2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
36 #include <linux/smp_lock.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
45 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 time_t nfsd4_lease = 90; /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
55 static stateid_t zerostateid; /* bits all 0 */
56 static stateid_t onestateid; /* bits all 1 */
57 static u64 current_sessionid = 1;
59 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
60 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
62 /* forward declarations */
63 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
64 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
65 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
66 static void nfs4_set_recdir(char *recdir);
70 /* Currently used for almost all code touching nfsv4 state: */
71 static DEFINE_MUTEX(client_mutex);
74 * Currently used for the del_recall_lru and file hash table. In an
75 * effort to decrease the scope of the client_mutex, this spinlock may
76 * eventually cover more:
78 static DEFINE_SPINLOCK(recall_lock);
80 static struct kmem_cache *stateowner_slab = NULL;
81 static struct kmem_cache *file_slab = NULL;
82 static struct kmem_cache *stateid_slab = NULL;
83 static struct kmem_cache *deleg_slab = NULL;
88 mutex_lock(&client_mutex);
92 nfs4_unlock_state(void)
94 mutex_unlock(&client_mutex);
98 opaque_hashval(const void *ptr, int nbytes)
100 unsigned char *cptr = (unsigned char *) ptr;
110 static struct list_head del_recall_lru;
113 put_nfs4_file(struct nfs4_file *fi)
115 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
116 list_del(&fi->fi_hash);
117 spin_unlock(&recall_lock);
119 kmem_cache_free(file_slab, fi);
124 get_nfs4_file(struct nfs4_file *fi)
126 atomic_inc(&fi->fi_ref);
129 static int num_delegations;
130 unsigned int max_delegations;
133 * Open owner state (share locks)
136 /* hash tables for nfs4_stateowner */
137 #define OWNER_HASH_BITS 8
138 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
139 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
141 #define ownerid_hashval(id) \
142 ((id) & OWNER_HASH_MASK)
143 #define ownerstr_hashval(clientid, ownername) \
144 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
146 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
147 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
149 /* hash table for nfs4_file */
150 #define FILE_HASH_BITS 8
151 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
152 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
153 /* hash table for (open)nfs4_stateid */
154 #define STATEID_HASH_BITS 10
155 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
156 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
158 #define file_hashval(x) \
159 hash_ptr(x, FILE_HASH_BITS)
160 #define stateid_hashval(owner_id, file_id) \
161 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
163 static struct list_head file_hashtbl[FILE_HASH_SIZE];
164 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
166 static struct nfs4_delegation *
167 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
169 struct nfs4_delegation *dp;
170 struct nfs4_file *fp = stp->st_file;
171 struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
173 dprintk("NFSD alloc_init_deleg\n");
174 if (fp->fi_had_conflict)
176 if (num_delegations > max_delegations)
178 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
182 INIT_LIST_HEAD(&dp->dl_perfile);
183 INIT_LIST_HEAD(&dp->dl_perclnt);
184 INIT_LIST_HEAD(&dp->dl_recall_lru);
189 get_file(stp->st_vfs_file);
190 dp->dl_vfs_file = stp->st_vfs_file;
192 dp->dl_ident = cb->cb_ident;
193 dp->dl_stateid.si_boot = get_seconds();
194 dp->dl_stateid.si_stateownerid = current_delegid++;
195 dp->dl_stateid.si_fileid = 0;
196 dp->dl_stateid.si_generation = 0;
197 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
199 atomic_set(&dp->dl_count, 1);
200 list_add(&dp->dl_perfile, &fp->fi_delegations);
201 list_add(&dp->dl_perclnt, &clp->cl_delegations);
202 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
207 nfs4_put_delegation(struct nfs4_delegation *dp)
209 if (atomic_dec_and_test(&dp->dl_count)) {
210 dprintk("NFSD: freeing dp %p\n",dp);
211 put_nfs4_file(dp->dl_file);
212 kmem_cache_free(deleg_slab, dp);
217 /* Remove the associated file_lock first, then remove the delegation.
218 * lease_modify() is called to remove the FS_LEASE file_lock from
219 * the i_flock list, eventually calling nfsd's lock_manager
220 * fl_release_callback.
223 nfs4_close_delegation(struct nfs4_delegation *dp)
225 struct file *filp = dp->dl_vfs_file;
227 dprintk("NFSD: close_delegation dp %p\n",dp);
228 dp->dl_vfs_file = NULL;
229 /* The following nfsd_close may not actually close the file,
230 * but we want to remove the lease in any case. */
232 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
236 /* Called under the state lock. */
238 unhash_delegation(struct nfs4_delegation *dp)
240 list_del_init(&dp->dl_perfile);
241 list_del_init(&dp->dl_perclnt);
242 spin_lock(&recall_lock);
243 list_del_init(&dp->dl_recall_lru);
244 spin_unlock(&recall_lock);
245 nfs4_close_delegation(dp);
246 nfs4_put_delegation(dp);
253 /* client_lock protects the client lru list and session hash table */
254 static DEFINE_SPINLOCK(client_lock);
256 /* Hash tables for nfs4_clientid state */
257 #define CLIENT_HASH_BITS 4
258 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
259 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
261 #define clientid_hashval(id) \
262 ((id) & CLIENT_HASH_MASK)
263 #define clientstr_hashval(name) \
264 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
266 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
267 * used in reboot/reset lease grace period processing
269 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
270 * setclientid_confirmed info.
272 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
275 * client_lru holds client queue ordered by nfs4_client.cl_time
278 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
279 * for last close replay.
281 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
282 static int reclaim_str_hashtbl_size = 0;
283 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
284 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
285 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
286 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
287 static struct list_head client_lru;
288 static struct list_head close_lru;
290 static void unhash_generic_stateid(struct nfs4_stateid *stp)
292 list_del(&stp->st_hash);
293 list_del(&stp->st_perfile);
294 list_del(&stp->st_perstateowner);
297 static void free_generic_stateid(struct nfs4_stateid *stp)
299 put_nfs4_file(stp->st_file);
300 kmem_cache_free(stateid_slab, stp);
303 static void release_lock_stateid(struct nfs4_stateid *stp)
305 unhash_generic_stateid(stp);
306 locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
307 free_generic_stateid(stp);
310 static void unhash_lockowner(struct nfs4_stateowner *sop)
312 struct nfs4_stateid *stp;
314 list_del(&sop->so_idhash);
315 list_del(&sop->so_strhash);
316 list_del(&sop->so_perstateid);
317 while (!list_empty(&sop->so_stateids)) {
318 stp = list_first_entry(&sop->so_stateids,
319 struct nfs4_stateid, st_perstateowner);
320 release_lock_stateid(stp);
324 static void release_lockowner(struct nfs4_stateowner *sop)
326 unhash_lockowner(sop);
327 nfs4_put_stateowner(sop);
331 release_stateid_lockowners(struct nfs4_stateid *open_stp)
333 struct nfs4_stateowner *lock_sop;
335 while (!list_empty(&open_stp->st_lockowners)) {
336 lock_sop = list_entry(open_stp->st_lockowners.next,
337 struct nfs4_stateowner, so_perstateid);
338 /* list_del(&open_stp->st_lockowners); */
339 BUG_ON(lock_sop->so_is_open_owner);
340 release_lockowner(lock_sop);
344 static void release_open_stateid(struct nfs4_stateid *stp)
346 unhash_generic_stateid(stp);
347 release_stateid_lockowners(stp);
348 nfsd_close(stp->st_vfs_file);
349 free_generic_stateid(stp);
352 static void unhash_openowner(struct nfs4_stateowner *sop)
354 struct nfs4_stateid *stp;
356 list_del(&sop->so_idhash);
357 list_del(&sop->so_strhash);
358 list_del(&sop->so_perclient);
359 list_del(&sop->so_perstateid); /* XXX: necessary? */
360 while (!list_empty(&sop->so_stateids)) {
361 stp = list_first_entry(&sop->so_stateids,
362 struct nfs4_stateid, st_perstateowner);
363 release_open_stateid(stp);
367 static void release_openowner(struct nfs4_stateowner *sop)
369 unhash_openowner(sop);
370 list_del(&sop->so_close_lru);
371 nfs4_put_stateowner(sop);
374 #define SESSION_HASH_SIZE 512
375 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
378 hash_sessionid(struct nfs4_sessionid *sessionid)
380 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
382 return sid->sequence % SESSION_HASH_SIZE;
386 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
388 u32 *ptr = (u32 *)(&sessionid->data[0]);
389 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
393 gen_sessionid(struct nfsd4_session *ses)
395 struct nfs4_client *clp = ses->se_client;
396 struct nfsd4_sessionid *sid;
398 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
399 sid->clientid = clp->cl_clientid;
400 sid->sequence = current_sessionid++;
405 * The protocol defines ca_maxresponssize_cached to include the size of
406 * the rpc header, but all we need to cache is the data starting after
407 * the end of the initial SEQUENCE operation--the rest we regenerate
408 * each time. Therefore we can advertise a ca_maxresponssize_cached
409 * value that is the number of bytes in our cache plus a few additional
410 * bytes. In order to stay on the safe side, and not promise more than
411 * we can cache, those additional bytes must be the minimum possible: 24
412 * bytes of rpc header (xid through accept state, with AUTH_NULL
413 * verifier), 12 for the compound header (with zero-length tag), and 44
414 * for the SEQUENCE op response:
416 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
419 * Give the client the number of ca_maxresponsesize_cached slots it
420 * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
421 * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
422 * than NFSD_MAX_SLOTS_PER_SESSION.
424 * If we run out of reserved DRC memory we should (up to a point)
425 * re-negotiate active sessions and reduce their slot usage to make
426 * rooom for new connections. For now we just fail the create session.
428 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
430 int mem, size = fchan->maxresp_cached;
432 if (fchan->maxreqs < 1)
435 if (size < NFSD_MIN_HDR_SEQ_SZ)
436 size = NFSD_MIN_HDR_SEQ_SZ;
437 size -= NFSD_MIN_HDR_SEQ_SZ;
438 if (size > NFSD_SLOT_CACHE_SIZE)
439 size = NFSD_SLOT_CACHE_SIZE;
441 /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
442 mem = fchan->maxreqs * size;
443 if (mem > NFSD_MAX_MEM_PER_SESSION) {
444 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
445 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
446 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
447 mem = fchan->maxreqs * size;
450 spin_lock(&nfsd_drc_lock);
451 /* bound the total session drc memory ussage */
452 if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
453 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
454 mem = fchan->maxreqs * size;
456 nfsd_drc_mem_used += mem;
457 spin_unlock(&nfsd_drc_lock);
459 if (fchan->maxreqs == 0)
460 return nfserr_serverfault;
462 fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
467 * fchan holds the client values on input, and the server values on output
468 * sv_max_mesg is the maximum payload plus one page for overhead.
470 static int init_forechannel_attrs(struct svc_rqst *rqstp,
471 struct nfsd4_channel_attrs *session_fchan,
472 struct nfsd4_channel_attrs *fchan)
475 __u32 maxcount = nfsd_serv->sv_max_mesg;
477 /* headerpadsz set to zero in encode routine */
479 /* Use the client's max request and max response size if possible */
480 if (fchan->maxreq_sz > maxcount)
481 fchan->maxreq_sz = maxcount;
482 session_fchan->maxreq_sz = fchan->maxreq_sz;
484 if (fchan->maxresp_sz > maxcount)
485 fchan->maxresp_sz = maxcount;
486 session_fchan->maxresp_sz = fchan->maxresp_sz;
488 /* Use the client's maxops if possible */
489 if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
490 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
491 session_fchan->maxops = fchan->maxops;
493 /* FIXME: Error means no more DRC pages so the server should
494 * recover pages from existing sessions. For now fail session
497 status = set_forechannel_drc_size(fchan);
499 session_fchan->maxresp_cached = fchan->maxresp_cached;
500 session_fchan->maxreqs = fchan->maxreqs;
502 dprintk("%s status %d\n", __func__, status);
507 free_session_slots(struct nfsd4_session *ses)
511 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
512 kfree(ses->se_slots[i]);
516 * We don't actually need to cache the rpc and session headers, so we
517 * can allocate a little less for each slot:
519 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
521 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
525 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
526 struct nfsd4_create_session *cses)
528 struct nfsd4_session *new, tmp;
529 struct nfsd4_slot *sp;
530 int idx, slotsize, cachesize, i;
533 memset(&tmp, 0, sizeof(tmp));
535 /* FIXME: For now, we just accept the client back channel attributes. */
536 tmp.se_bchannel = cses->back_channel;
537 status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
538 &cses->fore_channel);
542 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
543 + sizeof(struct nfsd4_session) > PAGE_SIZE);
545 status = nfserr_serverfault;
546 /* allocate struct nfsd4_session and slot table pointers in one piece */
547 slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
548 new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
552 memcpy(new, &tmp, sizeof(*new));
554 /* allocate each struct nfsd4_slot and data cache in one piece */
555 cachesize = slot_bytes(&new->se_fchannel);
556 for (i = 0; i < new->se_fchannel.maxreqs; i++) {
557 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
560 new->se_slots[i] = sp;
563 new->se_client = clp;
565 idx = hash_sessionid(&new->se_sessionid);
566 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
567 NFS4_MAX_SESSIONID_LEN);
569 new->se_flags = cses->flags;
570 kref_init(&new->se_ref);
571 spin_lock(&client_lock);
572 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
573 list_add(&new->se_perclnt, &clp->cl_sessions);
574 spin_unlock(&client_lock);
580 free_session_slots(new);
585 /* caller must hold client_lock */
586 static struct nfsd4_session *
587 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
589 struct nfsd4_session *elem;
592 dump_sessionid(__func__, sessionid);
593 idx = hash_sessionid(sessionid);
594 dprintk("%s: idx is %d\n", __func__, idx);
595 /* Search in the appropriate list */
596 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
597 dump_sessionid("list traversal", &elem->se_sessionid);
598 if (!memcmp(elem->se_sessionid.data, sessionid->data,
599 NFS4_MAX_SESSIONID_LEN)) {
604 dprintk("%s: session not found\n", __func__);
608 /* caller must hold client_lock */
610 unhash_session(struct nfsd4_session *ses)
612 list_del(&ses->se_hash);
613 list_del(&ses->se_perclnt);
617 free_session(struct kref *kref)
619 struct nfsd4_session *ses;
622 ses = container_of(kref, struct nfsd4_session, se_ref);
623 spin_lock(&nfsd_drc_lock);
624 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
625 nfsd_drc_mem_used -= mem;
626 spin_unlock(&nfsd_drc_lock);
627 free_session_slots(ses);
631 /* must be called under the client_lock */
633 renew_client_locked(struct nfs4_client *clp)
636 * Move client to the end to the LRU list.
638 dprintk("renewing client (clientid %08x/%08x)\n",
639 clp->cl_clientid.cl_boot,
640 clp->cl_clientid.cl_id);
641 list_move_tail(&clp->cl_lru, &client_lru);
642 clp->cl_time = get_seconds();
646 renew_client(struct nfs4_client *clp)
648 spin_lock(&client_lock);
649 renew_client_locked(clp);
650 spin_unlock(&client_lock);
653 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
655 STALE_CLIENTID(clientid_t *clid)
657 if (clid->cl_boot == boot_time)
659 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
660 clid->cl_boot, clid->cl_id, boot_time);
665 * XXX Should we use a slab cache ?
666 * This type of memory management is somewhat inefficient, but we use it
667 * anyway since SETCLIENTID is not a common operation.
669 static struct nfs4_client *alloc_client(struct xdr_netobj name)
671 struct nfs4_client *clp;
673 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
676 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
677 if (clp->cl_name.data == NULL) {
681 memcpy(clp->cl_name.data, name.data, name.len);
682 clp->cl_name.len = name.len;
687 free_client(struct nfs4_client *clp)
689 if (clp->cl_cred.cr_group_info)
690 put_group_info(clp->cl_cred.cr_group_info);
691 kfree(clp->cl_principal);
692 kfree(clp->cl_name.data);
697 expire_client(struct nfs4_client *clp)
699 struct nfs4_stateowner *sop;
700 struct nfs4_delegation *dp;
701 struct list_head reaplist;
703 INIT_LIST_HEAD(&reaplist);
704 spin_lock(&recall_lock);
705 while (!list_empty(&clp->cl_delegations)) {
706 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
707 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
709 list_del_init(&dp->dl_perclnt);
710 list_move(&dp->dl_recall_lru, &reaplist);
712 spin_unlock(&recall_lock);
713 while (!list_empty(&reaplist)) {
714 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
715 list_del_init(&dp->dl_recall_lru);
716 unhash_delegation(dp);
718 while (!list_empty(&clp->cl_openowners)) {
719 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
720 release_openowner(sop);
722 list_del(&clp->cl_idhash);
723 list_del(&clp->cl_strhash);
724 spin_lock(&client_lock);
725 list_del(&clp->cl_lru);
726 while (!list_empty(&clp->cl_sessions)) {
727 struct nfsd4_session *ses;
728 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
731 nfsd4_put_session(ses);
733 spin_unlock(&client_lock);
734 nfsd4_set_callback_client(clp, NULL);
735 if (clp->cl_cb_conn.cb_xprt)
736 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
740 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
742 memcpy(target->cl_verifier.data, source->data,
743 sizeof(target->cl_verifier.data));
746 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
748 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
749 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
752 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
754 target->cr_uid = source->cr_uid;
755 target->cr_gid = source->cr_gid;
756 target->cr_group_info = source->cr_group_info;
757 get_group_info(target->cr_group_info);
760 static int same_name(const char *n1, const char *n2)
762 return 0 == memcmp(n1, n2, HEXDIR_LEN);
766 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
768 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
772 same_clid(clientid_t *cl1, clientid_t *cl2)
774 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
777 /* XXX what about NGROUP */
779 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
781 return cr1->cr_uid == cr2->cr_uid;
784 static void gen_clid(struct nfs4_client *clp)
786 static u32 current_clientid = 1;
788 clp->cl_clientid.cl_boot = boot_time;
789 clp->cl_clientid.cl_id = current_clientid++;
792 static void gen_confirm(struct nfs4_client *clp)
797 p = (u32 *)clp->cl_confirm.data;
798 *p++ = get_seconds();
802 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
803 struct svc_rqst *rqstp, nfs4_verifier *verf)
805 struct nfs4_client *clp;
806 struct sockaddr *sa = svc_addr(rqstp);
809 clp = alloc_client(name);
813 princ = svc_gss_principal(rqstp);
815 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
816 if (clp->cl_principal == NULL) {
822 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
823 atomic_set(&clp->cl_cb_set, 0);
824 INIT_LIST_HEAD(&clp->cl_idhash);
825 INIT_LIST_HEAD(&clp->cl_strhash);
826 INIT_LIST_HEAD(&clp->cl_openowners);
827 INIT_LIST_HEAD(&clp->cl_delegations);
828 INIT_LIST_HEAD(&clp->cl_sessions);
829 INIT_LIST_HEAD(&clp->cl_lru);
830 clear_bit(0, &clp->cl_cb_slot_busy);
831 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
832 copy_verf(clp, verf);
833 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
834 clp->cl_flavor = rqstp->rq_flavor;
835 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
841 static int check_name(struct xdr_netobj name)
845 if (name.len > NFS4_OPAQUE_LIMIT) {
846 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
853 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
855 unsigned int idhashval;
857 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
858 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
859 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
864 move_to_confirmed(struct nfs4_client *clp)
866 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
867 unsigned int strhashval;
869 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
870 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
871 strhashval = clientstr_hashval(clp->cl_recdir);
872 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
876 static struct nfs4_client *
877 find_confirmed_client(clientid_t *clid)
879 struct nfs4_client *clp;
880 unsigned int idhashval = clientid_hashval(clid->cl_id);
882 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
883 if (same_clid(&clp->cl_clientid, clid))
889 static struct nfs4_client *
890 find_unconfirmed_client(clientid_t *clid)
892 struct nfs4_client *clp;
893 unsigned int idhashval = clientid_hashval(clid->cl_id);
895 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
896 if (same_clid(&clp->cl_clientid, clid))
903 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
904 * parameter. Matching is based on the fact the at least one of the
905 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
907 * FIXME: we need to unify the clientid namespaces for nfsv4.x
908 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
909 * and SET_CLIENTID{,_CONFIRM}
912 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
914 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
915 return use_exchange_id == has_exchange_flags;
918 static struct nfs4_client *
919 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
920 bool use_exchange_id)
922 struct nfs4_client *clp;
924 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
925 if (same_name(clp->cl_recdir, dname) &&
926 match_clientid_establishment(clp, use_exchange_id))
932 static struct nfs4_client *
933 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
934 bool use_exchange_id)
936 struct nfs4_client *clp;
938 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
939 if (same_name(clp->cl_recdir, dname) &&
940 match_clientid_establishment(clp, use_exchange_id))
947 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
949 struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
950 unsigned short expected_family;
952 /* Currently, we only support tcp and tcp6 for the callback channel */
953 if (se->se_callback_netid_len == 3 &&
954 !memcmp(se->se_callback_netid_val, "tcp", 3))
955 expected_family = AF_INET;
956 else if (se->se_callback_netid_len == 4 &&
957 !memcmp(se->se_callback_netid_val, "tcp6", 4))
958 expected_family = AF_INET6;
962 cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
963 se->se_callback_addr_len,
964 (struct sockaddr *) &cb->cb_addr,
965 sizeof(cb->cb_addr));
967 if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
970 if (cb->cb_addr.ss_family == AF_INET6)
971 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
973 cb->cb_minorversion = 0;
974 cb->cb_prog = se->se_callback_prog;
975 cb->cb_ident = se->se_callback_ident;
978 cb->cb_addr.ss_family = AF_UNSPEC;
980 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
981 "will not receive delegations\n",
982 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
988 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
991 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
993 struct nfsd4_slot *slot = resp->cstate.slot;
996 dprintk("--> %s slot %p\n", __func__, slot);
998 slot->sl_opcnt = resp->opcnt;
999 slot->sl_status = resp->cstate.status;
1001 if (nfsd4_not_cached(resp)) {
1002 slot->sl_datalen = 0;
1005 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1006 base = (char *)resp->cstate.datap -
1007 (char *)resp->xbuf->head[0].iov_base;
1008 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1010 WARN("%s: sessions DRC could not cache compound\n", __func__);
1015 * Encode the replay sequence operation from the slot values.
1016 * If cachethis is FALSE encode the uncached rep error on the next
1017 * operation which sets resp->p and increments resp->opcnt for
1018 * nfs4svc_encode_compoundres.
1022 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1023 struct nfsd4_compoundres *resp)
1025 struct nfsd4_op *op;
1026 struct nfsd4_slot *slot = resp->cstate.slot;
1028 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1029 resp->opcnt, resp->cstate.slot->sl_cachethis);
1031 /* Encode the replayed sequence operation */
1032 op = &args->ops[resp->opcnt - 1];
1033 nfsd4_encode_operation(resp, op);
1035 /* Return nfserr_retry_uncached_rep in next operation. */
1036 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1037 op = &args->ops[resp->opcnt++];
1038 op->status = nfserr_retry_uncached_rep;
1039 nfsd4_encode_operation(resp, op);
1045 * The sequence operation is not cached because we can use the slot and
1049 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1050 struct nfsd4_sequence *seq)
1052 struct nfsd4_slot *slot = resp->cstate.slot;
1055 dprintk("--> %s slot %p\n", __func__, slot);
1057 /* Either returns 0 or nfserr_retry_uncached */
1058 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1059 if (status == nfserr_retry_uncached_rep)
1062 /* The sequence operation has been encoded, cstate->datap set. */
1063 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1065 resp->opcnt = slot->sl_opcnt;
1066 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1067 status = slot->sl_status;
1073 * Set the exchange_id flags returned by the server.
1076 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1078 /* pNFS is not supported */
1079 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1081 /* Referrals are supported, Migration is not. */
1082 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1084 /* set the wire flags to return to client. */
1085 clid->flags = new->cl_exchange_flags;
1089 nfsd4_exchange_id(struct svc_rqst *rqstp,
1090 struct nfsd4_compound_state *cstate,
1091 struct nfsd4_exchange_id *exid)
1093 struct nfs4_client *unconf, *conf, *new;
1095 unsigned int strhashval;
1096 char dname[HEXDIR_LEN];
1097 char addr_str[INET6_ADDRSTRLEN];
1098 nfs4_verifier verf = exid->verifier;
1099 struct sockaddr *sa = svc_addr(rqstp);
1101 rpc_ntop(sa, addr_str, sizeof(addr_str));
1102 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1103 "ip_addr=%s flags %x, spa_how %d\n",
1104 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1105 addr_str, exid->flags, exid->spa_how);
1107 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1108 return nfserr_inval;
1110 /* Currently only support SP4_NONE */
1111 switch (exid->spa_how) {
1115 return nfserr_encr_alg_unsupp;
1117 BUG(); /* checked by xdr code */
1119 return nfserr_serverfault; /* no excuse :-/ */
1122 status = nfs4_make_rec_clidname(dname, &exid->clname);
1127 strhashval = clientstr_hashval(dname);
1132 conf = find_confirmed_client_by_str(dname, strhashval, true);
1134 if (!same_verf(&verf, &conf->cl_verifier)) {
1135 /* 18.35.4 case 8 */
1136 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1137 status = nfserr_not_same;
1140 /* Client reboot: destroy old state */
1141 expire_client(conf);
1144 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1145 /* 18.35.4 case 9 */
1146 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1147 status = nfserr_perm;
1150 expire_client(conf);
1154 * Set bit when the owner id and verifier map to an already
1155 * confirmed client id (18.35.3).
1157 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1160 * Falling into 18.35.4 case 2, possible router replay.
1161 * Leave confirmed record intact and return same result.
1163 copy_verf(conf, &verf);
1168 /* 18.35.4 case 7 */
1169 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1170 status = nfserr_noent;
1174 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1177 * Possible retry or client restart. Per 18.35.4 case 4,
1178 * a new unconfirmed record should be generated regardless
1179 * of whether any properties have changed.
1181 expire_client(unconf);
1186 new = create_client(exid->clname, dname, rqstp, &verf);
1188 status = nfserr_serverfault;
1193 add_to_unconfirmed(new, strhashval);
1195 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1196 exid->clientid.cl_id = new->cl_clientid.cl_id;
1199 nfsd4_set_ex_flags(new, exid);
1201 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1202 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1206 nfs4_unlock_state();
1208 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1213 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1215 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1218 /* The slot is in use, and no response has been sent. */
1220 if (seqid == slot_seqid)
1221 return nfserr_jukebox;
1223 return nfserr_seq_misordered;
1226 if (likely(seqid == slot_seqid + 1))
1229 if (seqid == slot_seqid)
1230 return nfserr_replay_cache;
1232 if (seqid == 1 && (slot_seqid + 1) == 0)
1234 /* Misordered replay or misordered new request */
1235 return nfserr_seq_misordered;
1239 * Cache the create session result into the create session single DRC
1240 * slot cache by saving the xdr structure. sl_seqid has been set.
1241 * Do this for solo or embedded create session operations.
1244 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1245 struct nfsd4_clid_slot *slot, int nfserr)
1247 slot->sl_status = nfserr;
1248 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1252 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1253 struct nfsd4_clid_slot *slot)
1255 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1256 return slot->sl_status;
1260 nfsd4_create_session(struct svc_rqst *rqstp,
1261 struct nfsd4_compound_state *cstate,
1262 struct nfsd4_create_session *cr_ses)
1264 struct sockaddr *sa = svc_addr(rqstp);
1265 struct nfs4_client *conf, *unconf;
1266 struct nfsd4_clid_slot *cs_slot = NULL;
1270 unconf = find_unconfirmed_client(&cr_ses->clientid);
1271 conf = find_confirmed_client(&cr_ses->clientid);
1274 cs_slot = &conf->cl_cs_slot;
1275 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1276 if (status == nfserr_replay_cache) {
1277 dprintk("Got a create_session replay! seqid= %d\n",
1279 /* Return the cached reply status */
1280 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1282 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1283 status = nfserr_seq_misordered;
1284 dprintk("Sequence misordered!\n");
1285 dprintk("Expected seqid= %d but got seqid= %d\n",
1286 cs_slot->sl_seqid, cr_ses->seqid);
1289 cs_slot->sl_seqid++;
1290 } else if (unconf) {
1291 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1292 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1293 status = nfserr_clid_inuse;
1297 cs_slot = &unconf->cl_cs_slot;
1298 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1300 /* an unconfirmed replay returns misordered */
1301 status = nfserr_seq_misordered;
1305 cs_slot->sl_seqid++; /* from 0 to 1 */
1306 move_to_confirmed(unconf);
1308 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1309 unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1310 svc_xprt_get(rqstp->rq_xprt);
1312 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1314 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1315 unconf->cl_cb_conn.cb_minorversion =
1316 cstate->minorversion;
1317 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1318 unconf->cl_cb_seq_nr = 1;
1319 nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1323 status = nfserr_stale_clientid;
1328 * We do not support RDMA or persistent sessions
1330 cr_ses->flags &= ~SESSION4_PERSIST;
1331 cr_ses->flags &= ~SESSION4_RDMA;
1333 status = alloc_init_session(rqstp, conf, cr_ses);
1337 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1338 NFS4_MAX_SESSIONID_LEN);
1339 cr_ses->seqid = cs_slot->sl_seqid;
1342 /* cache solo and embedded create sessions under the state lock */
1343 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1345 nfs4_unlock_state();
1346 dprintk("%s returns %d\n", __func__, ntohl(status));
1350 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1352 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1353 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1355 return argp->opcnt == resp->opcnt;
1358 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1362 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1366 nfsd4_destroy_session(struct svc_rqst *r,
1367 struct nfsd4_compound_state *cstate,
1368 struct nfsd4_destroy_session *sessionid)
1370 struct nfsd4_session *ses;
1371 u32 status = nfserr_badsession;
1374 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1375 * - Should we return nfserr_back_chan_busy if waiting for
1376 * callbacks on to-be-destroyed session?
1377 * - Do we need to clear any callback info from previous session?
1380 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1381 if (!nfsd4_last_compound_op(r))
1382 return nfserr_not_only_op;
1384 dump_sessionid(__func__, &sessionid->sessionid);
1385 spin_lock(&client_lock);
1386 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1388 spin_unlock(&client_lock);
1392 unhash_session(ses);
1393 spin_unlock(&client_lock);
1395 /* wait for callbacks */
1396 nfsd4_set_callback_client(ses->se_client, NULL);
1397 nfsd4_put_session(ses);
1400 dprintk("%s returns %d\n", __func__, ntohl(status));
1405 nfsd4_sequence(struct svc_rqst *rqstp,
1406 struct nfsd4_compound_state *cstate,
1407 struct nfsd4_sequence *seq)
1409 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1410 struct nfsd4_session *session;
1411 struct nfsd4_slot *slot;
1414 if (resp->opcnt != 1)
1415 return nfserr_sequence_pos;
1417 spin_lock(&client_lock);
1418 status = nfserr_badsession;
1419 session = find_in_sessionid_hashtbl(&seq->sessionid);
1423 status = nfserr_badslot;
1424 if (seq->slotid >= session->se_fchannel.maxreqs)
1427 slot = session->se_slots[seq->slotid];
1428 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1430 /* We do not negotiate the number of slots yet, so set the
1431 * maxslots to the session maxreqs which is used to encode
1432 * sr_highest_slotid and the sr_target_slot id to maxslots */
1433 seq->maxslots = session->se_fchannel.maxreqs;
1435 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1436 if (status == nfserr_replay_cache) {
1437 cstate->slot = slot;
1438 cstate->session = session;
1439 /* Return the cached reply status and set cstate->status
1440 * for nfsd4_proc_compound processing */
1441 status = nfsd4_replay_cache_entry(resp, seq);
1442 cstate->status = nfserr_replay_cache;
1448 /* Success! bump slot seqid */
1449 slot->sl_inuse = true;
1450 slot->sl_seqid = seq->seqid;
1451 slot->sl_cachethis = seq->cachethis;
1453 cstate->slot = slot;
1454 cstate->session = session;
1457 /* Hold a session reference until done processing the compound. */
1458 if (cstate->session) {
1459 nfsd4_get_session(cstate->session);
1460 /* Renew the clientid on success and on replay */
1461 renew_client_locked(session->se_client);
1463 spin_unlock(&client_lock);
1464 dprintk("%s: return %d\n", __func__, ntohl(status));
1469 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1470 struct nfsd4_setclientid *setclid)
1472 struct sockaddr *sa = svc_addr(rqstp);
1473 struct xdr_netobj clname = {
1474 .len = setclid->se_namelen,
1475 .data = setclid->se_name,
1477 nfs4_verifier clverifier = setclid->se_verf;
1478 unsigned int strhashval;
1479 struct nfs4_client *conf, *unconf, *new;
1481 char dname[HEXDIR_LEN];
1483 if (!check_name(clname))
1484 return nfserr_inval;
1486 status = nfs4_make_rec_clidname(dname, &clname);
1491 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1492 * We get here on a DRC miss.
1495 strhashval = clientstr_hashval(dname);
1498 conf = find_confirmed_client_by_str(dname, strhashval, false);
1500 /* RFC 3530 14.2.33 CASE 0: */
1501 status = nfserr_clid_inuse;
1502 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1503 char addr_str[INET6_ADDRSTRLEN];
1504 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1506 dprintk("NFSD: setclientid: string in use by client "
1507 "at %s\n", addr_str);
1512 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1513 * has a description of SETCLIENTID request processing consisting
1514 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1516 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1517 status = nfserr_resource;
1520 * RFC 3530 14.2.33 CASE 4:
1521 * placed first, because it is the normal case
1524 expire_client(unconf);
1525 new = create_client(clname, dname, rqstp, &clverifier);
1529 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1531 * RFC 3530 14.2.33 CASE 1:
1532 * probable callback update
1535 /* Note this is removing unconfirmed {*x***},
1536 * which is stronger than RFC recommended {vxc**}.
1537 * This has the advantage that there is at most
1538 * one {*x***} in either list at any time.
1540 expire_client(unconf);
1542 new = create_client(clname, dname, rqstp, &clverifier);
1545 copy_clid(new, conf);
1546 } else if (!unconf) {
1548 * RFC 3530 14.2.33 CASE 2:
1549 * probable client reboot; state will be removed if
1552 new = create_client(clname, dname, rqstp, &clverifier);
1558 * RFC 3530 14.2.33 CASE 3:
1559 * probable client reboot; state will be removed if
1562 expire_client(unconf);
1563 new = create_client(clname, dname, rqstp, &clverifier);
1568 gen_callback(new, setclid, rpc_get_scope_id(sa));
1569 add_to_unconfirmed(new, strhashval);
1570 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1571 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1572 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1575 nfs4_unlock_state();
1581 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1582 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1583 * bullets, labeled as CASE1 - CASE4 below.
1586 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1587 struct nfsd4_compound_state *cstate,
1588 struct nfsd4_setclientid_confirm *setclientid_confirm)
1590 struct sockaddr *sa = svc_addr(rqstp);
1591 struct nfs4_client *conf, *unconf;
1592 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1593 clientid_t * clid = &setclientid_confirm->sc_clientid;
1596 if (STALE_CLIENTID(clid))
1597 return nfserr_stale_clientid;
1599 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1600 * We get here on a DRC miss.
1605 conf = find_confirmed_client(clid);
1606 unconf = find_unconfirmed_client(clid);
1608 status = nfserr_clid_inuse;
1609 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1611 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1615 * section 14.2.34 of RFC 3530 has a description of
1616 * SETCLIENTID_CONFIRM request processing consisting
1617 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1619 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1621 * RFC 3530 14.2.34 CASE 1:
1624 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1625 status = nfserr_clid_inuse;
1627 atomic_set(&conf->cl_cb_set, 0);
1628 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1629 expire_client(unconf);
1633 } else if (conf && !unconf) {
1635 * RFC 3530 14.2.34 CASE 2:
1636 * probable retransmitted request; play it safe and
1639 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1640 status = nfserr_clid_inuse;
1643 } else if (!conf && unconf
1644 && same_verf(&unconf->cl_confirm, &confirm)) {
1646 * RFC 3530 14.2.34 CASE 3:
1647 * Normal case; new or rebooted client:
1649 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1650 status = nfserr_clid_inuse;
1653 clientstr_hashval(unconf->cl_recdir);
1654 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1657 nfsd4_remove_clid_dir(conf);
1658 expire_client(conf);
1660 move_to_confirmed(unconf);
1662 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1665 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1666 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1669 * RFC 3530 14.2.34 CASE 4:
1670 * Client probably hasn't noticed that we rebooted yet.
1672 status = nfserr_stale_clientid;
1674 /* check that we have hit one of the cases...*/
1675 status = nfserr_clid_inuse;
1678 nfs4_unlock_state();
1682 /* OPEN Share state helper functions */
1683 static inline struct nfs4_file *
1684 alloc_init_file(struct inode *ino)
1686 struct nfs4_file *fp;
1687 unsigned int hashval = file_hashval(ino);
1689 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1691 atomic_set(&fp->fi_ref, 1);
1692 INIT_LIST_HEAD(&fp->fi_hash);
1693 INIT_LIST_HEAD(&fp->fi_stateids);
1694 INIT_LIST_HEAD(&fp->fi_delegations);
1695 spin_lock(&recall_lock);
1696 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1697 spin_unlock(&recall_lock);
1698 fp->fi_inode = igrab(ino);
1699 fp->fi_id = current_fileid++;
1700 fp->fi_had_conflict = false;
1707 nfsd4_free_slab(struct kmem_cache **slab)
1711 kmem_cache_destroy(*slab);
1716 nfsd4_free_slabs(void)
1718 nfsd4_free_slab(&stateowner_slab);
1719 nfsd4_free_slab(&file_slab);
1720 nfsd4_free_slab(&stateid_slab);
1721 nfsd4_free_slab(&deleg_slab);
1725 nfsd4_init_slabs(void)
1727 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1728 sizeof(struct nfs4_stateowner), 0, 0, NULL);
1729 if (stateowner_slab == NULL)
1731 file_slab = kmem_cache_create("nfsd4_files",
1732 sizeof(struct nfs4_file), 0, 0, NULL);
1733 if (file_slab == NULL)
1735 stateid_slab = kmem_cache_create("nfsd4_stateids",
1736 sizeof(struct nfs4_stateid), 0, 0, NULL);
1737 if (stateid_slab == NULL)
1739 deleg_slab = kmem_cache_create("nfsd4_delegations",
1740 sizeof(struct nfs4_delegation), 0, 0, NULL);
1741 if (deleg_slab == NULL)
1746 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1751 nfs4_free_stateowner(struct kref *kref)
1753 struct nfs4_stateowner *sop =
1754 container_of(kref, struct nfs4_stateowner, so_ref);
1755 kfree(sop->so_owner.data);
1756 kmem_cache_free(stateowner_slab, sop);
1759 static inline struct nfs4_stateowner *
1760 alloc_stateowner(struct xdr_netobj *owner)
1762 struct nfs4_stateowner *sop;
1764 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1765 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1766 memcpy(sop->so_owner.data, owner->data, owner->len);
1767 sop->so_owner.len = owner->len;
1768 kref_init(&sop->so_ref);
1771 kmem_cache_free(stateowner_slab, sop);
1776 static struct nfs4_stateowner *
1777 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1778 struct nfs4_stateowner *sop;
1779 struct nfs4_replay *rp;
1780 unsigned int idhashval;
1782 if (!(sop = alloc_stateowner(&open->op_owner)))
1784 idhashval = ownerid_hashval(current_ownerid);
1785 INIT_LIST_HEAD(&sop->so_idhash);
1786 INIT_LIST_HEAD(&sop->so_strhash);
1787 INIT_LIST_HEAD(&sop->so_perclient);
1788 INIT_LIST_HEAD(&sop->so_stateids);
1789 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
1790 INIT_LIST_HEAD(&sop->so_close_lru);
1792 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1793 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1794 list_add(&sop->so_perclient, &clp->cl_openowners);
1795 sop->so_is_open_owner = 1;
1796 sop->so_id = current_ownerid++;
1797 sop->so_client = clp;
1798 sop->so_seqid = open->op_seqid;
1799 sop->so_confirmed = 0;
1800 rp = &sop->so_replay;
1801 rp->rp_status = nfserr_serverfault;
1803 rp->rp_buf = rp->rp_ibuf;
1808 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1809 struct nfs4_stateowner *sop = open->op_stateowner;
1810 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1812 INIT_LIST_HEAD(&stp->st_hash);
1813 INIT_LIST_HEAD(&stp->st_perstateowner);
1814 INIT_LIST_HEAD(&stp->st_lockowners);
1815 INIT_LIST_HEAD(&stp->st_perfile);
1816 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1817 list_add(&stp->st_perstateowner, &sop->so_stateids);
1818 list_add(&stp->st_perfile, &fp->fi_stateids);
1819 stp->st_stateowner = sop;
1822 stp->st_stateid.si_boot = get_seconds();
1823 stp->st_stateid.si_stateownerid = sop->so_id;
1824 stp->st_stateid.si_fileid = fp->fi_id;
1825 stp->st_stateid.si_generation = 0;
1826 stp->st_access_bmap = 0;
1827 stp->st_deny_bmap = 0;
1828 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1829 &stp->st_access_bmap);
1830 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1831 stp->st_openstp = NULL;
1835 move_to_close_lru(struct nfs4_stateowner *sop)
1837 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1839 list_move_tail(&sop->so_close_lru, &close_lru);
1840 sop->so_time = get_seconds();
1844 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1847 return (sop->so_owner.len == owner->len) &&
1848 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1849 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1852 static struct nfs4_stateowner *
1853 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1855 struct nfs4_stateowner *so = NULL;
1857 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1858 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1864 /* search file_hashtbl[] for file */
1865 static struct nfs4_file *
1866 find_file(struct inode *ino)
1868 unsigned int hashval = file_hashval(ino);
1869 struct nfs4_file *fp;
1871 spin_lock(&recall_lock);
1872 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1873 if (fp->fi_inode == ino) {
1875 spin_unlock(&recall_lock);
1879 spin_unlock(&recall_lock);
1883 static inline int access_valid(u32 x, u32 minorversion)
1885 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1887 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1889 x &= ~NFS4_SHARE_ACCESS_MASK;
1890 if (minorversion && x) {
1891 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1893 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1895 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1902 static inline int deny_valid(u32 x)
1904 /* Note: unlike access bits, deny bits may be zero. */
1905 return x <= NFS4_SHARE_DENY_BOTH;
1909 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1910 * st_{access,deny}_bmap field of the stateid, in order to track not
1911 * only what share bits are currently in force, but also what
1912 * combinations of share bits previous opens have used. This allows us
1913 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1914 * return an error if the client attempt to downgrade to a combination
1915 * of share bits not explicable by closing some of its previous opens.
1917 * XXX: This enforcement is actually incomplete, since we don't keep
1918 * track of access/deny bit combinations; so, e.g., we allow:
1920 * OPEN allow read, deny write
1921 * OPEN allow both, deny none
1922 * DOWNGRADE allow read, deny none
1924 * which we should reject.
1927 set_access(unsigned int *access, unsigned long bmap) {
1931 for (i = 1; i < 4; i++) {
1932 if (test_bit(i, &bmap))
1938 set_deny(unsigned int *deny, unsigned long bmap) {
1942 for (i = 0; i < 4; i++) {
1943 if (test_bit(i, &bmap))
1949 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1950 unsigned int access, deny;
1952 set_access(&access, stp->st_access_bmap);
1953 set_deny(&deny, stp->st_deny_bmap);
1954 if ((access & open->op_share_deny) || (deny & open->op_share_access))
1960 * Called to check deny when READ with all zero stateid or
1961 * WRITE with all zero or all one stateid
1964 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1966 struct inode *ino = current_fh->fh_dentry->d_inode;
1967 struct nfs4_file *fp;
1968 struct nfs4_stateid *stp;
1971 dprintk("NFSD: nfs4_share_conflict\n");
1973 fp = find_file(ino);
1976 ret = nfserr_locked;
1977 /* Search for conflicting share reservations */
1978 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1979 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1980 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1990 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1992 if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1993 drop_file_write_access(filp);
1994 spin_lock(&filp->f_lock);
1995 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1996 spin_unlock(&filp->f_lock);
2001 * Spawn a thread to perform a recall on the delegation represented
2002 * by the lease (file_lock)
2004 * Called from break_lease() with lock_kernel() held.
2005 * Note: we assume break_lease will only call this *once* for any given
2009 void nfsd_break_deleg_cb(struct file_lock *fl)
2011 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2013 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2017 /* We're assuming the state code never drops its reference
2018 * without first removing the lease. Since we're in this lease
2019 * callback (and since the lease code is serialized by the kernel
2020 * lock) we know the server hasn't removed the lease yet, we know
2021 * it's safe to take a reference: */
2022 atomic_inc(&dp->dl_count);
2024 spin_lock(&recall_lock);
2025 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2026 spin_unlock(&recall_lock);
2028 /* only place dl_time is set. protected by lock_kernel*/
2029 dp->dl_time = get_seconds();
2032 * We don't want the locks code to timeout the lease for us;
2033 * we'll remove it ourself if the delegation isn't returned
2036 fl->fl_break_time = 0;
2038 dp->dl_file->fi_had_conflict = true;
2039 nfsd4_cb_recall(dp);
2043 * The file_lock is being reapd.
2045 * Called by locks_free_lock() with lock_kernel() held.
2048 void nfsd_release_deleg_cb(struct file_lock *fl)
2050 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2052 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2054 if (!(fl->fl_flags & FL_LEASE) || !dp)
2056 dp->dl_flock = NULL;
2060 * Set the delegation file_lock back pointer.
2062 * Called from setlease() with lock_kernel() held.
2065 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2067 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2069 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2076 * Called from setlease() with lock_kernel() held
2079 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2081 struct nfs4_delegation *onlistd =
2082 (struct nfs4_delegation *)onlist->fl_owner;
2083 struct nfs4_delegation *tryd =
2084 (struct nfs4_delegation *)try->fl_owner;
2086 if (onlist->fl_lmops != try->fl_lmops)
2089 return onlistd->dl_client == tryd->dl_client;
2094 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2097 return lease_modify(onlist, arg);
2102 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2103 .fl_break = nfsd_break_deleg_cb,
2104 .fl_release_private = nfsd_release_deleg_cb,
2105 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2106 .fl_mylease = nfsd_same_client_deleg_cb,
2107 .fl_change = nfsd_change_deleg_cb,
2112 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2113 struct nfsd4_open *open)
2115 clientid_t *clientid = &open->op_clientid;
2116 struct nfs4_client *clp = NULL;
2117 unsigned int strhashval;
2118 struct nfs4_stateowner *sop = NULL;
2120 if (!check_name(open->op_owner))
2121 return nfserr_inval;
2123 if (STALE_CLIENTID(&open->op_clientid))
2124 return nfserr_stale_clientid;
2126 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2127 sop = find_openstateowner_str(strhashval, open);
2128 open->op_stateowner = sop;
2130 /* Make sure the client's lease hasn't expired. */
2131 clp = find_confirmed_client(clientid);
2133 return nfserr_expired;
2136 /* When sessions are used, skip open sequenceid processing */
2137 if (nfsd4_has_session(cstate))
2139 if (!sop->so_confirmed) {
2140 /* Replace unconfirmed owners without checking for replay. */
2141 clp = sop->so_client;
2142 release_openowner(sop);
2143 open->op_stateowner = NULL;
2146 if (open->op_seqid == sop->so_seqid - 1) {
2147 if (sop->so_replay.rp_buflen)
2148 return nfserr_replay_me;
2149 /* The original OPEN failed so spectacularly
2150 * that we don't even have replay data saved!
2151 * Therefore, we have no choice but to continue
2152 * processing this OPEN; presumably, we'll
2153 * fail again for the same reason.
2155 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2158 if (open->op_seqid != sop->so_seqid)
2159 return nfserr_bad_seqid;
2161 if (open->op_stateowner == NULL) {
2162 sop = alloc_init_open_stateowner(strhashval, clp, open);
2164 return nfserr_resource;
2165 open->op_stateowner = sop;
2167 list_del_init(&sop->so_close_lru);
2168 renew_client(sop->so_client);
2172 static inline __be32
2173 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2175 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2176 return nfserr_openmode;
2181 static struct nfs4_delegation *
2182 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2184 struct nfs4_delegation *dp;
2186 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2187 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2194 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2195 struct nfs4_delegation **dp)
2198 __be32 status = nfserr_bad_stateid;
2200 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2203 flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2204 RD_STATE : WR_STATE;
2205 status = nfs4_check_delegmode(*dp, flags);
2209 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2213 open->op_stateowner->so_confirmed = 1;
2218 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2220 struct nfs4_stateid *local;
2221 __be32 status = nfserr_share_denied;
2222 struct nfs4_stateowner *sop = open->op_stateowner;
2224 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2225 /* ignore lock owners */
2226 if (local->st_stateowner->so_is_open_owner == 0)
2228 /* remember if we have seen this open owner */
2229 if (local->st_stateowner == sop)
2231 /* check for conflicting share reservations */
2232 if (!test_share(local, open))
2240 static inline struct nfs4_stateid *
2241 nfs4_alloc_stateid(void)
2243 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2247 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2248 struct nfs4_delegation *dp,
2249 struct svc_fh *cur_fh, int flags)
2251 struct nfs4_stateid *stp;
2253 stp = nfs4_alloc_stateid();
2255 return nfserr_resource;
2258 get_file(dp->dl_vfs_file);
2259 stp->st_vfs_file = dp->dl_vfs_file;
2262 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2265 if (status == nfserr_dropit)
2266 status = nfserr_jukebox;
2267 kmem_cache_free(stateid_slab, stp);
2275 static inline __be32
2276 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2277 struct nfsd4_open *open)
2279 struct iattr iattr = {
2280 .ia_valid = ATTR_SIZE,
2283 if (!open->op_truncate)
2285 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2286 return nfserr_inval;
2287 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2291 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2293 struct file *filp = stp->st_vfs_file;
2294 struct inode *inode = filp->f_path.dentry->d_inode;
2295 unsigned int share_access, new_writer;
2298 set_access(&share_access, stp->st_access_bmap);
2299 new_writer = (~share_access) & open->op_share_access
2300 & NFS4_SHARE_ACCESS_WRITE;
2303 int err = get_write_access(inode);
2305 return nfserrno(err);
2306 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2308 return nfserrno(err);
2309 file_take_write(filp);
2311 status = nfsd4_truncate(rqstp, cur_fh, open);
2314 put_write_access(inode);
2317 /* remember the open */
2318 filp->f_mode |= open->op_share_access;
2319 __set_bit(open->op_share_access, &stp->st_access_bmap);
2320 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2327 nfs4_set_claim_prev(struct nfsd4_open *open)
2329 open->op_stateowner->so_confirmed = 1;
2330 open->op_stateowner->so_client->cl_firststate = 1;
2334 * Attempt to hand out a delegation.
2337 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2339 struct nfs4_delegation *dp;
2340 struct nfs4_stateowner *sop = stp->st_stateowner;
2341 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2342 struct file_lock fl, *flp = &fl;
2343 int status, flag = 0;
2345 flag = NFS4_OPEN_DELEGATE_NONE;
2346 open->op_recall = 0;
2347 switch (open->op_claim_type) {
2348 case NFS4_OPEN_CLAIM_PREVIOUS:
2350 open->op_recall = 1;
2351 flag = open->op_delegate_type;
2352 if (flag == NFS4_OPEN_DELEGATE_NONE)
2355 case NFS4_OPEN_CLAIM_NULL:
2356 /* Let's not give out any delegations till everyone's
2357 * had the chance to reclaim theirs.... */
2358 if (locks_in_grace())
2360 if (!cb_up || !sop->so_confirmed)
2362 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2363 flag = NFS4_OPEN_DELEGATE_WRITE;
2365 flag = NFS4_OPEN_DELEGATE_READ;
2371 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2373 flag = NFS4_OPEN_DELEGATE_NONE;
2376 locks_init_lock(&fl);
2377 fl.fl_lmops = &nfsd_lease_mng_ops;
2378 fl.fl_flags = FL_LEASE;
2379 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2380 fl.fl_end = OFFSET_MAX;
2381 fl.fl_owner = (fl_owner_t)dp;
2382 fl.fl_file = stp->st_vfs_file;
2383 fl.fl_pid = current->tgid;
2385 /* vfs_setlease checks to see if delegation should be handed out.
2386 * the lock_manager callbacks fl_mylease and fl_change are used
2388 if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2389 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2390 unhash_delegation(dp);
2391 flag = NFS4_OPEN_DELEGATE_NONE;
2395 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2397 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2398 STATEID_VAL(&dp->dl_stateid));
2400 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2401 && flag == NFS4_OPEN_DELEGATE_NONE
2402 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2403 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2404 open->op_delegate_type = flag;
2408 * called with nfs4_lock_state() held.
2411 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2413 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2414 struct nfs4_file *fp = NULL;
2415 struct inode *ino = current_fh->fh_dentry->d_inode;
2416 struct nfs4_stateid *stp = NULL;
2417 struct nfs4_delegation *dp = NULL;
2420 status = nfserr_inval;
2421 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2422 || !deny_valid(open->op_share_deny))
2425 * Lookup file; if found, lookup stateid and check open request,
2426 * and check for delegations in the process of being recalled.
2427 * If not found, create the nfs4_file struct
2429 fp = find_file(ino);
2431 if ((status = nfs4_check_open(fp, open, &stp)))
2433 status = nfs4_check_deleg(fp, open, &dp);
2437 status = nfserr_bad_stateid;
2438 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2440 status = nfserr_resource;
2441 fp = alloc_init_file(ino);
2447 * OPEN the file, or upgrade an existing OPEN.
2448 * If truncate fails, the OPEN fails.
2451 /* Stateid was found, this is an OPEN upgrade */
2452 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2455 update_stateid(&stp->st_stateid);
2457 /* Stateid was not found, this is a new OPEN */
2459 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2460 flags |= NFSD_MAY_READ;
2461 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2462 flags |= NFSD_MAY_WRITE;
2463 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2466 init_stateid(stp, fp, open);
2467 status = nfsd4_truncate(rqstp, current_fh, open);
2469 release_open_stateid(stp);
2472 if (nfsd4_has_session(&resp->cstate))
2473 update_stateid(&stp->st_stateid);
2475 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2477 if (nfsd4_has_session(&resp->cstate)) {
2478 open->op_stateowner->so_confirmed = 1;
2479 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2483 * Attempt to hand out a delegation. No error return, because the
2484 * OPEN succeeds even if we fail.
2486 nfs4_open_delegation(current_fh, open, stp);
2490 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2491 STATEID_VAL(&stp->st_stateid));
2495 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2496 nfs4_set_claim_prev(open);
2498 * To finish the open response, we just need to set the rflags.
2500 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2501 if (!open->op_stateowner->so_confirmed &&
2502 !nfsd4_has_session(&resp->cstate))
2503 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2509 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2512 struct nfs4_client *clp;
2516 dprintk("process_renew(%08x/%08x): starting\n",
2517 clid->cl_boot, clid->cl_id);
2518 status = nfserr_stale_clientid;
2519 if (STALE_CLIENTID(clid))
2521 clp = find_confirmed_client(clid);
2522 status = nfserr_expired;
2524 /* We assume the client took too long to RENEW. */
2525 dprintk("nfsd4_renew: clientid not found!\n");
2529 status = nfserr_cb_path_down;
2530 if (!list_empty(&clp->cl_delegations)
2531 && !atomic_read(&clp->cl_cb_set))
2535 nfs4_unlock_state();
2539 struct lock_manager nfsd4_manager = {
2543 nfsd4_end_grace(void)
2545 dprintk("NFSD: end of grace period\n");
2546 nfsd4_recdir_purge_old();
2547 locks_end_grace(&nfsd4_manager);
2549 * Now that every NFSv4 client has had the chance to recover and
2550 * to see the (possibly new, possibly shorter) lease time, we
2551 * can safely set the next grace time to the current lease time:
2553 nfsd4_grace = nfsd4_lease;
2557 nfs4_laundromat(void)
2559 struct nfs4_client *clp;
2560 struct nfs4_stateowner *sop;
2561 struct nfs4_delegation *dp;
2562 struct list_head *pos, *next, reaplist;
2563 time_t cutoff = get_seconds() - nfsd4_lease;
2564 time_t t, clientid_val = nfsd4_lease;
2565 time_t u, test_val = nfsd4_lease;
2569 dprintk("NFSD: laundromat service - starting\n");
2570 if (locks_in_grace())
2572 INIT_LIST_HEAD(&reaplist);
2573 spin_lock(&client_lock);
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 list_move(&clp->cl_lru, &reaplist);
2584 spin_unlock(&client_lock);
2585 list_for_each_safe(pos, next, &reaplist) {
2586 clp = list_entry(pos, struct nfs4_client, cl_lru);
2587 dprintk("NFSD: purging unused client (clientid %08x)\n",
2588 clp->cl_clientid.cl_id);
2589 nfsd4_remove_clid_dir(clp);
2592 spin_lock(&recall_lock);
2593 list_for_each_safe(pos, next, &del_recall_lru) {
2594 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2595 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2596 u = dp->dl_time - cutoff;
2601 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2603 list_move(&dp->dl_recall_lru, &reaplist);
2605 spin_unlock(&recall_lock);
2606 list_for_each_safe(pos, next, &reaplist) {
2607 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2608 list_del_init(&dp->dl_recall_lru);
2609 unhash_delegation(dp);
2611 test_val = nfsd4_lease;
2612 list_for_each_safe(pos, next, &close_lru) {
2613 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2614 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2615 u = sop->so_time - cutoff;
2620 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2622 release_openowner(sop);
2624 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2625 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2626 nfs4_unlock_state();
2627 return clientid_val;
2630 static struct workqueue_struct *laundry_wq;
2631 static void laundromat_main(struct work_struct *);
2632 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2635 laundromat_main(struct work_struct *not_used)
2639 t = nfs4_laundromat();
2640 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2641 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2644 static struct nfs4_stateowner *
2645 search_close_lru(u32 st_id, int flags)
2647 struct nfs4_stateowner *local = NULL;
2649 if (flags & CLOSE_STATE) {
2650 list_for_each_entry(local, &close_lru, so_close_lru) {
2651 if (local->so_id == st_id)
2659 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2661 return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2665 STALE_STATEID(stateid_t *stateid)
2667 if (time_after((unsigned long)boot_time,
2668 (unsigned long)stateid->si_boot)) {
2669 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2670 STATEID_VAL(stateid));
2677 EXPIRED_STATEID(stateid_t *stateid)
2679 if (time_before((unsigned long)boot_time,
2680 ((unsigned long)stateid->si_boot)) &&
2681 time_before((unsigned long)(stateid->si_boot + nfsd4_lease), get_seconds())) {
2682 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2683 STATEID_VAL(stateid));
2690 stateid_error_map(stateid_t *stateid)
2692 if (STALE_STATEID(stateid))
2693 return nfserr_stale_stateid;
2694 if (EXPIRED_STATEID(stateid))
2695 return nfserr_expired;
2697 dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2698 STATEID_VAL(stateid));
2699 return nfserr_bad_stateid;
2703 access_permit_read(unsigned long access_bmap)
2705 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2706 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2707 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2711 access_permit_write(unsigned long access_bmap)
2713 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2714 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2718 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2720 __be32 status = nfserr_openmode;
2722 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2724 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2731 static inline __be32
2732 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2734 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2736 else if (locks_in_grace()) {
2737 /* Answer in remaining cases depends on existance of
2738 * conflicting state; so we must wait out the grace period. */
2739 return nfserr_grace;
2740 } else if (flags & WR_STATE)
2741 return nfs4_share_conflict(current_fh,
2742 NFS4_SHARE_DENY_WRITE);
2743 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2744 return nfs4_share_conflict(current_fh,
2745 NFS4_SHARE_DENY_READ);
2749 * Allow READ/WRITE during grace period on recovered state only for files
2750 * that are not able to provide mandatory locking.
2753 grace_disallows_io(struct inode *inode)
2755 return locks_in_grace() && mandatory_lock(inode);
2758 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2761 * When sessions are used the stateid generation number is ignored
2764 if ((flags & HAS_SESSION) && in->si_generation == 0)
2767 /* If the client sends us a stateid from the future, it's buggy: */
2768 if (in->si_generation > ref->si_generation)
2769 return nfserr_bad_stateid;
2771 * The following, however, can happen. For example, if the
2772 * client sends an open and some IO at the same time, the open
2773 * may bump si_generation while the IO is still in flight.
2774 * Thanks to hard links and renames, the client never knows what
2775 * file an open will affect. So it could avoid that situation
2776 * only by serializing all opens and IO from the same open
2777 * owner. To recover from the old_stateid error, the client
2778 * will just have to retry the IO:
2780 if (in->si_generation < ref->si_generation)
2781 return nfserr_old_stateid;
2786 static int is_delegation_stateid(stateid_t *stateid)
2788 return stateid->si_fileid == 0;
2792 * Checks for stateid operations
2795 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2796 stateid_t *stateid, int flags, struct file **filpp)
2798 struct nfs4_stateid *stp = NULL;
2799 struct nfs4_delegation *dp = NULL;
2800 struct svc_fh *current_fh = &cstate->current_fh;
2801 struct inode *ino = current_fh->fh_dentry->d_inode;
2807 if (grace_disallows_io(ino))
2808 return nfserr_grace;
2810 if (nfsd4_has_session(cstate))
2811 flags |= HAS_SESSION;
2813 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2814 return check_special_stateids(current_fh, stateid, flags);
2816 status = nfserr_stale_stateid;
2817 if (STALE_STATEID(stateid))
2820 status = nfserr_bad_stateid;
2821 if (is_delegation_stateid(stateid)) {
2822 dp = find_delegation_stateid(ino, stateid);
2824 status = stateid_error_map(stateid);
2827 status = check_stateid_generation(stateid, &dp->dl_stateid,
2831 status = nfs4_check_delegmode(dp, flags);
2834 renew_client(dp->dl_client);
2836 *filpp = dp->dl_vfs_file;
2837 } else { /* open or lock stateid */
2838 stp = find_stateid(stateid, flags);
2840 status = stateid_error_map(stateid);
2843 if (nfs4_check_fh(current_fh, stp))
2845 if (!stp->st_stateowner->so_confirmed)
2847 status = check_stateid_generation(stateid, &stp->st_stateid,
2851 status = nfs4_check_openmode(stp, flags);
2854 renew_client(stp->st_stateowner->so_client);
2856 *filpp = stp->st_vfs_file;
2866 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2867 RD_STATE : WR_STATE;
2871 * Checks for sequence id mutating operations.
2874 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2875 stateid_t *stateid, int flags,
2876 struct nfs4_stateowner **sopp,
2877 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2879 struct nfs4_stateid *stp;
2880 struct nfs4_stateowner *sop;
2881 struct svc_fh *current_fh = &cstate->current_fh;
2884 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2885 seqid, STATEID_VAL(stateid));
2890 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2891 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2892 return nfserr_bad_stateid;
2895 if (STALE_STATEID(stateid))
2896 return nfserr_stale_stateid;
2898 if (nfsd4_has_session(cstate))
2899 flags |= HAS_SESSION;
2902 * We return BAD_STATEID if filehandle doesn't match stateid,
2903 * the confirmed flag is incorrecly set, or the generation
2904 * number is incorrect.
2906 stp = find_stateid(stateid, flags);
2909 * Also, we should make sure this isn't just the result of
2912 sop = search_close_lru(stateid->si_stateownerid, flags);
2914 return stateid_error_map(stateid);
2920 *sopp = sop = stp->st_stateowner;
2923 clientid_t *lockclid = &lock->v.new.clientid;
2924 struct nfs4_client *clp = sop->so_client;
2928 lkflg = setlkflg(lock->lk_type);
2930 if (lock->lk_is_new) {
2931 if (!sop->so_is_open_owner)
2932 return nfserr_bad_stateid;
2933 if (!(flags & HAS_SESSION) &&
2934 !same_clid(&clp->cl_clientid, lockclid))
2935 return nfserr_bad_stateid;
2936 /* stp is the open stateid */
2937 status = nfs4_check_openmode(stp, lkflg);
2941 /* stp is the lock stateid */
2942 status = nfs4_check_openmode(stp->st_openstp, lkflg);
2948 if (nfs4_check_fh(current_fh, stp)) {
2949 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2950 return nfserr_bad_stateid;
2954 * We now validate the seqid and stateid generation numbers.
2955 * For the moment, we ignore the possibility of
2956 * generation number wraparound.
2958 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2961 if (sop->so_confirmed && flags & CONFIRM) {
2962 dprintk("NFSD: preprocess_seqid_op: expected"
2963 " unconfirmed stateowner!\n");
2964 return nfserr_bad_stateid;
2966 if (!sop->so_confirmed && !(flags & CONFIRM)) {
2967 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2968 " confirmed yet!\n");
2969 return nfserr_bad_stateid;
2971 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2974 renew_client(sop->so_client);
2978 if (seqid == sop->so_seqid - 1) {
2979 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2980 /* indicate replay to calling function */
2981 return nfserr_replay_me;
2983 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2984 sop->so_seqid, seqid);
2986 return nfserr_bad_seqid;
2990 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2991 struct nfsd4_open_confirm *oc)
2994 struct nfs4_stateowner *sop;
2995 struct nfs4_stateid *stp;
2997 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2998 (int)cstate->current_fh.fh_dentry->d_name.len,
2999 cstate->current_fh.fh_dentry->d_name.name);
3001 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3007 if ((status = nfs4_preprocess_seqid_op(cstate,
3008 oc->oc_seqid, &oc->oc_req_stateid,
3009 CONFIRM | OPEN_STATE,
3010 &oc->oc_stateowner, &stp, NULL)))
3013 sop = oc->oc_stateowner;
3014 sop->so_confirmed = 1;
3015 update_stateid(&stp->st_stateid);
3016 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3017 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3018 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3020 nfsd4_create_clid_dir(sop->so_client);
3022 if (oc->oc_stateowner) {
3023 nfs4_get_stateowner(oc->oc_stateowner);
3024 cstate->replay_owner = oc->oc_stateowner;
3026 nfs4_unlock_state();
3032 * unset all bits in union bitmap (bmap) that
3033 * do not exist in share (from successful OPEN_DOWNGRADE)
3036 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3039 for (i = 1; i < 4; i++) {
3040 if ((i & access) != i)
3041 __clear_bit(i, bmap);
3046 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3049 for (i = 0; i < 4; i++) {
3050 if ((i & deny) != i)
3051 __clear_bit(i, bmap);
3056 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3057 struct nfsd4_compound_state *cstate,
3058 struct nfsd4_open_downgrade *od)
3061 struct nfs4_stateid *stp;
3062 unsigned int share_access;
3064 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3065 (int)cstate->current_fh.fh_dentry->d_name.len,
3066 cstate->current_fh.fh_dentry->d_name.name);
3068 if (!access_valid(od->od_share_access, cstate->minorversion)
3069 || !deny_valid(od->od_share_deny))
3070 return nfserr_inval;
3073 if ((status = nfs4_preprocess_seqid_op(cstate,
3077 &od->od_stateowner, &stp, NULL)))
3080 status = nfserr_inval;
3081 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3082 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3083 stp->st_access_bmap, od->od_share_access);
3086 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3087 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3088 stp->st_deny_bmap, od->od_share_deny);
3091 set_access(&share_access, stp->st_access_bmap);
3092 nfs4_file_downgrade(stp->st_vfs_file,
3093 share_access & ~od->od_share_access);
3095 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3096 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3098 update_stateid(&stp->st_stateid);
3099 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3102 if (od->od_stateowner) {
3103 nfs4_get_stateowner(od->od_stateowner);
3104 cstate->replay_owner = od->od_stateowner;
3106 nfs4_unlock_state();
3111 * nfs4_unlock_state() called after encode
3114 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3115 struct nfsd4_close *close)
3118 struct nfs4_stateid *stp;
3120 dprintk("NFSD: nfsd4_close on file %.*s\n",
3121 (int)cstate->current_fh.fh_dentry->d_name.len,
3122 cstate->current_fh.fh_dentry->d_name.name);
3125 /* check close_lru for replay */
3126 if ((status = nfs4_preprocess_seqid_op(cstate,
3129 OPEN_STATE | CLOSE_STATE,
3130 &close->cl_stateowner, &stp, NULL)))
3133 update_stateid(&stp->st_stateid);
3134 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3136 /* release_stateid() calls nfsd_close() if needed */
3137 release_open_stateid(stp);
3139 /* place unused nfs4_stateowners on so_close_lru list to be
3140 * released by the laundromat service after the lease period
3141 * to enable us to handle CLOSE replay
3143 if (list_empty(&close->cl_stateowner->so_stateids))
3144 move_to_close_lru(close->cl_stateowner);
3146 if (close->cl_stateowner) {
3147 nfs4_get_stateowner(close->cl_stateowner);
3148 cstate->replay_owner = close->cl_stateowner;
3150 nfs4_unlock_state();
3155 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3156 struct nfsd4_delegreturn *dr)
3158 struct nfs4_delegation *dp;
3159 stateid_t *stateid = &dr->dr_stateid;
3160 struct inode *inode;
3164 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3166 inode = cstate->current_fh.fh_dentry->d_inode;
3168 if (nfsd4_has_session(cstate))
3169 flags |= HAS_SESSION;
3171 status = nfserr_bad_stateid;
3172 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3174 status = nfserr_stale_stateid;
3175 if (STALE_STATEID(stateid))
3177 status = nfserr_bad_stateid;
3178 if (!is_delegation_stateid(stateid))
3180 dp = find_delegation_stateid(inode, stateid);
3182 status = stateid_error_map(stateid);
3185 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3188 renew_client(dp->dl_client);
3190 unhash_delegation(dp);
3192 nfs4_unlock_state();
3199 * Lock owner state (byte-range locks)
3201 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3202 #define LOCK_HASH_BITS 8
3203 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3204 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3207 end_offset(u64 start, u64 len)
3212 return end >= start ? end: NFS4_MAX_UINT64;
3215 /* last octet in a range */
3217 last_byte_offset(u64 start, u64 len)
3223 return end > start ? end - 1: NFS4_MAX_UINT64;
3226 #define lockownerid_hashval(id) \
3227 ((id) & LOCK_HASH_MASK)
3229 static inline unsigned int
3230 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3231 struct xdr_netobj *ownername)
3233 return (file_hashval(inode) + cl_id
3234 + opaque_hashval(ownername->data, ownername->len))
3238 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3239 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3240 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3242 static struct nfs4_stateid *
3243 find_stateid(stateid_t *stid, int flags)
3245 struct nfs4_stateid *local;
3246 u32 st_id = stid->si_stateownerid;
3247 u32 f_id = stid->si_fileid;
3248 unsigned int hashval;
3250 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3251 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3252 hashval = stateid_hashval(st_id, f_id);
3253 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3254 if ((local->st_stateid.si_stateownerid == st_id) &&
3255 (local->st_stateid.si_fileid == f_id))
3260 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3261 hashval = stateid_hashval(st_id, f_id);
3262 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3263 if ((local->st_stateid.si_stateownerid == st_id) &&
3264 (local->st_stateid.si_fileid == f_id))
3271 static struct nfs4_delegation *
3272 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3274 struct nfs4_file *fp;
3275 struct nfs4_delegation *dl;
3277 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3280 fp = find_file(ino);
3283 dl = find_delegation_file(fp, stid);
3289 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3290 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3291 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3292 * locking, this prevents us from being completely protocol-compliant. The
3293 * real solution to this problem is to start using unsigned file offsets in
3294 * the VFS, but this is a very deep change!
3297 nfs4_transform_lock_offset(struct file_lock *lock)
3299 if (lock->fl_start < 0)
3300 lock->fl_start = OFFSET_MAX;
3301 if (lock->fl_end < 0)
3302 lock->fl_end = OFFSET_MAX;
3305 /* Hack!: For now, we're defining this just so we can use a pointer to it
3306 * as a unique cookie to identify our (NFSv4's) posix locks. */
3307 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3311 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3313 struct nfs4_stateowner *sop;
3316 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3317 sop = (struct nfs4_stateowner *) fl->fl_owner;
3318 hval = lockownerid_hashval(sop->so_id);
3319 kref_get(&sop->so_ref);
3321 deny->ld_clientid = sop->so_client->cl_clientid;
3323 deny->ld_sop = NULL;
3324 deny->ld_clientid.cl_boot = 0;
3325 deny->ld_clientid.cl_id = 0;
3327 deny->ld_start = fl->fl_start;
3328 deny->ld_length = NFS4_MAX_UINT64;
3329 if (fl->fl_end != NFS4_MAX_UINT64)
3330 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3331 deny->ld_type = NFS4_READ_LT;
3332 if (fl->fl_type != F_RDLCK)
3333 deny->ld_type = NFS4_WRITE_LT;
3336 static struct nfs4_stateowner *
3337 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3338 struct xdr_netobj *owner)
3340 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3341 struct nfs4_stateowner *op;
3343 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3344 if (same_owner_str(op, owner, clid))
3351 * Alloc a lock owner structure.
3352 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3355 * strhashval = lock_ownerstr_hashval
3358 static struct nfs4_stateowner *
3359 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3360 struct nfs4_stateowner *sop;
3361 struct nfs4_replay *rp;
3362 unsigned int idhashval;
3364 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3366 idhashval = lockownerid_hashval(current_ownerid);
3367 INIT_LIST_HEAD(&sop->so_idhash);
3368 INIT_LIST_HEAD(&sop->so_strhash);
3369 INIT_LIST_HEAD(&sop->so_perclient);
3370 INIT_LIST_HEAD(&sop->so_stateids);
3371 INIT_LIST_HEAD(&sop->so_perstateid);
3372 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3374 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3375 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3376 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3377 sop->so_is_open_owner = 0;
3378 sop->so_id = current_ownerid++;
3379 sop->so_client = clp;
3380 /* It is the openowner seqid that will be incremented in encode in the
3381 * case of new lockowners; so increment the lock seqid manually: */
3382 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3383 sop->so_confirmed = 1;
3384 rp = &sop->so_replay;
3385 rp->rp_status = nfserr_serverfault;
3387 rp->rp_buf = rp->rp_ibuf;
3391 static struct nfs4_stateid *
3392 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3394 struct nfs4_stateid *stp;
3395 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3397 stp = nfs4_alloc_stateid();
3400 INIT_LIST_HEAD(&stp->st_hash);
3401 INIT_LIST_HEAD(&stp->st_perfile);
3402 INIT_LIST_HEAD(&stp->st_perstateowner);
3403 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3404 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3405 list_add(&stp->st_perfile, &fp->fi_stateids);
3406 list_add(&stp->st_perstateowner, &sop->so_stateids);
3407 stp->st_stateowner = sop;
3410 stp->st_stateid.si_boot = get_seconds();
3411 stp->st_stateid.si_stateownerid = sop->so_id;
3412 stp->st_stateid.si_fileid = fp->fi_id;
3413 stp->st_stateid.si_generation = 0;
3414 stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3415 stp->st_access_bmap = open_stp->st_access_bmap;
3416 stp->st_deny_bmap = open_stp->st_deny_bmap;
3417 stp->st_openstp = open_stp;
3424 check_lock_length(u64 offset, u64 length)
3426 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3427 LOFF_OVERFLOW(offset, length)));
3434 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3435 struct nfsd4_lock *lock)
3437 struct nfs4_stateowner *open_sop = NULL;
3438 struct nfs4_stateowner *lock_sop = NULL;
3439 struct nfs4_stateid *lock_stp;
3441 struct file_lock file_lock;
3442 struct file_lock conflock;
3444 unsigned int strhashval;
3448 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3449 (long long) lock->lk_offset,
3450 (long long) lock->lk_length);
3452 if (check_lock_length(lock->lk_offset, lock->lk_length))
3453 return nfserr_inval;
3455 if ((status = fh_verify(rqstp, &cstate->current_fh,
3456 S_IFREG, NFSD_MAY_LOCK))) {
3457 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3463 if (lock->lk_is_new) {
3465 * Client indicates that this is a new lockowner.
3466 * Use open owner and open stateid to create lock owner and
3469 struct nfs4_stateid *open_stp = NULL;
3470 struct nfs4_file *fp;
3472 status = nfserr_stale_clientid;
3473 if (!nfsd4_has_session(cstate) &&
3474 STALE_CLIENTID(&lock->lk_new_clientid))
3477 /* validate and update open stateid and open seqid */
3478 status = nfs4_preprocess_seqid_op(cstate,
3479 lock->lk_new_open_seqid,
3480 &lock->lk_new_open_stateid,
3482 &lock->lk_replay_owner, &open_stp,
3486 open_sop = lock->lk_replay_owner;
3487 /* create lockowner and lock stateid */
3488 fp = open_stp->st_file;
3489 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3490 open_sop->so_client->cl_clientid.cl_id,
3491 &lock->v.new.owner);
3492 /* XXX: Do we need to check for duplicate stateowners on
3493 * the same file, or should they just be allowed (and
3494 * create new stateids)? */
3495 status = nfserr_resource;
3496 lock_sop = alloc_init_lock_stateowner(strhashval,
3497 open_sop->so_client, open_stp, lock);
3498 if (lock_sop == NULL)
3500 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3501 if (lock_stp == NULL)
3504 /* lock (lock owner + lock stateid) already exists */
3505 status = nfs4_preprocess_seqid_op(cstate,
3506 lock->lk_old_lock_seqid,
3507 &lock->lk_old_lock_stateid,
3509 &lock->lk_replay_owner, &lock_stp, lock);
3512 lock_sop = lock->lk_replay_owner;
3514 /* lock->lk_replay_owner and lock_stp have been created or found */
3515 filp = lock_stp->st_vfs_file;
3517 status = nfserr_grace;
3518 if (locks_in_grace() && !lock->lk_reclaim)
3520 status = nfserr_no_grace;
3521 if (!locks_in_grace() && lock->lk_reclaim)
3524 locks_init_lock(&file_lock);
3525 switch (lock->lk_type) {
3528 file_lock.fl_type = F_RDLCK;
3532 case NFS4_WRITEW_LT:
3533 file_lock.fl_type = F_WRLCK;
3537 status = nfserr_inval;
3540 file_lock.fl_owner = (fl_owner_t)lock_sop;
3541 file_lock.fl_pid = current->tgid;
3542 file_lock.fl_file = filp;
3543 file_lock.fl_flags = FL_POSIX;
3544 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3546 file_lock.fl_start = lock->lk_offset;
3547 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3548 nfs4_transform_lock_offset(&file_lock);
3551 * Try to lock the file in the VFS.
3552 * Note: locks.c uses the BKL to protect the inode's lock list.
3555 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3557 case 0: /* success! */
3558 update_stateid(&lock_stp->st_stateid);
3559 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3563 case (EAGAIN): /* conflock holds conflicting lock */
3564 status = nfserr_denied;
3565 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3566 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3569 status = nfserr_deadlock;
3572 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3573 status = nfserr_resource;
3577 if (status && lock->lk_is_new && lock_sop)
3578 release_lockowner(lock_sop);
3579 if (lock->lk_replay_owner) {
3580 nfs4_get_stateowner(lock->lk_replay_owner);
3581 cstate->replay_owner = lock->lk_replay_owner;
3583 nfs4_unlock_state();
3588 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3589 * so we do a temporary open here just to get an open file to pass to
3590 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3593 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3598 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3601 err = vfs_test_lock(file, lock);
3610 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3611 struct nfsd4_lockt *lockt)
3613 struct inode *inode;
3614 struct file_lock file_lock;
3618 if (locks_in_grace())
3619 return nfserr_grace;
3621 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3622 return nfserr_inval;
3624 lockt->lt_stateowner = NULL;
3627 status = nfserr_stale_clientid;
3628 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3631 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3632 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3633 if (status == nfserr_symlink)
3634 status = nfserr_inval;
3638 inode = cstate->current_fh.fh_dentry->d_inode;
3639 locks_init_lock(&file_lock);
3640 switch (lockt->lt_type) {
3643 file_lock.fl_type = F_RDLCK;
3646 case NFS4_WRITEW_LT:
3647 file_lock.fl_type = F_WRLCK;
3650 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3651 status = nfserr_inval;
3655 lockt->lt_stateowner = find_lockstateowner_str(inode,
3656 &lockt->lt_clientid, &lockt->lt_owner);
3657 if (lockt->lt_stateowner)
3658 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3659 file_lock.fl_pid = current->tgid;
3660 file_lock.fl_flags = FL_POSIX;
3662 file_lock.fl_start = lockt->lt_offset;
3663 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3665 nfs4_transform_lock_offset(&file_lock);
3668 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3670 status = nfserrno(error);
3673 if (file_lock.fl_type != F_UNLCK) {
3674 status = nfserr_denied;
3675 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3678 nfs4_unlock_state();
3683 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3684 struct nfsd4_locku *locku)
3686 struct nfs4_stateid *stp;
3687 struct file *filp = NULL;
3688 struct file_lock file_lock;
3692 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3693 (long long) locku->lu_offset,
3694 (long long) locku->lu_length);
3696 if (check_lock_length(locku->lu_offset, locku->lu_length))
3697 return nfserr_inval;
3701 if ((status = nfs4_preprocess_seqid_op(cstate,
3705 &locku->lu_stateowner, &stp, NULL)))
3708 filp = stp->st_vfs_file;
3710 locks_init_lock(&file_lock);
3711 file_lock.fl_type = F_UNLCK;
3712 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3713 file_lock.fl_pid = current->tgid;
3714 file_lock.fl_file = filp;
3715 file_lock.fl_flags = FL_POSIX;
3716 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3717 file_lock.fl_start = locku->lu_offset;
3719 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3720 nfs4_transform_lock_offset(&file_lock);
3723 * Try to unlock the file in the VFS.
3725 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3727 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3731 * OK, unlock succeeded; the only thing left to do is update the stateid.
3733 update_stateid(&stp->st_stateid);
3734 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3737 if (locku->lu_stateowner) {
3738 nfs4_get_stateowner(locku->lu_stateowner);
3739 cstate->replay_owner = locku->lu_stateowner;
3741 nfs4_unlock_state();
3745 status = nfserrno(err);
3751 * 1: locks held by lockowner
3752 * 0: no locks held by lockowner
3755 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3757 struct file_lock **flpp;
3758 struct inode *inode = filp->f_path.dentry->d_inode;
3762 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3763 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3774 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3775 struct nfsd4_compound_state *cstate,
3776 struct nfsd4_release_lockowner *rlockowner)
3778 clientid_t *clid = &rlockowner->rl_clientid;
3779 struct nfs4_stateowner *sop;
3780 struct nfs4_stateid *stp;
3781 struct xdr_netobj *owner = &rlockowner->rl_owner;
3782 struct list_head matches;
3786 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3787 clid->cl_boot, clid->cl_id);
3789 /* XXX check for lease expiration */
3791 status = nfserr_stale_clientid;
3792 if (STALE_CLIENTID(clid))
3797 status = nfserr_locks_held;
3798 /* XXX: we're doing a linear search through all the lockowners.
3799 * Yipes! For now we'll just hope clients aren't really using
3800 * release_lockowner much, but eventually we have to fix these
3801 * data structures. */
3802 INIT_LIST_HEAD(&matches);
3803 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3804 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3805 if (!same_owner_str(sop, owner, clid))
3807 list_for_each_entry(stp, &sop->so_stateids,
3809 if (check_for_locks(stp->st_vfs_file, sop))
3811 /* Note: so_perclient unused for lockowners,
3812 * so it's OK to fool with here. */
3813 list_add(&sop->so_perclient, &matches);
3817 /* Clients probably won't expect us to return with some (but not all)
3818 * of the lockowner state released; so don't release any until all
3819 * have been checked. */
3821 while (!list_empty(&matches)) {
3822 sop = list_entry(matches.next, struct nfs4_stateowner,
3824 /* unhash_stateowner deletes so_perclient only
3825 * for openowners. */
3826 list_del(&sop->so_perclient);
3827 release_lockowner(sop);
3830 nfs4_unlock_state();
3834 static inline struct nfs4_client_reclaim *
3837 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3841 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3843 unsigned int strhashval = clientstr_hashval(name);
3844 struct nfs4_client *clp;
3846 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3851 * failure => all reset bets are off, nfserr_no_grace...
3854 nfs4_client_to_reclaim(const char *name)
3856 unsigned int strhashval;
3857 struct nfs4_client_reclaim *crp = NULL;
3859 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3860 crp = alloc_reclaim();
3863 strhashval = clientstr_hashval(name);
3864 INIT_LIST_HEAD(&crp->cr_strhash);
3865 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3866 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3867 reclaim_str_hashtbl_size++;
3872 nfs4_release_reclaim(void)
3874 struct nfs4_client_reclaim *crp = NULL;
3877 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3878 while (!list_empty(&reclaim_str_hashtbl[i])) {
3879 crp = list_entry(reclaim_str_hashtbl[i].next,
3880 struct nfs4_client_reclaim, cr_strhash);
3881 list_del(&crp->cr_strhash);
3883 reclaim_str_hashtbl_size--;
3886 BUG_ON(reclaim_str_hashtbl_size);
3890 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3891 static struct nfs4_client_reclaim *
3892 nfs4_find_reclaim_client(clientid_t *clid)
3894 unsigned int strhashval;
3895 struct nfs4_client *clp;
3896 struct nfs4_client_reclaim *crp = NULL;
3899 /* find clientid in conf_id_hashtbl */
3900 clp = find_confirmed_client(clid);
3904 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3905 clp->cl_name.len, clp->cl_name.data,
3908 /* find clp->cl_name in reclaim_str_hashtbl */
3909 strhashval = clientstr_hashval(clp->cl_recdir);
3910 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3911 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3919 * Called from OPEN. Look for clientid in reclaim list.
3922 nfs4_check_open_reclaim(clientid_t *clid)
3924 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3927 /* initialization to perform at module load time: */
3930 nfs4_state_init(void)
3934 status = nfsd4_init_slabs();
3937 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3938 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3939 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3940 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3941 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3942 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3944 for (i = 0; i < SESSION_HASH_SIZE; i++)
3945 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3946 for (i = 0; i < FILE_HASH_SIZE; i++) {
3947 INIT_LIST_HEAD(&file_hashtbl[i]);
3949 for (i = 0; i < OWNER_HASH_SIZE; i++) {
3950 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3951 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3953 for (i = 0; i < STATEID_HASH_SIZE; i++) {
3954 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3955 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3957 for (i = 0; i < LOCK_HASH_SIZE; i++) {
3958 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3959 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3961 memset(&onestateid, ~0, sizeof(stateid_t));
3962 INIT_LIST_HEAD(&close_lru);
3963 INIT_LIST_HEAD(&client_lru);
3964 INIT_LIST_HEAD(&del_recall_lru);
3965 reclaim_str_hashtbl_size = 0;
3970 nfsd4_load_reboot_recovery_data(void)
3975 nfsd4_init_recdir(user_recovery_dirname);
3976 status = nfsd4_recdir_load();
3977 nfs4_unlock_state();
3979 printk("NFSD: Failure reading reboot recovery data\n");
3983 * Since the lifetime of a delegation isn't limited to that of an open, a
3984 * client may quite reasonably hang on to a delegation as long as it has
3985 * the inode cached. This becomes an obvious problem the first time a
3986 * client's inode cache approaches the size of the server's total memory.
3988 * For now we avoid this problem by imposing a hard limit on the number
3989 * of delegations, which varies according to the server's memory size.
3992 set_max_delegations(void)
3995 * Allow at most 4 delegations per megabyte of RAM. Quick
3996 * estimates suggest that in the worst case (where every delegation
3997 * is for a different inode), a delegation could take about 1.5K,
3998 * giving a worst case usage of about 6% of memory.
4000 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4003 /* initialization to perform when the nfsd service is started: */
4006 __nfs4_state_start(void)
4010 boot_time = get_seconds();
4011 locks_start_grace(&nfsd4_manager);
4012 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4014 ret = set_callback_cred();
4017 laundry_wq = create_singlethread_workqueue("nfsd4");
4018 if (laundry_wq == NULL)
4020 ret = nfsd4_create_callback_queue();
4022 goto out_free_laundry;
4023 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4024 set_max_delegations();
4027 destroy_workqueue(laundry_wq);
4032 nfs4_state_start(void)
4038 nfsd4_load_reboot_recovery_data();
4039 ret = __nfs4_state_start();
4047 __nfs4_state_shutdown(void)
4050 struct nfs4_client *clp = NULL;
4051 struct nfs4_delegation *dp = NULL;
4052 struct list_head *pos, *next, reaplist;
4054 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4055 while (!list_empty(&conf_id_hashtbl[i])) {
4056 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4059 while (!list_empty(&unconf_str_hashtbl[i])) {
4060 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4064 INIT_LIST_HEAD(&reaplist);
4065 spin_lock(&recall_lock);
4066 list_for_each_safe(pos, next, &del_recall_lru) {
4067 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4068 list_move(&dp->dl_recall_lru, &reaplist);
4070 spin_unlock(&recall_lock);
4071 list_for_each_safe(pos, next, &reaplist) {
4072 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4073 list_del_init(&dp->dl_recall_lru);
4074 unhash_delegation(dp);
4077 nfsd4_shutdown_recdir();
4082 nfs4_state_shutdown(void)
4084 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4085 destroy_workqueue(laundry_wq);
4086 locks_end_grace(&nfsd4_manager);
4088 nfs4_release_reclaim();
4089 __nfs4_state_shutdown();
4090 nfsd4_destroy_callback_queue();
4091 nfs4_unlock_state();
4095 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4096 * accessed when nfsd is starting.
4099 nfs4_set_recdir(char *recdir)
4101 strcpy(user_recovery_dirname, recdir);
4105 * Change the NFSv4 recovery directory to recdir.
4108 nfs4_reset_recoverydir(char *recdir)
4113 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4117 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4118 nfs4_set_recdir(recdir);
4126 nfs4_recoverydir(void)
4128 return user_recovery_dirname;