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)
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);
234 dp->dl_stateid.si_boot = boot_time;
235 dp->dl_stateid.si_stateownerid = current_delegid++;
236 dp->dl_stateid.si_fileid = 0;
237 dp->dl_stateid.si_generation = 0;
238 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
240 atomic_set(&dp->dl_count, 1);
241 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
246 nfs4_put_delegation(struct nfs4_delegation *dp)
248 if (atomic_dec_and_test(&dp->dl_count)) {
249 dprintk("NFSD: freeing dp %p\n",dp);
250 put_nfs4_file(dp->dl_file);
251 kmem_cache_free(deleg_slab, dp);
256 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
258 if (atomic_dec_and_test(&fp->fi_delegees)) {
259 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
261 fp->fi_deleg_file = NULL;
265 /* Called under the state lock. */
267 unhash_delegation(struct nfs4_delegation *dp)
269 list_del_init(&dp->dl_perclnt);
270 spin_lock(&recall_lock);
271 list_del_init(&dp->dl_perfile);
272 list_del_init(&dp->dl_recall_lru);
273 spin_unlock(&recall_lock);
274 nfs4_put_deleg_lease(dp->dl_file);
275 nfs4_put_delegation(dp);
282 /* client_lock protects the client lru list and session hash table */
283 static DEFINE_SPINLOCK(client_lock);
285 /* Hash tables for nfs4_clientid state */
286 #define CLIENT_HASH_BITS 4
287 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
288 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
290 #define clientid_hashval(id) \
291 ((id) & CLIENT_HASH_MASK)
292 #define clientstr_hashval(name) \
293 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
295 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
296 * used in reboot/reset lease grace period processing
298 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
299 * setclientid_confirmed info.
301 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
304 * client_lru holds client queue ordered by nfs4_client.cl_time
307 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
308 * for last close replay.
310 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
311 static int reclaim_str_hashtbl_size = 0;
312 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
313 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
314 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
315 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
316 static struct list_head client_lru;
317 static struct list_head close_lru;
320 * We store the NONE, READ, WRITE, and BOTH bits separately in the
321 * st_{access,deny}_bmap field of the stateid, in order to track not
322 * only what share bits are currently in force, but also what
323 * combinations of share bits previous opens have used. This allows us
324 * to enforce the recommendation of rfc 3530 14.2.19 that the server
325 * return an error if the client attempt to downgrade to a combination
326 * of share bits not explicable by closing some of its previous opens.
328 * XXX: This enforcement is actually incomplete, since we don't keep
329 * track of access/deny bit combinations; so, e.g., we allow:
331 * OPEN allow read, deny write
332 * OPEN allow both, deny none
333 * DOWNGRADE allow read, deny none
335 * which we should reject.
338 set_access(unsigned int *access, unsigned long bmap) {
342 for (i = 1; i < 4; i++) {
343 if (test_bit(i, &bmap))
349 set_deny(unsigned int *deny, unsigned long bmap) {
353 for (i = 0; i < 4; i++) {
354 if (test_bit(i, &bmap))
360 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
361 unsigned int access, deny;
363 set_access(&access, stp->st_access_bmap);
364 set_deny(&deny, stp->st_deny_bmap);
365 if ((access & open->op_share_deny) || (deny & open->op_share_access))
370 static int nfs4_access_to_omode(u32 access)
372 switch (access & NFS4_SHARE_ACCESS_BOTH) {
373 case NFS4_SHARE_ACCESS_READ:
375 case NFS4_SHARE_ACCESS_WRITE:
377 case NFS4_SHARE_ACCESS_BOTH:
383 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
387 set_access(&access, stp->st_access_bmap);
388 return nfs4_access_to_omode(access);
391 static void unhash_generic_stateid(struct nfs4_stateid *stp)
393 list_del(&stp->st_hash);
394 list_del(&stp->st_perfile);
395 list_del(&stp->st_perstateowner);
398 static void free_generic_stateid(struct nfs4_stateid *stp)
400 int oflag = nfs4_access_bmap_to_omode(stp);
402 nfs4_file_put_access(stp->st_file, oflag);
403 put_nfs4_file(stp->st_file);
404 kmem_cache_free(stateid_slab, stp);
407 static void release_lock_stateid(struct nfs4_stateid *stp)
411 unhash_generic_stateid(stp);
412 file = find_any_file(stp->st_file);
414 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
415 free_generic_stateid(stp);
418 static void unhash_lockowner(struct nfs4_stateowner *sop)
420 struct nfs4_stateid *stp;
422 list_del(&sop->so_idhash);
423 list_del(&sop->so_strhash);
424 list_del(&sop->so_perstateid);
425 while (!list_empty(&sop->so_stateids)) {
426 stp = list_first_entry(&sop->so_stateids,
427 struct nfs4_stateid, st_perstateowner);
428 release_lock_stateid(stp);
432 static void release_lockowner(struct nfs4_stateowner *sop)
434 unhash_lockowner(sop);
435 nfs4_put_stateowner(sop);
439 release_stateid_lockowners(struct nfs4_stateid *open_stp)
441 struct nfs4_stateowner *lock_sop;
443 while (!list_empty(&open_stp->st_lockowners)) {
444 lock_sop = list_entry(open_stp->st_lockowners.next,
445 struct nfs4_stateowner, so_perstateid);
446 /* list_del(&open_stp->st_lockowners); */
447 BUG_ON(lock_sop->so_is_open_owner);
448 release_lockowner(lock_sop);
452 static void release_open_stateid(struct nfs4_stateid *stp)
454 unhash_generic_stateid(stp);
455 release_stateid_lockowners(stp);
456 free_generic_stateid(stp);
459 static void unhash_openowner(struct nfs4_stateowner *sop)
461 struct nfs4_stateid *stp;
463 list_del(&sop->so_idhash);
464 list_del(&sop->so_strhash);
465 list_del(&sop->so_perclient);
466 list_del(&sop->so_perstateid); /* XXX: necessary? */
467 while (!list_empty(&sop->so_stateids)) {
468 stp = list_first_entry(&sop->so_stateids,
469 struct nfs4_stateid, st_perstateowner);
470 release_open_stateid(stp);
474 static void release_openowner(struct nfs4_stateowner *sop)
476 unhash_openowner(sop);
477 list_del(&sop->so_close_lru);
478 nfs4_put_stateowner(sop);
481 #define SESSION_HASH_SIZE 512
482 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
485 hash_sessionid(struct nfs4_sessionid *sessionid)
487 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
489 return sid->sequence % SESSION_HASH_SIZE;
493 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
495 u32 *ptr = (u32 *)(&sessionid->data[0]);
496 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
500 gen_sessionid(struct nfsd4_session *ses)
502 struct nfs4_client *clp = ses->se_client;
503 struct nfsd4_sessionid *sid;
505 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
506 sid->clientid = clp->cl_clientid;
507 sid->sequence = current_sessionid++;
512 * The protocol defines ca_maxresponssize_cached to include the size of
513 * the rpc header, but all we need to cache is the data starting after
514 * the end of the initial SEQUENCE operation--the rest we regenerate
515 * each time. Therefore we can advertise a ca_maxresponssize_cached
516 * value that is the number of bytes in our cache plus a few additional
517 * bytes. In order to stay on the safe side, and not promise more than
518 * we can cache, those additional bytes must be the minimum possible: 24
519 * bytes of rpc header (xid through accept state, with AUTH_NULL
520 * verifier), 12 for the compound header (with zero-length tag), and 44
521 * for the SEQUENCE op response:
523 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
526 free_session_slots(struct nfsd4_session *ses)
530 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
531 kfree(ses->se_slots[i]);
535 * We don't actually need to cache the rpc and session headers, so we
536 * can allocate a little less for each slot:
538 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
540 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
543 static int nfsd4_sanitize_slot_size(u32 size)
545 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
546 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
552 * XXX: If we run out of reserved DRC memory we could (up to a point)
553 * re-negotiate active sessions and reduce their slot usage to make
554 * rooom for new connections. For now we just fail the create session.
556 static int nfsd4_get_drc_mem(int slotsize, u32 num)
560 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
562 spin_lock(&nfsd_drc_lock);
563 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
564 nfsd_drc_max_mem - nfsd_drc_mem_used);
565 num = min_t(int, num, avail / slotsize);
566 nfsd_drc_mem_used += num * slotsize;
567 spin_unlock(&nfsd_drc_lock);
572 static void nfsd4_put_drc_mem(int slotsize, int num)
574 spin_lock(&nfsd_drc_lock);
575 nfsd_drc_mem_used -= slotsize * num;
576 spin_unlock(&nfsd_drc_lock);
579 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
581 struct nfsd4_session *new;
584 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
585 + sizeof(struct nfsd4_session) > PAGE_SIZE);
586 mem = numslots * sizeof(struct nfsd4_slot *);
588 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
591 /* allocate each struct nfsd4_slot and data cache in one piece */
592 for (i = 0; i < numslots; i++) {
593 mem = sizeof(struct nfsd4_slot) + slotsize;
594 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
595 if (!new->se_slots[i])
601 kfree(new->se_slots[i]);
606 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
608 u32 maxrpc = nfsd_serv->sv_max_mesg;
610 new->maxreqs = numslots;
611 new->maxresp_cached = min_t(u32, req->maxresp_cached,
612 slotsize + NFSD_MIN_HDR_SEQ_SZ);
613 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
614 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
615 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
618 static void free_conn(struct nfsd4_conn *c)
620 svc_xprt_put(c->cn_xprt);
624 static void nfsd4_conn_lost(struct svc_xpt_user *u)
626 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
627 struct nfs4_client *clp = c->cn_session->se_client;
629 spin_lock(&clp->cl_lock);
630 if (!list_empty(&c->cn_persession)) {
631 list_del(&c->cn_persession);
634 spin_unlock(&clp->cl_lock);
635 nfsd4_probe_callback(clp);
638 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
640 struct nfsd4_conn *conn;
642 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
645 svc_xprt_get(rqstp->rq_xprt);
646 conn->cn_xprt = rqstp->rq_xprt;
647 conn->cn_flags = flags;
648 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
652 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
654 conn->cn_session = ses;
655 list_add(&conn->cn_persession, &ses->se_conns);
658 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
660 struct nfs4_client *clp = ses->se_client;
662 spin_lock(&clp->cl_lock);
663 __nfsd4_hash_conn(conn, ses);
664 spin_unlock(&clp->cl_lock);
667 static int nfsd4_register_conn(struct nfsd4_conn *conn)
669 conn->cn_xpt_user.callback = nfsd4_conn_lost;
670 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
673 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
675 struct nfsd4_conn *conn;
678 conn = alloc_conn(rqstp, dir);
680 return nfserr_jukebox;
681 nfsd4_hash_conn(conn, ses);
682 ret = nfsd4_register_conn(conn);
684 /* oops; xprt is already down: */
685 nfsd4_conn_lost(&conn->cn_xpt_user);
689 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
691 u32 dir = NFS4_CDFC4_FORE;
693 if (ses->se_flags & SESSION4_BACK_CHAN)
694 dir |= NFS4_CDFC4_BACK;
696 return nfsd4_new_conn(rqstp, ses, dir);
699 /* must be called under client_lock */
700 static void nfsd4_del_conns(struct nfsd4_session *s)
702 struct nfs4_client *clp = s->se_client;
703 struct nfsd4_conn *c;
705 spin_lock(&clp->cl_lock);
706 while (!list_empty(&s->se_conns)) {
707 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
708 list_del_init(&c->cn_persession);
709 spin_unlock(&clp->cl_lock);
711 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
714 spin_lock(&clp->cl_lock);
716 spin_unlock(&clp->cl_lock);
719 void free_session(struct kref *kref)
721 struct nfsd4_session *ses;
724 ses = container_of(kref, struct nfsd4_session, se_ref);
725 nfsd4_del_conns(ses);
726 spin_lock(&nfsd_drc_lock);
727 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
728 nfsd_drc_mem_used -= mem;
729 spin_unlock(&nfsd_drc_lock);
730 free_session_slots(ses);
734 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
736 struct nfsd4_session *new;
737 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
738 int numslots, slotsize;
743 * Note decreasing slot size below client's request may
744 * make it difficult for client to function correctly, whereas
745 * decreasing the number of slots will (just?) affect
746 * performance. When short on memory we therefore prefer to
747 * decrease number of slots instead of their size.
749 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
750 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
754 new = alloc_session(slotsize, numslots);
756 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
759 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
761 new->se_client = clp;
764 INIT_LIST_HEAD(&new->se_conns);
766 new->se_cb_seq_nr = 1;
767 new->se_flags = cses->flags;
768 new->se_cb_prog = cses->callback_prog;
769 kref_init(&new->se_ref);
770 idx = hash_sessionid(&new->se_sessionid);
771 spin_lock(&client_lock);
772 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
773 spin_lock(&clp->cl_lock);
774 list_add(&new->se_perclnt, &clp->cl_sessions);
775 spin_unlock(&clp->cl_lock);
776 spin_unlock(&client_lock);
778 status = nfsd4_new_conn_from_crses(rqstp, new);
779 /* whoops: benny points out, status is ignored! (err, or bogus) */
781 free_session(&new->se_ref);
784 if (cses->flags & SESSION4_BACK_CHAN) {
785 struct sockaddr *sa = svc_addr(rqstp);
787 * This is a little silly; with sessions there's no real
788 * use for the callback address. Use the peer address
789 * as a reasonable default for now, but consider fixing
790 * the rpc client not to require an address in the
793 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
794 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
796 nfsd4_probe_callback(clp);
800 /* caller must hold client_lock */
801 static struct nfsd4_session *
802 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
804 struct nfsd4_session *elem;
807 dump_sessionid(__func__, sessionid);
808 idx = hash_sessionid(sessionid);
809 /* Search in the appropriate list */
810 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
811 if (!memcmp(elem->se_sessionid.data, sessionid->data,
812 NFS4_MAX_SESSIONID_LEN)) {
817 dprintk("%s: session not found\n", __func__);
821 /* caller must hold client_lock */
823 unhash_session(struct nfsd4_session *ses)
825 list_del(&ses->se_hash);
826 spin_lock(&ses->se_client->cl_lock);
827 list_del(&ses->se_perclnt);
828 spin_unlock(&ses->se_client->cl_lock);
831 /* must be called under the client_lock */
833 renew_client_locked(struct nfs4_client *clp)
835 if (is_client_expired(clp)) {
836 dprintk("%s: client (clientid %08x/%08x) already expired\n",
838 clp->cl_clientid.cl_boot,
839 clp->cl_clientid.cl_id);
844 * Move client to the end to the LRU list.
846 dprintk("renewing client (clientid %08x/%08x)\n",
847 clp->cl_clientid.cl_boot,
848 clp->cl_clientid.cl_id);
849 list_move_tail(&clp->cl_lru, &client_lru);
850 clp->cl_time = get_seconds();
854 renew_client(struct nfs4_client *clp)
856 spin_lock(&client_lock);
857 renew_client_locked(clp);
858 spin_unlock(&client_lock);
861 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
863 STALE_CLIENTID(clientid_t *clid)
865 if (clid->cl_boot == boot_time)
867 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
868 clid->cl_boot, clid->cl_id, boot_time);
873 * XXX Should we use a slab cache ?
874 * This type of memory management is somewhat inefficient, but we use it
875 * anyway since SETCLIENTID is not a common operation.
877 static struct nfs4_client *alloc_client(struct xdr_netobj name)
879 struct nfs4_client *clp;
881 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
884 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
885 if (clp->cl_name.data == NULL) {
889 memcpy(clp->cl_name.data, name.data, name.len);
890 clp->cl_name.len = name.len;
895 free_client(struct nfs4_client *clp)
897 while (!list_empty(&clp->cl_sessions)) {
898 struct nfsd4_session *ses;
899 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
901 list_del(&ses->se_perclnt);
902 nfsd4_put_session(ses);
904 if (clp->cl_cred.cr_group_info)
905 put_group_info(clp->cl_cred.cr_group_info);
906 kfree(clp->cl_principal);
907 kfree(clp->cl_name.data);
912 release_session_client(struct nfsd4_session *session)
914 struct nfs4_client *clp = session->se_client;
916 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
918 if (is_client_expired(clp)) {
920 session->se_client = NULL;
922 renew_client_locked(clp);
923 spin_unlock(&client_lock);
926 /* must be called under the client_lock */
928 unhash_client_locked(struct nfs4_client *clp)
930 struct nfsd4_session *ses;
932 mark_client_expired(clp);
933 list_del(&clp->cl_lru);
934 spin_lock(&clp->cl_lock);
935 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
936 list_del_init(&ses->se_hash);
937 spin_unlock(&clp->cl_lock);
941 expire_client(struct nfs4_client *clp)
943 struct nfs4_stateowner *sop;
944 struct nfs4_delegation *dp;
945 struct list_head reaplist;
947 INIT_LIST_HEAD(&reaplist);
948 spin_lock(&recall_lock);
949 while (!list_empty(&clp->cl_delegations)) {
950 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
951 list_del_init(&dp->dl_perclnt);
952 list_move(&dp->dl_recall_lru, &reaplist);
954 spin_unlock(&recall_lock);
955 while (!list_empty(&reaplist)) {
956 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
957 list_del_init(&dp->dl_recall_lru);
958 unhash_delegation(dp);
960 while (!list_empty(&clp->cl_openowners)) {
961 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
962 release_openowner(sop);
964 nfsd4_shutdown_callback(clp);
965 if (clp->cl_cb_conn.cb_xprt)
966 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
967 list_del(&clp->cl_idhash);
968 list_del(&clp->cl_strhash);
969 spin_lock(&client_lock);
970 unhash_client_locked(clp);
971 if (atomic_read(&clp->cl_refcount) == 0)
973 spin_unlock(&client_lock);
976 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
978 memcpy(target->cl_verifier.data, source->data,
979 sizeof(target->cl_verifier.data));
982 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
984 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
985 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
988 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
990 target->cr_uid = source->cr_uid;
991 target->cr_gid = source->cr_gid;
992 target->cr_group_info = source->cr_group_info;
993 get_group_info(target->cr_group_info);
996 static int same_name(const char *n1, const char *n2)
998 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1002 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1004 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1008 same_clid(clientid_t *cl1, clientid_t *cl2)
1010 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1013 /* XXX what about NGROUP */
1015 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1017 return cr1->cr_uid == cr2->cr_uid;
1020 static void gen_clid(struct nfs4_client *clp)
1022 static u32 current_clientid = 1;
1024 clp->cl_clientid.cl_boot = boot_time;
1025 clp->cl_clientid.cl_id = current_clientid++;
1028 static void gen_confirm(struct nfs4_client *clp)
1033 p = (u32 *)clp->cl_confirm.data;
1034 *p++ = get_seconds();
1038 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1039 struct svc_rqst *rqstp, nfs4_verifier *verf)
1041 struct nfs4_client *clp;
1042 struct sockaddr *sa = svc_addr(rqstp);
1045 clp = alloc_client(name);
1049 INIT_LIST_HEAD(&clp->cl_sessions);
1051 princ = svc_gss_principal(rqstp);
1053 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1054 if (clp->cl_principal == NULL) {
1060 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1061 atomic_set(&clp->cl_refcount, 0);
1062 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1063 INIT_LIST_HEAD(&clp->cl_idhash);
1064 INIT_LIST_HEAD(&clp->cl_strhash);
1065 INIT_LIST_HEAD(&clp->cl_openowners);
1066 INIT_LIST_HEAD(&clp->cl_delegations);
1067 INIT_LIST_HEAD(&clp->cl_lru);
1068 INIT_LIST_HEAD(&clp->cl_callbacks);
1069 spin_lock_init(&clp->cl_lock);
1070 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1071 clp->cl_time = get_seconds();
1072 clear_bit(0, &clp->cl_cb_slot_busy);
1073 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1074 copy_verf(clp, verf);
1075 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1076 clp->cl_flavor = rqstp->rq_flavor;
1077 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1079 clp->cl_cb_session = NULL;
1083 static int check_name(struct xdr_netobj name)
1087 if (name.len > NFS4_OPAQUE_LIMIT) {
1088 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1095 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1097 unsigned int idhashval;
1099 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1100 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1101 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1106 move_to_confirmed(struct nfs4_client *clp)
1108 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1109 unsigned int strhashval;
1111 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1112 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1113 strhashval = clientstr_hashval(clp->cl_recdir);
1114 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1118 static struct nfs4_client *
1119 find_confirmed_client(clientid_t *clid)
1121 struct nfs4_client *clp;
1122 unsigned int idhashval = clientid_hashval(clid->cl_id);
1124 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1125 if (same_clid(&clp->cl_clientid, clid))
1131 static struct nfs4_client *
1132 find_unconfirmed_client(clientid_t *clid)
1134 struct nfs4_client *clp;
1135 unsigned int idhashval = clientid_hashval(clid->cl_id);
1137 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1138 if (same_clid(&clp->cl_clientid, clid))
1144 static bool clp_used_exchangeid(struct nfs4_client *clp)
1146 return clp->cl_exchange_flags != 0;
1149 static struct nfs4_client *
1150 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1152 struct nfs4_client *clp;
1154 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1155 if (same_name(clp->cl_recdir, dname))
1161 static struct nfs4_client *
1162 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1164 struct nfs4_client *clp;
1166 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1167 if (same_name(clp->cl_recdir, dname))
1173 static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1177 ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1178 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1181 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1182 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1188 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1190 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1191 struct sockaddr *sa = svc_addr(rqstp);
1192 u32 scopeid = rpc_get_scope_id(sa);
1193 unsigned short expected_family;
1195 /* Currently, we only support tcp and tcp6 for the callback channel */
1196 if (se->se_callback_netid_len == 3 &&
1197 !memcmp(se->se_callback_netid_val, "tcp", 3))
1198 expected_family = AF_INET;
1199 else if (se->se_callback_netid_len == 4 &&
1200 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1201 expected_family = AF_INET6;
1205 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1206 se->se_callback_addr_len,
1207 (struct sockaddr *)&conn->cb_addr,
1208 sizeof(conn->cb_addr));
1210 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1213 if (conn->cb_addr.ss_family == AF_INET6)
1214 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1216 conn->cb_prog = se->se_callback_prog;
1217 conn->cb_ident = se->se_callback_ident;
1218 rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1221 conn->cb_addr.ss_family = AF_UNSPEC;
1222 conn->cb_addrlen = 0;
1223 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1224 "will not receive delegations\n",
1225 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1231 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1234 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1236 struct nfsd4_slot *slot = resp->cstate.slot;
1239 dprintk("--> %s slot %p\n", __func__, slot);
1241 slot->sl_opcnt = resp->opcnt;
1242 slot->sl_status = resp->cstate.status;
1244 if (nfsd4_not_cached(resp)) {
1245 slot->sl_datalen = 0;
1248 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1249 base = (char *)resp->cstate.datap -
1250 (char *)resp->xbuf->head[0].iov_base;
1251 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1253 WARN("%s: sessions DRC could not cache compound\n", __func__);
1258 * Encode the replay sequence operation from the slot values.
1259 * If cachethis is FALSE encode the uncached rep error on the next
1260 * operation which sets resp->p and increments resp->opcnt for
1261 * nfs4svc_encode_compoundres.
1265 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1266 struct nfsd4_compoundres *resp)
1268 struct nfsd4_op *op;
1269 struct nfsd4_slot *slot = resp->cstate.slot;
1271 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1272 resp->opcnt, resp->cstate.slot->sl_cachethis);
1274 /* Encode the replayed sequence operation */
1275 op = &args->ops[resp->opcnt - 1];
1276 nfsd4_encode_operation(resp, op);
1278 /* Return nfserr_retry_uncached_rep in next operation. */
1279 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1280 op = &args->ops[resp->opcnt++];
1281 op->status = nfserr_retry_uncached_rep;
1282 nfsd4_encode_operation(resp, op);
1288 * The sequence operation is not cached because we can use the slot and
1292 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1293 struct nfsd4_sequence *seq)
1295 struct nfsd4_slot *slot = resp->cstate.slot;
1298 dprintk("--> %s slot %p\n", __func__, slot);
1300 /* Either returns 0 or nfserr_retry_uncached */
1301 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1302 if (status == nfserr_retry_uncached_rep)
1305 /* The sequence operation has been encoded, cstate->datap set. */
1306 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1308 resp->opcnt = slot->sl_opcnt;
1309 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1310 status = slot->sl_status;
1316 * Set the exchange_id flags returned by the server.
1319 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1321 /* pNFS is not supported */
1322 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1324 /* Referrals are supported, Migration is not. */
1325 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1327 /* set the wire flags to return to client. */
1328 clid->flags = new->cl_exchange_flags;
1332 nfsd4_exchange_id(struct svc_rqst *rqstp,
1333 struct nfsd4_compound_state *cstate,
1334 struct nfsd4_exchange_id *exid)
1336 struct nfs4_client *unconf, *conf, *new;
1338 unsigned int strhashval;
1339 char dname[HEXDIR_LEN];
1340 char addr_str[INET6_ADDRSTRLEN];
1341 nfs4_verifier verf = exid->verifier;
1342 struct sockaddr *sa = svc_addr(rqstp);
1344 rpc_ntop(sa, addr_str, sizeof(addr_str));
1345 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1346 "ip_addr=%s flags %x, spa_how %d\n",
1347 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1348 addr_str, exid->flags, exid->spa_how);
1350 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1351 return nfserr_inval;
1353 /* Currently only support SP4_NONE */
1354 switch (exid->spa_how) {
1358 return nfserr_serverfault;
1360 BUG(); /* checked by xdr code */
1362 return nfserr_serverfault; /* no excuse :-/ */
1365 status = nfs4_make_rec_clidname(dname, &exid->clname);
1370 strhashval = clientstr_hashval(dname);
1375 conf = find_confirmed_client_by_str(dname, strhashval);
1377 if (!clp_used_exchangeid(conf)) {
1378 status = nfserr_clid_inuse; /* XXX: ? */
1381 if (!same_verf(&verf, &conf->cl_verifier)) {
1382 /* 18.35.4 case 8 */
1383 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1384 status = nfserr_not_same;
1387 /* Client reboot: destroy old state */
1388 expire_client(conf);
1391 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1392 /* 18.35.4 case 9 */
1393 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1394 status = nfserr_perm;
1397 expire_client(conf);
1401 * Set bit when the owner id and verifier map to an already
1402 * confirmed client id (18.35.3).
1404 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1407 * Falling into 18.35.4 case 2, possible router replay.
1408 * Leave confirmed record intact and return same result.
1410 copy_verf(conf, &verf);
1415 /* 18.35.4 case 7 */
1416 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1417 status = nfserr_noent;
1421 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1424 * Possible retry or client restart. Per 18.35.4 case 4,
1425 * a new unconfirmed record should be generated regardless
1426 * of whether any properties have changed.
1428 expire_client(unconf);
1433 new = create_client(exid->clname, dname, rqstp, &verf);
1435 status = nfserr_jukebox;
1440 add_to_unconfirmed(new, strhashval);
1442 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1443 exid->clientid.cl_id = new->cl_clientid.cl_id;
1446 nfsd4_set_ex_flags(new, exid);
1448 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1449 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1453 nfs4_unlock_state();
1455 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1460 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1462 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1465 /* The slot is in use, and no response has been sent. */
1467 if (seqid == slot_seqid)
1468 return nfserr_jukebox;
1470 return nfserr_seq_misordered;
1473 if (likely(seqid == slot_seqid + 1))
1476 if (seqid == slot_seqid)
1477 return nfserr_replay_cache;
1479 if (seqid == 1 && (slot_seqid + 1) == 0)
1481 /* Misordered replay or misordered new request */
1482 return nfserr_seq_misordered;
1486 * Cache the create session result into the create session single DRC
1487 * slot cache by saving the xdr structure. sl_seqid has been set.
1488 * Do this for solo or embedded create session operations.
1491 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1492 struct nfsd4_clid_slot *slot, int nfserr)
1494 slot->sl_status = nfserr;
1495 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1499 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1500 struct nfsd4_clid_slot *slot)
1502 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1503 return slot->sl_status;
1507 nfsd4_create_session(struct svc_rqst *rqstp,
1508 struct nfsd4_compound_state *cstate,
1509 struct nfsd4_create_session *cr_ses)
1511 struct sockaddr *sa = svc_addr(rqstp);
1512 struct nfs4_client *conf, *unconf;
1513 struct nfsd4_session *new;
1514 struct nfsd4_clid_slot *cs_slot = NULL;
1515 bool confirm_me = false;
1518 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1519 return nfserr_inval;
1522 unconf = find_unconfirmed_client(&cr_ses->clientid);
1523 conf = find_confirmed_client(&cr_ses->clientid);
1526 cs_slot = &conf->cl_cs_slot;
1527 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1528 if (status == nfserr_replay_cache) {
1529 dprintk("Got a create_session replay! seqid= %d\n",
1531 /* Return the cached reply status */
1532 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1534 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1535 status = nfserr_seq_misordered;
1536 dprintk("Sequence misordered!\n");
1537 dprintk("Expected seqid= %d but got seqid= %d\n",
1538 cs_slot->sl_seqid, cr_ses->seqid);
1541 } else if (unconf) {
1542 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1543 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1544 status = nfserr_clid_inuse;
1548 cs_slot = &unconf->cl_cs_slot;
1549 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1551 /* an unconfirmed replay returns misordered */
1552 status = nfserr_seq_misordered;
1559 status = nfserr_stale_clientid;
1564 * XXX: we should probably set this at creation time, and check
1565 * for consistent minorversion use throughout:
1567 conf->cl_minorversion = 1;
1569 * We do not support RDMA or persistent sessions
1571 cr_ses->flags &= ~SESSION4_PERSIST;
1572 cr_ses->flags &= ~SESSION4_RDMA;
1574 status = nfserr_jukebox;
1575 new = alloc_init_session(rqstp, conf, cr_ses);
1579 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1580 NFS4_MAX_SESSIONID_LEN);
1581 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1582 sizeof(struct nfsd4_channel_attrs));
1583 cs_slot->sl_seqid++;
1584 cr_ses->seqid = cs_slot->sl_seqid;
1586 /* cache solo and embedded create sessions under the state lock */
1587 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1589 move_to_confirmed(conf);
1591 nfs4_unlock_state();
1592 dprintk("%s returns %d\n", __func__, ntohl(status));
1596 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1598 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1599 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1601 return argp->opcnt == resp->opcnt;
1604 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1607 case NFS4_CDFC4_FORE:
1608 case NFS4_CDFC4_BACK:
1610 case NFS4_CDFC4_FORE_OR_BOTH:
1611 case NFS4_CDFC4_BACK_OR_BOTH:
1612 *dir = NFS4_CDFC4_BOTH;
1615 return nfserr_inval;
1618 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1619 struct nfsd4_compound_state *cstate,
1620 struct nfsd4_bind_conn_to_session *bcts)
1624 if (!nfsd4_last_compound_op(rqstp))
1625 return nfserr_not_only_op;
1626 spin_lock(&client_lock);
1627 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1628 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1629 * client_lock iself: */
1630 if (cstate->session) {
1631 nfsd4_get_session(cstate->session);
1632 atomic_inc(&cstate->session->se_client->cl_refcount);
1634 spin_unlock(&client_lock);
1635 if (!cstate->session)
1636 return nfserr_badsession;
1638 status = nfsd4_map_bcts_dir(&bcts->dir);
1639 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1643 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1647 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1651 nfsd4_destroy_session(struct svc_rqst *r,
1652 struct nfsd4_compound_state *cstate,
1653 struct nfsd4_destroy_session *sessionid)
1655 struct nfsd4_session *ses;
1656 u32 status = nfserr_badsession;
1659 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1660 * - Should we return nfserr_back_chan_busy if waiting for
1661 * callbacks on to-be-destroyed session?
1662 * - Do we need to clear any callback info from previous session?
1665 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1666 if (!nfsd4_last_compound_op(r))
1667 return nfserr_not_only_op;
1669 dump_sessionid(__func__, &sessionid->sessionid);
1670 spin_lock(&client_lock);
1671 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1673 spin_unlock(&client_lock);
1677 unhash_session(ses);
1678 spin_unlock(&client_lock);
1681 nfsd4_probe_callback_sync(ses->se_client);
1682 nfs4_unlock_state();
1684 nfsd4_del_conns(ses);
1686 nfsd4_put_session(ses);
1689 dprintk("%s returns %d\n", __func__, ntohl(status));
1693 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1695 struct nfsd4_conn *c;
1697 list_for_each_entry(c, &s->se_conns, cn_persession) {
1698 if (c->cn_xprt == xpt) {
1705 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1707 struct nfs4_client *clp = ses->se_client;
1708 struct nfsd4_conn *c;
1711 spin_lock(&clp->cl_lock);
1712 c = __nfsd4_find_conn(new->cn_xprt, ses);
1714 spin_unlock(&clp->cl_lock);
1718 __nfsd4_hash_conn(new, ses);
1719 spin_unlock(&clp->cl_lock);
1720 ret = nfsd4_register_conn(new);
1722 /* oops; xprt is already down: */
1723 nfsd4_conn_lost(&new->cn_xpt_user);
1727 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1729 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1731 return args->opcnt > session->se_fchannel.maxops;
1735 nfsd4_sequence(struct svc_rqst *rqstp,
1736 struct nfsd4_compound_state *cstate,
1737 struct nfsd4_sequence *seq)
1739 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1740 struct nfsd4_session *session;
1741 struct nfsd4_slot *slot;
1742 struct nfsd4_conn *conn;
1745 if (resp->opcnt != 1)
1746 return nfserr_sequence_pos;
1749 * Will be either used or freed by nfsd4_sequence_check_conn
1752 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1754 return nfserr_jukebox;
1756 spin_lock(&client_lock);
1757 status = nfserr_badsession;
1758 session = find_in_sessionid_hashtbl(&seq->sessionid);
1762 status = nfserr_too_many_ops;
1763 if (nfsd4_session_too_many_ops(rqstp, session))
1766 status = nfserr_badslot;
1767 if (seq->slotid >= session->se_fchannel.maxreqs)
1770 slot = session->se_slots[seq->slotid];
1771 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1773 /* We do not negotiate the number of slots yet, so set the
1774 * maxslots to the session maxreqs which is used to encode
1775 * sr_highest_slotid and the sr_target_slot id to maxslots */
1776 seq->maxslots = session->se_fchannel.maxreqs;
1778 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1779 if (status == nfserr_replay_cache) {
1780 cstate->slot = slot;
1781 cstate->session = session;
1782 /* Return the cached reply status and set cstate->status
1783 * for nfsd4_proc_compound processing */
1784 status = nfsd4_replay_cache_entry(resp, seq);
1785 cstate->status = nfserr_replay_cache;
1791 nfsd4_sequence_check_conn(conn, session);
1794 /* Success! bump slot seqid */
1795 slot->sl_inuse = true;
1796 slot->sl_seqid = seq->seqid;
1797 slot->sl_cachethis = seq->cachethis;
1799 cstate->slot = slot;
1800 cstate->session = session;
1803 /* Hold a session reference until done processing the compound. */
1804 if (cstate->session) {
1805 struct nfs4_client *clp = session->se_client;
1807 nfsd4_get_session(cstate->session);
1808 atomic_inc(&clp->cl_refcount);
1809 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1810 seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1813 spin_unlock(&client_lock);
1814 dprintk("%s: return %d\n", __func__, ntohl(status));
1819 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1823 if (rc->rca_one_fs) {
1824 if (!cstate->current_fh.fh_dentry)
1825 return nfserr_nofilehandle;
1827 * We don't take advantage of the rca_one_fs case.
1828 * That's OK, it's optional, we can safely ignore it.
1834 status = nfserr_complete_already;
1835 if (cstate->session->se_client->cl_firststate)
1838 status = nfserr_stale_clientid;
1839 if (is_client_expired(cstate->session->se_client))
1841 * The following error isn't really legal.
1842 * But we only get here if the client just explicitly
1843 * destroyed the client. Surely it no longer cares what
1844 * error it gets back on an operation for the dead
1850 nfsd4_create_clid_dir(cstate->session->se_client);
1852 nfs4_unlock_state();
1857 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1858 struct nfsd4_setclientid *setclid)
1860 struct xdr_netobj clname = {
1861 .len = setclid->se_namelen,
1862 .data = setclid->se_name,
1864 nfs4_verifier clverifier = setclid->se_verf;
1865 unsigned int strhashval;
1866 struct nfs4_client *conf, *unconf, *new;
1868 char dname[HEXDIR_LEN];
1870 if (!check_name(clname))
1871 return nfserr_inval;
1873 status = nfs4_make_rec_clidname(dname, &clname);
1878 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1879 * We get here on a DRC miss.
1882 strhashval = clientstr_hashval(dname);
1885 conf = find_confirmed_client_by_str(dname, strhashval);
1887 /* RFC 3530 14.2.33 CASE 0: */
1888 status = nfserr_clid_inuse;
1889 if (clp_used_exchangeid(conf))
1891 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1892 char addr_str[INET6_ADDRSTRLEN];
1893 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1895 dprintk("NFSD: setclientid: string in use by client "
1896 "at %s\n", addr_str);
1901 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1902 * has a description of SETCLIENTID request processing consisting
1903 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1905 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1906 status = nfserr_resource;
1909 * RFC 3530 14.2.33 CASE 4:
1910 * placed first, because it is the normal case
1913 expire_client(unconf);
1914 new = create_client(clname, dname, rqstp, &clverifier);
1918 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1920 * RFC 3530 14.2.33 CASE 1:
1921 * probable callback update
1924 /* Note this is removing unconfirmed {*x***},
1925 * which is stronger than RFC recommended {vxc**}.
1926 * This has the advantage that there is at most
1927 * one {*x***} in either list at any time.
1929 expire_client(unconf);
1931 new = create_client(clname, dname, rqstp, &clverifier);
1934 copy_clid(new, conf);
1935 } else if (!unconf) {
1937 * RFC 3530 14.2.33 CASE 2:
1938 * probable client reboot; state will be removed if
1941 new = create_client(clname, dname, rqstp, &clverifier);
1947 * RFC 3530 14.2.33 CASE 3:
1948 * probable client reboot; state will be removed if
1951 expire_client(unconf);
1952 new = create_client(clname, dname, rqstp, &clverifier);
1958 * XXX: we should probably set this at creation time, and check
1959 * for consistent minorversion use throughout:
1961 new->cl_minorversion = 0;
1962 gen_callback(new, setclid, rqstp);
1963 add_to_unconfirmed(new, strhashval);
1964 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1965 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1966 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1969 nfs4_unlock_state();
1975 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1976 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1977 * bullets, labeled as CASE1 - CASE4 below.
1980 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1981 struct nfsd4_compound_state *cstate,
1982 struct nfsd4_setclientid_confirm *setclientid_confirm)
1984 struct sockaddr *sa = svc_addr(rqstp);
1985 struct nfs4_client *conf, *unconf;
1986 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1987 clientid_t * clid = &setclientid_confirm->sc_clientid;
1990 if (STALE_CLIENTID(clid))
1991 return nfserr_stale_clientid;
1993 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1994 * We get here on a DRC miss.
1999 conf = find_confirmed_client(clid);
2000 unconf = find_unconfirmed_client(clid);
2002 status = nfserr_clid_inuse;
2003 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2005 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2009 * section 14.2.34 of RFC 3530 has a description of
2010 * SETCLIENTID_CONFIRM request processing consisting
2011 * of 4 bullet points, labeled as CASE1 - CASE4 below.
2013 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2015 * RFC 3530 14.2.34 CASE 1:
2018 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2019 status = nfserr_clid_inuse;
2021 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2022 nfsd4_probe_callback(conf);
2023 expire_client(unconf);
2027 } else if (conf && !unconf) {
2029 * RFC 3530 14.2.34 CASE 2:
2030 * probable retransmitted request; play it safe and
2033 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2034 status = nfserr_clid_inuse;
2037 } else if (!conf && unconf
2038 && same_verf(&unconf->cl_confirm, &confirm)) {
2040 * RFC 3530 14.2.34 CASE 3:
2041 * Normal case; new or rebooted client:
2043 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2044 status = nfserr_clid_inuse;
2047 clientstr_hashval(unconf->cl_recdir);
2048 conf = find_confirmed_client_by_str(unconf->cl_recdir,
2051 nfsd4_remove_clid_dir(conf);
2052 expire_client(conf);
2054 move_to_confirmed(unconf);
2056 nfsd4_probe_callback(conf);
2059 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2060 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2063 * RFC 3530 14.2.34 CASE 4:
2064 * Client probably hasn't noticed that we rebooted yet.
2066 status = nfserr_stale_clientid;
2068 /* check that we have hit one of the cases...*/
2069 status = nfserr_clid_inuse;
2072 nfs4_unlock_state();
2076 /* OPEN Share state helper functions */
2077 static inline struct nfs4_file *
2078 alloc_init_file(struct inode *ino)
2080 struct nfs4_file *fp;
2081 unsigned int hashval = file_hashval(ino);
2083 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2085 atomic_set(&fp->fi_ref, 1);
2086 INIT_LIST_HEAD(&fp->fi_hash);
2087 INIT_LIST_HEAD(&fp->fi_stateids);
2088 INIT_LIST_HEAD(&fp->fi_delegations);
2089 fp->fi_inode = igrab(ino);
2090 fp->fi_id = current_fileid++;
2091 fp->fi_had_conflict = false;
2092 fp->fi_lease = NULL;
2093 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2094 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2095 spin_lock(&recall_lock);
2096 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2097 spin_unlock(&recall_lock);
2104 nfsd4_free_slab(struct kmem_cache **slab)
2108 kmem_cache_destroy(*slab);
2113 nfsd4_free_slabs(void)
2115 nfsd4_free_slab(&stateowner_slab);
2116 nfsd4_free_slab(&file_slab);
2117 nfsd4_free_slab(&stateid_slab);
2118 nfsd4_free_slab(&deleg_slab);
2122 nfsd4_init_slabs(void)
2124 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2125 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2126 if (stateowner_slab == NULL)
2128 file_slab = kmem_cache_create("nfsd4_files",
2129 sizeof(struct nfs4_file), 0, 0, NULL);
2130 if (file_slab == NULL)
2132 stateid_slab = kmem_cache_create("nfsd4_stateids",
2133 sizeof(struct nfs4_stateid), 0, 0, NULL);
2134 if (stateid_slab == NULL)
2136 deleg_slab = kmem_cache_create("nfsd4_delegations",
2137 sizeof(struct nfs4_delegation), 0, 0, NULL);
2138 if (deleg_slab == NULL)
2143 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2148 nfs4_free_stateowner(struct kref *kref)
2150 struct nfs4_stateowner *sop =
2151 container_of(kref, struct nfs4_stateowner, so_ref);
2152 kfree(sop->so_owner.data);
2153 kmem_cache_free(stateowner_slab, sop);
2156 static inline struct nfs4_stateowner *
2157 alloc_stateowner(struct xdr_netobj *owner)
2159 struct nfs4_stateowner *sop;
2161 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2162 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2163 memcpy(sop->so_owner.data, owner->data, owner->len);
2164 sop->so_owner.len = owner->len;
2165 kref_init(&sop->so_ref);
2168 kmem_cache_free(stateowner_slab, sop);
2173 static struct nfs4_stateowner *
2174 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2175 struct nfs4_stateowner *sop;
2176 struct nfs4_replay *rp;
2177 unsigned int idhashval;
2179 if (!(sop = alloc_stateowner(&open->op_owner)))
2181 idhashval = ownerid_hashval(current_ownerid);
2182 INIT_LIST_HEAD(&sop->so_idhash);
2183 INIT_LIST_HEAD(&sop->so_strhash);
2184 INIT_LIST_HEAD(&sop->so_perclient);
2185 INIT_LIST_HEAD(&sop->so_stateids);
2186 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2187 INIT_LIST_HEAD(&sop->so_close_lru);
2189 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2190 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2191 list_add(&sop->so_perclient, &clp->cl_openowners);
2192 sop->so_is_open_owner = 1;
2193 sop->so_id = current_ownerid++;
2194 sop->so_client = clp;
2195 sop->so_seqid = open->op_seqid;
2196 sop->so_confirmed = 0;
2197 rp = &sop->so_replay;
2198 rp->rp_status = nfserr_serverfault;
2200 rp->rp_buf = rp->rp_ibuf;
2205 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2206 struct nfs4_stateowner *sop = open->op_stateowner;
2207 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2209 INIT_LIST_HEAD(&stp->st_hash);
2210 INIT_LIST_HEAD(&stp->st_perstateowner);
2211 INIT_LIST_HEAD(&stp->st_lockowners);
2212 INIT_LIST_HEAD(&stp->st_perfile);
2213 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2214 list_add(&stp->st_perstateowner, &sop->so_stateids);
2215 list_add(&stp->st_perfile, &fp->fi_stateids);
2216 stp->st_stateowner = sop;
2219 stp->st_stateid.si_boot = boot_time;
2220 stp->st_stateid.si_stateownerid = sop->so_id;
2221 stp->st_stateid.si_fileid = fp->fi_id;
2222 stp->st_stateid.si_generation = 0;
2223 stp->st_access_bmap = 0;
2224 stp->st_deny_bmap = 0;
2225 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2226 &stp->st_access_bmap);
2227 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2228 stp->st_openstp = NULL;
2232 move_to_close_lru(struct nfs4_stateowner *sop)
2234 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2236 list_move_tail(&sop->so_close_lru, &close_lru);
2237 sop->so_time = get_seconds();
2241 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2244 return (sop->so_owner.len == owner->len) &&
2245 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2246 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2249 static struct nfs4_stateowner *
2250 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2252 struct nfs4_stateowner *so = NULL;
2254 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2255 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2261 /* search file_hashtbl[] for file */
2262 static struct nfs4_file *
2263 find_file(struct inode *ino)
2265 unsigned int hashval = file_hashval(ino);
2266 struct nfs4_file *fp;
2268 spin_lock(&recall_lock);
2269 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2270 if (fp->fi_inode == ino) {
2272 spin_unlock(&recall_lock);
2276 spin_unlock(&recall_lock);
2280 static inline int access_valid(u32 x, u32 minorversion)
2282 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2284 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2286 x &= ~NFS4_SHARE_ACCESS_MASK;
2287 if (minorversion && x) {
2288 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2290 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2292 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2299 static inline int deny_valid(u32 x)
2301 /* Note: unlike access bits, deny bits may be zero. */
2302 return x <= NFS4_SHARE_DENY_BOTH;
2306 * Called to check deny when READ with all zero stateid or
2307 * WRITE with all zero or all one stateid
2310 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2312 struct inode *ino = current_fh->fh_dentry->d_inode;
2313 struct nfs4_file *fp;
2314 struct nfs4_stateid *stp;
2317 dprintk("NFSD: nfs4_share_conflict\n");
2319 fp = find_file(ino);
2322 ret = nfserr_locked;
2323 /* Search for conflicting share reservations */
2324 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2325 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2326 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2336 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2338 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2339 nfs4_file_put_access(fp, O_WRONLY);
2340 if (share_access & NFS4_SHARE_ACCESS_READ)
2341 nfs4_file_put_access(fp, O_RDONLY);
2344 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
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 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2355 /* only place dl_time is set. protected by lock_flocks*/
2356 dp->dl_time = get_seconds();
2358 nfsd4_cb_recall(dp);
2361 /* Called from break_lease() with lock_flocks() held. */
2362 static void nfsd_break_deleg_cb(struct file_lock *fl)
2364 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2365 struct nfs4_delegation *dp;
2368 /* We assume break_lease is only called once per lease: */
2369 BUG_ON(fp->fi_had_conflict);
2371 * We don't want the locks code to timeout the lease for us;
2372 * we'll remove it ourself if a delegation isn't returned
2375 fl->fl_break_time = 0;
2377 spin_lock(&recall_lock);
2378 fp->fi_had_conflict = true;
2379 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2380 nfsd_break_one_deleg(dp);
2381 spin_unlock(&recall_lock);
2385 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2388 return lease_modify(onlist, arg);
2393 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2394 .fl_break = nfsd_break_deleg_cb,
2395 .fl_change = nfsd_change_deleg_cb,
2400 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2401 struct nfsd4_open *open)
2403 clientid_t *clientid = &open->op_clientid;
2404 struct nfs4_client *clp = NULL;
2405 unsigned int strhashval;
2406 struct nfs4_stateowner *sop = NULL;
2408 if (!check_name(open->op_owner))
2409 return nfserr_inval;
2411 if (STALE_CLIENTID(&open->op_clientid))
2412 return nfserr_stale_clientid;
2414 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2415 sop = find_openstateowner_str(strhashval, open);
2416 open->op_stateowner = sop;
2418 /* Make sure the client's lease hasn't expired. */
2419 clp = find_confirmed_client(clientid);
2421 return nfserr_expired;
2424 /* When sessions are used, skip open sequenceid processing */
2425 if (nfsd4_has_session(cstate))
2427 if (!sop->so_confirmed) {
2428 /* Replace unconfirmed owners without checking for replay. */
2429 clp = sop->so_client;
2430 release_openowner(sop);
2431 open->op_stateowner = NULL;
2434 if (open->op_seqid == sop->so_seqid - 1) {
2435 if (sop->so_replay.rp_buflen)
2436 return nfserr_replay_me;
2437 /* The original OPEN failed so spectacularly
2438 * that we don't even have replay data saved!
2439 * Therefore, we have no choice but to continue
2440 * processing this OPEN; presumably, we'll
2441 * fail again for the same reason.
2443 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2446 if (open->op_seqid != sop->so_seqid)
2447 return nfserr_bad_seqid;
2449 if (open->op_stateowner == NULL) {
2450 sop = alloc_init_open_stateowner(strhashval, clp, open);
2452 return nfserr_resource;
2453 open->op_stateowner = sop;
2455 list_del_init(&sop->so_close_lru);
2456 renew_client(sop->so_client);
2460 static inline __be32
2461 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2463 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2464 return nfserr_openmode;
2469 static struct nfs4_delegation *
2470 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2472 struct nfs4_delegation *dp;
2474 spin_lock(&recall_lock);
2475 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2476 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid) {
2477 spin_unlock(&recall_lock);
2480 spin_unlock(&recall_lock);
2484 int share_access_to_flags(u32 share_access)
2486 share_access &= ~NFS4_SHARE_WANT_MASK;
2488 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2492 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2493 struct nfs4_delegation **dp)
2496 __be32 status = nfserr_bad_stateid;
2498 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2501 flags = share_access_to_flags(open->op_share_access);
2502 status = nfs4_check_delegmode(*dp, flags);
2506 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2510 open->op_stateowner->so_confirmed = 1;
2515 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2517 struct nfs4_stateid *local;
2518 __be32 status = nfserr_share_denied;
2519 struct nfs4_stateowner *sop = open->op_stateowner;
2521 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2522 /* ignore lock owners */
2523 if (local->st_stateowner->so_is_open_owner == 0)
2525 /* remember if we have seen this open owner */
2526 if (local->st_stateowner == sop)
2528 /* check for conflicting share reservations */
2529 if (!test_share(local, open))
2537 static inline struct nfs4_stateid *
2538 nfs4_alloc_stateid(void)
2540 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2543 static inline int nfs4_access_to_access(u32 nfs4_access)
2547 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2548 flags |= NFSD_MAY_READ;
2549 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2550 flags |= NFSD_MAY_WRITE;
2554 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2555 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2558 int oflag = nfs4_access_to_omode(nfs4_access);
2559 int access = nfs4_access_to_access(nfs4_access);
2561 if (!fp->fi_fds[oflag]) {
2562 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2563 &fp->fi_fds[oflag]);
2567 nfs4_file_get_access(fp, oflag);
2573 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2574 struct nfs4_file *fp, struct svc_fh *cur_fh,
2575 struct nfsd4_open *open)
2577 struct nfs4_stateid *stp;
2580 stp = nfs4_alloc_stateid();
2582 return nfserr_resource;
2584 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2586 kmem_cache_free(stateid_slab, stp);
2593 static inline __be32
2594 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2595 struct nfsd4_open *open)
2597 struct iattr iattr = {
2598 .ia_valid = ATTR_SIZE,
2601 if (!open->op_truncate)
2603 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2604 return nfserr_inval;
2605 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2609 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2611 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2615 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2617 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2621 status = nfsd4_truncate(rqstp, cur_fh, open);
2624 int oflag = nfs4_access_to_omode(new_access);
2625 nfs4_file_put_access(fp, oflag);
2629 /* remember the open */
2630 __set_bit(op_share_access, &stp->st_access_bmap);
2631 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2638 nfs4_set_claim_prev(struct nfsd4_open *open)
2640 open->op_stateowner->so_confirmed = 1;
2641 open->op_stateowner->so_client->cl_firststate = 1;
2644 /* Should we give out recallable state?: */
2645 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2647 if (clp->cl_cb_state == NFSD4_CB_UP)
2650 * In the sessions case, since we don't have to establish a
2651 * separate connection for callbacks, we assume it's OK
2652 * until we hear otherwise:
2654 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2657 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2659 struct file_lock *fl;
2661 fl = locks_alloc_lock();
2664 locks_init_lock(fl);
2665 fl->fl_lmops = &nfsd_lease_mng_ops;
2666 fl->fl_flags = FL_LEASE;
2667 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2668 fl->fl_end = OFFSET_MAX;
2669 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2670 fl->fl_pid = current->tgid;
2674 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2676 struct nfs4_file *fp = dp->dl_file;
2677 struct file_lock *fl;
2680 fl = nfs4_alloc_init_lease(dp, flag);
2683 fl->fl_file = find_readable_file(fp);
2684 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2685 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2687 list_del_init(&dp->dl_perclnt);
2688 locks_free_lock(fl);
2692 fp->fi_deleg_file = fl->fl_file;
2693 get_file(fp->fi_deleg_file);
2694 atomic_set(&fp->fi_delegees, 1);
2695 list_add(&dp->dl_perfile, &fp->fi_delegations);
2699 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2701 struct nfs4_file *fp = dp->dl_file;
2704 return nfs4_setlease(dp, flag);
2705 spin_lock(&recall_lock);
2706 if (fp->fi_had_conflict) {
2707 spin_unlock(&recall_lock);
2710 atomic_inc(&fp->fi_delegees);
2711 list_add(&dp->dl_perfile, &fp->fi_delegations);
2712 spin_unlock(&recall_lock);
2713 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2718 * Attempt to hand out a delegation.
2721 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2723 struct nfs4_delegation *dp;
2724 struct nfs4_stateowner *sop = stp->st_stateowner;
2726 int status, flag = 0;
2728 cb_up = nfsd4_cb_channel_good(sop->so_client);
2729 flag = NFS4_OPEN_DELEGATE_NONE;
2730 open->op_recall = 0;
2731 switch (open->op_claim_type) {
2732 case NFS4_OPEN_CLAIM_PREVIOUS:
2734 open->op_recall = 1;
2735 flag = open->op_delegate_type;
2736 if (flag == NFS4_OPEN_DELEGATE_NONE)
2739 case NFS4_OPEN_CLAIM_NULL:
2740 /* Let's not give out any delegations till everyone's
2741 * had the chance to reclaim theirs.... */
2742 if (locks_in_grace())
2744 if (!cb_up || !sop->so_confirmed)
2746 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2747 flag = NFS4_OPEN_DELEGATE_WRITE;
2749 flag = NFS4_OPEN_DELEGATE_READ;
2755 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2758 status = nfs4_set_delegation(dp, flag);
2762 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2764 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2765 STATEID_VAL(&dp->dl_stateid));
2767 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2768 && flag == NFS4_OPEN_DELEGATE_NONE
2769 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2770 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2771 open->op_delegate_type = flag;
2774 nfs4_put_delegation(dp);
2776 flag = NFS4_OPEN_DELEGATE_NONE;
2781 * called with nfs4_lock_state() held.
2784 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2786 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2787 struct nfs4_file *fp = NULL;
2788 struct inode *ino = current_fh->fh_dentry->d_inode;
2789 struct nfs4_stateid *stp = NULL;
2790 struct nfs4_delegation *dp = NULL;
2793 status = nfserr_inval;
2794 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2795 || !deny_valid(open->op_share_deny))
2798 * Lookup file; if found, lookup stateid and check open request,
2799 * and check for delegations in the process of being recalled.
2800 * If not found, create the nfs4_file struct
2802 fp = find_file(ino);
2804 if ((status = nfs4_check_open(fp, open, &stp)))
2806 status = nfs4_check_deleg(fp, open, &dp);
2810 status = nfserr_bad_stateid;
2811 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2813 status = nfserr_resource;
2814 fp = alloc_init_file(ino);
2820 * OPEN the file, or upgrade an existing OPEN.
2821 * If truncate fails, the OPEN fails.
2824 /* Stateid was found, this is an OPEN upgrade */
2825 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2828 update_stateid(&stp->st_stateid);
2830 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2833 init_stateid(stp, fp, open);
2834 status = nfsd4_truncate(rqstp, current_fh, open);
2836 release_open_stateid(stp);
2839 if (nfsd4_has_session(&resp->cstate))
2840 update_stateid(&stp->st_stateid);
2842 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2844 if (nfsd4_has_session(&resp->cstate))
2845 open->op_stateowner->so_confirmed = 1;
2848 * Attempt to hand out a delegation. No error return, because the
2849 * OPEN succeeds even if we fail.
2851 nfs4_open_delegation(current_fh, open, stp);
2855 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2856 STATEID_VAL(&stp->st_stateid));
2860 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2861 nfs4_set_claim_prev(open);
2863 * To finish the open response, we just need to set the rflags.
2865 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2866 if (!open->op_stateowner->so_confirmed &&
2867 !nfsd4_has_session(&resp->cstate))
2868 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2874 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2877 struct nfs4_client *clp;
2881 dprintk("process_renew(%08x/%08x): starting\n",
2882 clid->cl_boot, clid->cl_id);
2883 status = nfserr_stale_clientid;
2884 if (STALE_CLIENTID(clid))
2886 clp = find_confirmed_client(clid);
2887 status = nfserr_expired;
2889 /* We assume the client took too long to RENEW. */
2890 dprintk("nfsd4_renew: clientid not found!\n");
2894 status = nfserr_cb_path_down;
2895 if (!list_empty(&clp->cl_delegations)
2896 && clp->cl_cb_state != NFSD4_CB_UP)
2900 nfs4_unlock_state();
2904 struct lock_manager nfsd4_manager = {
2908 nfsd4_end_grace(void)
2910 dprintk("NFSD: end of grace period\n");
2911 nfsd4_recdir_purge_old();
2912 locks_end_grace(&nfsd4_manager);
2914 * Now that every NFSv4 client has had the chance to recover and
2915 * to see the (possibly new, possibly shorter) lease time, we
2916 * can safely set the next grace time to the current lease time:
2918 nfsd4_grace = nfsd4_lease;
2922 nfs4_laundromat(void)
2924 struct nfs4_client *clp;
2925 struct nfs4_stateowner *sop;
2926 struct nfs4_delegation *dp;
2927 struct list_head *pos, *next, reaplist;
2928 time_t cutoff = get_seconds() - nfsd4_lease;
2929 time_t t, clientid_val = nfsd4_lease;
2930 time_t u, test_val = nfsd4_lease;
2934 dprintk("NFSD: laundromat service - starting\n");
2935 if (locks_in_grace())
2937 INIT_LIST_HEAD(&reaplist);
2938 spin_lock(&client_lock);
2939 list_for_each_safe(pos, next, &client_lru) {
2940 clp = list_entry(pos, struct nfs4_client, cl_lru);
2941 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2942 t = clp->cl_time - cutoff;
2943 if (clientid_val > t)
2947 if (atomic_read(&clp->cl_refcount)) {
2948 dprintk("NFSD: client in use (clientid %08x)\n",
2949 clp->cl_clientid.cl_id);
2952 unhash_client_locked(clp);
2953 list_add(&clp->cl_lru, &reaplist);
2955 spin_unlock(&client_lock);
2956 list_for_each_safe(pos, next, &reaplist) {
2957 clp = list_entry(pos, struct nfs4_client, cl_lru);
2958 dprintk("NFSD: purging unused client (clientid %08x)\n",
2959 clp->cl_clientid.cl_id);
2960 nfsd4_remove_clid_dir(clp);
2963 spin_lock(&recall_lock);
2964 list_for_each_safe(pos, next, &del_recall_lru) {
2965 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2966 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2967 u = dp->dl_time - cutoff;
2972 list_move(&dp->dl_recall_lru, &reaplist);
2974 spin_unlock(&recall_lock);
2975 list_for_each_safe(pos, next, &reaplist) {
2976 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2977 list_del_init(&dp->dl_recall_lru);
2978 unhash_delegation(dp);
2980 test_val = nfsd4_lease;
2981 list_for_each_safe(pos, next, &close_lru) {
2982 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2983 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2984 u = sop->so_time - cutoff;
2989 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2991 release_openowner(sop);
2993 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2994 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2995 nfs4_unlock_state();
2996 return clientid_val;
2999 static struct workqueue_struct *laundry_wq;
3000 static void laundromat_main(struct work_struct *);
3001 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3004 laundromat_main(struct work_struct *not_used)
3008 t = nfs4_laundromat();
3009 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3010 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3013 static struct nfs4_stateowner *
3014 search_close_lru(u32 st_id, int flags)
3016 struct nfs4_stateowner *local = NULL;
3018 if (flags & CLOSE_STATE) {
3019 list_for_each_entry(local, &close_lru, so_close_lru) {
3020 if (local->so_id == st_id)
3028 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
3030 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
3034 STALE_STATEID(stateid_t *stateid)
3036 if (stateid->si_boot == boot_time)
3038 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3039 STATEID_VAL(stateid));
3044 access_permit_read(unsigned long access_bmap)
3046 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3047 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3048 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3052 access_permit_write(unsigned long access_bmap)
3054 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3055 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3059 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
3061 __be32 status = nfserr_openmode;
3063 /* For lock stateid's, we test the parent open, not the lock: */
3064 if (stp->st_openstp)
3065 stp = stp->st_openstp;
3066 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3068 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3075 static inline __be32
3076 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3078 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3080 else if (locks_in_grace()) {
3081 /* Answer in remaining cases depends on existance of
3082 * conflicting state; so we must wait out the grace period. */
3083 return nfserr_grace;
3084 } else if (flags & WR_STATE)
3085 return nfs4_share_conflict(current_fh,
3086 NFS4_SHARE_DENY_WRITE);
3087 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3088 return nfs4_share_conflict(current_fh,
3089 NFS4_SHARE_DENY_READ);
3093 * Allow READ/WRITE during grace period on recovered state only for files
3094 * that are not able to provide mandatory locking.
3097 grace_disallows_io(struct inode *inode)
3099 return locks_in_grace() && mandatory_lock(inode);
3102 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3105 * When sessions are used the stateid generation number is ignored
3108 if ((flags & HAS_SESSION) && in->si_generation == 0)
3111 /* If the client sends us a stateid from the future, it's buggy: */
3112 if (in->si_generation > ref->si_generation)
3113 return nfserr_bad_stateid;
3115 * The following, however, can happen. For example, if the
3116 * client sends an open and some IO at the same time, the open
3117 * may bump si_generation while the IO is still in flight.
3118 * Thanks to hard links and renames, the client never knows what
3119 * file an open will affect. So it could avoid that situation
3120 * only by serializing all opens and IO from the same open
3121 * owner. To recover from the old_stateid error, the client
3122 * will just have to retry the IO:
3124 if (in->si_generation < ref->si_generation)
3125 return nfserr_old_stateid;
3130 static int is_delegation_stateid(stateid_t *stateid)
3132 return stateid->si_fileid == 0;
3136 * Checks for stateid operations
3139 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3140 stateid_t *stateid, int flags, struct file **filpp)
3142 struct nfs4_stateid *stp = NULL;
3143 struct nfs4_delegation *dp = NULL;
3144 struct svc_fh *current_fh = &cstate->current_fh;
3145 struct inode *ino = current_fh->fh_dentry->d_inode;
3151 if (grace_disallows_io(ino))
3152 return nfserr_grace;
3154 if (nfsd4_has_session(cstate))
3155 flags |= HAS_SESSION;
3157 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3158 return check_special_stateids(current_fh, stateid, flags);
3160 status = nfserr_stale_stateid;
3161 if (STALE_STATEID(stateid))
3165 * We assume that any stateid that has the current boot time,
3166 * but that we can't find, is expired:
3168 status = nfserr_expired;
3169 if (is_delegation_stateid(stateid)) {
3170 dp = find_delegation_stateid(ino, stateid);
3173 status = check_stateid_generation(stateid, &dp->dl_stateid,
3177 status = nfs4_check_delegmode(dp, flags);
3180 renew_client(dp->dl_client);
3182 *filpp = dp->dl_file->fi_deleg_file;
3185 } else { /* open or lock stateid */
3186 stp = find_stateid(stateid, flags);
3189 status = nfserr_bad_stateid;
3190 if (nfs4_check_fh(current_fh, stp))
3192 if (!stp->st_stateowner->so_confirmed)
3194 status = check_stateid_generation(stateid, &stp->st_stateid,
3198 status = nfs4_check_openmode(stp, flags);
3201 renew_client(stp->st_stateowner->so_client);
3203 if (flags & RD_STATE)
3204 *filpp = find_readable_file(stp->st_file);
3206 *filpp = find_writeable_file(stp->st_file);
3217 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3218 RD_STATE : WR_STATE;
3222 * Checks for sequence id mutating operations.
3225 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3226 stateid_t *stateid, int flags,
3227 struct nfs4_stateowner **sopp,
3228 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3230 struct nfs4_stateid *stp;
3231 struct nfs4_stateowner *sop;
3232 struct svc_fh *current_fh = &cstate->current_fh;
3235 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3236 seqid, STATEID_VAL(stateid));
3241 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3242 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3243 return nfserr_bad_stateid;
3246 if (STALE_STATEID(stateid))
3247 return nfserr_stale_stateid;
3249 if (nfsd4_has_session(cstate))
3250 flags |= HAS_SESSION;
3253 * We return BAD_STATEID if filehandle doesn't match stateid,
3254 * the confirmed flag is incorrecly set, or the generation
3255 * number is incorrect.
3257 stp = find_stateid(stateid, flags);
3260 * Also, we should make sure this isn't just the result of
3263 sop = search_close_lru(stateid->si_stateownerid, flags);
3264 /* It's not stale; let's assume it's expired: */
3266 return nfserr_expired;
3272 *sopp = sop = stp->st_stateowner;
3275 clientid_t *lockclid = &lock->v.new.clientid;
3276 struct nfs4_client *clp = sop->so_client;
3280 lkflg = setlkflg(lock->lk_type);
3282 if (lock->lk_is_new) {
3283 if (!sop->so_is_open_owner)
3284 return nfserr_bad_stateid;
3285 if (!(flags & HAS_SESSION) &&
3286 !same_clid(&clp->cl_clientid, lockclid))
3287 return nfserr_bad_stateid;
3288 /* stp is the open stateid */
3289 status = nfs4_check_openmode(stp, lkflg);
3293 /* stp is the lock stateid */
3294 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3300 if (nfs4_check_fh(current_fh, stp)) {
3301 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3302 return nfserr_bad_stateid;
3306 * We now validate the seqid and stateid generation numbers.
3307 * For the moment, we ignore the possibility of
3308 * generation number wraparound.
3310 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3313 if (sop->so_confirmed && flags & CONFIRM) {
3314 dprintk("NFSD: preprocess_seqid_op: expected"
3315 " unconfirmed stateowner!\n");
3316 return nfserr_bad_stateid;
3318 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3319 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3320 " confirmed yet!\n");
3321 return nfserr_bad_stateid;
3323 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3326 renew_client(sop->so_client);
3330 if (seqid == sop->so_seqid - 1) {
3331 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3332 /* indicate replay to calling function */
3333 return nfserr_replay_me;
3335 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3336 sop->so_seqid, seqid);
3338 return nfserr_bad_seqid;
3342 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3343 struct nfsd4_open_confirm *oc)
3346 struct nfs4_stateowner *sop;
3347 struct nfs4_stateid *stp;
3349 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3350 (int)cstate->current_fh.fh_dentry->d_name.len,
3351 cstate->current_fh.fh_dentry->d_name.name);
3353 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3359 if ((status = nfs4_preprocess_seqid_op(cstate,
3360 oc->oc_seqid, &oc->oc_req_stateid,
3361 CONFIRM | OPEN_STATE,
3362 &oc->oc_stateowner, &stp, NULL)))
3365 sop = oc->oc_stateowner;
3366 sop->so_confirmed = 1;
3367 update_stateid(&stp->st_stateid);
3368 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3369 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3370 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3372 nfsd4_create_clid_dir(sop->so_client);
3374 if (oc->oc_stateowner) {
3375 nfs4_get_stateowner(oc->oc_stateowner);
3376 cstate->replay_owner = oc->oc_stateowner;
3378 nfs4_unlock_state();
3384 * unset all bits in union bitmap (bmap) that
3385 * do not exist in share (from successful OPEN_DOWNGRADE)
3388 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3391 for (i = 1; i < 4; i++) {
3392 if ((i & access) != i)
3393 __clear_bit(i, bmap);
3398 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3401 for (i = 0; i < 4; i++) {
3402 if ((i & deny) != i)
3403 __clear_bit(i, bmap);
3408 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3409 struct nfsd4_compound_state *cstate,
3410 struct nfsd4_open_downgrade *od)
3413 struct nfs4_stateid *stp;
3414 unsigned int share_access;
3416 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3417 (int)cstate->current_fh.fh_dentry->d_name.len,
3418 cstate->current_fh.fh_dentry->d_name.name);
3420 if (!access_valid(od->od_share_access, cstate->minorversion)
3421 || !deny_valid(od->od_share_deny))
3422 return nfserr_inval;
3425 if ((status = nfs4_preprocess_seqid_op(cstate,
3429 &od->od_stateowner, &stp, NULL)))
3432 status = nfserr_inval;
3433 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3434 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3435 stp->st_access_bmap, od->od_share_access);
3438 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3439 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3440 stp->st_deny_bmap, od->od_share_deny);
3443 set_access(&share_access, stp->st_access_bmap);
3444 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3446 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3447 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3449 update_stateid(&stp->st_stateid);
3450 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3453 if (od->od_stateowner) {
3454 nfs4_get_stateowner(od->od_stateowner);
3455 cstate->replay_owner = od->od_stateowner;
3457 nfs4_unlock_state();
3462 * nfs4_unlock_state() called after encode
3465 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3466 struct nfsd4_close *close)
3469 struct nfs4_stateid *stp;
3471 dprintk("NFSD: nfsd4_close on file %.*s\n",
3472 (int)cstate->current_fh.fh_dentry->d_name.len,
3473 cstate->current_fh.fh_dentry->d_name.name);
3476 /* check close_lru for replay */
3477 if ((status = nfs4_preprocess_seqid_op(cstate,
3480 OPEN_STATE | CLOSE_STATE,
3481 &close->cl_stateowner, &stp, NULL)))
3484 update_stateid(&stp->st_stateid);
3485 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3487 /* release_stateid() calls nfsd_close() if needed */
3488 release_open_stateid(stp);
3490 /* place unused nfs4_stateowners on so_close_lru list to be
3491 * released by the laundromat service after the lease period
3492 * to enable us to handle CLOSE replay
3494 if (list_empty(&close->cl_stateowner->so_stateids))
3495 move_to_close_lru(close->cl_stateowner);
3497 if (close->cl_stateowner) {
3498 nfs4_get_stateowner(close->cl_stateowner);
3499 cstate->replay_owner = close->cl_stateowner;
3501 nfs4_unlock_state();
3506 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3507 struct nfsd4_delegreturn *dr)
3509 struct nfs4_delegation *dp;
3510 stateid_t *stateid = &dr->dr_stateid;
3511 struct inode *inode;
3515 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3517 inode = cstate->current_fh.fh_dentry->d_inode;
3519 if (nfsd4_has_session(cstate))
3520 flags |= HAS_SESSION;
3522 status = nfserr_bad_stateid;
3523 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3525 status = nfserr_stale_stateid;
3526 if (STALE_STATEID(stateid))
3528 status = nfserr_bad_stateid;
3529 if (!is_delegation_stateid(stateid))
3531 status = nfserr_expired;
3532 dp = find_delegation_stateid(inode, stateid);
3535 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3538 renew_client(dp->dl_client);
3540 unhash_delegation(dp);
3542 nfs4_unlock_state();
3549 * Lock owner state (byte-range locks)
3551 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3552 #define LOCK_HASH_BITS 8
3553 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3554 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3557 end_offset(u64 start, u64 len)
3562 return end >= start ? end: NFS4_MAX_UINT64;
3565 /* last octet in a range */
3567 last_byte_offset(u64 start, u64 len)
3573 return end > start ? end - 1: NFS4_MAX_UINT64;
3576 #define lockownerid_hashval(id) \
3577 ((id) & LOCK_HASH_MASK)
3579 static inline unsigned int
3580 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3581 struct xdr_netobj *ownername)
3583 return (file_hashval(inode) + cl_id
3584 + opaque_hashval(ownername->data, ownername->len))
3588 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3589 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3590 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3592 static struct nfs4_stateid *
3593 find_stateid(stateid_t *stid, int flags)
3595 struct nfs4_stateid *local;
3596 u32 st_id = stid->si_stateownerid;
3597 u32 f_id = stid->si_fileid;
3598 unsigned int hashval;
3600 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3601 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3602 hashval = stateid_hashval(st_id, f_id);
3603 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3604 if ((local->st_stateid.si_stateownerid == st_id) &&
3605 (local->st_stateid.si_fileid == f_id))
3610 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3611 hashval = stateid_hashval(st_id, f_id);
3612 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3613 if ((local->st_stateid.si_stateownerid == st_id) &&
3614 (local->st_stateid.si_fileid == f_id))
3621 static struct nfs4_delegation *
3622 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3624 struct nfs4_file *fp;
3625 struct nfs4_delegation *dl;
3627 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3630 fp = find_file(ino);
3633 dl = find_delegation_file(fp, stid);
3639 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3640 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3641 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3642 * locking, this prevents us from being completely protocol-compliant. The
3643 * real solution to this problem is to start using unsigned file offsets in
3644 * the VFS, but this is a very deep change!
3647 nfs4_transform_lock_offset(struct file_lock *lock)
3649 if (lock->fl_start < 0)
3650 lock->fl_start = OFFSET_MAX;
3651 if (lock->fl_end < 0)
3652 lock->fl_end = OFFSET_MAX;
3655 /* Hack!: For now, we're defining this just so we can use a pointer to it
3656 * as a unique cookie to identify our (NFSv4's) posix locks. */
3657 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3661 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3663 struct nfs4_stateowner *sop;
3665 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3666 sop = (struct nfs4_stateowner *) fl->fl_owner;
3667 kref_get(&sop->so_ref);
3669 deny->ld_clientid = sop->so_client->cl_clientid;
3671 deny->ld_sop = NULL;
3672 deny->ld_clientid.cl_boot = 0;
3673 deny->ld_clientid.cl_id = 0;
3675 deny->ld_start = fl->fl_start;
3676 deny->ld_length = NFS4_MAX_UINT64;
3677 if (fl->fl_end != NFS4_MAX_UINT64)
3678 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3679 deny->ld_type = NFS4_READ_LT;
3680 if (fl->fl_type != F_RDLCK)
3681 deny->ld_type = NFS4_WRITE_LT;
3684 static struct nfs4_stateowner *
3685 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3686 struct xdr_netobj *owner)
3688 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3689 struct nfs4_stateowner *op;
3691 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3692 if (same_owner_str(op, owner, clid))
3699 * Alloc a lock owner structure.
3700 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3703 * strhashval = lock_ownerstr_hashval
3706 static struct nfs4_stateowner *
3707 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3708 struct nfs4_stateowner *sop;
3709 struct nfs4_replay *rp;
3710 unsigned int idhashval;
3712 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3714 idhashval = lockownerid_hashval(current_ownerid);
3715 INIT_LIST_HEAD(&sop->so_idhash);
3716 INIT_LIST_HEAD(&sop->so_strhash);
3717 INIT_LIST_HEAD(&sop->so_perclient);
3718 INIT_LIST_HEAD(&sop->so_stateids);
3719 INIT_LIST_HEAD(&sop->so_perstateid);
3720 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3722 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3723 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3724 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3725 sop->so_is_open_owner = 0;
3726 sop->so_id = current_ownerid++;
3727 sop->so_client = clp;
3728 /* It is the openowner seqid that will be incremented in encode in the
3729 * case of new lockowners; so increment the lock seqid manually: */
3730 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3731 sop->so_confirmed = 1;
3732 rp = &sop->so_replay;
3733 rp->rp_status = nfserr_serverfault;
3735 rp->rp_buf = rp->rp_ibuf;
3739 static struct nfs4_stateid *
3740 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3742 struct nfs4_stateid *stp;
3743 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3745 stp = nfs4_alloc_stateid();
3748 INIT_LIST_HEAD(&stp->st_hash);
3749 INIT_LIST_HEAD(&stp->st_perfile);
3750 INIT_LIST_HEAD(&stp->st_perstateowner);
3751 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3752 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3753 list_add(&stp->st_perfile, &fp->fi_stateids);
3754 list_add(&stp->st_perstateowner, &sop->so_stateids);
3755 stp->st_stateowner = sop;
3758 stp->st_stateid.si_boot = boot_time;
3759 stp->st_stateid.si_stateownerid = sop->so_id;
3760 stp->st_stateid.si_fileid = fp->fi_id;
3761 stp->st_stateid.si_generation = 0;
3762 stp->st_access_bmap = 0;
3763 stp->st_deny_bmap = open_stp->st_deny_bmap;
3764 stp->st_openstp = open_stp;
3771 check_lock_length(u64 offset, u64 length)
3773 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3774 LOFF_OVERFLOW(offset, length)));
3777 static void get_lock_access(struct nfs4_stateid *lock_stp, u32 access)
3779 struct nfs4_file *fp = lock_stp->st_file;
3780 int oflag = nfs4_access_to_omode(access);
3782 if (test_bit(access, &lock_stp->st_access_bmap))
3784 nfs4_file_get_access(fp, oflag);
3785 __set_bit(access, &lock_stp->st_access_bmap);
3792 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3793 struct nfsd4_lock *lock)
3795 struct nfs4_stateowner *open_sop = NULL;
3796 struct nfs4_stateowner *lock_sop = NULL;
3797 struct nfs4_stateid *lock_stp;
3798 struct nfs4_file *fp;
3799 struct file *filp = NULL;
3800 struct file_lock file_lock;
3801 struct file_lock conflock;
3803 unsigned int strhashval;
3806 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3807 (long long) lock->lk_offset,
3808 (long long) lock->lk_length);
3810 if (check_lock_length(lock->lk_offset, lock->lk_length))
3811 return nfserr_inval;
3813 if ((status = fh_verify(rqstp, &cstate->current_fh,
3814 S_IFREG, NFSD_MAY_LOCK))) {
3815 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3821 if (lock->lk_is_new) {
3823 * Client indicates that this is a new lockowner.
3824 * Use open owner and open stateid to create lock owner and
3827 struct nfs4_stateid *open_stp = NULL;
3829 status = nfserr_stale_clientid;
3830 if (!nfsd4_has_session(cstate) &&
3831 STALE_CLIENTID(&lock->lk_new_clientid))
3834 /* validate and update open stateid and open seqid */
3835 status = nfs4_preprocess_seqid_op(cstate,
3836 lock->lk_new_open_seqid,
3837 &lock->lk_new_open_stateid,
3839 &lock->lk_replay_owner, &open_stp,
3843 open_sop = lock->lk_replay_owner;
3844 /* create lockowner and lock stateid */
3845 fp = open_stp->st_file;
3846 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3847 open_sop->so_client->cl_clientid.cl_id,
3848 &lock->v.new.owner);
3849 /* XXX: Do we need to check for duplicate stateowners on
3850 * the same file, or should they just be allowed (and
3851 * create new stateids)? */
3852 status = nfserr_resource;
3853 lock_sop = alloc_init_lock_stateowner(strhashval,
3854 open_sop->so_client, open_stp, lock);
3855 if (lock_sop == NULL)
3857 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3858 if (lock_stp == NULL)
3861 /* lock (lock owner + lock stateid) already exists */
3862 status = nfs4_preprocess_seqid_op(cstate,
3863 lock->lk_old_lock_seqid,
3864 &lock->lk_old_lock_stateid,
3866 &lock->lk_replay_owner, &lock_stp, lock);
3869 lock_sop = lock->lk_replay_owner;
3870 fp = lock_stp->st_file;
3872 /* lock->lk_replay_owner and lock_stp have been created or found */
3874 status = nfserr_grace;
3875 if (locks_in_grace() && !lock->lk_reclaim)
3877 status = nfserr_no_grace;
3878 if (!locks_in_grace() && lock->lk_reclaim)
3881 locks_init_lock(&file_lock);
3882 switch (lock->lk_type) {
3885 filp = find_readable_file(lock_stp->st_file);
3887 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
3888 file_lock.fl_type = F_RDLCK;
3891 case NFS4_WRITEW_LT:
3892 filp = find_writeable_file(lock_stp->st_file);
3894 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
3895 file_lock.fl_type = F_WRLCK;
3898 status = nfserr_inval;
3902 status = nfserr_openmode;
3905 file_lock.fl_owner = (fl_owner_t)lock_sop;
3906 file_lock.fl_pid = current->tgid;
3907 file_lock.fl_file = filp;
3908 file_lock.fl_flags = FL_POSIX;
3909 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3911 file_lock.fl_start = lock->lk_offset;
3912 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3913 nfs4_transform_lock_offset(&file_lock);
3916 * Try to lock the file in the VFS.
3917 * Note: locks.c uses the BKL to protect the inode's lock list.
3920 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
3922 case 0: /* success! */
3923 update_stateid(&lock_stp->st_stateid);
3924 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3928 case (EAGAIN): /* conflock holds conflicting lock */
3929 status = nfserr_denied;
3930 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3931 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3934 status = nfserr_deadlock;
3937 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3938 status = nfserr_resource;
3942 if (status && lock->lk_is_new && lock_sop)
3943 release_lockowner(lock_sop);
3944 if (lock->lk_replay_owner) {
3945 nfs4_get_stateowner(lock->lk_replay_owner);
3946 cstate->replay_owner = lock->lk_replay_owner;
3948 nfs4_unlock_state();
3953 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3954 * so we do a temporary open here just to get an open file to pass to
3955 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3958 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3963 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3966 err = vfs_test_lock(file, lock);
3975 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3976 struct nfsd4_lockt *lockt)
3978 struct inode *inode;
3979 struct file_lock file_lock;
3983 if (locks_in_grace())
3984 return nfserr_grace;
3986 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3987 return nfserr_inval;
3989 lockt->lt_stateowner = NULL;
3992 status = nfserr_stale_clientid;
3993 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3996 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3997 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3998 if (status == nfserr_symlink)
3999 status = nfserr_inval;
4003 inode = cstate->current_fh.fh_dentry->d_inode;
4004 locks_init_lock(&file_lock);
4005 switch (lockt->lt_type) {
4008 file_lock.fl_type = F_RDLCK;
4011 case NFS4_WRITEW_LT:
4012 file_lock.fl_type = F_WRLCK;
4015 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4016 status = nfserr_inval;
4020 lockt->lt_stateowner = find_lockstateowner_str(inode,
4021 &lockt->lt_clientid, &lockt->lt_owner);
4022 if (lockt->lt_stateowner)
4023 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
4024 file_lock.fl_pid = current->tgid;
4025 file_lock.fl_flags = FL_POSIX;
4027 file_lock.fl_start = lockt->lt_offset;
4028 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4030 nfs4_transform_lock_offset(&file_lock);
4033 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4035 status = nfserrno(error);
4038 if (file_lock.fl_type != F_UNLCK) {
4039 status = nfserr_denied;
4040 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4043 nfs4_unlock_state();
4048 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4049 struct nfsd4_locku *locku)
4051 struct nfs4_stateid *stp;
4052 struct file *filp = NULL;
4053 struct file_lock file_lock;
4057 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4058 (long long) locku->lu_offset,
4059 (long long) locku->lu_length);
4061 if (check_lock_length(locku->lu_offset, locku->lu_length))
4062 return nfserr_inval;
4066 if ((status = nfs4_preprocess_seqid_op(cstate,
4070 &locku->lu_stateowner, &stp, NULL)))
4073 filp = find_any_file(stp->st_file);
4075 status = nfserr_lock_range;
4079 locks_init_lock(&file_lock);
4080 file_lock.fl_type = F_UNLCK;
4081 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
4082 file_lock.fl_pid = current->tgid;
4083 file_lock.fl_file = filp;
4084 file_lock.fl_flags = FL_POSIX;
4085 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4086 file_lock.fl_start = locku->lu_offset;
4088 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4089 nfs4_transform_lock_offset(&file_lock);
4092 * Try to unlock the file in the VFS.
4094 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4096 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4100 * OK, unlock succeeded; the only thing left to do is update the stateid.
4102 update_stateid(&stp->st_stateid);
4103 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4106 if (locku->lu_stateowner) {
4107 nfs4_get_stateowner(locku->lu_stateowner);
4108 cstate->replay_owner = locku->lu_stateowner;
4110 nfs4_unlock_state();
4114 status = nfserrno(err);
4120 * 1: locks held by lockowner
4121 * 0: no locks held by lockowner
4124 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4126 struct file_lock **flpp;
4127 struct inode *inode = filp->fi_inode;
4131 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4132 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4143 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4144 struct nfsd4_compound_state *cstate,
4145 struct nfsd4_release_lockowner *rlockowner)
4147 clientid_t *clid = &rlockowner->rl_clientid;
4148 struct nfs4_stateowner *sop;
4149 struct nfs4_stateid *stp;
4150 struct xdr_netobj *owner = &rlockowner->rl_owner;
4151 struct list_head matches;
4155 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4156 clid->cl_boot, clid->cl_id);
4158 /* XXX check for lease expiration */
4160 status = nfserr_stale_clientid;
4161 if (STALE_CLIENTID(clid))
4166 status = nfserr_locks_held;
4167 /* XXX: we're doing a linear search through all the lockowners.
4168 * Yipes! For now we'll just hope clients aren't really using
4169 * release_lockowner much, but eventually we have to fix these
4170 * data structures. */
4171 INIT_LIST_HEAD(&matches);
4172 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4173 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4174 if (!same_owner_str(sop, owner, clid))
4176 list_for_each_entry(stp, &sop->so_stateids,
4178 if (check_for_locks(stp->st_file, sop))
4180 /* Note: so_perclient unused for lockowners,
4181 * so it's OK to fool with here. */
4182 list_add(&sop->so_perclient, &matches);
4186 /* Clients probably won't expect us to return with some (but not all)
4187 * of the lockowner state released; so don't release any until all
4188 * have been checked. */
4190 while (!list_empty(&matches)) {
4191 sop = list_entry(matches.next, struct nfs4_stateowner,
4193 /* unhash_stateowner deletes so_perclient only
4194 * for openowners. */
4195 list_del(&sop->so_perclient);
4196 release_lockowner(sop);
4199 nfs4_unlock_state();
4203 static inline struct nfs4_client_reclaim *
4206 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4210 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4212 unsigned int strhashval = clientstr_hashval(name);
4213 struct nfs4_client *clp;
4215 clp = find_confirmed_client_by_str(name, strhashval);
4220 * failure => all reset bets are off, nfserr_no_grace...
4223 nfs4_client_to_reclaim(const char *name)
4225 unsigned int strhashval;
4226 struct nfs4_client_reclaim *crp = NULL;
4228 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4229 crp = alloc_reclaim();
4232 strhashval = clientstr_hashval(name);
4233 INIT_LIST_HEAD(&crp->cr_strhash);
4234 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4235 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4236 reclaim_str_hashtbl_size++;
4241 nfs4_release_reclaim(void)
4243 struct nfs4_client_reclaim *crp = NULL;
4246 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4247 while (!list_empty(&reclaim_str_hashtbl[i])) {
4248 crp = list_entry(reclaim_str_hashtbl[i].next,
4249 struct nfs4_client_reclaim, cr_strhash);
4250 list_del(&crp->cr_strhash);
4252 reclaim_str_hashtbl_size--;
4255 BUG_ON(reclaim_str_hashtbl_size);
4259 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4260 static struct nfs4_client_reclaim *
4261 nfs4_find_reclaim_client(clientid_t *clid)
4263 unsigned int strhashval;
4264 struct nfs4_client *clp;
4265 struct nfs4_client_reclaim *crp = NULL;
4268 /* find clientid in conf_id_hashtbl */
4269 clp = find_confirmed_client(clid);
4273 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4274 clp->cl_name.len, clp->cl_name.data,
4277 /* find clp->cl_name in reclaim_str_hashtbl */
4278 strhashval = clientstr_hashval(clp->cl_recdir);
4279 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4280 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4288 * Called from OPEN. Look for clientid in reclaim list.
4291 nfs4_check_open_reclaim(clientid_t *clid)
4293 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4296 /* initialization to perform at module load time: */
4299 nfs4_state_init(void)
4303 status = nfsd4_init_slabs();
4306 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4307 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4308 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4309 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4310 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4311 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4313 for (i = 0; i < SESSION_HASH_SIZE; i++)
4314 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4315 for (i = 0; i < FILE_HASH_SIZE; i++) {
4316 INIT_LIST_HEAD(&file_hashtbl[i]);
4318 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4319 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4320 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4322 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4323 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4324 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4326 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4327 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4328 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4330 memset(&onestateid, ~0, sizeof(stateid_t));
4331 INIT_LIST_HEAD(&close_lru);
4332 INIT_LIST_HEAD(&client_lru);
4333 INIT_LIST_HEAD(&del_recall_lru);
4334 reclaim_str_hashtbl_size = 0;
4339 nfsd4_load_reboot_recovery_data(void)
4344 nfsd4_init_recdir(user_recovery_dirname);
4345 status = nfsd4_recdir_load();
4346 nfs4_unlock_state();
4348 printk("NFSD: Failure reading reboot recovery data\n");
4352 * Since the lifetime of a delegation isn't limited to that of an open, a
4353 * client may quite reasonably hang on to a delegation as long as it has
4354 * the inode cached. This becomes an obvious problem the first time a
4355 * client's inode cache approaches the size of the server's total memory.
4357 * For now we avoid this problem by imposing a hard limit on the number
4358 * of delegations, which varies according to the server's memory size.
4361 set_max_delegations(void)
4364 * Allow at most 4 delegations per megabyte of RAM. Quick
4365 * estimates suggest that in the worst case (where every delegation
4366 * is for a different inode), a delegation could take about 1.5K,
4367 * giving a worst case usage of about 6% of memory.
4369 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4372 /* initialization to perform when the nfsd service is started: */
4375 __nfs4_state_start(void)
4379 boot_time = get_seconds();
4380 locks_start_grace(&nfsd4_manager);
4381 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4383 ret = set_callback_cred();
4386 laundry_wq = create_singlethread_workqueue("nfsd4");
4387 if (laundry_wq == NULL)
4389 ret = nfsd4_create_callback_queue();
4391 goto out_free_laundry;
4392 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4393 set_max_delegations();
4396 destroy_workqueue(laundry_wq);
4401 nfs4_state_start(void)
4403 nfsd4_load_reboot_recovery_data();
4404 return __nfs4_state_start();
4408 __nfs4_state_shutdown(void)
4411 struct nfs4_client *clp = NULL;
4412 struct nfs4_delegation *dp = NULL;
4413 struct list_head *pos, *next, reaplist;
4415 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4416 while (!list_empty(&conf_id_hashtbl[i])) {
4417 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4420 while (!list_empty(&unconf_str_hashtbl[i])) {
4421 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4425 INIT_LIST_HEAD(&reaplist);
4426 spin_lock(&recall_lock);
4427 list_for_each_safe(pos, next, &del_recall_lru) {
4428 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4429 list_move(&dp->dl_recall_lru, &reaplist);
4431 spin_unlock(&recall_lock);
4432 list_for_each_safe(pos, next, &reaplist) {
4433 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4434 list_del_init(&dp->dl_recall_lru);
4435 unhash_delegation(dp);
4438 nfsd4_shutdown_recdir();
4442 nfs4_state_shutdown(void)
4444 cancel_delayed_work_sync(&laundromat_work);
4445 destroy_workqueue(laundry_wq);
4446 locks_end_grace(&nfsd4_manager);
4448 nfs4_release_reclaim();
4449 __nfs4_state_shutdown();
4450 nfs4_unlock_state();
4451 nfsd4_destroy_callback_queue();
4455 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4456 * accessed when nfsd is starting.
4459 nfs4_set_recdir(char *recdir)
4461 strcpy(user_recovery_dirname, recdir);
4465 * Change the NFSv4 recovery directory to recdir.
4468 nfs4_reset_recoverydir(char *recdir)
4473 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4477 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4478 nfs4_set_recdir(recdir);
4486 nfs4_recoverydir(void)
4488 return user_recovery_dirname;