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 struct nfs4_delegation *
166 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
168 struct nfs4_delegation *dp;
169 struct nfs4_file *fp = stp->st_file;
170 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
172 dprintk("NFSD alloc_init_deleg\n");
174 * Major work on the lease subsystem (for example, to support
175 * calbacks on stat) will be required before we can support
176 * write delegations properly.
178 if (type != NFS4_OPEN_DELEGATE_READ)
180 if (fp->fi_had_conflict)
182 if (num_delegations > max_delegations)
184 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
188 INIT_LIST_HEAD(&dp->dl_perfile);
189 INIT_LIST_HEAD(&dp->dl_perclnt);
190 INIT_LIST_HEAD(&dp->dl_recall_lru);
195 get_file(stp->st_vfs_file);
196 dp->dl_vfs_file = stp->st_vfs_file;
198 dp->dl_ident = cb->cb_ident;
199 dp->dl_stateid.si_boot = boot_time;
200 dp->dl_stateid.si_stateownerid = current_delegid++;
201 dp->dl_stateid.si_fileid = 0;
202 dp->dl_stateid.si_generation = 0;
203 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
205 atomic_set(&dp->dl_count, 1);
206 list_add(&dp->dl_perfile, &fp->fi_delegations);
207 list_add(&dp->dl_perclnt, &clp->cl_delegations);
208 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
213 nfs4_put_delegation(struct nfs4_delegation *dp)
215 if (atomic_dec_and_test(&dp->dl_count)) {
216 dprintk("NFSD: freeing dp %p\n",dp);
217 put_nfs4_file(dp->dl_file);
218 kmem_cache_free(deleg_slab, dp);
223 /* Remove the associated file_lock first, then remove the delegation.
224 * lease_modify() is called to remove the FS_LEASE file_lock from
225 * the i_flock list, eventually calling nfsd's lock_manager
226 * fl_release_callback.
229 nfs4_close_delegation(struct nfs4_delegation *dp)
231 struct file *filp = dp->dl_vfs_file;
233 dprintk("NFSD: close_delegation dp %p\n",dp);
234 dp->dl_vfs_file = NULL;
235 /* The following nfsd_close may not actually close the file,
236 * but we want to remove the lease in any case. */
238 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
242 /* Called under the state lock. */
244 unhash_delegation(struct nfs4_delegation *dp)
246 list_del_init(&dp->dl_perfile);
247 list_del_init(&dp->dl_perclnt);
248 spin_lock(&recall_lock);
249 list_del_init(&dp->dl_recall_lru);
250 spin_unlock(&recall_lock);
251 nfs4_close_delegation(dp);
252 nfs4_put_delegation(dp);
259 /* client_lock protects the client lru list and session hash table */
260 static DEFINE_SPINLOCK(client_lock);
262 /* Hash tables for nfs4_clientid state */
263 #define CLIENT_HASH_BITS 4
264 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
265 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
267 #define clientid_hashval(id) \
268 ((id) & CLIENT_HASH_MASK)
269 #define clientstr_hashval(name) \
270 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
272 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
273 * used in reboot/reset lease grace period processing
275 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
276 * setclientid_confirmed info.
278 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
281 * client_lru holds client queue ordered by nfs4_client.cl_time
284 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
285 * for last close replay.
287 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
288 static int reclaim_str_hashtbl_size = 0;
289 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
290 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
291 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
292 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
293 static struct list_head client_lru;
294 static struct list_head close_lru;
296 static void unhash_generic_stateid(struct nfs4_stateid *stp)
298 list_del(&stp->st_hash);
299 list_del(&stp->st_perfile);
300 list_del(&stp->st_perstateowner);
303 static void free_generic_stateid(struct nfs4_stateid *stp)
305 put_nfs4_file(stp->st_file);
306 kmem_cache_free(stateid_slab, stp);
309 static void release_lock_stateid(struct nfs4_stateid *stp)
311 unhash_generic_stateid(stp);
312 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
313 free_generic_stateid(stp);
316 static void unhash_lockowner(struct nfs4_stateowner *sop)
318 struct nfs4_stateid *stp;
320 list_del(&sop->so_idhash);
321 list_del(&sop->so_strhash);
322 list_del(&sop->so_perstateid);
323 while (!list_empty(&sop->so_stateids)) {
324 stp = list_first_entry(&sop->so_stateids,
325 struct nfs4_stateid, st_perstateowner);
326 release_lock_stateid(stp);
330 static void release_lockowner(struct nfs4_stateowner *sop)
332 unhash_lockowner(sop);
333 nfs4_put_stateowner(sop);
337 release_stateid_lockowners(struct nfs4_stateid *open_stp)
339 struct nfs4_stateowner *lock_sop;
341 while (!list_empty(&open_stp->st_lockowners)) {
342 lock_sop = list_entry(open_stp->st_lockowners.next,
343 struct nfs4_stateowner, so_perstateid);
344 /* list_del(&open_stp->st_lockowners); */
345 BUG_ON(lock_sop->so_is_open_owner);
346 release_lockowner(lock_sop);
350 static void release_open_stateid(struct nfs4_stateid *stp)
352 unhash_generic_stateid(stp);
353 release_stateid_lockowners(stp);
354 nfsd_close(stp->st_vfs_file);
355 free_generic_stateid(stp);
358 static void unhash_openowner(struct nfs4_stateowner *sop)
360 struct nfs4_stateid *stp;
362 list_del(&sop->so_idhash);
363 list_del(&sop->so_strhash);
364 list_del(&sop->so_perclient);
365 list_del(&sop->so_perstateid); /* XXX: necessary? */
366 while (!list_empty(&sop->so_stateids)) {
367 stp = list_first_entry(&sop->so_stateids,
368 struct nfs4_stateid, st_perstateowner);
369 release_open_stateid(stp);
373 static void release_openowner(struct nfs4_stateowner *sop)
375 unhash_openowner(sop);
376 list_del(&sop->so_close_lru);
377 nfs4_put_stateowner(sop);
380 #define SESSION_HASH_SIZE 512
381 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
384 hash_sessionid(struct nfs4_sessionid *sessionid)
386 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
388 return sid->sequence % SESSION_HASH_SIZE;
392 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
394 u32 *ptr = (u32 *)(&sessionid->data[0]);
395 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
399 gen_sessionid(struct nfsd4_session *ses)
401 struct nfs4_client *clp = ses->se_client;
402 struct nfsd4_sessionid *sid;
404 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
405 sid->clientid = clp->cl_clientid;
406 sid->sequence = current_sessionid++;
411 * The protocol defines ca_maxresponssize_cached to include the size of
412 * the rpc header, but all we need to cache is the data starting after
413 * the end of the initial SEQUENCE operation--the rest we regenerate
414 * each time. Therefore we can advertise a ca_maxresponssize_cached
415 * value that is the number of bytes in our cache plus a few additional
416 * bytes. In order to stay on the safe side, and not promise more than
417 * we can cache, those additional bytes must be the minimum possible: 24
418 * bytes of rpc header (xid through accept state, with AUTH_NULL
419 * verifier), 12 for the compound header (with zero-length tag), and 44
420 * for the SEQUENCE op response:
422 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
425 * Give the client the number of ca_maxresponsesize_cached slots it
426 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
427 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
428 * than NFSD_MAX_SLOTS_PER_SESSION.
430 * If we run out of reserved DRC memory we should (up to a point)
431 * re-negotiate active sessions and reduce their slot usage to make
432 * rooom for new connections. For now we just fail the create session.
434 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
436 int mem, size = fchan->maxresp_cached;
438 if (fchan->maxreqs < 1)
441 if (size < NFSD_MIN_HDR_SEQ_SZ)
442 size = NFSD_MIN_HDR_SEQ_SZ;
443 size -= NFSD_MIN_HDR_SEQ_SZ;
444 if (size > NFSD_SLOT_CACHE_SIZE)
445 size = NFSD_SLOT_CACHE_SIZE;
447 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
448 mem = fchan->maxreqs * size;
449 if (mem > NFSD_MAX_MEM_PER_SESSION) {
450 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
451 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
452 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
453 mem = fchan->maxreqs * size;
456 spin_lock(&nfsd_drc_lock);
457 /* bound the total session drc memory ussage */
458 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
459 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
460 mem = fchan->maxreqs * size;
462 nfsd_drc_mem_used += mem;
463 spin_unlock(&nfsd_drc_lock);
465 if (fchan->maxreqs == 0)
466 return nfserr_jukebox;
468 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
473 * fchan holds the client values on input, and the server values on output
474 * sv_max_mesg is the maximum payload plus one page for overhead.
476 static int init_forechannel_attrs(struct svc_rqst *rqstp,
477 struct nfsd4_channel_attrs *session_fchan,
478 struct nfsd4_channel_attrs *fchan)
481 __u32 maxcount = nfsd_serv->sv_max_mesg;
483 /* headerpadsz set to zero in encode routine */
485 /* Use the client's max request and max response size if possible */
486 if (fchan->maxreq_sz > maxcount)
487 fchan->maxreq_sz = maxcount;
488 session_fchan->maxreq_sz = fchan->maxreq_sz;
490 if (fchan->maxresp_sz > maxcount)
491 fchan->maxresp_sz = maxcount;
492 session_fchan->maxresp_sz = fchan->maxresp_sz;
494 /* Use the client's maxops if possible */
495 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
496 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
497 session_fchan->maxops = fchan->maxops;
499 /* FIXME: Error means no more DRC pages so the server should
500 * recover pages from existing sessions. For now fail session
503 status = set_forechannel_drc_size(fchan);
505 session_fchan->maxresp_cached = fchan->maxresp_cached;
506 session_fchan->maxreqs = fchan->maxreqs;
508 dprintk("%s status %d\n", __func__, status);
513 free_session_slots(struct nfsd4_session *ses)
517 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
518 kfree(ses->se_slots[i]);
522 * We don't actually need to cache the rpc and session headers, so we
523 * can allocate a little less for each slot:
525 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
527 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
531 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
532 struct nfsd4_create_session *cses)
534 struct nfsd4_session *new, tmp;
535 struct nfsd4_slot *sp;
536 int idx, slotsize, cachesize, i;
539 memset(&tmp, 0, sizeof(tmp));
541 /* FIXME: For now, we just accept the client back channel attributes. */
542 tmp.se_bchannel = cses->back_channel;
543 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
544 &cses->fore_channel);
548 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
549 + sizeof(struct nfsd4_session) > PAGE_SIZE);
551 status = nfserr_jukebox;
552 /* allocate struct nfsd4_session and slot table pointers in one piece */
553 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
554 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
558 memcpy(new, &tmp, sizeof(*new));
560 /* allocate each struct nfsd4_slot and data cache in one piece */
561 cachesize = slot_bytes(&new->se_fchannel);
562 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
563 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
566 new->se_slots[i] = sp;
569 new->se_client = clp;
571 idx = hash_sessionid(&new->se_sessionid);
572 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
573 NFS4_MAX_SESSIONID_LEN);
575 new->se_flags = cses->flags;
576 kref_init(&new->se_ref);
577 spin_lock(&client_lock);
578 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
579 list_add(&new->se_perclnt, &clp->cl_sessions);
580 spin_unlock(&client_lock);
586 free_session_slots(new);
591 /* caller must hold client_lock */
592 static struct nfsd4_session *
593 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
595 struct nfsd4_session *elem;
598 dump_sessionid(__func__, sessionid);
599 idx = hash_sessionid(sessionid);
600 /* Search in the appropriate list */
601 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
602 if (!memcmp(elem->se_sessionid.data, sessionid->data,
603 NFS4_MAX_SESSIONID_LEN)) {
608 dprintk("%s: session not found\n", __func__);
612 /* caller must hold client_lock */
614 unhash_session(struct nfsd4_session *ses)
616 list_del(&ses->se_hash);
617 list_del(&ses->se_perclnt);
621 free_session(struct kref *kref)
623 struct nfsd4_session *ses;
626 ses = container_of(kref, struct nfsd4_session, se_ref);
627 spin_lock(&nfsd_drc_lock);
628 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
629 nfsd_drc_mem_used -= mem;
630 spin_unlock(&nfsd_drc_lock);
631 free_session_slots(ses);
635 /* must be called under the client_lock */
637 renew_client_locked(struct nfs4_client *clp)
639 if (is_client_expired(clp)) {
640 dprintk("%s: client (clientid %08x/%08x) already expired\n",
642 clp->cl_clientid.cl_boot,
643 clp->cl_clientid.cl_id);
648 * Move client to the end to the LRU list.
650 dprintk("renewing client (clientid %08x/%08x)\n",
651 clp->cl_clientid.cl_boot,
652 clp->cl_clientid.cl_id);
653 list_move_tail(&clp->cl_lru, &client_lru);
654 clp->cl_time = get_seconds();
658 renew_client(struct nfs4_client *clp)
660 spin_lock(&client_lock);
661 renew_client_locked(clp);
662 spin_unlock(&client_lock);
665 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
667 STALE_CLIENTID(clientid_t *clid)
669 if (clid->cl_boot == boot_time)
671 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
672 clid->cl_boot, clid->cl_id, boot_time);
677 * XXX Should we use a slab cache ?
678 * This type of memory management is somewhat inefficient, but we use it
679 * anyway since SETCLIENTID is not a common operation.
681 static struct nfs4_client *alloc_client(struct xdr_netobj name)
683 struct nfs4_client *clp;
685 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
688 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
689 if (clp->cl_name.data == NULL) {
693 memcpy(clp->cl_name.data, name.data, name.len);
694 clp->cl_name.len = name.len;
699 free_client(struct nfs4_client *clp)
701 if (clp->cl_cred.cr_group_info)
702 put_group_info(clp->cl_cred.cr_group_info);
703 kfree(clp->cl_principal);
704 kfree(clp->cl_name.data);
709 release_session_client(struct nfsd4_session *session)
711 struct nfs4_client *clp = session->se_client;
713 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
715 if (is_client_expired(clp)) {
717 session->se_client = NULL;
719 renew_client_locked(clp);
720 spin_unlock(&client_lock);
723 /* must be called under the client_lock */
725 unhash_client_locked(struct nfs4_client *clp)
727 mark_client_expired(clp);
728 list_del(&clp->cl_lru);
729 while (!list_empty(&clp->cl_sessions)) {
730 struct nfsd4_session *ses;
731 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
734 nfsd4_put_session(ses);
739 expire_client(struct nfs4_client *clp)
741 struct nfs4_stateowner *sop;
742 struct nfs4_delegation *dp;
743 struct list_head reaplist;
745 INIT_LIST_HEAD(&reaplist);
746 spin_lock(&recall_lock);
747 while (!list_empty(&clp->cl_delegations)) {
748 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
749 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
751 list_del_init(&dp->dl_perclnt);
752 list_move(&dp->dl_recall_lru, &reaplist);
754 spin_unlock(&recall_lock);
755 while (!list_empty(&reaplist)) {
756 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
757 list_del_init(&dp->dl_recall_lru);
758 unhash_delegation(dp);
760 while (!list_empty(&clp->cl_openowners)) {
761 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
762 release_openowner(sop);
764 nfsd4_set_callback_client(clp, NULL);
765 if (clp->cl_cb_conn.cb_xprt)
766 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
767 list_del(&clp->cl_idhash);
768 list_del(&clp->cl_strhash);
769 spin_lock(&client_lock);
770 unhash_client_locked(clp);
771 if (atomic_read(&clp->cl_refcount) == 0)
773 spin_unlock(&client_lock);
776 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
778 memcpy(target->cl_verifier.data, source->data,
779 sizeof(target->cl_verifier.data));
782 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
784 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
785 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
788 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
790 target->cr_uid = source->cr_uid;
791 target->cr_gid = source->cr_gid;
792 target->cr_group_info = source->cr_group_info;
793 get_group_info(target->cr_group_info);
796 static int same_name(const char *n1, const char *n2)
798 return 0 == memcmp(n1, n2, HEXDIR_LEN);
802 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
804 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
808 same_clid(clientid_t *cl1, clientid_t *cl2)
810 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
813 /* XXX what about NGROUP */
815 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
817 return cr1->cr_uid == cr2->cr_uid;
820 static void gen_clid(struct nfs4_client *clp)
822 static u32 current_clientid = 1;
824 clp->cl_clientid.cl_boot = boot_time;
825 clp->cl_clientid.cl_id = current_clientid++;
828 static void gen_confirm(struct nfs4_client *clp)
833 p = (u32 *)clp->cl_confirm.data;
834 *p++ = get_seconds();
838 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
839 struct svc_rqst *rqstp, nfs4_verifier *verf)
841 struct nfs4_client *clp;
842 struct sockaddr *sa = svc_addr(rqstp);
845 clp = alloc_client(name);
849 princ = svc_gss_principal(rqstp);
851 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
852 if (clp->cl_principal == NULL) {
858 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
859 atomic_set(&clp->cl_refcount, 0);
860 atomic_set(&clp->cl_cb_set, 0);
861 INIT_LIST_HEAD(&clp->cl_idhash);
862 INIT_LIST_HEAD(&clp->cl_strhash);
863 INIT_LIST_HEAD(&clp->cl_openowners);
864 INIT_LIST_HEAD(&clp->cl_delegations);
865 INIT_LIST_HEAD(&clp->cl_sessions);
866 INIT_LIST_HEAD(&clp->cl_lru);
867 clp->cl_time = get_seconds();
868 clear_bit(0, &clp->cl_cb_slot_busy);
869 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
870 copy_verf(clp, verf);
871 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
872 clp->cl_flavor = rqstp->rq_flavor;
873 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
879 static int check_name(struct xdr_netobj name)
883 if (name.len > NFS4_OPAQUE_LIMIT) {
884 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
891 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
893 unsigned int idhashval;
895 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
896 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
897 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
902 move_to_confirmed(struct nfs4_client *clp)
904 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
905 unsigned int strhashval;
907 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
908 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
909 strhashval = clientstr_hashval(clp->cl_recdir);
910 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
914 static struct nfs4_client *
915 find_confirmed_client(clientid_t *clid)
917 struct nfs4_client *clp;
918 unsigned int idhashval = clientid_hashval(clid->cl_id);
920 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
921 if (same_clid(&clp->cl_clientid, clid))
927 static struct nfs4_client *
928 find_unconfirmed_client(clientid_t *clid)
930 struct nfs4_client *clp;
931 unsigned int idhashval = clientid_hashval(clid->cl_id);
933 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
934 if (same_clid(&clp->cl_clientid, clid))
941 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
942 * parameter. Matching is based on the fact the at least one of the
943 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
945 * FIXME: we need to unify the clientid namespaces for nfsv4.x
946 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
947 * and SET_CLIENTID{,_CONFIRM}
950 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
952 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
953 return use_exchange_id == has_exchange_flags;
956 static struct nfs4_client *
957 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
958 bool use_exchange_id)
960 struct nfs4_client *clp;
962 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
963 if (same_name(clp->cl_recdir, dname) &&
964 match_clientid_establishment(clp, use_exchange_id))
970 static struct nfs4_client *
971 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
972 bool use_exchange_id)
974 struct nfs4_client *clp;
976 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
977 if (same_name(clp->cl_recdir, dname) &&
978 match_clientid_establishment(clp, use_exchange_id))
985 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
987 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
988 unsigned short expected_family;
990 /* Currently, we only support tcp and tcp6 for the callback channel */
991 if (se->se_callback_netid_len == 3 &&
992 !memcmp(se->se_callback_netid_val, "tcp", 3))
993 expected_family = AF_INET;
994 else if (se->se_callback_netid_len == 4 &&
995 !memcmp(se->se_callback_netid_val, "tcp6", 4))
996 expected_family = AF_INET6;
1000 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1001 se->se_callback_addr_len,
1002 (struct sockaddr *) &cb->cb_addr,
1003 sizeof(cb->cb_addr));
1005 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
1008 if (cb->cb_addr.ss_family == AF_INET6)
1009 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
1011 cb->cb_minorversion = 0;
1012 cb->cb_prog = se->se_callback_prog;
1013 cb->cb_ident = se->se_callback_ident;
1016 cb->cb_addr.ss_family = AF_UNSPEC;
1018 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1019 "will not receive delegations\n",
1020 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1026 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1029 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1031 struct nfsd4_slot *slot = resp->cstate.slot;
1034 dprintk("--> %s slot %p\n", __func__, slot);
1036 slot->sl_opcnt = resp->opcnt;
1037 slot->sl_status = resp->cstate.status;
1039 if (nfsd4_not_cached(resp)) {
1040 slot->sl_datalen = 0;
1043 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1044 base = (char *)resp->cstate.datap -
1045 (char *)resp->xbuf->head[0].iov_base;
1046 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1048 WARN("%s: sessions DRC could not cache compound\n", __func__);
1053 * Encode the replay sequence operation from the slot values.
1054 * If cachethis is FALSE encode the uncached rep error on the next
1055 * operation which sets resp->p and increments resp->opcnt for
1056 * nfs4svc_encode_compoundres.
1060 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1061 struct nfsd4_compoundres *resp)
1063 struct nfsd4_op *op;
1064 struct nfsd4_slot *slot = resp->cstate.slot;
1066 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1067 resp->opcnt, resp->cstate.slot->sl_cachethis);
1069 /* Encode the replayed sequence operation */
1070 op = &args->ops[resp->opcnt - 1];
1071 nfsd4_encode_operation(resp, op);
1073 /* Return nfserr_retry_uncached_rep in next operation. */
1074 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1075 op = &args->ops[resp->opcnt++];
1076 op->status = nfserr_retry_uncached_rep;
1077 nfsd4_encode_operation(resp, op);
1083 * The sequence operation is not cached because we can use the slot and
1087 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1088 struct nfsd4_sequence *seq)
1090 struct nfsd4_slot *slot = resp->cstate.slot;
1093 dprintk("--> %s slot %p\n", __func__, slot);
1095 /* Either returns 0 or nfserr_retry_uncached */
1096 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1097 if (status == nfserr_retry_uncached_rep)
1100 /* The sequence operation has been encoded, cstate->datap set. */
1101 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1103 resp->opcnt = slot->sl_opcnt;
1104 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1105 status = slot->sl_status;
1111 * Set the exchange_id flags returned by the server.
1114 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1116 /* pNFS is not supported */
1117 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1119 /* Referrals are supported, Migration is not. */
1120 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1122 /* set the wire flags to return to client. */
1123 clid->flags = new->cl_exchange_flags;
1127 nfsd4_exchange_id(struct svc_rqst *rqstp,
1128 struct nfsd4_compound_state *cstate,
1129 struct nfsd4_exchange_id *exid)
1131 struct nfs4_client *unconf, *conf, *new;
1133 unsigned int strhashval;
1134 char dname[HEXDIR_LEN];
1135 char addr_str[INET6_ADDRSTRLEN];
1136 nfs4_verifier verf = exid->verifier;
1137 struct sockaddr *sa = svc_addr(rqstp);
1139 rpc_ntop(sa, addr_str, sizeof(addr_str));
1140 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1141 "ip_addr=%s flags %x, spa_how %d\n",
1142 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1143 addr_str, exid->flags, exid->spa_how);
1145 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1146 return nfserr_inval;
1148 /* Currently only support SP4_NONE */
1149 switch (exid->spa_how) {
1153 return nfserr_encr_alg_unsupp;
1155 BUG(); /* checked by xdr code */
1157 return nfserr_serverfault; /* no excuse :-/ */
1160 status = nfs4_make_rec_clidname(dname, &exid->clname);
1165 strhashval = clientstr_hashval(dname);
1170 conf = find_confirmed_client_by_str(dname, strhashval, true);
1172 if (!same_verf(&verf, &conf->cl_verifier)) {
1173 /* 18.35.4 case 8 */
1174 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1175 status = nfserr_not_same;
1178 /* Client reboot: destroy old state */
1179 expire_client(conf);
1182 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1183 /* 18.35.4 case 9 */
1184 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1185 status = nfserr_perm;
1188 expire_client(conf);
1192 * Set bit when the owner id and verifier map to an already
1193 * confirmed client id (18.35.3).
1195 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1198 * Falling into 18.35.4 case 2, possible router replay.
1199 * Leave confirmed record intact and return same result.
1201 copy_verf(conf, &verf);
1206 /* 18.35.4 case 7 */
1207 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1208 status = nfserr_noent;
1212 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1215 * Possible retry or client restart. Per 18.35.4 case 4,
1216 * a new unconfirmed record should be generated regardless
1217 * of whether any properties have changed.
1219 expire_client(unconf);
1224 new = create_client(exid->clname, dname, rqstp, &verf);
1226 status = nfserr_jukebox;
1231 add_to_unconfirmed(new, strhashval);
1233 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1234 exid->clientid.cl_id = new->cl_clientid.cl_id;
1237 nfsd4_set_ex_flags(new, exid);
1239 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1240 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1244 nfs4_unlock_state();
1246 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1251 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1253 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1256 /* The slot is in use, and no response has been sent. */
1258 if (seqid == slot_seqid)
1259 return nfserr_jukebox;
1261 return nfserr_seq_misordered;
1264 if (likely(seqid == slot_seqid + 1))
1267 if (seqid == slot_seqid)
1268 return nfserr_replay_cache;
1270 if (seqid == 1 && (slot_seqid + 1) == 0)
1272 /* Misordered replay or misordered new request */
1273 return nfserr_seq_misordered;
1277 * Cache the create session result into the create session single DRC
1278 * slot cache by saving the xdr structure. sl_seqid has been set.
1279 * Do this for solo or embedded create session operations.
1282 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1283 struct nfsd4_clid_slot *slot, int nfserr)
1285 slot->sl_status = nfserr;
1286 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1290 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1291 struct nfsd4_clid_slot *slot)
1293 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1294 return slot->sl_status;
1298 nfsd4_create_session(struct svc_rqst *rqstp,
1299 struct nfsd4_compound_state *cstate,
1300 struct nfsd4_create_session *cr_ses)
1302 struct sockaddr *sa = svc_addr(rqstp);
1303 struct nfs4_client *conf, *unconf;
1304 struct nfsd4_clid_slot *cs_slot = NULL;
1308 unconf = find_unconfirmed_client(&cr_ses->clientid);
1309 conf = find_confirmed_client(&cr_ses->clientid);
1312 cs_slot = &conf->cl_cs_slot;
1313 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1314 if (status == nfserr_replay_cache) {
1315 dprintk("Got a create_session replay! seqid= %d\n",
1317 /* Return the cached reply status */
1318 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1320 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1321 status = nfserr_seq_misordered;
1322 dprintk("Sequence misordered!\n");
1323 dprintk("Expected seqid= %d but got seqid= %d\n",
1324 cs_slot->sl_seqid, cr_ses->seqid);
1327 cs_slot->sl_seqid++;
1328 } else if (unconf) {
1329 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1330 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1331 status = nfserr_clid_inuse;
1335 cs_slot = &unconf->cl_cs_slot;
1336 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1338 /* an unconfirmed replay returns misordered */
1339 status = nfserr_seq_misordered;
1343 cs_slot->sl_seqid++; /* from 0 to 1 */
1344 move_to_confirmed(unconf);
1346 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1347 unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1348 svc_xprt_get(rqstp->rq_xprt);
1350 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1352 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1353 unconf->cl_cb_conn.cb_minorversion =
1354 cstate->minorversion;
1355 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1356 unconf->cl_cb_seq_nr = 1;
1357 nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1361 status = nfserr_stale_clientid;
1366 * We do not support RDMA or persistent sessions
1368 cr_ses->flags &= ~SESSION4_PERSIST;
1369 cr_ses->flags &= ~SESSION4_RDMA;
1371 status = alloc_init_session(rqstp, conf, cr_ses);
1375 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1376 NFS4_MAX_SESSIONID_LEN);
1377 cr_ses->seqid = cs_slot->sl_seqid;
1380 /* cache solo and embedded create sessions under the state lock */
1381 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1383 nfs4_unlock_state();
1384 dprintk("%s returns %d\n", __func__, ntohl(status));
1388 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1390 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1391 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1393 return argp->opcnt == resp->opcnt;
1396 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1400 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1404 nfsd4_destroy_session(struct svc_rqst *r,
1405 struct nfsd4_compound_state *cstate,
1406 struct nfsd4_destroy_session *sessionid)
1408 struct nfsd4_session *ses;
1409 u32 status = nfserr_badsession;
1412 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1413 * - Should we return nfserr_back_chan_busy if waiting for
1414 * callbacks on to-be-destroyed session?
1415 * - Do we need to clear any callback info from previous session?
1418 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1419 if (!nfsd4_last_compound_op(r))
1420 return nfserr_not_only_op;
1422 dump_sessionid(__func__, &sessionid->sessionid);
1423 spin_lock(&client_lock);
1424 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1426 spin_unlock(&client_lock);
1430 unhash_session(ses);
1431 spin_unlock(&client_lock);
1434 /* wait for callbacks */
1435 nfsd4_set_callback_client(ses->se_client, NULL);
1436 nfs4_unlock_state();
1437 nfsd4_put_session(ses);
1440 dprintk("%s returns %d\n", __func__, ntohl(status));
1445 nfsd4_sequence(struct svc_rqst *rqstp,
1446 struct nfsd4_compound_state *cstate,
1447 struct nfsd4_sequence *seq)
1449 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1450 struct nfsd4_session *session;
1451 struct nfsd4_slot *slot;
1454 if (resp->opcnt != 1)
1455 return nfserr_sequence_pos;
1457 spin_lock(&client_lock);
1458 status = nfserr_badsession;
1459 session = find_in_sessionid_hashtbl(&seq->sessionid);
1463 status = nfserr_badslot;
1464 if (seq->slotid >= session->se_fchannel.maxreqs)
1467 slot = session->se_slots[seq->slotid];
1468 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1470 /* We do not negotiate the number of slots yet, so set the
1471 * maxslots to the session maxreqs which is used to encode
1472 * sr_highest_slotid and the sr_target_slot id to maxslots */
1473 seq->maxslots = session->se_fchannel.maxreqs;
1475 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1476 if (status == nfserr_replay_cache) {
1477 cstate->slot = slot;
1478 cstate->session = session;
1479 /* Return the cached reply status and set cstate->status
1480 * for nfsd4_proc_compound processing */
1481 status = nfsd4_replay_cache_entry(resp, seq);
1482 cstate->status = nfserr_replay_cache;
1488 /* Success! bump slot seqid */
1489 slot->sl_inuse = true;
1490 slot->sl_seqid = seq->seqid;
1491 slot->sl_cachethis = seq->cachethis;
1493 cstate->slot = slot;
1494 cstate->session = session;
1497 /* Hold a session reference until done processing the compound. */
1498 if (cstate->session) {
1499 nfsd4_get_session(cstate->session);
1500 atomic_inc(&session->se_client->cl_refcount);
1502 spin_unlock(&client_lock);
1503 dprintk("%s: return %d\n", __func__, ntohl(status));
1508 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1510 if (rc->rca_one_fs) {
1511 if (!cstate->current_fh.fh_dentry)
1512 return nfserr_nofilehandle;
1514 * We don't take advantage of the rca_one_fs case.
1515 * That's OK, it's optional, we can safely ignore it.
1520 if (is_client_expired(cstate->session->se_client)) {
1521 nfs4_unlock_state();
1523 * The following error isn't really legal.
1524 * But we only get here if the client just explicitly
1525 * destroyed the client. Surely it no longer cares what
1526 * error it gets back on an operation for the dead
1529 return nfserr_stale_clientid;
1531 nfsd4_create_clid_dir(cstate->session->se_client);
1532 nfs4_unlock_state();
1537 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1538 struct nfsd4_setclientid *setclid)
1540 struct sockaddr *sa = svc_addr(rqstp);
1541 struct xdr_netobj clname = {
1542 .len = setclid->se_namelen,
1543 .data = setclid->se_name,
1545 nfs4_verifier clverifier = setclid->se_verf;
1546 unsigned int strhashval;
1547 struct nfs4_client *conf, *unconf, *new;
1549 char dname[HEXDIR_LEN];
1551 if (!check_name(clname))
1552 return nfserr_inval;
1554 status = nfs4_make_rec_clidname(dname, &clname);
1559 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1560 * We get here on a DRC miss.
1563 strhashval = clientstr_hashval(dname);
1566 conf = find_confirmed_client_by_str(dname, strhashval, false);
1568 /* RFC 3530 14.2.33 CASE 0: */
1569 status = nfserr_clid_inuse;
1570 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1571 char addr_str[INET6_ADDRSTRLEN];
1572 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1574 dprintk("NFSD: setclientid: string in use by client "
1575 "at %s\n", addr_str);
1580 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1581 * has a description of SETCLIENTID request processing consisting
1582 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1584 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1585 status = nfserr_resource;
1588 * RFC 3530 14.2.33 CASE 4:
1589 * placed first, because it is the normal case
1592 expire_client(unconf);
1593 new = create_client(clname, dname, rqstp, &clverifier);
1597 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1599 * RFC 3530 14.2.33 CASE 1:
1600 * probable callback update
1603 /* Note this is removing unconfirmed {*x***},
1604 * which is stronger than RFC recommended {vxc**}.
1605 * This has the advantage that there is at most
1606 * one {*x***} in either list at any time.
1608 expire_client(unconf);
1610 new = create_client(clname, dname, rqstp, &clverifier);
1613 copy_clid(new, conf);
1614 } else if (!unconf) {
1616 * RFC 3530 14.2.33 CASE 2:
1617 * probable client reboot; state will be removed if
1620 new = create_client(clname, dname, rqstp, &clverifier);
1626 * RFC 3530 14.2.33 CASE 3:
1627 * probable client reboot; state will be removed if
1630 expire_client(unconf);
1631 new = create_client(clname, dname, rqstp, &clverifier);
1636 gen_callback(new, setclid, rpc_get_scope_id(sa));
1637 add_to_unconfirmed(new, strhashval);
1638 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1639 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1640 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1643 nfs4_unlock_state();
1649 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1650 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1651 * bullets, labeled as CASE1 - CASE4 below.
1654 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1655 struct nfsd4_compound_state *cstate,
1656 struct nfsd4_setclientid_confirm *setclientid_confirm)
1658 struct sockaddr *sa = svc_addr(rqstp);
1659 struct nfs4_client *conf, *unconf;
1660 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1661 clientid_t * clid = &setclientid_confirm->sc_clientid;
1664 if (STALE_CLIENTID(clid))
1665 return nfserr_stale_clientid;
1667 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1668 * We get here on a DRC miss.
1673 conf = find_confirmed_client(clid);
1674 unconf = find_unconfirmed_client(clid);
1676 status = nfserr_clid_inuse;
1677 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1679 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1683 * section 14.2.34 of RFC 3530 has a description of
1684 * SETCLIENTID_CONFIRM request processing consisting
1685 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1687 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1689 * RFC 3530 14.2.34 CASE 1:
1692 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1693 status = nfserr_clid_inuse;
1695 atomic_set(&conf->cl_cb_set, 0);
1696 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1697 expire_client(unconf);
1701 } else if (conf && !unconf) {
1703 * RFC 3530 14.2.34 CASE 2:
1704 * probable retransmitted request; play it safe and
1707 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1708 status = nfserr_clid_inuse;
1711 } else if (!conf && unconf
1712 && same_verf(&unconf->cl_confirm, &confirm)) {
1714 * RFC 3530 14.2.34 CASE 3:
1715 * Normal case; new or rebooted client:
1717 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1718 status = nfserr_clid_inuse;
1721 clientstr_hashval(unconf->cl_recdir);
1722 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1725 nfsd4_remove_clid_dir(conf);
1726 expire_client(conf);
1728 move_to_confirmed(unconf);
1730 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1733 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1734 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1737 * RFC 3530 14.2.34 CASE 4:
1738 * Client probably hasn't noticed that we rebooted yet.
1740 status = nfserr_stale_clientid;
1742 /* check that we have hit one of the cases...*/
1743 status = nfserr_clid_inuse;
1746 nfs4_unlock_state();
1750 /* OPEN Share state helper functions */
1751 static inline struct nfs4_file *
1752 alloc_init_file(struct inode *ino)
1754 struct nfs4_file *fp;
1755 unsigned int hashval = file_hashval(ino);
1757 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1759 atomic_set(&fp->fi_ref, 1);
1760 INIT_LIST_HEAD(&fp->fi_hash);
1761 INIT_LIST_HEAD(&fp->fi_stateids);
1762 INIT_LIST_HEAD(&fp->fi_delegations);
1763 fp->fi_inode = igrab(ino);
1764 fp->fi_id = current_fileid++;
1765 fp->fi_had_conflict = false;
1766 spin_lock(&recall_lock);
1767 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1768 spin_unlock(&recall_lock);
1775 nfsd4_free_slab(struct kmem_cache **slab)
1779 kmem_cache_destroy(*slab);
1784 nfsd4_free_slabs(void)
1786 nfsd4_free_slab(&stateowner_slab);
1787 nfsd4_free_slab(&file_slab);
1788 nfsd4_free_slab(&stateid_slab);
1789 nfsd4_free_slab(&deleg_slab);
1793 nfsd4_init_slabs(void)
1795 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1796 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1797 if (stateowner_slab == NULL)
1799 file_slab = kmem_cache_create("nfsd4_files",
1800 sizeof(struct nfs4_file), 0, 0, NULL);
1801 if (file_slab == NULL)
1803 stateid_slab = kmem_cache_create("nfsd4_stateids",
1804 sizeof(struct nfs4_stateid), 0, 0, NULL);
1805 if (stateid_slab == NULL)
1807 deleg_slab = kmem_cache_create("nfsd4_delegations",
1808 sizeof(struct nfs4_delegation), 0, 0, NULL);
1809 if (deleg_slab == NULL)
1814 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1819 nfs4_free_stateowner(struct kref *kref)
1821 struct nfs4_stateowner *sop =
1822 container_of(kref, struct nfs4_stateowner, so_ref);
1823 kfree(sop->so_owner.data);
1824 kmem_cache_free(stateowner_slab, sop);
1827 static inline struct nfs4_stateowner *
1828 alloc_stateowner(struct xdr_netobj *owner)
1830 struct nfs4_stateowner *sop;
1832 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1833 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1834 memcpy(sop->so_owner.data, owner->data, owner->len);
1835 sop->so_owner.len = owner->len;
1836 kref_init(&sop->so_ref);
1839 kmem_cache_free(stateowner_slab, sop);
1844 static struct nfs4_stateowner *
1845 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1846 struct nfs4_stateowner *sop;
1847 struct nfs4_replay *rp;
1848 unsigned int idhashval;
1850 if (!(sop = alloc_stateowner(&open->op_owner)))
1852 idhashval = ownerid_hashval(current_ownerid);
1853 INIT_LIST_HEAD(&sop->so_idhash);
1854 INIT_LIST_HEAD(&sop->so_strhash);
1855 INIT_LIST_HEAD(&sop->so_perclient);
1856 INIT_LIST_HEAD(&sop->so_stateids);
1857 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1858 INIT_LIST_HEAD(&sop->so_close_lru);
1860 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1861 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1862 list_add(&sop->so_perclient, &clp->cl_openowners);
1863 sop->so_is_open_owner = 1;
1864 sop->so_id = current_ownerid++;
1865 sop->so_client = clp;
1866 sop->so_seqid = open->op_seqid;
1867 sop->so_confirmed = 0;
1868 rp = &sop->so_replay;
1869 rp->rp_status = nfserr_serverfault;
1871 rp->rp_buf = rp->rp_ibuf;
1876 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1877 struct nfs4_stateowner *sop = open->op_stateowner;
1878 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1880 INIT_LIST_HEAD(&stp->st_hash);
1881 INIT_LIST_HEAD(&stp->st_perstateowner);
1882 INIT_LIST_HEAD(&stp->st_lockowners);
1883 INIT_LIST_HEAD(&stp->st_perfile);
1884 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1885 list_add(&stp->st_perstateowner, &sop->so_stateids);
1886 list_add(&stp->st_perfile, &fp->fi_stateids);
1887 stp->st_stateowner = sop;
1890 stp->st_stateid.si_boot = boot_time;
1891 stp->st_stateid.si_stateownerid = sop->so_id;
1892 stp->st_stateid.si_fileid = fp->fi_id;
1893 stp->st_stateid.si_generation = 0;
1894 stp->st_access_bmap = 0;
1895 stp->st_deny_bmap = 0;
1896 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1897 &stp->st_access_bmap);
1898 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1899 stp->st_openstp = NULL;
1903 move_to_close_lru(struct nfs4_stateowner *sop)
1905 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1907 list_move_tail(&sop->so_close_lru, &close_lru);
1908 sop->so_time = get_seconds();
1912 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1915 return (sop->so_owner.len == owner->len) &&
1916 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1917 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1920 static struct nfs4_stateowner *
1921 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1923 struct nfs4_stateowner *so = NULL;
1925 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1926 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1932 /* search file_hashtbl[] for file */
1933 static struct nfs4_file *
1934 find_file(struct inode *ino)
1936 unsigned int hashval = file_hashval(ino);
1937 struct nfs4_file *fp;
1939 spin_lock(&recall_lock);
1940 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1941 if (fp->fi_inode == ino) {
1943 spin_unlock(&recall_lock);
1947 spin_unlock(&recall_lock);
1951 static inline int access_valid(u32 x, u32 minorversion)
1953 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1955 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1957 x &= ~NFS4_SHARE_ACCESS_MASK;
1958 if (minorversion && x) {
1959 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1961 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1963 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1970 static inline int deny_valid(u32 x)
1972 /* Note: unlike access bits, deny bits may be zero. */
1973 return x <= NFS4_SHARE_DENY_BOTH;
1977 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1978 * st_{access,deny}_bmap field of the stateid, in order to track not
1979 * only what share bits are currently in force, but also what
1980 * combinations of share bits previous opens have used. This allows us
1981 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1982 * return an error if the client attempt to downgrade to a combination
1983 * of share bits not explicable by closing some of its previous opens.
1985 * XXX: This enforcement is actually incomplete, since we don't keep
1986 * track of access/deny bit combinations; so, e.g., we allow:
1988 * OPEN allow read, deny write
1989 * OPEN allow both, deny none
1990 * DOWNGRADE allow read, deny none
1992 * which we should reject.
1995 set_access(unsigned int *access, unsigned long bmap) {
1999 for (i = 1; i < 4; i++) {
2000 if (test_bit(i, &bmap))
2006 set_deny(unsigned int *deny, unsigned long bmap) {
2010 for (i = 0; i < 4; i++) {
2011 if (test_bit(i, &bmap))
2017 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2018 unsigned int access, deny;
2020 set_access(&access, stp->st_access_bmap);
2021 set_deny(&deny, stp->st_deny_bmap);
2022 if ((access & open->op_share_deny) || (deny & open->op_share_access))
2028 * Called to check deny when READ with all zero stateid or
2029 * WRITE with all zero or all one stateid
2032 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2034 struct inode *ino = current_fh->fh_dentry->d_inode;
2035 struct nfs4_file *fp;
2036 struct nfs4_stateid *stp;
2039 dprintk("NFSD: nfs4_share_conflict\n");
2041 fp = find_file(ino);
2044 ret = nfserr_locked;
2045 /* Search for conflicting share reservations */
2046 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2047 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2048 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2058 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2060 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2061 drop_file_write_access(filp);
2062 spin_lock(&filp->f_lock);
2063 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2064 spin_unlock(&filp->f_lock);
2069 * Spawn a thread to perform a recall on the delegation represented
2070 * by the lease (file_lock)
2072 * Called from break_lease() with lock_kernel() held.
2073 * Note: we assume break_lease will only call this *once* for any given
2077 void nfsd_break_deleg_cb(struct file_lock *fl)
2079 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2081 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2085 /* We're assuming the state code never drops its reference
2086 * without first removing the lease. Since we're in this lease
2087 * callback (and since the lease code is serialized by the kernel
2088 * lock) we know the server hasn't removed the lease yet, we know
2089 * it's safe to take a reference: */
2090 atomic_inc(&dp->dl_count);
2092 spin_lock(&recall_lock);
2093 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2094 spin_unlock(&recall_lock);
2096 /* only place dl_time is set. protected by lock_kernel*/
2097 dp->dl_time = get_seconds();
2100 * We don't want the locks code to timeout the lease for us;
2101 * we'll remove it ourself if the delegation isn't returned
2104 fl->fl_break_time = 0;
2106 dp->dl_file->fi_had_conflict = true;
2107 nfsd4_cb_recall(dp);
2111 * The file_lock is being reapd.
2113 * Called by locks_free_lock() with lock_kernel() held.
2116 void nfsd_release_deleg_cb(struct file_lock *fl)
2118 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2120 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2122 if (!(fl->fl_flags & FL_LEASE) || !dp)
2124 dp->dl_flock = NULL;
2128 * Set the delegation file_lock back pointer.
2130 * Called from setlease() with lock_kernel() held.
2133 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2135 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2137 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2144 * Called from setlease() with lock_kernel() held
2147 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2149 struct nfs4_delegation *onlistd =
2150 (struct nfs4_delegation *)onlist->fl_owner;
2151 struct nfs4_delegation *tryd =
2152 (struct nfs4_delegation *)try->fl_owner;
2154 if (onlist->fl_lmops != try->fl_lmops)
2157 return onlistd->dl_client == tryd->dl_client;
2162 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2165 return lease_modify(onlist, arg);
2170 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2171 .fl_break = nfsd_break_deleg_cb,
2172 .fl_release_private = nfsd_release_deleg_cb,
2173 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2174 .fl_mylease = nfsd_same_client_deleg_cb,
2175 .fl_change = nfsd_change_deleg_cb,
2180 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2181 struct nfsd4_open *open)
2183 clientid_t *clientid = &open->op_clientid;
2184 struct nfs4_client *clp = NULL;
2185 unsigned int strhashval;
2186 struct nfs4_stateowner *sop = NULL;
2188 if (!check_name(open->op_owner))
2189 return nfserr_inval;
2191 if (STALE_CLIENTID(&open->op_clientid))
2192 return nfserr_stale_clientid;
2194 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2195 sop = find_openstateowner_str(strhashval, open);
2196 open->op_stateowner = sop;
2198 /* Make sure the client's lease hasn't expired. */
2199 clp = find_confirmed_client(clientid);
2201 return nfserr_expired;
2204 /* When sessions are used, skip open sequenceid processing */
2205 if (nfsd4_has_session(cstate))
2207 if (!sop->so_confirmed) {
2208 /* Replace unconfirmed owners without checking for replay. */
2209 clp = sop->so_client;
2210 release_openowner(sop);
2211 open->op_stateowner = NULL;
2214 if (open->op_seqid == sop->so_seqid - 1) {
2215 if (sop->so_replay.rp_buflen)
2216 return nfserr_replay_me;
2217 /* The original OPEN failed so spectacularly
2218 * that we don't even have replay data saved!
2219 * Therefore, we have no choice but to continue
2220 * processing this OPEN; presumably, we'll
2221 * fail again for the same reason.
2223 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2226 if (open->op_seqid != sop->so_seqid)
2227 return nfserr_bad_seqid;
2229 if (open->op_stateowner == NULL) {
2230 sop = alloc_init_open_stateowner(strhashval, clp, open);
2232 return nfserr_resource;
2233 open->op_stateowner = sop;
2235 list_del_init(&sop->so_close_lru);
2236 renew_client(sop->so_client);
2240 static inline __be32
2241 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2243 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2244 return nfserr_openmode;
2249 static struct nfs4_delegation *
2250 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2252 struct nfs4_delegation *dp;
2254 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2255 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2261 int share_access_to_flags(u32 share_access)
2263 share_access &= ~NFS4_SHARE_WANT_MASK;
2265 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2269 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2270 struct nfs4_delegation **dp)
2273 __be32 status = nfserr_bad_stateid;
2275 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2278 flags = share_access_to_flags(open->op_share_access);
2279 status = nfs4_check_delegmode(*dp, flags);
2283 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2287 open->op_stateowner->so_confirmed = 1;
2292 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2294 struct nfs4_stateid *local;
2295 __be32 status = nfserr_share_denied;
2296 struct nfs4_stateowner *sop = open->op_stateowner;
2298 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2299 /* ignore lock owners */
2300 if (local->st_stateowner->so_is_open_owner == 0)
2302 /* remember if we have seen this open owner */
2303 if (local->st_stateowner == sop)
2305 /* check for conflicting share reservations */
2306 if (!test_share(local, open))
2314 static inline struct nfs4_stateid *
2315 nfs4_alloc_stateid(void)
2317 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2320 static inline int nfs4_access_to_access(u32 nfs4_access)
2324 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2325 flags |= NFSD_MAY_READ;
2326 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2327 flags |= NFSD_MAY_WRITE;
2332 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2333 struct nfs4_delegation *dp,
2334 struct svc_fh *cur_fh, struct nfsd4_open *open)
2336 struct nfs4_stateid *stp;
2338 stp = nfs4_alloc_stateid();
2340 return nfserr_resource;
2343 get_file(dp->dl_vfs_file);
2344 stp->st_vfs_file = dp->dl_vfs_file;
2347 int access = nfs4_access_to_access(open->op_share_access);
2349 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2352 if (status == nfserr_dropit)
2353 status = nfserr_jukebox;
2354 kmem_cache_free(stateid_slab, stp);
2362 static inline __be32
2363 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2364 struct nfsd4_open *open)
2366 struct iattr iattr = {
2367 .ia_valid = ATTR_SIZE,
2370 if (!open->op_truncate)
2372 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2373 return nfserr_inval;
2374 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2378 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2380 struct file *filp = stp->st_vfs_file;
2381 struct inode *inode = filp->f_path.dentry->d_inode;
2382 unsigned int share_access, new_writer;
2383 u32 op_share_access;
2386 set_access(&share_access, stp->st_access_bmap);
2387 new_writer = (~share_access) & open->op_share_access
2388 & NFS4_SHARE_ACCESS_WRITE;
2391 int err = get_write_access(inode);
2393 return nfserrno(err);
2394 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2396 return nfserrno(err);
2397 file_take_write(filp);
2399 status = nfsd4_truncate(rqstp, cur_fh, open);
2402 put_write_access(inode);
2405 /* remember the open */
2406 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2407 filp->f_mode |= op_share_access;
2408 __set_bit(op_share_access, &stp->st_access_bmap);
2409 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2416 nfs4_set_claim_prev(struct nfsd4_open *open)
2418 open->op_stateowner->so_confirmed = 1;
2419 open->op_stateowner->so_client->cl_firststate = 1;
2423 * Attempt to hand out a delegation.
2426 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2428 struct nfs4_delegation *dp;
2429 struct nfs4_stateowner *sop = stp->st_stateowner;
2430 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2431 struct file_lock fl, *flp = &fl;
2432 int status, flag = 0;
2434 flag = NFS4_OPEN_DELEGATE_NONE;
2435 open->op_recall = 0;
2436 switch (open->op_claim_type) {
2437 case NFS4_OPEN_CLAIM_PREVIOUS:
2439 open->op_recall = 1;
2440 flag = open->op_delegate_type;
2441 if (flag == NFS4_OPEN_DELEGATE_NONE)
2444 case NFS4_OPEN_CLAIM_NULL:
2445 /* Let's not give out any delegations till everyone's
2446 * had the chance to reclaim theirs.... */
2447 if (locks_in_grace())
2449 if (!cb_up || !sop->so_confirmed)
2451 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2452 flag = NFS4_OPEN_DELEGATE_WRITE;
2454 flag = NFS4_OPEN_DELEGATE_READ;
2460 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2462 flag = NFS4_OPEN_DELEGATE_NONE;
2465 locks_init_lock(&fl);
2466 fl.fl_lmops = &nfsd_lease_mng_ops;
2467 fl.fl_flags = FL_LEASE;
2468 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2469 fl.fl_end = OFFSET_MAX;
2470 fl.fl_owner = (fl_owner_t)dp;
2471 fl.fl_file = stp->st_vfs_file;
2472 fl.fl_pid = current->tgid;
2474 /* vfs_setlease checks to see if delegation should be handed out.
2475 * the lock_manager callbacks fl_mylease and fl_change are used
2477 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2478 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2479 unhash_delegation(dp);
2480 flag = NFS4_OPEN_DELEGATE_NONE;
2484 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2486 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2487 STATEID_VAL(&dp->dl_stateid));
2489 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2490 && flag == NFS4_OPEN_DELEGATE_NONE
2491 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2492 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2493 open->op_delegate_type = flag;
2497 * called with nfs4_lock_state() held.
2500 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2502 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2503 struct nfs4_file *fp = NULL;
2504 struct inode *ino = current_fh->fh_dentry->d_inode;
2505 struct nfs4_stateid *stp = NULL;
2506 struct nfs4_delegation *dp = NULL;
2509 status = nfserr_inval;
2510 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2511 || !deny_valid(open->op_share_deny))
2514 * Lookup file; if found, lookup stateid and check open request,
2515 * and check for delegations in the process of being recalled.
2516 * If not found, create the nfs4_file struct
2518 fp = find_file(ino);
2520 if ((status = nfs4_check_open(fp, open, &stp)))
2522 status = nfs4_check_deleg(fp, open, &dp);
2526 status = nfserr_bad_stateid;
2527 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2529 status = nfserr_resource;
2530 fp = alloc_init_file(ino);
2536 * OPEN the file, or upgrade an existing OPEN.
2537 * If truncate fails, the OPEN fails.
2540 /* Stateid was found, this is an OPEN upgrade */
2541 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2544 update_stateid(&stp->st_stateid);
2546 /* Stateid was not found, this is a new OPEN */
2547 status = nfs4_new_open(rqstp, &stp, dp, current_fh, open);
2550 init_stateid(stp, fp, open);
2551 status = nfsd4_truncate(rqstp, current_fh, open);
2553 release_open_stateid(stp);
2556 if (nfsd4_has_session(&resp->cstate))
2557 update_stateid(&stp->st_stateid);
2559 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2561 if (nfsd4_has_session(&resp->cstate))
2562 open->op_stateowner->so_confirmed = 1;
2565 * Attempt to hand out a delegation. No error return, because the
2566 * OPEN succeeds even if we fail.
2568 nfs4_open_delegation(current_fh, open, stp);
2572 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2573 STATEID_VAL(&stp->st_stateid));
2577 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2578 nfs4_set_claim_prev(open);
2580 * To finish the open response, we just need to set the rflags.
2582 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2583 if (!open->op_stateowner->so_confirmed &&
2584 !nfsd4_has_session(&resp->cstate))
2585 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2591 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2594 struct nfs4_client *clp;
2598 dprintk("process_renew(%08x/%08x): starting\n",
2599 clid->cl_boot, clid->cl_id);
2600 status = nfserr_stale_clientid;
2601 if (STALE_CLIENTID(clid))
2603 clp = find_confirmed_client(clid);
2604 status = nfserr_expired;
2606 /* We assume the client took too long to RENEW. */
2607 dprintk("nfsd4_renew: clientid not found!\n");
2611 status = nfserr_cb_path_down;
2612 if (!list_empty(&clp->cl_delegations)
2613 && !atomic_read(&clp->cl_cb_set))
2617 nfs4_unlock_state();
2621 struct lock_manager nfsd4_manager = {
2625 nfsd4_end_grace(void)
2627 dprintk("NFSD: end of grace period\n");
2628 nfsd4_recdir_purge_old();
2629 locks_end_grace(&nfsd4_manager);
2631 * Now that every NFSv4 client has had the chance to recover and
2632 * to see the (possibly new, possibly shorter) lease time, we
2633 * can safely set the next grace time to the current lease time:
2635 nfsd4_grace = nfsd4_lease;
2639 nfs4_laundromat(void)
2641 struct nfs4_client *clp;
2642 struct nfs4_stateowner *sop;
2643 struct nfs4_delegation *dp;
2644 struct list_head *pos, *next, reaplist;
2645 time_t cutoff = get_seconds() - nfsd4_lease;
2646 time_t t, clientid_val = nfsd4_lease;
2647 time_t u, test_val = nfsd4_lease;
2651 dprintk("NFSD: laundromat service - starting\n");
2652 if (locks_in_grace())
2654 INIT_LIST_HEAD(&reaplist);
2655 spin_lock(&client_lock);
2656 list_for_each_safe(pos, next, &client_lru) {
2657 clp = list_entry(pos, struct nfs4_client, cl_lru);
2658 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2659 t = clp->cl_time - cutoff;
2660 if (clientid_val > t)
2664 if (atomic_read(&clp->cl_refcount)) {
2665 dprintk("NFSD: client in use (clientid %08x)\n",
2666 clp->cl_clientid.cl_id);
2669 unhash_client_locked(clp);
2670 list_add(&clp->cl_lru, &reaplist);
2672 spin_unlock(&client_lock);
2673 list_for_each_safe(pos, next, &reaplist) {
2674 clp = list_entry(pos, struct nfs4_client, cl_lru);
2675 dprintk("NFSD: purging unused client (clientid %08x)\n",
2676 clp->cl_clientid.cl_id);
2677 nfsd4_remove_clid_dir(clp);
2680 spin_lock(&recall_lock);
2681 list_for_each_safe(pos, next, &del_recall_lru) {
2682 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2683 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2684 u = dp->dl_time - cutoff;
2689 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2691 list_move(&dp->dl_recall_lru, &reaplist);
2693 spin_unlock(&recall_lock);
2694 list_for_each_safe(pos, next, &reaplist) {
2695 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2696 list_del_init(&dp->dl_recall_lru);
2697 unhash_delegation(dp);
2699 test_val = nfsd4_lease;
2700 list_for_each_safe(pos, next, &close_lru) {
2701 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2702 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2703 u = sop->so_time - cutoff;
2708 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2710 release_openowner(sop);
2712 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2713 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2714 nfs4_unlock_state();
2715 return clientid_val;
2718 static struct workqueue_struct *laundry_wq;
2719 static void laundromat_main(struct work_struct *);
2720 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2723 laundromat_main(struct work_struct *not_used)
2727 t = nfs4_laundromat();
2728 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2729 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2732 static struct nfs4_stateowner *
2733 search_close_lru(u32 st_id, int flags)
2735 struct nfs4_stateowner *local = NULL;
2737 if (flags & CLOSE_STATE) {
2738 list_for_each_entry(local, &close_lru, so_close_lru) {
2739 if (local->so_id == st_id)
2747 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2749 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2753 STALE_STATEID(stateid_t *stateid)
2755 if (stateid->si_boot == boot_time)
2757 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2758 STATEID_VAL(stateid));
2763 access_permit_read(unsigned long access_bmap)
2765 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2766 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2767 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2771 access_permit_write(unsigned long access_bmap)
2773 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2774 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2778 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2780 __be32 status = nfserr_openmode;
2782 /* For lock stateid's, we test the parent open, not the lock: */
2783 if (stp->st_openstp)
2784 stp = stp->st_openstp;
2785 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2787 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2794 static inline __be32
2795 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2797 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2799 else if (locks_in_grace()) {
2800 /* Answer in remaining cases depends on existance of
2801 * conflicting state; so we must wait out the grace period. */
2802 return nfserr_grace;
2803 } else if (flags & WR_STATE)
2804 return nfs4_share_conflict(current_fh,
2805 NFS4_SHARE_DENY_WRITE);
2806 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2807 return nfs4_share_conflict(current_fh,
2808 NFS4_SHARE_DENY_READ);
2812 * Allow READ/WRITE during grace period on recovered state only for files
2813 * that are not able to provide mandatory locking.
2816 grace_disallows_io(struct inode *inode)
2818 return locks_in_grace() && mandatory_lock(inode);
2821 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2824 * When sessions are used the stateid generation number is ignored
2827 if ((flags & HAS_SESSION) && in->si_generation == 0)
2830 /* If the client sends us a stateid from the future, it's buggy: */
2831 if (in->si_generation > ref->si_generation)
2832 return nfserr_bad_stateid;
2834 * The following, however, can happen. For example, if the
2835 * client sends an open and some IO at the same time, the open
2836 * may bump si_generation while the IO is still in flight.
2837 * Thanks to hard links and renames, the client never knows what
2838 * file an open will affect. So it could avoid that situation
2839 * only by serializing all opens and IO from the same open
2840 * owner. To recover from the old_stateid error, the client
2841 * will just have to retry the IO:
2843 if (in->si_generation < ref->si_generation)
2844 return nfserr_old_stateid;
2849 static int is_delegation_stateid(stateid_t *stateid)
2851 return stateid->si_fileid == 0;
2855 * Checks for stateid operations
2858 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2859 stateid_t *stateid, int flags, struct file **filpp)
2861 struct nfs4_stateid *stp = NULL;
2862 struct nfs4_delegation *dp = NULL;
2863 struct svc_fh *current_fh = &cstate->current_fh;
2864 struct inode *ino = current_fh->fh_dentry->d_inode;
2870 if (grace_disallows_io(ino))
2871 return nfserr_grace;
2873 if (nfsd4_has_session(cstate))
2874 flags |= HAS_SESSION;
2876 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2877 return check_special_stateids(current_fh, stateid, flags);
2879 status = nfserr_stale_stateid;
2880 if (STALE_STATEID(stateid))
2883 status = nfserr_bad_stateid;
2884 if (is_delegation_stateid(stateid)) {
2885 dp = find_delegation_stateid(ino, stateid);
2888 status = check_stateid_generation(stateid, &dp->dl_stateid,
2892 status = nfs4_check_delegmode(dp, flags);
2895 renew_client(dp->dl_client);
2897 *filpp = dp->dl_vfs_file;
2898 } else { /* open or lock stateid */
2899 stp = find_stateid(stateid, flags);
2902 if (nfs4_check_fh(current_fh, stp))
2904 if (!stp->st_stateowner->so_confirmed)
2906 status = check_stateid_generation(stateid, &stp->st_stateid,
2910 status = nfs4_check_openmode(stp, flags);
2913 renew_client(stp->st_stateowner->so_client);
2915 *filpp = stp->st_vfs_file;
2925 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2926 RD_STATE : WR_STATE;
2930 * Checks for sequence id mutating operations.
2933 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2934 stateid_t *stateid, int flags,
2935 struct nfs4_stateowner **sopp,
2936 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2938 struct nfs4_stateid *stp;
2939 struct nfs4_stateowner *sop;
2940 struct svc_fh *current_fh = &cstate->current_fh;
2943 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2944 seqid, STATEID_VAL(stateid));
2949 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2950 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2951 return nfserr_bad_stateid;
2954 if (STALE_STATEID(stateid))
2955 return nfserr_stale_stateid;
2957 if (nfsd4_has_session(cstate))
2958 flags |= HAS_SESSION;
2961 * We return BAD_STATEID if filehandle doesn't match stateid,
2962 * the confirmed flag is incorrecly set, or the generation
2963 * number is incorrect.
2965 stp = find_stateid(stateid, flags);
2968 * Also, we should make sure this isn't just the result of
2971 sop = search_close_lru(stateid->si_stateownerid, flags);
2973 return nfserr_bad_stateid;
2979 *sopp = sop = stp->st_stateowner;
2982 clientid_t *lockclid = &lock->v.new.clientid;
2983 struct nfs4_client *clp = sop->so_client;
2987 lkflg = setlkflg(lock->lk_type);
2989 if (lock->lk_is_new) {
2990 if (!sop->so_is_open_owner)
2991 return nfserr_bad_stateid;
2992 if (!(flags & HAS_SESSION) &&
2993 !same_clid(&clp->cl_clientid, lockclid))
2994 return nfserr_bad_stateid;
2995 /* stp is the open stateid */
2996 status = nfs4_check_openmode(stp, lkflg);
3000 /* stp is the lock stateid */
3001 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3007 if (nfs4_check_fh(current_fh, stp)) {
3008 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3009 return nfserr_bad_stateid;
3013 * We now validate the seqid and stateid generation numbers.
3014 * For the moment, we ignore the possibility of
3015 * generation number wraparound.
3017 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3020 if (sop->so_confirmed && flags & CONFIRM) {
3021 dprintk("NFSD: preprocess_seqid_op: expected"
3022 " unconfirmed stateowner!\n");
3023 return nfserr_bad_stateid;
3025 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3026 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3027 " confirmed yet!\n");
3028 return nfserr_bad_stateid;
3030 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3033 renew_client(sop->so_client);
3037 if (seqid == sop->so_seqid - 1) {
3038 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3039 /* indicate replay to calling function */
3040 return nfserr_replay_me;
3042 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3043 sop->so_seqid, seqid);
3045 return nfserr_bad_seqid;
3049 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3050 struct nfsd4_open_confirm *oc)
3053 struct nfs4_stateowner *sop;
3054 struct nfs4_stateid *stp;
3056 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3057 (int)cstate->current_fh.fh_dentry->d_name.len,
3058 cstate->current_fh.fh_dentry->d_name.name);
3060 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3066 if ((status = nfs4_preprocess_seqid_op(cstate,
3067 oc->oc_seqid, &oc->oc_req_stateid,
3068 CONFIRM | OPEN_STATE,
3069 &oc->oc_stateowner, &stp, NULL)))
3072 sop = oc->oc_stateowner;
3073 sop->so_confirmed = 1;
3074 update_stateid(&stp->st_stateid);
3075 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3076 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3077 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3079 nfsd4_create_clid_dir(sop->so_client);
3081 if (oc->oc_stateowner) {
3082 nfs4_get_stateowner(oc->oc_stateowner);
3083 cstate->replay_owner = oc->oc_stateowner;
3085 nfs4_unlock_state();
3091 * unset all bits in union bitmap (bmap) that
3092 * do not exist in share (from successful OPEN_DOWNGRADE)
3095 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3098 for (i = 1; i < 4; i++) {
3099 if ((i & access) != i)
3100 __clear_bit(i, bmap);
3105 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3108 for (i = 0; i < 4; i++) {
3109 if ((i & deny) != i)
3110 __clear_bit(i, bmap);
3115 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3116 struct nfsd4_compound_state *cstate,
3117 struct nfsd4_open_downgrade *od)
3120 struct nfs4_stateid *stp;
3121 unsigned int share_access;
3123 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3124 (int)cstate->current_fh.fh_dentry->d_name.len,
3125 cstate->current_fh.fh_dentry->d_name.name);
3127 if (!access_valid(od->od_share_access, cstate->minorversion)
3128 || !deny_valid(od->od_share_deny))
3129 return nfserr_inval;
3132 if ((status = nfs4_preprocess_seqid_op(cstate,
3136 &od->od_stateowner, &stp, NULL)))
3139 status = nfserr_inval;
3140 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3141 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3142 stp->st_access_bmap, od->od_share_access);
3145 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3146 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3147 stp->st_deny_bmap, od->od_share_deny);
3150 set_access(&share_access, stp->st_access_bmap);
3151 nfs4_file_downgrade(stp->st_vfs_file,
3152 share_access & ~od->od_share_access);
3154 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3155 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3157 update_stateid(&stp->st_stateid);
3158 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3161 if (od->od_stateowner) {
3162 nfs4_get_stateowner(od->od_stateowner);
3163 cstate->replay_owner = od->od_stateowner;
3165 nfs4_unlock_state();
3170 * nfs4_unlock_state() called after encode
3173 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3174 struct nfsd4_close *close)
3177 struct nfs4_stateid *stp;
3179 dprintk("NFSD: nfsd4_close on file %.*s\n",
3180 (int)cstate->current_fh.fh_dentry->d_name.len,
3181 cstate->current_fh.fh_dentry->d_name.name);
3184 /* check close_lru for replay */
3185 if ((status = nfs4_preprocess_seqid_op(cstate,
3188 OPEN_STATE | CLOSE_STATE,
3189 &close->cl_stateowner, &stp, NULL)))
3192 update_stateid(&stp->st_stateid);
3193 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3195 /* release_stateid() calls nfsd_close() if needed */
3196 release_open_stateid(stp);
3198 /* place unused nfs4_stateowners on so_close_lru list to be
3199 * released by the laundromat service after the lease period
3200 * to enable us to handle CLOSE replay
3202 if (list_empty(&close->cl_stateowner->so_stateids))
3203 move_to_close_lru(close->cl_stateowner);
3205 if (close->cl_stateowner) {
3206 nfs4_get_stateowner(close->cl_stateowner);
3207 cstate->replay_owner = close->cl_stateowner;
3209 nfs4_unlock_state();
3214 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3215 struct nfsd4_delegreturn *dr)
3217 struct nfs4_delegation *dp;
3218 stateid_t *stateid = &dr->dr_stateid;
3219 struct inode *inode;
3223 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3225 inode = cstate->current_fh.fh_dentry->d_inode;
3227 if (nfsd4_has_session(cstate))
3228 flags |= HAS_SESSION;
3230 status = nfserr_bad_stateid;
3231 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3233 status = nfserr_stale_stateid;
3234 if (STALE_STATEID(stateid))
3236 status = nfserr_bad_stateid;
3237 if (!is_delegation_stateid(stateid))
3239 dp = find_delegation_stateid(inode, stateid);
3242 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3245 renew_client(dp->dl_client);
3247 unhash_delegation(dp);
3249 nfs4_unlock_state();
3256 * Lock owner state (byte-range locks)
3258 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3259 #define LOCK_HASH_BITS 8
3260 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3261 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3264 end_offset(u64 start, u64 len)
3269 return end >= start ? end: NFS4_MAX_UINT64;
3272 /* last octet in a range */
3274 last_byte_offset(u64 start, u64 len)
3280 return end > start ? end - 1: NFS4_MAX_UINT64;
3283 #define lockownerid_hashval(id) \
3284 ((id) & LOCK_HASH_MASK)
3286 static inline unsigned int
3287 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3288 struct xdr_netobj *ownername)
3290 return (file_hashval(inode) + cl_id
3291 + opaque_hashval(ownername->data, ownername->len))
3295 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3296 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3297 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3299 static struct nfs4_stateid *
3300 find_stateid(stateid_t *stid, int flags)
3302 struct nfs4_stateid *local;
3303 u32 st_id = stid->si_stateownerid;
3304 u32 f_id = stid->si_fileid;
3305 unsigned int hashval;
3307 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3308 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3309 hashval = stateid_hashval(st_id, f_id);
3310 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3311 if ((local->st_stateid.si_stateownerid == st_id) &&
3312 (local->st_stateid.si_fileid == f_id))
3317 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3318 hashval = stateid_hashval(st_id, f_id);
3319 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3320 if ((local->st_stateid.si_stateownerid == st_id) &&
3321 (local->st_stateid.si_fileid == f_id))
3328 static struct nfs4_delegation *
3329 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3331 struct nfs4_file *fp;
3332 struct nfs4_delegation *dl;
3334 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3337 fp = find_file(ino);
3340 dl = find_delegation_file(fp, stid);
3346 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3347 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3348 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3349 * locking, this prevents us from being completely protocol-compliant. The
3350 * real solution to this problem is to start using unsigned file offsets in
3351 * the VFS, but this is a very deep change!
3354 nfs4_transform_lock_offset(struct file_lock *lock)
3356 if (lock->fl_start < 0)
3357 lock->fl_start = OFFSET_MAX;
3358 if (lock->fl_end < 0)
3359 lock->fl_end = OFFSET_MAX;
3362 /* Hack!: For now, we're defining this just so we can use a pointer to it
3363 * as a unique cookie to identify our (NFSv4's) posix locks. */
3364 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3368 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3370 struct nfs4_stateowner *sop;
3373 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3374 sop = (struct nfs4_stateowner *) fl->fl_owner;
3375 hval = lockownerid_hashval(sop->so_id);
3376 kref_get(&sop->so_ref);
3378 deny->ld_clientid = sop->so_client->cl_clientid;
3380 deny->ld_sop = NULL;
3381 deny->ld_clientid.cl_boot = 0;
3382 deny->ld_clientid.cl_id = 0;
3384 deny->ld_start = fl->fl_start;
3385 deny->ld_length = NFS4_MAX_UINT64;
3386 if (fl->fl_end != NFS4_MAX_UINT64)
3387 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3388 deny->ld_type = NFS4_READ_LT;
3389 if (fl->fl_type != F_RDLCK)
3390 deny->ld_type = NFS4_WRITE_LT;
3393 static struct nfs4_stateowner *
3394 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3395 struct xdr_netobj *owner)
3397 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3398 struct nfs4_stateowner *op;
3400 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3401 if (same_owner_str(op, owner, clid))
3408 * Alloc a lock owner structure.
3409 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3412 * strhashval = lock_ownerstr_hashval
3415 static struct nfs4_stateowner *
3416 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3417 struct nfs4_stateowner *sop;
3418 struct nfs4_replay *rp;
3419 unsigned int idhashval;
3421 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3423 idhashval = lockownerid_hashval(current_ownerid);
3424 INIT_LIST_HEAD(&sop->so_idhash);
3425 INIT_LIST_HEAD(&sop->so_strhash);
3426 INIT_LIST_HEAD(&sop->so_perclient);
3427 INIT_LIST_HEAD(&sop->so_stateids);
3428 INIT_LIST_HEAD(&sop->so_perstateid);
3429 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3431 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3432 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3433 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3434 sop->so_is_open_owner = 0;
3435 sop->so_id = current_ownerid++;
3436 sop->so_client = clp;
3437 /* It is the openowner seqid that will be incremented in encode in the
3438 * case of new lockowners; so increment the lock seqid manually: */
3439 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3440 sop->so_confirmed = 1;
3441 rp = &sop->so_replay;
3442 rp->rp_status = nfserr_serverfault;
3444 rp->rp_buf = rp->rp_ibuf;
3448 static struct nfs4_stateid *
3449 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3451 struct nfs4_stateid *stp;
3452 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3454 stp = nfs4_alloc_stateid();
3457 INIT_LIST_HEAD(&stp->st_hash);
3458 INIT_LIST_HEAD(&stp->st_perfile);
3459 INIT_LIST_HEAD(&stp->st_perstateowner);
3460 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3461 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3462 list_add(&stp->st_perfile, &fp->fi_stateids);
3463 list_add(&stp->st_perstateowner, &sop->so_stateids);
3464 stp->st_stateowner = sop;
3467 stp->st_stateid.si_boot = boot_time;
3468 stp->st_stateid.si_stateownerid = sop->so_id;
3469 stp->st_stateid.si_fileid = fp->fi_id;
3470 stp->st_stateid.si_generation = 0;
3471 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3472 stp->st_deny_bmap = open_stp->st_deny_bmap;
3473 stp->st_openstp = open_stp;
3480 check_lock_length(u64 offset, u64 length)
3482 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3483 LOFF_OVERFLOW(offset, length)));
3490 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3491 struct nfsd4_lock *lock)
3493 struct nfs4_stateowner *open_sop = NULL;
3494 struct nfs4_stateowner *lock_sop = NULL;
3495 struct nfs4_stateid *lock_stp;
3497 struct file_lock file_lock;
3498 struct file_lock conflock;
3500 unsigned int strhashval;
3504 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3505 (long long) lock->lk_offset,
3506 (long long) lock->lk_length);
3508 if (check_lock_length(lock->lk_offset, lock->lk_length))
3509 return nfserr_inval;
3511 if ((status = fh_verify(rqstp, &cstate->current_fh,
3512 S_IFREG, NFSD_MAY_LOCK))) {
3513 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3519 if (lock->lk_is_new) {
3521 * Client indicates that this is a new lockowner.
3522 * Use open owner and open stateid to create lock owner and
3525 struct nfs4_stateid *open_stp = NULL;
3526 struct nfs4_file *fp;
3528 status = nfserr_stale_clientid;
3529 if (!nfsd4_has_session(cstate) &&
3530 STALE_CLIENTID(&lock->lk_new_clientid))
3533 /* validate and update open stateid and open seqid */
3534 status = nfs4_preprocess_seqid_op(cstate,
3535 lock->lk_new_open_seqid,
3536 &lock->lk_new_open_stateid,
3538 &lock->lk_replay_owner, &open_stp,
3542 open_sop = lock->lk_replay_owner;
3543 /* create lockowner and lock stateid */
3544 fp = open_stp->st_file;
3545 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3546 open_sop->so_client->cl_clientid.cl_id,
3547 &lock->v.new.owner);
3548 /* XXX: Do we need to check for duplicate stateowners on
3549 * the same file, or should they just be allowed (and
3550 * create new stateids)? */
3551 status = nfserr_resource;
3552 lock_sop = alloc_init_lock_stateowner(strhashval,
3553 open_sop->so_client, open_stp, lock);
3554 if (lock_sop == NULL)
3556 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3557 if (lock_stp == NULL)
3560 /* lock (lock owner + lock stateid) already exists */
3561 status = nfs4_preprocess_seqid_op(cstate,
3562 lock->lk_old_lock_seqid,
3563 &lock->lk_old_lock_stateid,
3565 &lock->lk_replay_owner, &lock_stp, lock);
3568 lock_sop = lock->lk_replay_owner;
3570 /* lock->lk_replay_owner and lock_stp have been created or found */
3571 filp = lock_stp->st_vfs_file;
3573 status = nfserr_grace;
3574 if (locks_in_grace() && !lock->lk_reclaim)
3576 status = nfserr_no_grace;
3577 if (!locks_in_grace() && lock->lk_reclaim)
3580 locks_init_lock(&file_lock);
3581 switch (lock->lk_type) {
3584 file_lock.fl_type = F_RDLCK;
3588 case NFS4_WRITEW_LT:
3589 file_lock.fl_type = F_WRLCK;
3593 status = nfserr_inval;
3596 file_lock.fl_owner = (fl_owner_t)lock_sop;
3597 file_lock.fl_pid = current->tgid;
3598 file_lock.fl_file = filp;
3599 file_lock.fl_flags = FL_POSIX;
3600 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3602 file_lock.fl_start = lock->lk_offset;
3603 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3604 nfs4_transform_lock_offset(&file_lock);
3607 * Try to lock the file in the VFS.
3608 * Note: locks.c uses the BKL to protect the inode's lock list.
3611 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3613 case 0: /* success! */
3614 update_stateid(&lock_stp->st_stateid);
3615 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3619 case (EAGAIN): /* conflock holds conflicting lock */
3620 status = nfserr_denied;
3621 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3622 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3625 status = nfserr_deadlock;
3628 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3629 status = nfserr_resource;
3633 if (status && lock->lk_is_new && lock_sop)
3634 release_lockowner(lock_sop);
3635 if (lock->lk_replay_owner) {
3636 nfs4_get_stateowner(lock->lk_replay_owner);
3637 cstate->replay_owner = lock->lk_replay_owner;
3639 nfs4_unlock_state();
3644 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3645 * so we do a temporary open here just to get an open file to pass to
3646 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3649 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3654 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3657 err = vfs_test_lock(file, lock);
3666 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3667 struct nfsd4_lockt *lockt)
3669 struct inode *inode;
3670 struct file_lock file_lock;
3674 if (locks_in_grace())
3675 return nfserr_grace;
3677 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3678 return nfserr_inval;
3680 lockt->lt_stateowner = NULL;
3683 status = nfserr_stale_clientid;
3684 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3687 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3688 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3689 if (status == nfserr_symlink)
3690 status = nfserr_inval;
3694 inode = cstate->current_fh.fh_dentry->d_inode;
3695 locks_init_lock(&file_lock);
3696 switch (lockt->lt_type) {
3699 file_lock.fl_type = F_RDLCK;
3702 case NFS4_WRITEW_LT:
3703 file_lock.fl_type = F_WRLCK;
3706 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3707 status = nfserr_inval;
3711 lockt->lt_stateowner = find_lockstateowner_str(inode,
3712 &lockt->lt_clientid, &lockt->lt_owner);
3713 if (lockt->lt_stateowner)
3714 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3715 file_lock.fl_pid = current->tgid;
3716 file_lock.fl_flags = FL_POSIX;
3718 file_lock.fl_start = lockt->lt_offset;
3719 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3721 nfs4_transform_lock_offset(&file_lock);
3724 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3726 status = nfserrno(error);
3729 if (file_lock.fl_type != F_UNLCK) {
3730 status = nfserr_denied;
3731 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3734 nfs4_unlock_state();
3739 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3740 struct nfsd4_locku *locku)
3742 struct nfs4_stateid *stp;
3743 struct file *filp = NULL;
3744 struct file_lock file_lock;
3748 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3749 (long long) locku->lu_offset,
3750 (long long) locku->lu_length);
3752 if (check_lock_length(locku->lu_offset, locku->lu_length))
3753 return nfserr_inval;
3757 if ((status = nfs4_preprocess_seqid_op(cstate,
3761 &locku->lu_stateowner, &stp, NULL)))
3764 filp = stp->st_vfs_file;
3766 locks_init_lock(&file_lock);
3767 file_lock.fl_type = F_UNLCK;
3768 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3769 file_lock.fl_pid = current->tgid;
3770 file_lock.fl_file = filp;
3771 file_lock.fl_flags = FL_POSIX;
3772 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3773 file_lock.fl_start = locku->lu_offset;
3775 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3776 nfs4_transform_lock_offset(&file_lock);
3779 * Try to unlock the file in the VFS.
3781 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3783 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3787 * OK, unlock succeeded; the only thing left to do is update the stateid.
3789 update_stateid(&stp->st_stateid);
3790 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3793 if (locku->lu_stateowner) {
3794 nfs4_get_stateowner(locku->lu_stateowner);
3795 cstate->replay_owner = locku->lu_stateowner;
3797 nfs4_unlock_state();
3801 status = nfserrno(err);
3807 * 1: locks held by lockowner
3808 * 0: no locks held by lockowner
3811 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3813 struct file_lock **flpp;
3814 struct inode *inode = filp->f_path.dentry->d_inode;
3818 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3819 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3830 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3831 struct nfsd4_compound_state *cstate,
3832 struct nfsd4_release_lockowner *rlockowner)
3834 clientid_t *clid = &rlockowner->rl_clientid;
3835 struct nfs4_stateowner *sop;
3836 struct nfs4_stateid *stp;
3837 struct xdr_netobj *owner = &rlockowner->rl_owner;
3838 struct list_head matches;
3842 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3843 clid->cl_boot, clid->cl_id);
3845 /* XXX check for lease expiration */
3847 status = nfserr_stale_clientid;
3848 if (STALE_CLIENTID(clid))
3853 status = nfserr_locks_held;
3854 /* XXX: we're doing a linear search through all the lockowners.
3855 * Yipes! For now we'll just hope clients aren't really using
3856 * release_lockowner much, but eventually we have to fix these
3857 * data structures. */
3858 INIT_LIST_HEAD(&matches);
3859 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3860 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3861 if (!same_owner_str(sop, owner, clid))
3863 list_for_each_entry(stp, &sop->so_stateids,
3865 if (check_for_locks(stp->st_vfs_file, sop))
3867 /* Note: so_perclient unused for lockowners,
3868 * so it's OK to fool with here. */
3869 list_add(&sop->so_perclient, &matches);
3873 /* Clients probably won't expect us to return with some (but not all)
3874 * of the lockowner state released; so don't release any until all
3875 * have been checked. */
3877 while (!list_empty(&matches)) {
3878 sop = list_entry(matches.next, struct nfs4_stateowner,
3880 /* unhash_stateowner deletes so_perclient only
3881 * for openowners. */
3882 list_del(&sop->so_perclient);
3883 release_lockowner(sop);
3886 nfs4_unlock_state();
3890 static inline struct nfs4_client_reclaim *
3893 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3897 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3899 unsigned int strhashval = clientstr_hashval(name);
3900 struct nfs4_client *clp;
3902 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3907 * failure => all reset bets are off, nfserr_no_grace...
3910 nfs4_client_to_reclaim(const char *name)
3912 unsigned int strhashval;
3913 struct nfs4_client_reclaim *crp = NULL;
3915 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3916 crp = alloc_reclaim();
3919 strhashval = clientstr_hashval(name);
3920 INIT_LIST_HEAD(&crp->cr_strhash);
3921 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3922 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3923 reclaim_str_hashtbl_size++;
3928 nfs4_release_reclaim(void)
3930 struct nfs4_client_reclaim *crp = NULL;
3933 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3934 while (!list_empty(&reclaim_str_hashtbl[i])) {
3935 crp = list_entry(reclaim_str_hashtbl[i].next,
3936 struct nfs4_client_reclaim, cr_strhash);
3937 list_del(&crp->cr_strhash);
3939 reclaim_str_hashtbl_size--;
3942 BUG_ON(reclaim_str_hashtbl_size);
3946 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3947 static struct nfs4_client_reclaim *
3948 nfs4_find_reclaim_client(clientid_t *clid)
3950 unsigned int strhashval;
3951 struct nfs4_client *clp;
3952 struct nfs4_client_reclaim *crp = NULL;
3955 /* find clientid in conf_id_hashtbl */
3956 clp = find_confirmed_client(clid);
3960 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3961 clp->cl_name.len, clp->cl_name.data,
3964 /* find clp->cl_name in reclaim_str_hashtbl */
3965 strhashval = clientstr_hashval(clp->cl_recdir);
3966 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3967 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3975 * Called from OPEN. Look for clientid in reclaim list.
3978 nfs4_check_open_reclaim(clientid_t *clid)
3980 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3983 /* initialization to perform at module load time: */
3986 nfs4_state_init(void)
3990 status = nfsd4_init_slabs();
3993 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3994 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3995 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3996 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3997 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3998 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4000 for (i = 0; i < SESSION_HASH_SIZE; i++)
4001 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4002 for (i = 0; i < FILE_HASH_SIZE; i++) {
4003 INIT_LIST_HEAD(&file_hashtbl[i]);
4005 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4006 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4007 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4009 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4010 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4011 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4013 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4014 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4015 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4017 memset(&onestateid, ~0, sizeof(stateid_t));
4018 INIT_LIST_HEAD(&close_lru);
4019 INIT_LIST_HEAD(&client_lru);
4020 INIT_LIST_HEAD(&del_recall_lru);
4021 reclaim_str_hashtbl_size = 0;
4026 nfsd4_load_reboot_recovery_data(void)
4031 nfsd4_init_recdir(user_recovery_dirname);
4032 status = nfsd4_recdir_load();
4033 nfs4_unlock_state();
4035 printk("NFSD: Failure reading reboot recovery data\n");
4039 * Since the lifetime of a delegation isn't limited to that of an open, a
4040 * client may quite reasonably hang on to a delegation as long as it has
4041 * the inode cached. This becomes an obvious problem the first time a
4042 * client's inode cache approaches the size of the server's total memory.
4044 * For now we avoid this problem by imposing a hard limit on the number
4045 * of delegations, which varies according to the server's memory size.
4048 set_max_delegations(void)
4051 * Allow at most 4 delegations per megabyte of RAM. Quick
4052 * estimates suggest that in the worst case (where every delegation
4053 * is for a different inode), a delegation could take about 1.5K,
4054 * giving a worst case usage of about 6% of memory.
4056 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4059 /* initialization to perform when the nfsd service is started: */
4062 __nfs4_state_start(void)
4066 boot_time = get_seconds();
4067 locks_start_grace(&nfsd4_manager);
4068 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4070 ret = set_callback_cred();
4073 laundry_wq = create_singlethread_workqueue("nfsd4");
4074 if (laundry_wq == NULL)
4076 ret = nfsd4_create_callback_queue();
4078 goto out_free_laundry;
4079 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4080 set_max_delegations();
4083 destroy_workqueue(laundry_wq);
4088 nfs4_state_start(void)
4090 nfsd4_load_reboot_recovery_data();
4091 return __nfs4_state_start();
4095 __nfs4_state_shutdown(void)
4098 struct nfs4_client *clp = NULL;
4099 struct nfs4_delegation *dp = NULL;
4100 struct list_head *pos, *next, reaplist;
4102 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4103 while (!list_empty(&conf_id_hashtbl[i])) {
4104 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4107 while (!list_empty(&unconf_str_hashtbl[i])) {
4108 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4112 INIT_LIST_HEAD(&reaplist);
4113 spin_lock(&recall_lock);
4114 list_for_each_safe(pos, next, &del_recall_lru) {
4115 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4116 list_move(&dp->dl_recall_lru, &reaplist);
4118 spin_unlock(&recall_lock);
4119 list_for_each_safe(pos, next, &reaplist) {
4120 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4121 list_del_init(&dp->dl_recall_lru);
4122 unhash_delegation(dp);
4125 nfsd4_shutdown_recdir();
4129 nfs4_state_shutdown(void)
4131 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4132 destroy_workqueue(laundry_wq);
4133 locks_end_grace(&nfsd4_manager);
4135 nfs4_release_reclaim();
4136 __nfs4_state_shutdown();
4137 nfsd4_destroy_callback_queue();
4138 nfs4_unlock_state();
4142 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4143 * accessed when nfsd is starting.
4146 nfs4_set_recdir(char *recdir)
4148 strcpy(user_recovery_dirname, recdir);
4152 * Change the NFSv4 recovery directory to recdir.
4155 nfs4_reset_recoverydir(char *recdir)
4160 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4164 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4165 nfs4_set_recdir(recdir);
4173 nfs4_recoverydir(void)
4175 return user_recovery_dirname;