unconf->cl_cb_xprt);
1341 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1343 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1344 unconf->cl_cb_conn.cb_minorversion =
1345 cstate->minorversion;
1346 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1347 unconf->cl_cb_seq_nr = 1;
1348 nfsd4_probe_callback(unconf);
1352 status = nfserr_stale_clientid;
1356 status = alloc_init_session(rqstp, conf, cr_ses);
1360 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1361 NFS4_MAX_SESSIONID_LEN);
1362 cr_ses->seqid = cs_slot->sl_seqid;
1365 /* cache solo and embedded create sessions under the state lock */
1366 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1368 nfs4_unlock_state();
1369 dprintk("%s returns %d\n", __func__, ntohl(status));
1374 nfsd4_destroy_session(struct svc_rqst *r,
1375 struct nfsd4_compound_state *cstate,
1376 struct nfsd4_destroy_session *sessionid)
1378 struct nfsd4_session *ses;
1379 u32 status = nfserr_badsession;
1382 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1383 * - Should we return nfserr_back_chan_busy if waiting for
1384 * callbacks on to-be-destroyed session?
1385 * - Do we need to clear any callback info from previous session?
1388 dump_sessionid(__func__, &sessionid->sessionid);
1389 spin_lock(&sessionid_lock);
1390 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1392 spin_unlock(&sessionid_lock);
1396 unhash_session(ses);
1397 spin_unlock(&sessionid_lock);
1399 /* wait for callbacks */
1400 shutdown_callback_client(ses->se_client);
1401 nfsd4_put_session(ses);
1404 dprintk("%s returns %d\n", __func__, ntohl(status));
1409 nfsd4_sequence(struct svc_rqst *rqstp,
1410 struct nfsd4_compound_state *cstate,
1411 struct nfsd4_sequence *seq)
1413 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1414 struct nfsd4_session *session;
1415 struct nfsd4_slot *slot;
1418 if (resp->opcnt != 1)
1419 return nfserr_sequence_pos;
1421 spin_lock(&sessionid_lock);
1422 status = nfserr_badsession;
1423 session = find_in_sessionid_hashtbl(&seq->sessionid);
1427 status = nfserr_badslot;
1428 if (seq->slotid >= session->se_fchannel.maxreqs)
1431 slot = session->se_slots[seq->slotid];
1432 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1434 /* We do not negotiate the number of slots yet, so set the
1435 * maxslots to the session maxreqs which is used to encode
1436 * sr_highest_slotid and the sr_target_slot id to maxslots */
1437 seq->maxslots = session->se_fchannel.maxreqs;
1439 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1440 if (status == nfserr_replay_cache) {
1441 cstate->slot = slot;
1442 cstate->session = session;
1443 /* Return the cached reply status and set cstate->status
1444 * for nfsd4_proc_compound processing */
1445 status = nfsd4_replay_cache_entry(resp, seq);
1446 cstate->status = nfserr_replay_cache;
1452 /* Success! bump slot seqid */
1453 slot->sl_inuse = true;
1454 slot->sl_seqid = seq->seqid;
1455 slot->sl_cachethis = seq->cachethis;
1457 cstate->slot = slot;
1458 cstate->session = session;
1460 /* Hold a session reference until done processing the compound:
1461 * nfsd4_put_session called only if the cstate slot is set.
1463 nfsd4_get_session(session);
1465 spin_unlock(&sessionid_lock);
1466 /* Renew the clientid on success and on replay */
1467 if (cstate->session) {
1469 renew_client(session->se_client);
1470 nfs4_unlock_state();
1472 dprintk("%s: return %d\n", __func__, ntohl(status));
1477 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1478 struct nfsd4_setclientid *setclid)
1480 struct sockaddr *sa = svc_addr(rqstp);
1481 struct xdr_netobj clname = {
1482 .len = setclid->se_namelen,
1483 .data = setclid->se_name,
1485 nfs4_verifier clverifier = setclid->se_verf;
1486 unsigned int strhashval;
1487 struct nfs4_client *conf, *unconf, *new;
1489 char dname[HEXDIR_LEN];
1491 if (!check_name(clname))
1492 return nfserr_inval;
1494 status = nfs4_make_rec_clidname(dname, &clname);
1499 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1500 * We get here on a DRC miss.
1503 strhashval = clientstr_hashval(dname);
1506 conf = find_confirmed_client_by_str(dname, strhashval, false);
1508 /* RFC 3530 14.2.33 CASE 0: */
1509 status = nfserr_clid_inuse;
1510 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1511 char addr_str[INET6_ADDRSTRLEN];
1512 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1514 dprintk("NFSD: setclientid: string in use by client "
1515 "at %s\n", addr_str);
1520 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1521 * has a description of SETCLIENTID request processing consisting
1522 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1524 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1525 status = nfserr_resource;
1528 * RFC 3530 14.2.33 CASE 4:
1529 * placed first, because it is the normal case
1532 expire_client(unconf);
1533 new = create_client(clname, dname, rqstp, &clverifier);
1537 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1539 * RFC 3530 14.2.33 CASE 1:
1540 * probable callback update
1543 /* Note this is removing unconfirmed {*x***},
1544 * which is stronger than RFC recommended {vxc**}.
1545 * This has the advantage that there is at most
1546 * one {*x***} in either list at any time.
1548 expire_client(unconf);
1550 new = create_client(clname, dname, rqstp, &clverifier);
1553 copy_clid(new, conf);
1554 } else if (!unconf) {
1556 * RFC 3530 14.2.33 CASE 2:
1557 * probable client reboot; state will be removed if
1560 new = create_client(clname, dname, rqstp, &clverifier);
1566 * RFC 3530 14.2.33 CASE 3:
1567 * probable client reboot; state will be removed if
1570 expire_client(unconf);
1571 new = create_client(clname, dname, rqstp, &clverifier);
1576 gen_callback(new, setclid, rpc_get_scope_id(sa));
1577 add_to_unconfirmed(new, strhashval);
1578 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1579 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1580 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1583 nfs4_unlock_state();
1589 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1590 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1591 * bullets, labeled as CASE1 - CASE4 below.
1594 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1595 struct nfsd4_compound_state *cstate,
1596 struct nfsd4_setclientid_confirm *setclientid_confirm)
1598 struct sockaddr *sa = svc_addr(rqstp);
1599 struct nfs4_client *conf, *unconf;
1600 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1601 clientid_t * clid = &setclientid_confirm->sc_clientid;
1604 if (STALE_CLIENTID(clid))
1605 return nfserr_stale_clientid;
1607 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1608 * We get here on a DRC miss.
1613 conf = find_confirmed_client(clid);
1614 unconf = find_unconfirmed_client(clid);
1616 status = nfserr_clid_inuse;
1617 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1619 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1623 * section 14.2.34 of RFC 3530 has a description of
1624 * SETCLIENTID_CONFIRM request processing consisting
1625 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1627 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1629 * RFC 3530 14.2.34 CASE 1:
1632 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1633 status = nfserr_clid_inuse;
1635 /* XXX: We just turn off callbacks until we can handle
1636 * change request correctly. */
1637 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1638 expire_client(unconf);
1642 } else if (conf && !unconf) {
1644 * RFC 3530 14.2.34 CASE 2:
1645 * probable retransmitted request; play it safe and
1648 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1649 status = nfserr_clid_inuse;
1652 } else if (!conf && unconf
1653 && same_verf(&unconf->cl_confirm, &confirm)) {
1655 * RFC 3530 14.2.34 CASE 3:
1656 * Normal case; new or rebooted client:
1658 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1659 status = nfserr_clid_inuse;
1662 clientstr_hashval(unconf->cl_recdir);
1663 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1666 nfsd4_remove_clid_dir(conf);
1667 expire_client(conf);
1669 move_to_confirmed(unconf);
1671 nfsd4_probe_callback(conf);
1674 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1675 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1678 * RFC 3530 14.2.34 CASE 4:
1679 * Client probably hasn't noticed that we rebooted yet.
1681 status = nfserr_stale_clientid;
1683 /* check that we have hit one of the cases...*/
1684 status = nfserr_clid_inuse;
1687 nfs4_unlock_state();
1691 /* OPEN Share state helper functions */
1692 static inline struct nfs4_file *
1693 alloc_init_file(struct inode *ino)
1695 struct nfs4_file *fp;
1696 unsigned int hashval = file_hashval(ino);
1698 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1700 atomic_set(&fp->fi_ref, 1);
1701 INIT_LIST_HEAD(&fp->fi_hash);
1702 INIT_LIST_HEAD(&fp->fi_stateids);
1703 INIT_LIST_HEAD(&fp->fi_delegations);
1704 spin_lock(&recall_lock);
1705 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1706 spin_unlock(&recall_lock);
1707 fp->fi_inode = igrab(ino);
1708 fp->fi_id = current_fileid++;
1709 fp->fi_had_conflict = false;
1716 nfsd4_free_slab(struct kmem_cache **slab)
1720 kmem_cache_destroy(*slab);
1725 nfsd4_free_slabs(void)
1727 nfsd4_free_slab(&stateowner_slab);
1728 nfsd4_free_slab(&file_slab);
1729 nfsd4_free_slab(&stateid_slab);
1730 nfsd4_free_slab(&deleg_slab);
1734 nfsd4_init_slabs(void)
1736 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1737 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1738 if (stateowner_slab == NULL)
1740 file_slab = kmem_cache_create("nfsd4_files",
1741 sizeof(struct nfs4_file), 0, 0, NULL);
1742 if (file_slab == NULL)
1744 stateid_slab = kmem_cache_create("nfsd4_stateids",
1745 sizeof(struct nfs4_stateid), 0, 0, NULL);
1746 if (stateid_slab == NULL)
1748 deleg_slab = kmem_cache_create("nfsd4_delegations",
1749 sizeof(struct nfs4_delegation), 0, 0, NULL);
1750 if (deleg_slab == NULL)
1755 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1760 nfs4_free_stateowner(struct kref *kref)
1762 struct nfs4_stateowner *sop =
1763 container_of(kref, struct nfs4_stateowner, so_ref);
1764 kfree(sop->so_owner.data);
1765 kmem_cache_free(stateowner_slab, sop);
1768 static inline struct nfs4_stateowner *
1769 alloc_stateowner(struct xdr_netobj *owner)
1771 struct nfs4_stateowner *sop;
1773 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1774 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1775 memcpy(sop->so_owner.data, owner->data, owner->len);
1776 sop->so_owner.len = owner->len;
1777 kref_init(&sop->so_ref);
1780 kmem_cache_free(stateowner_slab, sop);
1785 static struct nfs4_stateowner *
1786 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1787 struct nfs4_stateowner *sop;
1788 struct nfs4_replay *rp;
1789 unsigned int idhashval;
1791 if (!(sop = alloc_stateowner(&open->op_owner)))
1793 idhashval = ownerid_hashval(current_ownerid);
1794 INIT_LIST_HEAD(&sop->so_idhash);
1795 INIT_LIST_HEAD(&sop->so_strhash);
1796 INIT_LIST_HEAD(&sop->so_perclient);
1797 INIT_LIST_HEAD(&sop->so_stateids);
1798 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1799 INIT_LIST_HEAD(&sop->so_close_lru);
1801 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1802 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1803 list_add(&sop->so_perclient, &clp->cl_openowners);
1804 sop->so_is_open_owner = 1;
1805 sop->so_id = current_ownerid++;
1806 sop->so_client = clp;
1807 sop->so_seqid = open->op_seqid;
1808 sop->so_confirmed = 0;
1809 rp = &sop->so_replay;
1810 rp->rp_status = nfserr_serverfault;
1812 rp->rp_buf = rp->rp_ibuf;
1817 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1818 struct nfs4_stateowner *sop = open->op_stateowner;
1819 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1821 INIT_LIST_HEAD(&stp->st_hash);
1822 INIT_LIST_HEAD(&stp->st_perstateowner);
1823 INIT_LIST_HEAD(&stp->st_lockowners);
1824 INIT_LIST_HEAD(&stp->st_perfile);
1825 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1826 list_add(&stp->st_perstateowner, &sop->so_stateids);
1827 list_add(&stp->st_perfile, &fp->fi_stateids);
1828 stp->st_stateowner = sop;
1831 stp->st_stateid.si_boot = get_seconds();
1832 stp->st_stateid.si_stateownerid = sop->so_id;
1833 stp->st_stateid.si_fileid = fp->fi_id;
1834 stp->st_stateid.si_generation = 0;
1835 stp->st_access_bmap = 0;
1836 stp->st_deny_bmap = 0;
1837 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1838 &stp->st_access_bmap);
1839 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1840 stp->st_openstp = NULL;
1844 move_to_close_lru(struct nfs4_stateowner *sop)
1846 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1848 list_move_tail(&sop->so_close_lru, &close_lru);
1849 sop->so_time = get_seconds();
1853 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1856 return (sop->so_owner.len == owner->len) &&
1857 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1858 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1861 static struct nfs4_stateowner *
1862 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1864 struct nfs4_stateowner *so = NULL;
1866 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1867 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1873 /* search file_hashtbl[] for file */
1874 static struct nfs4_file *
1875 find_file(struct inode *ino)
1877 unsigned int hashval = file_hashval(ino);
1878 struct nfs4_file *fp;
1880 spin_lock(&recall_lock);
1881 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1882 if (fp->fi_inode == ino) {
1884 spin_unlock(&recall_lock);
1888 spin_unlock(&recall_lock);
1892 static inline int access_valid(u32 x, u32 minorversion)
1894 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1896 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1898 x &= ~NFS4_SHARE_ACCESS_MASK;
1899 if (minorversion && x) {
1900 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1902 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1904 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1911 static inline int deny_valid(u32 x)
1913 /* Note: unlike access bits, deny bits may be zero. */
1914 return x <= NFS4_SHARE_DENY_BOTH;
1918 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1919 * st_{access,deny}_bmap field of the stateid, in order to track not
1920 * only what share bits are currently in force, but also what
1921 * combinations of share bits previous opens have used. This allows us
1922 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1923 * return an error if the client attempt to downgrade to a combination
1924 * of share bits not explicable by closing some of its previous opens.
1926 * XXX: This enforcement is actually incomplete, since we don't keep
1927 * track of access/deny bit combinations; so, e.g., we allow:
1929 * OPEN allow read, deny write
1930 * OPEN allow both, deny none
1931 * DOWNGRADE allow read, deny none
1933 * which we should reject.
1936 set_access(unsigned int *access, unsigned long bmap) {
1940 for (i = 1; i < 4; i++) {
1941 if (test_bit(i, &bmap))
1947 set_deny(unsigned int *deny, unsigned long bmap) {
1951 for (i = 0; i < 4; i++) {
1952 if (test_bit(i, &bmap))
1958 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1959 unsigned int access, deny;
1961 set_access(&access, stp->st_access_bmap);
1962 set_deny(&deny, stp->st_deny_bmap);
1963 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1969 * Called to check deny when READ with all zero stateid or
1970 * WRITE with all zero or all one stateid
1973 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1975 struct inode *ino = current_fh->fh_dentry->d_inode;
1976 struct nfs4_file *fp;
1977 struct nfs4_stateid *stp;
1980 dprintk("NFSD: nfs4_share_conflict\n");
1982 fp = find_file(ino);
1985 ret = nfserr_locked;
1986 /* Search for conflicting share reservations */
1987 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1988 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1989 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1999 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2001 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2002 drop_file_write_access(filp);
2003 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2008 * Spawn a thread to perform a recall on the delegation represented
2009 * by the lease (file_lock)
2011 * Called from break_lease() with lock_kernel() held.
2012 * Note: we assume break_lease will only call this *once* for any given
2016 void nfsd_break_deleg_cb(struct file_lock *fl)
2018 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2020 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2024 /* We're assuming the state code never drops its reference
2025 * without first removing the lease. Since we're in this lease
2026 * callback (and since the lease code is serialized by the kernel
2027 * lock) we know the server hasn't removed the lease yet, we know
2028 * it's safe to take a reference: */
2029 atomic_inc(&dp->dl_count);
2030 atomic_inc(&dp->dl_client->cl_count);
2032 spin_lock(&recall_lock);
2033 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2034 spin_unlock(&recall_lock);
2036 /* only place dl_time is set. protected by lock_kernel*/
2037 dp->dl_time = get_seconds();
2040 * We don't want the locks code to timeout the lease for us;
2041 * we'll remove it ourself if the delegation isn't returned
2044 fl->fl_break_time = 0;
2046 dp->dl_file->fi_had_conflict = true;
2047 nfsd4_cb_recall(dp);
2051 * The file_lock is being reapd.
2053 * Called by locks_free_lock() with lock_kernel() held.
2056 void nfsd_release_deleg_cb(struct file_lock *fl)
2058 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2060 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2062 if (!(fl->fl_flags & FL_LEASE) || !dp)
2064 dp->dl_flock = NULL;
2068 * Set the delegation file_lock back pointer.
2070 * Called from setlease() with lock_kernel() held.
2073 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2075 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2077 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2084 * Called from setlease() with lock_kernel() held
2087 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2089 struct nfs4_delegation *onlistd =
2090 (struct nfs4_delegation *)onlist->fl_owner;
2091 struct nfs4_delegation *tryd =
2092 (struct nfs4_delegation *)try->fl_owner;
2094 if (onlist->fl_lmops != try->fl_lmops)
2097 return onlistd->dl_client == tryd->dl_client;
2102 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2105 return lease_modify(onlist, arg);
2110 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2111 .fl_break = nfsd_break_deleg_cb,
2112 .fl_release_private = nfsd_release_deleg_cb,
2113 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2114 .fl_mylease = nfsd_same_client_deleg_cb,
2115 .fl_change = nfsd_change_deleg_cb,
2120 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2121 struct nfsd4_open *open)
2123 clientid_t *clientid = &open->op_clientid;
2124 struct nfs4_client *clp = NULL;
2125 unsigned int strhashval;
2126 struct nfs4_stateowner *sop = NULL;
2128 if (!check_name(open->op_owner))
2129 return nfserr_inval;
2131 if (STALE_CLIENTID(&open->op_clientid))
2132 return nfserr_stale_clientid;
2134 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2135 sop = find_openstateowner_str(strhashval, open);
2136 open->op_stateowner = sop;
2138 /* Make sure the client's lease hasn't expired. */
2139 clp = find_confirmed_client(clientid);
2141 return nfserr_expired;
2144 /* When sessions are used, skip open sequenceid processing */
2145 if (nfsd4_has_session(cstate))
2147 if (!sop->so_confirmed) {
2148 /* Replace unconfirmed owners without checking for replay. */
2149 clp = sop->so_client;
2150 release_openowner(sop);
2151 open->op_stateowner = NULL;
2154 if (open->op_seqid == sop->so_seqid - 1) {
2155 if (sop->so_replay.rp_buflen)
2156 return nfserr_replay_me;
2157 /* The original OPEN failed so spectacularly
2158 * that we don't even have replay data saved!
2159 * Therefore, we have no choice but to continue
2160 * processing this OPEN; presumably, we'll
2161 * fail again for the same reason.
2163 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2166 if (open->op_seqid != sop->so_seqid)
2167 return nfserr_bad_seqid;
2169 if (open->op_stateowner == NULL) {
2170 sop = alloc_init_open_stateowner(strhashval, clp, open);
2172 return nfserr_resource;
2173 open->op_stateowner = sop;
2175 list_del_init(&sop->so_close_lru);
2176 renew_client(sop->so_client);
2180 static inline __be32
2181 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2183 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2184 return nfserr_openmode;
2189 static struct nfs4_delegation *
2190 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2192 struct nfs4_delegation *dp;
2194 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2195 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2202 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2203 struct nfs4_delegation **dp)
2206 __be32 status = nfserr_bad_stateid;
2208 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2211 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2212 RD_STATE : WR_STATE;
2213 status = nfs4_check_delegmode(*dp, flags);
2217 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2221 open->op_stateowner->so_confirmed = 1;
2226 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2228 struct nfs4_stateid *local;
2229 __be32 status = nfserr_share_denied;
2230 struct nfs4_stateowner *sop = open->op_stateowner;
2232 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2233 /* ignore lock owners */
2234 if (local->st_stateowner->so_is_open_owner == 0)
2236 /* remember if we have seen this open owner */
2237 if (local->st_stateowner == sop)
2239 /* check for conflicting share reservations */
2240 if (!test_share(local, open))
2248 static inline struct nfs4_stateid *
2249 nfs4_alloc_stateid(void)
2251 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2255 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2256 struct nfs4_delegation *dp,
2257 struct svc_fh *cur_fh, int flags)
2259 struct nfs4_stateid *stp;
2261 stp = nfs4_alloc_stateid();
2263 return nfserr_resource;
2266 get_file(dp->dl_vfs_file);
2267 stp->st_vfs_file = dp->dl_vfs_file;
2270 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2273 if (status == nfserr_dropit)
2274 status = nfserr_jukebox;
2275 kmem_cache_free(stateid_slab, stp);
2283 static inline __be32
2284 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2285 struct nfsd4_open *open)
2287 struct iattr iattr = {
2288 .ia_valid = ATTR_SIZE,
2291 if (!open->op_truncate)
2293 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2294 return nfserr_inval;
2295 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2299 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2301 struct file *filp = stp->st_vfs_file;
2302 struct inode *inode = filp->f_path.dentry->d_inode;
2303 unsigned int share_access, new_writer;
2306 set_access(&share_access, stp->st_access_bmap);
2307 new_writer = (~share_access) & open->op_share_access
2308 & NFS4_SHARE_ACCESS_WRITE;
2311 int err = get_write_access(inode);
2313 return nfserrno(err);
2314 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2316 return nfserrno(err);
2317 file_take_write(filp);
2319 status = nfsd4_truncate(rqstp, cur_fh, open);
2322 put_write_access(inode);
2325 /* remember the open */
2326 filp->f_mode |= open->op_share_access;
2327 __set_bit(open->op_share_access, &stp->st_access_bmap);
2328 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2335 nfs4_set_claim_prev(struct nfsd4_open *open)
2337 open->op_stateowner->so_confirmed = 1;
2338 open->op_stateowner->so_client->cl_firststate = 1;
2342 * Attempt to hand out a delegation.
2345 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2347 struct nfs4_delegation *dp;
2348 struct nfs4_stateowner *sop = stp->st_stateowner;
2349 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2350 struct file_lock fl, *flp = &fl;
2351 int status, flag = 0;
2353 flag = NFS4_OPEN_DELEGATE_NONE;
2354 open->op_recall = 0;
2355 switch (open->op_claim_type) {
2356 case NFS4_OPEN_CLAIM_PREVIOUS:
2357 if (!atomic_read(&cb->cb_set))
2358 open->op_recall = 1;
2359 flag = open->op_delegate_type;
2360 if (flag == NFS4_OPEN_DELEGATE_NONE)
2363 case NFS4_OPEN_CLAIM_NULL:
2364 /* Let's not give out any delegations till everyone's
2365 * had the chance to reclaim theirs.... */
2366 if (locks_in_grace())
2368 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2370 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2371 flag = NFS4_OPEN_DELEGATE_WRITE;
2373 flag = NFS4_OPEN_DELEGATE_READ;
2379 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2381 flag = NFS4_OPEN_DELEGATE_NONE;
2384 locks_init_lock(&fl);
2385 fl.fl_lmops = &nfsd_lease_mng_ops;
2386 fl.fl_flags = FL_LEASE;
2387 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2388 fl.fl_end = OFFSET_MAX;
2389 fl.fl_owner = (fl_owner_t)dp;
2390 fl.fl_file = stp->st_vfs_file;
2391 fl.fl_pid = current->tgid;
2393 /* vfs_setlease checks to see if delegation should be handed out.
2394 * the lock_manager callbacks fl_mylease and fl_change are used
2396 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2397 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2398 unhash_delegation(dp);
2399 flag = NFS4_OPEN_DELEGATE_NONE;
2403 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2405 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2406 STATEID_VAL(&dp->dl_stateid));
2408 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2409 && flag == NFS4_OPEN_DELEGATE_NONE
2410 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2411 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2412 open->op_delegate_type = flag;
2416 * called with nfs4_lock_state() held.
2419 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2421 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2422 struct nfs4_file *fp = NULL;
2423 struct inode *ino = current_fh->fh_dentry->d_inode;
2424 struct nfs4_stateid *stp = NULL;
2425 struct nfs4_delegation *dp = NULL;
2428 status = nfserr_inval;
2429 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2430 || !deny_valid(open->op_share_deny))
2433 * Lookup file; if found, lookup stateid and check open request,
2434 * and check for delegations in the process of being recalled.
2435 * If not found, create the nfs4_file struct
2437 fp = find_file(ino);
2439 if ((status = nfs4_check_open(fp, open, &stp)))
2441 status = nfs4_check_deleg(fp, open, &dp);
2445 status = nfserr_bad_stateid;
2446 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2448 status = nfserr_resource;
2449 fp = alloc_init_file(ino);
2455 * OPEN the file, or upgrade an existing OPEN.
2456 * If truncate fails, the OPEN fails.
2459 /* Stateid was found, this is an OPEN upgrade */
2460 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2463 update_stateid(&stp->st_stateid);
2465 /* Stateid was not found, this is a new OPEN */
2467 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2468 flags |= NFSD_MAY_READ;
2469 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2470 flags |= NFSD_MAY_WRITE;
2471 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2474 init_stateid(stp, fp, open);
2475 status = nfsd4_truncate(rqstp, current_fh, open);
2477 release_open_stateid(stp);
2480 if (nfsd4_has_session(&resp->cstate))
2481 update_stateid(&stp->st_stateid);
2483 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2485 if (nfsd4_has_session(&resp->cstate))
2486 open->op_stateowner->so_confirmed = 1;
2489 * Attempt to hand out a delegation. No error return, because the
2490 * OPEN succeeds even if we fail.
2492 nfs4_open_delegation(current_fh, open, stp);
2496 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2497 STATEID_VAL(&stp->st_stateid));
2501 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2502 nfs4_set_claim_prev(open);
2504 * To finish the open response, we just need to set the rflags.
2506 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2507 if (!open->op_stateowner->so_confirmed &&
2508 !nfsd4_has_session(&resp->cstate))
2509 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2515 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2518 struct nfs4_client *clp;
2522 dprintk("process_renew(%08x/%08x): starting\n",
2523 clid->cl_boot, clid->cl_id);
2524 status = nfserr_stale_clientid;
2525 if (STALE_CLIENTID(clid))
2527 clp = find_confirmed_client(clid);
2528 status = nfserr_expired;
2530 /* We assume the client took too long to RENEW. */
2531 dprintk("nfsd4_renew: clientid not found!\n");
2535 status = nfserr_cb_path_down;
2536 if (!list_empty(&clp->cl_delegations)
2537 && !atomic_read(&clp->cl_cb_conn.cb_set))
2541 nfs4_unlock_state();
2545 struct lock_manager nfsd4_manager = {
2549 nfsd4_end_grace(void)
2551 dprintk("NFSD: end of grace period\n");
2552 nfsd4_recdir_purge_old();
2553 locks_end_grace(&nfsd4_manager);
2557 nfs4_laundromat(void)
2559 struct nfs4_client *clp;
2560 struct nfs4_stateowner *sop;
2561 struct nfs4_delegation *dp;
2562 struct list_head *pos, *next, reaplist;
2563 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2564 time_t t, clientid_val = NFSD_LEASE_TIME;
2565 time_t u, test_val = NFSD_LEASE_TIME;
2569 dprintk("NFSD: laundromat service - starting\n");
2570 if (locks_in_grace())
2572 list_for_each_safe(pos, next, &client_lru) {
2573 clp = list_entry(pos, struct nfs4_client, cl_lru);
2574 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2575 t = clp->cl_time - cutoff;
2576 if (clientid_val > t)
2580 dprintk("NFSD: purging unused client (clientid %08x)\n",
2581 clp->cl_clientid.cl_id);
2582 nfsd4_remove_clid_dir(clp);
2585 INIT_LIST_HEAD(&reaplist);
2586 spin_lock(&recall_lock);
2587 list_for_each_safe(pos, next, &del_recall_lru) {
2588 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2589 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2590 u = dp->dl_time - cutoff;
2595 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2597 list_move(&dp->dl_recall_lru, &reaplist);
2599 spin_unlock(&recall_lock);
2600 list_for_each_safe(pos, next, &reaplist) {
2601 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2602 list_del_init(&dp->dl_recall_lru);
2603 unhash_delegation(dp);
2605 test_val = NFSD_LEASE_TIME;
2606 list_for_each_safe(pos, next, &close_lru) {
2607 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2608 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2609 u = sop->so_time - cutoff;
2614 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2616 release_openowner(sop);
2618 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2619 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2620 nfs4_unlock_state();
2621 return clientid_val;
2624 static struct workqueue_struct *laundry_wq;
2625 static void laundromat_main(struct work_struct *);
2626 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2629 laundromat_main(struct work_struct *not_used)
2633 t = nfs4_laundromat();
2634 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2635 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2638 static struct nfs4_stateowner *
2639 search_close_lru(u32 st_id, int flags)
2641 struct nfs4_stateowner *local = NULL;
2643 if (flags & CLOSE_STATE) {
2644 list_for_each_entry(local, &close_lru, so_close_lru) {
2645 if (local->so_id == st_id)
2653 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2655 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2659 STALE_STATEID(stateid_t *stateid)
2661 if (time_after((unsigned long)boot_time,
2662 (unsigned long)stateid->si_boot)) {
2663 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2664 STATEID_VAL(stateid));
2671 EXPIRED_STATEID(stateid_t *stateid)
2673 if (time_before((unsigned long)boot_time,
2674 ((unsigned long)stateid->si_boot)) &&
2675 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2676 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2677 STATEID_VAL(stateid));
2684 stateid_error_map(stateid_t *stateid)
2686 if (STALE_STATEID(stateid))
2687 return nfserr_stale_stateid;
2688 if (EXPIRED_STATEID(stateid))
2689 return nfserr_expired;
2691 dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2692 STATEID_VAL(stateid));
2693 return nfserr_bad_stateid;
2697 access_permit_read(unsigned long access_bmap)
2699 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2700 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2701 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2705 access_permit_write(unsigned long access_bmap)
2707 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2708 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2712 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2714 __be32 status = nfserr_openmode;
2716 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2718 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2725 static inline __be32
2726 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2728 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2730 else if (locks_in_grace()) {
2731 /* Answer in remaining cases depends on existance of
2732 * conflicting state; so we must wait out the grace period. */
2733 return nfserr_grace;
2734 } else if (flags & WR_STATE)
2735 return nfs4_share_conflict(current_fh,
2736 NFS4_SHARE_DENY_WRITE);
2737 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2738 return nfs4_share_conflict(current_fh,
2739 NFS4_SHARE_DENY_READ);
2743 * Allow READ/WRITE during grace period on recovered state only for files
2744 * that are not able to provide mandatory locking.
2747 grace_disallows_io(struct inode *inode)
2749 return locks_in_grace() && mandatory_lock(inode);
2752 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2755 * When sessions are used the stateid generation number is ignored
2758 if ((flags & HAS_SESSION) && in->si_generation == 0)
2761 /* If the client sends us a stateid from the future, it's buggy: */
2762 if (in->si_generation > ref->si_generation)
2763 return nfserr_bad_stateid;
2765 * The following, however, can happen. For example, if the
2766 * client sends an open and some IO at the same time, the open
2767 * may bump si_generation while the IO is still in flight.
2768 * Thanks to hard links and renames, the client never knows what
2769 * file an open will affect. So it could avoid that situation
2770 * only by serializing all opens and IO from the same open
2771 * owner. To recover from the old_stateid error, the client
2772 * will just have to retry the IO:
2774 if (in->si_generation < ref->si_generation)
2775 return nfserr_old_stateid;
2780 static int is_delegation_stateid(stateid_t *stateid)
2782 return stateid->si_fileid == 0;
2786 * Checks for stateid operations
2789 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2790 stateid_t *stateid, int flags, struct file **filpp)
2792 struct nfs4_stateid *stp = NULL;
2793 struct nfs4_delegation *dp = NULL;
2794 struct svc_fh *current_fh = &cstate->current_fh;
2795 struct inode *ino = current_fh->fh_dentry->d_inode;
2801 if (grace_disallows_io(ino))
2802 return nfserr_grace;
2804 if (nfsd4_has_session(cstate))
2805 flags |= HAS_SESSION;
2807 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2808 return check_special_stateids(current_fh, stateid, flags);
2810 status = nfserr_stale_stateid;
2811 if (STALE_STATEID(stateid))
2814 status = nfserr_bad_stateid;
2815 if (is_delegation_stateid(stateid)) {
2816 dp = find_delegation_stateid(ino, stateid);
2818 status = stateid_error_map(stateid);
2821 status = check_stateid_generation(stateid, &dp->dl_stateid,
2825 status = nfs4_check_delegmode(dp, flags);
2828 renew_client(dp->dl_client);
2830 *filpp = dp->dl_vfs_file;
2831 } else { /* open or lock stateid */
2832 stp = find_stateid(stateid, flags);
2834 status = stateid_error_map(stateid);
2837 if (nfs4_check_fh(current_fh, stp))
2839 if (!stp->st_stateowner->so_confirmed)
2841 status = check_stateid_generation(stateid, &stp->st_stateid,
2845 status = nfs4_check_openmode(stp, flags);
2848 renew_client(stp->st_stateowner->so_client);
2850 *filpp = stp->st_vfs_file;
2860 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2861 RD_STATE : WR_STATE;
2865 * Checks for sequence id mutating operations.
2868 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2869 stateid_t *stateid, int flags,
2870 struct nfs4_stateowner **sopp,
2871 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2873 struct nfs4_stateid *stp;
2874 struct nfs4_stateowner *sop;
2875 struct svc_fh *current_fh = &cstate->current_fh;
2878 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2879 seqid, STATEID_VAL(stateid));
2884 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2885 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2886 return nfserr_bad_stateid;
2889 if (STALE_STATEID(stateid))
2890 return nfserr_stale_stateid;
2892 if (nfsd4_has_session(cstate))
2893 flags |= HAS_SESSION;
2896 * We return BAD_STATEID if filehandle doesn't match stateid,
2897 * the confirmed flag is incorrecly set, or the generation
2898 * number is incorrect.
2900 stp = find_stateid(stateid, flags);
2903 * Also, we should make sure this isn't just the result of
2906 sop = search_close_lru(stateid->si_stateownerid, flags);
2908 return stateid_error_map(stateid);
2914 *sopp = sop = stp->st_stateowner;
2917 clientid_t *lockclid = &lock->v.new.clientid;
2918 struct nfs4_client *clp = sop->so_client;
2922 lkflg = setlkflg(lock->lk_type);
2924 if (lock->lk_is_new) {
2925 if (!sop->so_is_open_owner)
2926 return nfserr_bad_stateid;
2927 if (!(flags & HAS_SESSION) &&
2928 !same_clid(&clp->cl_clientid, lockclid))
2929 return nfserr_bad_stateid;
2930 /* stp is the open stateid */
2931 status = nfs4_check_openmode(stp, lkflg);
2935 /* stp is the lock stateid */
2936 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2942 if (nfs4_check_fh(current_fh, stp)) {
2943 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2944 return nfserr_bad_stateid;
2948 * We now validate the seqid and stateid generation numbers.
2949 * For the moment, we ignore the possibility of
2950 * generation number wraparound.
2952 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2955 if (sop->so_confirmed && flags & CONFIRM) {
2956 dprintk("NFSD: preprocess_seqid_op: expected"
2957 " unconfirmed stateowner!\n");
2958 return nfserr_bad_stateid;
2960 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2961 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2962 " confirmed yet!\n");
2963 return nfserr_bad_stateid;
2965 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2968 renew_client(sop->so_client);
2972 if (seqid == sop->so_seqid - 1) {
2973 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2974 /* indicate replay to calling function */
2975 return nfserr_replay_me;
2977 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2978 sop->so_seqid, seqid);
2980 return nfserr_bad_seqid;
2984 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2985 struct nfsd4_open_confirm *oc)
2988 struct nfs4_stateowner *sop;
2989 struct nfs4_stateid *stp;
2991 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2992 (int)cstate->current_fh.fh_dentry->d_name.len,
2993 cstate->current_fh.fh_dentry->d_name.name);
2995 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3001 if ((status = nfs4_preprocess_seqid_op(cstate,
3002 oc->oc_seqid, &oc->oc_req_stateid,
3003 CONFIRM | OPEN_STATE,
3004 &oc->oc_stateowner, &stp, NULL)))
3007 sop = oc->oc_stateowner;
3008 sop->so_confirmed = 1;
3009 update_stateid(&stp->st_stateid);
3010 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3011 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3012 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3014 nfsd4_create_clid_dir(sop->so_client);
3016 if (oc->oc_stateowner) {
3017 nfs4_get_stateowner(oc->oc_stateowner);
3018 cstate->replay_owner = oc->oc_stateowner;
3020 nfs4_unlock_state();
3026 * unset all bits in union bitmap (bmap) that
3027 * do not exist in share (from successful OPEN_DOWNGRADE)
3030 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3033 for (i = 1; i < 4; i++) {
3034 if ((i & access) != i)
3035 __clear_bit(i, bmap);
3040 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3043 for (i = 0; i < 4; i++) {
3044 if ((i & deny) != i)
3045 __clear_bit(i, bmap);
3050 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3051 struct nfsd4_compound_state *cstate,
3052 struct nfsd4_open_downgrade *od)
3055 struct nfs4_stateid *stp;
3056 unsigned int share_access;
3058 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3059 (int)cstate->current_fh.fh_dentry->d_name.len,
3060 cstate->current_fh.fh_dentry->d_name.name);
3062 if (!access_valid(od->od_share_access, cstate->minorversion)
3063 || !deny_valid(od->od_share_deny))
3064 return nfserr_inval;
3067 if ((status = nfs4_preprocess_seqid_op(cstate,
3071 &od->od_stateowner, &stp, NULL)))
3074 status = nfserr_inval;
3075 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3076 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3077 stp->st_access_bmap, od->od_share_access);
3080 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3081 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3082 stp->st_deny_bmap, od->od_share_deny);
3085 set_access(&share_access, stp->st_access_bmap);
3086 nfs4_file_downgrade(stp->st_vfs_file,
3087 share_access & ~od->od_share_access);
3089 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3090 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3092 update_stateid(&stp->st_stateid);
3093 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3096 if (od->od_stateowner) {
3097 nfs4_get_stateowner(od->od_stateowner);
3098 cstate->replay_owner = od->od_stateowner;
3100 nfs4_unlock_state();
3105 * nfs4_unlock_state() called after encode
3108 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3109 struct nfsd4_close *close)
3112 struct nfs4_stateid *stp;
3114 dprintk("NFSD: nfsd4_close on file %.*s\n",
3115 (int)cstate->current_fh.fh_dentry->d_name.len,
3116 cstate->current_fh.fh_dentry->d_name.name);
3119 /* check close_lru for replay */
3120 if ((status = nfs4_preprocess_seqid_op(cstate,
3123 OPEN_STATE | CLOSE_STATE,
3124 &close->cl_stateowner, &stp, NULL)))
3127 update_stateid(&stp->st_stateid);
3128 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3130 /* release_stateid() calls nfsd_close() if needed */
3131 release_open_stateid(stp);
3133 /* place unused nfs4_stateowners on so_close_lru list to be
3134 * released by the laundromat service after the lease period
3135 * to enable us to handle CLOSE replay
3137 if (list_empty(&close->cl_stateowner->so_stateids))
3138 move_to_close_lru(close->cl_stateowner);
3140 if (close->cl_stateowner) {
3141 nfs4_get_stateowner(close->cl_stateowner);
3142 cstate->replay_owner = close->cl_stateowner;
3144 nfs4_unlock_state();
3149 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3150 struct nfsd4_delegreturn *dr)
3152 struct nfs4_delegation *dp;
3153 stateid_t *stateid = &dr->dr_stateid;
3154 struct inode *inode;
3158 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3160 inode = cstate->current_fh.fh_dentry->d_inode;
3162 if (nfsd4_has_session(cstate))
3163 flags |= HAS_SESSION;
3165 status = nfserr_bad_stateid;
3166 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3168 status = nfserr_stale_stateid;
3169 if (STALE_STATEID(stateid))
3171 status = nfserr_bad_stateid;
3172 if (!is_delegation_stateid(stateid))
3174 dp = find_delegation_stateid(inode, stateid);
3176 status = stateid_error_map(stateid);
3179 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3182 renew_client(dp->dl_client);
3184 unhash_delegation(dp);
3186 nfs4_unlock_state();
3193 * Lock owner state (byte-range locks)
3195 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3196 #define LOCK_HASH_BITS 8
3197 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3198 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3201 end_offset(u64 start, u64 len)
3206 return end >= start ? end: NFS4_MAX_UINT64;
3209 /* last octet in a range */
3211 last_byte_offset(u64 start, u64 len)
3217 return end > start ? end - 1: NFS4_MAX_UINT64;
3220 #define lockownerid_hashval(id) \
3221 ((id) & LOCK_HASH_MASK)
3223 static inline unsigned int
3224 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3225 struct xdr_netobj *ownername)
3227 return (file_hashval(inode) + cl_id
3228 + opaque_hashval(ownername->data, ownername->len))
3232 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3233 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3234 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3236 static struct nfs4_stateid *
3237 find_stateid(stateid_t *stid, int flags)
3239 struct nfs4_stateid *local;
3240 u32 st_id = stid->si_stateownerid;
3241 u32 f_id = stid->si_fileid;
3242 unsigned int hashval;
3244 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3245 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3246 hashval = stateid_hashval(st_id, f_id);
3247 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3248 if ((local->st_stateid.si_stateownerid == st_id) &&
3249 (local->st_stateid.si_fileid == f_id))
3254 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3255 hashval = stateid_hashval(st_id, f_id);
3256 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3257 if ((local->st_stateid.si_stateownerid == st_id) &&
3258 (local->st_stateid.si_fileid == f_id))
3265 static struct nfs4_delegation *
3266 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3268 struct nfs4_file *fp;
3269 struct nfs4_delegation *dl;
3271 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3274 fp = find_file(ino);
3277 dl = find_delegation_file(fp, stid);
3283 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3284 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3285 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3286 * locking, this prevents us from being completely protocol-compliant. The
3287 * real solution to this problem is to start using unsigned file offsets in
3288 * the VFS, but this is a very deep change!
3291 nfs4_transform_lock_offset(struct file_lock *lock)
3293 if (lock->fl_start < 0)
3294 lock->fl_start = OFFSET_MAX;
3295 if (lock->fl_end < 0)
3296 lock->fl_end = OFFSET_MAX;
3299 /* Hack!: For now, we're defining this just so we can use a pointer to it
3300 * as a unique cookie to identify our (NFSv4's) posix locks. */
3301 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3305 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3307 struct nfs4_stateowner *sop;
3310 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3311 sop = (struct nfs4_stateowner *) fl->fl_owner;
3312 hval = lockownerid_hashval(sop->so_id);
3313 kref_get(&sop->so_ref);
3315 deny->ld_clientid = sop->so_client->cl_clientid;
3317 deny->ld_sop = NULL;
3318 deny->ld_clientid.cl_boot = 0;
3319 deny->ld_clientid.cl_id = 0;
3321 deny->ld_start = fl->fl_start;
3322 deny->ld_length = NFS4_MAX_UINT64;
3323 if (fl->fl_end != NFS4_MAX_UINT64)
3324 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3325 deny->ld_type = NFS4_READ_LT;
3326 if (fl->fl_type != F_RDLCK)
3327 deny->ld_type = NFS4_WRITE_LT;
3330 static struct nfs4_stateowner *
3331 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3332 struct xdr_netobj *owner)
3334 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3335 struct nfs4_stateowner *op;
3337 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3338 if (same_owner_str(op, owner, clid))
3345 * Alloc a lock owner structure.
3346 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3349 * strhashval = lock_ownerstr_hashval
3352 static struct nfs4_stateowner *
3353 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3354 struct nfs4_stateowner *sop;
3355 struct nfs4_replay *rp;
3356 unsigned int idhashval;
3358 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3360 idhashval = lockownerid_hashval(current_ownerid);
3361 INIT_LIST_HEAD(&sop->so_idhash);
3362 INIT_LIST_HEAD(&sop->so_strhash);
3363 INIT_LIST_HEAD(&sop->so_perclient);
3364 INIT_LIST_HEAD(&sop->so_stateids);
3365 INIT_LIST_HEAD(&sop->so_perstateid);
3366 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3368 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3369 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3370 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3371 sop->so_is_open_owner = 0;
3372 sop->so_id = current_ownerid++;
3373 sop->so_client = clp;
3374 /* It is the openowner seqid that will be incremented in encode in the
3375 * case of new lockowners; so increment the lock seqid manually: */
3376 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3377 sop->so_confirmed = 1;
3378 rp = &sop->so_replay;
3379 rp->rp_status = nfserr_serverfault;
3381 rp->rp_buf = rp->rp_ibuf;
3385 static struct nfs4_stateid *
3386 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3388 struct nfs4_stateid *stp;
3389 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3391 stp = nfs4_alloc_stateid();
3394 INIT_LIST_HEAD(&stp->st_hash);
3395 INIT_LIST_HEAD(&stp->st_perfile);
3396 INIT_LIST_HEAD(&stp->st_perstateowner);
3397 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3398 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3399 list_add(&stp->st_perfile, &fp->fi_stateids);
3400 list_add(&stp->st_perstateowner, &sop->so_stateids);
3401 stp->st_stateowner = sop;
3404 stp->st_stateid.si_boot = get_seconds();
3405 stp->st_stateid.si_stateownerid = sop->so_id;
3406 stp->st_stateid.si_fileid = fp->fi_id;
3407 stp->st_stateid.si_generation = 0;
3408 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3409 stp->st_access_bmap = open_stp->st_access_bmap;
3410 stp->st_deny_bmap = open_stp->st_deny_bmap;
3411 stp->st_openstp = open_stp;
3418 check_lock_length(u64 offset, u64 length)
3420 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3421 LOFF_OVERFLOW(offset, length)));
3428 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3429 struct nfsd4_lock *lock)
3431 struct nfs4_stateowner *open_sop = NULL;
3432 struct nfs4_stateowner *lock_sop = NULL;
3433 struct nfs4_stateid *lock_stp;
3435 struct file_lock file_lock;
3436 struct file_lock conflock;
3438 unsigned int strhashval;
3442 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3443 (long long) lock->lk_offset,
3444 (long long) lock->lk_length);
3446 if (check_lock_length(lock->lk_offset, lock->lk_length))
3447 return nfserr_inval;
3449 if ((status = fh_verify(rqstp, &cstate->current_fh,
3450 S_IFREG, NFSD_MAY_LOCK))) {
3451 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3457 if (lock->lk_is_new) {
3459 * Client indicates that this is a new lockowner.
3460 * Use open owner and open stateid to create lock owner and
3463 struct nfs4_stateid *open_stp = NULL;
3464 struct nfs4_file *fp;
3466 status = nfserr_stale_clientid;
3467 if (!nfsd4_has_session(cstate) &&
3468 STALE_CLIENTID(&lock->lk_new_clientid))
3471 /* validate and update open stateid and open seqid */
3472 status = nfs4_preprocess_seqid_op(cstate,
3473 lock->lk_new_open_seqid,
3474 &lock->lk_new_open_stateid,
3476 &lock->lk_replay_owner, &open_stp,
3480 open_sop = lock->lk_replay_owner;
3481 /* create lockowner and lock stateid */
3482 fp = open_stp->st_file;
3483 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3484 open_sop->so_client->cl_clientid.cl_id,
3485 &lock->v.new.owner);
3486 /* XXX: Do we need to check for duplicate stateowners on
3487 * the same file, or should they just be allowed (and
3488 * create new stateids)? */
3489 status = nfserr_resource;
3490 lock_sop = alloc_init_lock_stateowner(strhashval,
3491 open_sop->so_client, open_stp, lock);
3492 if (lock_sop == NULL)
3494 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3495 if (lock_stp == NULL)
3498 /* lock (lock owner + lock stateid) already exists */
3499 status = nfs4_preprocess_seqid_op(cstate,
3500 lock->lk_old_lock_seqid,
3501 &lock->lk_old_lock_stateid,
3503 &lock->lk_replay_owner, &lock_stp, lock);
3506 lock_sop = lock->lk_replay_owner;
3508 /* lock->lk_replay_owner and lock_stp have been created or found */
3509 filp = lock_stp->st_vfs_file;
3511 status = nfserr_grace;
3512 if (locks_in_grace() && !lock->lk_reclaim)
3514 status = nfserr_no_grace;
3515 if (!locks_in_grace() && lock->lk_reclaim)
3518 locks_init_lock(&file_lock);
3519 switch (lock->lk_type) {
3522 file_lock.fl_type = F_RDLCK;
3526 case NFS4_WRITEW_LT:
3527 file_lock.fl_type = F_WRLCK;
3531 status = nfserr_inval;
3534 file_lock.fl_owner = (fl_owner_t)lock_sop;
3535 file_lock.fl_pid = current->tgid;
3536 file_lock.fl_file = filp;
3537 file_lock.fl_flags = FL_POSIX;
3538 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3540 file_lock.fl_start = lock->lk_offset;
3541 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3542 nfs4_transform_lock_offset(&file_lock);
3545 * Try to lock the file in the VFS.
3546 * Note: locks.c uses the BKL to protect the inode's lock list.
3549 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3551 case 0: /* success! */
3552 update_stateid(&lock_stp->st_stateid);
3553 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3557 case (EAGAIN): /* conflock holds conflicting lock */
3558 status = nfserr_denied;
3559 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3560 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3563 status = nfserr_deadlock;
3566 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3567 status = nfserr_resource;
3571 if (status && lock->lk_is_new && lock_sop)
3572 release_lockowner(lock_sop);
3573 if (lock->lk_replay_owner) {
3574 nfs4_get_stateowner(lock->lk_replay_owner);
3575 cstate->replay_owner = lock->lk_replay_owner;
3577 nfs4_unlock_state();
3582 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3583 * so we do a temporary open here just to get an open file to pass to
3584 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3587 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3592 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3595 err = vfs_test_lock(file, lock);
3604 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3605 struct nfsd4_lockt *lockt)
3607 struct inode *inode;
3608 struct file_lock file_lock;
3612 if (locks_in_grace())
3613 return nfserr_grace;
3615 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3616 return nfserr_inval;
3618 lockt->lt_stateowner = NULL;
3621 status = nfserr_stale_clientid;
3622 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3625 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3626 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3627 if (status == nfserr_symlink)
3628 status = nfserr_inval;
3632 inode = cstate->current_fh.fh_dentry->d_inode;
3633 locks_init_lock(&file_lock);
3634 switch (lockt->lt_type) {
3637 file_lock.fl_type = F_RDLCK;
3640 case NFS4_WRITEW_LT:
3641 file_lock.fl_type = F_WRLCK;
3644 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3645 status = nfserr_inval;
3649 lockt->lt_stateowner = find_lockstateowner_str(inode,
3650 &lockt->lt_clientid, &lockt->lt_owner);
3651 if (lockt->lt_stateowner)
3652 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3653 file_lock.fl_pid = current->tgid;
3654 file_lock.fl_flags = FL_POSIX;
3656 file_lock.fl_start = lockt->lt_offset;
3657 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3659 nfs4_transform_lock_offset(&file_lock);
3662 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3664 status = nfserrno(error);
3667 if (file_lock.fl_type != F_UNLCK) {
3668 status = nfserr_denied;
3669 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3672 nfs4_unlock_state();
3677 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3678 struct nfsd4_locku *locku)
3680 struct nfs4_stateid *stp;
3681 struct file *filp = NULL;
3682 struct file_lock file_lock;
3686 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3687 (long long) locku->lu_offset,
3688 (long long) locku->lu_length);
3690 if (check_lock_length(locku->lu_offset, locku->lu_length))
3691 return nfserr_inval;
3695 if ((status = nfs4_preprocess_seqid_op(cstate,
3699 &locku->lu_stateowner, &stp, NULL)))
3702 filp = stp->st_vfs_file;
3704 locks_init_lock(&file_lock);
3705 file_lock.fl_type = F_UNLCK;
3706 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3707 file_lock.fl_pid = current->tgid;
3708 file_lock.fl_file = filp;
3709 file_lock.fl_flags = FL_POSIX;
3710 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3711 file_lock.fl_start = locku->lu_offset;
3713 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3714 nfs4_transform_lock_offset(&file_lock);
3717 * Try to unlock the file in the VFS.
3719 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3721 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3725 * OK, unlock succeeded; the only thing left to do is update the stateid.
3727 update_stateid(&stp->st_stateid);
3728 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3731 if (locku->lu_stateowner) {
3732 nfs4_get_stateowner(locku->lu_stateowner);
3733 cstate->replay_owner = locku->lu_stateowner;
3735 nfs4_unlock_state();
3739 status = nfserrno(err);
3745 * 1: locks held by lockowner
3746 * 0: no locks held by lockowner
3749 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3751 struct file_lock **flpp;
3752 struct inode *inode = filp->f_path.dentry->d_inode;
3756 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3757 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3768 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3769 struct nfsd4_compound_state *cstate,
3770 struct nfsd4_release_lockowner *rlockowner)
3772 clientid_t *clid = &rlockowner->rl_clientid;
3773 struct nfs4_stateowner *sop;
3774 struct nfs4_stateid *stp;
3775 struct xdr_netobj *owner = &rlockowner->rl_owner;
3776 struct list_head matches;
3780 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3781 clid->cl_boot, clid->cl_id);
3783 /* XXX check for lease expiration */
3785 status = nfserr_stale_clientid;
3786 if (STALE_CLIENTID(clid))
3791 status = nfserr_locks_held;
3792 /* XXX: we're doing a linear search through all the lockowners.
3793 * Yipes! For now we'll just hope clients aren't really using
3794 * release_lockowner much, but eventually we have to fix these
3795 * data structures. */
3796 INIT_LIST_HEAD(&matches);
3797 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3798 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3799 if (!same_owner_str(sop, owner, clid))
3801 list_for_each_entry(stp, &sop->so_stateids,
3803 if (check_for_locks(stp->st_vfs_file, sop))
3805 /* Note: so_perclient unused for lockowners,
3806 * so it's OK to fool with here. */
3807 list_add(&sop->so_perclient, &matches);
3811 /* Clients probably won't expect us to return with some (but not all)
3812 * of the lockowner state released; so don't release any until all
3813 * have been checked. */
3815 while (!list_empty(&matches)) {
3816 sop = list_entry(matches.next, struct nfs4_stateowner,
3818 /* unhash_stateowner deletes so_perclient only
3819 * for openowners. */
3820 list_del(&sop->so_perclient);
3821 release_lockowner(sop);
3824 nfs4_unlock_state();
3828 static inline struct nfs4_client_reclaim *
3831 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3835 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3837 unsigned int strhashval = clientstr_hashval(name);
3838 struct nfs4_client *clp;
3840 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3845 * failure => all reset bets are off, nfserr_no_grace...
3848 nfs4_client_to_reclaim(const char *name)
3850 unsigned int strhashval;
3851 struct nfs4_client_reclaim *crp = NULL;
3853 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3854 crp = alloc_reclaim();
3857 strhashval = clientstr_hashval(name);
3858 INIT_LIST_HEAD(&crp->cr_strhash);
3859 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3860 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3861 reclaim_str_hashtbl_size++;
3866 nfs4_release_reclaim(void)
3868 struct nfs4_client_reclaim *crp = NULL;
3871 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3872 while (!list_empty(&reclaim_str_hashtbl[i])) {
3873 crp = list_entry(reclaim_str_hashtbl[i].next,
3874 struct nfs4_client_reclaim, cr_strhash);
3875 list_del(&crp->cr_strhash);
3877 reclaim_str_hashtbl_size--;
3880 BUG_ON(reclaim_str_hashtbl_size);
3884 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3885 static struct nfs4_client_reclaim *
3886 nfs4_find_reclaim_client(clientid_t *clid)
3888 unsigned int strhashval;
3889 struct nfs4_client *clp;
3890 struct nfs4_client_reclaim *crp = NULL;
3893 /* find clientid in conf_id_hashtbl */
3894 clp = find_confirmed_client(clid);
3898 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3899 clp->cl_name.len, clp->cl_name.data,
3902 /* find clp->cl_name in reclaim_str_hashtbl */
3903 strhashval = clientstr_hashval(clp->cl_recdir);
3904 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3905 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3913 * Called from OPEN. Look for clientid in reclaim list.
3916 nfs4_check_open_reclaim(clientid_t *clid)
3918 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3921 /* initialization to perform at module load time: */
3924 nfs4_state_init(void)
3928 status = nfsd4_init_slabs();
3931 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3932 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3933 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3934 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3935 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3936 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3938 for (i = 0; i < SESSION_HASH_SIZE; i++)
3939 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3940 for (i = 0; i < FILE_HASH_SIZE; i++) {
3941 INIT_LIST_HEAD(&file_hashtbl[i]);
3943 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3944 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3945 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3947 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3948 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3949 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3951 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3952 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3953 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3955 memset(&onestateid, ~0, sizeof(stateid_t));
3956 INIT_LIST_HEAD(&close_lru);
3957 INIT_LIST_HEAD(&client_lru);
3958 INIT_LIST_HEAD(&del_recall_lru);
3959 reclaim_str_hashtbl_size = 0;
3964 nfsd4_load_reboot_recovery_data(void)
3969 nfsd4_init_recdir(user_recovery_dirname);
3970 status = nfsd4_recdir_load();
3971 nfs4_unlock_state();
3973 printk("NFSD: Failure reading reboot recovery data\n");
3977 get_nfs4_grace_period(void)
3979 return max(user_lease_time, lease_time) * HZ;
3983 * Since the lifetime of a delegation isn't limited to that of an open, a
3984 * client may quite reasonably hang on to a delegation as long as it has
3985 * the inode cached. This becomes an obvious problem the first time a
3986 * client's inode cache approaches the size of the server's total memory.
3988 * For now we avoid this problem by imposing a hard limit on the number
3989 * of delegations, which varies according to the server's memory size.
3992 set_max_delegations(void)
3995 * Allow at most 4 delegations per megabyte of RAM. Quick
3996 * estimates suggest that in the worst case (where every delegation
3997 * is for a different inode), a delegation could take about 1.5K,
3998 * giving a worst case usage of about 6% of memory.
4000 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4003 /* initialization to perform when the nfsd service is started: */
4006 __nfs4_state_start(void)
4008 unsigned long grace_time;
4010 boot_time = get_seconds();
4011 grace_time = get_nfs4_grace_period();
4012 lease_time = user_lease_time;
4013 locks_start_grace(&nfsd4_manager);
4014 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4016 laundry_wq = create_singlethread_workqueue("nfsd4");
4017 if (laundry_wq == NULL)
4019 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4020 set_max_delegations();
4021 return set_callback_cred();
4025 nfs4_state_start(void)
4031 nfsd4_load_reboot_recovery_data();
4032 ret = __nfs4_state_start();
4040 nfs4_lease_time(void)
4046 __nfs4_state_shutdown(void)
4049 struct nfs4_client *clp = NULL;
4050 struct nfs4_delegation *dp = NULL;
4051 struct list_head *pos, *next, reaplist;
4053 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4054 while (!list_empty(&conf_id_hashtbl[i])) {
4055 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4058 while (!list_empty(&unconf_str_hashtbl[i])) {
4059 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4063 INIT_LIST_HEAD(&reaplist);
4064 spin_lock(&recall_lock);
4065 list_for_each_safe(pos, next, &del_recall_lru) {
4066 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4067 list_move(&dp->dl_recall_lru, &reaplist);
4069 spin_unlock(&recall_lock);
4070 list_for_each_safe(pos, next, &reaplist) {
4071 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4072 list_del_init(&dp->dl_recall_lru);
4073 unhash_delegation(dp);
4076 nfsd4_shutdown_recdir();
4081 nfs4_state_shutdown(void)
4083 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4084 destroy_workqueue(laundry_wq);
4085 locks_end_grace(&nfsd4_manager);
4087 nfs4_release_reclaim();
4088 __nfs4_state_shutdown();
4089 nfs4_unlock_state();
4093 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4094 * accessed when nfsd is starting.
4097 nfs4_set_recdir(char *recdir)
4099 strcpy(user_recovery_dirname, recdir);
4103 * Change the NFSv4 recovery directory to recdir.
4106 nfs4_reset_recoverydir(char *recdir)
4111 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4115 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4116 nfs4_set_recdir(recdir);
4124 nfs4_recoverydir(void)
4126 return user_recovery_dirname;
4130 * Called when leasetime is changed.
4132 * The only way the protocol gives us to handle on-the-fly lease changes is to
4133 * simulate a reboot. Instead of doing that, we just wait till the next time
4134 * we start to register any changes in lease time. If the administrator
4135 * really wants to change the lease time *now*, they can go ahead and bring
4136 * nfsd down and then back up again after changing the lease time.
4138 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4139 * when nfsd is starting
4142 nfs4_reset_lease(time_t leasetime)
4144 user_lease_time = leasetime;