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>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
45 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 time_t nfsd4_lease = 90; /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static stateid_t zerostateid; /* bits all 0 */
55 static stateid_t onestateid; /* bits all 1 */
56 static u64 current_sessionid = 1;
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
73 * Currently used for the del_recall_lru and file hash table. In an
74 * effort to decrease the scope of the client_mutex, this spinlock may
75 * eventually cover more:
77 static DEFINE_SPINLOCK(recall_lock);
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
87 mutex_lock(&client_mutex);
91 nfs4_unlock_state(void)
93 mutex_unlock(&client_mutex);
97 opaque_hashval(const void *ptr, int nbytes)
99 unsigned char *cptr = (unsigned char *) ptr;
109 static struct list_head del_recall_lru;
112 put_nfs4_file(struct nfs4_file *fi)
114 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115 list_del(&fi->fi_hash);
116 spin_unlock(&recall_lock);
118 kmem_cache_free(file_slab, fi);
123 get_nfs4_file(struct nfs4_file *fi)
125 atomic_inc(&fi->fi_ref);
128 static int num_delegations;
129 unsigned int max_delegations;
132 * Open owner state (share locks)
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS 8
137 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
140 #define ownerid_hashval(id) \
141 ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS 8
150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS 10
154 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
157 #define file_hashval(x) \
158 hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id) \
160 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
167 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168 atomic_inc(&fp->fi_access[oflag]);
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
173 if (oflag == O_RDWR) {
174 __nfs4_file_get_access(fp, O_RDONLY);
175 __nfs4_file_get_access(fp, O_WRONLY);
177 __nfs4_file_get_access(fp, oflag);
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
182 if (fp->fi_fds[oflag]) {
183 fput(fp->fi_fds[oflag]);
184 fp->fi_fds[oflag] = NULL;
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
190 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191 nfs4_file_put_fd(fp, O_RDWR);
192 nfs4_file_put_fd(fp, oflag);
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
198 if (oflag == O_RDWR) {
199 __nfs4_file_put_access(fp, O_RDONLY);
200 __nfs4_file_put_access(fp, O_WRONLY);
202 __nfs4_file_put_access(fp, oflag);
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
208 struct nfs4_delegation *dp;
209 struct nfs4_file *fp = stp->st_file;
211 dprintk("NFSD alloc_init_deleg\n");
213 * Major work on the lease subsystem (for example, to support
214 * calbacks on stat) will be required before we can support
215 * write delegations properly.
217 if (type != NFS4_OPEN_DELEGATE_READ)
219 if (fp->fi_had_conflict)
221 if (num_delegations > max_delegations)
223 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
227 INIT_LIST_HEAD(&dp->dl_perfile);
228 INIT_LIST_HEAD(&dp->dl_perclnt);
229 INIT_LIST_HEAD(&dp->dl_recall_lru);
233 dp->dl_vfs_file = find_readable_file(fp);
234 get_file(dp->dl_vfs_file);
237 dp->dl_stateid.si_boot = boot_time;
238 dp->dl_stateid.si_stateownerid = current_delegid++;
239 dp->dl_stateid.si_fileid = 0;
240 dp->dl_stateid.si_generation = 0;
241 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
243 atomic_set(&dp->dl_count, 1);
244 list_add(&dp->dl_perfile, &fp->fi_delegations);
245 list_add(&dp->dl_perclnt, &clp->cl_delegations);
246 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
251 nfs4_put_delegation(struct nfs4_delegation *dp)
253 if (atomic_dec_and_test(&dp->dl_count)) {
254 dprintk("NFSD: freeing dp %p\n",dp);
255 put_nfs4_file(dp->dl_file);
256 fput(dp->dl_vfs_file);
257 kmem_cache_free(deleg_slab, dp);
262 /* Remove the associated file_lock first, then remove the delegation.
263 * lease_modify() is called to remove the FS_LEASE file_lock from
264 * the i_flock list, eventually calling nfsd's lock_manager
265 * fl_release_callback.
268 nfs4_close_delegation(struct nfs4_delegation *dp)
270 dprintk("NFSD: close_delegation dp %p\n",dp);
271 /* XXX: do we even need this check?: */
273 vfs_setlease(dp->dl_vfs_file, F_UNLCK, &dp->dl_flock);
276 /* Called under the state lock. */
278 unhash_delegation(struct nfs4_delegation *dp)
280 list_del_init(&dp->dl_perfile);
281 list_del_init(&dp->dl_perclnt);
282 spin_lock(&recall_lock);
283 list_del_init(&dp->dl_recall_lru);
284 spin_unlock(&recall_lock);
285 nfs4_close_delegation(dp);
286 nfs4_put_delegation(dp);
293 /* client_lock protects the client lru list and session hash table */
294 static DEFINE_SPINLOCK(client_lock);
296 /* Hash tables for nfs4_clientid state */
297 #define CLIENT_HASH_BITS 4
298 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
299 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
301 #define clientid_hashval(id) \
302 ((id) & CLIENT_HASH_MASK)
303 #define clientstr_hashval(name) \
304 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
306 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
307 * used in reboot/reset lease grace period processing
309 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
310 * setclientid_confirmed info.
312 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
315 * client_lru holds client queue ordered by nfs4_client.cl_time
318 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
319 * for last close replay.
321 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
322 static int reclaim_str_hashtbl_size = 0;
323 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
324 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
325 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
326 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
327 static struct list_head client_lru;
328 static struct list_head close_lru;
330 static void unhash_generic_stateid(struct nfs4_stateid *stp)
332 list_del(&stp->st_hash);
333 list_del(&stp->st_perfile);
334 list_del(&stp->st_perstateowner);
337 static void free_generic_stateid(struct nfs4_stateid *stp)
339 put_nfs4_file(stp->st_file);
340 kmem_cache_free(stateid_slab, stp);
343 static void release_lock_stateid(struct nfs4_stateid *stp)
347 unhash_generic_stateid(stp);
348 file = find_any_file(stp->st_file);
350 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
351 free_generic_stateid(stp);
354 static void unhash_lockowner(struct nfs4_stateowner *sop)
356 struct nfs4_stateid *stp;
358 list_del(&sop->so_idhash);
359 list_del(&sop->so_strhash);
360 list_del(&sop->so_perstateid);
361 while (!list_empty(&sop->so_stateids)) {
362 stp = list_first_entry(&sop->so_stateids,
363 struct nfs4_stateid, st_perstateowner);
364 release_lock_stateid(stp);
368 static void release_lockowner(struct nfs4_stateowner *sop)
370 unhash_lockowner(sop);
371 nfs4_put_stateowner(sop);
375 release_stateid_lockowners(struct nfs4_stateid *open_stp)
377 struct nfs4_stateowner *lock_sop;
379 while (!list_empty(&open_stp->st_lockowners)) {
380 lock_sop = list_entry(open_stp->st_lockowners.next,
381 struct nfs4_stateowner, so_perstateid);
382 /* list_del(&open_stp->st_lockowners); */
383 BUG_ON(lock_sop->so_is_open_owner);
384 release_lockowner(lock_sop);
389 * We store the NONE, READ, WRITE, and BOTH bits separately in the
390 * st_{access,deny}_bmap field of the stateid, in order to track not
391 * only what share bits are currently in force, but also what
392 * combinations of share bits previous opens have used. This allows us
393 * to enforce the recommendation of rfc 3530 14.2.19 that the server
394 * return an error if the client attempt to downgrade to a combination
395 * of share bits not explicable by closing some of its previous opens.
397 * XXX: This enforcement is actually incomplete, since we don't keep
398 * track of access/deny bit combinations; so, e.g., we allow:
400 * OPEN allow read, deny write
401 * OPEN allow both, deny none
402 * DOWNGRADE allow read, deny none
404 * which we should reject.
407 set_access(unsigned int *access, unsigned long bmap) {
411 for (i = 1; i < 4; i++) {
412 if (test_bit(i, &bmap))
418 set_deny(unsigned int *deny, unsigned long bmap) {
422 for (i = 0; i < 4; i++) {
423 if (test_bit(i, &bmap))
429 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
430 unsigned int access, deny;
432 set_access(&access, stp->st_access_bmap);
433 set_deny(&deny, stp->st_deny_bmap);
434 if ((access & open->op_share_deny) || (deny & open->op_share_access))
439 static int nfs4_access_to_omode(u32 access)
441 switch (access & NFS4_SHARE_ACCESS_BOTH) {
442 case NFS4_SHARE_ACCESS_READ:
444 case NFS4_SHARE_ACCESS_WRITE:
446 case NFS4_SHARE_ACCESS_BOTH:
452 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
456 set_access(&access, stp->st_access_bmap);
457 return nfs4_access_to_omode(access);
460 static void release_open_stateid(struct nfs4_stateid *stp)
462 int oflag = nfs4_access_bmap_to_omode(stp);
464 unhash_generic_stateid(stp);
465 release_stateid_lockowners(stp);
466 nfs4_file_put_access(stp->st_file, oflag);
467 free_generic_stateid(stp);
470 static void unhash_openowner(struct nfs4_stateowner *sop)
472 struct nfs4_stateid *stp;
474 list_del(&sop->so_idhash);
475 list_del(&sop->so_strhash);
476 list_del(&sop->so_perclient);
477 list_del(&sop->so_perstateid); /* XXX: necessary? */
478 while (!list_empty(&sop->so_stateids)) {
479 stp = list_first_entry(&sop->so_stateids,
480 struct nfs4_stateid, st_perstateowner);
481 release_open_stateid(stp);
485 static void release_openowner(struct nfs4_stateowner *sop)
487 unhash_openowner(sop);
488 list_del(&sop->so_close_lru);
489 nfs4_put_stateowner(sop);
492 #define SESSION_HASH_SIZE 512
493 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
496 hash_sessionid(struct nfs4_sessionid *sessionid)
498 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
500 return sid->sequence % SESSION_HASH_SIZE;
504 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
506 u32 *ptr = (u32 *)(&sessionid->data[0]);
507 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
511 gen_sessionid(struct nfsd4_session *ses)
513 struct nfs4_client *clp = ses->se_client;
514 struct nfsd4_sessionid *sid;
516 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
517 sid->clientid = clp->cl_clientid;
518 sid->sequence = current_sessionid++;
523 * The protocol defines ca_maxresponssize_cached to include the size of
524 * the rpc header, but all we need to cache is the data starting after
525 * the end of the initial SEQUENCE operation--the rest we regenerate
526 * each time. Therefore we can advertise a ca_maxresponssize_cached
527 * value that is the number of bytes in our cache plus a few additional
528 * bytes. In order to stay on the safe side, and not promise more than
529 * we can cache, those additional bytes must be the minimum possible: 24
530 * bytes of rpc header (xid through accept state, with AUTH_NULL
531 * verifier), 12 for the compound header (with zero-length tag), and 44
532 * for the SEQUENCE op response:
534 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
537 free_session_slots(struct nfsd4_session *ses)
541 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
542 kfree(ses->se_slots[i]);
546 * We don't actually need to cache the rpc and session headers, so we
547 * can allocate a little less for each slot:
549 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
551 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
554 static int nfsd4_sanitize_slot_size(u32 size)
556 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
557 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
563 * XXX: If we run out of reserved DRC memory we could (up to a point)
564 * re-negotiate active sessions and reduce their slot usage to make
565 * rooom for new connections. For now we just fail the create session.
567 static int nfsd4_get_drc_mem(int slotsize, u32 num)
571 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
573 spin_lock(&nfsd_drc_lock);
574 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
575 nfsd_drc_max_mem - nfsd_drc_mem_used);
576 num = min_t(int, num, avail / slotsize);
577 nfsd_drc_mem_used += num * slotsize;
578 spin_unlock(&nfsd_drc_lock);
583 static void nfsd4_put_drc_mem(int slotsize, int num)
585 spin_lock(&nfsd_drc_lock);
586 nfsd_drc_mem_used -= slotsize * num;
587 spin_unlock(&nfsd_drc_lock);
590 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
592 struct nfsd4_session *new;
595 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
596 + sizeof(struct nfsd4_session) > PAGE_SIZE);
597 mem = numslots * sizeof(struct nfsd4_slot *);
599 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
602 /* allocate each struct nfsd4_slot and data cache in one piece */
603 for (i = 0; i < numslots; i++) {
604 mem = sizeof(struct nfsd4_slot) + slotsize;
605 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
606 if (!new->se_slots[i])
612 kfree(new->se_slots[i]);
617 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
619 u32 maxrpc = nfsd_serv->sv_max_mesg;
621 new->maxreqs = numslots;
622 new->maxresp_cached = slotsize + NFSD_MIN_HDR_SEQ_SZ;
623 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
624 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
625 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
628 static void free_conn(struct nfsd4_conn *c)
630 svc_xprt_put(c->cn_xprt);
634 static void nfsd4_conn_lost(struct svc_xpt_user *u)
636 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
637 struct nfs4_client *clp = c->cn_session->se_client;
639 spin_lock(&clp->cl_lock);
640 if (!list_empty(&c->cn_persession)) {
641 list_del(&c->cn_persession);
644 spin_unlock(&clp->cl_lock);
645 /* XXX: mark callback for update, probe callback */
648 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
650 struct nfsd4_conn *conn;
652 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
655 svc_xprt_get(rqstp->rq_xprt);
656 conn->cn_xprt = rqstp->rq_xprt;
657 conn->cn_flags = flags;
658 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
662 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
664 conn->cn_session = ses;
665 list_add(&conn->cn_persession, &ses->se_conns);
668 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
670 struct nfs4_client *clp = ses->se_client;
672 spin_lock(&clp->cl_lock);
673 __nfsd4_hash_conn(conn, ses);
674 spin_unlock(&clp->cl_lock);
677 static int nfsd4_register_conn(struct nfsd4_conn *conn)
679 conn->cn_xpt_user.callback = nfsd4_conn_lost;
680 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
683 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
685 struct nfsd4_conn *conn;
688 conn = alloc_conn(rqstp, dir);
690 return nfserr_jukebox;
691 nfsd4_hash_conn(conn, ses);
692 ret = nfsd4_register_conn(conn);
694 /* oops; xprt is already down: */
695 nfsd4_conn_lost(&conn->cn_xpt_user);
699 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
701 u32 dir = NFS4_CDFC4_FORE;
703 if (ses->se_flags & SESSION4_BACK_CHAN)
704 dir |= NFS4_CDFC4_BACK;
706 return nfsd4_new_conn(rqstp, ses, dir);
709 /* must be called under client_lock */
710 static void nfsd4_del_conns(struct nfsd4_session *s)
712 struct nfs4_client *clp = s->se_client;
713 struct nfsd4_conn *c;
715 spin_lock(&clp->cl_lock);
716 while (!list_empty(&s->se_conns)) {
717 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
718 list_del_init(&c->cn_persession);
719 spin_unlock(&clp->cl_lock);
721 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
724 spin_lock(&clp->cl_lock);
726 spin_unlock(&clp->cl_lock);
729 void free_session(struct kref *kref)
731 struct nfsd4_session *ses;
734 ses = container_of(kref, struct nfsd4_session, se_ref);
735 nfsd4_del_conns(ses);
736 spin_lock(&nfsd_drc_lock);
737 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
738 nfsd_drc_mem_used -= mem;
739 spin_unlock(&nfsd_drc_lock);
740 free_session_slots(ses);
744 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
746 struct nfsd4_session *new;
747 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
748 int numslots, slotsize;
753 * Note decreasing slot size below client's request may
754 * make it difficult for client to function correctly, whereas
755 * decreasing the number of slots will (just?) affect
756 * performance. When short on memory we therefore prefer to
757 * decrease number of slots instead of their size.
759 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
760 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
764 new = alloc_session(slotsize, numslots);
766 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
769 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
771 new->se_client = clp;
774 INIT_LIST_HEAD(&new->se_conns);
776 new->se_cb_seq_nr = 1;
777 new->se_flags = cses->flags;
778 new->se_cb_prog = cses->callback_prog;
779 kref_init(&new->se_ref);
780 idx = hash_sessionid(&new->se_sessionid);
781 spin_lock(&client_lock);
782 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
783 spin_lock(&clp->cl_lock);
784 list_add(&new->se_perclnt, &clp->cl_sessions);
785 spin_unlock(&clp->cl_lock);
786 spin_unlock(&client_lock);
788 status = nfsd4_new_conn_from_crses(rqstp, new);
789 /* whoops: benny points out, status is ignored! (err, or bogus) */
791 free_session(&new->se_ref);
794 if (cses->flags & SESSION4_BACK_CHAN) {
795 struct sockaddr *sa = svc_addr(rqstp);
797 * This is a little silly; with sessions there's no real
798 * use for the callback address. Use the peer address
799 * as a reasonable default for now, but consider fixing
800 * the rpc client not to require an address in the
803 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
804 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
806 nfsd4_probe_callback(clp);
810 /* caller must hold client_lock */
811 static struct nfsd4_session *
812 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
814 struct nfsd4_session *elem;
817 dump_sessionid(__func__, sessionid);
818 idx = hash_sessionid(sessionid);
819 /* Search in the appropriate list */
820 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
821 if (!memcmp(elem->se_sessionid.data, sessionid->data,
822 NFS4_MAX_SESSIONID_LEN)) {
827 dprintk("%s: session not found\n", __func__);
831 /* caller must hold client_lock */
833 unhash_session(struct nfsd4_session *ses)
835 list_del(&ses->se_hash);
836 spin_lock(&ses->se_client->cl_lock);
837 list_del(&ses->se_perclnt);
838 spin_unlock(&ses->se_client->cl_lock);
841 /* must be called under the client_lock */
843 renew_client_locked(struct nfs4_client *clp)
845 if (is_client_expired(clp)) {
846 dprintk("%s: client (clientid %08x/%08x) already expired\n",
848 clp->cl_clientid.cl_boot,
849 clp->cl_clientid.cl_id);
854 * Move client to the end to the LRU list.
856 dprintk("renewing client (clientid %08x/%08x)\n",
857 clp->cl_clientid.cl_boot,
858 clp->cl_clientid.cl_id);
859 list_move_tail(&clp->cl_lru, &client_lru);
860 clp->cl_time = get_seconds();
864 renew_client(struct nfs4_client *clp)
866 spin_lock(&client_lock);
867 renew_client_locked(clp);
868 spin_unlock(&client_lock);
871 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
873 STALE_CLIENTID(clientid_t *clid)
875 if (clid->cl_boot == boot_time)
877 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
878 clid->cl_boot, clid->cl_id, boot_time);
883 * XXX Should we use a slab cache ?
884 * This type of memory management is somewhat inefficient, but we use it
885 * anyway since SETCLIENTID is not a common operation.
887 static struct nfs4_client *alloc_client(struct xdr_netobj name)
889 struct nfs4_client *clp;
891 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
894 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
895 if (clp->cl_name.data == NULL) {
899 memcpy(clp->cl_name.data, name.data, name.len);
900 clp->cl_name.len = name.len;
905 free_client(struct nfs4_client *clp)
907 while (!list_empty(&clp->cl_sessions)) {
908 struct nfsd4_session *ses;
909 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
911 list_del(&ses->se_perclnt);
912 nfsd4_put_session(ses);
914 if (clp->cl_cred.cr_group_info)
915 put_group_info(clp->cl_cred.cr_group_info);
916 kfree(clp->cl_principal);
917 kfree(clp->cl_name.data);
922 release_session_client(struct nfsd4_session *session)
924 struct nfs4_client *clp = session->se_client;
926 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
928 if (is_client_expired(clp)) {
930 session->se_client = NULL;
932 renew_client_locked(clp);
933 spin_unlock(&client_lock);
936 /* must be called under the client_lock */
938 unhash_client_locked(struct nfs4_client *clp)
940 struct nfsd4_session *ses;
942 mark_client_expired(clp);
943 list_del(&clp->cl_lru);
944 spin_lock(&clp->cl_lock);
945 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
946 list_del_init(&ses->se_hash);
947 spin_unlock(&clp->cl_lock);
951 expire_client(struct nfs4_client *clp)
953 struct nfs4_stateowner *sop;
954 struct nfs4_delegation *dp;
955 struct list_head reaplist;
957 INIT_LIST_HEAD(&reaplist);
958 spin_lock(&recall_lock);
959 while (!list_empty(&clp->cl_delegations)) {
960 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
961 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
963 list_del_init(&dp->dl_perclnt);
964 list_move(&dp->dl_recall_lru, &reaplist);
966 spin_unlock(&recall_lock);
967 while (!list_empty(&reaplist)) {
968 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
969 list_del_init(&dp->dl_recall_lru);
970 unhash_delegation(dp);
972 while (!list_empty(&clp->cl_openowners)) {
973 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
974 release_openowner(sop);
976 nfsd4_shutdown_callback(clp);
977 if (clp->cl_cb_conn.cb_xprt)
978 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
979 list_del(&clp->cl_idhash);
980 list_del(&clp->cl_strhash);
981 spin_lock(&client_lock);
982 unhash_client_locked(clp);
983 if (atomic_read(&clp->cl_refcount) == 0)
985 spin_unlock(&client_lock);
988 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
990 memcpy(target->cl_verifier.data, source->data,
991 sizeof(target->cl_verifier.data));
994 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
996 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
997 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1000 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1002 target->cr_uid = source->cr_uid;
1003 target->cr_gid = source->cr_gid;
1004 target->cr_group_info = source->cr_group_info;
1005 get_group_info(target->cr_group_info);
1008 static int same_name(const char *n1, const char *n2)
1010 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1014 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1016 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1020 same_clid(clientid_t *cl1, clientid_t *cl2)
1022 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1025 /* XXX what about NGROUP */
1027 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1029 return cr1->cr_uid == cr2->cr_uid;
1032 static void gen_clid(struct nfs4_client *clp)
1034 static u32 current_clientid = 1;
1036 clp->cl_clientid.cl_boot = boot_time;
1037 clp->cl_clientid.cl_id = current_clientid++;
1040 static void gen_confirm(struct nfs4_client *clp)
1045 p = (u32 *)clp->cl_confirm.data;
1046 *p++ = get_seconds();
1050 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1051 struct svc_rqst *rqstp, nfs4_verifier *verf)
1053 struct nfs4_client *clp;
1054 struct sockaddr *sa = svc_addr(rqstp);
1057 clp = alloc_client(name);
1061 INIT_LIST_HEAD(&clp->cl_sessions);
1063 princ = svc_gss_principal(rqstp);
1065 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1066 if (clp->cl_principal == NULL) {
1072 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1073 atomic_set(&clp->cl_refcount, 0);
1074 atomic_set(&clp->cl_cb_set, 0);
1075 INIT_LIST_HEAD(&clp->cl_idhash);
1076 INIT_LIST_HEAD(&clp->cl_strhash);
1077 INIT_LIST_HEAD(&clp->cl_openowners);
1078 INIT_LIST_HEAD(&clp->cl_delegations);
1079 INIT_LIST_HEAD(&clp->cl_lru);
1080 spin_lock_init(&clp->cl_lock);
1081 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1082 clp->cl_time = get_seconds();
1083 clear_bit(0, &clp->cl_cb_slot_busy);
1084 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1085 copy_verf(clp, verf);
1086 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1087 clp->cl_flavor = rqstp->rq_flavor;
1088 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1090 clp->cl_cb_session = NULL;
1094 static int check_name(struct xdr_netobj name)
1098 if (name.len > NFS4_OPAQUE_LIMIT) {
1099 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1106 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1108 unsigned int idhashval;
1110 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1111 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1112 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1117 move_to_confirmed(struct nfs4_client *clp)
1119 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1120 unsigned int strhashval;
1122 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1123 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1124 strhashval = clientstr_hashval(clp->cl_recdir);
1125 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1129 static struct nfs4_client *
1130 find_confirmed_client(clientid_t *clid)
1132 struct nfs4_client *clp;
1133 unsigned int idhashval = clientid_hashval(clid->cl_id);
1135 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1136 if (same_clid(&clp->cl_clientid, clid))
1142 static struct nfs4_client *
1143 find_unconfirmed_client(clientid_t *clid)
1145 struct nfs4_client *clp;
1146 unsigned int idhashval = clientid_hashval(clid->cl_id);
1148 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1149 if (same_clid(&clp->cl_clientid, clid))
1155 static bool clp_used_exchangeid(struct nfs4_client *clp)
1157 return clp->cl_exchange_flags != 0;
1160 static struct nfs4_client *
1161 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1163 struct nfs4_client *clp;
1165 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1166 if (same_name(clp->cl_recdir, dname))
1172 static struct nfs4_client *
1173 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1175 struct nfs4_client *clp;
1177 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1178 if (same_name(clp->cl_recdir, dname))
1184 static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1188 ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1189 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1192 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1193 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1199 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1201 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1202 struct sockaddr *sa = svc_addr(rqstp);
1203 u32 scopeid = rpc_get_scope_id(sa);
1204 unsigned short expected_family;
1206 /* Currently, we only support tcp and tcp6 for the callback channel */
1207 if (se->se_callback_netid_len == 3 &&
1208 !memcmp(se->se_callback_netid_val, "tcp", 3))
1209 expected_family = AF_INET;
1210 else if (se->se_callback_netid_len == 4 &&
1211 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1212 expected_family = AF_INET6;
1216 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1217 se->se_callback_addr_len,
1218 (struct sockaddr *)&conn->cb_addr,
1219 sizeof(conn->cb_addr));
1221 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1224 if (conn->cb_addr.ss_family == AF_INET6)
1225 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1227 conn->cb_prog = se->se_callback_prog;
1228 conn->cb_ident = se->se_callback_ident;
1229 rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1232 conn->cb_addr.ss_family = AF_UNSPEC;
1233 conn->cb_addrlen = 0;
1234 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1235 "will not receive delegations\n",
1236 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1242 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1245 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1247 struct nfsd4_slot *slot = resp->cstate.slot;
1250 dprintk("--> %s slot %p\n", __func__, slot);
1252 slot->sl_opcnt = resp->opcnt;
1253 slot->sl_status = resp->cstate.status;
1255 if (nfsd4_not_cached(resp)) {
1256 slot->sl_datalen = 0;
1259 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1260 base = (char *)resp->cstate.datap -
1261 (char *)resp->xbuf->head[0].iov_base;
1262 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1264 WARN("%s: sessions DRC could not cache compound\n", __func__);
1269 * Encode the replay sequence operation from the slot values.
1270 * If cachethis is FALSE encode the uncached rep error on the next
1271 * operation which sets resp->p and increments resp->opcnt for
1272 * nfs4svc_encode_compoundres.
1276 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1277 struct nfsd4_compoundres *resp)
1279 struct nfsd4_op *op;
1280 struct nfsd4_slot *slot = resp->cstate.slot;
1282 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1283 resp->opcnt, resp->cstate.slot->sl_cachethis);
1285 /* Encode the replayed sequence operation */
1286 op = &args->ops[resp->opcnt - 1];
1287 nfsd4_encode_operation(resp, op);
1289 /* Return nfserr_retry_uncached_rep in next operation. */
1290 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1291 op = &args->ops[resp->opcnt++];
1292 op->status = nfserr_retry_uncached_rep;
1293 nfsd4_encode_operation(resp, op);
1299 * The sequence operation is not cached because we can use the slot and
1303 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1304 struct nfsd4_sequence *seq)
1306 struct nfsd4_slot *slot = resp->cstate.slot;
1309 dprintk("--> %s slot %p\n", __func__, slot);
1311 /* Either returns 0 or nfserr_retry_uncached */
1312 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1313 if (status == nfserr_retry_uncached_rep)
1316 /* The sequence operation has been encoded, cstate->datap set. */
1317 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1319 resp->opcnt = slot->sl_opcnt;
1320 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1321 status = slot->sl_status;
1327 * Set the exchange_id flags returned by the server.
1330 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1332 /* pNFS is not supported */
1333 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1335 /* Referrals are supported, Migration is not. */
1336 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1338 /* set the wire flags to return to client. */
1339 clid->flags = new->cl_exchange_flags;
1343 nfsd4_exchange_id(struct svc_rqst *rqstp,
1344 struct nfsd4_compound_state *cstate,
1345 struct nfsd4_exchange_id *exid)
1347 struct nfs4_client *unconf, *conf, *new;
1349 unsigned int strhashval;
1350 char dname[HEXDIR_LEN];
1351 char addr_str[INET6_ADDRSTRLEN];
1352 nfs4_verifier verf = exid->verifier;
1353 struct sockaddr *sa = svc_addr(rqstp);
1355 rpc_ntop(sa, addr_str, sizeof(addr_str));
1356 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1357 "ip_addr=%s flags %x, spa_how %d\n",
1358 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1359 addr_str, exid->flags, exid->spa_how);
1361 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1362 return nfserr_inval;
1364 /* Currently only support SP4_NONE */
1365 switch (exid->spa_how) {
1369 return nfserr_serverfault;
1371 BUG(); /* checked by xdr code */
1373 return nfserr_serverfault; /* no excuse :-/ */
1376 status = nfs4_make_rec_clidname(dname, &exid->clname);
1381 strhashval = clientstr_hashval(dname);
1386 conf = find_confirmed_client_by_str(dname, strhashval);
1388 if (!clp_used_exchangeid(conf)) {
1389 status = nfserr_clid_inuse; /* XXX: ? */
1392 if (!same_verf(&verf, &conf->cl_verifier)) {
1393 /* 18.35.4 case 8 */
1394 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1395 status = nfserr_not_same;
1398 /* Client reboot: destroy old state */
1399 expire_client(conf);
1402 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1403 /* 18.35.4 case 9 */
1404 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1405 status = nfserr_perm;
1408 expire_client(conf);
1412 * Set bit when the owner id and verifier map to an already
1413 * confirmed client id (18.35.3).
1415 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1418 * Falling into 18.35.4 case 2, possible router replay.
1419 * Leave confirmed record intact and return same result.
1421 copy_verf(conf, &verf);
1426 /* 18.35.4 case 7 */
1427 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1428 status = nfserr_noent;
1432 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1435 * Possible retry or client restart. Per 18.35.4 case 4,
1436 * a new unconfirmed record should be generated regardless
1437 * of whether any properties have changed.
1439 expire_client(unconf);
1444 new = create_client(exid->clname, dname, rqstp, &verf);
1446 status = nfserr_jukebox;
1451 add_to_unconfirmed(new, strhashval);
1453 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1454 exid->clientid.cl_id = new->cl_clientid.cl_id;
1457 nfsd4_set_ex_flags(new, exid);
1459 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1460 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1464 nfs4_unlock_state();
1466 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1471 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1473 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1476 /* The slot is in use, and no response has been sent. */
1478 if (seqid == slot_seqid)
1479 return nfserr_jukebox;
1481 return nfserr_seq_misordered;
1484 if (likely(seqid == slot_seqid + 1))
1487 if (seqid == slot_seqid)
1488 return nfserr_replay_cache;
1490 if (seqid == 1 && (slot_seqid + 1) == 0)
1492 /* Misordered replay or misordered new request */
1493 return nfserr_seq_misordered;
1497 * Cache the create session result into the create session single DRC
1498 * slot cache by saving the xdr structure. sl_seqid has been set.
1499 * Do this for solo or embedded create session operations.
1502 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1503 struct nfsd4_clid_slot *slot, int nfserr)
1505 slot->sl_status = nfserr;
1506 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1510 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1511 struct nfsd4_clid_slot *slot)
1513 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1514 return slot->sl_status;
1518 nfsd4_create_session(struct svc_rqst *rqstp,
1519 struct nfsd4_compound_state *cstate,
1520 struct nfsd4_create_session *cr_ses)
1522 struct sockaddr *sa = svc_addr(rqstp);
1523 struct nfs4_client *conf, *unconf;
1524 struct nfsd4_session *new;
1525 struct nfsd4_clid_slot *cs_slot = NULL;
1526 bool confirm_me = false;
1530 unconf = find_unconfirmed_client(&cr_ses->clientid);
1531 conf = find_confirmed_client(&cr_ses->clientid);
1534 cs_slot = &conf->cl_cs_slot;
1535 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1536 if (status == nfserr_replay_cache) {
1537 dprintk("Got a create_session replay! seqid= %d\n",
1539 /* Return the cached reply status */
1540 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1542 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1543 status = nfserr_seq_misordered;
1544 dprintk("Sequence misordered!\n");
1545 dprintk("Expected seqid= %d but got seqid= %d\n",
1546 cs_slot->sl_seqid, cr_ses->seqid);
1549 } else if (unconf) {
1550 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1551 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1552 status = nfserr_clid_inuse;
1556 cs_slot = &unconf->cl_cs_slot;
1557 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1559 /* an unconfirmed replay returns misordered */
1560 status = nfserr_seq_misordered;
1567 status = nfserr_stale_clientid;
1572 * XXX: we should probably set this at creation time, and check
1573 * for consistent minorversion use throughout:
1575 conf->cl_minorversion = 1;
1577 * We do not support RDMA or persistent sessions
1579 cr_ses->flags &= ~SESSION4_PERSIST;
1580 cr_ses->flags &= ~SESSION4_RDMA;
1582 status = nfserr_jukebox;
1583 new = alloc_init_session(rqstp, conf, cr_ses);
1587 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1588 NFS4_MAX_SESSIONID_LEN);
1589 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1590 sizeof(struct nfsd4_channel_attrs));
1591 cs_slot->sl_seqid++;
1592 cr_ses->seqid = cs_slot->sl_seqid;
1594 /* cache solo and embedded create sessions under the state lock */
1595 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1597 move_to_confirmed(conf);
1599 nfs4_unlock_state();
1600 dprintk("%s returns %d\n", __func__, ntohl(status));
1604 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1606 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1607 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1609 return argp->opcnt == resp->opcnt;
1612 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1615 case NFS4_CDFC4_FORE:
1616 case NFS4_CDFC4_BACK:
1618 case NFS4_CDFC4_FORE_OR_BOTH:
1619 case NFS4_CDFC4_BACK_OR_BOTH:
1620 *dir = NFS4_CDFC4_BOTH;
1623 return nfserr_inval;
1626 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1627 struct nfsd4_compound_state *cstate,
1628 struct nfsd4_bind_conn_to_session *bcts)
1632 if (!nfsd4_last_compound_op(rqstp))
1633 return nfserr_not_only_op;
1634 spin_lock(&client_lock);
1635 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1636 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1637 * client_lock iself: */
1638 if (cstate->session) {
1639 nfsd4_get_session(cstate->session);
1640 atomic_inc(&cstate->session->se_client->cl_refcount);
1642 spin_unlock(&client_lock);
1643 if (!cstate->session)
1644 return nfserr_badsession;
1646 status = nfsd4_map_bcts_dir(&bcts->dir);
1647 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1651 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1655 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1659 nfsd4_destroy_session(struct svc_rqst *r,
1660 struct nfsd4_compound_state *cstate,
1661 struct nfsd4_destroy_session *sessionid)
1663 struct nfsd4_session *ses;
1664 u32 status = nfserr_badsession;
1667 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1668 * - Should we return nfserr_back_chan_busy if waiting for
1669 * callbacks on to-be-destroyed session?
1670 * - Do we need to clear any callback info from previous session?
1673 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1674 if (!nfsd4_last_compound_op(r))
1675 return nfserr_not_only_op;
1677 dump_sessionid(__func__, &sessionid->sessionid);
1678 spin_lock(&client_lock);
1679 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1681 spin_unlock(&client_lock);
1685 unhash_session(ses);
1686 spin_unlock(&client_lock);
1689 /* wait for callbacks */
1690 nfsd4_shutdown_callback(ses->se_client);
1691 nfs4_unlock_state();
1693 nfsd4_del_conns(ses);
1695 nfsd4_put_session(ses);
1698 dprintk("%s returns %d\n", __func__, ntohl(status));
1702 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1704 struct nfsd4_conn *c;
1706 list_for_each_entry(c, &s->se_conns, cn_persession) {
1707 if (c->cn_xprt == xpt) {
1714 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1716 struct nfs4_client *clp = ses->se_client;
1717 struct nfsd4_conn *c;
1720 spin_lock(&clp->cl_lock);
1721 c = __nfsd4_find_conn(new->cn_xprt, ses);
1723 spin_unlock(&clp->cl_lock);
1727 __nfsd4_hash_conn(new, ses);
1728 spin_unlock(&clp->cl_lock);
1729 ret = nfsd4_register_conn(new);
1731 /* oops; xprt is already down: */
1732 nfsd4_conn_lost(&new->cn_xpt_user);
1737 nfsd4_sequence(struct svc_rqst *rqstp,
1738 struct nfsd4_compound_state *cstate,
1739 struct nfsd4_sequence *seq)
1741 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1742 struct nfsd4_session *session;
1743 struct nfsd4_slot *slot;
1744 struct nfsd4_conn *conn;
1747 if (resp->opcnt != 1)
1748 return nfserr_sequence_pos;
1751 * Will be either used or freed by nfsd4_sequence_check_conn
1754 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1756 return nfserr_jukebox;
1758 spin_lock(&client_lock);
1759 status = nfserr_badsession;
1760 session = find_in_sessionid_hashtbl(&seq->sessionid);
1764 status = nfserr_badslot;
1765 if (seq->slotid >= session->se_fchannel.maxreqs)
1768 slot = session->se_slots[seq->slotid];
1769 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1771 /* We do not negotiate the number of slots yet, so set the
1772 * maxslots to the session maxreqs which is used to encode
1773 * sr_highest_slotid and the sr_target_slot id to maxslots */
1774 seq->maxslots = session->se_fchannel.maxreqs;
1776 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1777 if (status == nfserr_replay_cache) {
1778 cstate->slot = slot;
1779 cstate->session = session;
1780 /* Return the cached reply status and set cstate->status
1781 * for nfsd4_proc_compound processing */
1782 status = nfsd4_replay_cache_entry(resp, seq);
1783 cstate->status = nfserr_replay_cache;
1789 nfsd4_sequence_check_conn(conn, session);
1792 /* Success! bump slot seqid */
1793 slot->sl_inuse = true;
1794 slot->sl_seqid = seq->seqid;
1795 slot->sl_cachethis = seq->cachethis;
1797 cstate->slot = slot;
1798 cstate->session = session;
1801 /* Hold a session reference until done processing the compound. */
1802 if (cstate->session) {
1803 nfsd4_get_session(cstate->session);
1804 atomic_inc(&session->se_client->cl_refcount);
1807 spin_unlock(&client_lock);
1808 dprintk("%s: return %d\n", __func__, ntohl(status));
1813 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1815 if (rc->rca_one_fs) {
1816 if (!cstate->current_fh.fh_dentry)
1817 return nfserr_nofilehandle;
1819 * We don't take advantage of the rca_one_fs case.
1820 * That's OK, it's optional, we can safely ignore it.
1825 if (is_client_expired(cstate->session->se_client)) {
1826 nfs4_unlock_state();
1828 * The following error isn't really legal.
1829 * But we only get here if the client just explicitly
1830 * destroyed the client. Surely it no longer cares what
1831 * error it gets back on an operation for the dead
1834 return nfserr_stale_clientid;
1836 nfsd4_create_clid_dir(cstate->session->se_client);
1837 nfs4_unlock_state();
1842 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1843 struct nfsd4_setclientid *setclid)
1845 struct xdr_netobj clname = {
1846 .len = setclid->se_namelen,
1847 .data = setclid->se_name,
1849 nfs4_verifier clverifier = setclid->se_verf;
1850 unsigned int strhashval;
1851 struct nfs4_client *conf, *unconf, *new;
1853 char dname[HEXDIR_LEN];
1855 if (!check_name(clname))
1856 return nfserr_inval;
1858 status = nfs4_make_rec_clidname(dname, &clname);
1863 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1864 * We get here on a DRC miss.
1867 strhashval = clientstr_hashval(dname);
1870 conf = find_confirmed_client_by_str(dname, strhashval);
1872 /* RFC 3530 14.2.33 CASE 0: */
1873 status = nfserr_clid_inuse;
1874 if (clp_used_exchangeid(conf))
1876 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1877 char addr_str[INET6_ADDRSTRLEN];
1878 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1880 dprintk("NFSD: setclientid: string in use by client "
1881 "at %s\n", addr_str);
1886 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1887 * has a description of SETCLIENTID request processing consisting
1888 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1890 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1891 status = nfserr_resource;
1894 * RFC 3530 14.2.33 CASE 4:
1895 * placed first, because it is the normal case
1898 expire_client(unconf);
1899 new = create_client(clname, dname, rqstp, &clverifier);
1903 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1905 * RFC 3530 14.2.33 CASE 1:
1906 * probable callback update
1909 /* Note this is removing unconfirmed {*x***},
1910 * which is stronger than RFC recommended {vxc**}.
1911 * This has the advantage that there is at most
1912 * one {*x***} in either list at any time.
1914 expire_client(unconf);
1916 new = create_client(clname, dname, rqstp, &clverifier);
1919 copy_clid(new, conf);
1920 } else if (!unconf) {
1922 * RFC 3530 14.2.33 CASE 2:
1923 * probable client reboot; state will be removed if
1926 new = create_client(clname, dname, rqstp, &clverifier);
1932 * RFC 3530 14.2.33 CASE 3:
1933 * probable client reboot; state will be removed if
1936 expire_client(unconf);
1937 new = create_client(clname, dname, rqstp, &clverifier);
1943 * XXX: we should probably set this at creation time, and check
1944 * for consistent minorversion use throughout:
1946 new->cl_minorversion = 0;
1947 gen_callback(new, setclid, rqstp);
1948 add_to_unconfirmed(new, strhashval);
1949 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1950 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1951 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1954 nfs4_unlock_state();
1960 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1961 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1962 * bullets, labeled as CASE1 - CASE4 below.
1965 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1966 struct nfsd4_compound_state *cstate,
1967 struct nfsd4_setclientid_confirm *setclientid_confirm)
1969 struct sockaddr *sa = svc_addr(rqstp);
1970 struct nfs4_client *conf, *unconf;
1971 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1972 clientid_t * clid = &setclientid_confirm->sc_clientid;
1975 if (STALE_CLIENTID(clid))
1976 return nfserr_stale_clientid;
1978 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1979 * We get here on a DRC miss.
1984 conf = find_confirmed_client(clid);
1985 unconf = find_unconfirmed_client(clid);
1987 status = nfserr_clid_inuse;
1988 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1990 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1994 * section 14.2.34 of RFC 3530 has a description of
1995 * SETCLIENTID_CONFIRM request processing consisting
1996 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1998 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2000 * RFC 3530 14.2.34 CASE 1:
2003 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2004 status = nfserr_clid_inuse;
2006 atomic_set(&conf->cl_cb_set, 0);
2007 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2008 nfsd4_probe_callback(conf);
2009 expire_client(unconf);
2013 } else if (conf && !unconf) {
2015 * RFC 3530 14.2.34 CASE 2:
2016 * probable retransmitted request; play it safe and
2019 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2020 status = nfserr_clid_inuse;
2023 } else if (!conf && unconf
2024 && same_verf(&unconf->cl_confirm, &confirm)) {
2026 * RFC 3530 14.2.34 CASE 3:
2027 * Normal case; new or rebooted client:
2029 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2030 status = nfserr_clid_inuse;
2033 clientstr_hashval(unconf->cl_recdir);
2034 conf = find_confirmed_client_by_str(unconf->cl_recdir,
2037 nfsd4_remove_clid_dir(conf);
2038 expire_client(conf);
2040 move_to_confirmed(unconf);
2042 nfsd4_probe_callback(conf);
2045 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2046 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2049 * RFC 3530 14.2.34 CASE 4:
2050 * Client probably hasn't noticed that we rebooted yet.
2052 status = nfserr_stale_clientid;
2054 /* check that we have hit one of the cases...*/
2055 status = nfserr_clid_inuse;
2058 nfs4_unlock_state();
2062 /* OPEN Share state helper functions */
2063 static inline struct nfs4_file *
2064 alloc_init_file(struct inode *ino)
2066 struct nfs4_file *fp;
2067 unsigned int hashval = file_hashval(ino);
2069 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2071 atomic_set(&fp->fi_ref, 1);
2072 INIT_LIST_HEAD(&fp->fi_hash);
2073 INIT_LIST_HEAD(&fp->fi_stateids);
2074 INIT_LIST_HEAD(&fp->fi_delegations);
2075 fp->fi_inode = igrab(ino);
2076 fp->fi_id = current_fileid++;
2077 fp->fi_had_conflict = false;
2078 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2079 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2080 spin_lock(&recall_lock);
2081 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2082 spin_unlock(&recall_lock);
2089 nfsd4_free_slab(struct kmem_cache **slab)
2093 kmem_cache_destroy(*slab);
2098 nfsd4_free_slabs(void)
2100 nfsd4_free_slab(&stateowner_slab);
2101 nfsd4_free_slab(&file_slab);
2102 nfsd4_free_slab(&stateid_slab);
2103 nfsd4_free_slab(&deleg_slab);
2107 nfsd4_init_slabs(void)
2109 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2110 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2111 if (stateowner_slab == NULL)
2113 file_slab = kmem_cache_create("nfsd4_files",
2114 sizeof(struct nfs4_file), 0, 0, NULL);
2115 if (file_slab == NULL)
2117 stateid_slab = kmem_cache_create("nfsd4_stateids",
2118 sizeof(struct nfs4_stateid), 0, 0, NULL);
2119 if (stateid_slab == NULL)
2121 deleg_slab = kmem_cache_create("nfsd4_delegations",
2122 sizeof(struct nfs4_delegation), 0, 0, NULL);
2123 if (deleg_slab == NULL)
2128 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2133 nfs4_free_stateowner(struct kref *kref)
2135 struct nfs4_stateowner *sop =
2136 container_of(kref, struct nfs4_stateowner, so_ref);
2137 kfree(sop->so_owner.data);
2138 kmem_cache_free(stateowner_slab, sop);
2141 static inline struct nfs4_stateowner *
2142 alloc_stateowner(struct xdr_netobj *owner)
2144 struct nfs4_stateowner *sop;
2146 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2147 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2148 memcpy(sop->so_owner.data, owner->data, owner->len);
2149 sop->so_owner.len = owner->len;
2150 kref_init(&sop->so_ref);
2153 kmem_cache_free(stateowner_slab, sop);
2158 static struct nfs4_stateowner *
2159 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2160 struct nfs4_stateowner *sop;
2161 struct nfs4_replay *rp;
2162 unsigned int idhashval;
2164 if (!(sop = alloc_stateowner(&open->op_owner)))
2166 idhashval = ownerid_hashval(current_ownerid);
2167 INIT_LIST_HEAD(&sop->so_idhash);
2168 INIT_LIST_HEAD(&sop->so_strhash);
2169 INIT_LIST_HEAD(&sop->so_perclient);
2170 INIT_LIST_HEAD(&sop->so_stateids);
2171 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2172 INIT_LIST_HEAD(&sop->so_close_lru);
2174 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2175 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2176 list_add(&sop->so_perclient, &clp->cl_openowners);
2177 sop->so_is_open_owner = 1;
2178 sop->so_id = current_ownerid++;
2179 sop->so_client = clp;
2180 sop->so_seqid = open->op_seqid;
2181 sop->so_confirmed = 0;
2182 rp = &sop->so_replay;
2183 rp->rp_status = nfserr_serverfault;
2185 rp->rp_buf = rp->rp_ibuf;
2190 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2191 struct nfs4_stateowner *sop = open->op_stateowner;
2192 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2194 INIT_LIST_HEAD(&stp->st_hash);
2195 INIT_LIST_HEAD(&stp->st_perstateowner);
2196 INIT_LIST_HEAD(&stp->st_lockowners);
2197 INIT_LIST_HEAD(&stp->st_perfile);
2198 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2199 list_add(&stp->st_perstateowner, &sop->so_stateids);
2200 list_add(&stp->st_perfile, &fp->fi_stateids);
2201 stp->st_stateowner = sop;
2204 stp->st_stateid.si_boot = boot_time;
2205 stp->st_stateid.si_stateownerid = sop->so_id;
2206 stp->st_stateid.si_fileid = fp->fi_id;
2207 stp->st_stateid.si_generation = 0;
2208 stp->st_access_bmap = 0;
2209 stp->st_deny_bmap = 0;
2210 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2211 &stp->st_access_bmap);
2212 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2213 stp->st_openstp = NULL;
2217 move_to_close_lru(struct nfs4_stateowner *sop)
2219 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2221 list_move_tail(&sop->so_close_lru, &close_lru);
2222 sop->so_time = get_seconds();
2226 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2229 return (sop->so_owner.len == owner->len) &&
2230 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2231 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2234 static struct nfs4_stateowner *
2235 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2237 struct nfs4_stateowner *so = NULL;
2239 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2240 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2246 /* search file_hashtbl[] for file */
2247 static struct nfs4_file *
2248 find_file(struct inode *ino)
2250 unsigned int hashval = file_hashval(ino);
2251 struct nfs4_file *fp;
2253 spin_lock(&recall_lock);
2254 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2255 if (fp->fi_inode == ino) {
2257 spin_unlock(&recall_lock);
2261 spin_unlock(&recall_lock);
2265 static inline int access_valid(u32 x, u32 minorversion)
2267 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2269 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2271 x &= ~NFS4_SHARE_ACCESS_MASK;
2272 if (minorversion && x) {
2273 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2275 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2277 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2284 static inline int deny_valid(u32 x)
2286 /* Note: unlike access bits, deny bits may be zero. */
2287 return x <= NFS4_SHARE_DENY_BOTH;
2291 * Called to check deny when READ with all zero stateid or
2292 * WRITE with all zero or all one stateid
2295 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2297 struct inode *ino = current_fh->fh_dentry->d_inode;
2298 struct nfs4_file *fp;
2299 struct nfs4_stateid *stp;
2302 dprintk("NFSD: nfs4_share_conflict\n");
2304 fp = find_file(ino);
2307 ret = nfserr_locked;
2308 /* Search for conflicting share reservations */
2309 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2310 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2311 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2321 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2323 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2324 nfs4_file_put_access(fp, O_WRONLY);
2325 if (share_access & NFS4_SHARE_ACCESS_READ)
2326 nfs4_file_put_access(fp, O_RDONLY);
2330 * Spawn a thread to perform a recall on the delegation represented
2331 * by the lease (file_lock)
2333 * Called from break_lease() with lock_flocks() held.
2334 * Note: we assume break_lease will only call this *once* for any given
2338 void nfsd_break_deleg_cb(struct file_lock *fl)
2340 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2342 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2346 /* We're assuming the state code never drops its reference
2347 * without first removing the lease. Since we're in this lease
2348 * callback (and since the lease code is serialized by the kernel
2349 * lock) we know the server hasn't removed the lease yet, we know
2350 * it's safe to take a reference: */
2351 atomic_inc(&dp->dl_count);
2353 spin_lock(&recall_lock);
2354 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2355 spin_unlock(&recall_lock);
2357 /* only place dl_time is set. protected by lock_flocks*/
2358 dp->dl_time = get_seconds();
2361 * We don't want the locks code to timeout the lease for us;
2362 * we'll remove it ourself if the delegation isn't returned
2365 fl->fl_break_time = 0;
2367 dp->dl_file->fi_had_conflict = true;
2368 nfsd4_cb_recall(dp);
2372 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2375 return lease_modify(onlist, arg);
2380 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2381 .fl_break = nfsd_break_deleg_cb,
2382 .fl_change = nfsd_change_deleg_cb,
2387 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2388 struct nfsd4_open *open)
2390 clientid_t *clientid = &open->op_clientid;
2391 struct nfs4_client *clp = NULL;
2392 unsigned int strhashval;
2393 struct nfs4_stateowner *sop = NULL;
2395 if (!check_name(open->op_owner))
2396 return nfserr_inval;
2398 if (STALE_CLIENTID(&open->op_clientid))
2399 return nfserr_stale_clientid;
2401 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2402 sop = find_openstateowner_str(strhashval, open);
2403 open->op_stateowner = sop;
2405 /* Make sure the client's lease hasn't expired. */
2406 clp = find_confirmed_client(clientid);
2408 return nfserr_expired;
2411 /* When sessions are used, skip open sequenceid processing */
2412 if (nfsd4_has_session(cstate))
2414 if (!sop->so_confirmed) {
2415 /* Replace unconfirmed owners without checking for replay. */
2416 clp = sop->so_client;
2417 release_openowner(sop);
2418 open->op_stateowner = NULL;
2421 if (open->op_seqid == sop->so_seqid - 1) {
2422 if (sop->so_replay.rp_buflen)
2423 return nfserr_replay_me;
2424 /* The original OPEN failed so spectacularly
2425 * that we don't even have replay data saved!
2426 * Therefore, we have no choice but to continue
2427 * processing this OPEN; presumably, we'll
2428 * fail again for the same reason.
2430 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2433 if (open->op_seqid != sop->so_seqid)
2434 return nfserr_bad_seqid;
2436 if (open->op_stateowner == NULL) {
2437 sop = alloc_init_open_stateowner(strhashval, clp, open);
2439 return nfserr_resource;
2440 open->op_stateowner = sop;
2442 list_del_init(&sop->so_close_lru);
2443 renew_client(sop->so_client);
2447 static inline __be32
2448 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2450 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2451 return nfserr_openmode;
2456 static struct nfs4_delegation *
2457 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2459 struct nfs4_delegation *dp;
2461 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2462 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2468 int share_access_to_flags(u32 share_access)
2470 share_access &= ~NFS4_SHARE_WANT_MASK;
2472 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2476 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2477 struct nfs4_delegation **dp)
2480 __be32 status = nfserr_bad_stateid;
2482 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2485 flags = share_access_to_flags(open->op_share_access);
2486 status = nfs4_check_delegmode(*dp, flags);
2490 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2494 open->op_stateowner->so_confirmed = 1;
2499 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2501 struct nfs4_stateid *local;
2502 __be32 status = nfserr_share_denied;
2503 struct nfs4_stateowner *sop = open->op_stateowner;
2505 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2506 /* ignore lock owners */
2507 if (local->st_stateowner->so_is_open_owner == 0)
2509 /* remember if we have seen this open owner */
2510 if (local->st_stateowner == sop)
2512 /* check for conflicting share reservations */
2513 if (!test_share(local, open))
2521 static inline struct nfs4_stateid *
2522 nfs4_alloc_stateid(void)
2524 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2527 static inline int nfs4_access_to_access(u32 nfs4_access)
2531 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2532 flags |= NFSD_MAY_READ;
2533 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2534 flags |= NFSD_MAY_WRITE;
2538 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2539 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2542 int oflag = nfs4_access_to_omode(nfs4_access);
2543 int access = nfs4_access_to_access(nfs4_access);
2545 if (!fp->fi_fds[oflag]) {
2546 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2547 &fp->fi_fds[oflag]);
2551 nfs4_file_get_access(fp, oflag);
2557 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2558 struct nfs4_file *fp, struct svc_fh *cur_fh,
2559 struct nfsd4_open *open)
2561 struct nfs4_stateid *stp;
2564 stp = nfs4_alloc_stateid();
2566 return nfserr_resource;
2568 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2570 kmem_cache_free(stateid_slab, stp);
2577 static inline __be32
2578 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2579 struct nfsd4_open *open)
2581 struct iattr iattr = {
2582 .ia_valid = ATTR_SIZE,
2585 if (!open->op_truncate)
2587 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2588 return nfserr_inval;
2589 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2593 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2595 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2599 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2601 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2605 status = nfsd4_truncate(rqstp, cur_fh, open);
2608 int oflag = nfs4_access_to_omode(new_access);
2609 nfs4_file_put_access(fp, oflag);
2613 /* remember the open */
2614 __set_bit(op_share_access, &stp->st_access_bmap);
2615 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2622 nfs4_set_claim_prev(struct nfsd4_open *open)
2624 open->op_stateowner->so_confirmed = 1;
2625 open->op_stateowner->so_client->cl_firststate = 1;
2629 * Attempt to hand out a delegation.
2632 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2634 struct nfs4_delegation *dp;
2635 struct nfs4_stateowner *sop = stp->st_stateowner;
2636 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2637 struct file_lock *fl;
2638 int status, flag = 0;
2640 flag = NFS4_OPEN_DELEGATE_NONE;
2641 open->op_recall = 0;
2642 switch (open->op_claim_type) {
2643 case NFS4_OPEN_CLAIM_PREVIOUS:
2645 open->op_recall = 1;
2646 flag = open->op_delegate_type;
2647 if (flag == NFS4_OPEN_DELEGATE_NONE)
2650 case NFS4_OPEN_CLAIM_NULL:
2651 /* Let's not give out any delegations till everyone's
2652 * had the chance to reclaim theirs.... */
2653 if (locks_in_grace())
2655 if (!cb_up || !sop->so_confirmed)
2657 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2658 flag = NFS4_OPEN_DELEGATE_WRITE;
2660 flag = NFS4_OPEN_DELEGATE_READ;
2666 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2668 flag = NFS4_OPEN_DELEGATE_NONE;
2672 fl = locks_alloc_lock();
2675 locks_init_lock(fl);
2676 fl->fl_lmops = &nfsd_lease_mng_ops;
2677 fl->fl_flags = FL_LEASE;
2678 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2679 fl->fl_end = OFFSET_MAX;
2680 fl->fl_owner = (fl_owner_t)dp;
2681 fl->fl_file = find_readable_file(stp->st_file);
2682 BUG_ON(!fl->fl_file);
2683 fl->fl_pid = current->tgid;
2686 /* vfs_setlease checks to see if delegation should be handed out.
2687 * the lock_manager callback fl_change is used
2689 if ((status = vfs_setlease(fl->fl_file, fl->fl_type, &fl))) {
2690 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2691 dp->dl_flock = NULL;
2692 locks_free_lock(fl);
2693 unhash_delegation(dp);
2694 flag = NFS4_OPEN_DELEGATE_NONE;
2698 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2700 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2701 STATEID_VAL(&dp->dl_stateid));
2703 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2704 && flag == NFS4_OPEN_DELEGATE_NONE
2705 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2706 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2707 open->op_delegate_type = flag;
2711 * called with nfs4_lock_state() held.
2714 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2716 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2717 struct nfs4_file *fp = NULL;
2718 struct inode *ino = current_fh->fh_dentry->d_inode;
2719 struct nfs4_stateid *stp = NULL;
2720 struct nfs4_delegation *dp = NULL;
2723 status = nfserr_inval;
2724 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2725 || !deny_valid(open->op_share_deny))
2728 * Lookup file; if found, lookup stateid and check open request,
2729 * and check for delegations in the process of being recalled.
2730 * If not found, create the nfs4_file struct
2732 fp = find_file(ino);
2734 if ((status = nfs4_check_open(fp, open, &stp)))
2736 status = nfs4_check_deleg(fp, open, &dp);
2740 status = nfserr_bad_stateid;
2741 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2743 status = nfserr_resource;
2744 fp = alloc_init_file(ino);
2750 * OPEN the file, or upgrade an existing OPEN.
2751 * If truncate fails, the OPEN fails.
2754 /* Stateid was found, this is an OPEN upgrade */
2755 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2758 update_stateid(&stp->st_stateid);
2760 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2763 init_stateid(stp, fp, open);
2764 status = nfsd4_truncate(rqstp, current_fh, open);
2766 release_open_stateid(stp);
2769 if (nfsd4_has_session(&resp->cstate))
2770 update_stateid(&stp->st_stateid);
2772 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2774 if (nfsd4_has_session(&resp->cstate))
2775 open->op_stateowner->so_confirmed = 1;
2778 * Attempt to hand out a delegation. No error return, because the
2779 * OPEN succeeds even if we fail.
2781 nfs4_open_delegation(current_fh, open, stp);
2785 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2786 STATEID_VAL(&stp->st_stateid));
2790 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2791 nfs4_set_claim_prev(open);
2793 * To finish the open response, we just need to set the rflags.
2795 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2796 if (!open->op_stateowner->so_confirmed &&
2797 !nfsd4_has_session(&resp->cstate))
2798 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2804 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2807 struct nfs4_client *clp;
2811 dprintk("process_renew(%08x/%08x): starting\n",
2812 clid->cl_boot, clid->cl_id);
2813 status = nfserr_stale_clientid;
2814 if (STALE_CLIENTID(clid))
2816 clp = find_confirmed_client(clid);
2817 status = nfserr_expired;
2819 /* We assume the client took too long to RENEW. */
2820 dprintk("nfsd4_renew: clientid not found!\n");
2824 status = nfserr_cb_path_down;
2825 if (!list_empty(&clp->cl_delegations)
2826 && !atomic_read(&clp->cl_cb_set))
2830 nfs4_unlock_state();
2834 struct lock_manager nfsd4_manager = {
2838 nfsd4_end_grace(void)
2840 dprintk("NFSD: end of grace period\n");
2841 nfsd4_recdir_purge_old();
2842 locks_end_grace(&nfsd4_manager);
2844 * Now that every NFSv4 client has had the chance to recover and
2845 * to see the (possibly new, possibly shorter) lease time, we
2846 * can safely set the next grace time to the current lease time:
2848 nfsd4_grace = nfsd4_lease;
2852 nfs4_laundromat(void)
2854 struct nfs4_client *clp;
2855 struct nfs4_stateowner *sop;
2856 struct nfs4_delegation *dp;
2857 struct list_head *pos, *next, reaplist;
2858 time_t cutoff = get_seconds() - nfsd4_lease;
2859 time_t t, clientid_val = nfsd4_lease;
2860 time_t u, test_val = nfsd4_lease;
2864 dprintk("NFSD: laundromat service - starting\n");
2865 if (locks_in_grace())
2867 INIT_LIST_HEAD(&reaplist);
2868 spin_lock(&client_lock);
2869 list_for_each_safe(pos, next, &client_lru) {
2870 clp = list_entry(pos, struct nfs4_client, cl_lru);
2871 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2872 t = clp->cl_time - cutoff;
2873 if (clientid_val > t)
2877 if (atomic_read(&clp->cl_refcount)) {
2878 dprintk("NFSD: client in use (clientid %08x)\n",
2879 clp->cl_clientid.cl_id);
2882 unhash_client_locked(clp);
2883 list_add(&clp->cl_lru, &reaplist);
2885 spin_unlock(&client_lock);
2886 list_for_each_safe(pos, next, &reaplist) {
2887 clp = list_entry(pos, struct nfs4_client, cl_lru);
2888 dprintk("NFSD: purging unused client (clientid %08x)\n",
2889 clp->cl_clientid.cl_id);
2890 nfsd4_remove_clid_dir(clp);
2893 spin_lock(&recall_lock);
2894 list_for_each_safe(pos, next, &del_recall_lru) {
2895 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2896 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2897 u = dp->dl_time - cutoff;
2902 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2904 list_move(&dp->dl_recall_lru, &reaplist);
2906 spin_unlock(&recall_lock);
2907 list_for_each_safe(pos, next, &reaplist) {
2908 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2909 list_del_init(&dp->dl_recall_lru);
2910 unhash_delegation(dp);
2912 test_val = nfsd4_lease;
2913 list_for_each_safe(pos, next, &close_lru) {
2914 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2915 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2916 u = sop->so_time - cutoff;
2921 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2923 release_openowner(sop);
2925 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2926 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2927 nfs4_unlock_state();
2928 return clientid_val;
2931 static struct workqueue_struct *laundry_wq;
2932 static void laundromat_main(struct work_struct *);
2933 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2936 laundromat_main(struct work_struct *not_used)
2940 t = nfs4_laundromat();
2941 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2942 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2945 static struct nfs4_stateowner *
2946 search_close_lru(u32 st_id, int flags)
2948 struct nfs4_stateowner *local = NULL;
2950 if (flags & CLOSE_STATE) {
2951 list_for_each_entry(local, &close_lru, so_close_lru) {
2952 if (local->so_id == st_id)
2960 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2962 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2966 STALE_STATEID(stateid_t *stateid)
2968 if (stateid->si_boot == boot_time)
2970 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2971 STATEID_VAL(stateid));
2976 access_permit_read(unsigned long access_bmap)
2978 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2979 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2980 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2984 access_permit_write(unsigned long access_bmap)
2986 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2987 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2991 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2993 __be32 status = nfserr_openmode;
2995 /* For lock stateid's, we test the parent open, not the lock: */
2996 if (stp->st_openstp)
2997 stp = stp->st_openstp;
2998 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3000 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3007 static inline __be32
3008 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3010 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3012 else if (locks_in_grace()) {
3013 /* Answer in remaining cases depends on existance of
3014 * conflicting state; so we must wait out the grace period. */
3015 return nfserr_grace;
3016 } else if (flags & WR_STATE)
3017 return nfs4_share_conflict(current_fh,
3018 NFS4_SHARE_DENY_WRITE);
3019 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3020 return nfs4_share_conflict(current_fh,
3021 NFS4_SHARE_DENY_READ);
3025 * Allow READ/WRITE during grace period on recovered state only for files
3026 * that are not able to provide mandatory locking.
3029 grace_disallows_io(struct inode *inode)
3031 return locks_in_grace() && mandatory_lock(inode);
3034 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3037 * When sessions are used the stateid generation number is ignored
3040 if ((flags & HAS_SESSION) && in->si_generation == 0)
3043 /* If the client sends us a stateid from the future, it's buggy: */
3044 if (in->si_generation > ref->si_generation)
3045 return nfserr_bad_stateid;
3047 * The following, however, can happen. For example, if the
3048 * client sends an open and some IO at the same time, the open
3049 * may bump si_generation while the IO is still in flight.
3050 * Thanks to hard links and renames, the client never knows what
3051 * file an open will affect. So it could avoid that situation
3052 * only by serializing all opens and IO from the same open
3053 * owner. To recover from the old_stateid error, the client
3054 * will just have to retry the IO:
3056 if (in->si_generation < ref->si_generation)
3057 return nfserr_old_stateid;
3062 static int is_delegation_stateid(stateid_t *stateid)
3064 return stateid->si_fileid == 0;
3068 * Checks for stateid operations
3071 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3072 stateid_t *stateid, int flags, struct file **filpp)
3074 struct nfs4_stateid *stp = NULL;
3075 struct nfs4_delegation *dp = NULL;
3076 struct svc_fh *current_fh = &cstate->current_fh;
3077 struct inode *ino = current_fh->fh_dentry->d_inode;
3083 if (grace_disallows_io(ino))
3084 return nfserr_grace;
3086 if (nfsd4_has_session(cstate))
3087 flags |= HAS_SESSION;
3089 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3090 return check_special_stateids(current_fh, stateid, flags);
3092 status = nfserr_stale_stateid;
3093 if (STALE_STATEID(stateid))
3097 * We assume that any stateid that has the current boot time,
3098 * but that we can't find, is expired:
3100 status = nfserr_expired;
3101 if (is_delegation_stateid(stateid)) {
3102 dp = find_delegation_stateid(ino, stateid);
3105 status = check_stateid_generation(stateid, &dp->dl_stateid,
3109 status = nfs4_check_delegmode(dp, flags);
3112 renew_client(dp->dl_client);
3114 *filpp = find_readable_file(dp->dl_file);
3117 } else { /* open or lock stateid */
3118 stp = find_stateid(stateid, flags);
3121 status = nfserr_bad_stateid;
3122 if (nfs4_check_fh(current_fh, stp))
3124 if (!stp->st_stateowner->so_confirmed)
3126 status = check_stateid_generation(stateid, &stp->st_stateid,
3130 status = nfs4_check_openmode(stp, flags);
3133 renew_client(stp->st_stateowner->so_client);
3135 if (flags & RD_STATE)
3136 *filpp = find_readable_file(stp->st_file);
3138 *filpp = find_writeable_file(stp->st_file);
3149 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3150 RD_STATE : WR_STATE;
3154 * Checks for sequence id mutating operations.
3157 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3158 stateid_t *stateid, int flags,
3159 struct nfs4_stateowner **sopp,
3160 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3162 struct nfs4_stateid *stp;
3163 struct nfs4_stateowner *sop;
3164 struct svc_fh *current_fh = &cstate->current_fh;
3167 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3168 seqid, STATEID_VAL(stateid));
3173 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3174 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3175 return nfserr_bad_stateid;
3178 if (STALE_STATEID(stateid))
3179 return nfserr_stale_stateid;
3181 if (nfsd4_has_session(cstate))
3182 flags |= HAS_SESSION;
3185 * We return BAD_STATEID if filehandle doesn't match stateid,
3186 * the confirmed flag is incorrecly set, or the generation
3187 * number is incorrect.
3189 stp = find_stateid(stateid, flags);
3192 * Also, we should make sure this isn't just the result of
3195 sop = search_close_lru(stateid->si_stateownerid, flags);
3196 /* It's not stale; let's assume it's expired: */
3198 return nfserr_expired;
3204 *sopp = sop = stp->st_stateowner;
3207 clientid_t *lockclid = &lock->v.new.clientid;
3208 struct nfs4_client *clp = sop->so_client;
3212 lkflg = setlkflg(lock->lk_type);
3214 if (lock->lk_is_new) {
3215 if (!sop->so_is_open_owner)
3216 return nfserr_bad_stateid;
3217 if (!(flags & HAS_SESSION) &&
3218 !same_clid(&clp->cl_clientid, lockclid))
3219 return nfserr_bad_stateid;
3220 /* stp is the open stateid */
3221 status = nfs4_check_openmode(stp, lkflg);
3225 /* stp is the lock stateid */
3226 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3232 if (nfs4_check_fh(current_fh, stp)) {
3233 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3234 return nfserr_bad_stateid;
3238 * We now validate the seqid and stateid generation numbers.
3239 * For the moment, we ignore the possibility of
3240 * generation number wraparound.
3242 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3245 if (sop->so_confirmed && flags & CONFIRM) {
3246 dprintk("NFSD: preprocess_seqid_op: expected"
3247 " unconfirmed stateowner!\n");
3248 return nfserr_bad_stateid;
3250 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3251 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3252 " confirmed yet!\n");
3253 return nfserr_bad_stateid;
3255 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3258 renew_client(sop->so_client);
3262 if (seqid == sop->so_seqid - 1) {
3263 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3264 /* indicate replay to calling function */
3265 return nfserr_replay_me;
3267 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3268 sop->so_seqid, seqid);
3270 return nfserr_bad_seqid;
3274 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3275 struct nfsd4_open_confirm *oc)
3278 struct nfs4_stateowner *sop;
3279 struct nfs4_stateid *stp;
3281 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3282 (int)cstate->current_fh.fh_dentry->d_name.len,
3283 cstate->current_fh.fh_dentry->d_name.name);
3285 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3291 if ((status = nfs4_preprocess_seqid_op(cstate,
3292 oc->oc_seqid, &oc->oc_req_stateid,
3293 CONFIRM | OPEN_STATE,
3294 &oc->oc_stateowner, &stp, NULL)))
3297 sop = oc->oc_stateowner;
3298 sop->so_confirmed = 1;
3299 update_stateid(&stp->st_stateid);
3300 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3301 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3302 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3304 nfsd4_create_clid_dir(sop->so_client);
3306 if (oc->oc_stateowner) {
3307 nfs4_get_stateowner(oc->oc_stateowner);
3308 cstate->replay_owner = oc->oc_stateowner;
3310 nfs4_unlock_state();
3316 * unset all bits in union bitmap (bmap) that
3317 * do not exist in share (from successful OPEN_DOWNGRADE)
3320 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3323 for (i = 1; i < 4; i++) {
3324 if ((i & access) != i)
3325 __clear_bit(i, bmap);
3330 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3333 for (i = 0; i < 4; i++) {
3334 if ((i & deny) != i)
3335 __clear_bit(i, bmap);
3340 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3341 struct nfsd4_compound_state *cstate,
3342 struct nfsd4_open_downgrade *od)
3345 struct nfs4_stateid *stp;
3346 unsigned int share_access;
3348 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3349 (int)cstate->current_fh.fh_dentry->d_name.len,
3350 cstate->current_fh.fh_dentry->d_name.name);
3352 if (!access_valid(od->od_share_access, cstate->minorversion)
3353 || !deny_valid(od->od_share_deny))
3354 return nfserr_inval;
3357 if ((status = nfs4_preprocess_seqid_op(cstate,
3361 &od->od_stateowner, &stp, NULL)))
3364 status = nfserr_inval;
3365 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3366 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3367 stp->st_access_bmap, od->od_share_access);
3370 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3371 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3372 stp->st_deny_bmap, od->od_share_deny);
3375 set_access(&share_access, stp->st_access_bmap);
3376 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3378 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3379 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3381 update_stateid(&stp->st_stateid);
3382 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3385 if (od->od_stateowner) {
3386 nfs4_get_stateowner(od->od_stateowner);
3387 cstate->replay_owner = od->od_stateowner;
3389 nfs4_unlock_state();
3394 * nfs4_unlock_state() called after encode
3397 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3398 struct nfsd4_close *close)
3401 struct nfs4_stateid *stp;
3403 dprintk("NFSD: nfsd4_close on file %.*s\n",
3404 (int)cstate->current_fh.fh_dentry->d_name.len,
3405 cstate->current_fh.fh_dentry->d_name.name);
3408 /* check close_lru for replay */
3409 if ((status = nfs4_preprocess_seqid_op(cstate,
3412 OPEN_STATE | CLOSE_STATE,
3413 &close->cl_stateowner, &stp, NULL)))
3416 update_stateid(&stp->st_stateid);
3417 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3419 /* release_stateid() calls nfsd_close() if needed */
3420 release_open_stateid(stp);
3422 /* place unused nfs4_stateowners on so_close_lru list to be
3423 * released by the laundromat service after the lease period
3424 * to enable us to handle CLOSE replay
3426 if (list_empty(&close->cl_stateowner->so_stateids))
3427 move_to_close_lru(close->cl_stateowner);
3429 if (close->cl_stateowner) {
3430 nfs4_get_stateowner(close->cl_stateowner);
3431 cstate->replay_owner = close->cl_stateowner;
3433 nfs4_unlock_state();
3438 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3439 struct nfsd4_delegreturn *dr)
3441 struct nfs4_delegation *dp;
3442 stateid_t *stateid = &dr->dr_stateid;
3443 struct inode *inode;
3447 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3449 inode = cstate->current_fh.fh_dentry->d_inode;
3451 if (nfsd4_has_session(cstate))
3452 flags |= HAS_SESSION;
3454 status = nfserr_bad_stateid;
3455 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3457 status = nfserr_stale_stateid;
3458 if (STALE_STATEID(stateid))
3460 status = nfserr_bad_stateid;
3461 if (!is_delegation_stateid(stateid))
3463 status = nfserr_expired;
3464 dp = find_delegation_stateid(inode, stateid);
3467 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3470 renew_client(dp->dl_client);
3472 unhash_delegation(dp);
3474 nfs4_unlock_state();
3481 * Lock owner state (byte-range locks)
3483 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3484 #define LOCK_HASH_BITS 8
3485 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3486 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3489 end_offset(u64 start, u64 len)
3494 return end >= start ? end: NFS4_MAX_UINT64;
3497 /* last octet in a range */
3499 last_byte_offset(u64 start, u64 len)
3505 return end > start ? end - 1: NFS4_MAX_UINT64;
3508 #define lockownerid_hashval(id) \
3509 ((id) & LOCK_HASH_MASK)
3511 static inline unsigned int
3512 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3513 struct xdr_netobj *ownername)
3515 return (file_hashval(inode) + cl_id
3516 + opaque_hashval(ownername->data, ownername->len))
3520 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3521 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3522 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3524 static struct nfs4_stateid *
3525 find_stateid(stateid_t *stid, int flags)
3527 struct nfs4_stateid *local;
3528 u32 st_id = stid->si_stateownerid;
3529 u32 f_id = stid->si_fileid;
3530 unsigned int hashval;
3532 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3533 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3534 hashval = stateid_hashval(st_id, f_id);
3535 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3536 if ((local->st_stateid.si_stateownerid == st_id) &&
3537 (local->st_stateid.si_fileid == f_id))
3542 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3543 hashval = stateid_hashval(st_id, f_id);
3544 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3545 if ((local->st_stateid.si_stateownerid == st_id) &&
3546 (local->st_stateid.si_fileid == f_id))
3553 static struct nfs4_delegation *
3554 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3556 struct nfs4_file *fp;
3557 struct nfs4_delegation *dl;
3559 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3562 fp = find_file(ino);
3565 dl = find_delegation_file(fp, stid);
3571 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3572 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3573 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3574 * locking, this prevents us from being completely protocol-compliant. The
3575 * real solution to this problem is to start using unsigned file offsets in
3576 * the VFS, but this is a very deep change!
3579 nfs4_transform_lock_offset(struct file_lock *lock)
3581 if (lock->fl_start < 0)
3582 lock->fl_start = OFFSET_MAX;
3583 if (lock->fl_end < 0)
3584 lock->fl_end = OFFSET_MAX;
3587 /* Hack!: For now, we're defining this just so we can use a pointer to it
3588 * as a unique cookie to identify our (NFSv4's) posix locks. */
3589 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3593 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3595 struct nfs4_stateowner *sop;
3597 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3598 sop = (struct nfs4_stateowner *) fl->fl_owner;
3599 kref_get(&sop->so_ref);
3601 deny->ld_clientid = sop->so_client->cl_clientid;
3603 deny->ld_sop = NULL;
3604 deny->ld_clientid.cl_boot = 0;
3605 deny->ld_clientid.cl_id = 0;
3607 deny->ld_start = fl->fl_start;
3608 deny->ld_length = NFS4_MAX_UINT64;
3609 if (fl->fl_end != NFS4_MAX_UINT64)
3610 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3611 deny->ld_type = NFS4_READ_LT;
3612 if (fl->fl_type != F_RDLCK)
3613 deny->ld_type = NFS4_WRITE_LT;
3616 static struct nfs4_stateowner *
3617 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3618 struct xdr_netobj *owner)
3620 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3621 struct nfs4_stateowner *op;
3623 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3624 if (same_owner_str(op, owner, clid))
3631 * Alloc a lock owner structure.
3632 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3635 * strhashval = lock_ownerstr_hashval
3638 static struct nfs4_stateowner *
3639 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3640 struct nfs4_stateowner *sop;
3641 struct nfs4_replay *rp;
3642 unsigned int idhashval;
3644 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3646 idhashval = lockownerid_hashval(current_ownerid);
3647 INIT_LIST_HEAD(&sop->so_idhash);
3648 INIT_LIST_HEAD(&sop->so_strhash);
3649 INIT_LIST_HEAD(&sop->so_perclient);
3650 INIT_LIST_HEAD(&sop->so_stateids);
3651 INIT_LIST_HEAD(&sop->so_perstateid);
3652 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3654 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3655 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3656 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3657 sop->so_is_open_owner = 0;
3658 sop->so_id = current_ownerid++;
3659 sop->so_client = clp;
3660 /* It is the openowner seqid that will be incremented in encode in the
3661 * case of new lockowners; so increment the lock seqid manually: */
3662 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3663 sop->so_confirmed = 1;
3664 rp = &sop->so_replay;
3665 rp->rp_status = nfserr_serverfault;
3667 rp->rp_buf = rp->rp_ibuf;
3671 static struct nfs4_stateid *
3672 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3674 struct nfs4_stateid *stp;
3675 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3677 stp = nfs4_alloc_stateid();
3680 INIT_LIST_HEAD(&stp->st_hash);
3681 INIT_LIST_HEAD(&stp->st_perfile);
3682 INIT_LIST_HEAD(&stp->st_perstateowner);
3683 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3684 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3685 list_add(&stp->st_perfile, &fp->fi_stateids);
3686 list_add(&stp->st_perstateowner, &sop->so_stateids);
3687 stp->st_stateowner = sop;
3690 stp->st_stateid.si_boot = boot_time;
3691 stp->st_stateid.si_stateownerid = sop->so_id;
3692 stp->st_stateid.si_fileid = fp->fi_id;
3693 stp->st_stateid.si_generation = 0;
3694 stp->st_deny_bmap = open_stp->st_deny_bmap;
3695 stp->st_openstp = open_stp;
3702 check_lock_length(u64 offset, u64 length)
3704 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3705 LOFF_OVERFLOW(offset, length)));
3712 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3713 struct nfsd4_lock *lock)
3715 struct nfs4_stateowner *open_sop = NULL;
3716 struct nfs4_stateowner *lock_sop = NULL;
3717 struct nfs4_stateid *lock_stp;
3718 struct nfs4_file *fp;
3719 struct file *filp = NULL;
3720 struct file_lock file_lock;
3721 struct file_lock conflock;
3723 unsigned int strhashval;
3727 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3728 (long long) lock->lk_offset,
3729 (long long) lock->lk_length);
3731 if (check_lock_length(lock->lk_offset, lock->lk_length))
3732 return nfserr_inval;
3734 if ((status = fh_verify(rqstp, &cstate->current_fh,
3735 S_IFREG, NFSD_MAY_LOCK))) {
3736 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3742 if (lock->lk_is_new) {
3744 * Client indicates that this is a new lockowner.
3745 * Use open owner and open stateid to create lock owner and
3748 struct nfs4_stateid *open_stp = NULL;
3750 status = nfserr_stale_clientid;
3751 if (!nfsd4_has_session(cstate) &&
3752 STALE_CLIENTID(&lock->lk_new_clientid))
3755 /* validate and update open stateid and open seqid */
3756 status = nfs4_preprocess_seqid_op(cstate,
3757 lock->lk_new_open_seqid,
3758 &lock->lk_new_open_stateid,
3760 &lock->lk_replay_owner, &open_stp,
3764 open_sop = lock->lk_replay_owner;
3765 /* create lockowner and lock stateid */
3766 fp = open_stp->st_file;
3767 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3768 open_sop->so_client->cl_clientid.cl_id,
3769 &lock->v.new.owner);
3770 /* XXX: Do we need to check for duplicate stateowners on
3771 * the same file, or should they just be allowed (and
3772 * create new stateids)? */
3773 status = nfserr_resource;
3774 lock_sop = alloc_init_lock_stateowner(strhashval,
3775 open_sop->so_client, open_stp, lock);
3776 if (lock_sop == NULL)
3778 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3779 if (lock_stp == NULL)
3782 /* lock (lock owner + lock stateid) already exists */
3783 status = nfs4_preprocess_seqid_op(cstate,
3784 lock->lk_old_lock_seqid,
3785 &lock->lk_old_lock_stateid,
3787 &lock->lk_replay_owner, &lock_stp, lock);
3790 lock_sop = lock->lk_replay_owner;
3791 fp = lock_stp->st_file;
3793 /* lock->lk_replay_owner and lock_stp have been created or found */
3795 status = nfserr_grace;
3796 if (locks_in_grace() && !lock->lk_reclaim)
3798 status = nfserr_no_grace;
3799 if (!locks_in_grace() && lock->lk_reclaim)
3802 locks_init_lock(&file_lock);
3803 switch (lock->lk_type) {
3806 if (find_readable_file(lock_stp->st_file)) {
3807 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3808 filp = find_readable_file(lock_stp->st_file);
3810 file_lock.fl_type = F_RDLCK;
3814 case NFS4_WRITEW_LT:
3815 if (find_writeable_file(lock_stp->st_file)) {
3816 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3817 filp = find_writeable_file(lock_stp->st_file);
3819 file_lock.fl_type = F_WRLCK;
3823 status = nfserr_inval;
3827 status = nfserr_openmode;
3830 file_lock.fl_owner = (fl_owner_t)lock_sop;
3831 file_lock.fl_pid = current->tgid;
3832 file_lock.fl_file = filp;
3833 file_lock.fl_flags = FL_POSIX;
3834 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3836 file_lock.fl_start = lock->lk_offset;
3837 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3838 nfs4_transform_lock_offset(&file_lock);
3841 * Try to lock the file in the VFS.
3842 * Note: locks.c uses the BKL to protect the inode's lock list.
3845 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3847 case 0: /* success! */
3848 update_stateid(&lock_stp->st_stateid);
3849 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3853 case (EAGAIN): /* conflock holds conflicting lock */
3854 status = nfserr_denied;
3855 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3856 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3859 status = nfserr_deadlock;
3862 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3863 status = nfserr_resource;
3867 if (status && lock->lk_is_new && lock_sop)
3868 release_lockowner(lock_sop);
3869 if (lock->lk_replay_owner) {
3870 nfs4_get_stateowner(lock->lk_replay_owner);
3871 cstate->replay_owner = lock->lk_replay_owner;
3873 nfs4_unlock_state();
3878 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3879 * so we do a temporary open here just to get an open file to pass to
3880 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3883 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3888 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3891 err = vfs_test_lock(file, lock);
3900 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3901 struct nfsd4_lockt *lockt)
3903 struct inode *inode;
3904 struct file_lock file_lock;
3908 if (locks_in_grace())
3909 return nfserr_grace;
3911 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3912 return nfserr_inval;
3914 lockt->lt_stateowner = NULL;
3917 status = nfserr_stale_clientid;
3918 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3921 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3922 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3923 if (status == nfserr_symlink)
3924 status = nfserr_inval;
3928 inode = cstate->current_fh.fh_dentry->d_inode;
3929 locks_init_lock(&file_lock);
3930 switch (lockt->lt_type) {
3933 file_lock.fl_type = F_RDLCK;
3936 case NFS4_WRITEW_LT:
3937 file_lock.fl_type = F_WRLCK;
3940 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3941 status = nfserr_inval;
3945 lockt->lt_stateowner = find_lockstateowner_str(inode,
3946 &lockt->lt_clientid, &lockt->lt_owner);
3947 if (lockt->lt_stateowner)
3948 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3949 file_lock.fl_pid = current->tgid;
3950 file_lock.fl_flags = FL_POSIX;
3952 file_lock.fl_start = lockt->lt_offset;
3953 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3955 nfs4_transform_lock_offset(&file_lock);
3958 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3960 status = nfserrno(error);
3963 if (file_lock.fl_type != F_UNLCK) {
3964 status = nfserr_denied;
3965 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3968 nfs4_unlock_state();
3973 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3974 struct nfsd4_locku *locku)
3976 struct nfs4_stateid *stp;
3977 struct file *filp = NULL;
3978 struct file_lock file_lock;
3982 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3983 (long long) locku->lu_offset,
3984 (long long) locku->lu_length);
3986 if (check_lock_length(locku->lu_offset, locku->lu_length))
3987 return nfserr_inval;
3991 if ((status = nfs4_preprocess_seqid_op(cstate,
3995 &locku->lu_stateowner, &stp, NULL)))
3998 filp = find_any_file(stp->st_file);
4000 status = nfserr_lock_range;
4004 locks_init_lock(&file_lock);
4005 file_lock.fl_type = F_UNLCK;
4006 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
4007 file_lock.fl_pid = current->tgid;
4008 file_lock.fl_file = filp;
4009 file_lock.fl_flags = FL_POSIX;
4010 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4011 file_lock.fl_start = locku->lu_offset;
4013 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4014 nfs4_transform_lock_offset(&file_lock);
4017 * Try to unlock the file in the VFS.
4019 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4021 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4025 * OK, unlock succeeded; the only thing left to do is update the stateid.
4027 update_stateid(&stp->st_stateid);
4028 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4031 if (locku->lu_stateowner) {
4032 nfs4_get_stateowner(locku->lu_stateowner);
4033 cstate->replay_owner = locku->lu_stateowner;
4035 nfs4_unlock_state();
4039 status = nfserrno(err);
4045 * 1: locks held by lockowner
4046 * 0: no locks held by lockowner
4049 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4051 struct file_lock **flpp;
4052 struct inode *inode = filp->fi_inode;
4056 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4057 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4068 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4069 struct nfsd4_compound_state *cstate,
4070 struct nfsd4_release_lockowner *rlockowner)
4072 clientid_t *clid = &rlockowner->rl_clientid;
4073 struct nfs4_stateowner *sop;
4074 struct nfs4_stateid *stp;
4075 struct xdr_netobj *owner = &rlockowner->rl_owner;
4076 struct list_head matches;
4080 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4081 clid->cl_boot, clid->cl_id);
4083 /* XXX check for lease expiration */
4085 status = nfserr_stale_clientid;
4086 if (STALE_CLIENTID(clid))
4091 status = nfserr_locks_held;
4092 /* XXX: we're doing a linear search through all the lockowners.
4093 * Yipes! For now we'll just hope clients aren't really using
4094 * release_lockowner much, but eventually we have to fix these
4095 * data structures. */
4096 INIT_LIST_HEAD(&matches);
4097 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4098 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4099 if (!same_owner_str(sop, owner, clid))
4101 list_for_each_entry(stp, &sop->so_stateids,
4103 if (check_for_locks(stp->st_file, sop))
4105 /* Note: so_perclient unused for lockowners,
4106 * so it's OK to fool with here. */
4107 list_add(&sop->so_perclient, &matches);
4111 /* Clients probably won't expect us to return with some (but not all)
4112 * of the lockowner state released; so don't release any until all
4113 * have been checked. */
4115 while (!list_empty(&matches)) {
4116 sop = list_entry(matches.next, struct nfs4_stateowner,
4118 /* unhash_stateowner deletes so_perclient only
4119 * for openowners. */
4120 list_del(&sop->so_perclient);
4121 release_lockowner(sop);
4124 nfs4_unlock_state();
4128 static inline struct nfs4_client_reclaim *
4131 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4135 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4137 unsigned int strhashval = clientstr_hashval(name);
4138 struct nfs4_client *clp;
4140 clp = find_confirmed_client_by_str(name, strhashval);
4145 * failure => all reset bets are off, nfserr_no_grace...
4148 nfs4_client_to_reclaim(const char *name)
4150 unsigned int strhashval;
4151 struct nfs4_client_reclaim *crp = NULL;
4153 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4154 crp = alloc_reclaim();
4157 strhashval = clientstr_hashval(name);
4158 INIT_LIST_HEAD(&crp->cr_strhash);
4159 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4160 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4161 reclaim_str_hashtbl_size++;
4166 nfs4_release_reclaim(void)
4168 struct nfs4_client_reclaim *crp = NULL;
4171 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4172 while (!list_empty(&reclaim_str_hashtbl[i])) {
4173 crp = list_entry(reclaim_str_hashtbl[i].next,
4174 struct nfs4_client_reclaim, cr_strhash);
4175 list_del(&crp->cr_strhash);
4177 reclaim_str_hashtbl_size--;
4180 BUG_ON(reclaim_str_hashtbl_size);
4184 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4185 static struct nfs4_client_reclaim *
4186 nfs4_find_reclaim_client(clientid_t *clid)
4188 unsigned int strhashval;
4189 struct nfs4_client *clp;
4190 struct nfs4_client_reclaim *crp = NULL;
4193 /* find clientid in conf_id_hashtbl */
4194 clp = find_confirmed_client(clid);
4198 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4199 clp->cl_name.len, clp->cl_name.data,
4202 /* find clp->cl_name in reclaim_str_hashtbl */
4203 strhashval = clientstr_hashval(clp->cl_recdir);
4204 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4205 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4213 * Called from OPEN. Look for clientid in reclaim list.
4216 nfs4_check_open_reclaim(clientid_t *clid)
4218 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4221 /* initialization to perform at module load time: */
4224 nfs4_state_init(void)
4228 status = nfsd4_init_slabs();
4231 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4232 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4233 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4234 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4235 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4236 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4238 for (i = 0; i < SESSION_HASH_SIZE; i++)
4239 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4240 for (i = 0; i < FILE_HASH_SIZE; i++) {
4241 INIT_LIST_HEAD(&file_hashtbl[i]);
4243 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4244 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4245 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4247 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4248 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4249 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4251 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4252 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4253 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4255 memset(&onestateid, ~0, sizeof(stateid_t));
4256 INIT_LIST_HEAD(&close_lru);
4257 INIT_LIST_HEAD(&client_lru);
4258 INIT_LIST_HEAD(&del_recall_lru);
4259 reclaim_str_hashtbl_size = 0;
4264 nfsd4_load_reboot_recovery_data(void)
4269 nfsd4_init_recdir(user_recovery_dirname);
4270 status = nfsd4_recdir_load();
4271 nfs4_unlock_state();
4273 printk("NFSD: Failure reading reboot recovery data\n");
4277 * Since the lifetime of a delegation isn't limited to that of an open, a
4278 * client may quite reasonably hang on to a delegation as long as it has
4279 * the inode cached. This becomes an obvious problem the first time a
4280 * client's inode cache approaches the size of the server's total memory.
4282 * For now we avoid this problem by imposing a hard limit on the number
4283 * of delegations, which varies according to the server's memory size.
4286 set_max_delegations(void)
4289 * Allow at most 4 delegations per megabyte of RAM. Quick
4290 * estimates suggest that in the worst case (where every delegation
4291 * is for a different inode), a delegation could take about 1.5K,
4292 * giving a worst case usage of about 6% of memory.
4294 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4297 /* initialization to perform when the nfsd service is started: */
4300 __nfs4_state_start(void)
4304 boot_time = get_seconds();
4305 locks_start_grace(&nfsd4_manager);
4306 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4308 ret = set_callback_cred();
4311 laundry_wq = create_singlethread_workqueue("nfsd4");
4312 if (laundry_wq == NULL)
4314 ret = nfsd4_create_callback_queue();
4316 goto out_free_laundry;
4317 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4318 set_max_delegations();
4321 destroy_workqueue(laundry_wq);
4326 nfs4_state_start(void)
4328 nfsd4_load_reboot_recovery_data();
4329 return __nfs4_state_start();
4333 __nfs4_state_shutdown(void)
4336 struct nfs4_client *clp = NULL;
4337 struct nfs4_delegation *dp = NULL;
4338 struct list_head *pos, *next, reaplist;
4340 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4341 while (!list_empty(&conf_id_hashtbl[i])) {
4342 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4345 while (!list_empty(&unconf_str_hashtbl[i])) {
4346 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4350 INIT_LIST_HEAD(&reaplist);
4351 spin_lock(&recall_lock);
4352 list_for_each_safe(pos, next, &del_recall_lru) {
4353 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4354 list_move(&dp->dl_recall_lru, &reaplist);
4356 spin_unlock(&recall_lock);
4357 list_for_each_safe(pos, next, &reaplist) {
4358 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4359 list_del_init(&dp->dl_recall_lru);
4360 unhash_delegation(dp);
4363 nfsd4_shutdown_recdir();
4367 nfs4_state_shutdown(void)
4369 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4370 destroy_workqueue(laundry_wq);
4371 locks_end_grace(&nfsd4_manager);
4373 nfs4_release_reclaim();
4374 __nfs4_state_shutdown();
4375 nfs4_unlock_state();
4376 nfsd4_destroy_callback_queue();
4380 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4381 * accessed when nfsd is starting.
4384 nfs4_set_recdir(char *recdir)
4386 strcpy(user_recovery_dirname, recdir);
4390 * Change the NFSv4 recovery directory to recdir.
4393 nfs4_reset_recoverydir(char *recdir)
4398 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4402 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4403 nfs4_set_recdir(recdir);
4411 nfs4_recoverydir(void)
4413 return user_recovery_dirname;