2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58 #include <linux/sunrpc/clnt.h>
60 #define NFSDDBG_FACILITY NFSDDBG_PROC
63 static time_t lease_time = 90; /* default lease time */
64 static time_t user_lease_time = 90;
65 static time_t boot_time;
66 static u32 current_ownerid = 1;
67 static u32 current_fileid = 1;
68 static u32 current_delegid = 1;
70 static stateid_t zerostateid; /* bits all 0 */
71 static stateid_t onestateid; /* bits all 1 */
72 static u64 current_sessionid = 1;
74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
75 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
77 /* forward declarations */
78 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
79 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
80 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
81 static void nfs4_set_recdir(char *recdir);
85 /* Currently used for almost all code touching nfsv4 state: */
86 static DEFINE_MUTEX(client_mutex);
89 * Currently used for the del_recall_lru and file hash table. In an
90 * effort to decrease the scope of the client_mutex, this spinlock may
91 * eventually cover more:
93 static DEFINE_SPINLOCK(recall_lock);
95 static struct kmem_cache *stateowner_slab = NULL;
96 static struct kmem_cache *file_slab = NULL;
97 static struct kmem_cache *stateid_slab = NULL;
98 static struct kmem_cache *deleg_slab = NULL;
101 nfs4_lock_state(void)
103 mutex_lock(&client_mutex);
107 nfs4_unlock_state(void)
109 mutex_unlock(&client_mutex);
113 opaque_hashval(const void *ptr, int nbytes)
115 unsigned char *cptr = (unsigned char *) ptr;
125 static struct list_head del_recall_lru;
128 put_nfs4_file(struct nfs4_file *fi)
130 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
131 list_del(&fi->fi_hash);
132 spin_unlock(&recall_lock);
134 kmem_cache_free(file_slab, fi);
139 get_nfs4_file(struct nfs4_file *fi)
141 atomic_inc(&fi->fi_ref);
144 static int num_delegations;
145 unsigned int max_delegations;
148 * Open owner state (share locks)
151 /* hash tables for nfs4_stateowner */
152 #define OWNER_HASH_BITS 8
153 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
154 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
156 #define ownerid_hashval(id) \
157 ((id) & OWNER_HASH_MASK)
158 #define ownerstr_hashval(clientid, ownername) \
159 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
161 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
162 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
164 /* hash table for nfs4_file */
165 #define FILE_HASH_BITS 8
166 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
167 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
168 /* hash table for (open)nfs4_stateid */
169 #define STATEID_HASH_BITS 10
170 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
171 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
173 #define file_hashval(x) \
174 hash_ptr(x, FILE_HASH_BITS)
175 #define stateid_hashval(owner_id, file_id) \
176 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
178 static struct list_head file_hashtbl[FILE_HASH_SIZE];
179 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
181 static struct nfs4_delegation *
182 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
184 struct nfs4_delegation *dp;
185 struct nfs4_file *fp = stp->st_file;
186 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
188 dprintk("NFSD alloc_init_deleg\n");
189 if (fp->fi_had_conflict)
191 if (num_delegations > max_delegations)
193 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
197 INIT_LIST_HEAD(&dp->dl_perfile);
198 INIT_LIST_HEAD(&dp->dl_perclnt);
199 INIT_LIST_HEAD(&dp->dl_recall_lru);
204 get_file(stp->st_vfs_file);
205 dp->dl_vfs_file = stp->st_vfs_file;
207 dp->dl_ident = cb->cb_ident;
208 dp->dl_stateid.si_boot = get_seconds();
209 dp->dl_stateid.si_stateownerid = current_delegid++;
210 dp->dl_stateid.si_fileid = 0;
211 dp->dl_stateid.si_generation = 0;
212 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
214 atomic_set(&dp->dl_count, 1);
215 list_add(&dp->dl_perfile, &fp->fi_delegations);
216 list_add(&dp->dl_perclnt, &clp->cl_delegations);
221 nfs4_put_delegation(struct nfs4_delegation *dp)
223 if (atomic_dec_and_test(&dp->dl_count)) {
224 dprintk("NFSD: freeing dp %p\n",dp);
225 put_nfs4_file(dp->dl_file);
226 kmem_cache_free(deleg_slab, dp);
231 /* Remove the associated file_lock first, then remove the delegation.
232 * lease_modify() is called to remove the FS_LEASE file_lock from
233 * the i_flock list, eventually calling nfsd's lock_manager
234 * fl_release_callback.
237 nfs4_close_delegation(struct nfs4_delegation *dp)
239 struct file *filp = dp->dl_vfs_file;
241 dprintk("NFSD: close_delegation dp %p\n",dp);
242 dp->dl_vfs_file = NULL;
243 /* The following nfsd_close may not actually close the file,
244 * but we want to remove the lease in any case. */
246 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
250 /* Called under the state lock. */
252 unhash_delegation(struct nfs4_delegation *dp)
254 list_del_init(&dp->dl_perfile);
255 list_del_init(&dp->dl_perclnt);
256 spin_lock(&recall_lock);
257 list_del_init(&dp->dl_recall_lru);
258 spin_unlock(&recall_lock);
259 nfs4_close_delegation(dp);
260 nfs4_put_delegation(dp);
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS 4
269 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
272 #define clientid_hashval(id) \
273 ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
277 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278 * used in reboot/reset lease grace period processing
280 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281 * setclientid_confirmed info.
283 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
286 * client_lru holds client queue ordered by nfs4_client.cl_time
289 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290 * for last close replay.
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
303 list_del(&stp->st_hash);
304 list_del(&stp->st_perfile);
305 list_del(&stp->st_perstateowner);
308 static void free_generic_stateid(struct nfs4_stateid *stp)
310 put_nfs4_file(stp->st_file);
311 kmem_cache_free(stateid_slab, stp);
314 static void release_lock_stateid(struct nfs4_stateid *stp)
316 unhash_generic_stateid(stp);
317 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318 free_generic_stateid(stp);
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
323 struct nfs4_stateid *stp;
325 list_del(&sop->so_idhash);
326 list_del(&sop->so_strhash);
327 list_del(&sop->so_perstateid);
328 while (!list_empty(&sop->so_stateids)) {
329 stp = list_first_entry(&sop->so_stateids,
330 struct nfs4_stateid, st_perstateowner);
331 release_lock_stateid(stp);
335 static void release_lockowner(struct nfs4_stateowner *sop)
337 unhash_lockowner(sop);
338 nfs4_put_stateowner(sop);
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
344 struct nfs4_stateowner *lock_sop;
346 while (!list_empty(&open_stp->st_lockowners)) {
347 lock_sop = list_entry(open_stp->st_lockowners.next,
348 struct nfs4_stateowner, so_perstateid);
349 /* list_del(&open_stp->st_lockowners); */
350 BUG_ON(lock_sop->so_is_open_owner);
351 release_lockowner(lock_sop);
355 static void release_open_stateid(struct nfs4_stateid *stp)
357 unhash_generic_stateid(stp);
358 release_stateid_lockowners(stp);
359 nfsd_close(stp->st_vfs_file);
360 free_generic_stateid(stp);
363 static void unhash_openowner(struct nfs4_stateowner *sop)
365 struct nfs4_stateid *stp;
367 list_del(&sop->so_idhash);
368 list_del(&sop->so_strhash);
369 list_del(&sop->so_perclient);
370 list_del(&sop->so_perstateid); /* XXX: necessary? */
371 while (!list_empty(&sop->so_stateids)) {
372 stp = list_first_entry(&sop->so_stateids,
373 struct nfs4_stateid, st_perstateowner);
374 release_open_stateid(stp);
378 static void release_openowner(struct nfs4_stateowner *sop)
380 unhash_openowner(sop);
381 list_del(&sop->so_close_lru);
382 nfs4_put_stateowner(sop);
385 static DEFINE_SPINLOCK(sessionid_lock);
386 #define SESSION_HASH_SIZE 512
387 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
390 hash_sessionid(struct nfs4_sessionid *sessionid)
392 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
394 return sid->sequence % SESSION_HASH_SIZE;
398 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
400 u32 *ptr = (u32 *)(&sessionid->data[0]);
401 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
405 gen_sessionid(struct nfsd4_session *ses)
407 struct nfs4_client *clp = ses->se_client;
408 struct nfsd4_sessionid *sid;
410 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
411 sid->clientid = clp->cl_clientid;
412 sid->sequence = current_sessionid++;
417 * The protocol defines ca_maxresponssize_cached to include the size of
418 * the rpc header, but all we need to cache is the data starting after
419 * the end of the initial SEQUENCE operation--the rest we regenerate
420 * each time. Therefore we can advertise a ca_maxresponssize_cached
421 * value that is the number of bytes in our cache plus a few additional
422 * bytes. In order to stay on the safe side, and not promise more than
423 * we can cache, those additional bytes must be the minimum possible: 24
424 * bytes of rpc header (xid through accept state, with AUTH_NULL
425 * verifier), 12 for the compound header (with zero-length tag), and 44
426 * for the SEQUENCE op response:
428 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
431 * Give the client the number of ca_maxresponsesize_cached slots it
432 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
433 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
434 * than NFSD_MAX_SLOTS_PER_SESSION.
436 * If we run out of reserved DRC memory we should (up to a point)
437 * re-negotiate active sessions and reduce their slot usage to make
438 * rooom for new connections. For now we just fail the create session.
440 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
442 int mem, size = fchan->maxresp_cached;
444 if (fchan->maxreqs < 1)
447 if (size < NFSD_MIN_HDR_SEQ_SZ)
448 size = NFSD_MIN_HDR_SEQ_SZ;
449 size -= NFSD_MIN_HDR_SEQ_SZ;
450 if (size > NFSD_SLOT_CACHE_SIZE)
451 size = NFSD_SLOT_CACHE_SIZE;
453 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
454 mem = fchan->maxreqs * size;
455 if (mem > NFSD_MAX_MEM_PER_SESSION) {
456 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
457 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
458 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
459 mem = fchan->maxreqs * size;
462 spin_lock(&nfsd_drc_lock);
463 /* bound the total session drc memory ussage */
464 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
465 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
466 mem = fchan->maxreqs * size;
468 nfsd_drc_mem_used += mem;
469 spin_unlock(&nfsd_drc_lock);
471 if (fchan->maxreqs == 0)
472 return nfserr_serverfault;
474 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
479 * fchan holds the client values on input, and the server values on output
481 static int init_forechannel_attrs(struct svc_rqst *rqstp,
482 struct nfsd4_channel_attrs *session_fchan,
483 struct nfsd4_channel_attrs *fchan)
486 __u32 maxcount = svc_max_payload(rqstp);
488 /* headerpadsz set to zero in encode routine */
490 /* Use the client's max request and max response size if possible */
491 if (fchan->maxreq_sz > maxcount)
492 fchan->maxreq_sz = maxcount;
493 session_fchan->maxreq_sz = fchan->maxreq_sz;
495 if (fchan->maxresp_sz > maxcount)
496 fchan->maxresp_sz = maxcount;
497 session_fchan->maxresp_sz = fchan->maxresp_sz;
499 /* Use the client's maxops if possible */
500 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
501 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
502 session_fchan->maxops = fchan->maxops;
504 /* FIXME: Error means no more DRC pages so the server should
505 * recover pages from existing sessions. For now fail session
508 status = set_forechannel_drc_size(fchan);
510 session_fchan->maxresp_cached = fchan->maxresp_cached;
511 session_fchan->maxreqs = fchan->maxreqs;
513 dprintk("%s status %d\n", __func__, status);
518 free_session_slots(struct nfsd4_session *ses)
522 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
523 kfree(ses->se_slots[i]);
527 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
528 struct nfsd4_create_session *cses)
530 struct nfsd4_session *new, tmp;
531 struct nfsd4_slot *sp;
532 int idx, slotsize, cachesize, i;
535 memset(&tmp, 0, sizeof(tmp));
537 /* FIXME: For now, we just accept the client back channel attributes. */
538 tmp.se_bchannel = cses->back_channel;
539 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
540 &cses->fore_channel);
544 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
545 + sizeof(struct nfsd4_session) > PAGE_SIZE);
547 status = nfserr_serverfault;
548 /* allocate struct nfsd4_session and slot table pointers in one piece */
549 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
550 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
554 memcpy(new, &tmp, sizeof(*new));
556 /* allocate each struct nfsd4_slot and data cache in one piece */
557 cachesize = new->se_fchannel.maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
558 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
559 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
562 new->se_slots[i] = sp;
565 new->se_client = clp;
567 idx = hash_sessionid(&new->se_sessionid);
568 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
569 NFS4_MAX_SESSIONID_LEN);
571 new->se_flags = cses->flags;
572 kref_init(&new->se_ref);
573 spin_lock(&sessionid_lock);
574 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
575 list_add(&new->se_perclnt, &clp->cl_sessions);
576 spin_unlock(&sessionid_lock);
582 free_session_slots(new);
587 /* caller must hold sessionid_lock */
588 static struct nfsd4_session *
589 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
591 struct nfsd4_session *elem;
594 dump_sessionid(__func__, sessionid);
595 idx = hash_sessionid(sessionid);
596 dprintk("%s: idx is %d\n", __func__, idx);
597 /* Search in the appropriate list */
598 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
599 dump_sessionid("list traversal", &elem->se_sessionid);
600 if (!memcmp(elem->se_sessionid.data, sessionid->data,
601 NFS4_MAX_SESSIONID_LEN)) {
606 dprintk("%s: session not found\n", __func__);
610 /* caller must hold sessionid_lock */
612 unhash_session(struct nfsd4_session *ses)
614 list_del(&ses->se_hash);
615 list_del(&ses->se_perclnt);
619 release_session(struct nfsd4_session *ses)
621 spin_lock(&sessionid_lock);
623 spin_unlock(&sessionid_lock);
624 nfsd4_put_session(ses);
628 free_session(struct kref *kref)
630 struct nfsd4_session *ses;
632 ses = container_of(kref, struct nfsd4_session, se_ref);
633 spin_lock(&nfsd_drc_lock);
634 nfsd_drc_mem_used -= ses->se_fchannel.maxreqs * NFSD_SLOT_CACHE_SIZE;
635 spin_unlock(&nfsd_drc_lock);
636 free_session_slots(ses);
641 renew_client(struct nfs4_client *clp)
644 * Move client to the end to the LRU list.
646 dprintk("renewing client (clientid %08x/%08x)\n",
647 clp->cl_clientid.cl_boot,
648 clp->cl_clientid.cl_id);
649 list_move_tail(&clp->cl_lru, &client_lru);
650 clp->cl_time = get_seconds();
653 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
655 STALE_CLIENTID(clientid_t *clid)
657 if (clid->cl_boot == boot_time)
659 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
660 clid->cl_boot, clid->cl_id, boot_time);
665 * XXX Should we use a slab cache ?
666 * This type of memory management is somewhat inefficient, but we use it
667 * anyway since SETCLIENTID is not a common operation.
669 static struct nfs4_client *alloc_client(struct xdr_netobj name)
671 struct nfs4_client *clp;
673 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
676 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
677 if (clp->cl_name.data == NULL) {
681 memcpy(clp->cl_name.data, name.data, name.len);
682 clp->cl_name.len = name.len;
687 shutdown_callback_client(struct nfs4_client *clp)
689 struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
693 * Callback threads take a reference on the client, so there
694 * should be no outstanding callbacks at this point.
696 clp->cl_cb_conn.cb_client = NULL;
697 rpc_shutdown_client(clnt);
702 free_client(struct nfs4_client *clp)
704 shutdown_callback_client(clp);
706 svc_xprt_put(clp->cl_cb_xprt);
707 if (clp->cl_cred.cr_group_info)
708 put_group_info(clp->cl_cred.cr_group_info);
709 kfree(clp->cl_principal);
710 kfree(clp->cl_name.data);
715 put_nfs4_client(struct nfs4_client *clp)
717 if (atomic_dec_and_test(&clp->cl_count))
722 expire_client(struct nfs4_client *clp)
724 struct nfs4_stateowner *sop;
725 struct nfs4_delegation *dp;
726 struct list_head reaplist;
728 dprintk("NFSD: expire_client cl_count %d\n",
729 atomic_read(&clp->cl_count));
731 INIT_LIST_HEAD(&reaplist);
732 spin_lock(&recall_lock);
733 while (!list_empty(&clp->cl_delegations)) {
734 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
735 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
737 list_del_init(&dp->dl_perclnt);
738 list_move(&dp->dl_recall_lru, &reaplist);
740 spin_unlock(&recall_lock);
741 while (!list_empty(&reaplist)) {
742 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
743 list_del_init(&dp->dl_recall_lru);
744 unhash_delegation(dp);
746 list_del(&clp->cl_idhash);
747 list_del(&clp->cl_strhash);
748 list_del(&clp->cl_lru);
749 while (!list_empty(&clp->cl_openowners)) {
750 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
751 release_openowner(sop);
753 while (!list_empty(&clp->cl_sessions)) {
754 struct nfsd4_session *ses;
755 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
757 release_session(ses);
759 put_nfs4_client(clp);
762 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
764 struct nfs4_client *clp;
766 clp = alloc_client(name);
769 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
770 atomic_set(&clp->cl_count, 1);
771 atomic_set(&clp->cl_cb_conn.cb_set, 0);
772 INIT_LIST_HEAD(&clp->cl_idhash);
773 INIT_LIST_HEAD(&clp->cl_strhash);
774 INIT_LIST_HEAD(&clp->cl_openowners);
775 INIT_LIST_HEAD(&clp->cl_delegations);
776 INIT_LIST_HEAD(&clp->cl_sessions);
777 INIT_LIST_HEAD(&clp->cl_lru);
781 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
783 memcpy(target->cl_verifier.data, source->data,
784 sizeof(target->cl_verifier.data));
787 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
789 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
790 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
793 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
795 target->cr_uid = source->cr_uid;
796 target->cr_gid = source->cr_gid;
797 target->cr_group_info = source->cr_group_info;
798 get_group_info(target->cr_group_info);
801 static int same_name(const char *n1, const char *n2)
803 return 0 == memcmp(n1, n2, HEXDIR_LEN);
807 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
809 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
813 same_clid(clientid_t *cl1, clientid_t *cl2)
815 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
818 /* XXX what about NGROUP */
820 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
822 return cr1->cr_uid == cr2->cr_uid;
825 static void gen_clid(struct nfs4_client *clp)
827 static u32 current_clientid = 1;
829 clp->cl_clientid.cl_boot = boot_time;
830 clp->cl_clientid.cl_id = current_clientid++;
833 static void gen_confirm(struct nfs4_client *clp)
838 p = (u32 *)clp->cl_confirm.data;
839 *p++ = get_seconds();
843 static int check_name(struct xdr_netobj name)
847 if (name.len > NFS4_OPAQUE_LIMIT) {
848 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
855 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
857 unsigned int idhashval;
859 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
860 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
861 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
862 list_add_tail(&clp->cl_lru, &client_lru);
863 clp->cl_time = get_seconds();
867 move_to_confirmed(struct nfs4_client *clp)
869 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
870 unsigned int strhashval;
872 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
873 list_del_init(&clp->cl_strhash);
874 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
875 strhashval = clientstr_hashval(clp->cl_recdir);
876 list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
880 static struct nfs4_client *
881 find_confirmed_client(clientid_t *clid)
883 struct nfs4_client *clp;
884 unsigned int idhashval = clientid_hashval(clid->cl_id);
886 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
887 if (same_clid(&clp->cl_clientid, clid))
893 static struct nfs4_client *
894 find_unconfirmed_client(clientid_t *clid)
896 struct nfs4_client *clp;
897 unsigned int idhashval = clientid_hashval(clid->cl_id);
899 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
900 if (same_clid(&clp->cl_clientid, clid))
907 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
908 * parameter. Matching is based on the fact the at least one of the
909 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
911 * FIXME: we need to unify the clientid namespaces for nfsv4.x
912 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
913 * and SET_CLIENTID{,_CONFIRM}
916 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
918 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
919 return use_exchange_id == has_exchange_flags;
922 static struct nfs4_client *
923 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
924 bool use_exchange_id)
926 struct nfs4_client *clp;
928 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
929 if (same_name(clp->cl_recdir, dname) &&
930 match_clientid_establishment(clp, use_exchange_id))
936 static struct nfs4_client *
937 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
938 bool use_exchange_id)
940 struct nfs4_client *clp;
942 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
943 if (same_name(clp->cl_recdir, dname) &&
944 match_clientid_establishment(clp, use_exchange_id))
951 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
953 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
954 unsigned short expected_family;
956 /* Currently, we only support tcp and tcp6 for the callback channel */
957 if (se->se_callback_netid_len == 3 &&
958 !memcmp(se->se_callback_netid_val, "tcp", 3))
959 expected_family = AF_INET;
960 else if (se->se_callback_netid_len == 4 &&
961 !memcmp(se->se_callback_netid_val, "tcp6", 4))
962 expected_family = AF_INET6;
966 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
967 se->se_callback_addr_len,
968 (struct sockaddr *) &cb->cb_addr,
969 sizeof(cb->cb_addr));
971 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
974 if (cb->cb_addr.ss_family == AF_INET6)
975 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
977 cb->cb_minorversion = 0;
978 cb->cb_prog = se->se_callback_prog;
979 cb->cb_ident = se->se_callback_ident;
982 cb->cb_addr.ss_family = AF_UNSPEC;
984 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
985 "will not receive delegations\n",
986 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
992 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
995 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
997 struct nfsd4_slot *slot = resp->cstate.slot;
1000 dprintk("--> %s slot %p\n", __func__, slot);
1002 slot->sl_opcnt = resp->opcnt;
1003 slot->sl_status = resp->cstate.status;
1005 if (nfsd4_not_cached(resp)) {
1006 slot->sl_datalen = 0;
1009 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1010 base = (char *)resp->cstate.datap -
1011 (char *)resp->xbuf->head[0].iov_base;
1012 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1014 WARN("%s: sessions DRC could not cache compound\n", __func__);
1019 * Encode the replay sequence operation from the slot values.
1020 * If cachethis is FALSE encode the uncached rep error on the next
1021 * operation which sets resp->p and increments resp->opcnt for
1022 * nfs4svc_encode_compoundres.
1026 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1027 struct nfsd4_compoundres *resp)
1029 struct nfsd4_op *op;
1030 struct nfsd4_slot *slot = resp->cstate.slot;
1032 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1033 resp->opcnt, resp->cstate.slot->sl_cachethis);
1035 /* Encode the replayed sequence operation */
1036 op = &args->ops[resp->opcnt - 1];
1037 nfsd4_encode_operation(resp, op);
1039 /* Return nfserr_retry_uncached_rep in next operation. */
1040 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1041 op = &args->ops[resp->opcnt++];
1042 op->status = nfserr_retry_uncached_rep;
1043 nfsd4_encode_operation(resp, op);
1049 * The sequence operation is not cached because we can use the slot and
1053 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1054 struct nfsd4_sequence *seq)
1056 struct nfsd4_slot *slot = resp->cstate.slot;
1059 dprintk("--> %s slot %p\n", __func__, slot);
1061 /* Either returns 0 or nfserr_retry_uncached */
1062 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1063 if (status == nfserr_retry_uncached_rep)
1066 /* The sequence operation has been encoded, cstate->datap set. */
1067 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1069 resp->opcnt = slot->sl_opcnt;
1070 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1071 status = slot->sl_status;
1077 * Set the exchange_id flags returned by the server.
1080 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1082 /* pNFS is not supported */
1083 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1085 /* Referrals are supported, Migration is not. */
1086 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1088 /* set the wire flags to return to client. */
1089 clid->flags = new->cl_exchange_flags;
1093 nfsd4_exchange_id(struct svc_rqst *rqstp,
1094 struct nfsd4_compound_state *cstate,
1095 struct nfsd4_exchange_id *exid)
1097 struct nfs4_client *unconf, *conf, *new;
1099 unsigned int strhashval;
1100 char dname[HEXDIR_LEN];
1101 char addr_str[INET6_ADDRSTRLEN];
1102 nfs4_verifier verf = exid->verifier;
1103 struct sockaddr *sa = svc_addr(rqstp);
1105 rpc_ntop(sa, addr_str, sizeof(addr_str));
1106 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1107 "ip_addr=%s flags %x, spa_how %d\n",
1108 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1109 addr_str, exid->flags, exid->spa_how);
1111 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1112 return nfserr_inval;
1114 /* Currently only support SP4_NONE */
1115 switch (exid->spa_how) {
1119 return nfserr_encr_alg_unsupp;
1121 BUG(); /* checked by xdr code */
1123 return nfserr_serverfault; /* no excuse :-/ */
1126 status = nfs4_make_rec_clidname(dname, &exid->clname);
1131 strhashval = clientstr_hashval(dname);
1136 conf = find_confirmed_client_by_str(dname, strhashval, true);
1138 if (!same_verf(&verf, &conf->cl_verifier)) {
1139 /* 18.35.4 case 8 */
1140 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1141 status = nfserr_not_same;
1144 /* Client reboot: destroy old state */
1145 expire_client(conf);
1148 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1149 /* 18.35.4 case 9 */
1150 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1151 status = nfserr_perm;
1154 expire_client(conf);
1158 * Set bit when the owner id and verifier map to an already
1159 * confirmed client id (18.35.3).
1161 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1164 * Falling into 18.35.4 case 2, possible router replay.
1165 * Leave confirmed record intact and return same result.
1167 copy_verf(conf, &verf);
1172 /* 18.35.4 case 7 */
1173 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1174 status = nfserr_noent;
1178 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1181 * Possible retry or client restart. Per 18.35.4 case 4,
1182 * a new unconfirmed record should be generated regardless
1183 * of whether any properties have changed.
1185 expire_client(unconf);
1190 new = create_client(exid->clname, dname);
1192 status = nfserr_serverfault;
1196 copy_verf(new, &verf);
1197 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1198 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1201 add_to_unconfirmed(new, strhashval);
1203 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1204 exid->clientid.cl_id = new->cl_clientid.cl_id;
1207 nfsd4_set_ex_flags(new, exid);
1209 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1210 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1214 nfs4_unlock_state();
1216 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1221 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1223 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1226 /* The slot is in use, and no response has been sent. */
1228 if (seqid == slot_seqid)
1229 return nfserr_jukebox;
1231 return nfserr_seq_misordered;
1234 if (likely(seqid == slot_seqid + 1))
1237 if (seqid == slot_seqid)
1238 return nfserr_replay_cache;
1240 if (seqid == 1 && (slot_seqid + 1) == 0)
1242 /* Misordered replay or misordered new request */
1243 return nfserr_seq_misordered;
1247 * Cache the create session result into the create session single DRC
1248 * slot cache by saving the xdr structure. sl_seqid has been set.
1249 * Do this for solo or embedded create session operations.
1252 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1253 struct nfsd4_clid_slot *slot, int nfserr)
1255 slot->sl_status = nfserr;
1256 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1260 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1261 struct nfsd4_clid_slot *slot)
1263 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1264 return slot->sl_status;
1268 nfsd4_create_session(struct svc_rqst *rqstp,
1269 struct nfsd4_compound_state *cstate,
1270 struct nfsd4_create_session *cr_ses)
1272 struct sockaddr *sa = svc_addr(rqstp);
1273 struct nfs4_client *conf, *unconf;
1274 struct nfsd4_clid_slot *cs_slot = NULL;
1278 unconf = find_unconfirmed_client(&cr_ses->clientid);
1279 conf = find_confirmed_client(&cr_ses->clientid);
1282 cs_slot = &conf->cl_cs_slot;
1283 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1284 if (status == nfserr_replay_cache) {
1285 dprintk("Got a create_session replay! seqid= %d\n",
1287 /* Return the cached reply status */
1288 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1290 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1291 status = nfserr_seq_misordered;
1292 dprintk("Sequence misordered!\n");
1293 dprintk("Expected seqid= %d but got seqid= %d\n",
1294 cs_slot->sl_seqid, cr_ses->seqid);
1297 cs_slot->sl_seqid++;
1298 } else if (unconf) {
1299 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1300 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1301 status = nfserr_clid_inuse;
1305 cs_slot = &unconf->cl_cs_slot;
1306 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1308 /* an unconfirmed replay returns misordered */
1309 status = nfserr_seq_misordered;
1313 cs_slot->sl_seqid++; /* from 0 to 1 */
1314 move_to_confirmed(unconf);
1317 * We do not support RDMA or persistent sessions
1319 cr_ses->flags &= ~SESSION4_PERSIST;
1320 cr_ses->flags &= ~SESSION4_RDMA;
1322 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1323 unconf->cl_cb_xprt = rqstp->rq_xprt;
1324 svc_xprt_get(unconf->cl_cb_xprt);
1326 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1328 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1329 unconf->cl_cb_conn.cb_minorversion =
1330 cstate->minorversion;
1331 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1332 nfsd4_probe_callback(unconf);
1336 status = nfserr_stale_clientid;
1340 status = alloc_init_session(rqstp, conf, cr_ses);
1344 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1345 NFS4_MAX_SESSIONID_LEN);
1346 cr_ses->seqid = cs_slot->sl_seqid;
1349 /* cache solo and embedded create sessions under the state lock */
1350 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1352 nfs4_unlock_state();
1353 dprintk("%s returns %d\n", __func__, ntohl(status));
1358 nfsd4_destroy_session(struct svc_rqst *r,
1359 struct nfsd4_compound_state *cstate,
1360 struct nfsd4_destroy_session *sessionid)
1362 struct nfsd4_session *ses;
1363 u32 status = nfserr_badsession;
1366 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1367 * - Should we return nfserr_back_chan_busy if waiting for
1368 * callbacks on to-be-destroyed session?
1369 * - Do we need to clear any callback info from previous session?
1372 dump_sessionid(__func__, &sessionid->sessionid);
1373 spin_lock(&sessionid_lock);
1374 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1376 spin_unlock(&sessionid_lock);
1380 unhash_session(ses);
1381 spin_unlock(&sessionid_lock);
1383 /* wait for callbacks */
1384 shutdown_callback_client(ses->se_client);
1385 nfsd4_put_session(ses);
1388 dprintk("%s returns %d\n", __func__, ntohl(status));
1393 nfsd4_sequence(struct svc_rqst *rqstp,
1394 struct nfsd4_compound_state *cstate,
1395 struct nfsd4_sequence *seq)
1397 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1398 struct nfsd4_session *session;
1399 struct nfsd4_slot *slot;
1402 if (resp->opcnt != 1)
1403 return nfserr_sequence_pos;
1405 spin_lock(&sessionid_lock);
1406 status = nfserr_badsession;
1407 session = find_in_sessionid_hashtbl(&seq->sessionid);
1411 status = nfserr_badslot;
1412 if (seq->slotid >= session->se_fchannel.maxreqs)
1415 slot = session->se_slots[seq->slotid];
1416 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1418 /* We do not negotiate the number of slots yet, so set the
1419 * maxslots to the session maxreqs which is used to encode
1420 * sr_highest_slotid and the sr_target_slot id to maxslots */
1421 seq->maxslots = session->se_fchannel.maxreqs;
1423 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1424 if (status == nfserr_replay_cache) {
1425 cstate->slot = slot;
1426 cstate->session = session;
1427 /* Return the cached reply status and set cstate->status
1428 * for nfsd4_proc_compound processing */
1429 status = nfsd4_replay_cache_entry(resp, seq);
1430 cstate->status = nfserr_replay_cache;
1436 /* Success! bump slot seqid */
1437 slot->sl_inuse = true;
1438 slot->sl_seqid = seq->seqid;
1439 slot->sl_cachethis = seq->cachethis;
1441 cstate->slot = slot;
1442 cstate->session = session;
1444 /* Hold a session reference until done processing the compound:
1445 * nfsd4_put_session called only if the cstate slot is set.
1447 nfsd4_get_session(session);
1449 spin_unlock(&sessionid_lock);
1450 /* Renew the clientid on success and on replay */
1451 if (cstate->session) {
1453 renew_client(session->se_client);
1454 nfs4_unlock_state();
1456 dprintk("%s: return %d\n", __func__, ntohl(status));
1461 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1462 struct nfsd4_setclientid *setclid)
1464 struct sockaddr *sa = svc_addr(rqstp);
1465 struct xdr_netobj clname = {
1466 .len = setclid->se_namelen,
1467 .data = setclid->se_name,
1469 nfs4_verifier clverifier = setclid->se_verf;
1470 unsigned int strhashval;
1471 struct nfs4_client *conf, *unconf, *new;
1474 char dname[HEXDIR_LEN];
1476 if (!check_name(clname))
1477 return nfserr_inval;
1479 status = nfs4_make_rec_clidname(dname, &clname);
1484 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1485 * We get here on a DRC miss.
1488 strhashval = clientstr_hashval(dname);
1491 conf = find_confirmed_client_by_str(dname, strhashval, false);
1493 /* RFC 3530 14.2.33 CASE 0: */
1494 status = nfserr_clid_inuse;
1495 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1496 char addr_str[INET6_ADDRSTRLEN];
1497 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1499 dprintk("NFSD: setclientid: string in use by client "
1500 "at %s\n", addr_str);
1505 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1506 * has a description of SETCLIENTID request processing consisting
1507 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1509 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1510 status = nfserr_resource;
1513 * RFC 3530 14.2.33 CASE 4:
1514 * placed first, because it is the normal case
1517 expire_client(unconf);
1518 new = create_client(clname, dname);
1522 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1524 * RFC 3530 14.2.33 CASE 1:
1525 * probable callback update
1528 /* Note this is removing unconfirmed {*x***},
1529 * which is stronger than RFC recommended {vxc**}.
1530 * This has the advantage that there is at most
1531 * one {*x***} in either list at any time.
1533 expire_client(unconf);
1535 new = create_client(clname, dname);
1538 copy_clid(new, conf);
1539 } else if (!unconf) {
1541 * RFC 3530 14.2.33 CASE 2:
1542 * probable client reboot; state will be removed if
1545 new = create_client(clname, dname);
1551 * RFC 3530 14.2.33 CASE 3:
1552 * probable client reboot; state will be removed if
1555 expire_client(unconf);
1556 new = create_client(clname, dname);
1561 copy_verf(new, &clverifier);
1562 rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1563 new->cl_flavor = rqstp->rq_flavor;
1564 princ = svc_gss_principal(rqstp);
1566 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1567 if (new->cl_principal == NULL) {
1572 copy_cred(&new->cl_cred, &rqstp->rq_cred);
1574 gen_callback(new, setclid, rpc_get_scope_id(sa));
1575 add_to_unconfirmed(new, strhashval);
1576 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1577 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1578 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1581 nfs4_unlock_state();
1587 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1588 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1589 * bullets, labeled as CASE1 - CASE4 below.
1592 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1593 struct nfsd4_compound_state *cstate,
1594 struct nfsd4_setclientid_confirm *setclientid_confirm)
1596 struct sockaddr *sa = svc_addr(rqstp);
1597 struct nfs4_client *conf, *unconf;
1598 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1599 clientid_t * clid = &setclientid_confirm->sc_clientid;
1602 if (STALE_CLIENTID(clid))
1603 return nfserr_stale_clientid;
1605 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1606 * We get here on a DRC miss.
1611 conf = find_confirmed_client(clid);
1612 unconf = find_unconfirmed_client(clid);
1614 status = nfserr_clid_inuse;
1615 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1617 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1621 * section 14.2.34 of RFC 3530 has a description of
1622 * SETCLIENTID_CONFIRM request processing consisting
1623 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1625 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1627 * RFC 3530 14.2.34 CASE 1:
1630 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1631 status = nfserr_clid_inuse;
1633 /* XXX: We just turn off callbacks until we can handle
1634 * change request correctly. */
1635 atomic_set(&conf->cl_cb_conn.cb_set, 0);
1636 expire_client(unconf);
1640 } else if (conf && !unconf) {
1642 * RFC 3530 14.2.34 CASE 2:
1643 * probable retransmitted request; play it safe and
1646 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1647 status = nfserr_clid_inuse;
1650 } else if (!conf && unconf
1651 && same_verf(&unconf->cl_confirm, &confirm)) {
1653 * RFC 3530 14.2.34 CASE 3:
1654 * Normal case; new or rebooted client:
1656 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1657 status = nfserr_clid_inuse;
1660 clientstr_hashval(unconf->cl_recdir);
1661 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1664 nfsd4_remove_clid_dir(conf);
1665 expire_client(conf);
1667 move_to_confirmed(unconf);
1669 nfsd4_probe_callback(conf);
1672 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1673 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1676 * RFC 3530 14.2.34 CASE 4:
1677 * Client probably hasn't noticed that we rebooted yet.
1679 status = nfserr_stale_clientid;
1681 /* check that we have hit one of the cases...*/
1682 status = nfserr_clid_inuse;
1685 nfs4_unlock_state();
1689 /* OPEN Share state helper functions */
1690 static inline struct nfs4_file *
1691 alloc_init_file(struct inode *ino)
1693 struct nfs4_file *fp;
1694 unsigned int hashval = file_hashval(ino);
1696 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1698 atomic_set(&fp->fi_ref, 1);
1699 INIT_LIST_HEAD(&fp->fi_hash);
1700 INIT_LIST_HEAD(&fp->fi_stateids);
1701 INIT_LIST_HEAD(&fp->fi_delegations);
1702 spin_lock(&recall_lock);
1703 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1704 spin_unlock(&recall_lock);
1705 fp->fi_inode = igrab(ino);
1706 fp->fi_id = current_fileid++;
1707 fp->fi_had_conflict = false;
1714 nfsd4_free_slab(struct kmem_cache **slab)
1718 kmem_cache_destroy(*slab);
1723 nfsd4_free_slabs(void)
1725 nfsd4_free_slab(&stateowner_slab);
1726 nfsd4_free_slab(&file_slab);
1727 nfsd4_free_slab(&stateid_slab);
1728 nfsd4_free_slab(&deleg_slab);
1732 nfsd4_init_slabs(void)
1734 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1735 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1736 if (stateowner_slab == NULL)
1738 file_slab = kmem_cache_create("nfsd4_files",
1739 sizeof(struct nfs4_file), 0, 0, NULL);
1740 if (file_slab == NULL)
1742 stateid_slab = kmem_cache_create("nfsd4_stateids",
1743 sizeof(struct nfs4_stateid), 0, 0, NULL);
1744 if (stateid_slab == NULL)
1746 deleg_slab = kmem_cache_create("nfsd4_delegations",
1747 sizeof(struct nfs4_delegation), 0, 0, NULL);
1748 if (deleg_slab == NULL)
1753 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1758 nfs4_free_stateowner(struct kref *kref)
1760 struct nfs4_stateowner *sop =
1761 container_of(kref, struct nfs4_stateowner, so_ref);
1762 kfree(sop->so_owner.data);
1763 kmem_cache_free(stateowner_slab, sop);
1766 static inline struct nfs4_stateowner *
1767 alloc_stateowner(struct xdr_netobj *owner)
1769 struct nfs4_stateowner *sop;
1771 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1772 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1773 memcpy(sop->so_owner.data, owner->data, owner->len);
1774 sop->so_owner.len = owner->len;
1775 kref_init(&sop->so_ref);
1778 kmem_cache_free(stateowner_slab, sop);
1783 static struct nfs4_stateowner *
1784 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1785 struct nfs4_stateowner *sop;
1786 struct nfs4_replay *rp;
1787 unsigned int idhashval;
1789 if (!(sop = alloc_stateowner(&open->op_owner)))
1791 idhashval = ownerid_hashval(current_ownerid);
1792 INIT_LIST_HEAD(&sop->so_idhash);
1793 INIT_LIST_HEAD(&sop->so_strhash);
1794 INIT_LIST_HEAD(&sop->so_perclient);
1795 INIT_LIST_HEAD(&sop->so_stateids);
1796 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1797 INIT_LIST_HEAD(&sop->so_close_lru);
1799 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1800 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1801 list_add(&sop->so_perclient, &clp->cl_openowners);
1802 sop->so_is_open_owner = 1;
1803 sop->so_id = current_ownerid++;
1804 sop->so_client = clp;
1805 sop->so_seqid = open->op_seqid;
1806 sop->so_confirmed = 0;
1807 rp = &sop->so_replay;
1808 rp->rp_status = nfserr_serverfault;
1810 rp->rp_buf = rp->rp_ibuf;
1815 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1816 struct nfs4_stateowner *sop = open->op_stateowner;
1817 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1819 INIT_LIST_HEAD(&stp->st_hash);
1820 INIT_LIST_HEAD(&stp->st_perstateowner);
1821 INIT_LIST_HEAD(&stp->st_lockowners);
1822 INIT_LIST_HEAD(&stp->st_perfile);
1823 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1824 list_add(&stp->st_perstateowner, &sop->so_stateids);
1825 list_add(&stp->st_perfile, &fp->fi_stateids);
1826 stp->st_stateowner = sop;
1829 stp->st_stateid.si_boot = get_seconds();
1830 stp->st_stateid.si_stateownerid = sop->so_id;
1831 stp->st_stateid.si_fileid = fp->fi_id;
1832 stp->st_stateid.si_generation = 0;
1833 stp->st_access_bmap = 0;
1834 stp->st_deny_bmap = 0;
1835 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1836 &stp->st_access_bmap);
1837 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1838 stp->st_openstp = NULL;
1842 move_to_close_lru(struct nfs4_stateowner *sop)
1844 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1846 list_move_tail(&sop->so_close_lru, &close_lru);
1847 sop->so_time = get_seconds();
1851 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1854 return (sop->so_owner.len == owner->len) &&
1855 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1856 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1859 static struct nfs4_stateowner *
1860 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1862 struct nfs4_stateowner *so = NULL;
1864 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1865 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1871 /* search file_hashtbl[] for file */
1872 static struct nfs4_file *
1873 find_file(struct inode *ino)
1875 unsigned int hashval = file_hashval(ino);
1876 struct nfs4_file *fp;
1878 spin_lock(&recall_lock);
1879 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1880 if (fp->fi_inode == ino) {
1882 spin_unlock(&recall_lock);
1886 spin_unlock(&recall_lock);
1890 static inline int access_valid(u32 x, u32 minorversion)
1892 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1894 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1896 x &= ~NFS4_SHARE_ACCESS_MASK;
1897 if (minorversion && x) {
1898 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1900 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1902 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1909 static inline int deny_valid(u32 x)
1911 /* Note: unlike access bits, deny bits may be zero. */
1912 return x <= NFS4_SHARE_DENY_BOTH;
1916 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1917 * st_{access,deny}_bmap field of the stateid, in order to track not
1918 * only what share bits are currently in force, but also what
1919 * combinations of share bits previous opens have used. This allows us
1920 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1921 * return an error if the client attempt to downgrade to a combination
1922 * of share bits not explicable by closing some of its previous opens.
1924 * XXX: This enforcement is actually incomplete, since we don't keep
1925 * track of access/deny bit combinations; so, e.g., we allow:
1927 * OPEN allow read, deny write
1928 * OPEN allow both, deny none
1929 * DOWNGRADE allow read, deny none
1931 * which we should reject.
1934 set_access(unsigned int *access, unsigned long bmap) {
1938 for (i = 1; i < 4; i++) {
1939 if (test_bit(i, &bmap))
1945 set_deny(unsigned int *deny, unsigned long bmap) {
1949 for (i = 0; i < 4; i++) {
1950 if (test_bit(i, &bmap))
1956 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1957 unsigned int access, deny;
1959 set_access(&access, stp->st_access_bmap);
1960 set_deny(&deny, stp->st_deny_bmap);
1961 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1967 * Called to check deny when READ with all zero stateid or
1968 * WRITE with all zero or all one stateid
1971 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1973 struct inode *ino = current_fh->fh_dentry->d_inode;
1974 struct nfs4_file *fp;
1975 struct nfs4_stateid *stp;
1978 dprintk("NFSD: nfs4_share_conflict\n");
1980 fp = find_file(ino);
1983 ret = nfserr_locked;
1984 /* Search for conflicting share reservations */
1985 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1986 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1987 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1997 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1999 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2000 drop_file_write_access(filp);
2001 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2006 * Spawn a thread to perform a recall on the delegation represented
2007 * by the lease (file_lock)
2009 * Called from break_lease() with lock_kernel() held.
2010 * Note: we assume break_lease will only call this *once* for any given
2014 void nfsd_break_deleg_cb(struct file_lock *fl)
2016 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2018 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2022 /* We're assuming the state code never drops its reference
2023 * without first removing the lease. Since we're in this lease
2024 * callback (and since the lease code is serialized by the kernel
2025 * lock) we know the server hasn't removed the lease yet, we know
2026 * it's safe to take a reference: */
2027 atomic_inc(&dp->dl_count);
2028 atomic_inc(&dp->dl_client->cl_count);
2030 spin_lock(&recall_lock);
2031 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2032 spin_unlock(&recall_lock);
2034 /* only place dl_time is set. protected by lock_kernel*/
2035 dp->dl_time = get_seconds();
2038 * We don't want the locks code to timeout the lease for us;
2039 * we'll remove it ourself if the delegation isn't returned
2042 fl->fl_break_time = 0;
2044 dp->dl_file->fi_had_conflict = true;
2045 nfsd4_cb_recall(dp);
2049 * The file_lock is being reapd.
2051 * Called by locks_free_lock() with lock_kernel() held.
2054 void nfsd_release_deleg_cb(struct file_lock *fl)
2056 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2058 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2060 if (!(fl->fl_flags & FL_LEASE) || !dp)
2062 dp->dl_flock = NULL;
2066 * Set the delegation file_lock back pointer.
2068 * Called from setlease() with lock_kernel() held.
2071 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2073 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2075 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2082 * Called from setlease() with lock_kernel() held
2085 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2087 struct nfs4_delegation *onlistd =
2088 (struct nfs4_delegation *)onlist->fl_owner;
2089 struct nfs4_delegation *tryd =
2090 (struct nfs4_delegation *)try->fl_owner;
2092 if (onlist->fl_lmops != try->fl_lmops)
2095 return onlistd->dl_client == tryd->dl_client;
2100 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2103 return lease_modify(onlist, arg);
2108 static struct lock_manager_operations nfsd_lease_mng_ops = {
2109 .fl_break = nfsd_break_deleg_cb,
2110 .fl_release_private = nfsd_release_deleg_cb,
2111 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2112 .fl_mylease = nfsd_same_client_deleg_cb,
2113 .fl_change = nfsd_change_deleg_cb,
2118 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2119 struct nfsd4_open *open)
2121 clientid_t *clientid = &open->op_clientid;
2122 struct nfs4_client *clp = NULL;
2123 unsigned int strhashval;
2124 struct nfs4_stateowner *sop = NULL;
2126 if (!check_name(open->op_owner))
2127 return nfserr_inval;
2129 if (STALE_CLIENTID(&open->op_clientid))
2130 return nfserr_stale_clientid;
2132 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2133 sop = find_openstateowner_str(strhashval, open);
2134 open->op_stateowner = sop;
2136 /* Make sure the client's lease hasn't expired. */
2137 clp = find_confirmed_client(clientid);
2139 return nfserr_expired;
2142 /* When sessions are used, skip open sequenceid processing */
2143 if (nfsd4_has_session(cstate))
2145 if (!sop->so_confirmed) {
2146 /* Replace unconfirmed owners without checking for replay. */
2147 clp = sop->so_client;
2148 release_openowner(sop);
2149 open->op_stateowner = NULL;
2152 if (open->op_seqid == sop->so_seqid - 1) {
2153 if (sop->so_replay.rp_buflen)
2154 return nfserr_replay_me;
2155 /* The original OPEN failed so spectacularly
2156 * that we don't even have replay data saved!
2157 * Therefore, we have no choice but to continue
2158 * processing this OPEN; presumably, we'll
2159 * fail again for the same reason.
2161 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2164 if (open->op_seqid != sop->so_seqid)
2165 return nfserr_bad_seqid;
2167 if (open->op_stateowner == NULL) {
2168 sop = alloc_init_open_stateowner(strhashval, clp, open);
2170 return nfserr_resource;
2171 open->op_stateowner = sop;
2173 list_del_init(&sop->so_close_lru);
2174 renew_client(sop->so_client);
2178 static inline __be32
2179 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2181 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2182 return nfserr_openmode;
2187 static struct nfs4_delegation *
2188 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2190 struct nfs4_delegation *dp;
2192 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2193 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2200 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2201 struct nfs4_delegation **dp)
2204 __be32 status = nfserr_bad_stateid;
2206 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2209 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2210 RD_STATE : WR_STATE;
2211 status = nfs4_check_delegmode(*dp, flags);
2215 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2219 open->op_stateowner->so_confirmed = 1;
2224 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2226 struct nfs4_stateid *local;
2227 __be32 status = nfserr_share_denied;
2228 struct nfs4_stateowner *sop = open->op_stateowner;
2230 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2231 /* ignore lock owners */
2232 if (local->st_stateowner->so_is_open_owner == 0)
2234 /* remember if we have seen this open owner */
2235 if (local->st_stateowner == sop)
2237 /* check for conflicting share reservations */
2238 if (!test_share(local, open))
2246 static inline struct nfs4_stateid *
2247 nfs4_alloc_stateid(void)
2249 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2253 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2254 struct nfs4_delegation *dp,
2255 struct svc_fh *cur_fh, int flags)
2257 struct nfs4_stateid *stp;
2259 stp = nfs4_alloc_stateid();
2261 return nfserr_resource;
2264 get_file(dp->dl_vfs_file);
2265 stp->st_vfs_file = dp->dl_vfs_file;
2268 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2271 if (status == nfserr_dropit)
2272 status = nfserr_jukebox;
2273 kmem_cache_free(stateid_slab, stp);
2281 static inline __be32
2282 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2283 struct nfsd4_open *open)
2285 struct iattr iattr = {
2286 .ia_valid = ATTR_SIZE,
2289 if (!open->op_truncate)
2291 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2292 return nfserr_inval;
2293 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2297 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2299 struct file *filp = stp->st_vfs_file;
2300 struct inode *inode = filp->f_path.dentry->d_inode;
2301 unsigned int share_access, new_writer;
2304 set_access(&share_access, stp->st_access_bmap);
2305 new_writer = (~share_access) & open->op_share_access
2306 & NFS4_SHARE_ACCESS_WRITE;
2309 int err = get_write_access(inode);
2311 return nfserrno(err);
2312 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2314 return nfserrno(err);
2315 file_take_write(filp);
2317 status = nfsd4_truncate(rqstp, cur_fh, open);
2320 put_write_access(inode);
2323 /* remember the open */
2324 filp->f_mode |= open->op_share_access;
2325 __set_bit(open->op_share_access, &stp->st_access_bmap);
2326 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2333 nfs4_set_claim_prev(struct nfsd4_open *open)
2335 open->op_stateowner->so_confirmed = 1;
2336 open->op_stateowner->so_client->cl_firststate = 1;
2340 * Attempt to hand out a delegation.
2343 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2345 struct nfs4_delegation *dp;
2346 struct nfs4_stateowner *sop = stp->st_stateowner;
2347 struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2348 struct file_lock fl, *flp = &fl;
2349 int status, flag = 0;
2351 flag = NFS4_OPEN_DELEGATE_NONE;
2352 open->op_recall = 0;
2353 switch (open->op_claim_type) {
2354 case NFS4_OPEN_CLAIM_PREVIOUS:
2355 if (!atomic_read(&cb->cb_set))
2356 open->op_recall = 1;
2357 flag = open->op_delegate_type;
2358 if (flag == NFS4_OPEN_DELEGATE_NONE)
2361 case NFS4_OPEN_CLAIM_NULL:
2362 /* Let's not give out any delegations till everyone's
2363 * had the chance to reclaim theirs.... */
2364 if (locks_in_grace())
2366 if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2368 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2369 flag = NFS4_OPEN_DELEGATE_WRITE;
2371 flag = NFS4_OPEN_DELEGATE_READ;
2377 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2379 flag = NFS4_OPEN_DELEGATE_NONE;
2382 locks_init_lock(&fl);
2383 fl.fl_lmops = &nfsd_lease_mng_ops;
2384 fl.fl_flags = FL_LEASE;
2385 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2386 fl.fl_end = OFFSET_MAX;
2387 fl.fl_owner = (fl_owner_t)dp;
2388 fl.fl_file = stp->st_vfs_file;
2389 fl.fl_pid = current->tgid;
2391 /* vfs_setlease checks to see if delegation should be handed out.
2392 * the lock_manager callbacks fl_mylease and fl_change are used
2394 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2395 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2396 unhash_delegation(dp);
2397 flag = NFS4_OPEN_DELEGATE_NONE;
2401 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2403 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2404 dp->dl_stateid.si_boot,
2405 dp->dl_stateid.si_stateownerid,
2406 dp->dl_stateid.si_fileid,
2407 dp->dl_stateid.si_generation);
2409 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2410 && flag == NFS4_OPEN_DELEGATE_NONE
2411 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2412 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2413 open->op_delegate_type = flag;
2417 * called with nfs4_lock_state() held.
2420 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2422 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2423 struct nfs4_file *fp = NULL;
2424 struct inode *ino = current_fh->fh_dentry->d_inode;
2425 struct nfs4_stateid *stp = NULL;
2426 struct nfs4_delegation *dp = NULL;
2429 status = nfserr_inval;
2430 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2431 || !deny_valid(open->op_share_deny))
2434 * Lookup file; if found, lookup stateid and check open request,
2435 * and check for delegations in the process of being recalled.
2436 * If not found, create the nfs4_file struct
2438 fp = find_file(ino);
2440 if ((status = nfs4_check_open(fp, open, &stp)))
2442 status = nfs4_check_deleg(fp, open, &dp);
2446 status = nfserr_bad_stateid;
2447 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2449 status = nfserr_resource;
2450 fp = alloc_init_file(ino);
2456 * OPEN the file, or upgrade an existing OPEN.
2457 * If truncate fails, the OPEN fails.
2460 /* Stateid was found, this is an OPEN upgrade */
2461 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2464 update_stateid(&stp->st_stateid);
2466 /* Stateid was not found, this is a new OPEN */
2468 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2469 flags |= NFSD_MAY_READ;
2470 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2471 flags |= NFSD_MAY_WRITE;
2472 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2475 init_stateid(stp, fp, open);
2476 status = nfsd4_truncate(rqstp, current_fh, open);
2478 release_open_stateid(stp);
2481 if (nfsd4_has_session(&resp->cstate))
2482 update_stateid(&stp->st_stateid);
2484 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2486 if (nfsd4_has_session(&resp->cstate))
2487 open->op_stateowner->so_confirmed = 1;
2490 * Attempt to hand out a delegation. No error return, because the
2491 * OPEN succeeds even if we fail.
2493 nfs4_open_delegation(current_fh, open, stp);
2497 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2498 stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2499 stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2503 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2504 nfs4_set_claim_prev(open);
2506 * To finish the open response, we just need to set the rflags.
2508 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2509 if (!open->op_stateowner->so_confirmed &&
2510 !nfsd4_has_session(&resp->cstate))
2511 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2517 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2520 struct nfs4_client *clp;
2524 dprintk("process_renew(%08x/%08x): starting\n",
2525 clid->cl_boot, clid->cl_id);
2526 status = nfserr_stale_clientid;
2527 if (STALE_CLIENTID(clid))
2529 clp = find_confirmed_client(clid);
2530 status = nfserr_expired;
2532 /* We assume the client took too long to RENEW. */
2533 dprintk("nfsd4_renew: clientid not found!\n");
2537 status = nfserr_cb_path_down;
2538 if (!list_empty(&clp->cl_delegations)
2539 && !atomic_read(&clp->cl_cb_conn.cb_set))
2543 nfs4_unlock_state();
2547 struct lock_manager nfsd4_manager = {
2551 nfsd4_end_grace(void)
2553 dprintk("NFSD: end of grace period\n");
2554 nfsd4_recdir_purge_old();
2555 locks_end_grace(&nfsd4_manager);
2559 nfs4_laundromat(void)
2561 struct nfs4_client *clp;
2562 struct nfs4_stateowner *sop;
2563 struct nfs4_delegation *dp;
2564 struct list_head *pos, *next, reaplist;
2565 time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2566 time_t t, clientid_val = NFSD_LEASE_TIME;
2567 time_t u, test_val = NFSD_LEASE_TIME;
2571 dprintk("NFSD: laundromat service - starting\n");
2572 if (locks_in_grace())
2574 list_for_each_safe(pos, next, &client_lru) {
2575 clp = list_entry(pos, struct nfs4_client, cl_lru);
2576 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2577 t = clp->cl_time - cutoff;
2578 if (clientid_val > t)
2582 dprintk("NFSD: purging unused client (clientid %08x)\n",
2583 clp->cl_clientid.cl_id);
2584 nfsd4_remove_clid_dir(clp);
2587 INIT_LIST_HEAD(&reaplist);
2588 spin_lock(&recall_lock);
2589 list_for_each_safe(pos, next, &del_recall_lru) {
2590 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2591 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2592 u = dp->dl_time - cutoff;
2597 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2599 list_move(&dp->dl_recall_lru, &reaplist);
2601 spin_unlock(&recall_lock);
2602 list_for_each_safe(pos, next, &reaplist) {
2603 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2604 list_del_init(&dp->dl_recall_lru);
2605 unhash_delegation(dp);
2607 test_val = NFSD_LEASE_TIME;
2608 list_for_each_safe(pos, next, &close_lru) {
2609 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2610 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2611 u = sop->so_time - cutoff;
2616 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2618 release_openowner(sop);
2620 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2621 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2622 nfs4_unlock_state();
2623 return clientid_val;
2626 static struct workqueue_struct *laundry_wq;
2627 static void laundromat_main(struct work_struct *);
2628 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2631 laundromat_main(struct work_struct *not_used)
2635 t = nfs4_laundromat();
2636 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2637 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2640 static struct nfs4_stateowner *
2641 search_close_lru(u32 st_id, int flags)
2643 struct nfs4_stateowner *local = NULL;
2645 if (flags & CLOSE_STATE) {
2646 list_for_each_entry(local, &close_lru, so_close_lru) {
2647 if (local->so_id == st_id)
2655 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2657 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2661 STALE_STATEID(stateid_t *stateid)
2663 if (time_after((unsigned long)boot_time,
2664 (unsigned long)stateid->si_boot)) {
2665 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2666 stateid->si_boot, stateid->si_stateownerid,
2667 stateid->si_fileid, stateid->si_generation);
2674 EXPIRED_STATEID(stateid_t *stateid)
2676 if (time_before((unsigned long)boot_time,
2677 ((unsigned long)stateid->si_boot)) &&
2678 time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2679 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2680 stateid->si_boot, stateid->si_stateownerid,
2681 stateid->si_fileid, stateid->si_generation);
2688 stateid_error_map(stateid_t *stateid)
2690 if (STALE_STATEID(stateid))
2691 return nfserr_stale_stateid;
2692 if (EXPIRED_STATEID(stateid))
2693 return nfserr_expired;
2695 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2696 stateid->si_boot, stateid->si_stateownerid,
2697 stateid->si_fileid, stateid->si_generation);
2698 return nfserr_bad_stateid;
2702 access_permit_read(unsigned long access_bmap)
2704 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2705 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2706 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2710 access_permit_write(unsigned long access_bmap)
2712 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2713 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2717 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2719 __be32 status = nfserr_openmode;
2721 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2723 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2730 static inline __be32
2731 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2733 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2735 else if (locks_in_grace()) {
2736 /* Answer in remaining cases depends on existance of
2737 * conflicting state; so we must wait out the grace period. */
2738 return nfserr_grace;
2739 } else if (flags & WR_STATE)
2740 return nfs4_share_conflict(current_fh,
2741 NFS4_SHARE_DENY_WRITE);
2742 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2743 return nfs4_share_conflict(current_fh,
2744 NFS4_SHARE_DENY_READ);
2748 * Allow READ/WRITE during grace period on recovered state only for files
2749 * that are not able to provide mandatory locking.
2752 grace_disallows_io(struct inode *inode)
2754 return locks_in_grace() && mandatory_lock(inode);
2757 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2760 * When sessions are used the stateid generation number is ignored
2763 if ((flags & HAS_SESSION) && in->si_generation == 0)
2766 /* If the client sends us a stateid from the future, it's buggy: */
2767 if (in->si_generation > ref->si_generation)
2768 return nfserr_bad_stateid;
2770 * The following, however, can happen. For example, if the
2771 * client sends an open and some IO at the same time, the open
2772 * may bump si_generation while the IO is still in flight.
2773 * Thanks to hard links and renames, the client never knows what
2774 * file an open will affect. So it could avoid that situation
2775 * only by serializing all opens and IO from the same open
2776 * owner. To recover from the old_stateid error, the client
2777 * will just have to retry the IO:
2779 if (in->si_generation < ref->si_generation)
2780 return nfserr_old_stateid;
2785 static int is_delegation_stateid(stateid_t *stateid)
2787 return stateid->si_fileid == 0;
2791 * Checks for stateid operations
2794 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2795 stateid_t *stateid, int flags, struct file **filpp)
2797 struct nfs4_stateid *stp = NULL;
2798 struct nfs4_delegation *dp = NULL;
2799 struct svc_fh *current_fh = &cstate->current_fh;
2800 struct inode *ino = current_fh->fh_dentry->d_inode;
2806 if (grace_disallows_io(ino))
2807 return nfserr_grace;
2809 if (nfsd4_has_session(cstate))
2810 flags |= HAS_SESSION;
2812 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2813 return check_special_stateids(current_fh, stateid, flags);
2815 status = nfserr_stale_stateid;
2816 if (STALE_STATEID(stateid))
2819 status = nfserr_bad_stateid;
2820 if (is_delegation_stateid(stateid)) {
2821 dp = find_delegation_stateid(ino, stateid);
2823 status = stateid_error_map(stateid);
2826 status = check_stateid_generation(stateid, &dp->dl_stateid,
2830 status = nfs4_check_delegmode(dp, flags);
2833 renew_client(dp->dl_client);
2835 *filpp = dp->dl_vfs_file;
2836 } else { /* open or lock stateid */
2837 stp = find_stateid(stateid, flags);
2839 status = stateid_error_map(stateid);
2842 if (nfs4_check_fh(current_fh, stp))
2844 if (!stp->st_stateowner->so_confirmed)
2846 status = check_stateid_generation(stateid, &stp->st_stateid,
2850 status = nfs4_check_openmode(stp, flags);
2853 renew_client(stp->st_stateowner->so_client);
2855 *filpp = stp->st_vfs_file;
2865 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2866 RD_STATE : WR_STATE;
2870 * Checks for sequence id mutating operations.
2873 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2874 stateid_t *stateid, int flags,
2875 struct nfs4_stateowner **sopp,
2876 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2878 struct nfs4_stateid *stp;
2879 struct nfs4_stateowner *sop;
2880 struct svc_fh *current_fh = &cstate->current_fh;
2883 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2884 "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2885 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2886 stateid->si_generation);
2891 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2892 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2893 return nfserr_bad_stateid;
2896 if (STALE_STATEID(stateid))
2897 return nfserr_stale_stateid;
2899 if (nfsd4_has_session(cstate))
2900 flags |= HAS_SESSION;
2903 * We return BAD_STATEID if filehandle doesn't match stateid,
2904 * the confirmed flag is incorrecly set, or the generation
2905 * number is incorrect.
2907 stp = find_stateid(stateid, flags);
2910 * Also, we should make sure this isn't just the result of
2913 sop = search_close_lru(stateid->si_stateownerid, flags);
2915 return stateid_error_map(stateid);
2921 *sopp = sop = stp->st_stateowner;
2924 clientid_t *lockclid = &lock->v.new.clientid;
2925 struct nfs4_client *clp = sop->so_client;
2929 lkflg = setlkflg(lock->lk_type);
2931 if (lock->lk_is_new) {
2932 if (!sop->so_is_open_owner)
2933 return nfserr_bad_stateid;
2934 if (!(flags & HAS_SESSION) &&
2935 !same_clid(&clp->cl_clientid, lockclid))
2936 return nfserr_bad_stateid;
2937 /* stp is the open stateid */
2938 status = nfs4_check_openmode(stp, lkflg);
2942 /* stp is the lock stateid */
2943 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2949 if (nfs4_check_fh(current_fh, stp)) {
2950 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2951 return nfserr_bad_stateid;
2955 * We now validate the seqid and stateid generation numbers.
2956 * For the moment, we ignore the possibility of
2957 * generation number wraparound.
2959 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2962 if (sop->so_confirmed && flags & CONFIRM) {
2963 dprintk("NFSD: preprocess_seqid_op: expected"
2964 " unconfirmed stateowner!\n");
2965 return nfserr_bad_stateid;
2967 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2968 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2969 " confirmed yet!\n");
2970 return nfserr_bad_stateid;
2972 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2975 renew_client(sop->so_client);
2979 if (seqid == sop->so_seqid - 1) {
2980 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2981 /* indicate replay to calling function */
2982 return nfserr_replay_me;
2984 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2985 sop->so_seqid, seqid);
2987 return nfserr_bad_seqid;
2991 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2992 struct nfsd4_open_confirm *oc)
2995 struct nfs4_stateowner *sop;
2996 struct nfs4_stateid *stp;
2998 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2999 (int)cstate->current_fh.fh_dentry->d_name.len,
3000 cstate->current_fh.fh_dentry->d_name.name);
3002 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3008 if ((status = nfs4_preprocess_seqid_op(cstate,
3009 oc->oc_seqid, &oc->oc_req_stateid,
3010 CONFIRM | OPEN_STATE,
3011 &oc->oc_stateowner, &stp, NULL)))
3014 sop = oc->oc_stateowner;
3015 sop->so_confirmed = 1;
3016 update_stateid(&stp->st_stateid);
3017 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3018 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3019 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3020 stp->st_stateid.si_boot,
3021 stp->st_stateid.si_stateownerid,
3022 stp->st_stateid.si_fileid,
3023 stp->st_stateid.si_generation);
3025 nfsd4_create_clid_dir(sop->so_client);
3027 if (oc->oc_stateowner) {
3028 nfs4_get_stateowner(oc->oc_stateowner);
3029 cstate->replay_owner = oc->oc_stateowner;
3031 nfs4_unlock_state();
3037 * unset all bits in union bitmap (bmap) that
3038 * do not exist in share (from successful OPEN_DOWNGRADE)
3041 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3044 for (i = 1; i < 4; i++) {
3045 if ((i & access) != i)
3046 __clear_bit(i, bmap);
3051 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3054 for (i = 0; i < 4; i++) {
3055 if ((i & deny) != i)
3056 __clear_bit(i, bmap);
3061 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3062 struct nfsd4_compound_state *cstate,
3063 struct nfsd4_open_downgrade *od)
3066 struct nfs4_stateid *stp;
3067 unsigned int share_access;
3069 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3070 (int)cstate->current_fh.fh_dentry->d_name.len,
3071 cstate->current_fh.fh_dentry->d_name.name);
3073 if (!access_valid(od->od_share_access, cstate->minorversion)
3074 || !deny_valid(od->od_share_deny))
3075 return nfserr_inval;
3078 if ((status = nfs4_preprocess_seqid_op(cstate,
3082 &od->od_stateowner, &stp, NULL)))
3085 status = nfserr_inval;
3086 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3087 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3088 stp->st_access_bmap, od->od_share_access);
3091 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3092 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3093 stp->st_deny_bmap, od->od_share_deny);
3096 set_access(&share_access, stp->st_access_bmap);
3097 nfs4_file_downgrade(stp->st_vfs_file,
3098 share_access & ~od->od_share_access);
3100 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3101 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3103 update_stateid(&stp->st_stateid);
3104 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3107 if (od->od_stateowner) {
3108 nfs4_get_stateowner(od->od_stateowner);
3109 cstate->replay_owner = od->od_stateowner;
3111 nfs4_unlock_state();
3116 * nfs4_unlock_state() called after encode
3119 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3120 struct nfsd4_close *close)
3123 struct nfs4_stateid *stp;
3125 dprintk("NFSD: nfsd4_close on file %.*s\n",
3126 (int)cstate->current_fh.fh_dentry->d_name.len,
3127 cstate->current_fh.fh_dentry->d_name.name);
3130 /* check close_lru for replay */
3131 if ((status = nfs4_preprocess_seqid_op(cstate,
3134 OPEN_STATE | CLOSE_STATE,
3135 &close->cl_stateowner, &stp, NULL)))
3138 update_stateid(&stp->st_stateid);
3139 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3141 /* release_stateid() calls nfsd_close() if needed */
3142 release_open_stateid(stp);
3144 /* place unused nfs4_stateowners on so_close_lru list to be
3145 * released by the laundromat service after the lease period
3146 * to enable us to handle CLOSE replay
3148 if (list_empty(&close->cl_stateowner->so_stateids))
3149 move_to_close_lru(close->cl_stateowner);
3151 if (close->cl_stateowner) {
3152 nfs4_get_stateowner(close->cl_stateowner);
3153 cstate->replay_owner = close->cl_stateowner;
3155 nfs4_unlock_state();
3160 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3161 struct nfsd4_delegreturn *dr)
3163 struct nfs4_delegation *dp;
3164 stateid_t *stateid = &dr->dr_stateid;
3165 struct inode *inode;
3169 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3171 inode = cstate->current_fh.fh_dentry->d_inode;
3173 if (nfsd4_has_session(cstate))
3174 flags |= HAS_SESSION;
3176 status = nfserr_bad_stateid;
3177 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3179 status = nfserr_stale_stateid;
3180 if (STALE_STATEID(stateid))
3182 status = nfserr_bad_stateid;
3183 if (!is_delegation_stateid(stateid))
3185 dp = find_delegation_stateid(inode, stateid);
3187 status = stateid_error_map(stateid);
3190 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3193 renew_client(dp->dl_client);
3195 unhash_delegation(dp);
3197 nfs4_unlock_state();
3204 * Lock owner state (byte-range locks)
3206 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3207 #define LOCK_HASH_BITS 8
3208 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3209 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3212 end_offset(u64 start, u64 len)
3217 return end >= start ? end: NFS4_MAX_UINT64;
3220 /* last octet in a range */
3222 last_byte_offset(u64 start, u64 len)
3228 return end > start ? end - 1: NFS4_MAX_UINT64;
3231 #define lockownerid_hashval(id) \
3232 ((id) & LOCK_HASH_MASK)
3234 static inline unsigned int
3235 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3236 struct xdr_netobj *ownername)
3238 return (file_hashval(inode) + cl_id
3239 + opaque_hashval(ownername->data, ownername->len))
3243 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3244 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3245 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3247 static struct nfs4_stateid *
3248 find_stateid(stateid_t *stid, int flags)
3250 struct nfs4_stateid *local;
3251 u32 st_id = stid->si_stateownerid;
3252 u32 f_id = stid->si_fileid;
3253 unsigned int hashval;
3255 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3256 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3257 hashval = stateid_hashval(st_id, f_id);
3258 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3259 if ((local->st_stateid.si_stateownerid == st_id) &&
3260 (local->st_stateid.si_fileid == f_id))
3265 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3266 hashval = stateid_hashval(st_id, f_id);
3267 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3268 if ((local->st_stateid.si_stateownerid == st_id) &&
3269 (local->st_stateid.si_fileid == f_id))
3276 static struct nfs4_delegation *
3277 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3279 struct nfs4_file *fp;
3280 struct nfs4_delegation *dl;
3282 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3283 stid->si_boot, stid->si_stateownerid,
3284 stid->si_fileid, stid->si_generation);
3286 fp = find_file(ino);
3289 dl = find_delegation_file(fp, stid);
3295 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3296 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3297 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3298 * locking, this prevents us from being completely protocol-compliant. The
3299 * real solution to this problem is to start using unsigned file offsets in
3300 * the VFS, but this is a very deep change!
3303 nfs4_transform_lock_offset(struct file_lock *lock)
3305 if (lock->fl_start < 0)
3306 lock->fl_start = OFFSET_MAX;
3307 if (lock->fl_end < 0)
3308 lock->fl_end = OFFSET_MAX;
3311 /* Hack!: For now, we're defining this just so we can use a pointer to it
3312 * as a unique cookie to identify our (NFSv4's) posix locks. */
3313 static struct lock_manager_operations nfsd_posix_mng_ops = {
3317 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3319 struct nfs4_stateowner *sop;
3322 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3323 sop = (struct nfs4_stateowner *) fl->fl_owner;
3324 hval = lockownerid_hashval(sop->so_id);
3325 kref_get(&sop->so_ref);
3327 deny->ld_clientid = sop->so_client->cl_clientid;
3329 deny->ld_sop = NULL;
3330 deny->ld_clientid.cl_boot = 0;
3331 deny->ld_clientid.cl_id = 0;
3333 deny->ld_start = fl->fl_start;
3334 deny->ld_length = NFS4_MAX_UINT64;
3335 if (fl->fl_end != NFS4_MAX_UINT64)
3336 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3337 deny->ld_type = NFS4_READ_LT;
3338 if (fl->fl_type != F_RDLCK)
3339 deny->ld_type = NFS4_WRITE_LT;
3342 static struct nfs4_stateowner *
3343 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3344 struct xdr_netobj *owner)
3346 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3347 struct nfs4_stateowner *op;
3349 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3350 if (same_owner_str(op, owner, clid))
3357 * Alloc a lock owner structure.
3358 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3361 * strhashval = lock_ownerstr_hashval
3364 static struct nfs4_stateowner *
3365 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3366 struct nfs4_stateowner *sop;
3367 struct nfs4_replay *rp;
3368 unsigned int idhashval;
3370 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3372 idhashval = lockownerid_hashval(current_ownerid);
3373 INIT_LIST_HEAD(&sop->so_idhash);
3374 INIT_LIST_HEAD(&sop->so_strhash);
3375 INIT_LIST_HEAD(&sop->so_perclient);
3376 INIT_LIST_HEAD(&sop->so_stateids);
3377 INIT_LIST_HEAD(&sop->so_perstateid);
3378 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3380 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3381 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3382 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3383 sop->so_is_open_owner = 0;
3384 sop->so_id = current_ownerid++;
3385 sop->so_client = clp;
3386 /* It is the openowner seqid that will be incremented in encode in the
3387 * case of new lockowners; so increment the lock seqid manually: */
3388 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3389 sop->so_confirmed = 1;
3390 rp = &sop->so_replay;
3391 rp->rp_status = nfserr_serverfault;
3393 rp->rp_buf = rp->rp_ibuf;
3397 static struct nfs4_stateid *
3398 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3400 struct nfs4_stateid *stp;
3401 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3403 stp = nfs4_alloc_stateid();
3406 INIT_LIST_HEAD(&stp->st_hash);
3407 INIT_LIST_HEAD(&stp->st_perfile);
3408 INIT_LIST_HEAD(&stp->st_perstateowner);
3409 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3410 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3411 list_add(&stp->st_perfile, &fp->fi_stateids);
3412 list_add(&stp->st_perstateowner, &sop->so_stateids);
3413 stp->st_stateowner = sop;
3416 stp->st_stateid.si_boot = get_seconds();
3417 stp->st_stateid.si_stateownerid = sop->so_id;
3418 stp->st_stateid.si_fileid = fp->fi_id;
3419 stp->st_stateid.si_generation = 0;
3420 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3421 stp->st_access_bmap = open_stp->st_access_bmap;
3422 stp->st_deny_bmap = open_stp->st_deny_bmap;
3423 stp->st_openstp = open_stp;
3430 check_lock_length(u64 offset, u64 length)
3432 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3433 LOFF_OVERFLOW(offset, length)));
3440 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3441 struct nfsd4_lock *lock)
3443 struct nfs4_stateowner *open_sop = NULL;
3444 struct nfs4_stateowner *lock_sop = NULL;
3445 struct nfs4_stateid *lock_stp;
3447 struct file_lock file_lock;
3448 struct file_lock conflock;
3450 unsigned int strhashval;
3454 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3455 (long long) lock->lk_offset,
3456 (long long) lock->lk_length);
3458 if (check_lock_length(lock->lk_offset, lock->lk_length))
3459 return nfserr_inval;
3461 if ((status = fh_verify(rqstp, &cstate->current_fh,
3462 S_IFREG, NFSD_MAY_LOCK))) {
3463 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3469 if (lock->lk_is_new) {
3471 * Client indicates that this is a new lockowner.
3472 * Use open owner and open stateid to create lock owner and
3475 struct nfs4_stateid *open_stp = NULL;
3476 struct nfs4_file *fp;
3478 status = nfserr_stale_clientid;
3479 if (!nfsd4_has_session(cstate) &&
3480 STALE_CLIENTID(&lock->lk_new_clientid))
3483 /* validate and update open stateid and open seqid */
3484 status = nfs4_preprocess_seqid_op(cstate,
3485 lock->lk_new_open_seqid,
3486 &lock->lk_new_open_stateid,
3488 &lock->lk_replay_owner, &open_stp,
3492 open_sop = lock->lk_replay_owner;
3493 /* create lockowner and lock stateid */
3494 fp = open_stp->st_file;
3495 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3496 open_sop->so_client->cl_clientid.cl_id,
3497 &lock->v.new.owner);
3498 /* XXX: Do we need to check for duplicate stateowners on
3499 * the same file, or should they just be allowed (and
3500 * create new stateids)? */
3501 status = nfserr_resource;
3502 lock_sop = alloc_init_lock_stateowner(strhashval,
3503 open_sop->so_client, open_stp, lock);
3504 if (lock_sop == NULL)
3506 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3507 if (lock_stp == NULL)
3510 /* lock (lock owner + lock stateid) already exists */
3511 status = nfs4_preprocess_seqid_op(cstate,
3512 lock->lk_old_lock_seqid,
3513 &lock->lk_old_lock_stateid,
3515 &lock->lk_replay_owner, &lock_stp, lock);
3518 lock_sop = lock->lk_replay_owner;
3520 /* lock->lk_replay_owner and lock_stp have been created or found */
3521 filp = lock_stp->st_vfs_file;
3523 status = nfserr_grace;
3524 if (locks_in_grace() && !lock->lk_reclaim)
3526 status = nfserr_no_grace;
3527 if (!locks_in_grace() && lock->lk_reclaim)
3530 locks_init_lock(&file_lock);
3531 switch (lock->lk_type) {
3534 file_lock.fl_type = F_RDLCK;
3538 case NFS4_WRITEW_LT:
3539 file_lock.fl_type = F_WRLCK;
3543 status = nfserr_inval;
3546 file_lock.fl_owner = (fl_owner_t)lock_sop;
3547 file_lock.fl_pid = current->tgid;
3548 file_lock.fl_file = filp;
3549 file_lock.fl_flags = FL_POSIX;
3550 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3552 file_lock.fl_start = lock->lk_offset;
3553 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3554 nfs4_transform_lock_offset(&file_lock);
3557 * Try to lock the file in the VFS.
3558 * Note: locks.c uses the BKL to protect the inode's lock list.
3561 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3563 case 0: /* success! */
3564 update_stateid(&lock_stp->st_stateid);
3565 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3569 case (EAGAIN): /* conflock holds conflicting lock */
3570 status = nfserr_denied;
3571 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3572 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3575 status = nfserr_deadlock;
3578 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3579 status = nfserr_resource;
3583 if (status && lock->lk_is_new && lock_sop)
3584 release_lockowner(lock_sop);
3585 if (lock->lk_replay_owner) {
3586 nfs4_get_stateowner(lock->lk_replay_owner);
3587 cstate->replay_owner = lock->lk_replay_owner;
3589 nfs4_unlock_state();
3594 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3595 * so we do a temporary open here just to get an open file to pass to
3596 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3599 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3604 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3607 err = vfs_test_lock(file, lock);
3616 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3617 struct nfsd4_lockt *lockt)
3619 struct inode *inode;
3620 struct file_lock file_lock;
3624 if (locks_in_grace())
3625 return nfserr_grace;
3627 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3628 return nfserr_inval;
3630 lockt->lt_stateowner = NULL;
3633 status = nfserr_stale_clientid;
3634 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3637 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3638 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3639 if (status == nfserr_symlink)
3640 status = nfserr_inval;
3644 inode = cstate->current_fh.fh_dentry->d_inode;
3645 locks_init_lock(&file_lock);
3646 switch (lockt->lt_type) {
3649 file_lock.fl_type = F_RDLCK;
3652 case NFS4_WRITEW_LT:
3653 file_lock.fl_type = F_WRLCK;
3656 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3657 status = nfserr_inval;
3661 lockt->lt_stateowner = find_lockstateowner_str(inode,
3662 &lockt->lt_clientid, &lockt->lt_owner);
3663 if (lockt->lt_stateowner)
3664 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3665 file_lock.fl_pid = current->tgid;
3666 file_lock.fl_flags = FL_POSIX;
3668 file_lock.fl_start = lockt->lt_offset;
3669 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3671 nfs4_transform_lock_offset(&file_lock);
3674 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3676 status = nfserrno(error);
3679 if (file_lock.fl_type != F_UNLCK) {
3680 status = nfserr_denied;
3681 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3684 nfs4_unlock_state();
3689 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3690 struct nfsd4_locku *locku)
3692 struct nfs4_stateid *stp;
3693 struct file *filp = NULL;
3694 struct file_lock file_lock;
3698 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3699 (long long) locku->lu_offset,
3700 (long long) locku->lu_length);
3702 if (check_lock_length(locku->lu_offset, locku->lu_length))
3703 return nfserr_inval;
3707 if ((status = nfs4_preprocess_seqid_op(cstate,
3711 &locku->lu_stateowner, &stp, NULL)))
3714 filp = stp->st_vfs_file;
3716 locks_init_lock(&file_lock);
3717 file_lock.fl_type = F_UNLCK;
3718 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3719 file_lock.fl_pid = current->tgid;
3720 file_lock.fl_file = filp;
3721 file_lock.fl_flags = FL_POSIX;
3722 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3723 file_lock.fl_start = locku->lu_offset;
3725 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3726 nfs4_transform_lock_offset(&file_lock);
3729 * Try to unlock the file in the VFS.
3731 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3733 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3737 * OK, unlock succeeded; the only thing left to do is update the stateid.
3739 update_stateid(&stp->st_stateid);
3740 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3743 if (locku->lu_stateowner) {
3744 nfs4_get_stateowner(locku->lu_stateowner);
3745 cstate->replay_owner = locku->lu_stateowner;
3747 nfs4_unlock_state();
3751 status = nfserrno(err);
3757 * 1: locks held by lockowner
3758 * 0: no locks held by lockowner
3761 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3763 struct file_lock **flpp;
3764 struct inode *inode = filp->f_path.dentry->d_inode;
3768 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3769 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3780 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3781 struct nfsd4_compound_state *cstate,
3782 struct nfsd4_release_lockowner *rlockowner)
3784 clientid_t *clid = &rlockowner->rl_clientid;
3785 struct nfs4_stateowner *sop;
3786 struct nfs4_stateid *stp;
3787 struct xdr_netobj *owner = &rlockowner->rl_owner;
3788 struct list_head matches;
3792 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3793 clid->cl_boot, clid->cl_id);
3795 /* XXX check for lease expiration */
3797 status = nfserr_stale_clientid;
3798 if (STALE_CLIENTID(clid))
3803 status = nfserr_locks_held;
3804 /* XXX: we're doing a linear search through all the lockowners.
3805 * Yipes! For now we'll just hope clients aren't really using
3806 * release_lockowner much, but eventually we have to fix these
3807 * data structures. */
3808 INIT_LIST_HEAD(&matches);
3809 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3810 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3811 if (!same_owner_str(sop, owner, clid))
3813 list_for_each_entry(stp, &sop->so_stateids,
3815 if (check_for_locks(stp->st_vfs_file, sop))
3817 /* Note: so_perclient unused for lockowners,
3818 * so it's OK to fool with here. */
3819 list_add(&sop->so_perclient, &matches);
3823 /* Clients probably won't expect us to return with some (but not all)
3824 * of the lockowner state released; so don't release any until all
3825 * have been checked. */
3827 while (!list_empty(&matches)) {
3828 sop = list_entry(matches.next, struct nfs4_stateowner,
3830 /* unhash_stateowner deletes so_perclient only
3831 * for openowners. */
3832 list_del(&sop->so_perclient);
3833 release_lockowner(sop);
3836 nfs4_unlock_state();
3840 static inline struct nfs4_client_reclaim *
3843 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3847 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3849 unsigned int strhashval = clientstr_hashval(name);
3850 struct nfs4_client *clp;
3852 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3857 * failure => all reset bets are off, nfserr_no_grace...
3860 nfs4_client_to_reclaim(const char *name)
3862 unsigned int strhashval;
3863 struct nfs4_client_reclaim *crp = NULL;
3865 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3866 crp = alloc_reclaim();
3869 strhashval = clientstr_hashval(name);
3870 INIT_LIST_HEAD(&crp->cr_strhash);
3871 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3872 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3873 reclaim_str_hashtbl_size++;
3878 nfs4_release_reclaim(void)
3880 struct nfs4_client_reclaim *crp = NULL;
3883 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3884 while (!list_empty(&reclaim_str_hashtbl[i])) {
3885 crp = list_entry(reclaim_str_hashtbl[i].next,
3886 struct nfs4_client_reclaim, cr_strhash);
3887 list_del(&crp->cr_strhash);
3889 reclaim_str_hashtbl_size--;
3892 BUG_ON(reclaim_str_hashtbl_size);
3896 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3897 static struct nfs4_client_reclaim *
3898 nfs4_find_reclaim_client(clientid_t *clid)
3900 unsigned int strhashval;
3901 struct nfs4_client *clp;
3902 struct nfs4_client_reclaim *crp = NULL;
3905 /* find clientid in conf_id_hashtbl */
3906 clp = find_confirmed_client(clid);
3910 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3911 clp->cl_name.len, clp->cl_name.data,
3914 /* find clp->cl_name in reclaim_str_hashtbl */
3915 strhashval = clientstr_hashval(clp->cl_recdir);
3916 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3917 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3925 * Called from OPEN. Look for clientid in reclaim list.
3928 nfs4_check_open_reclaim(clientid_t *clid)
3930 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3933 /* initialization to perform at module load time: */
3936 nfs4_state_init(void)
3940 status = nfsd4_init_slabs();
3943 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3944 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3945 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3946 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3947 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3948 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3950 for (i = 0; i < SESSION_HASH_SIZE; i++)
3951 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3952 for (i = 0; i < FILE_HASH_SIZE; i++) {
3953 INIT_LIST_HEAD(&file_hashtbl[i]);
3955 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3956 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3957 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3959 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3960 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3961 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3963 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3964 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3965 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3967 memset(&onestateid, ~0, sizeof(stateid_t));
3968 INIT_LIST_HEAD(&close_lru);
3969 INIT_LIST_HEAD(&client_lru);
3970 INIT_LIST_HEAD(&del_recall_lru);
3971 reclaim_str_hashtbl_size = 0;
3976 nfsd4_load_reboot_recovery_data(void)
3981 nfsd4_init_recdir(user_recovery_dirname);
3982 status = nfsd4_recdir_load();
3983 nfs4_unlock_state();
3985 printk("NFSD: Failure reading reboot recovery data\n");
3989 get_nfs4_grace_period(void)
3991 return max(user_lease_time, lease_time) * HZ;
3995 * Since the lifetime of a delegation isn't limited to that of an open, a
3996 * client may quite reasonably hang on to a delegation as long as it has
3997 * the inode cached. This becomes an obvious problem the first time a
3998 * client's inode cache approaches the size of the server's total memory.
4000 * For now we avoid this problem by imposing a hard limit on the number
4001 * of delegations, which varies according to the server's memory size.
4004 set_max_delegations(void)
4007 * Allow at most 4 delegations per megabyte of RAM. Quick
4008 * estimates suggest that in the worst case (where every delegation
4009 * is for a different inode), a delegation could take about 1.5K,
4010 * giving a worst case usage of about 6% of memory.
4012 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4015 /* initialization to perform when the nfsd service is started: */
4018 __nfs4_state_start(void)
4020 unsigned long grace_time;
4022 boot_time = get_seconds();
4023 grace_time = get_nfs4_grace_period();
4024 lease_time = user_lease_time;
4025 locks_start_grace(&nfsd4_manager);
4026 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4028 laundry_wq = create_singlethread_workqueue("nfsd4");
4029 if (laundry_wq == NULL)
4031 queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4032 set_max_delegations();
4033 return set_callback_cred();
4037 nfs4_state_start(void)
4043 nfsd4_load_reboot_recovery_data();
4044 ret = __nfs4_state_start();
4052 nfs4_lease_time(void)
4058 __nfs4_state_shutdown(void)
4061 struct nfs4_client *clp = NULL;
4062 struct nfs4_delegation *dp = NULL;
4063 struct list_head *pos, *next, reaplist;
4065 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4066 while (!list_empty(&conf_id_hashtbl[i])) {
4067 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4070 while (!list_empty(&unconf_str_hashtbl[i])) {
4071 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4075 INIT_LIST_HEAD(&reaplist);
4076 spin_lock(&recall_lock);
4077 list_for_each_safe(pos, next, &del_recall_lru) {
4078 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4079 list_move(&dp->dl_recall_lru, &reaplist);
4081 spin_unlock(&recall_lock);
4082 list_for_each_safe(pos, next, &reaplist) {
4083 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4084 list_del_init(&dp->dl_recall_lru);
4085 unhash_delegation(dp);
4088 nfsd4_shutdown_recdir();
4093 nfs4_state_shutdown(void)
4095 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4096 destroy_workqueue(laundry_wq);
4097 locks_end_grace(&nfsd4_manager);
4099 nfs4_release_reclaim();
4100 __nfs4_state_shutdown();
4101 nfs4_unlock_state();
4105 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4106 * accessed when nfsd is starting.
4109 nfs4_set_recdir(char *recdir)
4111 strcpy(user_recovery_dirname, recdir);
4115 * Change the NFSv4 recovery directory to recdir.
4118 nfs4_reset_recoverydir(char *recdir)
4123 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4127 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4128 nfs4_set_recdir(recdir);
4136 nfs4_recoverydir(void)
4138 return user_recovery_dirname;
4142 * Called when leasetime is changed.
4144 * The only way the protocol gives us to handle on-the-fly lease changes is to
4145 * simulate a reboot. Instead of doing that, we just wait till the next time
4146 * we start to register any changes in lease time. If the administrator
4147 * really wants to change the lease time *now*, they can go ahead and bring
4148 * nfsd down and then back up again after changing the lease time.
4150 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4151 * when nfsd is starting
4154 nfs4_reset_lease(time_t leasetime)
4156 user_lease_time = leasetime;