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;
54 static stateid_t zerostateid; /* bits all 0 */
55 static stateid_t onestateid; /* bits all 1 */
56 static u64 current_sessionid = 1;
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
73 * Currently used for the del_recall_lru and file hash table. In an
74 * effort to decrease the scope of the client_mutex, this spinlock may
75 * eventually cover more:
77 static DEFINE_SPINLOCK(recall_lock);
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
87 mutex_lock(&client_mutex);
91 nfs4_unlock_state(void)
93 mutex_unlock(&client_mutex);
97 opaque_hashval(const void *ptr, int nbytes)
99 unsigned char *cptr = (unsigned char *) ptr;
109 static struct list_head del_recall_lru;
112 put_nfs4_file(struct nfs4_file *fi)
114 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115 list_del(&fi->fi_hash);
116 spin_unlock(&recall_lock);
118 kmem_cache_free(file_slab, fi);
123 get_nfs4_file(struct nfs4_file *fi)
125 atomic_inc(&fi->fi_ref);
128 static int num_delegations;
129 unsigned int max_delegations;
132 * Open owner state (share locks)
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS 8
137 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
140 #define ownerid_hashval(id) \
141 ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS 8
150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS 10
154 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
157 #define file_hashval(x) \
158 hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id) \
160 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
167 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168 atomic_inc(&fp->fi_access[oflag]);
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
173 if (oflag == O_RDWR) {
174 __nfs4_file_get_access(fp, O_RDONLY);
175 __nfs4_file_get_access(fp, O_WRONLY);
177 __nfs4_file_get_access(fp, oflag);
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
182 if (fp->fi_fds[oflag]) {
183 fput(fp->fi_fds[oflag]);
184 fp->fi_fds[oflag] = NULL;
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
190 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191 nfs4_file_put_fd(fp, O_RDWR);
192 nfs4_file_put_fd(fp, oflag);
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
198 if (oflag == O_RDWR) {
199 __nfs4_file_put_access(fp, O_RDONLY);
200 __nfs4_file_put_access(fp, O_WRONLY);
202 __nfs4_file_put_access(fp, oflag);
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
208 struct nfs4_delegation *dp;
209 struct nfs4_file *fp = stp->st_file;
210 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
212 dprintk("NFSD alloc_init_deleg\n");
214 * Major work on the lease subsystem (for example, to support
215 * calbacks on stat) will be required before we can support
216 * write delegations properly.
218 if (type != NFS4_OPEN_DELEGATE_READ)
220 if (fp->fi_had_conflict)
222 if (num_delegations > max_delegations)
224 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
228 INIT_LIST_HEAD(&dp->dl_perfile);
229 INIT_LIST_HEAD(&dp->dl_perclnt);
230 INIT_LIST_HEAD(&dp->dl_recall_lru);
234 nfs4_file_get_access(fp, O_RDONLY);
237 dp->dl_ident = cb->cb_ident;
238 dp->dl_stateid.si_boot = boot_time;
239 dp->dl_stateid.si_stateownerid = current_delegid++;
240 dp->dl_stateid.si_fileid = 0;
241 dp->dl_stateid.si_generation = 0;
242 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
244 atomic_set(&dp->dl_count, 1);
245 list_add(&dp->dl_perfile, &fp->fi_delegations);
246 list_add(&dp->dl_perclnt, &clp->cl_delegations);
247 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
252 nfs4_put_delegation(struct nfs4_delegation *dp)
254 if (atomic_dec_and_test(&dp->dl_count)) {
255 dprintk("NFSD: freeing dp %p\n",dp);
256 put_nfs4_file(dp->dl_file);
257 kmem_cache_free(deleg_slab, dp);
262 /* Remove the associated file_lock first, then remove the delegation.
263 * lease_modify() is called to remove the FS_LEASE file_lock from
264 * the i_flock list, eventually calling nfsd's lock_manager
265 * fl_release_callback.
268 nfs4_close_delegation(struct nfs4_delegation *dp)
270 struct file *filp = find_readable_file(dp->dl_file);
272 dprintk("NFSD: close_delegation dp %p\n",dp);
274 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
275 nfs4_file_put_access(dp->dl_file, O_RDONLY);
278 /* Called under the state lock. */
280 unhash_delegation(struct nfs4_delegation *dp)
282 list_del_init(&dp->dl_perfile);
283 list_del_init(&dp->dl_perclnt);
284 spin_lock(&recall_lock);
285 list_del_init(&dp->dl_recall_lru);
286 spin_unlock(&recall_lock);
287 nfs4_close_delegation(dp);
288 nfs4_put_delegation(dp);
295 /* client_lock protects the client lru list and session hash table */
296 static DEFINE_SPINLOCK(client_lock);
298 /* Hash tables for nfs4_clientid state */
299 #define CLIENT_HASH_BITS 4
300 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
301 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
303 #define clientid_hashval(id) \
304 ((id) & CLIENT_HASH_MASK)
305 #define clientstr_hashval(name) \
306 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
308 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
309 * used in reboot/reset lease grace period processing
311 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
312 * setclientid_confirmed info.
314 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
317 * client_lru holds client queue ordered by nfs4_client.cl_time
320 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
321 * for last close replay.
323 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
324 static int reclaim_str_hashtbl_size = 0;
325 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
326 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
327 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
328 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
329 static struct list_head client_lru;
330 static struct list_head close_lru;
332 static void unhash_generic_stateid(struct nfs4_stateid *stp)
334 list_del(&stp->st_hash);
335 list_del(&stp->st_perfile);
336 list_del(&stp->st_perstateowner);
339 static void free_generic_stateid(struct nfs4_stateid *stp)
341 put_nfs4_file(stp->st_file);
342 kmem_cache_free(stateid_slab, stp);
345 static void release_lock_stateid(struct nfs4_stateid *stp)
349 unhash_generic_stateid(stp);
350 file = find_any_file(stp->st_file);
352 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
353 free_generic_stateid(stp);
356 static void unhash_lockowner(struct nfs4_stateowner *sop)
358 struct nfs4_stateid *stp;
360 list_del(&sop->so_idhash);
361 list_del(&sop->so_strhash);
362 list_del(&sop->so_perstateid);
363 while (!list_empty(&sop->so_stateids)) {
364 stp = list_first_entry(&sop->so_stateids,
365 struct nfs4_stateid, st_perstateowner);
366 release_lock_stateid(stp);
370 static void release_lockowner(struct nfs4_stateowner *sop)
372 unhash_lockowner(sop);
373 nfs4_put_stateowner(sop);
377 release_stateid_lockowners(struct nfs4_stateid *open_stp)
379 struct nfs4_stateowner *lock_sop;
381 while (!list_empty(&open_stp->st_lockowners)) {
382 lock_sop = list_entry(open_stp->st_lockowners.next,
383 struct nfs4_stateowner, so_perstateid);
384 /* list_del(&open_stp->st_lockowners); */
385 BUG_ON(lock_sop->so_is_open_owner);
386 release_lockowner(lock_sop);
391 * We store the NONE, READ, WRITE, and BOTH bits separately in the
392 * st_{access,deny}_bmap field of the stateid, in order to track not
393 * only what share bits are currently in force, but also what
394 * combinations of share bits previous opens have used. This allows us
395 * to enforce the recommendation of rfc 3530 14.2.19 that the server
396 * return an error if the client attempt to downgrade to a combination
397 * of share bits not explicable by closing some of its previous opens.
399 * XXX: This enforcement is actually incomplete, since we don't keep
400 * track of access/deny bit combinations; so, e.g., we allow:
402 * OPEN allow read, deny write
403 * OPEN allow both, deny none
404 * DOWNGRADE allow read, deny none
406 * which we should reject.
409 set_access(unsigned int *access, unsigned long bmap) {
413 for (i = 1; i < 4; i++) {
414 if (test_bit(i, &bmap))
420 set_deny(unsigned int *deny, unsigned long bmap) {
424 for (i = 0; i < 4; i++) {
425 if (test_bit(i, &bmap))
431 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
432 unsigned int access, deny;
434 set_access(&access, stp->st_access_bmap);
435 set_deny(&deny, stp->st_deny_bmap);
436 if ((access & open->op_share_deny) || (deny & open->op_share_access))
441 static int nfs4_access_to_omode(u32 access)
444 case NFS4_SHARE_ACCESS_READ:
446 case NFS4_SHARE_ACCESS_WRITE:
448 case NFS4_SHARE_ACCESS_BOTH:
454 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
458 set_access(&access, stp->st_access_bmap);
459 return nfs4_access_to_omode(access);
462 static void release_open_stateid(struct nfs4_stateid *stp)
464 int oflag = nfs4_access_bmap_to_omode(stp);
466 unhash_generic_stateid(stp);
467 release_stateid_lockowners(stp);
468 nfs4_file_put_access(stp->st_file, oflag);
469 free_generic_stateid(stp);
472 static void unhash_openowner(struct nfs4_stateowner *sop)
474 struct nfs4_stateid *stp;
476 list_del(&sop->so_idhash);
477 list_del(&sop->so_strhash);
478 list_del(&sop->so_perclient);
479 list_del(&sop->so_perstateid); /* XXX: necessary? */
480 while (!list_empty(&sop->so_stateids)) {
481 stp = list_first_entry(&sop->so_stateids,
482 struct nfs4_stateid, st_perstateowner);
483 release_open_stateid(stp);
487 static void release_openowner(struct nfs4_stateowner *sop)
489 unhash_openowner(sop);
490 list_del(&sop->so_close_lru);
491 nfs4_put_stateowner(sop);
494 #define SESSION_HASH_SIZE 512
495 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
498 hash_sessionid(struct nfs4_sessionid *sessionid)
500 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
502 return sid->sequence % SESSION_HASH_SIZE;
506 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
508 u32 *ptr = (u32 *)(&sessionid->data[0]);
509 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
513 gen_sessionid(struct nfsd4_session *ses)
515 struct nfs4_client *clp = ses->se_client;
516 struct nfsd4_sessionid *sid;
518 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
519 sid->clientid = clp->cl_clientid;
520 sid->sequence = current_sessionid++;
525 * The protocol defines ca_maxresponssize_cached to include the size of
526 * the rpc header, but all we need to cache is the data starting after
527 * the end of the initial SEQUENCE operation--the rest we regenerate
528 * each time. Therefore we can advertise a ca_maxresponssize_cached
529 * value that is the number of bytes in our cache plus a few additional
530 * bytes. In order to stay on the safe side, and not promise more than
531 * we can cache, those additional bytes must be the minimum possible: 24
532 * bytes of rpc header (xid through accept state, with AUTH_NULL
533 * verifier), 12 for the compound header (with zero-length tag), and 44
534 * for the SEQUENCE op response:
536 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
539 * Give the client the number of ca_maxresponsesize_cached slots it
540 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
541 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
542 * than NFSD_MAX_SLOTS_PER_SESSION.
544 * If we run out of reserved DRC memory we should (up to a point)
545 * re-negotiate active sessions and reduce their slot usage to make
546 * rooom for new connections. For now we just fail the create session.
548 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
550 int mem, size = fchan->maxresp_cached;
552 if (fchan->maxreqs < 1)
555 if (size < NFSD_MIN_HDR_SEQ_SZ)
556 size = NFSD_MIN_HDR_SEQ_SZ;
557 size -= NFSD_MIN_HDR_SEQ_SZ;
558 if (size > NFSD_SLOT_CACHE_SIZE)
559 size = NFSD_SLOT_CACHE_SIZE;
561 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
562 mem = fchan->maxreqs * size;
563 if (mem > NFSD_MAX_MEM_PER_SESSION) {
564 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
565 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
566 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
567 mem = fchan->maxreqs * size;
570 spin_lock(&nfsd_drc_lock);
571 /* bound the total session drc memory ussage */
572 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
573 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
574 mem = fchan->maxreqs * size;
576 nfsd_drc_mem_used += mem;
577 spin_unlock(&nfsd_drc_lock);
579 if (fchan->maxreqs == 0)
580 return nfserr_jukebox;
582 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
587 * fchan holds the client values on input, and the server values on output
588 * sv_max_mesg is the maximum payload plus one page for overhead.
590 static int init_forechannel_attrs(struct svc_rqst *rqstp,
591 struct nfsd4_channel_attrs *session_fchan,
592 struct nfsd4_channel_attrs *fchan)
595 __u32 maxcount = nfsd_serv->sv_max_mesg;
597 /* headerpadsz set to zero in encode routine */
599 /* Use the client's max request and max response size if possible */
600 if (fchan->maxreq_sz > maxcount)
601 fchan->maxreq_sz = maxcount;
602 session_fchan->maxreq_sz = fchan->maxreq_sz;
604 if (fchan->maxresp_sz > maxcount)
605 fchan->maxresp_sz = maxcount;
606 session_fchan->maxresp_sz = fchan->maxresp_sz;
608 /* Use the client's maxops if possible */
609 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
610 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
611 session_fchan->maxops = fchan->maxops;
613 /* FIXME: Error means no more DRC pages so the server should
614 * recover pages from existing sessions. For now fail session
617 status = set_forechannel_drc_size(fchan);
619 session_fchan->maxresp_cached = fchan->maxresp_cached;
620 session_fchan->maxreqs = fchan->maxreqs;
622 dprintk("%s status %d\n", __func__, status);
627 free_session_slots(struct nfsd4_session *ses)
631 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
632 kfree(ses->se_slots[i]);
636 * We don't actually need to cache the rpc and session headers, so we
637 * can allocate a little less for each slot:
639 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
641 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
645 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
646 struct nfsd4_create_session *cses)
648 struct nfsd4_session *new, tmp;
649 struct nfsd4_slot *sp;
650 int idx, slotsize, cachesize, i;
653 memset(&tmp, 0, sizeof(tmp));
655 /* FIXME: For now, we just accept the client back channel attributes. */
656 tmp.se_bchannel = cses->back_channel;
657 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
658 &cses->fore_channel);
662 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
663 + sizeof(struct nfsd4_session) > PAGE_SIZE);
665 status = nfserr_jukebox;
666 /* allocate struct nfsd4_session and slot table pointers in one piece */
667 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
668 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
672 memcpy(new, &tmp, sizeof(*new));
674 /* allocate each struct nfsd4_slot and data cache in one piece */
675 cachesize = slot_bytes(&new->se_fchannel);
676 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
677 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
680 new->se_slots[i] = sp;
683 new->se_client = clp;
685 idx = hash_sessionid(&new->se_sessionid);
686 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
687 NFS4_MAX_SESSIONID_LEN);
689 new->se_flags = cses->flags;
690 kref_init(&new->se_ref);
691 spin_lock(&client_lock);
692 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
693 list_add(&new->se_perclnt, &clp->cl_sessions);
694 spin_unlock(&client_lock);
700 free_session_slots(new);
705 /* caller must hold client_lock */
706 static struct nfsd4_session *
707 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
709 struct nfsd4_session *elem;
712 dump_sessionid(__func__, sessionid);
713 idx = hash_sessionid(sessionid);
714 /* Search in the appropriate list */
715 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
716 if (!memcmp(elem->se_sessionid.data, sessionid->data,
717 NFS4_MAX_SESSIONID_LEN)) {
722 dprintk("%s: session not found\n", __func__);
726 /* caller must hold client_lock */
728 unhash_session(struct nfsd4_session *ses)
730 list_del(&ses->se_hash);
731 list_del(&ses->se_perclnt);
735 free_session(struct kref *kref)
737 struct nfsd4_session *ses;
740 ses = container_of(kref, struct nfsd4_session, se_ref);
741 spin_lock(&nfsd_drc_lock);
742 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
743 nfsd_drc_mem_used -= mem;
744 spin_unlock(&nfsd_drc_lock);
745 free_session_slots(ses);
749 /* must be called under the client_lock */
751 renew_client_locked(struct nfs4_client *clp)
753 if (is_client_expired(clp)) {
754 dprintk("%s: client (clientid %08x/%08x) already expired\n",
756 clp->cl_clientid.cl_boot,
757 clp->cl_clientid.cl_id);
762 * Move client to the end to the LRU list.
764 dprintk("renewing client (clientid %08x/%08x)\n",
765 clp->cl_clientid.cl_boot,
766 clp->cl_clientid.cl_id);
767 list_move_tail(&clp->cl_lru, &client_lru);
768 clp->cl_time = get_seconds();
772 renew_client(struct nfs4_client *clp)
774 spin_lock(&client_lock);
775 renew_client_locked(clp);
776 spin_unlock(&client_lock);
779 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
781 STALE_CLIENTID(clientid_t *clid)
783 if (clid->cl_boot == boot_time)
785 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
786 clid->cl_boot, clid->cl_id, boot_time);
791 * XXX Should we use a slab cache ?
792 * This type of memory management is somewhat inefficient, but we use it
793 * anyway since SETCLIENTID is not a common operation.
795 static struct nfs4_client *alloc_client(struct xdr_netobj name)
797 struct nfs4_client *clp;
799 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
802 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
803 if (clp->cl_name.data == NULL) {
807 memcpy(clp->cl_name.data, name.data, name.len);
808 clp->cl_name.len = name.len;
813 free_client(struct nfs4_client *clp)
815 if (clp->cl_cred.cr_group_info)
816 put_group_info(clp->cl_cred.cr_group_info);
817 kfree(clp->cl_principal);
818 kfree(clp->cl_name.data);
823 release_session_client(struct nfsd4_session *session)
825 struct nfs4_client *clp = session->se_client;
827 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
829 if (is_client_expired(clp)) {
831 session->se_client = NULL;
833 renew_client_locked(clp);
834 spin_unlock(&client_lock);
837 /* must be called under the client_lock */
839 unhash_client_locked(struct nfs4_client *clp)
841 mark_client_expired(clp);
842 list_del(&clp->cl_lru);
843 while (!list_empty(&clp->cl_sessions)) {
844 struct nfsd4_session *ses;
845 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
848 nfsd4_put_session(ses);
853 expire_client(struct nfs4_client *clp)
855 struct nfs4_stateowner *sop;
856 struct nfs4_delegation *dp;
857 struct list_head reaplist;
859 INIT_LIST_HEAD(&reaplist);
860 spin_lock(&recall_lock);
861 while (!list_empty(&clp->cl_delegations)) {
862 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
863 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
865 list_del_init(&dp->dl_perclnt);
866 list_move(&dp->dl_recall_lru, &reaplist);
868 spin_unlock(&recall_lock);
869 while (!list_empty(&reaplist)) {
870 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
871 list_del_init(&dp->dl_recall_lru);
872 unhash_delegation(dp);
874 while (!list_empty(&clp->cl_openowners)) {
875 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
876 release_openowner(sop);
878 nfsd4_set_callback_client(clp, NULL);
879 if (clp->cl_cb_conn.cb_xprt)
880 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
881 list_del(&clp->cl_idhash);
882 list_del(&clp->cl_strhash);
883 spin_lock(&client_lock);
884 unhash_client_locked(clp);
885 if (atomic_read(&clp->cl_refcount) == 0)
887 spin_unlock(&client_lock);
890 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
892 memcpy(target->cl_verifier.data, source->data,
893 sizeof(target->cl_verifier.data));
896 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
898 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
899 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
902 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
904 target->cr_uid = source->cr_uid;
905 target->cr_gid = source->cr_gid;
906 target->cr_group_info = source->cr_group_info;
907 get_group_info(target->cr_group_info);
910 static int same_name(const char *n1, const char *n2)
912 return 0 == memcmp(n1, n2, HEXDIR_LEN);
916 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
918 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
922 same_clid(clientid_t *cl1, clientid_t *cl2)
924 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
927 /* XXX what about NGROUP */
929 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
931 return cr1->cr_uid == cr2->cr_uid;
934 static void gen_clid(struct nfs4_client *clp)
936 static u32 current_clientid = 1;
938 clp->cl_clientid.cl_boot = boot_time;
939 clp->cl_clientid.cl_id = current_clientid++;
942 static void gen_confirm(struct nfs4_client *clp)
947 p = (u32 *)clp->cl_confirm.data;
948 *p++ = get_seconds();
952 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
953 struct svc_rqst *rqstp, nfs4_verifier *verf)
955 struct nfs4_client *clp;
956 struct sockaddr *sa = svc_addr(rqstp);
959 clp = alloc_client(name);
963 princ = svc_gss_principal(rqstp);
965 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
966 if (clp->cl_principal == NULL) {
972 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
973 atomic_set(&clp->cl_refcount, 0);
974 atomic_set(&clp->cl_cb_set, 0);
975 INIT_LIST_HEAD(&clp->cl_idhash);
976 INIT_LIST_HEAD(&clp->cl_strhash);
977 INIT_LIST_HEAD(&clp->cl_openowners);
978 INIT_LIST_HEAD(&clp->cl_delegations);
979 INIT_LIST_HEAD(&clp->cl_sessions);
980 INIT_LIST_HEAD(&clp->cl_lru);
981 clp->cl_time = get_seconds();
982 clear_bit(0, &clp->cl_cb_slot_busy);
983 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
984 copy_verf(clp, verf);
985 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
986 clp->cl_flavor = rqstp->rq_flavor;
987 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
993 static int check_name(struct xdr_netobj name)
997 if (name.len > NFS4_OPAQUE_LIMIT) {
998 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1005 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1007 unsigned int idhashval;
1009 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1010 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1011 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1016 move_to_confirmed(struct nfs4_client *clp)
1018 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1019 unsigned int strhashval;
1021 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1022 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1023 strhashval = clientstr_hashval(clp->cl_recdir);
1024 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1028 static struct nfs4_client *
1029 find_confirmed_client(clientid_t *clid)
1031 struct nfs4_client *clp;
1032 unsigned int idhashval = clientid_hashval(clid->cl_id);
1034 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1035 if (same_clid(&clp->cl_clientid, clid))
1041 static struct nfs4_client *
1042 find_unconfirmed_client(clientid_t *clid)
1044 struct nfs4_client *clp;
1045 unsigned int idhashval = clientid_hashval(clid->cl_id);
1047 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1048 if (same_clid(&clp->cl_clientid, clid))
1055 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
1056 * parameter. Matching is based on the fact the at least one of the
1057 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
1059 * FIXME: we need to unify the clientid namespaces for nfsv4.x
1060 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
1061 * and SET_CLIENTID{,_CONFIRM}
1064 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
1066 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
1067 return use_exchange_id == has_exchange_flags;
1070 static struct nfs4_client *
1071 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
1072 bool use_exchange_id)
1074 struct nfs4_client *clp;
1076 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1077 if (same_name(clp->cl_recdir, dname) &&
1078 match_clientid_establishment(clp, use_exchange_id))
1084 static struct nfs4_client *
1085 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
1086 bool use_exchange_id)
1088 struct nfs4_client *clp;
1090 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1091 if (same_name(clp->cl_recdir, dname) &&
1092 match_clientid_establishment(clp, use_exchange_id))
1099 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1101 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
1102 unsigned short expected_family;
1104 /* Currently, we only support tcp and tcp6 for the callback channel */
1105 if (se->se_callback_netid_len == 3 &&
1106 !memcmp(se->se_callback_netid_val, "tcp", 3))
1107 expected_family = AF_INET;
1108 else if (se->se_callback_netid_len == 4 &&
1109 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1110 expected_family = AF_INET6;
1114 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1115 se->se_callback_addr_len,
1116 (struct sockaddr *) &cb->cb_addr,
1117 sizeof(cb->cb_addr));
1119 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
1122 if (cb->cb_addr.ss_family == AF_INET6)
1123 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
1125 cb->cb_minorversion = 0;
1126 cb->cb_prog = se->se_callback_prog;
1127 cb->cb_ident = se->se_callback_ident;
1130 cb->cb_addr.ss_family = AF_UNSPEC;
1132 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1133 "will not receive delegations\n",
1134 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1140 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1143 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1145 struct nfsd4_slot *slot = resp->cstate.slot;
1148 dprintk("--> %s slot %p\n", __func__, slot);
1150 slot->sl_opcnt = resp->opcnt;
1151 slot->sl_status = resp->cstate.status;
1153 if (nfsd4_not_cached(resp)) {
1154 slot->sl_datalen = 0;
1157 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1158 base = (char *)resp->cstate.datap -
1159 (char *)resp->xbuf->head[0].iov_base;
1160 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1162 WARN("%s: sessions DRC could not cache compound\n", __func__);
1167 * Encode the replay sequence operation from the slot values.
1168 * If cachethis is FALSE encode the uncached rep error on the next
1169 * operation which sets resp->p and increments resp->opcnt for
1170 * nfs4svc_encode_compoundres.
1174 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1175 struct nfsd4_compoundres *resp)
1177 struct nfsd4_op *op;
1178 struct nfsd4_slot *slot = resp->cstate.slot;
1180 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1181 resp->opcnt, resp->cstate.slot->sl_cachethis);
1183 /* Encode the replayed sequence operation */
1184 op = &args->ops[resp->opcnt - 1];
1185 nfsd4_encode_operation(resp, op);
1187 /* Return nfserr_retry_uncached_rep in next operation. */
1188 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1189 op = &args->ops[resp->opcnt++];
1190 op->status = nfserr_retry_uncached_rep;
1191 nfsd4_encode_operation(resp, op);
1197 * The sequence operation is not cached because we can use the slot and
1201 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1202 struct nfsd4_sequence *seq)
1204 struct nfsd4_slot *slot = resp->cstate.slot;
1207 dprintk("--> %s slot %p\n", __func__, slot);
1209 /* Either returns 0 or nfserr_retry_uncached */
1210 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1211 if (status == nfserr_retry_uncached_rep)
1214 /* The sequence operation has been encoded, cstate->datap set. */
1215 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1217 resp->opcnt = slot->sl_opcnt;
1218 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1219 status = slot->sl_status;
1225 * Set the exchange_id flags returned by the server.
1228 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1230 /* pNFS is not supported */
1231 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1233 /* Referrals are supported, Migration is not. */
1234 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1236 /* set the wire flags to return to client. */
1237 clid->flags = new->cl_exchange_flags;
1241 nfsd4_exchange_id(struct svc_rqst *rqstp,
1242 struct nfsd4_compound_state *cstate,
1243 struct nfsd4_exchange_id *exid)
1245 struct nfs4_client *unconf, *conf, *new;
1247 unsigned int strhashval;
1248 char dname[HEXDIR_LEN];
1249 char addr_str[INET6_ADDRSTRLEN];
1250 nfs4_verifier verf = exid->verifier;
1251 struct sockaddr *sa = svc_addr(rqstp);
1253 rpc_ntop(sa, addr_str, sizeof(addr_str));
1254 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1255 "ip_addr=%s flags %x, spa_how %d\n",
1256 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1257 addr_str, exid->flags, exid->spa_how);
1259 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1260 return nfserr_inval;
1262 /* Currently only support SP4_NONE */
1263 switch (exid->spa_how) {
1267 return nfserr_encr_alg_unsupp;
1269 BUG(); /* checked by xdr code */
1271 return nfserr_serverfault; /* no excuse :-/ */
1274 status = nfs4_make_rec_clidname(dname, &exid->clname);
1279 strhashval = clientstr_hashval(dname);
1284 conf = find_confirmed_client_by_str(dname, strhashval, true);
1286 if (!same_verf(&verf, &conf->cl_verifier)) {
1287 /* 18.35.4 case 8 */
1288 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1289 status = nfserr_not_same;
1292 /* Client reboot: destroy old state */
1293 expire_client(conf);
1296 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1297 /* 18.35.4 case 9 */
1298 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1299 status = nfserr_perm;
1302 expire_client(conf);
1306 * Set bit when the owner id and verifier map to an already
1307 * confirmed client id (18.35.3).
1309 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1312 * Falling into 18.35.4 case 2, possible router replay.
1313 * Leave confirmed record intact and return same result.
1315 copy_verf(conf, &verf);
1320 /* 18.35.4 case 7 */
1321 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1322 status = nfserr_noent;
1326 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1329 * Possible retry or client restart. Per 18.35.4 case 4,
1330 * a new unconfirmed record should be generated regardless
1331 * of whether any properties have changed.
1333 expire_client(unconf);
1338 new = create_client(exid->clname, dname, rqstp, &verf);
1340 status = nfserr_jukebox;
1345 add_to_unconfirmed(new, strhashval);
1347 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1348 exid->clientid.cl_id = new->cl_clientid.cl_id;
1351 nfsd4_set_ex_flags(new, exid);
1353 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1354 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1358 nfs4_unlock_state();
1360 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1365 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1367 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1370 /* The slot is in use, and no response has been sent. */
1372 if (seqid == slot_seqid)
1373 return nfserr_jukebox;
1375 return nfserr_seq_misordered;
1378 if (likely(seqid == slot_seqid + 1))
1381 if (seqid == slot_seqid)
1382 return nfserr_replay_cache;
1384 if (seqid == 1 && (slot_seqid + 1) == 0)
1386 /* Misordered replay or misordered new request */
1387 return nfserr_seq_misordered;
1391 * Cache the create session result into the create session single DRC
1392 * slot cache by saving the xdr structure. sl_seqid has been set.
1393 * Do this for solo or embedded create session operations.
1396 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1397 struct nfsd4_clid_slot *slot, int nfserr)
1399 slot->sl_status = nfserr;
1400 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1404 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1405 struct nfsd4_clid_slot *slot)
1407 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1408 return slot->sl_status;
1412 nfsd4_create_session(struct svc_rqst *rqstp,
1413 struct nfsd4_compound_state *cstate,
1414 struct nfsd4_create_session *cr_ses)
1416 struct sockaddr *sa = svc_addr(rqstp);
1417 struct nfs4_client *conf, *unconf;
1418 struct nfsd4_clid_slot *cs_slot = NULL;
1422 unconf = find_unconfirmed_client(&cr_ses->clientid);
1423 conf = find_confirmed_client(&cr_ses->clientid);
1426 cs_slot = &conf->cl_cs_slot;
1427 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1428 if (status == nfserr_replay_cache) {
1429 dprintk("Got a create_session replay! seqid= %d\n",
1431 /* Return the cached reply status */
1432 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1434 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1435 status = nfserr_seq_misordered;
1436 dprintk("Sequence misordered!\n");
1437 dprintk("Expected seqid= %d but got seqid= %d\n",
1438 cs_slot->sl_seqid, cr_ses->seqid);
1441 cs_slot->sl_seqid++;
1442 } else if (unconf) {
1443 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1444 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1445 status = nfserr_clid_inuse;
1449 cs_slot = &unconf->cl_cs_slot;
1450 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1452 /* an unconfirmed replay returns misordered */
1453 status = nfserr_seq_misordered;
1457 cs_slot->sl_seqid++; /* from 0 to 1 */
1458 move_to_confirmed(unconf);
1460 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1461 unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1462 svc_xprt_get(rqstp->rq_xprt);
1464 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1466 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1467 unconf->cl_cb_conn.cb_minorversion =
1468 cstate->minorversion;
1469 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1470 unconf->cl_cb_seq_nr = 1;
1471 nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1475 status = nfserr_stale_clientid;
1480 * We do not support RDMA or persistent sessions
1482 cr_ses->flags &= ~SESSION4_PERSIST;
1483 cr_ses->flags &= ~SESSION4_RDMA;
1485 status = alloc_init_session(rqstp, conf, cr_ses);
1489 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1490 NFS4_MAX_SESSIONID_LEN);
1491 cr_ses->seqid = cs_slot->sl_seqid;
1494 /* cache solo and embedded create sessions under the state lock */
1495 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1497 nfs4_unlock_state();
1498 dprintk("%s returns %d\n", __func__, ntohl(status));
1502 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1504 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1505 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1507 return argp->opcnt == resp->opcnt;
1510 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1514 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1518 nfsd4_destroy_session(struct svc_rqst *r,
1519 struct nfsd4_compound_state *cstate,
1520 struct nfsd4_destroy_session *sessionid)
1522 struct nfsd4_session *ses;
1523 u32 status = nfserr_badsession;
1526 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1527 * - Should we return nfserr_back_chan_busy if waiting for
1528 * callbacks on to-be-destroyed session?
1529 * - Do we need to clear any callback info from previous session?
1532 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1533 if (!nfsd4_last_compound_op(r))
1534 return nfserr_not_only_op;
1536 dump_sessionid(__func__, &sessionid->sessionid);
1537 spin_lock(&client_lock);
1538 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1540 spin_unlock(&client_lock);
1544 unhash_session(ses);
1545 spin_unlock(&client_lock);
1548 /* wait for callbacks */
1549 nfsd4_set_callback_client(ses->se_client, NULL);
1550 nfs4_unlock_state();
1551 nfsd4_put_session(ses);
1554 dprintk("%s returns %d\n", __func__, ntohl(status));
1559 nfsd4_sequence(struct svc_rqst *rqstp,
1560 struct nfsd4_compound_state *cstate,
1561 struct nfsd4_sequence *seq)
1563 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1564 struct nfsd4_session *session;
1565 struct nfsd4_slot *slot;
1568 if (resp->opcnt != 1)
1569 return nfserr_sequence_pos;
1571 spin_lock(&client_lock);
1572 status = nfserr_badsession;
1573 session = find_in_sessionid_hashtbl(&seq->sessionid);
1577 status = nfserr_badslot;
1578 if (seq->slotid >= session->se_fchannel.maxreqs)
1581 slot = session->se_slots[seq->slotid];
1582 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1584 /* We do not negotiate the number of slots yet, so set the
1585 * maxslots to the session maxreqs which is used to encode
1586 * sr_highest_slotid and the sr_target_slot id to maxslots */
1587 seq->maxslots = session->se_fchannel.maxreqs;
1589 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1590 if (status == nfserr_replay_cache) {
1591 cstate->slot = slot;
1592 cstate->session = session;
1593 /* Return the cached reply status and set cstate->status
1594 * for nfsd4_proc_compound processing */
1595 status = nfsd4_replay_cache_entry(resp, seq);
1596 cstate->status = nfserr_replay_cache;
1602 /* Success! bump slot seqid */
1603 slot->sl_inuse = true;
1604 slot->sl_seqid = seq->seqid;
1605 slot->sl_cachethis = seq->cachethis;
1607 cstate->slot = slot;
1608 cstate->session = session;
1611 /* Hold a session reference until done processing the compound. */
1612 if (cstate->session) {
1613 nfsd4_get_session(cstate->session);
1614 atomic_inc(&session->se_client->cl_refcount);
1616 spin_unlock(&client_lock);
1617 dprintk("%s: return %d\n", __func__, ntohl(status));
1622 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1624 if (rc->rca_one_fs) {
1625 if (!cstate->current_fh.fh_dentry)
1626 return nfserr_nofilehandle;
1628 * We don't take advantage of the rca_one_fs case.
1629 * That's OK, it's optional, we can safely ignore it.
1634 if (is_client_expired(cstate->session->se_client)) {
1635 nfs4_unlock_state();
1637 * The following error isn't really legal.
1638 * But we only get here if the client just explicitly
1639 * destroyed the client. Surely it no longer cares what
1640 * error it gets back on an operation for the dead
1643 return nfserr_stale_clientid;
1645 nfsd4_create_clid_dir(cstate->session->se_client);
1646 nfs4_unlock_state();
1651 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1652 struct nfsd4_setclientid *setclid)
1654 struct sockaddr *sa = svc_addr(rqstp);
1655 struct xdr_netobj clname = {
1656 .len = setclid->se_namelen,
1657 .data = setclid->se_name,
1659 nfs4_verifier clverifier = setclid->se_verf;
1660 unsigned int strhashval;
1661 struct nfs4_client *conf, *unconf, *new;
1663 char dname[HEXDIR_LEN];
1665 if (!check_name(clname))
1666 return nfserr_inval;
1668 status = nfs4_make_rec_clidname(dname, &clname);
1673 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1674 * We get here on a DRC miss.
1677 strhashval = clientstr_hashval(dname);
1680 conf = find_confirmed_client_by_str(dname, strhashval, false);
1682 /* RFC 3530 14.2.33 CASE 0: */
1683 status = nfserr_clid_inuse;
1684 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1685 char addr_str[INET6_ADDRSTRLEN];
1686 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1688 dprintk("NFSD: setclientid: string in use by client "
1689 "at %s\n", addr_str);
1694 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1695 * has a description of SETCLIENTID request processing consisting
1696 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1698 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1699 status = nfserr_resource;
1702 * RFC 3530 14.2.33 CASE 4:
1703 * placed first, because it is the normal case
1706 expire_client(unconf);
1707 new = create_client(clname, dname, rqstp, &clverifier);
1711 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1713 * RFC 3530 14.2.33 CASE 1:
1714 * probable callback update
1717 /* Note this is removing unconfirmed {*x***},
1718 * which is stronger than RFC recommended {vxc**}.
1719 * This has the advantage that there is at most
1720 * one {*x***} in either list at any time.
1722 expire_client(unconf);
1724 new = create_client(clname, dname, rqstp, &clverifier);
1727 copy_clid(new, conf);
1728 } else if (!unconf) {
1730 * RFC 3530 14.2.33 CASE 2:
1731 * probable client reboot; state will be removed if
1734 new = create_client(clname, dname, rqstp, &clverifier);
1740 * RFC 3530 14.2.33 CASE 3:
1741 * probable client reboot; state will be removed if
1744 expire_client(unconf);
1745 new = create_client(clname, dname, rqstp, &clverifier);
1750 gen_callback(new, setclid, rpc_get_scope_id(sa));
1751 add_to_unconfirmed(new, strhashval);
1752 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1753 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1754 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1757 nfs4_unlock_state();
1763 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1764 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1765 * bullets, labeled as CASE1 - CASE4 below.
1768 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1769 struct nfsd4_compound_state *cstate,
1770 struct nfsd4_setclientid_confirm *setclientid_confirm)
1772 struct sockaddr *sa = svc_addr(rqstp);
1773 struct nfs4_client *conf, *unconf;
1774 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1775 clientid_t * clid = &setclientid_confirm->sc_clientid;
1778 if (STALE_CLIENTID(clid))
1779 return nfserr_stale_clientid;
1781 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1782 * We get here on a DRC miss.
1787 conf = find_confirmed_client(clid);
1788 unconf = find_unconfirmed_client(clid);
1790 status = nfserr_clid_inuse;
1791 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1793 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1797 * section 14.2.34 of RFC 3530 has a description of
1798 * SETCLIENTID_CONFIRM request processing consisting
1799 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1801 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1803 * RFC 3530 14.2.34 CASE 1:
1806 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1807 status = nfserr_clid_inuse;
1809 atomic_set(&conf->cl_cb_set, 0);
1810 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1811 expire_client(unconf);
1815 } else if (conf && !unconf) {
1817 * RFC 3530 14.2.34 CASE 2:
1818 * probable retransmitted request; play it safe and
1821 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1822 status = nfserr_clid_inuse;
1825 } else if (!conf && unconf
1826 && same_verf(&unconf->cl_confirm, &confirm)) {
1828 * RFC 3530 14.2.34 CASE 3:
1829 * Normal case; new or rebooted client:
1831 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1832 status = nfserr_clid_inuse;
1835 clientstr_hashval(unconf->cl_recdir);
1836 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1839 nfsd4_remove_clid_dir(conf);
1840 expire_client(conf);
1842 move_to_confirmed(unconf);
1844 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1847 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1848 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1851 * RFC 3530 14.2.34 CASE 4:
1852 * Client probably hasn't noticed that we rebooted yet.
1854 status = nfserr_stale_clientid;
1856 /* check that we have hit one of the cases...*/
1857 status = nfserr_clid_inuse;
1860 nfs4_unlock_state();
1864 /* OPEN Share state helper functions */
1865 static inline struct nfs4_file *
1866 alloc_init_file(struct inode *ino)
1868 struct nfs4_file *fp;
1869 unsigned int hashval = file_hashval(ino);
1871 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1873 atomic_set(&fp->fi_ref, 1);
1874 INIT_LIST_HEAD(&fp->fi_hash);
1875 INIT_LIST_HEAD(&fp->fi_stateids);
1876 INIT_LIST_HEAD(&fp->fi_delegations);
1877 fp->fi_inode = igrab(ino);
1878 fp->fi_id = current_fileid++;
1879 fp->fi_had_conflict = false;
1880 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
1881 memset(fp->fi_access, 0, sizeof(fp->fi_access));
1882 spin_lock(&recall_lock);
1883 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1884 spin_unlock(&recall_lock);
1891 nfsd4_free_slab(struct kmem_cache **slab)
1895 kmem_cache_destroy(*slab);
1900 nfsd4_free_slabs(void)
1902 nfsd4_free_slab(&stateowner_slab);
1903 nfsd4_free_slab(&file_slab);
1904 nfsd4_free_slab(&stateid_slab);
1905 nfsd4_free_slab(&deleg_slab);
1909 nfsd4_init_slabs(void)
1911 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1912 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1913 if (stateowner_slab == NULL)
1915 file_slab = kmem_cache_create("nfsd4_files",
1916 sizeof(struct nfs4_file), 0, 0, NULL);
1917 if (file_slab == NULL)
1919 stateid_slab = kmem_cache_create("nfsd4_stateids",
1920 sizeof(struct nfs4_stateid), 0, 0, NULL);
1921 if (stateid_slab == NULL)
1923 deleg_slab = kmem_cache_create("nfsd4_delegations",
1924 sizeof(struct nfs4_delegation), 0, 0, NULL);
1925 if (deleg_slab == NULL)
1930 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1935 nfs4_free_stateowner(struct kref *kref)
1937 struct nfs4_stateowner *sop =
1938 container_of(kref, struct nfs4_stateowner, so_ref);
1939 kfree(sop->so_owner.data);
1940 kmem_cache_free(stateowner_slab, sop);
1943 static inline struct nfs4_stateowner *
1944 alloc_stateowner(struct xdr_netobj *owner)
1946 struct nfs4_stateowner *sop;
1948 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1949 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1950 memcpy(sop->so_owner.data, owner->data, owner->len);
1951 sop->so_owner.len = owner->len;
1952 kref_init(&sop->so_ref);
1955 kmem_cache_free(stateowner_slab, sop);
1960 static struct nfs4_stateowner *
1961 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1962 struct nfs4_stateowner *sop;
1963 struct nfs4_replay *rp;
1964 unsigned int idhashval;
1966 if (!(sop = alloc_stateowner(&open->op_owner)))
1968 idhashval = ownerid_hashval(current_ownerid);
1969 INIT_LIST_HEAD(&sop->so_idhash);
1970 INIT_LIST_HEAD(&sop->so_strhash);
1971 INIT_LIST_HEAD(&sop->so_perclient);
1972 INIT_LIST_HEAD(&sop->so_stateids);
1973 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1974 INIT_LIST_HEAD(&sop->so_close_lru);
1976 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1977 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1978 list_add(&sop->so_perclient, &clp->cl_openowners);
1979 sop->so_is_open_owner = 1;
1980 sop->so_id = current_ownerid++;
1981 sop->so_client = clp;
1982 sop->so_seqid = open->op_seqid;
1983 sop->so_confirmed = 0;
1984 rp = &sop->so_replay;
1985 rp->rp_status = nfserr_serverfault;
1987 rp->rp_buf = rp->rp_ibuf;
1992 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1993 struct nfs4_stateowner *sop = open->op_stateowner;
1994 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1996 INIT_LIST_HEAD(&stp->st_hash);
1997 INIT_LIST_HEAD(&stp->st_perstateowner);
1998 INIT_LIST_HEAD(&stp->st_lockowners);
1999 INIT_LIST_HEAD(&stp->st_perfile);
2000 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2001 list_add(&stp->st_perstateowner, &sop->so_stateids);
2002 list_add(&stp->st_perfile, &fp->fi_stateids);
2003 stp->st_stateowner = sop;
2006 stp->st_stateid.si_boot = boot_time;
2007 stp->st_stateid.si_stateownerid = sop->so_id;
2008 stp->st_stateid.si_fileid = fp->fi_id;
2009 stp->st_stateid.si_generation = 0;
2010 stp->st_access_bmap = 0;
2011 stp->st_deny_bmap = 0;
2012 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2013 &stp->st_access_bmap);
2014 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2015 stp->st_openstp = NULL;
2019 move_to_close_lru(struct nfs4_stateowner *sop)
2021 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2023 list_move_tail(&sop->so_close_lru, &close_lru);
2024 sop->so_time = get_seconds();
2028 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2031 return (sop->so_owner.len == owner->len) &&
2032 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2033 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2036 static struct nfs4_stateowner *
2037 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2039 struct nfs4_stateowner *so = NULL;
2041 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2042 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2048 /* search file_hashtbl[] for file */
2049 static struct nfs4_file *
2050 find_file(struct inode *ino)
2052 unsigned int hashval = file_hashval(ino);
2053 struct nfs4_file *fp;
2055 spin_lock(&recall_lock);
2056 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2057 if (fp->fi_inode == ino) {
2059 spin_unlock(&recall_lock);
2063 spin_unlock(&recall_lock);
2067 static inline int access_valid(u32 x, u32 minorversion)
2069 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2071 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2073 x &= ~NFS4_SHARE_ACCESS_MASK;
2074 if (minorversion && x) {
2075 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2077 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2079 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2086 static inline int deny_valid(u32 x)
2088 /* Note: unlike access bits, deny bits may be zero. */
2089 return x <= NFS4_SHARE_DENY_BOTH;
2093 * Called to check deny when READ with all zero stateid or
2094 * WRITE with all zero or all one stateid
2097 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2099 struct inode *ino = current_fh->fh_dentry->d_inode;
2100 struct nfs4_file *fp;
2101 struct nfs4_stateid *stp;
2104 dprintk("NFSD: nfs4_share_conflict\n");
2106 fp = find_file(ino);
2109 ret = nfserr_locked;
2110 /* Search for conflicting share reservations */
2111 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2112 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2113 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2123 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2125 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2126 nfs4_file_put_access(fp, O_WRONLY);
2127 if (share_access & NFS4_SHARE_ACCESS_READ)
2128 nfs4_file_put_access(fp, O_RDONLY);
2132 * Spawn a thread to perform a recall on the delegation represented
2133 * by the lease (file_lock)
2135 * Called from break_lease() with lock_kernel() held.
2136 * Note: we assume break_lease will only call this *once* for any given
2140 void nfsd_break_deleg_cb(struct file_lock *fl)
2142 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2144 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2148 /* We're assuming the state code never drops its reference
2149 * without first removing the lease. Since we're in this lease
2150 * callback (and since the lease code is serialized by the kernel
2151 * lock) we know the server hasn't removed the lease yet, we know
2152 * it's safe to take a reference: */
2153 atomic_inc(&dp->dl_count);
2155 spin_lock(&recall_lock);
2156 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2157 spin_unlock(&recall_lock);
2159 /* only place dl_time is set. protected by lock_kernel*/
2160 dp->dl_time = get_seconds();
2163 * We don't want the locks code to timeout the lease for us;
2164 * we'll remove it ourself if the delegation isn't returned
2167 fl->fl_break_time = 0;
2169 dp->dl_file->fi_had_conflict = true;
2170 nfsd4_cb_recall(dp);
2174 * The file_lock is being reapd.
2176 * Called by locks_free_lock() with lock_kernel() held.
2179 void nfsd_release_deleg_cb(struct file_lock *fl)
2181 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2183 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2185 if (!(fl->fl_flags & FL_LEASE) || !dp)
2187 dp->dl_flock = NULL;
2191 * Set the delegation file_lock back pointer.
2193 * Called from setlease() with lock_kernel() held.
2196 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2198 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2200 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2207 * Called from setlease() with lock_kernel() held
2210 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2212 struct nfs4_delegation *onlistd =
2213 (struct nfs4_delegation *)onlist->fl_owner;
2214 struct nfs4_delegation *tryd =
2215 (struct nfs4_delegation *)try->fl_owner;
2217 if (onlist->fl_lmops != try->fl_lmops)
2220 return onlistd->dl_client == tryd->dl_client;
2225 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2228 return lease_modify(onlist, arg);
2233 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2234 .fl_break = nfsd_break_deleg_cb,
2235 .fl_release_private = nfsd_release_deleg_cb,
2236 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2237 .fl_mylease = nfsd_same_client_deleg_cb,
2238 .fl_change = nfsd_change_deleg_cb,
2243 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2244 struct nfsd4_open *open)
2246 clientid_t *clientid = &open->op_clientid;
2247 struct nfs4_client *clp = NULL;
2248 unsigned int strhashval;
2249 struct nfs4_stateowner *sop = NULL;
2251 if (!check_name(open->op_owner))
2252 return nfserr_inval;
2254 if (STALE_CLIENTID(&open->op_clientid))
2255 return nfserr_stale_clientid;
2257 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2258 sop = find_openstateowner_str(strhashval, open);
2259 open->op_stateowner = sop;
2261 /* Make sure the client's lease hasn't expired. */
2262 clp = find_confirmed_client(clientid);
2264 return nfserr_expired;
2267 /* When sessions are used, skip open sequenceid processing */
2268 if (nfsd4_has_session(cstate))
2270 if (!sop->so_confirmed) {
2271 /* Replace unconfirmed owners without checking for replay. */
2272 clp = sop->so_client;
2273 release_openowner(sop);
2274 open->op_stateowner = NULL;
2277 if (open->op_seqid == sop->so_seqid - 1) {
2278 if (sop->so_replay.rp_buflen)
2279 return nfserr_replay_me;
2280 /* The original OPEN failed so spectacularly
2281 * that we don't even have replay data saved!
2282 * Therefore, we have no choice but to continue
2283 * processing this OPEN; presumably, we'll
2284 * fail again for the same reason.
2286 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2289 if (open->op_seqid != sop->so_seqid)
2290 return nfserr_bad_seqid;
2292 if (open->op_stateowner == NULL) {
2293 sop = alloc_init_open_stateowner(strhashval, clp, open);
2295 return nfserr_resource;
2296 open->op_stateowner = sop;
2298 list_del_init(&sop->so_close_lru);
2299 renew_client(sop->so_client);
2303 static inline __be32
2304 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2306 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2307 return nfserr_openmode;
2312 static struct nfs4_delegation *
2313 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2315 struct nfs4_delegation *dp;
2317 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2318 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2324 int share_access_to_flags(u32 share_access)
2326 share_access &= ~NFS4_SHARE_WANT_MASK;
2328 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2332 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2333 struct nfs4_delegation **dp)
2336 __be32 status = nfserr_bad_stateid;
2338 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2341 flags = share_access_to_flags(open->op_share_access);
2342 status = nfs4_check_delegmode(*dp, flags);
2346 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2350 open->op_stateowner->so_confirmed = 1;
2355 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2357 struct nfs4_stateid *local;
2358 __be32 status = nfserr_share_denied;
2359 struct nfs4_stateowner *sop = open->op_stateowner;
2361 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2362 /* ignore lock owners */
2363 if (local->st_stateowner->so_is_open_owner == 0)
2365 /* remember if we have seen this open owner */
2366 if (local->st_stateowner == sop)
2368 /* check for conflicting share reservations */
2369 if (!test_share(local, open))
2377 static inline struct nfs4_stateid *
2378 nfs4_alloc_stateid(void)
2380 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2383 static inline int nfs4_access_to_access(u32 nfs4_access)
2387 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2388 flags |= NFSD_MAY_READ;
2389 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2390 flags |= NFSD_MAY_WRITE;
2394 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2395 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2398 int oflag = nfs4_access_to_omode(nfs4_access);
2399 int access = nfs4_access_to_access(nfs4_access);
2401 if (!fp->fi_fds[oflag]) {
2402 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2403 &fp->fi_fds[oflag]);
2404 if (status == nfserr_dropit)
2405 status = nfserr_jukebox;
2409 nfs4_file_get_access(fp, oflag);
2415 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2416 struct nfs4_file *fp, struct svc_fh *cur_fh,
2417 struct nfsd4_open *open)
2419 struct nfs4_stateid *stp;
2422 stp = nfs4_alloc_stateid();
2424 return nfserr_resource;
2426 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2428 kmem_cache_free(stateid_slab, stp);
2435 static inline __be32
2436 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2437 struct nfsd4_open *open)
2439 struct iattr iattr = {
2440 .ia_valid = ATTR_SIZE,
2443 if (!open->op_truncate)
2445 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2446 return nfserr_inval;
2447 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2451 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2453 u32 op_share_access, new_access;
2456 set_access(&new_access, stp->st_access_bmap);
2457 new_access = (~new_access) & open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2460 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, new_access);
2464 status = nfsd4_truncate(rqstp, cur_fh, open);
2467 int oflag = nfs4_access_to_omode(new_access);
2468 nfs4_file_put_access(fp, oflag);
2472 /* remember the open */
2473 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2474 __set_bit(op_share_access, &stp->st_access_bmap);
2475 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2482 nfs4_set_claim_prev(struct nfsd4_open *open)
2484 open->op_stateowner->so_confirmed = 1;
2485 open->op_stateowner->so_client->cl_firststate = 1;
2489 * Attempt to hand out a delegation.
2492 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2494 struct nfs4_delegation *dp;
2495 struct nfs4_stateowner *sop = stp->st_stateowner;
2496 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2497 struct file_lock fl, *flp = &fl;
2498 int status, flag = 0;
2500 flag = NFS4_OPEN_DELEGATE_NONE;
2501 open->op_recall = 0;
2502 switch (open->op_claim_type) {
2503 case NFS4_OPEN_CLAIM_PREVIOUS:
2505 open->op_recall = 1;
2506 flag = open->op_delegate_type;
2507 if (flag == NFS4_OPEN_DELEGATE_NONE)
2510 case NFS4_OPEN_CLAIM_NULL:
2511 /* Let's not give out any delegations till everyone's
2512 * had the chance to reclaim theirs.... */
2513 if (locks_in_grace())
2515 if (!cb_up || !sop->so_confirmed)
2517 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2518 flag = NFS4_OPEN_DELEGATE_WRITE;
2520 flag = NFS4_OPEN_DELEGATE_READ;
2526 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2528 flag = NFS4_OPEN_DELEGATE_NONE;
2531 locks_init_lock(&fl);
2532 fl.fl_lmops = &nfsd_lease_mng_ops;
2533 fl.fl_flags = FL_LEASE;
2534 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2535 fl.fl_end = OFFSET_MAX;
2536 fl.fl_owner = (fl_owner_t)dp;
2537 fl.fl_file = find_readable_file(stp->st_file);
2538 BUG_ON(!fl.fl_file);
2539 fl.fl_pid = current->tgid;
2541 /* vfs_setlease checks to see if delegation should be handed out.
2542 * the lock_manager callbacks fl_mylease and fl_change are used
2544 if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
2545 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2546 unhash_delegation(dp);
2547 flag = NFS4_OPEN_DELEGATE_NONE;
2551 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2553 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2554 STATEID_VAL(&dp->dl_stateid));
2556 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2557 && flag == NFS4_OPEN_DELEGATE_NONE
2558 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2559 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2560 open->op_delegate_type = flag;
2564 * called with nfs4_lock_state() held.
2567 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2569 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2570 struct nfs4_file *fp = NULL;
2571 struct inode *ino = current_fh->fh_dentry->d_inode;
2572 struct nfs4_stateid *stp = NULL;
2573 struct nfs4_delegation *dp = NULL;
2576 status = nfserr_inval;
2577 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2578 || !deny_valid(open->op_share_deny))
2581 * Lookup file; if found, lookup stateid and check open request,
2582 * and check for delegations in the process of being recalled.
2583 * If not found, create the nfs4_file struct
2585 fp = find_file(ino);
2587 if ((status = nfs4_check_open(fp, open, &stp)))
2589 status = nfs4_check_deleg(fp, open, &dp);
2593 status = nfserr_bad_stateid;
2594 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2596 status = nfserr_resource;
2597 fp = alloc_init_file(ino);
2603 * OPEN the file, or upgrade an existing OPEN.
2604 * If truncate fails, the OPEN fails.
2607 /* Stateid was found, this is an OPEN upgrade */
2608 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2611 update_stateid(&stp->st_stateid);
2613 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2616 init_stateid(stp, fp, open);
2617 status = nfsd4_truncate(rqstp, current_fh, open);
2619 release_open_stateid(stp);
2622 if (nfsd4_has_session(&resp->cstate))
2623 update_stateid(&stp->st_stateid);
2625 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2627 if (nfsd4_has_session(&resp->cstate))
2628 open->op_stateowner->so_confirmed = 1;
2631 * Attempt to hand out a delegation. No error return, because the
2632 * OPEN succeeds even if we fail.
2634 nfs4_open_delegation(current_fh, open, stp);
2638 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2639 STATEID_VAL(&stp->st_stateid));
2643 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2644 nfs4_set_claim_prev(open);
2646 * To finish the open response, we just need to set the rflags.
2648 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2649 if (!open->op_stateowner->so_confirmed &&
2650 !nfsd4_has_session(&resp->cstate))
2651 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2657 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2660 struct nfs4_client *clp;
2664 dprintk("process_renew(%08x/%08x): starting\n",
2665 clid->cl_boot, clid->cl_id);
2666 status = nfserr_stale_clientid;
2667 if (STALE_CLIENTID(clid))
2669 clp = find_confirmed_client(clid);
2670 status = nfserr_expired;
2672 /* We assume the client took too long to RENEW. */
2673 dprintk("nfsd4_renew: clientid not found!\n");
2677 status = nfserr_cb_path_down;
2678 if (!list_empty(&clp->cl_delegations)
2679 && !atomic_read(&clp->cl_cb_set))
2683 nfs4_unlock_state();
2687 struct lock_manager nfsd4_manager = {
2691 nfsd4_end_grace(void)
2693 dprintk("NFSD: end of grace period\n");
2694 nfsd4_recdir_purge_old();
2695 locks_end_grace(&nfsd4_manager);
2697 * Now that every NFSv4 client has had the chance to recover and
2698 * to see the (possibly new, possibly shorter) lease time, we
2699 * can safely set the next grace time to the current lease time:
2701 nfsd4_grace = nfsd4_lease;
2705 nfs4_laundromat(void)
2707 struct nfs4_client *clp;
2708 struct nfs4_stateowner *sop;
2709 struct nfs4_delegation *dp;
2710 struct list_head *pos, *next, reaplist;
2711 time_t cutoff = get_seconds() - nfsd4_lease;
2712 time_t t, clientid_val = nfsd4_lease;
2713 time_t u, test_val = nfsd4_lease;
2717 dprintk("NFSD: laundromat service - starting\n");
2718 if (locks_in_grace())
2720 INIT_LIST_HEAD(&reaplist);
2721 spin_lock(&client_lock);
2722 list_for_each_safe(pos, next, &client_lru) {
2723 clp = list_entry(pos, struct nfs4_client, cl_lru);
2724 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2725 t = clp->cl_time - cutoff;
2726 if (clientid_val > t)
2730 if (atomic_read(&clp->cl_refcount)) {
2731 dprintk("NFSD: client in use (clientid %08x)\n",
2732 clp->cl_clientid.cl_id);
2735 unhash_client_locked(clp);
2736 list_add(&clp->cl_lru, &reaplist);
2738 spin_unlock(&client_lock);
2739 list_for_each_safe(pos, next, &reaplist) {
2740 clp = list_entry(pos, struct nfs4_client, cl_lru);
2741 dprintk("NFSD: purging unused client (clientid %08x)\n",
2742 clp->cl_clientid.cl_id);
2743 nfsd4_remove_clid_dir(clp);
2746 spin_lock(&recall_lock);
2747 list_for_each_safe(pos, next, &del_recall_lru) {
2748 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2749 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2750 u = dp->dl_time - cutoff;
2755 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2757 list_move(&dp->dl_recall_lru, &reaplist);
2759 spin_unlock(&recall_lock);
2760 list_for_each_safe(pos, next, &reaplist) {
2761 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2762 list_del_init(&dp->dl_recall_lru);
2763 unhash_delegation(dp);
2765 test_val = nfsd4_lease;
2766 list_for_each_safe(pos, next, &close_lru) {
2767 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2768 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2769 u = sop->so_time - cutoff;
2774 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2776 release_openowner(sop);
2778 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2779 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2780 nfs4_unlock_state();
2781 return clientid_val;
2784 static struct workqueue_struct *laundry_wq;
2785 static void laundromat_main(struct work_struct *);
2786 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2789 laundromat_main(struct work_struct *not_used)
2793 t = nfs4_laundromat();
2794 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2795 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2798 static struct nfs4_stateowner *
2799 search_close_lru(u32 st_id, int flags)
2801 struct nfs4_stateowner *local = NULL;
2803 if (flags & CLOSE_STATE) {
2804 list_for_each_entry(local, &close_lru, so_close_lru) {
2805 if (local->so_id == st_id)
2813 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2815 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2819 STALE_STATEID(stateid_t *stateid)
2821 if (stateid->si_boot == boot_time)
2823 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2824 STATEID_VAL(stateid));
2829 access_permit_read(unsigned long access_bmap)
2831 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2832 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2833 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2837 access_permit_write(unsigned long access_bmap)
2839 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2840 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2844 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2846 __be32 status = nfserr_openmode;
2848 /* For lock stateid's, we test the parent open, not the lock: */
2849 if (stp->st_openstp)
2850 stp = stp->st_openstp;
2851 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2853 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2860 static inline __be32
2861 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2863 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2865 else if (locks_in_grace()) {
2866 /* Answer in remaining cases depends on existance of
2867 * conflicting state; so we must wait out the grace period. */
2868 return nfserr_grace;
2869 } else if (flags & WR_STATE)
2870 return nfs4_share_conflict(current_fh,
2871 NFS4_SHARE_DENY_WRITE);
2872 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2873 return nfs4_share_conflict(current_fh,
2874 NFS4_SHARE_DENY_READ);
2878 * Allow READ/WRITE during grace period on recovered state only for files
2879 * that are not able to provide mandatory locking.
2882 grace_disallows_io(struct inode *inode)
2884 return locks_in_grace() && mandatory_lock(inode);
2887 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2890 * When sessions are used the stateid generation number is ignored
2893 if ((flags & HAS_SESSION) && in->si_generation == 0)
2896 /* If the client sends us a stateid from the future, it's buggy: */
2897 if (in->si_generation > ref->si_generation)
2898 return nfserr_bad_stateid;
2900 * The following, however, can happen. For example, if the
2901 * client sends an open and some IO at the same time, the open
2902 * may bump si_generation while the IO is still in flight.
2903 * Thanks to hard links and renames, the client never knows what
2904 * file an open will affect. So it could avoid that situation
2905 * only by serializing all opens and IO from the same open
2906 * owner. To recover from the old_stateid error, the client
2907 * will just have to retry the IO:
2909 if (in->si_generation < ref->si_generation)
2910 return nfserr_old_stateid;
2915 static int is_delegation_stateid(stateid_t *stateid)
2917 return stateid->si_fileid == 0;
2921 * Checks for stateid operations
2924 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2925 stateid_t *stateid, int flags, struct file **filpp)
2927 struct nfs4_stateid *stp = NULL;
2928 struct nfs4_delegation *dp = NULL;
2929 struct svc_fh *current_fh = &cstate->current_fh;
2930 struct inode *ino = current_fh->fh_dentry->d_inode;
2936 if (grace_disallows_io(ino))
2937 return nfserr_grace;
2939 if (nfsd4_has_session(cstate))
2940 flags |= HAS_SESSION;
2942 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2943 return check_special_stateids(current_fh, stateid, flags);
2945 status = nfserr_stale_stateid;
2946 if (STALE_STATEID(stateid))
2949 status = nfserr_bad_stateid;
2950 if (is_delegation_stateid(stateid)) {
2951 dp = find_delegation_stateid(ino, stateid);
2954 status = check_stateid_generation(stateid, &dp->dl_stateid,
2958 status = nfs4_check_delegmode(dp, flags);
2961 renew_client(dp->dl_client);
2963 *filpp = find_readable_file(dp->dl_file);
2965 } else { /* open or lock stateid */
2966 stp = find_stateid(stateid, flags);
2969 if (nfs4_check_fh(current_fh, stp))
2971 if (!stp->st_stateowner->so_confirmed)
2973 status = check_stateid_generation(stateid, &stp->st_stateid,
2977 status = nfs4_check_openmode(stp, flags);
2980 renew_client(stp->st_stateowner->so_client);
2982 if (flags & RD_STATE)
2983 *filpp = find_readable_file(stp->st_file);
2985 *filpp = find_writeable_file(stp->st_file);
2986 BUG_ON(!*filpp); /* assured by check_openmode */
2997 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2998 RD_STATE : WR_STATE;
3002 * Checks for sequence id mutating operations.
3005 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3006 stateid_t *stateid, int flags,
3007 struct nfs4_stateowner **sopp,
3008 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3010 struct nfs4_stateid *stp;
3011 struct nfs4_stateowner *sop;
3012 struct svc_fh *current_fh = &cstate->current_fh;
3015 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3016 seqid, STATEID_VAL(stateid));
3021 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3022 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3023 return nfserr_bad_stateid;
3026 if (STALE_STATEID(stateid))
3027 return nfserr_stale_stateid;
3029 if (nfsd4_has_session(cstate))
3030 flags |= HAS_SESSION;
3033 * We return BAD_STATEID if filehandle doesn't match stateid,
3034 * the confirmed flag is incorrecly set, or the generation
3035 * number is incorrect.
3037 stp = find_stateid(stateid, flags);
3040 * Also, we should make sure this isn't just the result of
3043 sop = search_close_lru(stateid->si_stateownerid, flags);
3045 return nfserr_bad_stateid;
3051 *sopp = sop = stp->st_stateowner;
3054 clientid_t *lockclid = &lock->v.new.clientid;
3055 struct nfs4_client *clp = sop->so_client;
3059 lkflg = setlkflg(lock->lk_type);
3061 if (lock->lk_is_new) {
3062 if (!sop->so_is_open_owner)
3063 return nfserr_bad_stateid;
3064 if (!(flags & HAS_SESSION) &&
3065 !same_clid(&clp->cl_clientid, lockclid))
3066 return nfserr_bad_stateid;
3067 /* stp is the open stateid */
3068 status = nfs4_check_openmode(stp, lkflg);
3072 /* stp is the lock stateid */
3073 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3079 if (nfs4_check_fh(current_fh, stp)) {
3080 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3081 return nfserr_bad_stateid;
3085 * We now validate the seqid and stateid generation numbers.
3086 * For the moment, we ignore the possibility of
3087 * generation number wraparound.
3089 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3092 if (sop->so_confirmed && flags & CONFIRM) {
3093 dprintk("NFSD: preprocess_seqid_op: expected"
3094 " unconfirmed stateowner!\n");
3095 return nfserr_bad_stateid;
3097 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3098 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3099 " confirmed yet!\n");
3100 return nfserr_bad_stateid;
3102 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3105 renew_client(sop->so_client);
3109 if (seqid == sop->so_seqid - 1) {
3110 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3111 /* indicate replay to calling function */
3112 return nfserr_replay_me;
3114 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3115 sop->so_seqid, seqid);
3117 return nfserr_bad_seqid;
3121 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3122 struct nfsd4_open_confirm *oc)
3125 struct nfs4_stateowner *sop;
3126 struct nfs4_stateid *stp;
3128 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3129 (int)cstate->current_fh.fh_dentry->d_name.len,
3130 cstate->current_fh.fh_dentry->d_name.name);
3132 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3138 if ((status = nfs4_preprocess_seqid_op(cstate,
3139 oc->oc_seqid, &oc->oc_req_stateid,
3140 CONFIRM | OPEN_STATE,
3141 &oc->oc_stateowner, &stp, NULL)))
3144 sop = oc->oc_stateowner;
3145 sop->so_confirmed = 1;
3146 update_stateid(&stp->st_stateid);
3147 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3148 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3149 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3151 nfsd4_create_clid_dir(sop->so_client);
3153 if (oc->oc_stateowner) {
3154 nfs4_get_stateowner(oc->oc_stateowner);
3155 cstate->replay_owner = oc->oc_stateowner;
3157 nfs4_unlock_state();
3163 * unset all bits in union bitmap (bmap) that
3164 * do not exist in share (from successful OPEN_DOWNGRADE)
3167 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3170 for (i = 1; i < 4; i++) {
3171 if ((i & access) != i)
3172 __clear_bit(i, bmap);
3177 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3180 for (i = 0; i < 4; i++) {
3181 if ((i & deny) != i)
3182 __clear_bit(i, bmap);
3187 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3188 struct nfsd4_compound_state *cstate,
3189 struct nfsd4_open_downgrade *od)
3192 struct nfs4_stateid *stp;
3193 unsigned int share_access;
3195 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3196 (int)cstate->current_fh.fh_dentry->d_name.len,
3197 cstate->current_fh.fh_dentry->d_name.name);
3199 if (!access_valid(od->od_share_access, cstate->minorversion)
3200 || !deny_valid(od->od_share_deny))
3201 return nfserr_inval;
3204 if ((status = nfs4_preprocess_seqid_op(cstate,
3208 &od->od_stateowner, &stp, NULL)))
3211 status = nfserr_inval;
3212 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3213 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3214 stp->st_access_bmap, od->od_share_access);
3217 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3218 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3219 stp->st_deny_bmap, od->od_share_deny);
3222 set_access(&share_access, stp->st_access_bmap);
3223 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3225 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3226 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3228 update_stateid(&stp->st_stateid);
3229 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3232 if (od->od_stateowner) {
3233 nfs4_get_stateowner(od->od_stateowner);
3234 cstate->replay_owner = od->od_stateowner;
3236 nfs4_unlock_state();
3241 * nfs4_unlock_state() called after encode
3244 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3245 struct nfsd4_close *close)
3248 struct nfs4_stateid *stp;
3250 dprintk("NFSD: nfsd4_close on file %.*s\n",
3251 (int)cstate->current_fh.fh_dentry->d_name.len,
3252 cstate->current_fh.fh_dentry->d_name.name);
3255 /* check close_lru for replay */
3256 if ((status = nfs4_preprocess_seqid_op(cstate,
3259 OPEN_STATE | CLOSE_STATE,
3260 &close->cl_stateowner, &stp, NULL)))
3263 update_stateid(&stp->st_stateid);
3264 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3266 /* release_stateid() calls nfsd_close() if needed */
3267 release_open_stateid(stp);
3269 /* place unused nfs4_stateowners on so_close_lru list to be
3270 * released by the laundromat service after the lease period
3271 * to enable us to handle CLOSE replay
3273 if (list_empty(&close->cl_stateowner->so_stateids))
3274 move_to_close_lru(close->cl_stateowner);
3276 if (close->cl_stateowner) {
3277 nfs4_get_stateowner(close->cl_stateowner);
3278 cstate->replay_owner = close->cl_stateowner;
3280 nfs4_unlock_state();
3285 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3286 struct nfsd4_delegreturn *dr)
3288 struct nfs4_delegation *dp;
3289 stateid_t *stateid = &dr->dr_stateid;
3290 struct inode *inode;
3294 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3296 inode = cstate->current_fh.fh_dentry->d_inode;
3298 if (nfsd4_has_session(cstate))
3299 flags |= HAS_SESSION;
3301 status = nfserr_bad_stateid;
3302 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3304 status = nfserr_stale_stateid;
3305 if (STALE_STATEID(stateid))
3307 status = nfserr_bad_stateid;
3308 if (!is_delegation_stateid(stateid))
3310 dp = find_delegation_stateid(inode, stateid);
3313 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3316 renew_client(dp->dl_client);
3318 unhash_delegation(dp);
3320 nfs4_unlock_state();
3327 * Lock owner state (byte-range locks)
3329 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3330 #define LOCK_HASH_BITS 8
3331 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3332 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3335 end_offset(u64 start, u64 len)
3340 return end >= start ? end: NFS4_MAX_UINT64;
3343 /* last octet in a range */
3345 last_byte_offset(u64 start, u64 len)
3351 return end > start ? end - 1: NFS4_MAX_UINT64;
3354 #define lockownerid_hashval(id) \
3355 ((id) & LOCK_HASH_MASK)
3357 static inline unsigned int
3358 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3359 struct xdr_netobj *ownername)
3361 return (file_hashval(inode) + cl_id
3362 + opaque_hashval(ownername->data, ownername->len))
3366 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3367 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3368 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3370 static struct nfs4_stateid *
3371 find_stateid(stateid_t *stid, int flags)
3373 struct nfs4_stateid *local;
3374 u32 st_id = stid->si_stateownerid;
3375 u32 f_id = stid->si_fileid;
3376 unsigned int hashval;
3378 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3379 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3380 hashval = stateid_hashval(st_id, f_id);
3381 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3382 if ((local->st_stateid.si_stateownerid == st_id) &&
3383 (local->st_stateid.si_fileid == f_id))
3388 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3389 hashval = stateid_hashval(st_id, f_id);
3390 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3391 if ((local->st_stateid.si_stateownerid == st_id) &&
3392 (local->st_stateid.si_fileid == f_id))
3399 static struct nfs4_delegation *
3400 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3402 struct nfs4_file *fp;
3403 struct nfs4_delegation *dl;
3405 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3408 fp = find_file(ino);
3411 dl = find_delegation_file(fp, stid);
3417 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3418 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3419 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3420 * locking, this prevents us from being completely protocol-compliant. The
3421 * real solution to this problem is to start using unsigned file offsets in
3422 * the VFS, but this is a very deep change!
3425 nfs4_transform_lock_offset(struct file_lock *lock)
3427 if (lock->fl_start < 0)
3428 lock->fl_start = OFFSET_MAX;
3429 if (lock->fl_end < 0)
3430 lock->fl_end = OFFSET_MAX;
3433 /* Hack!: For now, we're defining this just so we can use a pointer to it
3434 * as a unique cookie to identify our (NFSv4's) posix locks. */
3435 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3439 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3441 struct nfs4_stateowner *sop;
3443 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3444 sop = (struct nfs4_stateowner *) fl->fl_owner;
3445 kref_get(&sop->so_ref);
3447 deny->ld_clientid = sop->so_client->cl_clientid;
3449 deny->ld_sop = NULL;
3450 deny->ld_clientid.cl_boot = 0;
3451 deny->ld_clientid.cl_id = 0;
3453 deny->ld_start = fl->fl_start;
3454 deny->ld_length = NFS4_MAX_UINT64;
3455 if (fl->fl_end != NFS4_MAX_UINT64)
3456 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3457 deny->ld_type = NFS4_READ_LT;
3458 if (fl->fl_type != F_RDLCK)
3459 deny->ld_type = NFS4_WRITE_LT;
3462 static struct nfs4_stateowner *
3463 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3464 struct xdr_netobj *owner)
3466 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3467 struct nfs4_stateowner *op;
3469 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3470 if (same_owner_str(op, owner, clid))
3477 * Alloc a lock owner structure.
3478 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3481 * strhashval = lock_ownerstr_hashval
3484 static struct nfs4_stateowner *
3485 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3486 struct nfs4_stateowner *sop;
3487 struct nfs4_replay *rp;
3488 unsigned int idhashval;
3490 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3492 idhashval = lockownerid_hashval(current_ownerid);
3493 INIT_LIST_HEAD(&sop->so_idhash);
3494 INIT_LIST_HEAD(&sop->so_strhash);
3495 INIT_LIST_HEAD(&sop->so_perclient);
3496 INIT_LIST_HEAD(&sop->so_stateids);
3497 INIT_LIST_HEAD(&sop->so_perstateid);
3498 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3500 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3501 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3502 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3503 sop->so_is_open_owner = 0;
3504 sop->so_id = current_ownerid++;
3505 sop->so_client = clp;
3506 /* It is the openowner seqid that will be incremented in encode in the
3507 * case of new lockowners; so increment the lock seqid manually: */
3508 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3509 sop->so_confirmed = 1;
3510 rp = &sop->so_replay;
3511 rp->rp_status = nfserr_serverfault;
3513 rp->rp_buf = rp->rp_ibuf;
3517 static struct nfs4_stateid *
3518 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3520 struct nfs4_stateid *stp;
3521 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3523 stp = nfs4_alloc_stateid();
3526 INIT_LIST_HEAD(&stp->st_hash);
3527 INIT_LIST_HEAD(&stp->st_perfile);
3528 INIT_LIST_HEAD(&stp->st_perstateowner);
3529 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3530 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3531 list_add(&stp->st_perfile, &fp->fi_stateids);
3532 list_add(&stp->st_perstateowner, &sop->so_stateids);
3533 stp->st_stateowner = sop;
3536 stp->st_stateid.si_boot = boot_time;
3537 stp->st_stateid.si_stateownerid = sop->so_id;
3538 stp->st_stateid.si_fileid = fp->fi_id;
3539 stp->st_stateid.si_generation = 0;
3540 stp->st_deny_bmap = open_stp->st_deny_bmap;
3541 stp->st_openstp = open_stp;
3548 check_lock_length(u64 offset, u64 length)
3550 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3551 LOFF_OVERFLOW(offset, length)));
3558 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3559 struct nfsd4_lock *lock)
3561 struct nfs4_stateowner *open_sop = NULL;
3562 struct nfs4_stateowner *lock_sop = NULL;
3563 struct nfs4_stateid *lock_stp;
3565 struct file_lock file_lock;
3566 struct file_lock conflock;
3568 unsigned int strhashval;
3572 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3573 (long long) lock->lk_offset,
3574 (long long) lock->lk_length);
3576 if (check_lock_length(lock->lk_offset, lock->lk_length))
3577 return nfserr_inval;
3579 if ((status = fh_verify(rqstp, &cstate->current_fh,
3580 S_IFREG, NFSD_MAY_LOCK))) {
3581 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3587 if (lock->lk_is_new) {
3589 * Client indicates that this is a new lockowner.
3590 * Use open owner and open stateid to create lock owner and
3593 struct nfs4_stateid *open_stp = NULL;
3594 struct nfs4_file *fp;
3596 status = nfserr_stale_clientid;
3597 if (!nfsd4_has_session(cstate) &&
3598 STALE_CLIENTID(&lock->lk_new_clientid))
3601 /* validate and update open stateid and open seqid */
3602 status = nfs4_preprocess_seqid_op(cstate,
3603 lock->lk_new_open_seqid,
3604 &lock->lk_new_open_stateid,
3606 &lock->lk_replay_owner, &open_stp,
3610 open_sop = lock->lk_replay_owner;
3611 /* create lockowner and lock stateid */
3612 fp = open_stp->st_file;
3613 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3614 open_sop->so_client->cl_clientid.cl_id,
3615 &lock->v.new.owner);
3616 /* XXX: Do we need to check for duplicate stateowners on
3617 * the same file, or should they just be allowed (and
3618 * create new stateids)? */
3619 status = nfserr_resource;
3620 lock_sop = alloc_init_lock_stateowner(strhashval,
3621 open_sop->so_client, open_stp, lock);
3622 if (lock_sop == NULL)
3624 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3625 if (lock_stp == NULL)
3628 /* lock (lock owner + lock stateid) already exists */
3629 status = nfs4_preprocess_seqid_op(cstate,
3630 lock->lk_old_lock_seqid,
3631 &lock->lk_old_lock_stateid,
3633 &lock->lk_replay_owner, &lock_stp, lock);
3636 lock_sop = lock->lk_replay_owner;
3638 /* lock->lk_replay_owner and lock_stp have been created or found */
3640 status = nfserr_grace;
3641 if (locks_in_grace() && !lock->lk_reclaim)
3643 status = nfserr_no_grace;
3644 if (!locks_in_grace() && lock->lk_reclaim)
3647 locks_init_lock(&file_lock);
3648 switch (lock->lk_type) {
3651 filp = find_readable_file(lock_stp->st_file);
3652 file_lock.fl_type = F_RDLCK;
3656 case NFS4_WRITEW_LT:
3657 filp = find_writeable_file(lock_stp->st_file);
3658 file_lock.fl_type = F_WRLCK;
3662 status = nfserr_inval;
3666 status = nfserr_openmode;
3669 file_lock.fl_owner = (fl_owner_t)lock_sop;
3670 file_lock.fl_pid = current->tgid;
3671 file_lock.fl_file = filp;
3672 file_lock.fl_flags = FL_POSIX;
3673 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3675 file_lock.fl_start = lock->lk_offset;
3676 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3677 nfs4_transform_lock_offset(&file_lock);
3680 * Try to lock the file in the VFS.
3681 * Note: locks.c uses the BKL to protect the inode's lock list.
3684 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3686 case 0: /* success! */
3687 update_stateid(&lock_stp->st_stateid);
3688 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3692 case (EAGAIN): /* conflock holds conflicting lock */
3693 status = nfserr_denied;
3694 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3695 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3698 status = nfserr_deadlock;
3701 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3702 status = nfserr_resource;
3706 if (status && lock->lk_is_new && lock_sop)
3707 release_lockowner(lock_sop);
3708 if (lock->lk_replay_owner) {
3709 nfs4_get_stateowner(lock->lk_replay_owner);
3710 cstate->replay_owner = lock->lk_replay_owner;
3712 nfs4_unlock_state();
3717 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3718 * so we do a temporary open here just to get an open file to pass to
3719 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3722 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3727 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3730 err = vfs_test_lock(file, lock);
3739 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3740 struct nfsd4_lockt *lockt)
3742 struct inode *inode;
3743 struct file_lock file_lock;
3747 if (locks_in_grace())
3748 return nfserr_grace;
3750 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3751 return nfserr_inval;
3753 lockt->lt_stateowner = NULL;
3756 status = nfserr_stale_clientid;
3757 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3760 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3761 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3762 if (status == nfserr_symlink)
3763 status = nfserr_inval;
3767 inode = cstate->current_fh.fh_dentry->d_inode;
3768 locks_init_lock(&file_lock);
3769 switch (lockt->lt_type) {
3772 file_lock.fl_type = F_RDLCK;
3775 case NFS4_WRITEW_LT:
3776 file_lock.fl_type = F_WRLCK;
3779 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3780 status = nfserr_inval;
3784 lockt->lt_stateowner = find_lockstateowner_str(inode,
3785 &lockt->lt_clientid, &lockt->lt_owner);
3786 if (lockt->lt_stateowner)
3787 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3788 file_lock.fl_pid = current->tgid;
3789 file_lock.fl_flags = FL_POSIX;
3791 file_lock.fl_start = lockt->lt_offset;
3792 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3794 nfs4_transform_lock_offset(&file_lock);
3797 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3799 status = nfserrno(error);
3802 if (file_lock.fl_type != F_UNLCK) {
3803 status = nfserr_denied;
3804 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3807 nfs4_unlock_state();
3812 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3813 struct nfsd4_locku *locku)
3815 struct nfs4_stateid *stp;
3816 struct file *filp = NULL;
3817 struct file_lock file_lock;
3821 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3822 (long long) locku->lu_offset,
3823 (long long) locku->lu_length);
3825 if (check_lock_length(locku->lu_offset, locku->lu_length))
3826 return nfserr_inval;
3830 if ((status = nfs4_preprocess_seqid_op(cstate,
3834 &locku->lu_stateowner, &stp, NULL)))
3837 filp = find_any_file(stp->st_file);
3839 status = nfserr_lock_range;
3843 locks_init_lock(&file_lock);
3844 file_lock.fl_type = F_UNLCK;
3845 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3846 file_lock.fl_pid = current->tgid;
3847 file_lock.fl_file = filp;
3848 file_lock.fl_flags = FL_POSIX;
3849 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3850 file_lock.fl_start = locku->lu_offset;
3852 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3853 nfs4_transform_lock_offset(&file_lock);
3856 * Try to unlock the file in the VFS.
3858 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3860 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3864 * OK, unlock succeeded; the only thing left to do is update the stateid.
3866 update_stateid(&stp->st_stateid);
3867 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3870 if (locku->lu_stateowner) {
3871 nfs4_get_stateowner(locku->lu_stateowner);
3872 cstate->replay_owner = locku->lu_stateowner;
3874 nfs4_unlock_state();
3878 status = nfserrno(err);
3884 * 1: locks held by lockowner
3885 * 0: no locks held by lockowner
3888 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
3890 struct file_lock **flpp;
3891 struct inode *inode = filp->fi_inode;
3895 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3896 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3907 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3908 struct nfsd4_compound_state *cstate,
3909 struct nfsd4_release_lockowner *rlockowner)
3911 clientid_t *clid = &rlockowner->rl_clientid;
3912 struct nfs4_stateowner *sop;
3913 struct nfs4_stateid *stp;
3914 struct xdr_netobj *owner = &rlockowner->rl_owner;
3915 struct list_head matches;
3919 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3920 clid->cl_boot, clid->cl_id);
3922 /* XXX check for lease expiration */
3924 status = nfserr_stale_clientid;
3925 if (STALE_CLIENTID(clid))
3930 status = nfserr_locks_held;
3931 /* XXX: we're doing a linear search through all the lockowners.
3932 * Yipes! For now we'll just hope clients aren't really using
3933 * release_lockowner much, but eventually we have to fix these
3934 * data structures. */
3935 INIT_LIST_HEAD(&matches);
3936 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3937 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3938 if (!same_owner_str(sop, owner, clid))
3940 list_for_each_entry(stp, &sop->so_stateids,
3942 if (check_for_locks(stp->st_file, sop))
3944 /* Note: so_perclient unused for lockowners,
3945 * so it's OK to fool with here. */
3946 list_add(&sop->so_perclient, &matches);
3950 /* Clients probably won't expect us to return with some (but not all)
3951 * of the lockowner state released; so don't release any until all
3952 * have been checked. */
3954 while (!list_empty(&matches)) {
3955 sop = list_entry(matches.next, struct nfs4_stateowner,
3957 /* unhash_stateowner deletes so_perclient only
3958 * for openowners. */
3959 list_del(&sop->so_perclient);
3960 release_lockowner(sop);
3963 nfs4_unlock_state();
3967 static inline struct nfs4_client_reclaim *
3970 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3974 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3976 unsigned int strhashval = clientstr_hashval(name);
3977 struct nfs4_client *clp;
3979 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3984 * failure => all reset bets are off, nfserr_no_grace...
3987 nfs4_client_to_reclaim(const char *name)
3989 unsigned int strhashval;
3990 struct nfs4_client_reclaim *crp = NULL;
3992 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3993 crp = alloc_reclaim();
3996 strhashval = clientstr_hashval(name);
3997 INIT_LIST_HEAD(&crp->cr_strhash);
3998 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3999 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4000 reclaim_str_hashtbl_size++;
4005 nfs4_release_reclaim(void)
4007 struct nfs4_client_reclaim *crp = NULL;
4010 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4011 while (!list_empty(&reclaim_str_hashtbl[i])) {
4012 crp = list_entry(reclaim_str_hashtbl[i].next,
4013 struct nfs4_client_reclaim, cr_strhash);
4014 list_del(&crp->cr_strhash);
4016 reclaim_str_hashtbl_size--;
4019 BUG_ON(reclaim_str_hashtbl_size);
4023 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4024 static struct nfs4_client_reclaim *
4025 nfs4_find_reclaim_client(clientid_t *clid)
4027 unsigned int strhashval;
4028 struct nfs4_client *clp;
4029 struct nfs4_client_reclaim *crp = NULL;
4032 /* find clientid in conf_id_hashtbl */
4033 clp = find_confirmed_client(clid);
4037 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4038 clp->cl_name.len, clp->cl_name.data,
4041 /* find clp->cl_name in reclaim_str_hashtbl */
4042 strhashval = clientstr_hashval(clp->cl_recdir);
4043 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4044 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4052 * Called from OPEN. Look for clientid in reclaim list.
4055 nfs4_check_open_reclaim(clientid_t *clid)
4057 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4060 /* initialization to perform at module load time: */
4063 nfs4_state_init(void)
4067 status = nfsd4_init_slabs();
4070 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4071 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4072 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4073 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4074 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4075 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4077 for (i = 0; i < SESSION_HASH_SIZE; i++)
4078 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4079 for (i = 0; i < FILE_HASH_SIZE; i++) {
4080 INIT_LIST_HEAD(&file_hashtbl[i]);
4082 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4083 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4084 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4086 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4087 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4088 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4090 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4091 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4092 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4094 memset(&onestateid, ~0, sizeof(stateid_t));
4095 INIT_LIST_HEAD(&close_lru);
4096 INIT_LIST_HEAD(&client_lru);
4097 INIT_LIST_HEAD(&del_recall_lru);
4098 reclaim_str_hashtbl_size = 0;
4103 nfsd4_load_reboot_recovery_data(void)
4108 nfsd4_init_recdir(user_recovery_dirname);
4109 status = nfsd4_recdir_load();
4110 nfs4_unlock_state();
4112 printk("NFSD: Failure reading reboot recovery data\n");
4116 * Since the lifetime of a delegation isn't limited to that of an open, a
4117 * client may quite reasonably hang on to a delegation as long as it has
4118 * the inode cached. This becomes an obvious problem the first time a
4119 * client's inode cache approaches the size of the server's total memory.
4121 * For now we avoid this problem by imposing a hard limit on the number
4122 * of delegations, which varies according to the server's memory size.
4125 set_max_delegations(void)
4128 * Allow at most 4 delegations per megabyte of RAM. Quick
4129 * estimates suggest that in the worst case (where every delegation
4130 * is for a different inode), a delegation could take about 1.5K,
4131 * giving a worst case usage of about 6% of memory.
4133 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4136 /* initialization to perform when the nfsd service is started: */
4139 __nfs4_state_start(void)
4143 boot_time = get_seconds();
4144 locks_start_grace(&nfsd4_manager);
4145 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4147 ret = set_callback_cred();
4150 laundry_wq = create_singlethread_workqueue("nfsd4");
4151 if (laundry_wq == NULL)
4153 ret = nfsd4_create_callback_queue();
4155 goto out_free_laundry;
4156 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4157 set_max_delegations();
4160 destroy_workqueue(laundry_wq);
4165 nfs4_state_start(void)
4167 nfsd4_load_reboot_recovery_data();
4168 return __nfs4_state_start();
4172 __nfs4_state_shutdown(void)
4175 struct nfs4_client *clp = NULL;
4176 struct nfs4_delegation *dp = NULL;
4177 struct list_head *pos, *next, reaplist;
4179 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4180 while (!list_empty(&conf_id_hashtbl[i])) {
4181 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4184 while (!list_empty(&unconf_str_hashtbl[i])) {
4185 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4189 INIT_LIST_HEAD(&reaplist);
4190 spin_lock(&recall_lock);
4191 list_for_each_safe(pos, next, &del_recall_lru) {
4192 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4193 list_move(&dp->dl_recall_lru, &reaplist);
4195 spin_unlock(&recall_lock);
4196 list_for_each_safe(pos, next, &reaplist) {
4197 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4198 list_del_init(&dp->dl_recall_lru);
4199 unhash_delegation(dp);
4202 nfsd4_shutdown_recdir();
4206 nfs4_state_shutdown(void)
4208 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4209 destroy_workqueue(laundry_wq);
4210 locks_end_grace(&nfsd4_manager);
4212 nfs4_release_reclaim();
4213 __nfs4_state_shutdown();
4214 nfs4_unlock_state();
4215 nfsd4_destroy_callback_queue();
4219 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4220 * accessed when nfsd is starting.
4223 nfs4_set_recdir(char *recdir)
4225 strcpy(user_recovery_dirname, recdir);
4229 * Change the NFSv4 recovery directory to recdir.
4232 nfs4_reset_recoverydir(char *recdir)
4237 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4241 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4242 nfs4_set_recdir(recdir);
4250 nfs4_recoverydir(void)
4252 return user_recovery_dirname;