2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
36 #include <linux/smp_lock.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
45 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 time_t nfsd4_lease = 90; /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
55 static stateid_t zerostateid; /* bits all 0 */
56 static stateid_t onestateid; /* bits all 1 */
57 static u64 current_sessionid = 1;
59 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
60 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
62 /* forward declarations */
63 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
64 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
65 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
66 static void nfs4_set_recdir(char *recdir);
70 /* Currently used for almost all code touching nfsv4 state: */
71 static DEFINE_MUTEX(client_mutex);
74 * Currently used for the del_recall_lru and file hash table. In an
75 * effort to decrease the scope of the client_mutex, this spinlock may
76 * eventually cover more:
78 static DEFINE_SPINLOCK(recall_lock);
80 static struct kmem_cache *stateowner_slab = NULL;
81 static struct kmem_cache *file_slab = NULL;
82 static struct kmem_cache *stateid_slab = NULL;
83 static struct kmem_cache *deleg_slab = NULL;
88 mutex_lock(&client_mutex);
92 nfs4_unlock_state(void)
94 mutex_unlock(&client_mutex);
98 opaque_hashval(const void *ptr, int nbytes)
100 unsigned char *cptr = (unsigned char *) ptr;
110 static struct list_head del_recall_lru;
113 put_nfs4_file(struct nfs4_file *fi)
115 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
116 list_del(&fi->fi_hash);
117 spin_unlock(&recall_lock);
119 kmem_cache_free(file_slab, fi);
124 get_nfs4_file(struct nfs4_file *fi)
126 atomic_inc(&fi->fi_ref);
129 static int num_delegations;
130 unsigned int max_delegations;
133 * Open owner state (share locks)
136 /* hash tables for nfs4_stateowner */
137 #define OWNER_HASH_BITS 8
138 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
139 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
141 #define ownerid_hashval(id) \
142 ((id) & OWNER_HASH_MASK)
143 #define ownerstr_hashval(clientid, ownername) \
144 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
146 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
147 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
149 /* hash table for nfs4_file */
150 #define FILE_HASH_BITS 8
151 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
152 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
153 /* hash table for (open)nfs4_stateid */
154 #define STATEID_HASH_BITS 10
155 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
156 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
158 #define file_hashval(x) \
159 hash_ptr(x, FILE_HASH_BITS)
160 #define stateid_hashval(owner_id, file_id) \
161 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
163 static struct list_head file_hashtbl[FILE_HASH_SIZE];
164 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
166 static struct nfs4_delegation *
167 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
169 struct nfs4_delegation *dp;
170 struct nfs4_file *fp = stp->st_file;
171 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
173 dprintk("NFSD alloc_init_deleg\n");
174 if (fp->fi_had_conflict)
176 if (num_delegations > max_delegations)
178 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
182 INIT_LIST_HEAD(&dp->dl_perfile);
183 INIT_LIST_HEAD(&dp->dl_perclnt);
184 INIT_LIST_HEAD(&dp->dl_recall_lru);
189 get_file(stp->st_vfs_file);
190 dp->dl_vfs_file = stp->st_vfs_file;
192 dp->dl_ident = cb->cb_ident;
193 dp->dl_stateid.si_boot = get_seconds();
194 dp->dl_stateid.si_stateownerid = current_delegid++;
195 dp->dl_stateid.si_fileid = 0;
196 dp->dl_stateid.si_generation = 0;
197 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
199 atomic_set(&dp->dl_count, 1);
200 list_add(&dp->dl_perfile, &fp->fi_delegations);
201 list_add(&dp->dl_perclnt, &clp->cl_delegations);
202 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
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 /* client_lock protects the client lru list and session hash table */
254 static DEFINE_SPINLOCK(client_lock);
256 /* Hash tables for nfs4_clientid state */
257 #define CLIENT_HASH_BITS 4
258 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
259 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
261 #define clientid_hashval(id) \
262 ((id) & CLIENT_HASH_MASK)
263 #define clientstr_hashval(name) \
264 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
266 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
267 * used in reboot/reset lease grace period processing
269 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
270 * setclientid_confirmed info.
272 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
275 * client_lru holds client queue ordered by nfs4_client.cl_time
278 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
279 * for last close replay.
281 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
282 static int reclaim_str_hashtbl_size = 0;
283 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
284 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
285 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
286 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
287 static struct list_head client_lru;
288 static struct list_head close_lru;
290 static void unhash_generic_stateid(struct nfs4_stateid *stp)
292 list_del(&stp->st_hash);
293 list_del(&stp->st_perfile);
294 list_del(&stp->st_perstateowner);
297 static void free_generic_stateid(struct nfs4_stateid *stp)
299 put_nfs4_file(stp->st_file);
300 kmem_cache_free(stateid_slab, stp);
303 static void release_lock_stateid(struct nfs4_stateid *stp)
305 unhash_generic_stateid(stp);
306 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
307 free_generic_stateid(stp);
310 static void unhash_lockowner(struct nfs4_stateowner *sop)
312 struct nfs4_stateid *stp;
314 list_del(&sop->so_idhash);
315 list_del(&sop->so_strhash);
316 list_del(&sop->so_perstateid);
317 while (!list_empty(&sop->so_stateids)) {
318 stp = list_first_entry(&sop->so_stateids,
319 struct nfs4_stateid, st_perstateowner);
320 release_lock_stateid(stp);
324 static void release_lockowner(struct nfs4_stateowner *sop)
326 unhash_lockowner(sop);
327 nfs4_put_stateowner(sop);
331 release_stateid_lockowners(struct nfs4_stateid *open_stp)
333 struct nfs4_stateowner *lock_sop;
335 while (!list_empty(&open_stp->st_lockowners)) {
336 lock_sop = list_entry(open_stp->st_lockowners.next,
337 struct nfs4_stateowner, so_perstateid);
338 /* list_del(&open_stp->st_lockowners); */
339 BUG_ON(lock_sop->so_is_open_owner);
340 release_lockowner(lock_sop);
344 static void release_open_stateid(struct nfs4_stateid *stp)
346 unhash_generic_stateid(stp);
347 release_stateid_lockowners(stp);
348 nfsd_close(stp->st_vfs_file);
349 free_generic_stateid(stp);
352 static void unhash_openowner(struct nfs4_stateowner *sop)
354 struct nfs4_stateid *stp;
356 list_del(&sop->so_idhash);
357 list_del(&sop->so_strhash);
358 list_del(&sop->so_perclient);
359 list_del(&sop->so_perstateid); /* XXX: necessary? */
360 while (!list_empty(&sop->so_stateids)) {
361 stp = list_first_entry(&sop->so_stateids,
362 struct nfs4_stateid, st_perstateowner);
363 release_open_stateid(stp);
367 static void release_openowner(struct nfs4_stateowner *sop)
369 unhash_openowner(sop);
370 list_del(&sop->so_close_lru);
371 nfs4_put_stateowner(sop);
374 #define SESSION_HASH_SIZE 512
375 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
378 hash_sessionid(struct nfs4_sessionid *sessionid)
380 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
382 return sid->sequence % SESSION_HASH_SIZE;
386 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
388 u32 *ptr = (u32 *)(&sessionid->data[0]);
389 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
393 gen_sessionid(struct nfsd4_session *ses)
395 struct nfs4_client *clp = ses->se_client;
396 struct nfsd4_sessionid *sid;
398 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
399 sid->clientid = clp->cl_clientid;
400 sid->sequence = current_sessionid++;
405 * The protocol defines ca_maxresponssize_cached to include the size of
406 * the rpc header, but all we need to cache is the data starting after
407 * the end of the initial SEQUENCE operation--the rest we regenerate
408 * each time. Therefore we can advertise a ca_maxresponssize_cached
409 * value that is the number of bytes in our cache plus a few additional
410 * bytes. In order to stay on the safe side, and not promise more than
411 * we can cache, those additional bytes must be the minimum possible: 24
412 * bytes of rpc header (xid through accept state, with AUTH_NULL
413 * verifier), 12 for the compound header (with zero-length tag), and 44
414 * for the SEQUENCE op response:
416 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
419 * Give the client the number of ca_maxresponsesize_cached slots it
420 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
421 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
422 * than NFSD_MAX_SLOTS_PER_SESSION.
424 * If we run out of reserved DRC memory we should (up to a point)
425 * re-negotiate active sessions and reduce their slot usage to make
426 * rooom for new connections. For now we just fail the create session.
428 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
430 int mem, size = fchan->maxresp_cached;
432 if (fchan->maxreqs < 1)
435 if (size < NFSD_MIN_HDR_SEQ_SZ)
436 size = NFSD_MIN_HDR_SEQ_SZ;
437 size -= NFSD_MIN_HDR_SEQ_SZ;
438 if (size > NFSD_SLOT_CACHE_SIZE)
439 size = NFSD_SLOT_CACHE_SIZE;
441 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
442 mem = fchan->maxreqs * size;
443 if (mem > NFSD_MAX_MEM_PER_SESSION) {
444 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
445 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
446 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
447 mem = fchan->maxreqs * size;
450 spin_lock(&nfsd_drc_lock);
451 /* bound the total session drc memory ussage */
452 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
453 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
454 mem = fchan->maxreqs * size;
456 nfsd_drc_mem_used += mem;
457 spin_unlock(&nfsd_drc_lock);
459 if (fchan->maxreqs == 0)
460 return nfserr_serverfault;
462 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
467 * fchan holds the client values on input, and the server values on output
468 * sv_max_mesg is the maximum payload plus one page for overhead.
470 static int init_forechannel_attrs(struct svc_rqst *rqstp,
471 struct nfsd4_channel_attrs *session_fchan,
472 struct nfsd4_channel_attrs *fchan)
475 __u32 maxcount = nfsd_serv->sv_max_mesg;
477 /* headerpadsz set to zero in encode routine */
479 /* Use the client's max request and max response size if possible */
480 if (fchan->maxreq_sz > maxcount)
481 fchan->maxreq_sz = maxcount;
482 session_fchan->maxreq_sz = fchan->maxreq_sz;
484 if (fchan->maxresp_sz > maxcount)
485 fchan->maxresp_sz = maxcount;
486 session_fchan->maxresp_sz = fchan->maxresp_sz;
488 /* Use the client's maxops if possible */
489 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
490 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
491 session_fchan->maxops = fchan->maxops;
493 /* FIXME: Error means no more DRC pages so the server should
494 * recover pages from existing sessions. For now fail session
497 status = set_forechannel_drc_size(fchan);
499 session_fchan->maxresp_cached = fchan->maxresp_cached;
500 session_fchan->maxreqs = fchan->maxreqs;
502 dprintk("%s status %d\n", __func__, status);
507 free_session_slots(struct nfsd4_session *ses)
511 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
512 kfree(ses->se_slots[i]);
516 * We don't actually need to cache the rpc and session headers, so we
517 * can allocate a little less for each slot:
519 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
521 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
525 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
526 struct nfsd4_create_session *cses)
528 struct nfsd4_session *new, tmp;
529 struct nfsd4_slot *sp;
530 int idx, slotsize, cachesize, i;
533 memset(&tmp, 0, sizeof(tmp));
535 /* FIXME: For now, we just accept the client back channel attributes. */
536 tmp.se_bchannel = cses->back_channel;
537 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
538 &cses->fore_channel);
542 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
543 + sizeof(struct nfsd4_session) > PAGE_SIZE);
545 status = nfserr_serverfault;
546 /* allocate struct nfsd4_session and slot table pointers in one piece */
547 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
548 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
552 memcpy(new, &tmp, sizeof(*new));
554 /* allocate each struct nfsd4_slot and data cache in one piece */
555 cachesize = slot_bytes(&new->se_fchannel);
556 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
557 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
560 new->se_slots[i] = sp;
563 new->se_client = clp;
565 idx = hash_sessionid(&new->se_sessionid);
566 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
567 NFS4_MAX_SESSIONID_LEN);
569 new->se_flags = cses->flags;
570 kref_init(&new->se_ref);
571 spin_lock(&client_lock);
572 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
573 list_add(&new->se_perclnt, &clp->cl_sessions);
574 spin_unlock(&client_lock);
580 free_session_slots(new);
585 /* caller must hold client_lock */
586 static struct nfsd4_session *
587 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
589 struct nfsd4_session *elem;
592 dump_sessionid(__func__, sessionid);
593 idx = hash_sessionid(sessionid);
594 dprintk("%s: idx is %d\n", __func__, idx);
595 /* Search in the appropriate list */
596 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
597 dump_sessionid("list traversal", &elem->se_sessionid);
598 if (!memcmp(elem->se_sessionid.data, sessionid->data,
599 NFS4_MAX_SESSIONID_LEN)) {
604 dprintk("%s: session not found\n", __func__);
608 /* caller must hold client_lock */
610 unhash_session(struct nfsd4_session *ses)
612 list_del(&ses->se_hash);
613 list_del(&ses->se_perclnt);
617 free_session(struct kref *kref)
619 struct nfsd4_session *ses;
622 ses = container_of(kref, struct nfsd4_session, se_ref);
623 spin_lock(&nfsd_drc_lock);
624 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
625 nfsd_drc_mem_used -= mem;
626 spin_unlock(&nfsd_drc_lock);
627 free_session_slots(ses);
631 /* must be called under the client_lock */
633 renew_client_locked(struct nfs4_client *clp)
636 * Move client to the end to the LRU list.
638 dprintk("renewing client (clientid %08x/%08x)\n",
639 clp->cl_clientid.cl_boot,
640 clp->cl_clientid.cl_id);
641 list_move_tail(&clp->cl_lru, &client_lru);
642 clp->cl_time = get_seconds();
646 renew_client(struct nfs4_client *clp)
648 spin_lock(&client_lock);
649 renew_client_locked(clp);
650 spin_unlock(&client_lock);
653 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
655 STALE_CLIENTID(clientid_t *clid)
657 if (clid->cl_boot == boot_time)
659 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
660 clid->cl_boot, clid->cl_id, boot_time);
665 * XXX Should we use a slab cache ?
666 * This type of memory management is somewhat inefficient, but we use it
667 * anyway since SETCLIENTID is not a common operation.
669 static struct nfs4_client *alloc_client(struct xdr_netobj name)
671 struct nfs4_client *clp;
673 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
676 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
677 if (clp->cl_name.data == NULL) {
681 memcpy(clp->cl_name.data, name.data, name.len);
682 clp->cl_name.len = name.len;
687 free_client(struct nfs4_client *clp)
689 if (clp->cl_cred.cr_group_info)
690 put_group_info(clp->cl_cred.cr_group_info);
691 kfree(clp->cl_principal);
692 kfree(clp->cl_name.data);
696 /* must be called under the client_lock */
698 unhash_client_locked(struct nfs4_client *clp)
700 list_del(&clp->cl_lru);
701 while (!list_empty(&clp->cl_sessions)) {
702 struct nfsd4_session *ses;
703 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
706 nfsd4_put_session(ses);
711 expire_client(struct nfs4_client *clp)
713 struct nfs4_stateowner *sop;
714 struct nfs4_delegation *dp;
715 struct list_head reaplist;
717 INIT_LIST_HEAD(&reaplist);
718 spin_lock(&recall_lock);
719 while (!list_empty(&clp->cl_delegations)) {
720 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
721 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
723 list_del_init(&dp->dl_perclnt);
724 list_move(&dp->dl_recall_lru, &reaplist);
726 spin_unlock(&recall_lock);
727 while (!list_empty(&reaplist)) {
728 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
729 list_del_init(&dp->dl_recall_lru);
730 unhash_delegation(dp);
732 while (!list_empty(&clp->cl_openowners)) {
733 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
734 release_openowner(sop);
736 nfsd4_set_callback_client(clp, NULL);
737 if (clp->cl_cb_conn.cb_xprt)
738 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
739 list_del(&clp->cl_idhash);
740 list_del(&clp->cl_strhash);
741 spin_lock(&client_lock);
742 unhash_client_locked(clp);
743 spin_unlock(&client_lock);
747 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
749 memcpy(target->cl_verifier.data, source->data,
750 sizeof(target->cl_verifier.data));
753 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
755 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
756 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
759 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
761 target->cr_uid = source->cr_uid;
762 target->cr_gid = source->cr_gid;
763 target->cr_group_info = source->cr_group_info;
764 get_group_info(target->cr_group_info);
767 static int same_name(const char *n1, const char *n2)
769 return 0 == memcmp(n1, n2, HEXDIR_LEN);
773 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
775 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
779 same_clid(clientid_t *cl1, clientid_t *cl2)
781 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
784 /* XXX what about NGROUP */
786 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
788 return cr1->cr_uid == cr2->cr_uid;
791 static void gen_clid(struct nfs4_client *clp)
793 static u32 current_clientid = 1;
795 clp->cl_clientid.cl_boot = boot_time;
796 clp->cl_clientid.cl_id = current_clientid++;
799 static void gen_confirm(struct nfs4_client *clp)
804 p = (u32 *)clp->cl_confirm.data;
805 *p++ = get_seconds();
809 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
810 struct svc_rqst *rqstp, nfs4_verifier *verf)
812 struct nfs4_client *clp;
813 struct sockaddr *sa = svc_addr(rqstp);
816 clp = alloc_client(name);
820 princ = svc_gss_principal(rqstp);
822 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
823 if (clp->cl_principal == NULL) {
829 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
830 atomic_set(&clp->cl_cb_set, 0);
831 INIT_LIST_HEAD(&clp->cl_idhash);
832 INIT_LIST_HEAD(&clp->cl_strhash);
833 INIT_LIST_HEAD(&clp->cl_openowners);
834 INIT_LIST_HEAD(&clp->cl_delegations);
835 INIT_LIST_HEAD(&clp->cl_sessions);
836 INIT_LIST_HEAD(&clp->cl_lru);
837 clear_bit(0, &clp->cl_cb_slot_busy);
838 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
839 copy_verf(clp, verf);
840 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
841 clp->cl_flavor = rqstp->rq_flavor;
842 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
848 static int check_name(struct xdr_netobj name)
852 if (name.len > NFS4_OPAQUE_LIMIT) {
853 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
860 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
862 unsigned int idhashval;
864 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
865 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
866 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
871 move_to_confirmed(struct nfs4_client *clp)
873 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
874 unsigned int strhashval;
876 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
877 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
878 strhashval = clientstr_hashval(clp->cl_recdir);
879 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
883 static struct nfs4_client *
884 find_confirmed_client(clientid_t *clid)
886 struct nfs4_client *clp;
887 unsigned int idhashval = clientid_hashval(clid->cl_id);
889 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
890 if (same_clid(&clp->cl_clientid, clid))
896 static struct nfs4_client *
897 find_unconfirmed_client(clientid_t *clid)
899 struct nfs4_client *clp;
900 unsigned int idhashval = clientid_hashval(clid->cl_id);
902 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
903 if (same_clid(&clp->cl_clientid, clid))
910 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
911 * parameter. Matching is based on the fact the at least one of the
912 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
914 * FIXME: we need to unify the clientid namespaces for nfsv4.x
915 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
916 * and SET_CLIENTID{,_CONFIRM}
919 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
921 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
922 return use_exchange_id == has_exchange_flags;
925 static struct nfs4_client *
926 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
927 bool use_exchange_id)
929 struct nfs4_client *clp;
931 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
932 if (same_name(clp->cl_recdir, dname) &&
933 match_clientid_establishment(clp, use_exchange_id))
939 static struct nfs4_client *
940 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
941 bool use_exchange_id)
943 struct nfs4_client *clp;
945 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
946 if (same_name(clp->cl_recdir, dname) &&
947 match_clientid_establishment(clp, use_exchange_id))
954 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
956 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
957 unsigned short expected_family;
959 /* Currently, we only support tcp and tcp6 for the callback channel */
960 if (se->se_callback_netid_len == 3 &&
961 !memcmp(se->se_callback_netid_val, "tcp", 3))
962 expected_family = AF_INET;
963 else if (se->se_callback_netid_len == 4 &&
964 !memcmp(se->se_callback_netid_val, "tcp6", 4))
965 expected_family = AF_INET6;
969 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
970 se->se_callback_addr_len,
971 (struct sockaddr *) &cb->cb_addr,
972 sizeof(cb->cb_addr));
974 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
977 if (cb->cb_addr.ss_family == AF_INET6)
978 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
980 cb->cb_minorversion = 0;
981 cb->cb_prog = se->se_callback_prog;
982 cb->cb_ident = se->se_callback_ident;
985 cb->cb_addr.ss_family = AF_UNSPEC;
987 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
988 "will not receive delegations\n",
989 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
995 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
998 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1000 struct nfsd4_slot *slot = resp->cstate.slot;
1003 dprintk("--> %s slot %p\n", __func__, slot);
1005 slot->sl_opcnt = resp->opcnt;
1006 slot->sl_status = resp->cstate.status;
1008 if (nfsd4_not_cached(resp)) {
1009 slot->sl_datalen = 0;
1012 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1013 base = (char *)resp->cstate.datap -
1014 (char *)resp->xbuf->head[0].iov_base;
1015 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1017 WARN("%s: sessions DRC could not cache compound\n", __func__);
1022 * Encode the replay sequence operation from the slot values.
1023 * If cachethis is FALSE encode the uncached rep error on the next
1024 * operation which sets resp->p and increments resp->opcnt for
1025 * nfs4svc_encode_compoundres.
1029 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1030 struct nfsd4_compoundres *resp)
1032 struct nfsd4_op *op;
1033 struct nfsd4_slot *slot = resp->cstate.slot;
1035 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1036 resp->opcnt, resp->cstate.slot->sl_cachethis);
1038 /* Encode the replayed sequence operation */
1039 op = &args->ops[resp->opcnt - 1];
1040 nfsd4_encode_operation(resp, op);
1042 /* Return nfserr_retry_uncached_rep in next operation. */
1043 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1044 op = &args->ops[resp->opcnt++];
1045 op->status = nfserr_retry_uncached_rep;
1046 nfsd4_encode_operation(resp, op);
1052 * The sequence operation is not cached because we can use the slot and
1056 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1057 struct nfsd4_sequence *seq)
1059 struct nfsd4_slot *slot = resp->cstate.slot;
1062 dprintk("--> %s slot %p\n", __func__, slot);
1064 /* Either returns 0 or nfserr_retry_uncached */
1065 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1066 if (status == nfserr_retry_uncached_rep)
1069 /* The sequence operation has been encoded, cstate->datap set. */
1070 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1072 resp->opcnt = slot->sl_opcnt;
1073 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1074 status = slot->sl_status;
1080 * Set the exchange_id flags returned by the server.
1083 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1085 /* pNFS is not supported */
1086 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1088 /* Referrals are supported, Migration is not. */
1089 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1091 /* set the wire flags to return to client. */
1092 clid->flags = new->cl_exchange_flags;
1096 nfsd4_exchange_id(struct svc_rqst *rqstp,
1097 struct nfsd4_compound_state *cstate,
1098 struct nfsd4_exchange_id *exid)
1100 struct nfs4_client *unconf, *conf, *new;
1102 unsigned int strhashval;
1103 char dname[HEXDIR_LEN];
1104 char addr_str[INET6_ADDRSTRLEN];
1105 nfs4_verifier verf = exid->verifier;
1106 struct sockaddr *sa = svc_addr(rqstp);
1108 rpc_ntop(sa, addr_str, sizeof(addr_str));
1109 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1110 "ip_addr=%s flags %x, spa_how %d\n",
1111 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1112 addr_str, exid->flags, exid->spa_how);
1114 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1115 return nfserr_inval;
1117 /* Currently only support SP4_NONE */
1118 switch (exid->spa_how) {
1122 return nfserr_encr_alg_unsupp;
1124 BUG(); /* checked by xdr code */
1126 return nfserr_serverfault; /* no excuse :-/ */
1129 status = nfs4_make_rec_clidname(dname, &exid->clname);
1134 strhashval = clientstr_hashval(dname);
1139 conf = find_confirmed_client_by_str(dname, strhashval, true);
1141 if (!same_verf(&verf, &conf->cl_verifier)) {
1142 /* 18.35.4 case 8 */
1143 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1144 status = nfserr_not_same;
1147 /* Client reboot: destroy old state */
1148 expire_client(conf);
1151 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1152 /* 18.35.4 case 9 */
1153 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1154 status = nfserr_perm;
1157 expire_client(conf);
1161 * Set bit when the owner id and verifier map to an already
1162 * confirmed client id (18.35.3).
1164 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1167 * Falling into 18.35.4 case 2, possible router replay.
1168 * Leave confirmed record intact and return same result.
1170 copy_verf(conf, &verf);
1175 /* 18.35.4 case 7 */
1176 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1177 status = nfserr_noent;
1181 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1184 * Possible retry or client restart. Per 18.35.4 case 4,
1185 * a new unconfirmed record should be generated regardless
1186 * of whether any properties have changed.
1188 expire_client(unconf);
1193 new = create_client(exid->clname, dname, rqstp, &verf);
1195 status = nfserr_serverfault;
1200 add_to_unconfirmed(new, strhashval);
1202 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1203 exid->clientid.cl_id = new->cl_clientid.cl_id;
1206 nfsd4_set_ex_flags(new, exid);
1208 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1209 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1213 nfs4_unlock_state();
1215 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1220 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1222 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1225 /* The slot is in use, and no response has been sent. */
1227 if (seqid == slot_seqid)
1228 return nfserr_jukebox;
1230 return nfserr_seq_misordered;
1233 if (likely(seqid == slot_seqid + 1))
1236 if (seqid == slot_seqid)
1237 return nfserr_replay_cache;
1239 if (seqid == 1 && (slot_seqid + 1) == 0)
1241 /* Misordered replay or misordered new request */
1242 return nfserr_seq_misordered;
1246 * Cache the create session result into the create session single DRC
1247 * slot cache by saving the xdr structure. sl_seqid has been set.
1248 * Do this for solo or embedded create session operations.
1251 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1252 struct nfsd4_clid_slot *slot, int nfserr)
1254 slot->sl_status = nfserr;
1255 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1259 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1260 struct nfsd4_clid_slot *slot)
1262 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1263 return slot->sl_status;
1267 nfsd4_create_session(struct svc_rqst *rqstp,
1268 struct nfsd4_compound_state *cstate,
1269 struct nfsd4_create_session *cr_ses)
1271 struct sockaddr *sa = svc_addr(rqstp);
1272 struct nfs4_client *conf, *unconf;
1273 struct nfsd4_clid_slot *cs_slot = NULL;
1277 unconf = find_unconfirmed_client(&cr_ses->clientid);
1278 conf = find_confirmed_client(&cr_ses->clientid);
1281 cs_slot = &conf->cl_cs_slot;
1282 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1283 if (status == nfserr_replay_cache) {
1284 dprintk("Got a create_session replay! seqid= %d\n",
1286 /* Return the cached reply status */
1287 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1289 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1290 status = nfserr_seq_misordered;
1291 dprintk("Sequence misordered!\n");
1292 dprintk("Expected seqid= %d but got seqid= %d\n",
1293 cs_slot->sl_seqid, cr_ses->seqid);
1296 cs_slot->sl_seqid++;
1297 } else if (unconf) {
1298 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1299 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1300 status = nfserr_clid_inuse;
1304 cs_slot = &unconf->cl_cs_slot;
1305 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1307 /* an unconfirmed replay returns misordered */
1308 status = nfserr_seq_misordered;
1312 cs_slot->sl_seqid++; /* from 0 to 1 */
1313 move_to_confirmed(unconf);
1315 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1316 unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1317 svc_xprt_get(rqstp->rq_xprt);
1319 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1321 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1322 unconf->cl_cb_conn.cb_minorversion =
1323 cstate->minorversion;
1324 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1325 unconf->cl_cb_seq_nr = 1;
1326 nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1330 status = nfserr_stale_clientid;
1335 * We do not support RDMA or persistent sessions
1337 cr_ses->flags &= ~SESSION4_PERSIST;
1338 cr_ses->flags &= ~SESSION4_RDMA;
1340 status = alloc_init_session(rqstp, conf, cr_ses);
1344 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1345 NFS4_MAX_SESSIONID_LEN);
1346 cr_ses->seqid = cs_slot->sl_seqid;
1349 /* cache solo and embedded create sessions under the state lock */
1350 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1352 nfs4_unlock_state();
1353 dprintk("%s returns %d\n", __func__, ntohl(status));
1357 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1359 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1360 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1362 return argp->opcnt == resp->opcnt;
1365 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1369 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1373 nfsd4_destroy_session(struct svc_rqst *r,
1374 struct nfsd4_compound_state *cstate,
1375 struct nfsd4_destroy_session *sessionid)
1377 struct nfsd4_session *ses;
1378 u32 status = nfserr_badsession;
1381 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1382 * - Should we return nfserr_back_chan_busy if waiting for
1383 * callbacks on to-be-destroyed session?
1384 * - Do we need to clear any callback info from previous session?
1387 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1388 if (!nfsd4_last_compound_op(r))
1389 return nfserr_not_only_op;
1391 dump_sessionid(__func__, &sessionid->sessionid);
1392 spin_lock(&client_lock);
1393 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1395 spin_unlock(&client_lock);
1399 unhash_session(ses);
1400 spin_unlock(&client_lock);
1402 /* wait for callbacks */
1403 nfsd4_set_callback_client(ses->se_client, NULL);
1404 nfsd4_put_session(ses);
1407 dprintk("%s returns %d\n", __func__, ntohl(status));
1412 nfsd4_sequence(struct svc_rqst *rqstp,
1413 struct nfsd4_compound_state *cstate,
1414 struct nfsd4_sequence *seq)
1416 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1417 struct nfsd4_session *session;
1418 struct nfsd4_slot *slot;
1421 if (resp->opcnt != 1)
1422 return nfserr_sequence_pos;
1424 spin_lock(&client_lock);
1425 status = nfserr_badsession;
1426 session = find_in_sessionid_hashtbl(&seq->sessionid);
1430 status = nfserr_badslot;
1431 if (seq->slotid >= session->se_fchannel.maxreqs)
1434 slot = session->se_slots[seq->slotid];
1435 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1437 /* We do not negotiate the number of slots yet, so set the
1438 * maxslots to the session maxreqs which is used to encode
1439 * sr_highest_slotid and the sr_target_slot id to maxslots */
1440 seq->maxslots = session->se_fchannel.maxreqs;
1442 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1443 if (status == nfserr_replay_cache) {
1444 cstate->slot = slot;
1445 cstate->session = session;
1446 /* Return the cached reply status and set cstate->status
1447 * for nfsd4_proc_compound processing */
1448 status = nfsd4_replay_cache_entry(resp, seq);
1449 cstate->status = nfserr_replay_cache;
1455 /* Success! bump slot seqid */
1456 slot->sl_inuse = true;
1457 slot->sl_seqid = seq->seqid;
1458 slot->sl_cachethis = seq->cachethis;
1460 cstate->slot = slot;
1461 cstate->session = session;
1464 /* Hold a session reference until done processing the compound. */
1465 if (cstate->session) {
1466 nfsd4_get_session(cstate->session);
1467 /* Renew the clientid on success and on replay */
1468 renew_client_locked(session->se_client);
1470 spin_unlock(&client_lock);
1471 dprintk("%s: return %d\n", __func__, ntohl(status));
1476 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1477 struct nfsd4_setclientid *setclid)
1479 struct sockaddr *sa = svc_addr(rqstp);
1480 struct xdr_netobj clname = {
1481 .len = setclid->se_namelen,
1482 .data = setclid->se_name,
1484 nfs4_verifier clverifier = setclid->se_verf;
1485 unsigned int strhashval;
1486 struct nfs4_client *conf, *unconf, *new;
1488 char dname[HEXDIR_LEN];
1490 if (!check_name(clname))
1491 return nfserr_inval;
1493 status = nfs4_make_rec_clidname(dname, &clname);
1498 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1499 * We get here on a DRC miss.
1502 strhashval = clientstr_hashval(dname);
1505 conf = find_confirmed_client_by_str(dname, strhashval, false);
1507 /* RFC 3530 14.2.33 CASE 0: */
1508 status = nfserr_clid_inuse;
1509 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1510 char addr_str[INET6_ADDRSTRLEN];
1511 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1513 dprintk("NFSD: setclientid: string in use by client "
1514 "at %s\n", addr_str);
1519 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1520 * has a description of SETCLIENTID request processing consisting
1521 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1523 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1524 status = nfserr_resource;
1527 * RFC 3530 14.2.33 CASE 4:
1528 * placed first, because it is the normal case
1531 expire_client(unconf);
1532 new = create_client(clname, dname, rqstp, &clverifier);
1536 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1538 * RFC 3530 14.2.33 CASE 1:
1539 * probable callback update
1542 /* Note this is removing unconfirmed {*x***},
1543 * which is stronger than RFC recommended {vxc**}.
1544 * This has the advantage that there is at most
1545 * one {*x***} in either list at any time.
1547 expire_client(unconf);
1549 new = create_client(clname, dname, rqstp, &clverifier);
1552 copy_clid(new, conf);
1553 } else if (!unconf) {
1555 * RFC 3530 14.2.33 CASE 2:
1556 * probable client reboot; state will be removed if
1559 new = create_client(clname, dname, rqstp, &clverifier);
1565 * RFC 3530 14.2.33 CASE 3:
1566 * probable client reboot; state will be removed if
1569 expire_client(unconf);
1570 new = create_client(clname, dname, rqstp, &clverifier);
1575 gen_callback(new, setclid, rpc_get_scope_id(sa));
1576 add_to_unconfirmed(new, strhashval);
1577 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1578 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1579 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1582 nfs4_unlock_state();
1588 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1589 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1590 * bullets, labeled as CASE1 - CASE4 below.
1593 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1594 struct nfsd4_compound_state *cstate,
1595 struct nfsd4_setclientid_confirm *setclientid_confirm)
1597 struct sockaddr *sa = svc_addr(rqstp);
1598 struct nfs4_client *conf, *unconf;
1599 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1600 clientid_t * clid = &setclientid_confirm->sc_clientid;
1603 if (STALE_CLIENTID(clid))
1604 return nfserr_stale_clientid;
1606 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1607 * We get here on a DRC miss.
1612 conf = find_confirmed_client(clid);
1613 unconf = find_unconfirmed_client(clid);
1615 status = nfserr_clid_inuse;
1616 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1618 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1622 * section 14.2.34 of RFC 3530 has a description of
1623 * SETCLIENTID_CONFIRM request processing consisting
1624 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1626 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1628 * RFC 3530 14.2.34 CASE 1:
1631 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1632 status = nfserr_clid_inuse;
1634 atomic_set(&conf->cl_cb_set, 0);
1635 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1636 expire_client(unconf);
1640 } else if (conf && !unconf) {
1642 * RFC 3530 14.2.34 CASE 2:
1643 * probable retransmitted request; play it safe and
1646 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1647 status = nfserr_clid_inuse;
1650 } else if (!conf && unconf
1651 && same_verf(&unconf->cl_confirm, &confirm)) {
1653 * RFC 3530 14.2.34 CASE 3:
1654 * Normal case; new or rebooted client:
1656 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1657 status = nfserr_clid_inuse;
1660 clientstr_hashval(unconf->cl_recdir);
1661 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1664 nfsd4_remove_clid_dir(conf);
1665 expire_client(conf);
1667 move_to_confirmed(unconf);
1669 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1672 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1673 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1676 * RFC 3530 14.2.34 CASE 4:
1677 * Client probably hasn't noticed that we rebooted yet.
1679 status = nfserr_stale_clientid;
1681 /* check that we have hit one of the cases...*/
1682 status = nfserr_clid_inuse;
1685 nfs4_unlock_state();
1689 /* OPEN Share state helper functions */
1690 static inline struct nfs4_file *
1691 alloc_init_file(struct inode *ino)
1693 struct nfs4_file *fp;
1694 unsigned int hashval = file_hashval(ino);
1696 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1698 atomic_set(&fp->fi_ref, 1);
1699 INIT_LIST_HEAD(&fp->fi_hash);
1700 INIT_LIST_HEAD(&fp->fi_stateids);
1701 INIT_LIST_HEAD(&fp->fi_delegations);
1702 spin_lock(&recall_lock);
1703 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1704 spin_unlock(&recall_lock);
1705 fp->fi_inode = igrab(ino);
1706 fp->fi_id = current_fileid++;
1707 fp->fi_had_conflict = false;
1714 nfsd4_free_slab(struct kmem_cache **slab)
1718 kmem_cache_destroy(*slab);
1723 nfsd4_free_slabs(void)
1725 nfsd4_free_slab(&stateowner_slab);
1726 nfsd4_free_slab(&file_slab);
1727 nfsd4_free_slab(&stateid_slab);
1728 nfsd4_free_slab(&deleg_slab);
1732 nfsd4_init_slabs(void)
1734 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1735 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1736 if (stateowner_slab == NULL)
1738 file_slab = kmem_cache_create("nfsd4_files",
1739 sizeof(struct nfs4_file), 0, 0, NULL);
1740 if (file_slab == NULL)
1742 stateid_slab = kmem_cache_create("nfsd4_stateids",
1743 sizeof(struct nfs4_stateid), 0, 0, NULL);
1744 if (stateid_slab == NULL)
1746 deleg_slab = kmem_cache_create("nfsd4_delegations",
1747 sizeof(struct nfs4_delegation), 0, 0, NULL);
1748 if (deleg_slab == NULL)
1753 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1758 nfs4_free_stateowner(struct kref *kref)
1760 struct nfs4_stateowner *sop =
1761 container_of(kref, struct nfs4_stateowner, so_ref);
1762 kfree(sop->so_owner.data);
1763 kmem_cache_free(stateowner_slab, sop);
1766 static inline struct nfs4_stateowner *
1767 alloc_stateowner(struct xdr_netobj *owner)
1769 struct nfs4_stateowner *sop;
1771 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1772 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1773 memcpy(sop->so_owner.data, owner->data, owner->len);
1774 sop->so_owner.len = owner->len;
1775 kref_init(&sop->so_ref);
1778 kmem_cache_free(stateowner_slab, sop);
1783 static struct nfs4_stateowner *
1784 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1785 struct nfs4_stateowner *sop;
1786 struct nfs4_replay *rp;
1787 unsigned int idhashval;
1789 if (!(sop = alloc_stateowner(&open->op_owner)))
1791 idhashval = ownerid_hashval(current_ownerid);
1792 INIT_LIST_HEAD(&sop->so_idhash);
1793 INIT_LIST_HEAD(&sop->so_strhash);
1794 INIT_LIST_HEAD(&sop->so_perclient);
1795 INIT_LIST_HEAD(&sop->so_stateids);
1796 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1797 INIT_LIST_HEAD(&sop->so_close_lru);
1799 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1800 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1801 list_add(&sop->so_perclient, &clp->cl_openowners);
1802 sop->so_is_open_owner = 1;
1803 sop->so_id = current_ownerid++;
1804 sop->so_client = clp;
1805 sop->so_seqid = open->op_seqid;
1806 sop->so_confirmed = 0;
1807 rp = &sop->so_replay;
1808 rp->rp_status = nfserr_serverfault;
1810 rp->rp_buf = rp->rp_ibuf;
1815 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1816 struct nfs4_stateowner *sop = open->op_stateowner;
1817 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1819 INIT_LIST_HEAD(&stp->st_hash);
1820 INIT_LIST_HEAD(&stp->st_perstateowner);
1821 INIT_LIST_HEAD(&stp->st_lockowners);
1822 INIT_LIST_HEAD(&stp->st_perfile);
1823 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1824 list_add(&stp->st_perstateowner, &sop->so_stateids);
1825 list_add(&stp->st_perfile, &fp->fi_stateids);
1826 stp->st_stateowner = sop;
1829 stp->st_stateid.si_boot = get_seconds();
1830 stp->st_stateid.si_stateownerid = sop->so_id;
1831 stp->st_stateid.si_fileid = fp->fi_id;
1832 stp->st_stateid.si_generation = 0;
1833 stp->st_access_bmap = 0;
1834 stp->st_deny_bmap = 0;
1835 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1836 &stp->st_access_bmap);
1837 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1838 stp->st_openstp = NULL;
1842 move_to_close_lru(struct nfs4_stateowner *sop)
1844 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1846 list_move_tail(&sop->so_close_lru, &close_lru);
1847 sop->so_time = get_seconds();
1851 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1854 return (sop->so_owner.len == owner->len) &&
1855 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1856 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1859 static struct nfs4_stateowner *
1860 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1862 struct nfs4_stateowner *so = NULL;
1864 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1865 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1871 /* search file_hashtbl[] for file */
1872 static struct nfs4_file *
1873 find_file(struct inode *ino)
1875 unsigned int hashval = file_hashval(ino);
1876 struct nfs4_file *fp;
1878 spin_lock(&recall_lock);
1879 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1880 if (fp->fi_inode == ino) {
1882 spin_unlock(&recall_lock);
1886 spin_unlock(&recall_lock);
1890 static inline int access_valid(u32 x, u32 minorversion)
1892 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1894 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1896 x &= ~NFS4_SHARE_ACCESS_MASK;
1897 if (minorversion && x) {
1898 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1900 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1902 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1909 static inline int deny_valid(u32 x)
1911 /* Note: unlike access bits, deny bits may be zero. */
1912 return x <= NFS4_SHARE_DENY_BOTH;
1916 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1917 * st_{access,deny}_bmap field of the stateid, in order to track not
1918 * only what share bits are currently in force, but also what
1919 * combinations of share bits previous opens have used. This allows us
1920 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1921 * return an error if the client attempt to downgrade to a combination
1922 * of share bits not explicable by closing some of its previous opens.
1924 * XXX: This enforcement is actually incomplete, since we don't keep
1925 * track of access/deny bit combinations; so, e.g., we allow:
1927 * OPEN allow read, deny write
1928 * OPEN allow both, deny none
1929 * DOWNGRADE allow read, deny none
1931 * which we should reject.
1934 set_access(unsigned int *access, unsigned long bmap) {
1938 for (i = 1; i < 4; i++) {
1939 if (test_bit(i, &bmap))
1945 set_deny(unsigned int *deny, unsigned long bmap) {
1949 for (i = 0; i < 4; i++) {
1950 if (test_bit(i, &bmap))
1956 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1957 unsigned int access, deny;
1959 set_access(&access, stp->st_access_bmap);
1960 set_deny(&deny, stp->st_deny_bmap);
1961 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1967 * Called to check deny when READ with all zero stateid or
1968 * WRITE with all zero or all one stateid
1971 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1973 struct inode *ino = current_fh->fh_dentry->d_inode;
1974 struct nfs4_file *fp;
1975 struct nfs4_stateid *stp;
1978 dprintk("NFSD: nfs4_share_conflict\n");
1980 fp = find_file(ino);
1983 ret = nfserr_locked;
1984 /* Search for conflicting share reservations */
1985 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1986 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1987 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1997 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1999 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2000 drop_file_write_access(filp);
2001 spin_lock(&filp->f_lock);
2002 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2003 spin_unlock(&filp->f_lock);
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);
2031 spin_lock(&recall_lock);
2032 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2033 spin_unlock(&recall_lock);
2035 /* only place dl_time is set. protected by lock_kernel*/
2036 dp->dl_time = get_seconds();
2039 * We don't want the locks code to timeout the lease for us;
2040 * we'll remove it ourself if the delegation isn't returned
2043 fl->fl_break_time = 0;
2045 dp->dl_file->fi_had_conflict = true;
2046 nfsd4_cb_recall(dp);
2050 * The file_lock is being reapd.
2052 * Called by locks_free_lock() with lock_kernel() held.
2055 void nfsd_release_deleg_cb(struct file_lock *fl)
2057 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2059 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2061 if (!(fl->fl_flags & FL_LEASE) || !dp)
2063 dp->dl_flock = NULL;
2067 * Set the delegation file_lock back pointer.
2069 * Called from setlease() with lock_kernel() held.
2072 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2074 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2076 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2083 * Called from setlease() with lock_kernel() held
2086 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2088 struct nfs4_delegation *onlistd =
2089 (struct nfs4_delegation *)onlist->fl_owner;
2090 struct nfs4_delegation *tryd =
2091 (struct nfs4_delegation *)try->fl_owner;
2093 if (onlist->fl_lmops != try->fl_lmops)
2096 return onlistd->dl_client == tryd->dl_client;
2101 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2104 return lease_modify(onlist, arg);
2109 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2110 .fl_break = nfsd_break_deleg_cb,
2111 .fl_release_private = nfsd_release_deleg_cb,
2112 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2113 .fl_mylease = nfsd_same_client_deleg_cb,
2114 .fl_change = nfsd_change_deleg_cb,
2119 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2120 struct nfsd4_open *open)
2122 clientid_t *clientid = &open->op_clientid;
2123 struct nfs4_client *clp = NULL;
2124 unsigned int strhashval;
2125 struct nfs4_stateowner *sop = NULL;
2127 if (!check_name(open->op_owner))
2128 return nfserr_inval;
2130 if (STALE_CLIENTID(&open->op_clientid))
2131 return nfserr_stale_clientid;
2133 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2134 sop = find_openstateowner_str(strhashval, open);
2135 open->op_stateowner = sop;
2137 /* Make sure the client's lease hasn't expired. */
2138 clp = find_confirmed_client(clientid);
2140 return nfserr_expired;
2143 /* When sessions are used, skip open sequenceid processing */
2144 if (nfsd4_has_session(cstate))
2146 if (!sop->so_confirmed) {
2147 /* Replace unconfirmed owners without checking for replay. */
2148 clp = sop->so_client;
2149 release_openowner(sop);
2150 open->op_stateowner = NULL;
2153 if (open->op_seqid == sop->so_seqid - 1) {
2154 if (sop->so_replay.rp_buflen)
2155 return nfserr_replay_me;
2156 /* The original OPEN failed so spectacularly
2157 * that we don't even have replay data saved!
2158 * Therefore, we have no choice but to continue
2159 * processing this OPEN; presumably, we'll
2160 * fail again for the same reason.
2162 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2165 if (open->op_seqid != sop->so_seqid)
2166 return nfserr_bad_seqid;
2168 if (open->op_stateowner == NULL) {
2169 sop = alloc_init_open_stateowner(strhashval, clp, open);
2171 return nfserr_resource;
2172 open->op_stateowner = sop;
2174 list_del_init(&sop->so_close_lru);
2175 renew_client(sop->so_client);
2179 static inline __be32
2180 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2182 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2183 return nfserr_openmode;
2188 static struct nfs4_delegation *
2189 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2191 struct nfs4_delegation *dp;
2193 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2194 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2201 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2202 struct nfs4_delegation **dp)
2205 __be32 status = nfserr_bad_stateid;
2207 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2210 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2211 RD_STATE : WR_STATE;
2212 status = nfs4_check_delegmode(*dp, flags);
2216 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2220 open->op_stateowner->so_confirmed = 1;
2225 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2227 struct nfs4_stateid *local;
2228 __be32 status = nfserr_share_denied;
2229 struct nfs4_stateowner *sop = open->op_stateowner;
2231 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2232 /* ignore lock owners */
2233 if (local->st_stateowner->so_is_open_owner == 0)
2235 /* remember if we have seen this open owner */
2236 if (local->st_stateowner == sop)
2238 /* check for conflicting share reservations */
2239 if (!test_share(local, open))
2247 static inline struct nfs4_stateid *
2248 nfs4_alloc_stateid(void)
2250 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2254 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2255 struct nfs4_delegation *dp,
2256 struct svc_fh *cur_fh, int flags)
2258 struct nfs4_stateid *stp;
2260 stp = nfs4_alloc_stateid();
2262 return nfserr_resource;
2265 get_file(dp->dl_vfs_file);
2266 stp->st_vfs_file = dp->dl_vfs_file;
2269 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2272 if (status == nfserr_dropit)
2273 status = nfserr_jukebox;
2274 kmem_cache_free(stateid_slab, stp);
2282 static inline __be32
2283 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2284 struct nfsd4_open *open)
2286 struct iattr iattr = {
2287 .ia_valid = ATTR_SIZE,
2290 if (!open->op_truncate)
2292 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2293 return nfserr_inval;
2294 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2298 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2300 struct file *filp = stp->st_vfs_file;
2301 struct inode *inode = filp->f_path.dentry->d_inode;
2302 unsigned int share_access, new_writer;
2305 set_access(&share_access, stp->st_access_bmap);
2306 new_writer = (~share_access) & open->op_share_access
2307 & NFS4_SHARE_ACCESS_WRITE;
2310 int err = get_write_access(inode);
2312 return nfserrno(err);
2313 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2315 return nfserrno(err);
2316 file_take_write(filp);
2318 status = nfsd4_truncate(rqstp, cur_fh, open);
2321 put_write_access(inode);
2324 /* remember the open */
2325 filp->f_mode |= open->op_share_access;
2326 __set_bit(open->op_share_access, &stp->st_access_bmap);
2327 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2334 nfs4_set_claim_prev(struct nfsd4_open *open)
2336 open->op_stateowner->so_confirmed = 1;
2337 open->op_stateowner->so_client->cl_firststate = 1;
2341 * Attempt to hand out a delegation.
2344 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2346 struct nfs4_delegation *dp;
2347 struct nfs4_stateowner *sop = stp->st_stateowner;
2348 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2349 struct file_lock fl, *flp = &fl;
2350 int status, flag = 0;
2352 flag = NFS4_OPEN_DELEGATE_NONE;
2353 open->op_recall = 0;
2354 switch (open->op_claim_type) {
2355 case NFS4_OPEN_CLAIM_PREVIOUS:
2357 open->op_recall = 1;
2358 flag = open->op_delegate_type;
2359 if (flag == NFS4_OPEN_DELEGATE_NONE)
2362 case NFS4_OPEN_CLAIM_NULL:
2363 /* Let's not give out any delegations till everyone's
2364 * had the chance to reclaim theirs.... */
2365 if (locks_in_grace())
2367 if (!cb_up || !sop->so_confirmed)
2369 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2370 flag = NFS4_OPEN_DELEGATE_WRITE;
2372 flag = NFS4_OPEN_DELEGATE_READ;
2378 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2380 flag = NFS4_OPEN_DELEGATE_NONE;
2383 locks_init_lock(&fl);
2384 fl.fl_lmops = &nfsd_lease_mng_ops;
2385 fl.fl_flags = FL_LEASE;
2386 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2387 fl.fl_end = OFFSET_MAX;
2388 fl.fl_owner = (fl_owner_t)dp;
2389 fl.fl_file = stp->st_vfs_file;
2390 fl.fl_pid = current->tgid;
2392 /* vfs_setlease checks to see if delegation should be handed out.
2393 * the lock_manager callbacks fl_mylease and fl_change are used
2395 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2396 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2397 unhash_delegation(dp);
2398 flag = NFS4_OPEN_DELEGATE_NONE;
2402 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2404 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2405 STATEID_VAL(&dp->dl_stateid));
2407 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2408 && flag == NFS4_OPEN_DELEGATE_NONE
2409 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2410 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2411 open->op_delegate_type = flag;
2415 * called with nfs4_lock_state() held.
2418 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2420 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2421 struct nfs4_file *fp = NULL;
2422 struct inode *ino = current_fh->fh_dentry->d_inode;
2423 struct nfs4_stateid *stp = NULL;
2424 struct nfs4_delegation *dp = NULL;
2427 status = nfserr_inval;
2428 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2429 || !deny_valid(open->op_share_deny))
2432 * Lookup file; if found, lookup stateid and check open request,
2433 * and check for delegations in the process of being recalled.
2434 * If not found, create the nfs4_file struct
2436 fp = find_file(ino);
2438 if ((status = nfs4_check_open(fp, open, &stp)))
2440 status = nfs4_check_deleg(fp, open, &dp);
2444 status = nfserr_bad_stateid;
2445 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2447 status = nfserr_resource;
2448 fp = alloc_init_file(ino);
2454 * OPEN the file, or upgrade an existing OPEN.
2455 * If truncate fails, the OPEN fails.
2458 /* Stateid was found, this is an OPEN upgrade */
2459 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2462 update_stateid(&stp->st_stateid);
2464 /* Stateid was not found, this is a new OPEN */
2466 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2467 flags |= NFSD_MAY_READ;
2468 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2469 flags |= NFSD_MAY_WRITE;
2470 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2473 init_stateid(stp, fp, open);
2474 status = nfsd4_truncate(rqstp, current_fh, open);
2476 release_open_stateid(stp);
2479 if (nfsd4_has_session(&resp->cstate))
2480 update_stateid(&stp->st_stateid);
2482 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2484 if (nfsd4_has_session(&resp->cstate)) {
2485 open->op_stateowner->so_confirmed = 1;
2486 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2490 * Attempt to hand out a delegation. No error return, because the
2491 * OPEN succeeds even if we fail.
2493 nfs4_open_delegation(current_fh, open, stp);
2497 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2498 STATEID_VAL(&stp->st_stateid));
2502 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2503 nfs4_set_claim_prev(open);
2505 * To finish the open response, we just need to set the rflags.
2507 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2508 if (!open->op_stateowner->so_confirmed &&
2509 !nfsd4_has_session(&resp->cstate))
2510 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2516 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2519 struct nfs4_client *clp;
2523 dprintk("process_renew(%08x/%08x): starting\n",
2524 clid->cl_boot, clid->cl_id);
2525 status = nfserr_stale_clientid;
2526 if (STALE_CLIENTID(clid))
2528 clp = find_confirmed_client(clid);
2529 status = nfserr_expired;
2531 /* We assume the client took too long to RENEW. */
2532 dprintk("nfsd4_renew: clientid not found!\n");
2536 status = nfserr_cb_path_down;
2537 if (!list_empty(&clp->cl_delegations)
2538 && !atomic_read(&clp->cl_cb_set))
2542 nfs4_unlock_state();
2546 struct lock_manager nfsd4_manager = {
2550 nfsd4_end_grace(void)
2552 dprintk("NFSD: end of grace period\n");
2553 nfsd4_recdir_purge_old();
2554 locks_end_grace(&nfsd4_manager);
2556 * Now that every NFSv4 client has had the chance to recover and
2557 * to see the (possibly new, possibly shorter) lease time, we
2558 * can safely set the next grace time to the current lease time:
2560 nfsd4_grace = nfsd4_lease;
2564 nfs4_laundromat(void)
2566 struct nfs4_client *clp;
2567 struct nfs4_stateowner *sop;
2568 struct nfs4_delegation *dp;
2569 struct list_head *pos, *next, reaplist;
2570 time_t cutoff = get_seconds() - nfsd4_lease;
2571 time_t t, clientid_val = nfsd4_lease;
2572 time_t u, test_val = nfsd4_lease;
2576 dprintk("NFSD: laundromat service - starting\n");
2577 if (locks_in_grace())
2579 INIT_LIST_HEAD(&reaplist);
2580 spin_lock(&client_lock);
2581 list_for_each_safe(pos, next, &client_lru) {
2582 clp = list_entry(pos, struct nfs4_client, cl_lru);
2583 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2584 t = clp->cl_time - cutoff;
2585 if (clientid_val > t)
2589 list_move(&clp->cl_lru, &reaplist);
2591 spin_unlock(&client_lock);
2592 list_for_each_safe(pos, next, &reaplist) {
2593 clp = list_entry(pos, struct nfs4_client, cl_lru);
2594 dprintk("NFSD: purging unused client (clientid %08x)\n",
2595 clp->cl_clientid.cl_id);
2596 nfsd4_remove_clid_dir(clp);
2599 spin_lock(&recall_lock);
2600 list_for_each_safe(pos, next, &del_recall_lru) {
2601 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2602 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2603 u = dp->dl_time - cutoff;
2608 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2610 list_move(&dp->dl_recall_lru, &reaplist);
2612 spin_unlock(&recall_lock);
2613 list_for_each_safe(pos, next, &reaplist) {
2614 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2615 list_del_init(&dp->dl_recall_lru);
2616 unhash_delegation(dp);
2618 test_val = nfsd4_lease;
2619 list_for_each_safe(pos, next, &close_lru) {
2620 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2621 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2622 u = sop->so_time - cutoff;
2627 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2629 release_openowner(sop);
2631 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2632 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2633 nfs4_unlock_state();
2634 return clientid_val;
2637 static struct workqueue_struct *laundry_wq;
2638 static void laundromat_main(struct work_struct *);
2639 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2642 laundromat_main(struct work_struct *not_used)
2646 t = nfs4_laundromat();
2647 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2648 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2651 static struct nfs4_stateowner *
2652 search_close_lru(u32 st_id, int flags)
2654 struct nfs4_stateowner *local = NULL;
2656 if (flags & CLOSE_STATE) {
2657 list_for_each_entry(local, &close_lru, so_close_lru) {
2658 if (local->so_id == st_id)
2666 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2668 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2672 STALE_STATEID(stateid_t *stateid)
2674 if (time_after((unsigned long)boot_time,
2675 (unsigned long)stateid->si_boot)) {
2676 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2677 STATEID_VAL(stateid));
2684 EXPIRED_STATEID(stateid_t *stateid)
2686 if (time_before((unsigned long)boot_time,
2687 ((unsigned long)stateid->si_boot)) &&
2688 time_before((unsigned long)(stateid->si_boot + nfsd4_lease), get_seconds())) {
2689 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2690 STATEID_VAL(stateid));
2697 stateid_error_map(stateid_t *stateid)
2699 if (STALE_STATEID(stateid))
2700 return nfserr_stale_stateid;
2701 if (EXPIRED_STATEID(stateid))
2702 return nfserr_expired;
2704 dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2705 STATEID_VAL(stateid));
2706 return nfserr_bad_stateid;
2710 access_permit_read(unsigned long access_bmap)
2712 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2713 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2714 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2718 access_permit_write(unsigned long access_bmap)
2720 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2721 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2725 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2727 __be32 status = nfserr_openmode;
2729 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2731 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2738 static inline __be32
2739 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2741 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2743 else if (locks_in_grace()) {
2744 /* Answer in remaining cases depends on existance of
2745 * conflicting state; so we must wait out the grace period. */
2746 return nfserr_grace;
2747 } else if (flags & WR_STATE)
2748 return nfs4_share_conflict(current_fh,
2749 NFS4_SHARE_DENY_WRITE);
2750 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2751 return nfs4_share_conflict(current_fh,
2752 NFS4_SHARE_DENY_READ);
2756 * Allow READ/WRITE during grace period on recovered state only for files
2757 * that are not able to provide mandatory locking.
2760 grace_disallows_io(struct inode *inode)
2762 return locks_in_grace() && mandatory_lock(inode);
2765 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2768 * When sessions are used the stateid generation number is ignored
2771 if ((flags & HAS_SESSION) && in->si_generation == 0)
2774 /* If the client sends us a stateid from the future, it's buggy: */
2775 if (in->si_generation > ref->si_generation)
2776 return nfserr_bad_stateid;
2778 * The following, however, can happen. For example, if the
2779 * client sends an open and some IO at the same time, the open
2780 * may bump si_generation while the IO is still in flight.
2781 * Thanks to hard links and renames, the client never knows what
2782 * file an open will affect. So it could avoid that situation
2783 * only by serializing all opens and IO from the same open
2784 * owner. To recover from the old_stateid error, the client
2785 * will just have to retry the IO:
2787 if (in->si_generation < ref->si_generation)
2788 return nfserr_old_stateid;
2793 static int is_delegation_stateid(stateid_t *stateid)
2795 return stateid->si_fileid == 0;
2799 * Checks for stateid operations
2802 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2803 stateid_t *stateid, int flags, struct file **filpp)
2805 struct nfs4_stateid *stp = NULL;
2806 struct nfs4_delegation *dp = NULL;
2807 struct svc_fh *current_fh = &cstate->current_fh;
2808 struct inode *ino = current_fh->fh_dentry->d_inode;
2814 if (grace_disallows_io(ino))
2815 return nfserr_grace;
2817 if (nfsd4_has_session(cstate))
2818 flags |= HAS_SESSION;
2820 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2821 return check_special_stateids(current_fh, stateid, flags);
2823 status = nfserr_stale_stateid;
2824 if (STALE_STATEID(stateid))
2827 status = nfserr_bad_stateid;
2828 if (is_delegation_stateid(stateid)) {
2829 dp = find_delegation_stateid(ino, stateid);
2831 status = stateid_error_map(stateid);
2834 status = check_stateid_generation(stateid, &dp->dl_stateid,
2838 status = nfs4_check_delegmode(dp, flags);
2841 renew_client(dp->dl_client);
2843 *filpp = dp->dl_vfs_file;
2844 } else { /* open or lock stateid */
2845 stp = find_stateid(stateid, flags);
2847 status = stateid_error_map(stateid);
2850 if (nfs4_check_fh(current_fh, stp))
2852 if (!stp->st_stateowner->so_confirmed)
2854 status = check_stateid_generation(stateid, &stp->st_stateid,
2858 status = nfs4_check_openmode(stp, flags);
2861 renew_client(stp->st_stateowner->so_client);
2863 *filpp = stp->st_vfs_file;
2873 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2874 RD_STATE : WR_STATE;
2878 * Checks for sequence id mutating operations.
2881 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2882 stateid_t *stateid, int flags,
2883 struct nfs4_stateowner **sopp,
2884 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2886 struct nfs4_stateid *stp;
2887 struct nfs4_stateowner *sop;
2888 struct svc_fh *current_fh = &cstate->current_fh;
2891 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2892 seqid, STATEID_VAL(stateid));
2897 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2898 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2899 return nfserr_bad_stateid;
2902 if (STALE_STATEID(stateid))
2903 return nfserr_stale_stateid;
2905 if (nfsd4_has_session(cstate))
2906 flags |= HAS_SESSION;
2909 * We return BAD_STATEID if filehandle doesn't match stateid,
2910 * the confirmed flag is incorrecly set, or the generation
2911 * number is incorrect.
2913 stp = find_stateid(stateid, flags);
2916 * Also, we should make sure this isn't just the result of
2919 sop = search_close_lru(stateid->si_stateownerid, flags);
2921 return stateid_error_map(stateid);
2927 *sopp = sop = stp->st_stateowner;
2930 clientid_t *lockclid = &lock->v.new.clientid;
2931 struct nfs4_client *clp = sop->so_client;
2935 lkflg = setlkflg(lock->lk_type);
2937 if (lock->lk_is_new) {
2938 if (!sop->so_is_open_owner)
2939 return nfserr_bad_stateid;
2940 if (!(flags & HAS_SESSION) &&
2941 !same_clid(&clp->cl_clientid, lockclid))
2942 return nfserr_bad_stateid;
2943 /* stp is the open stateid */
2944 status = nfs4_check_openmode(stp, lkflg);
2948 /* stp is the lock stateid */
2949 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2955 if (nfs4_check_fh(current_fh, stp)) {
2956 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2957 return nfserr_bad_stateid;
2961 * We now validate the seqid and stateid generation numbers.
2962 * For the moment, we ignore the possibility of
2963 * generation number wraparound.
2965 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2968 if (sop->so_confirmed && flags & CONFIRM) {
2969 dprintk("NFSD: preprocess_seqid_op: expected"
2970 " unconfirmed stateowner!\n");
2971 return nfserr_bad_stateid;
2973 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2974 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2975 " confirmed yet!\n");
2976 return nfserr_bad_stateid;
2978 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2981 renew_client(sop->so_client);
2985 if (seqid == sop->so_seqid - 1) {
2986 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2987 /* indicate replay to calling function */
2988 return nfserr_replay_me;
2990 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2991 sop->so_seqid, seqid);
2993 return nfserr_bad_seqid;
2997 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2998 struct nfsd4_open_confirm *oc)
3001 struct nfs4_stateowner *sop;
3002 struct nfs4_stateid *stp;
3004 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3005 (int)cstate->current_fh.fh_dentry->d_name.len,
3006 cstate->current_fh.fh_dentry->d_name.name);
3008 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3014 if ((status = nfs4_preprocess_seqid_op(cstate,
3015 oc->oc_seqid, &oc->oc_req_stateid,
3016 CONFIRM | OPEN_STATE,
3017 &oc->oc_stateowner, &stp, NULL)))
3020 sop = oc->oc_stateowner;
3021 sop->so_confirmed = 1;
3022 update_stateid(&stp->st_stateid);
3023 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3024 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3025 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3027 nfsd4_create_clid_dir(sop->so_client);
3029 if (oc->oc_stateowner) {
3030 nfs4_get_stateowner(oc->oc_stateowner);
3031 cstate->replay_owner = oc->oc_stateowner;
3033 nfs4_unlock_state();
3039 * unset all bits in union bitmap (bmap) that
3040 * do not exist in share (from successful OPEN_DOWNGRADE)
3043 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3046 for (i = 1; i < 4; i++) {
3047 if ((i & access) != i)
3048 __clear_bit(i, bmap);
3053 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3056 for (i = 0; i < 4; i++) {
3057 if ((i & deny) != i)
3058 __clear_bit(i, bmap);
3063 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3064 struct nfsd4_compound_state *cstate,
3065 struct nfsd4_open_downgrade *od)
3068 struct nfs4_stateid *stp;
3069 unsigned int share_access;
3071 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3072 (int)cstate->current_fh.fh_dentry->d_name.len,
3073 cstate->current_fh.fh_dentry->d_name.name);
3075 if (!access_valid(od->od_share_access, cstate->minorversion)
3076 || !deny_valid(od->od_share_deny))
3077 return nfserr_inval;
3080 if ((status = nfs4_preprocess_seqid_op(cstate,
3084 &od->od_stateowner, &stp, NULL)))
3087 status = nfserr_inval;
3088 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3089 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3090 stp->st_access_bmap, od->od_share_access);
3093 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3094 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3095 stp->st_deny_bmap, od->od_share_deny);
3098 set_access(&share_access, stp->st_access_bmap);
3099 nfs4_file_downgrade(stp->st_vfs_file,
3100 share_access & ~od->od_share_access);
3102 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3103 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3105 update_stateid(&stp->st_stateid);
3106 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3109 if (od->od_stateowner) {
3110 nfs4_get_stateowner(od->od_stateowner);
3111 cstate->replay_owner = od->od_stateowner;
3113 nfs4_unlock_state();
3118 * nfs4_unlock_state() called after encode
3121 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3122 struct nfsd4_close *close)
3125 struct nfs4_stateid *stp;
3127 dprintk("NFSD: nfsd4_close on file %.*s\n",
3128 (int)cstate->current_fh.fh_dentry->d_name.len,
3129 cstate->current_fh.fh_dentry->d_name.name);
3132 /* check close_lru for replay */
3133 if ((status = nfs4_preprocess_seqid_op(cstate,
3136 OPEN_STATE | CLOSE_STATE,
3137 &close->cl_stateowner, &stp, NULL)))
3140 update_stateid(&stp->st_stateid);
3141 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3143 /* release_stateid() calls nfsd_close() if needed */
3144 release_open_stateid(stp);
3146 /* place unused nfs4_stateowners on so_close_lru list to be
3147 * released by the laundromat service after the lease period
3148 * to enable us to handle CLOSE replay
3150 if (list_empty(&close->cl_stateowner->so_stateids))
3151 move_to_close_lru(close->cl_stateowner);
3153 if (close->cl_stateowner) {
3154 nfs4_get_stateowner(close->cl_stateowner);
3155 cstate->replay_owner = close->cl_stateowner;
3157 nfs4_unlock_state();
3162 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3163 struct nfsd4_delegreturn *dr)
3165 struct nfs4_delegation *dp;
3166 stateid_t *stateid = &dr->dr_stateid;
3167 struct inode *inode;
3171 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3173 inode = cstate->current_fh.fh_dentry->d_inode;
3175 if (nfsd4_has_session(cstate))
3176 flags |= HAS_SESSION;
3178 status = nfserr_bad_stateid;
3179 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3181 status = nfserr_stale_stateid;
3182 if (STALE_STATEID(stateid))
3184 status = nfserr_bad_stateid;
3185 if (!is_delegation_stateid(stateid))
3187 dp = find_delegation_stateid(inode, stateid);
3189 status = stateid_error_map(stateid);
3192 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3195 renew_client(dp->dl_client);
3197 unhash_delegation(dp);
3199 nfs4_unlock_state();
3206 * Lock owner state (byte-range locks)
3208 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3209 #define LOCK_HASH_BITS 8
3210 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3211 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3214 end_offset(u64 start, u64 len)
3219 return end >= start ? end: NFS4_MAX_UINT64;
3222 /* last octet in a range */
3224 last_byte_offset(u64 start, u64 len)
3230 return end > start ? end - 1: NFS4_MAX_UINT64;
3233 #define lockownerid_hashval(id) \
3234 ((id) & LOCK_HASH_MASK)
3236 static inline unsigned int
3237 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3238 struct xdr_netobj *ownername)
3240 return (file_hashval(inode) + cl_id
3241 + opaque_hashval(ownername->data, ownername->len))
3245 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3246 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3247 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3249 static struct nfs4_stateid *
3250 find_stateid(stateid_t *stid, int flags)
3252 struct nfs4_stateid *local;
3253 u32 st_id = stid->si_stateownerid;
3254 u32 f_id = stid->si_fileid;
3255 unsigned int hashval;
3257 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3258 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3259 hashval = stateid_hashval(st_id, f_id);
3260 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3261 if ((local->st_stateid.si_stateownerid == st_id) &&
3262 (local->st_stateid.si_fileid == f_id))
3267 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3268 hashval = stateid_hashval(st_id, f_id);
3269 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3270 if ((local->st_stateid.si_stateownerid == st_id) &&
3271 (local->st_stateid.si_fileid == f_id))
3278 static struct nfs4_delegation *
3279 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3281 struct nfs4_file *fp;
3282 struct nfs4_delegation *dl;
3284 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3287 fp = find_file(ino);
3290 dl = find_delegation_file(fp, stid);
3296 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3297 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3298 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3299 * locking, this prevents us from being completely protocol-compliant. The
3300 * real solution to this problem is to start using unsigned file offsets in
3301 * the VFS, but this is a very deep change!
3304 nfs4_transform_lock_offset(struct file_lock *lock)
3306 if (lock->fl_start < 0)
3307 lock->fl_start = OFFSET_MAX;
3308 if (lock->fl_end < 0)
3309 lock->fl_end = OFFSET_MAX;
3312 /* Hack!: For now, we're defining this just so we can use a pointer to it
3313 * as a unique cookie to identify our (NFSv4's) posix locks. */
3314 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3318 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3320 struct nfs4_stateowner *sop;
3323 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3324 sop = (struct nfs4_stateowner *) fl->fl_owner;
3325 hval = lockownerid_hashval(sop->so_id);
3326 kref_get(&sop->so_ref);
3328 deny->ld_clientid = sop->so_client->cl_clientid;
3330 deny->ld_sop = NULL;
3331 deny->ld_clientid.cl_boot = 0;
3332 deny->ld_clientid.cl_id = 0;
3334 deny->ld_start = fl->fl_start;
3335 deny->ld_length = NFS4_MAX_UINT64;
3336 if (fl->fl_end != NFS4_MAX_UINT64)
3337 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3338 deny->ld_type = NFS4_READ_LT;
3339 if (fl->fl_type != F_RDLCK)
3340 deny->ld_type = NFS4_WRITE_LT;
3343 static struct nfs4_stateowner *
3344 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3345 struct xdr_netobj *owner)
3347 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3348 struct nfs4_stateowner *op;
3350 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3351 if (same_owner_str(op, owner, clid))
3358 * Alloc a lock owner structure.
3359 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3362 * strhashval = lock_ownerstr_hashval
3365 static struct nfs4_stateowner *
3366 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3367 struct nfs4_stateowner *sop;
3368 struct nfs4_replay *rp;
3369 unsigned int idhashval;
3371 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3373 idhashval = lockownerid_hashval(current_ownerid);
3374 INIT_LIST_HEAD(&sop->so_idhash);
3375 INIT_LIST_HEAD(&sop->so_strhash);
3376 INIT_LIST_HEAD(&sop->so_perclient);
3377 INIT_LIST_HEAD(&sop->so_stateids);
3378 INIT_LIST_HEAD(&sop->so_perstateid);
3379 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3381 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3382 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3383 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3384 sop->so_is_open_owner = 0;
3385 sop->so_id = current_ownerid++;
3386 sop->so_client = clp;
3387 /* It is the openowner seqid that will be incremented in encode in the
3388 * case of new lockowners; so increment the lock seqid manually: */
3389 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3390 sop->so_confirmed = 1;
3391 rp = &sop->so_replay;
3392 rp->rp_status = nfserr_serverfault;
3394 rp->rp_buf = rp->rp_ibuf;
3398 static struct nfs4_stateid *
3399 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3401 struct nfs4_stateid *stp;
3402 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3404 stp = nfs4_alloc_stateid();
3407 INIT_LIST_HEAD(&stp->st_hash);
3408 INIT_LIST_HEAD(&stp->st_perfile);
3409 INIT_LIST_HEAD(&stp->st_perstateowner);
3410 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3411 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3412 list_add(&stp->st_perfile, &fp->fi_stateids);
3413 list_add(&stp->st_perstateowner, &sop->so_stateids);
3414 stp->st_stateowner = sop;
3417 stp->st_stateid.si_boot = get_seconds();
3418 stp->st_stateid.si_stateownerid = sop->so_id;
3419 stp->st_stateid.si_fileid = fp->fi_id;
3420 stp->st_stateid.si_generation = 0;
3421 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3422 stp->st_access_bmap = open_stp->st_access_bmap;
3423 stp->st_deny_bmap = open_stp->st_deny_bmap;
3424 stp->st_openstp = open_stp;
3431 check_lock_length(u64 offset, u64 length)
3433 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3434 LOFF_OVERFLOW(offset, length)));
3441 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3442 struct nfsd4_lock *lock)
3444 struct nfs4_stateowner *open_sop = NULL;
3445 struct nfs4_stateowner *lock_sop = NULL;
3446 struct nfs4_stateid *lock_stp;
3448 struct file_lock file_lock;
3449 struct file_lock conflock;
3451 unsigned int strhashval;
3455 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3456 (long long) lock->lk_offset,
3457 (long long) lock->lk_length);
3459 if (check_lock_length(lock->lk_offset, lock->lk_length))
3460 return nfserr_inval;
3462 if ((status = fh_verify(rqstp, &cstate->current_fh,
3463 S_IFREG, NFSD_MAY_LOCK))) {
3464 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3470 if (lock->lk_is_new) {
3472 * Client indicates that this is a new lockowner.
3473 * Use open owner and open stateid to create lock owner and
3476 struct nfs4_stateid *open_stp = NULL;
3477 struct nfs4_file *fp;
3479 status = nfserr_stale_clientid;
3480 if (!nfsd4_has_session(cstate) &&
3481 STALE_CLIENTID(&lock->lk_new_clientid))
3484 /* validate and update open stateid and open seqid */
3485 status = nfs4_preprocess_seqid_op(cstate,
3486 lock->lk_new_open_seqid,
3487 &lock->lk_new_open_stateid,
3489 &lock->lk_replay_owner, &open_stp,
3493 open_sop = lock->lk_replay_owner;
3494 /* create lockowner and lock stateid */
3495 fp = open_stp->st_file;
3496 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3497 open_sop->so_client->cl_clientid.cl_id,
3498 &lock->v.new.owner);
3499 /* XXX: Do we need to check for duplicate stateowners on
3500 * the same file, or should they just be allowed (and
3501 * create new stateids)? */
3502 status = nfserr_resource;
3503 lock_sop = alloc_init_lock_stateowner(strhashval,
3504 open_sop->so_client, open_stp, lock);
3505 if (lock_sop == NULL)
3507 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3508 if (lock_stp == NULL)
3511 /* lock (lock owner + lock stateid) already exists */
3512 status = nfs4_preprocess_seqid_op(cstate,
3513 lock->lk_old_lock_seqid,
3514 &lock->lk_old_lock_stateid,
3516 &lock->lk_replay_owner, &lock_stp, lock);
3519 lock_sop = lock->lk_replay_owner;
3521 /* lock->lk_replay_owner and lock_stp have been created or found */
3522 filp = lock_stp->st_vfs_file;
3524 status = nfserr_grace;
3525 if (locks_in_grace() && !lock->lk_reclaim)
3527 status = nfserr_no_grace;
3528 if (!locks_in_grace() && lock->lk_reclaim)
3531 locks_init_lock(&file_lock);
3532 switch (lock->lk_type) {
3535 file_lock.fl_type = F_RDLCK;
3539 case NFS4_WRITEW_LT:
3540 file_lock.fl_type = F_WRLCK;
3544 status = nfserr_inval;
3547 file_lock.fl_owner = (fl_owner_t)lock_sop;
3548 file_lock.fl_pid = current->tgid;
3549 file_lock.fl_file = filp;
3550 file_lock.fl_flags = FL_POSIX;
3551 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3553 file_lock.fl_start = lock->lk_offset;
3554 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3555 nfs4_transform_lock_offset(&file_lock);
3558 * Try to lock the file in the VFS.
3559 * Note: locks.c uses the BKL to protect the inode's lock list.
3562 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3564 case 0: /* success! */
3565 update_stateid(&lock_stp->st_stateid);
3566 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3570 case (EAGAIN): /* conflock holds conflicting lock */
3571 status = nfserr_denied;
3572 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3573 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3576 status = nfserr_deadlock;
3579 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3580 status = nfserr_resource;
3584 if (status && lock->lk_is_new && lock_sop)
3585 release_lockowner(lock_sop);
3586 if (lock->lk_replay_owner) {
3587 nfs4_get_stateowner(lock->lk_replay_owner);
3588 cstate->replay_owner = lock->lk_replay_owner;
3590 nfs4_unlock_state();
3595 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3596 * so we do a temporary open here just to get an open file to pass to
3597 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3600 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3605 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3608 err = vfs_test_lock(file, lock);
3617 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3618 struct nfsd4_lockt *lockt)
3620 struct inode *inode;
3621 struct file_lock file_lock;
3625 if (locks_in_grace())
3626 return nfserr_grace;
3628 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3629 return nfserr_inval;
3631 lockt->lt_stateowner = NULL;
3634 status = nfserr_stale_clientid;
3635 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3638 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3639 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3640 if (status == nfserr_symlink)
3641 status = nfserr_inval;
3645 inode = cstate->current_fh.fh_dentry->d_inode;
3646 locks_init_lock(&file_lock);
3647 switch (lockt->lt_type) {
3650 file_lock.fl_type = F_RDLCK;
3653 case NFS4_WRITEW_LT:
3654 file_lock.fl_type = F_WRLCK;
3657 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3658 status = nfserr_inval;
3662 lockt->lt_stateowner = find_lockstateowner_str(inode,
3663 &lockt->lt_clientid, &lockt->lt_owner);
3664 if (lockt->lt_stateowner)
3665 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3666 file_lock.fl_pid = current->tgid;
3667 file_lock.fl_flags = FL_POSIX;
3669 file_lock.fl_start = lockt->lt_offset;
3670 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3672 nfs4_transform_lock_offset(&file_lock);
3675 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3677 status = nfserrno(error);
3680 if (file_lock.fl_type != F_UNLCK) {
3681 status = nfserr_denied;
3682 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3685 nfs4_unlock_state();
3690 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3691 struct nfsd4_locku *locku)
3693 struct nfs4_stateid *stp;
3694 struct file *filp = NULL;
3695 struct file_lock file_lock;
3699 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3700 (long long) locku->lu_offset,
3701 (long long) locku->lu_length);
3703 if (check_lock_length(locku->lu_offset, locku->lu_length))
3704 return nfserr_inval;
3708 if ((status = nfs4_preprocess_seqid_op(cstate,
3712 &locku->lu_stateowner, &stp, NULL)))
3715 filp = stp->st_vfs_file;
3717 locks_init_lock(&file_lock);
3718 file_lock.fl_type = F_UNLCK;
3719 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3720 file_lock.fl_pid = current->tgid;
3721 file_lock.fl_file = filp;
3722 file_lock.fl_flags = FL_POSIX;
3723 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3724 file_lock.fl_start = locku->lu_offset;
3726 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3727 nfs4_transform_lock_offset(&file_lock);
3730 * Try to unlock the file in the VFS.
3732 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3734 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3738 * OK, unlock succeeded; the only thing left to do is update the stateid.
3740 update_stateid(&stp->st_stateid);
3741 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3744 if (locku->lu_stateowner) {
3745 nfs4_get_stateowner(locku->lu_stateowner);
3746 cstate->replay_owner = locku->lu_stateowner;
3748 nfs4_unlock_state();
3752 status = nfserrno(err);
3758 * 1: locks held by lockowner
3759 * 0: no locks held by lockowner
3762 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3764 struct file_lock **flpp;
3765 struct inode *inode = filp->f_path.dentry->d_inode;
3769 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3770 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3781 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3782 struct nfsd4_compound_state *cstate,
3783 struct nfsd4_release_lockowner *rlockowner)
3785 clientid_t *clid = &rlockowner->rl_clientid;
3786 struct nfs4_stateowner *sop;
3787 struct nfs4_stateid *stp;
3788 struct xdr_netobj *owner = &rlockowner->rl_owner;
3789 struct list_head matches;
3793 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3794 clid->cl_boot, clid->cl_id);
3796 /* XXX check for lease expiration */
3798 status = nfserr_stale_clientid;
3799 if (STALE_CLIENTID(clid))
3804 status = nfserr_locks_held;
3805 /* XXX: we're doing a linear search through all the lockowners.
3806 * Yipes! For now we'll just hope clients aren't really using
3807 * release_lockowner much, but eventually we have to fix these
3808 * data structures. */
3809 INIT_LIST_HEAD(&matches);
3810 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3811 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3812 if (!same_owner_str(sop, owner, clid))
3814 list_for_each_entry(stp, &sop->so_stateids,
3816 if (check_for_locks(stp->st_vfs_file, sop))
3818 /* Note: so_perclient unused for lockowners,
3819 * so it's OK to fool with here. */
3820 list_add(&sop->so_perclient, &matches);
3824 /* Clients probably won't expect us to return with some (but not all)
3825 * of the lockowner state released; so don't release any until all
3826 * have been checked. */
3828 while (!list_empty(&matches)) {
3829 sop = list_entry(matches.next, struct nfs4_stateowner,
3831 /* unhash_stateowner deletes so_perclient only
3832 * for openowners. */
3833 list_del(&sop->so_perclient);
3834 release_lockowner(sop);
3837 nfs4_unlock_state();
3841 static inline struct nfs4_client_reclaim *
3844 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3848 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3850 unsigned int strhashval = clientstr_hashval(name);
3851 struct nfs4_client *clp;
3853 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3858 * failure => all reset bets are off, nfserr_no_grace...
3861 nfs4_client_to_reclaim(const char *name)
3863 unsigned int strhashval;
3864 struct nfs4_client_reclaim *crp = NULL;
3866 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3867 crp = alloc_reclaim();
3870 strhashval = clientstr_hashval(name);
3871 INIT_LIST_HEAD(&crp->cr_strhash);
3872 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3873 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3874 reclaim_str_hashtbl_size++;
3879 nfs4_release_reclaim(void)
3881 struct nfs4_client_reclaim *crp = NULL;
3884 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3885 while (!list_empty(&reclaim_str_hashtbl[i])) {
3886 crp = list_entry(reclaim_str_hashtbl[i].next,
3887 struct nfs4_client_reclaim, cr_strhash);
3888 list_del(&crp->cr_strhash);
3890 reclaim_str_hashtbl_size--;
3893 BUG_ON(reclaim_str_hashtbl_size);
3897 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3898 static struct nfs4_client_reclaim *
3899 nfs4_find_reclaim_client(clientid_t *clid)
3901 unsigned int strhashval;
3902 struct nfs4_client *clp;
3903 struct nfs4_client_reclaim *crp = NULL;
3906 /* find clientid in conf_id_hashtbl */
3907 clp = find_confirmed_client(clid);
3911 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3912 clp->cl_name.len, clp->cl_name.data,
3915 /* find clp->cl_name in reclaim_str_hashtbl */
3916 strhashval = clientstr_hashval(clp->cl_recdir);
3917 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3918 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3926 * Called from OPEN. Look for clientid in reclaim list.
3929 nfs4_check_open_reclaim(clientid_t *clid)
3931 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3934 /* initialization to perform at module load time: */
3937 nfs4_state_init(void)
3941 status = nfsd4_init_slabs();
3944 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3945 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3946 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3947 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3948 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3949 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3951 for (i = 0; i < SESSION_HASH_SIZE; i++)
3952 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3953 for (i = 0; i < FILE_HASH_SIZE; i++) {
3954 INIT_LIST_HEAD(&file_hashtbl[i]);
3956 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3957 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3958 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3960 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3961 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3962 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3964 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3965 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3966 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3968 memset(&onestateid, ~0, sizeof(stateid_t));
3969 INIT_LIST_HEAD(&close_lru);
3970 INIT_LIST_HEAD(&client_lru);
3971 INIT_LIST_HEAD(&del_recall_lru);
3972 reclaim_str_hashtbl_size = 0;
3977 nfsd4_load_reboot_recovery_data(void)
3982 nfsd4_init_recdir(user_recovery_dirname);
3983 status = nfsd4_recdir_load();
3984 nfs4_unlock_state();
3986 printk("NFSD: Failure reading reboot recovery data\n");
3990 * Since the lifetime of a delegation isn't limited to that of an open, a
3991 * client may quite reasonably hang on to a delegation as long as it has
3992 * the inode cached. This becomes an obvious problem the first time a
3993 * client's inode cache approaches the size of the server's total memory.
3995 * For now we avoid this problem by imposing a hard limit on the number
3996 * of delegations, which varies according to the server's memory size.
3999 set_max_delegations(void)
4002 * Allow at most 4 delegations per megabyte of RAM. Quick
4003 * estimates suggest that in the worst case (where every delegation
4004 * is for a different inode), a delegation could take about 1.5K,
4005 * giving a worst case usage of about 6% of memory.
4007 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4010 /* initialization to perform when the nfsd service is started: */
4013 __nfs4_state_start(void)
4017 boot_time = get_seconds();
4018 locks_start_grace(&nfsd4_manager);
4019 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4021 ret = set_callback_cred();
4024 laundry_wq = create_singlethread_workqueue("nfsd4");
4025 if (laundry_wq == NULL)
4027 ret = nfsd4_create_callback_queue();
4029 goto out_free_laundry;
4030 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4031 set_max_delegations();
4034 destroy_workqueue(laundry_wq);
4039 nfs4_state_start(void)
4045 nfsd4_load_reboot_recovery_data();
4046 ret = __nfs4_state_start();
4054 __nfs4_state_shutdown(void)
4057 struct nfs4_client *clp = NULL;
4058 struct nfs4_delegation *dp = NULL;
4059 struct list_head *pos, *next, reaplist;
4061 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4062 while (!list_empty(&conf_id_hashtbl[i])) {
4063 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4066 while (!list_empty(&unconf_str_hashtbl[i])) {
4067 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4071 INIT_LIST_HEAD(&reaplist);
4072 spin_lock(&recall_lock);
4073 list_for_each_safe(pos, next, &del_recall_lru) {
4074 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4075 list_move(&dp->dl_recall_lru, &reaplist);
4077 spin_unlock(&recall_lock);
4078 list_for_each_safe(pos, next, &reaplist) {
4079 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4080 list_del_init(&dp->dl_recall_lru);
4081 unhash_delegation(dp);
4084 nfsd4_shutdown_recdir();
4089 nfs4_state_shutdown(void)
4091 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4092 destroy_workqueue(laundry_wq);
4093 locks_end_grace(&nfsd4_manager);
4095 nfs4_release_reclaim();
4096 __nfs4_state_shutdown();
4097 nfsd4_destroy_callback_queue();
4098 nfs4_unlock_state();
4102 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4103 * accessed when nfsd is starting.
4106 nfs4_set_recdir(char *recdir)
4108 strcpy(user_recovery_dirname, recdir);
4112 * Change the NFSv4 recovery directory to recdir.
4115 nfs4_reset_recoverydir(char *recdir)
4120 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4124 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4125 nfs4_set_recdir(recdir);
4133 nfs4_recoverydir(void)
4135 return user_recovery_dirname;