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 if (atomic_read(&clp->cl_refcount) == 0)
745 spin_unlock(&client_lock);
748 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
750 memcpy(target->cl_verifier.data, source->data,
751 sizeof(target->cl_verifier.data));
754 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
756 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
757 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
760 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
762 target->cr_uid = source->cr_uid;
763 target->cr_gid = source->cr_gid;
764 target->cr_group_info = source->cr_group_info;
765 get_group_info(target->cr_group_info);
768 static int same_name(const char *n1, const char *n2)
770 return 0 == memcmp(n1, n2, HEXDIR_LEN);
774 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
776 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
780 same_clid(clientid_t *cl1, clientid_t *cl2)
782 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
785 /* XXX what about NGROUP */
787 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
789 return cr1->cr_uid == cr2->cr_uid;
792 static void gen_clid(struct nfs4_client *clp)
794 static u32 current_clientid = 1;
796 clp->cl_clientid.cl_boot = boot_time;
797 clp->cl_clientid.cl_id = current_clientid++;
800 static void gen_confirm(struct nfs4_client *clp)
805 p = (u32 *)clp->cl_confirm.data;
806 *p++ = get_seconds();
810 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
811 struct svc_rqst *rqstp, nfs4_verifier *verf)
813 struct nfs4_client *clp;
814 struct sockaddr *sa = svc_addr(rqstp);
817 clp = alloc_client(name);
821 princ = svc_gss_principal(rqstp);
823 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
824 if (clp->cl_principal == NULL) {
830 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
831 atomic_set(&clp->cl_refcount, 0);
832 atomic_set(&clp->cl_cb_set, 0);
833 INIT_LIST_HEAD(&clp->cl_idhash);
834 INIT_LIST_HEAD(&clp->cl_strhash);
835 INIT_LIST_HEAD(&clp->cl_openowners);
836 INIT_LIST_HEAD(&clp->cl_delegations);
837 INIT_LIST_HEAD(&clp->cl_sessions);
838 INIT_LIST_HEAD(&clp->cl_lru);
839 clear_bit(0, &clp->cl_cb_slot_busy);
840 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
841 copy_verf(clp, verf);
842 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
843 clp->cl_flavor = rqstp->rq_flavor;
844 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
850 static int check_name(struct xdr_netobj name)
854 if (name.len > NFS4_OPAQUE_LIMIT) {
855 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
862 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
864 unsigned int idhashval;
866 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
867 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
868 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
873 move_to_confirmed(struct nfs4_client *clp)
875 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
876 unsigned int strhashval;
878 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
879 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
880 strhashval = clientstr_hashval(clp->cl_recdir);
881 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
885 static struct nfs4_client *
886 find_confirmed_client(clientid_t *clid)
888 struct nfs4_client *clp;
889 unsigned int idhashval = clientid_hashval(clid->cl_id);
891 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
892 if (same_clid(&clp->cl_clientid, clid))
898 static struct nfs4_client *
899 find_unconfirmed_client(clientid_t *clid)
901 struct nfs4_client *clp;
902 unsigned int idhashval = clientid_hashval(clid->cl_id);
904 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
905 if (same_clid(&clp->cl_clientid, clid))
912 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
913 * parameter. Matching is based on the fact the at least one of the
914 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
916 * FIXME: we need to unify the clientid namespaces for nfsv4.x
917 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
918 * and SET_CLIENTID{,_CONFIRM}
921 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
923 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
924 return use_exchange_id == has_exchange_flags;
927 static struct nfs4_client *
928 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
929 bool use_exchange_id)
931 struct nfs4_client *clp;
933 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
934 if (same_name(clp->cl_recdir, dname) &&
935 match_clientid_establishment(clp, use_exchange_id))
941 static struct nfs4_client *
942 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
943 bool use_exchange_id)
945 struct nfs4_client *clp;
947 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
948 if (same_name(clp->cl_recdir, dname) &&
949 match_clientid_establishment(clp, use_exchange_id))
956 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
958 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
959 unsigned short expected_family;
961 /* Currently, we only support tcp and tcp6 for the callback channel */
962 if (se->se_callback_netid_len == 3 &&
963 !memcmp(se->se_callback_netid_val, "tcp", 3))
964 expected_family = AF_INET;
965 else if (se->se_callback_netid_len == 4 &&
966 !memcmp(se->se_callback_netid_val, "tcp6", 4))
967 expected_family = AF_INET6;
971 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
972 se->se_callback_addr_len,
973 (struct sockaddr *) &cb->cb_addr,
974 sizeof(cb->cb_addr));
976 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
979 if (cb->cb_addr.ss_family == AF_INET6)
980 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
982 cb->cb_minorversion = 0;
983 cb->cb_prog = se->se_callback_prog;
984 cb->cb_ident = se->se_callback_ident;
987 cb->cb_addr.ss_family = AF_UNSPEC;
989 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
990 "will not receive delegations\n",
991 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
997 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1000 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1002 struct nfsd4_slot *slot = resp->cstate.slot;
1005 dprintk("--> %s slot %p\n", __func__, slot);
1007 slot->sl_opcnt = resp->opcnt;
1008 slot->sl_status = resp->cstate.status;
1010 if (nfsd4_not_cached(resp)) {
1011 slot->sl_datalen = 0;
1014 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1015 base = (char *)resp->cstate.datap -
1016 (char *)resp->xbuf->head[0].iov_base;
1017 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1019 WARN("%s: sessions DRC could not cache compound\n", __func__);
1024 * Encode the replay sequence operation from the slot values.
1025 * If cachethis is FALSE encode the uncached rep error on the next
1026 * operation which sets resp->p and increments resp->opcnt for
1027 * nfs4svc_encode_compoundres.
1031 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1032 struct nfsd4_compoundres *resp)
1034 struct nfsd4_op *op;
1035 struct nfsd4_slot *slot = resp->cstate.slot;
1037 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1038 resp->opcnt, resp->cstate.slot->sl_cachethis);
1040 /* Encode the replayed sequence operation */
1041 op = &args->ops[resp->opcnt - 1];
1042 nfsd4_encode_operation(resp, op);
1044 /* Return nfserr_retry_uncached_rep in next operation. */
1045 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1046 op = &args->ops[resp->opcnt++];
1047 op->status = nfserr_retry_uncached_rep;
1048 nfsd4_encode_operation(resp, op);
1054 * The sequence operation is not cached because we can use the slot and
1058 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1059 struct nfsd4_sequence *seq)
1061 struct nfsd4_slot *slot = resp->cstate.slot;
1064 dprintk("--> %s slot %p\n", __func__, slot);
1066 /* Either returns 0 or nfserr_retry_uncached */
1067 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1068 if (status == nfserr_retry_uncached_rep)
1071 /* The sequence operation has been encoded, cstate->datap set. */
1072 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1074 resp->opcnt = slot->sl_opcnt;
1075 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1076 status = slot->sl_status;
1082 * Set the exchange_id flags returned by the server.
1085 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1087 /* pNFS is not supported */
1088 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1090 /* Referrals are supported, Migration is not. */
1091 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1093 /* set the wire flags to return to client. */
1094 clid->flags = new->cl_exchange_flags;
1098 nfsd4_exchange_id(struct svc_rqst *rqstp,
1099 struct nfsd4_compound_state *cstate,
1100 struct nfsd4_exchange_id *exid)
1102 struct nfs4_client *unconf, *conf, *new;
1104 unsigned int strhashval;
1105 char dname[HEXDIR_LEN];
1106 char addr_str[INET6_ADDRSTRLEN];
1107 nfs4_verifier verf = exid->verifier;
1108 struct sockaddr *sa = svc_addr(rqstp);
1110 rpc_ntop(sa, addr_str, sizeof(addr_str));
1111 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1112 "ip_addr=%s flags %x, spa_how %d\n",
1113 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1114 addr_str, exid->flags, exid->spa_how);
1116 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1117 return nfserr_inval;
1119 /* Currently only support SP4_NONE */
1120 switch (exid->spa_how) {
1124 return nfserr_encr_alg_unsupp;
1126 BUG(); /* checked by xdr code */
1128 return nfserr_serverfault; /* no excuse :-/ */
1131 status = nfs4_make_rec_clidname(dname, &exid->clname);
1136 strhashval = clientstr_hashval(dname);
1141 conf = find_confirmed_client_by_str(dname, strhashval, true);
1143 if (!same_verf(&verf, &conf->cl_verifier)) {
1144 /* 18.35.4 case 8 */
1145 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1146 status = nfserr_not_same;
1149 /* Client reboot: destroy old state */
1150 expire_client(conf);
1153 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1154 /* 18.35.4 case 9 */
1155 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1156 status = nfserr_perm;
1159 expire_client(conf);
1163 * Set bit when the owner id and verifier map to an already
1164 * confirmed client id (18.35.3).
1166 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1169 * Falling into 18.35.4 case 2, possible router replay.
1170 * Leave confirmed record intact and return same result.
1172 copy_verf(conf, &verf);
1177 /* 18.35.4 case 7 */
1178 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1179 status = nfserr_noent;
1183 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1186 * Possible retry or client restart. Per 18.35.4 case 4,
1187 * a new unconfirmed record should be generated regardless
1188 * of whether any properties have changed.
1190 expire_client(unconf);
1195 new = create_client(exid->clname, dname, rqstp, &verf);
1197 status = nfserr_serverfault;
1202 add_to_unconfirmed(new, strhashval);
1204 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1205 exid->clientid.cl_id = new->cl_clientid.cl_id;
1208 nfsd4_set_ex_flags(new, exid);
1210 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1211 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1215 nfs4_unlock_state();
1217 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1222 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1224 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1227 /* The slot is in use, and no response has been sent. */
1229 if (seqid == slot_seqid)
1230 return nfserr_jukebox;
1232 return nfserr_seq_misordered;
1235 if (likely(seqid == slot_seqid + 1))
1238 if (seqid == slot_seqid)
1239 return nfserr_replay_cache;
1241 if (seqid == 1 && (slot_seqid + 1) == 0)
1243 /* Misordered replay or misordered new request */
1244 return nfserr_seq_misordered;
1248 * Cache the create session result into the create session single DRC
1249 * slot cache by saving the xdr structure. sl_seqid has been set.
1250 * Do this for solo or embedded create session operations.
1253 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1254 struct nfsd4_clid_slot *slot, int nfserr)
1256 slot->sl_status = nfserr;
1257 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1261 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1262 struct nfsd4_clid_slot *slot)
1264 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1265 return slot->sl_status;
1269 nfsd4_create_session(struct svc_rqst *rqstp,
1270 struct nfsd4_compound_state *cstate,
1271 struct nfsd4_create_session *cr_ses)
1273 struct sockaddr *sa = svc_addr(rqstp);
1274 struct nfs4_client *conf, *unconf;
1275 struct nfsd4_clid_slot *cs_slot = NULL;
1279 unconf = find_unconfirmed_client(&cr_ses->clientid);
1280 conf = find_confirmed_client(&cr_ses->clientid);
1283 cs_slot = &conf->cl_cs_slot;
1284 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1285 if (status == nfserr_replay_cache) {
1286 dprintk("Got a create_session replay! seqid= %d\n",
1288 /* Return the cached reply status */
1289 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1291 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1292 status = nfserr_seq_misordered;
1293 dprintk("Sequence misordered!\n");
1294 dprintk("Expected seqid= %d but got seqid= %d\n",
1295 cs_slot->sl_seqid, cr_ses->seqid);
1298 cs_slot->sl_seqid++;
1299 } else if (unconf) {
1300 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1301 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1302 status = nfserr_clid_inuse;
1306 cs_slot = &unconf->cl_cs_slot;
1307 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1309 /* an unconfirmed replay returns misordered */
1310 status = nfserr_seq_misordered;
1314 cs_slot->sl_seqid++; /* from 0 to 1 */
1315 move_to_confirmed(unconf);
1317 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1318 unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1319 svc_xprt_get(rqstp->rq_xprt);
1321 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1323 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1324 unconf->cl_cb_conn.cb_minorversion =
1325 cstate->minorversion;
1326 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1327 unconf->cl_cb_seq_nr = 1;
1328 nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1332 status = nfserr_stale_clientid;
1337 * We do not support RDMA or persistent sessions
1339 cr_ses->flags &= ~SESSION4_PERSIST;
1340 cr_ses->flags &= ~SESSION4_RDMA;
1342 status = alloc_init_session(rqstp, conf, cr_ses);
1346 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1347 NFS4_MAX_SESSIONID_LEN);
1348 cr_ses->seqid = cs_slot->sl_seqid;
1351 /* cache solo and embedded create sessions under the state lock */
1352 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1354 nfs4_unlock_state();
1355 dprintk("%s returns %d\n", __func__, ntohl(status));
1359 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1361 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1362 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1364 return argp->opcnt == resp->opcnt;
1367 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1371 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1375 nfsd4_destroy_session(struct svc_rqst *r,
1376 struct nfsd4_compound_state *cstate,
1377 struct nfsd4_destroy_session *sessionid)
1379 struct nfsd4_session *ses;
1380 u32 status = nfserr_badsession;
1383 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1384 * - Should we return nfserr_back_chan_busy if waiting for
1385 * callbacks on to-be-destroyed session?
1386 * - Do we need to clear any callback info from previous session?
1389 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1390 if (!nfsd4_last_compound_op(r))
1391 return nfserr_not_only_op;
1393 dump_sessionid(__func__, &sessionid->sessionid);
1394 spin_lock(&client_lock);
1395 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1397 spin_unlock(&client_lock);
1401 unhash_session(ses);
1402 spin_unlock(&client_lock);
1404 /* wait for callbacks */
1405 nfsd4_set_callback_client(ses->se_client, NULL);
1406 nfsd4_put_session(ses);
1409 dprintk("%s returns %d\n", __func__, ntohl(status));
1414 nfsd4_sequence(struct svc_rqst *rqstp,
1415 struct nfsd4_compound_state *cstate,
1416 struct nfsd4_sequence *seq)
1418 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1419 struct nfsd4_session *session;
1420 struct nfsd4_slot *slot;
1423 if (resp->opcnt != 1)
1424 return nfserr_sequence_pos;
1426 spin_lock(&client_lock);
1427 status = nfserr_badsession;
1428 session = find_in_sessionid_hashtbl(&seq->sessionid);
1432 status = nfserr_badslot;
1433 if (seq->slotid >= session->se_fchannel.maxreqs)
1436 slot = session->se_slots[seq->slotid];
1437 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1439 /* We do not negotiate the number of slots yet, so set the
1440 * maxslots to the session maxreqs which is used to encode
1441 * sr_highest_slotid and the sr_target_slot id to maxslots */
1442 seq->maxslots = session->se_fchannel.maxreqs;
1444 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1445 if (status == nfserr_replay_cache) {
1446 cstate->slot = slot;
1447 cstate->session = session;
1448 /* Return the cached reply status and set cstate->status
1449 * for nfsd4_proc_compound processing */
1450 status = nfsd4_replay_cache_entry(resp, seq);
1451 cstate->status = nfserr_replay_cache;
1457 /* Success! bump slot seqid */
1458 slot->sl_inuse = true;
1459 slot->sl_seqid = seq->seqid;
1460 slot->sl_cachethis = seq->cachethis;
1462 cstate->slot = slot;
1463 cstate->session = session;
1466 /* Hold a session reference until done processing the compound. */
1467 if (cstate->session) {
1468 nfsd4_get_session(cstate->session);
1469 /* Renew the clientid on success and on replay */
1470 renew_client_locked(session->se_client);
1472 spin_unlock(&client_lock);
1473 dprintk("%s: return %d\n", __func__, ntohl(status));
1478 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1479 struct nfsd4_setclientid *setclid)
1481 struct sockaddr *sa = svc_addr(rqstp);
1482 struct xdr_netobj clname = {
1483 .len = setclid->se_namelen,
1484 .data = setclid->se_name,
1486 nfs4_verifier clverifier = setclid->se_verf;
1487 unsigned int strhashval;
1488 struct nfs4_client *conf, *unconf, *new;
1490 char dname[HEXDIR_LEN];
1492 if (!check_name(clname))
1493 return nfserr_inval;
1495 status = nfs4_make_rec_clidname(dname, &clname);
1500 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1501 * We get here on a DRC miss.
1504 strhashval = clientstr_hashval(dname);
1507 conf = find_confirmed_client_by_str(dname, strhashval, false);
1509 /* RFC 3530 14.2.33 CASE 0: */
1510 status = nfserr_clid_inuse;
1511 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1512 char addr_str[INET6_ADDRSTRLEN];
1513 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1515 dprintk("NFSD: setclientid: string in use by client "
1516 "at %s\n", addr_str);
1521 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1522 * has a description of SETCLIENTID request processing consisting
1523 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1525 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1526 status = nfserr_resource;
1529 * RFC 3530 14.2.33 CASE 4:
1530 * placed first, because it is the normal case
1533 expire_client(unconf);
1534 new = create_client(clname, dname, rqstp, &clverifier);
1538 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1540 * RFC 3530 14.2.33 CASE 1:
1541 * probable callback update
1544 /* Note this is removing unconfirmed {*x***},
1545 * which is stronger than RFC recommended {vxc**}.
1546 * This has the advantage that there is at most
1547 * one {*x***} in either list at any time.
1549 expire_client(unconf);
1551 new = create_client(clname, dname, rqstp, &clverifier);
1554 copy_clid(new, conf);
1555 } else if (!unconf) {
1557 * RFC 3530 14.2.33 CASE 2:
1558 * probable client reboot; state will be removed if
1561 new = create_client(clname, dname, rqstp, &clverifier);
1567 * RFC 3530 14.2.33 CASE 3:
1568 * probable client reboot; state will be removed if
1571 expire_client(unconf);
1572 new = create_client(clname, dname, rqstp, &clverifier);
1577 gen_callback(new, setclid, rpc_get_scope_id(sa));
1578 add_to_unconfirmed(new, strhashval);
1579 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1580 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1581 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1584 nfs4_unlock_state();
1590 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1591 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1592 * bullets, labeled as CASE1 - CASE4 below.
1595 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1596 struct nfsd4_compound_state *cstate,
1597 struct nfsd4_setclientid_confirm *setclientid_confirm)
1599 struct sockaddr *sa = svc_addr(rqstp);
1600 struct nfs4_client *conf, *unconf;
1601 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1602 clientid_t * clid = &setclientid_confirm->sc_clientid;
1605 if (STALE_CLIENTID(clid))
1606 return nfserr_stale_clientid;
1608 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1609 * We get here on a DRC miss.
1614 conf = find_confirmed_client(clid);
1615 unconf = find_unconfirmed_client(clid);
1617 status = nfserr_clid_inuse;
1618 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1620 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1624 * section 14.2.34 of RFC 3530 has a description of
1625 * SETCLIENTID_CONFIRM request processing consisting
1626 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1628 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1630 * RFC 3530 14.2.34 CASE 1:
1633 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1634 status = nfserr_clid_inuse;
1636 atomic_set(&conf->cl_cb_set, 0);
1637 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1638 expire_client(unconf);
1642 } else if (conf && !unconf) {
1644 * RFC 3530 14.2.34 CASE 2:
1645 * probable retransmitted request; play it safe and
1648 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1649 status = nfserr_clid_inuse;
1652 } else if (!conf && unconf
1653 && same_verf(&unconf->cl_confirm, &confirm)) {
1655 * RFC 3530 14.2.34 CASE 3:
1656 * Normal case; new or rebooted client:
1658 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1659 status = nfserr_clid_inuse;
1662 clientstr_hashval(unconf->cl_recdir);
1663 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1666 nfsd4_remove_clid_dir(conf);
1667 expire_client(conf);
1669 move_to_confirmed(unconf);
1671 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1674 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1675 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1678 * RFC 3530 14.2.34 CASE 4:
1679 * Client probably hasn't noticed that we rebooted yet.
1681 status = nfserr_stale_clientid;
1683 /* check that we have hit one of the cases...*/
1684 status = nfserr_clid_inuse;
1687 nfs4_unlock_state();
1691 /* OPEN Share state helper functions */
1692 static inline struct nfs4_file *
1693 alloc_init_file(struct inode *ino)
1695 struct nfs4_file *fp;
1696 unsigned int hashval = file_hashval(ino);
1698 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1700 atomic_set(&fp->fi_ref, 1);
1701 INIT_LIST_HEAD(&fp->fi_hash);
1702 INIT_LIST_HEAD(&fp->fi_stateids);
1703 INIT_LIST_HEAD(&fp->fi_delegations);
1704 spin_lock(&recall_lock);
1705 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1706 spin_unlock(&recall_lock);
1707 fp->fi_inode = igrab(ino);
1708 fp->fi_id = current_fileid++;
1709 fp->fi_had_conflict = false;
1716 nfsd4_free_slab(struct kmem_cache **slab)
1720 kmem_cache_destroy(*slab);
1725 nfsd4_free_slabs(void)
1727 nfsd4_free_slab(&stateowner_slab);
1728 nfsd4_free_slab(&file_slab);
1729 nfsd4_free_slab(&stateid_slab);
1730 nfsd4_free_slab(&deleg_slab);
1734 nfsd4_init_slabs(void)
1736 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1737 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1738 if (stateowner_slab == NULL)
1740 file_slab = kmem_cache_create("nfsd4_files",
1741 sizeof(struct nfs4_file), 0, 0, NULL);
1742 if (file_slab == NULL)
1744 stateid_slab = kmem_cache_create("nfsd4_stateids",
1745 sizeof(struct nfs4_stateid), 0, 0, NULL);
1746 if (stateid_slab == NULL)
1748 deleg_slab = kmem_cache_create("nfsd4_delegations",
1749 sizeof(struct nfs4_delegation), 0, 0, NULL);
1750 if (deleg_slab == NULL)
1755 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1760 nfs4_free_stateowner(struct kref *kref)
1762 struct nfs4_stateowner *sop =
1763 container_of(kref, struct nfs4_stateowner, so_ref);
1764 kfree(sop->so_owner.data);
1765 kmem_cache_free(stateowner_slab, sop);
1768 static inline struct nfs4_stateowner *
1769 alloc_stateowner(struct xdr_netobj *owner)
1771 struct nfs4_stateowner *sop;
1773 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1774 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1775 memcpy(sop->so_owner.data, owner->data, owner->len);
1776 sop->so_owner.len = owner->len;
1777 kref_init(&sop->so_ref);
1780 kmem_cache_free(stateowner_slab, sop);
1785 static struct nfs4_stateowner *
1786 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1787 struct nfs4_stateowner *sop;
1788 struct nfs4_replay *rp;
1789 unsigned int idhashval;
1791 if (!(sop = alloc_stateowner(&open->op_owner)))
1793 idhashval = ownerid_hashval(current_ownerid);
1794 INIT_LIST_HEAD(&sop->so_idhash);
1795 INIT_LIST_HEAD(&sop->so_strhash);
1796 INIT_LIST_HEAD(&sop->so_perclient);
1797 INIT_LIST_HEAD(&sop->so_stateids);
1798 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1799 INIT_LIST_HEAD(&sop->so_close_lru);
1801 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1802 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1803 list_add(&sop->so_perclient, &clp->cl_openowners);
1804 sop->so_is_open_owner = 1;
1805 sop->so_id = current_ownerid++;
1806 sop->so_client = clp;
1807 sop->so_seqid = open->op_seqid;
1808 sop->so_confirmed = 0;
1809 rp = &sop->so_replay;
1810 rp->rp_status = nfserr_serverfault;
1812 rp->rp_buf = rp->rp_ibuf;
1817 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1818 struct nfs4_stateowner *sop = open->op_stateowner;
1819 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1821 INIT_LIST_HEAD(&stp->st_hash);
1822 INIT_LIST_HEAD(&stp->st_perstateowner);
1823 INIT_LIST_HEAD(&stp->st_lockowners);
1824 INIT_LIST_HEAD(&stp->st_perfile);
1825 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1826 list_add(&stp->st_perstateowner, &sop->so_stateids);
1827 list_add(&stp->st_perfile, &fp->fi_stateids);
1828 stp->st_stateowner = sop;
1831 stp->st_stateid.si_boot = get_seconds();
1832 stp->st_stateid.si_stateownerid = sop->so_id;
1833 stp->st_stateid.si_fileid = fp->fi_id;
1834 stp->st_stateid.si_generation = 0;
1835 stp->st_access_bmap = 0;
1836 stp->st_deny_bmap = 0;
1837 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1838 &stp->st_access_bmap);
1839 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1840 stp->st_openstp = NULL;
1844 move_to_close_lru(struct nfs4_stateowner *sop)
1846 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1848 list_move_tail(&sop->so_close_lru, &close_lru);
1849 sop->so_time = get_seconds();
1853 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1856 return (sop->so_owner.len == owner->len) &&
1857 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1858 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1861 static struct nfs4_stateowner *
1862 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1864 struct nfs4_stateowner *so = NULL;
1866 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1867 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1873 /* search file_hashtbl[] for file */
1874 static struct nfs4_file *
1875 find_file(struct inode *ino)
1877 unsigned int hashval = file_hashval(ino);
1878 struct nfs4_file *fp;
1880 spin_lock(&recall_lock);
1881 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1882 if (fp->fi_inode == ino) {
1884 spin_unlock(&recall_lock);
1888 spin_unlock(&recall_lock);
1892 static inline int access_valid(u32 x, u32 minorversion)
1894 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1896 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1898 x &= ~NFS4_SHARE_ACCESS_MASK;
1899 if (minorversion && x) {
1900 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1902 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1904 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1911 static inline int deny_valid(u32 x)
1913 /* Note: unlike access bits, deny bits may be zero. */
1914 return x <= NFS4_SHARE_DENY_BOTH;
1918 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1919 * st_{access,deny}_bmap field of the stateid, in order to track not
1920 * only what share bits are currently in force, but also what
1921 * combinations of share bits previous opens have used. This allows us
1922 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1923 * return an error if the client attempt to downgrade to a combination
1924 * of share bits not explicable by closing some of its previous opens.
1926 * XXX: This enforcement is actually incomplete, since we don't keep
1927 * track of access/deny bit combinations; so, e.g., we allow:
1929 * OPEN allow read, deny write
1930 * OPEN allow both, deny none
1931 * DOWNGRADE allow read, deny none
1933 * which we should reject.
1936 set_access(unsigned int *access, unsigned long bmap) {
1940 for (i = 1; i < 4; i++) {
1941 if (test_bit(i, &bmap))
1947 set_deny(unsigned int *deny, unsigned long bmap) {
1951 for (i = 0; i < 4; i++) {
1952 if (test_bit(i, &bmap))
1958 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1959 unsigned int access, deny;
1961 set_access(&access, stp->st_access_bmap);
1962 set_deny(&deny, stp->st_deny_bmap);
1963 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1969 * Called to check deny when READ with all zero stateid or
1970 * WRITE with all zero or all one stateid
1973 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1975 struct inode *ino = current_fh->fh_dentry->d_inode;
1976 struct nfs4_file *fp;
1977 struct nfs4_stateid *stp;
1980 dprintk("NFSD: nfs4_share_conflict\n");
1982 fp = find_file(ino);
1985 ret = nfserr_locked;
1986 /* Search for conflicting share reservations */
1987 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1988 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1989 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1999 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2001 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2002 drop_file_write_access(filp);
2003 spin_lock(&filp->f_lock);
2004 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2005 spin_unlock(&filp->f_lock);
2010 * Spawn a thread to perform a recall on the delegation represented
2011 * by the lease (file_lock)
2013 * Called from break_lease() with lock_kernel() held.
2014 * Note: we assume break_lease will only call this *once* for any given
2018 void nfsd_break_deleg_cb(struct file_lock *fl)
2020 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2022 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2026 /* We're assuming the state code never drops its reference
2027 * without first removing the lease. Since we're in this lease
2028 * callback (and since the lease code is serialized by the kernel
2029 * lock) we know the server hasn't removed the lease yet, we know
2030 * it's safe to take a reference: */
2031 atomic_inc(&dp->dl_count);
2033 spin_lock(&recall_lock);
2034 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2035 spin_unlock(&recall_lock);
2037 /* only place dl_time is set. protected by lock_kernel*/
2038 dp->dl_time = get_seconds();
2041 * We don't want the locks code to timeout the lease for us;
2042 * we'll remove it ourself if the delegation isn't returned
2045 fl->fl_break_time = 0;
2047 dp->dl_file->fi_had_conflict = true;
2048 nfsd4_cb_recall(dp);
2052 * The file_lock is being reapd.
2054 * Called by locks_free_lock() with lock_kernel() held.
2057 void nfsd_release_deleg_cb(struct file_lock *fl)
2059 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2061 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2063 if (!(fl->fl_flags & FL_LEASE) || !dp)
2065 dp->dl_flock = NULL;
2069 * Set the delegation file_lock back pointer.
2071 * Called from setlease() with lock_kernel() held.
2074 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2076 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2078 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2085 * Called from setlease() with lock_kernel() held
2088 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2090 struct nfs4_delegation *onlistd =
2091 (struct nfs4_delegation *)onlist->fl_owner;
2092 struct nfs4_delegation *tryd =
2093 (struct nfs4_delegation *)try->fl_owner;
2095 if (onlist->fl_lmops != try->fl_lmops)
2098 return onlistd->dl_client == tryd->dl_client;
2103 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2106 return lease_modify(onlist, arg);
2111 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2112 .fl_break = nfsd_break_deleg_cb,
2113 .fl_release_private = nfsd_release_deleg_cb,
2114 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2115 .fl_mylease = nfsd_same_client_deleg_cb,
2116 .fl_change = nfsd_change_deleg_cb,
2121 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2122 struct nfsd4_open *open)
2124 clientid_t *clientid = &open->op_clientid;
2125 struct nfs4_client *clp = NULL;
2126 unsigned int strhashval;
2127 struct nfs4_stateowner *sop = NULL;
2129 if (!check_name(open->op_owner))
2130 return nfserr_inval;
2132 if (STALE_CLIENTID(&open->op_clientid))
2133 return nfserr_stale_clientid;
2135 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2136 sop = find_openstateowner_str(strhashval, open);
2137 open->op_stateowner = sop;
2139 /* Make sure the client's lease hasn't expired. */
2140 clp = find_confirmed_client(clientid);
2142 return nfserr_expired;
2145 /* When sessions are used, skip open sequenceid processing */
2146 if (nfsd4_has_session(cstate))
2148 if (!sop->so_confirmed) {
2149 /* Replace unconfirmed owners without checking for replay. */
2150 clp = sop->so_client;
2151 release_openowner(sop);
2152 open->op_stateowner = NULL;
2155 if (open->op_seqid == sop->so_seqid - 1) {
2156 if (sop->so_replay.rp_buflen)
2157 return nfserr_replay_me;
2158 /* The original OPEN failed so spectacularly
2159 * that we don't even have replay data saved!
2160 * Therefore, we have no choice but to continue
2161 * processing this OPEN; presumably, we'll
2162 * fail again for the same reason.
2164 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2167 if (open->op_seqid != sop->so_seqid)
2168 return nfserr_bad_seqid;
2170 if (open->op_stateowner == NULL) {
2171 sop = alloc_init_open_stateowner(strhashval, clp, open);
2173 return nfserr_resource;
2174 open->op_stateowner = sop;
2176 list_del_init(&sop->so_close_lru);
2177 renew_client(sop->so_client);
2181 static inline __be32
2182 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2184 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2185 return nfserr_openmode;
2190 static struct nfs4_delegation *
2191 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2193 struct nfs4_delegation *dp;
2195 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2196 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2203 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2204 struct nfs4_delegation **dp)
2207 __be32 status = nfserr_bad_stateid;
2209 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2212 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2213 RD_STATE : WR_STATE;
2214 status = nfs4_check_delegmode(*dp, flags);
2218 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2222 open->op_stateowner->so_confirmed = 1;
2227 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2229 struct nfs4_stateid *local;
2230 __be32 status = nfserr_share_denied;
2231 struct nfs4_stateowner *sop = open->op_stateowner;
2233 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2234 /* ignore lock owners */
2235 if (local->st_stateowner->so_is_open_owner == 0)
2237 /* remember if we have seen this open owner */
2238 if (local->st_stateowner == sop)
2240 /* check for conflicting share reservations */
2241 if (!test_share(local, open))
2249 static inline struct nfs4_stateid *
2250 nfs4_alloc_stateid(void)
2252 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2256 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2257 struct nfs4_delegation *dp,
2258 struct svc_fh *cur_fh, int flags)
2260 struct nfs4_stateid *stp;
2262 stp = nfs4_alloc_stateid();
2264 return nfserr_resource;
2267 get_file(dp->dl_vfs_file);
2268 stp->st_vfs_file = dp->dl_vfs_file;
2271 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2274 if (status == nfserr_dropit)
2275 status = nfserr_jukebox;
2276 kmem_cache_free(stateid_slab, stp);
2284 static inline __be32
2285 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2286 struct nfsd4_open *open)
2288 struct iattr iattr = {
2289 .ia_valid = ATTR_SIZE,
2292 if (!open->op_truncate)
2294 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2295 return nfserr_inval;
2296 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2300 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2302 struct file *filp = stp->st_vfs_file;
2303 struct inode *inode = filp->f_path.dentry->d_inode;
2304 unsigned int share_access, new_writer;
2307 set_access(&share_access, stp->st_access_bmap);
2308 new_writer = (~share_access) & open->op_share_access
2309 & NFS4_SHARE_ACCESS_WRITE;
2312 int err = get_write_access(inode);
2314 return nfserrno(err);
2315 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2317 return nfserrno(err);
2318 file_take_write(filp);
2320 status = nfsd4_truncate(rqstp, cur_fh, open);
2323 put_write_access(inode);
2326 /* remember the open */
2327 filp->f_mode |= open->op_share_access;
2328 __set_bit(open->op_share_access, &stp->st_access_bmap);
2329 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2336 nfs4_set_claim_prev(struct nfsd4_open *open)
2338 open->op_stateowner->so_confirmed = 1;
2339 open->op_stateowner->so_client->cl_firststate = 1;
2343 * Attempt to hand out a delegation.
2346 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2348 struct nfs4_delegation *dp;
2349 struct nfs4_stateowner *sop = stp->st_stateowner;
2350 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2351 struct file_lock fl, *flp = &fl;
2352 int status, flag = 0;
2354 flag = NFS4_OPEN_DELEGATE_NONE;
2355 open->op_recall = 0;
2356 switch (open->op_claim_type) {
2357 case NFS4_OPEN_CLAIM_PREVIOUS:
2359 open->op_recall = 1;
2360 flag = open->op_delegate_type;
2361 if (flag == NFS4_OPEN_DELEGATE_NONE)
2364 case NFS4_OPEN_CLAIM_NULL:
2365 /* Let's not give out any delegations till everyone's
2366 * had the chance to reclaim theirs.... */
2367 if (locks_in_grace())
2369 if (!cb_up || !sop->so_confirmed)
2371 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2372 flag = NFS4_OPEN_DELEGATE_WRITE;
2374 flag = NFS4_OPEN_DELEGATE_READ;
2380 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2382 flag = NFS4_OPEN_DELEGATE_NONE;
2385 locks_init_lock(&fl);
2386 fl.fl_lmops = &nfsd_lease_mng_ops;
2387 fl.fl_flags = FL_LEASE;
2388 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2389 fl.fl_end = OFFSET_MAX;
2390 fl.fl_owner = (fl_owner_t)dp;
2391 fl.fl_file = stp->st_vfs_file;
2392 fl.fl_pid = current->tgid;
2394 /* vfs_setlease checks to see if delegation should be handed out.
2395 * the lock_manager callbacks fl_mylease and fl_change are used
2397 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2398 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2399 unhash_delegation(dp);
2400 flag = NFS4_OPEN_DELEGATE_NONE;
2404 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2406 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2407 STATEID_VAL(&dp->dl_stateid));
2409 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2410 && flag == NFS4_OPEN_DELEGATE_NONE
2411 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2412 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2413 open->op_delegate_type = flag;
2417 * called with nfs4_lock_state() held.
2420 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2422 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2423 struct nfs4_file *fp = NULL;
2424 struct inode *ino = current_fh->fh_dentry->d_inode;
2425 struct nfs4_stateid *stp = NULL;
2426 struct nfs4_delegation *dp = NULL;
2429 status = nfserr_inval;
2430 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2431 || !deny_valid(open->op_share_deny))
2434 * Lookup file; if found, lookup stateid and check open request,
2435 * and check for delegations in the process of being recalled.
2436 * If not found, create the nfs4_file struct
2438 fp = find_file(ino);
2440 if ((status = nfs4_check_open(fp, open, &stp)))
2442 status = nfs4_check_deleg(fp, open, &dp);
2446 status = nfserr_bad_stateid;
2447 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2449 status = nfserr_resource;
2450 fp = alloc_init_file(ino);
2456 * OPEN the file, or upgrade an existing OPEN.
2457 * If truncate fails, the OPEN fails.
2460 /* Stateid was found, this is an OPEN upgrade */
2461 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2464 update_stateid(&stp->st_stateid);
2466 /* Stateid was not found, this is a new OPEN */
2468 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2469 flags |= NFSD_MAY_READ;
2470 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2471 flags |= NFSD_MAY_WRITE;
2472 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2475 init_stateid(stp, fp, open);
2476 status = nfsd4_truncate(rqstp, current_fh, open);
2478 release_open_stateid(stp);
2481 if (nfsd4_has_session(&resp->cstate))
2482 update_stateid(&stp->st_stateid);
2484 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2486 if (nfsd4_has_session(&resp->cstate)) {
2487 open->op_stateowner->so_confirmed = 1;
2488 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2492 * Attempt to hand out a delegation. No error return, because the
2493 * OPEN succeeds even if we fail.
2495 nfs4_open_delegation(current_fh, open, stp);
2499 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2500 STATEID_VAL(&stp->st_stateid));
2504 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2505 nfs4_set_claim_prev(open);
2507 * To finish the open response, we just need to set the rflags.
2509 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2510 if (!open->op_stateowner->so_confirmed &&
2511 !nfsd4_has_session(&resp->cstate))
2512 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2518 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2521 struct nfs4_client *clp;
2525 dprintk("process_renew(%08x/%08x): starting\n",
2526 clid->cl_boot, clid->cl_id);
2527 status = nfserr_stale_clientid;
2528 if (STALE_CLIENTID(clid))
2530 clp = find_confirmed_client(clid);
2531 status = nfserr_expired;
2533 /* We assume the client took too long to RENEW. */
2534 dprintk("nfsd4_renew: clientid not found!\n");
2538 status = nfserr_cb_path_down;
2539 if (!list_empty(&clp->cl_delegations)
2540 && !atomic_read(&clp->cl_cb_set))
2544 nfs4_unlock_state();
2548 struct lock_manager nfsd4_manager = {
2552 nfsd4_end_grace(void)
2554 dprintk("NFSD: end of grace period\n");
2555 nfsd4_recdir_purge_old();
2556 locks_end_grace(&nfsd4_manager);
2558 * Now that every NFSv4 client has had the chance to recover and
2559 * to see the (possibly new, possibly shorter) lease time, we
2560 * can safely set the next grace time to the current lease time:
2562 nfsd4_grace = nfsd4_lease;
2566 nfs4_laundromat(void)
2568 struct nfs4_client *clp;
2569 struct nfs4_stateowner *sop;
2570 struct nfs4_delegation *dp;
2571 struct list_head *pos, *next, reaplist;
2572 time_t cutoff = get_seconds() - nfsd4_lease;
2573 time_t t, clientid_val = nfsd4_lease;
2574 time_t u, test_val = nfsd4_lease;
2578 dprintk("NFSD: laundromat service - starting\n");
2579 if (locks_in_grace())
2581 INIT_LIST_HEAD(&reaplist);
2582 spin_lock(&client_lock);
2583 list_for_each_safe(pos, next, &client_lru) {
2584 clp = list_entry(pos, struct nfs4_client, cl_lru);
2585 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2586 t = clp->cl_time - cutoff;
2587 if (clientid_val > t)
2591 list_move(&clp->cl_lru, &reaplist);
2593 spin_unlock(&client_lock);
2594 list_for_each_safe(pos, next, &reaplist) {
2595 clp = list_entry(pos, struct nfs4_client, cl_lru);
2596 dprintk("NFSD: purging unused client (clientid %08x)\n",
2597 clp->cl_clientid.cl_id);
2598 nfsd4_remove_clid_dir(clp);
2601 spin_lock(&recall_lock);
2602 list_for_each_safe(pos, next, &del_recall_lru) {
2603 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2604 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2605 u = dp->dl_time - cutoff;
2610 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2612 list_move(&dp->dl_recall_lru, &reaplist);
2614 spin_unlock(&recall_lock);
2615 list_for_each_safe(pos, next, &reaplist) {
2616 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2617 list_del_init(&dp->dl_recall_lru);
2618 unhash_delegation(dp);
2620 test_val = nfsd4_lease;
2621 list_for_each_safe(pos, next, &close_lru) {
2622 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2623 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2624 u = sop->so_time - cutoff;
2629 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2631 release_openowner(sop);
2633 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2634 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2635 nfs4_unlock_state();
2636 return clientid_val;
2639 static struct workqueue_struct *laundry_wq;
2640 static void laundromat_main(struct work_struct *);
2641 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2644 laundromat_main(struct work_struct *not_used)
2648 t = nfs4_laundromat();
2649 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2650 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2653 static struct nfs4_stateowner *
2654 search_close_lru(u32 st_id, int flags)
2656 struct nfs4_stateowner *local = NULL;
2658 if (flags & CLOSE_STATE) {
2659 list_for_each_entry(local, &close_lru, so_close_lru) {
2660 if (local->so_id == st_id)
2668 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2670 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2674 STALE_STATEID(stateid_t *stateid)
2676 if (time_after((unsigned long)boot_time,
2677 (unsigned long)stateid->si_boot)) {
2678 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2679 STATEID_VAL(stateid));
2686 EXPIRED_STATEID(stateid_t *stateid)
2688 if (time_before((unsigned long)boot_time,
2689 ((unsigned long)stateid->si_boot)) &&
2690 time_before((unsigned long)(stateid->si_boot + nfsd4_lease), get_seconds())) {
2691 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2692 STATEID_VAL(stateid));
2699 stateid_error_map(stateid_t *stateid)
2701 if (STALE_STATEID(stateid))
2702 return nfserr_stale_stateid;
2703 if (EXPIRED_STATEID(stateid))
2704 return nfserr_expired;
2706 dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2707 STATEID_VAL(stateid));
2708 return nfserr_bad_stateid;
2712 access_permit_read(unsigned long access_bmap)
2714 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2715 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2716 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2720 access_permit_write(unsigned long access_bmap)
2722 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2723 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2727 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2729 __be32 status = nfserr_openmode;
2731 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2733 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2740 static inline __be32
2741 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2743 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2745 else if (locks_in_grace()) {
2746 /* Answer in remaining cases depends on existance of
2747 * conflicting state; so we must wait out the grace period. */
2748 return nfserr_grace;
2749 } else if (flags & WR_STATE)
2750 return nfs4_share_conflict(current_fh,
2751 NFS4_SHARE_DENY_WRITE);
2752 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2753 return nfs4_share_conflict(current_fh,
2754 NFS4_SHARE_DENY_READ);
2758 * Allow READ/WRITE during grace period on recovered state only for files
2759 * that are not able to provide mandatory locking.
2762 grace_disallows_io(struct inode *inode)
2764 return locks_in_grace() && mandatory_lock(inode);
2767 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2770 * When sessions are used the stateid generation number is ignored
2773 if ((flags & HAS_SESSION) && in->si_generation == 0)
2776 /* If the client sends us a stateid from the future, it's buggy: */
2777 if (in->si_generation > ref->si_generation)
2778 return nfserr_bad_stateid;
2780 * The following, however, can happen. For example, if the
2781 * client sends an open and some IO at the same time, the open
2782 * may bump si_generation while the IO is still in flight.
2783 * Thanks to hard links and renames, the client never knows what
2784 * file an open will affect. So it could avoid that situation
2785 * only by serializing all opens and IO from the same open
2786 * owner. To recover from the old_stateid error, the client
2787 * will just have to retry the IO:
2789 if (in->si_generation < ref->si_generation)
2790 return nfserr_old_stateid;
2795 static int is_delegation_stateid(stateid_t *stateid)
2797 return stateid->si_fileid == 0;
2801 * Checks for stateid operations
2804 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2805 stateid_t *stateid, int flags, struct file **filpp)
2807 struct nfs4_stateid *stp = NULL;
2808 struct nfs4_delegation *dp = NULL;
2809 struct svc_fh *current_fh = &cstate->current_fh;
2810 struct inode *ino = current_fh->fh_dentry->d_inode;
2816 if (grace_disallows_io(ino))
2817 return nfserr_grace;
2819 if (nfsd4_has_session(cstate))
2820 flags |= HAS_SESSION;
2822 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2823 return check_special_stateids(current_fh, stateid, flags);
2825 status = nfserr_stale_stateid;
2826 if (STALE_STATEID(stateid))
2829 status = nfserr_bad_stateid;
2830 if (is_delegation_stateid(stateid)) {
2831 dp = find_delegation_stateid(ino, stateid);
2833 status = stateid_error_map(stateid);
2836 status = check_stateid_generation(stateid, &dp->dl_stateid,
2840 status = nfs4_check_delegmode(dp, flags);
2843 renew_client(dp->dl_client);
2845 *filpp = dp->dl_vfs_file;
2846 } else { /* open or lock stateid */
2847 stp = find_stateid(stateid, flags);
2849 status = stateid_error_map(stateid);
2852 if (nfs4_check_fh(current_fh, stp))
2854 if (!stp->st_stateowner->so_confirmed)
2856 status = check_stateid_generation(stateid, &stp->st_stateid,
2860 status = nfs4_check_openmode(stp, flags);
2863 renew_client(stp->st_stateowner->so_client);
2865 *filpp = stp->st_vfs_file;
2875 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2876 RD_STATE : WR_STATE;
2880 * Checks for sequence id mutating operations.
2883 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2884 stateid_t *stateid, int flags,
2885 struct nfs4_stateowner **sopp,
2886 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2888 struct nfs4_stateid *stp;
2889 struct nfs4_stateowner *sop;
2890 struct svc_fh *current_fh = &cstate->current_fh;
2893 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2894 seqid, STATEID_VAL(stateid));
2899 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2900 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2901 return nfserr_bad_stateid;
2904 if (STALE_STATEID(stateid))
2905 return nfserr_stale_stateid;
2907 if (nfsd4_has_session(cstate))
2908 flags |= HAS_SESSION;
2911 * We return BAD_STATEID if filehandle doesn't match stateid,
2912 * the confirmed flag is incorrecly set, or the generation
2913 * number is incorrect.
2915 stp = find_stateid(stateid, flags);
2918 * Also, we should make sure this isn't just the result of
2921 sop = search_close_lru(stateid->si_stateownerid, flags);
2923 return stateid_error_map(stateid);
2929 *sopp = sop = stp->st_stateowner;
2932 clientid_t *lockclid = &lock->v.new.clientid;
2933 struct nfs4_client *clp = sop->so_client;
2937 lkflg = setlkflg(lock->lk_type);
2939 if (lock->lk_is_new) {
2940 if (!sop->so_is_open_owner)
2941 return nfserr_bad_stateid;
2942 if (!(flags & HAS_SESSION) &&
2943 !same_clid(&clp->cl_clientid, lockclid))
2944 return nfserr_bad_stateid;
2945 /* stp is the open stateid */
2946 status = nfs4_check_openmode(stp, lkflg);
2950 /* stp is the lock stateid */
2951 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2957 if (nfs4_check_fh(current_fh, stp)) {
2958 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2959 return nfserr_bad_stateid;
2963 * We now validate the seqid and stateid generation numbers.
2964 * For the moment, we ignore the possibility of
2965 * generation number wraparound.
2967 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2970 if (sop->so_confirmed && flags & CONFIRM) {
2971 dprintk("NFSD: preprocess_seqid_op: expected"
2972 " unconfirmed stateowner!\n");
2973 return nfserr_bad_stateid;
2975 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2976 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2977 " confirmed yet!\n");
2978 return nfserr_bad_stateid;
2980 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2983 renew_client(sop->so_client);
2987 if (seqid == sop->so_seqid - 1) {
2988 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2989 /* indicate replay to calling function */
2990 return nfserr_replay_me;
2992 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2993 sop->so_seqid, seqid);
2995 return nfserr_bad_seqid;
2999 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3000 struct nfsd4_open_confirm *oc)
3003 struct nfs4_stateowner *sop;
3004 struct nfs4_stateid *stp;
3006 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3007 (int)cstate->current_fh.fh_dentry->d_name.len,
3008 cstate->current_fh.fh_dentry->d_name.name);
3010 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3016 if ((status = nfs4_preprocess_seqid_op(cstate,
3017 oc->oc_seqid, &oc->oc_req_stateid,
3018 CONFIRM | OPEN_STATE,
3019 &oc->oc_stateowner, &stp, NULL)))
3022 sop = oc->oc_stateowner;
3023 sop->so_confirmed = 1;
3024 update_stateid(&stp->st_stateid);
3025 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3026 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3027 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3029 nfsd4_create_clid_dir(sop->so_client);
3031 if (oc->oc_stateowner) {
3032 nfs4_get_stateowner(oc->oc_stateowner);
3033 cstate->replay_owner = oc->oc_stateowner;
3035 nfs4_unlock_state();
3041 * unset all bits in union bitmap (bmap) that
3042 * do not exist in share (from successful OPEN_DOWNGRADE)
3045 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3048 for (i = 1; i < 4; i++) {
3049 if ((i & access) != i)
3050 __clear_bit(i, bmap);
3055 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3058 for (i = 0; i < 4; i++) {
3059 if ((i & deny) != i)
3060 __clear_bit(i, bmap);
3065 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3066 struct nfsd4_compound_state *cstate,
3067 struct nfsd4_open_downgrade *od)
3070 struct nfs4_stateid *stp;
3071 unsigned int share_access;
3073 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3074 (int)cstate->current_fh.fh_dentry->d_name.len,
3075 cstate->current_fh.fh_dentry->d_name.name);
3077 if (!access_valid(od->od_share_access, cstate->minorversion)
3078 || !deny_valid(od->od_share_deny))
3079 return nfserr_inval;
3082 if ((status = nfs4_preprocess_seqid_op(cstate,
3086 &od->od_stateowner, &stp, NULL)))
3089 status = nfserr_inval;
3090 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3091 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3092 stp->st_access_bmap, od->od_share_access);
3095 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3096 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3097 stp->st_deny_bmap, od->od_share_deny);
3100 set_access(&share_access, stp->st_access_bmap);
3101 nfs4_file_downgrade(stp->st_vfs_file,
3102 share_access & ~od->od_share_access);
3104 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3105 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3107 update_stateid(&stp->st_stateid);
3108 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3111 if (od->od_stateowner) {
3112 nfs4_get_stateowner(od->od_stateowner);
3113 cstate->replay_owner = od->od_stateowner;
3115 nfs4_unlock_state();
3120 * nfs4_unlock_state() called after encode
3123 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3124 struct nfsd4_close *close)
3127 struct nfs4_stateid *stp;
3129 dprintk("NFSD: nfsd4_close on file %.*s\n",
3130 (int)cstate->current_fh.fh_dentry->d_name.len,
3131 cstate->current_fh.fh_dentry->d_name.name);
3134 /* check close_lru for replay */
3135 if ((status = nfs4_preprocess_seqid_op(cstate,
3138 OPEN_STATE | CLOSE_STATE,
3139 &close->cl_stateowner, &stp, NULL)))
3142 update_stateid(&stp->st_stateid);
3143 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3145 /* release_stateid() calls nfsd_close() if needed */
3146 release_open_stateid(stp);
3148 /* place unused nfs4_stateowners on so_close_lru list to be
3149 * released by the laundromat service after the lease period
3150 * to enable us to handle CLOSE replay
3152 if (list_empty(&close->cl_stateowner->so_stateids))
3153 move_to_close_lru(close->cl_stateowner);
3155 if (close->cl_stateowner) {
3156 nfs4_get_stateowner(close->cl_stateowner);
3157 cstate->replay_owner = close->cl_stateowner;
3159 nfs4_unlock_state();
3164 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3165 struct nfsd4_delegreturn *dr)
3167 struct nfs4_delegation *dp;
3168 stateid_t *stateid = &dr->dr_stateid;
3169 struct inode *inode;
3173 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3175 inode = cstate->current_fh.fh_dentry->d_inode;
3177 if (nfsd4_has_session(cstate))
3178 flags |= HAS_SESSION;
3180 status = nfserr_bad_stateid;
3181 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3183 status = nfserr_stale_stateid;
3184 if (STALE_STATEID(stateid))
3186 status = nfserr_bad_stateid;
3187 if (!is_delegation_stateid(stateid))
3189 dp = find_delegation_stateid(inode, stateid);
3191 status = stateid_error_map(stateid);
3194 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3197 renew_client(dp->dl_client);
3199 unhash_delegation(dp);
3201 nfs4_unlock_state();
3208 * Lock owner state (byte-range locks)
3210 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3211 #define LOCK_HASH_BITS 8
3212 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3213 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3216 end_offset(u64 start, u64 len)
3221 return end >= start ? end: NFS4_MAX_UINT64;
3224 /* last octet in a range */
3226 last_byte_offset(u64 start, u64 len)
3232 return end > start ? end - 1: NFS4_MAX_UINT64;
3235 #define lockownerid_hashval(id) \
3236 ((id) & LOCK_HASH_MASK)
3238 static inline unsigned int
3239 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3240 struct xdr_netobj *ownername)
3242 return (file_hashval(inode) + cl_id
3243 + opaque_hashval(ownername->data, ownername->len))
3247 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3248 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3249 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3251 static struct nfs4_stateid *
3252 find_stateid(stateid_t *stid, int flags)
3254 struct nfs4_stateid *local;
3255 u32 st_id = stid->si_stateownerid;
3256 u32 f_id = stid->si_fileid;
3257 unsigned int hashval;
3259 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3260 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3261 hashval = stateid_hashval(st_id, f_id);
3262 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3263 if ((local->st_stateid.si_stateownerid == st_id) &&
3264 (local->st_stateid.si_fileid == f_id))
3269 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3270 hashval = stateid_hashval(st_id, f_id);
3271 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3272 if ((local->st_stateid.si_stateownerid == st_id) &&
3273 (local->st_stateid.si_fileid == f_id))
3280 static struct nfs4_delegation *
3281 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3283 struct nfs4_file *fp;
3284 struct nfs4_delegation *dl;
3286 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3289 fp = find_file(ino);
3292 dl = find_delegation_file(fp, stid);
3298 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3299 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3300 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3301 * locking, this prevents us from being completely protocol-compliant. The
3302 * real solution to this problem is to start using unsigned file offsets in
3303 * the VFS, but this is a very deep change!
3306 nfs4_transform_lock_offset(struct file_lock *lock)
3308 if (lock->fl_start < 0)
3309 lock->fl_start = OFFSET_MAX;
3310 if (lock->fl_end < 0)
3311 lock->fl_end = OFFSET_MAX;
3314 /* Hack!: For now, we're defining this just so we can use a pointer to it
3315 * as a unique cookie to identify our (NFSv4's) posix locks. */
3316 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3320 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3322 struct nfs4_stateowner *sop;
3325 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3326 sop = (struct nfs4_stateowner *) fl->fl_owner;
3327 hval = lockownerid_hashval(sop->so_id);
3328 kref_get(&sop->so_ref);
3330 deny->ld_clientid = sop->so_client->cl_clientid;
3332 deny->ld_sop = NULL;
3333 deny->ld_clientid.cl_boot = 0;
3334 deny->ld_clientid.cl_id = 0;
3336 deny->ld_start = fl->fl_start;
3337 deny->ld_length = NFS4_MAX_UINT64;
3338 if (fl->fl_end != NFS4_MAX_UINT64)
3339 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3340 deny->ld_type = NFS4_READ_LT;
3341 if (fl->fl_type != F_RDLCK)
3342 deny->ld_type = NFS4_WRITE_LT;
3345 static struct nfs4_stateowner *
3346 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3347 struct xdr_netobj *owner)
3349 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3350 struct nfs4_stateowner *op;
3352 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3353 if (same_owner_str(op, owner, clid))
3360 * Alloc a lock owner structure.
3361 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3364 * strhashval = lock_ownerstr_hashval
3367 static struct nfs4_stateowner *
3368 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3369 struct nfs4_stateowner *sop;
3370 struct nfs4_replay *rp;
3371 unsigned int idhashval;
3373 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3375 idhashval = lockownerid_hashval(current_ownerid);
3376 INIT_LIST_HEAD(&sop->so_idhash);
3377 INIT_LIST_HEAD(&sop->so_strhash);
3378 INIT_LIST_HEAD(&sop->so_perclient);
3379 INIT_LIST_HEAD(&sop->so_stateids);
3380 INIT_LIST_HEAD(&sop->so_perstateid);
3381 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3383 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3384 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3385 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3386 sop->so_is_open_owner = 0;
3387 sop->so_id = current_ownerid++;
3388 sop->so_client = clp;
3389 /* It is the openowner seqid that will be incremented in encode in the
3390 * case of new lockowners; so increment the lock seqid manually: */
3391 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3392 sop->so_confirmed = 1;
3393 rp = &sop->so_replay;
3394 rp->rp_status = nfserr_serverfault;
3396 rp->rp_buf = rp->rp_ibuf;
3400 static struct nfs4_stateid *
3401 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3403 struct nfs4_stateid *stp;
3404 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3406 stp = nfs4_alloc_stateid();
3409 INIT_LIST_HEAD(&stp->st_hash);
3410 INIT_LIST_HEAD(&stp->st_perfile);
3411 INIT_LIST_HEAD(&stp->st_perstateowner);
3412 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3413 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3414 list_add(&stp->st_perfile, &fp->fi_stateids);
3415 list_add(&stp->st_perstateowner, &sop->so_stateids);
3416 stp->st_stateowner = sop;
3419 stp->st_stateid.si_boot = get_seconds();
3420 stp->st_stateid.si_stateownerid = sop->so_id;
3421 stp->st_stateid.si_fileid = fp->fi_id;
3422 stp->st_stateid.si_generation = 0;
3423 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3424 stp->st_access_bmap = open_stp->st_access_bmap;
3425 stp->st_deny_bmap = open_stp->st_deny_bmap;
3426 stp->st_openstp = open_stp;
3433 check_lock_length(u64 offset, u64 length)
3435 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3436 LOFF_OVERFLOW(offset, length)));
3443 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3444 struct nfsd4_lock *lock)
3446 struct nfs4_stateowner *open_sop = NULL;
3447 struct nfs4_stateowner *lock_sop = NULL;
3448 struct nfs4_stateid *lock_stp;
3450 struct file_lock file_lock;
3451 struct file_lock conflock;
3453 unsigned int strhashval;
3457 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3458 (long long) lock->lk_offset,
3459 (long long) lock->lk_length);
3461 if (check_lock_length(lock->lk_offset, lock->lk_length))
3462 return nfserr_inval;
3464 if ((status = fh_verify(rqstp, &cstate->current_fh,
3465 S_IFREG, NFSD_MAY_LOCK))) {
3466 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3472 if (lock->lk_is_new) {
3474 * Client indicates that this is a new lockowner.
3475 * Use open owner and open stateid to create lock owner and
3478 struct nfs4_stateid *open_stp = NULL;
3479 struct nfs4_file *fp;
3481 status = nfserr_stale_clientid;
3482 if (!nfsd4_has_session(cstate) &&
3483 STALE_CLIENTID(&lock->lk_new_clientid))
3486 /* validate and update open stateid and open seqid */
3487 status = nfs4_preprocess_seqid_op(cstate,
3488 lock->lk_new_open_seqid,
3489 &lock->lk_new_open_stateid,
3491 &lock->lk_replay_owner, &open_stp,
3495 open_sop = lock->lk_replay_owner;
3496 /* create lockowner and lock stateid */
3497 fp = open_stp->st_file;
3498 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3499 open_sop->so_client->cl_clientid.cl_id,
3500 &lock->v.new.owner);
3501 /* XXX: Do we need to check for duplicate stateowners on
3502 * the same file, or should they just be allowed (and
3503 * create new stateids)? */
3504 status = nfserr_resource;
3505 lock_sop = alloc_init_lock_stateowner(strhashval,
3506 open_sop->so_client, open_stp, lock);
3507 if (lock_sop == NULL)
3509 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3510 if (lock_stp == NULL)
3513 /* lock (lock owner + lock stateid) already exists */
3514 status = nfs4_preprocess_seqid_op(cstate,
3515 lock->lk_old_lock_seqid,
3516 &lock->lk_old_lock_stateid,
3518 &lock->lk_replay_owner, &lock_stp, lock);
3521 lock_sop = lock->lk_replay_owner;
3523 /* lock->lk_replay_owner and lock_stp have been created or found */
3524 filp = lock_stp->st_vfs_file;
3526 status = nfserr_grace;
3527 if (locks_in_grace() && !lock->lk_reclaim)
3529 status = nfserr_no_grace;
3530 if (!locks_in_grace() && lock->lk_reclaim)
3533 locks_init_lock(&file_lock);
3534 switch (lock->lk_type) {
3537 file_lock.fl_type = F_RDLCK;
3541 case NFS4_WRITEW_LT:
3542 file_lock.fl_type = F_WRLCK;
3546 status = nfserr_inval;
3549 file_lock.fl_owner = (fl_owner_t)lock_sop;
3550 file_lock.fl_pid = current->tgid;
3551 file_lock.fl_file = filp;
3552 file_lock.fl_flags = FL_POSIX;
3553 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3555 file_lock.fl_start = lock->lk_offset;
3556 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3557 nfs4_transform_lock_offset(&file_lock);
3560 * Try to lock the file in the VFS.
3561 * Note: locks.c uses the BKL to protect the inode's lock list.
3564 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3566 case 0: /* success! */
3567 update_stateid(&lock_stp->st_stateid);
3568 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3572 case (EAGAIN): /* conflock holds conflicting lock */
3573 status = nfserr_denied;
3574 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3575 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3578 status = nfserr_deadlock;
3581 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3582 status = nfserr_resource;
3586 if (status && lock->lk_is_new && lock_sop)
3587 release_lockowner(lock_sop);
3588 if (lock->lk_replay_owner) {
3589 nfs4_get_stateowner(lock->lk_replay_owner);
3590 cstate->replay_owner = lock->lk_replay_owner;
3592 nfs4_unlock_state();
3597 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3598 * so we do a temporary open here just to get an open file to pass to
3599 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3602 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3607 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3610 err = vfs_test_lock(file, lock);
3619 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3620 struct nfsd4_lockt *lockt)
3622 struct inode *inode;
3623 struct file_lock file_lock;
3627 if (locks_in_grace())
3628 return nfserr_grace;
3630 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3631 return nfserr_inval;
3633 lockt->lt_stateowner = NULL;
3636 status = nfserr_stale_clientid;
3637 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3640 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3641 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3642 if (status == nfserr_symlink)
3643 status = nfserr_inval;
3647 inode = cstate->current_fh.fh_dentry->d_inode;
3648 locks_init_lock(&file_lock);
3649 switch (lockt->lt_type) {
3652 file_lock.fl_type = F_RDLCK;
3655 case NFS4_WRITEW_LT:
3656 file_lock.fl_type = F_WRLCK;
3659 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3660 status = nfserr_inval;
3664 lockt->lt_stateowner = find_lockstateowner_str(inode,
3665 &lockt->lt_clientid, &lockt->lt_owner);
3666 if (lockt->lt_stateowner)
3667 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3668 file_lock.fl_pid = current->tgid;
3669 file_lock.fl_flags = FL_POSIX;
3671 file_lock.fl_start = lockt->lt_offset;
3672 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3674 nfs4_transform_lock_offset(&file_lock);
3677 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3679 status = nfserrno(error);
3682 if (file_lock.fl_type != F_UNLCK) {
3683 status = nfserr_denied;
3684 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3687 nfs4_unlock_state();
3692 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3693 struct nfsd4_locku *locku)
3695 struct nfs4_stateid *stp;
3696 struct file *filp = NULL;
3697 struct file_lock file_lock;
3701 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3702 (long long) locku->lu_offset,
3703 (long long) locku->lu_length);
3705 if (check_lock_length(locku->lu_offset, locku->lu_length))
3706 return nfserr_inval;
3710 if ((status = nfs4_preprocess_seqid_op(cstate,
3714 &locku->lu_stateowner, &stp, NULL)))
3717 filp = stp->st_vfs_file;
3719 locks_init_lock(&file_lock);
3720 file_lock.fl_type = F_UNLCK;
3721 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3722 file_lock.fl_pid = current->tgid;
3723 file_lock.fl_file = filp;
3724 file_lock.fl_flags = FL_POSIX;
3725 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3726 file_lock.fl_start = locku->lu_offset;
3728 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3729 nfs4_transform_lock_offset(&file_lock);
3732 * Try to unlock the file in the VFS.
3734 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3736 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3740 * OK, unlock succeeded; the only thing left to do is update the stateid.
3742 update_stateid(&stp->st_stateid);
3743 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3746 if (locku->lu_stateowner) {
3747 nfs4_get_stateowner(locku->lu_stateowner);
3748 cstate->replay_owner = locku->lu_stateowner;
3750 nfs4_unlock_state();
3754 status = nfserrno(err);
3760 * 1: locks held by lockowner
3761 * 0: no locks held by lockowner
3764 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3766 struct file_lock **flpp;
3767 struct inode *inode = filp->f_path.dentry->d_inode;
3771 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3772 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3783 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3784 struct nfsd4_compound_state *cstate,
3785 struct nfsd4_release_lockowner *rlockowner)
3787 clientid_t *clid = &rlockowner->rl_clientid;
3788 struct nfs4_stateowner *sop;
3789 struct nfs4_stateid *stp;
3790 struct xdr_netobj *owner = &rlockowner->rl_owner;
3791 struct list_head matches;
3795 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3796 clid->cl_boot, clid->cl_id);
3798 /* XXX check for lease expiration */
3800 status = nfserr_stale_clientid;
3801 if (STALE_CLIENTID(clid))
3806 status = nfserr_locks_held;
3807 /* XXX: we're doing a linear search through all the lockowners.
3808 * Yipes! For now we'll just hope clients aren't really using
3809 * release_lockowner much, but eventually we have to fix these
3810 * data structures. */
3811 INIT_LIST_HEAD(&matches);
3812 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3813 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3814 if (!same_owner_str(sop, owner, clid))
3816 list_for_each_entry(stp, &sop->so_stateids,
3818 if (check_for_locks(stp->st_vfs_file, sop))
3820 /* Note: so_perclient unused for lockowners,
3821 * so it's OK to fool with here. */
3822 list_add(&sop->so_perclient, &matches);
3826 /* Clients probably won't expect us to return with some (but not all)
3827 * of the lockowner state released; so don't release any until all
3828 * have been checked. */
3830 while (!list_empty(&matches)) {
3831 sop = list_entry(matches.next, struct nfs4_stateowner,
3833 /* unhash_stateowner deletes so_perclient only
3834 * for openowners. */
3835 list_del(&sop->so_perclient);
3836 release_lockowner(sop);
3839 nfs4_unlock_state();
3843 static inline struct nfs4_client_reclaim *
3846 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3850 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3852 unsigned int strhashval = clientstr_hashval(name);
3853 struct nfs4_client *clp;
3855 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3860 * failure => all reset bets are off, nfserr_no_grace...
3863 nfs4_client_to_reclaim(const char *name)
3865 unsigned int strhashval;
3866 struct nfs4_client_reclaim *crp = NULL;
3868 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3869 crp = alloc_reclaim();
3872 strhashval = clientstr_hashval(name);
3873 INIT_LIST_HEAD(&crp->cr_strhash);
3874 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3875 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3876 reclaim_str_hashtbl_size++;
3881 nfs4_release_reclaim(void)
3883 struct nfs4_client_reclaim *crp = NULL;
3886 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3887 while (!list_empty(&reclaim_str_hashtbl[i])) {
3888 crp = list_entry(reclaim_str_hashtbl[i].next,
3889 struct nfs4_client_reclaim, cr_strhash);
3890 list_del(&crp->cr_strhash);
3892 reclaim_str_hashtbl_size--;
3895 BUG_ON(reclaim_str_hashtbl_size);
3899 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3900 static struct nfs4_client_reclaim *
3901 nfs4_find_reclaim_client(clientid_t *clid)
3903 unsigned int strhashval;
3904 struct nfs4_client *clp;
3905 struct nfs4_client_reclaim *crp = NULL;
3908 /* find clientid in conf_id_hashtbl */
3909 clp = find_confirmed_client(clid);
3913 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3914 clp->cl_name.len, clp->cl_name.data,
3917 /* find clp->cl_name in reclaim_str_hashtbl */
3918 strhashval = clientstr_hashval(clp->cl_recdir);
3919 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3920 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3928 * Called from OPEN. Look for clientid in reclaim list.
3931 nfs4_check_open_reclaim(clientid_t *clid)
3933 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3936 /* initialization to perform at module load time: */
3939 nfs4_state_init(void)
3943 status = nfsd4_init_slabs();
3946 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3947 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3948 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3949 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3950 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3951 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3953 for (i = 0; i < SESSION_HASH_SIZE; i++)
3954 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3955 for (i = 0; i < FILE_HASH_SIZE; i++) {
3956 INIT_LIST_HEAD(&file_hashtbl[i]);
3958 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3959 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3960 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3962 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3963 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3964 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3966 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3967 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3968 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3970 memset(&onestateid, ~0, sizeof(stateid_t));
3971 INIT_LIST_HEAD(&close_lru);
3972 INIT_LIST_HEAD(&client_lru);
3973 INIT_LIST_HEAD(&del_recall_lru);
3974 reclaim_str_hashtbl_size = 0;
3979 nfsd4_load_reboot_recovery_data(void)
3984 nfsd4_init_recdir(user_recovery_dirname);
3985 status = nfsd4_recdir_load();
3986 nfs4_unlock_state();
3988 printk("NFSD: Failure reading reboot recovery data\n");
3992 * Since the lifetime of a delegation isn't limited to that of an open, a
3993 * client may quite reasonably hang on to a delegation as long as it has
3994 * the inode cached. This becomes an obvious problem the first time a
3995 * client's inode cache approaches the size of the server's total memory.
3997 * For now we avoid this problem by imposing a hard limit on the number
3998 * of delegations, which varies according to the server's memory size.
4001 set_max_delegations(void)
4004 * Allow at most 4 delegations per megabyte of RAM. Quick
4005 * estimates suggest that in the worst case (where every delegation
4006 * is for a different inode), a delegation could take about 1.5K,
4007 * giving a worst case usage of about 6% of memory.
4009 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4012 /* initialization to perform when the nfsd service is started: */
4015 __nfs4_state_start(void)
4019 boot_time = get_seconds();
4020 locks_start_grace(&nfsd4_manager);
4021 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4023 ret = set_callback_cred();
4026 laundry_wq = create_singlethread_workqueue("nfsd4");
4027 if (laundry_wq == NULL)
4029 ret = nfsd4_create_callback_queue();
4031 goto out_free_laundry;
4032 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4033 set_max_delegations();
4036 destroy_workqueue(laundry_wq);
4041 nfs4_state_start(void)
4047 nfsd4_load_reboot_recovery_data();
4048 ret = __nfs4_state_start();
4056 __nfs4_state_shutdown(void)
4059 struct nfs4_client *clp = NULL;
4060 struct nfs4_delegation *dp = NULL;
4061 struct list_head *pos, *next, reaplist;
4063 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4064 while (!list_empty(&conf_id_hashtbl[i])) {
4065 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4068 while (!list_empty(&unconf_str_hashtbl[i])) {
4069 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4073 INIT_LIST_HEAD(&reaplist);
4074 spin_lock(&recall_lock);
4075 list_for_each_safe(pos, next, &del_recall_lru) {
4076 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4077 list_move(&dp->dl_recall_lru, &reaplist);
4079 spin_unlock(&recall_lock);
4080 list_for_each_safe(pos, next, &reaplist) {
4081 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4082 list_del_init(&dp->dl_recall_lru);
4083 unhash_delegation(dp);
4086 nfsd4_shutdown_recdir();
4091 nfs4_state_shutdown(void)
4093 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4094 destroy_workqueue(laundry_wq);
4095 locks_end_grace(&nfsd4_manager);
4097 nfs4_release_reclaim();
4098 __nfs4_state_shutdown();
4099 nfsd4_destroy_callback_queue();
4100 nfs4_unlock_state();
4104 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4105 * accessed when nfsd is starting.
4108 nfs4_set_recdir(char *recdir)
4110 strcpy(user_recovery_dirname, recdir);
4114 * Change the NFSv4 recovery directory to recdir.
4117 nfs4_reset_recoverydir(char *recdir)
4122 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4126 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4127 nfs4_set_recdir(recdir);
4135 nfs4_recoverydir(void)
4137 return user_recovery_dirname;