2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
36 #include <linux/smp_lock.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
45 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 time_t nfsd4_lease = 90; /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static stateid_t zerostateid; /* bits all 0 */
55 static stateid_t onestateid; /* bits all 1 */
56 static u64 current_sessionid = 1;
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
73 * Currently used for the del_recall_lru and file hash table. In an
74 * effort to decrease the scope of the client_mutex, this spinlock may
75 * eventually cover more:
77 static DEFINE_SPINLOCK(recall_lock);
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
87 mutex_lock(&client_mutex);
91 nfs4_unlock_state(void)
93 mutex_unlock(&client_mutex);
97 opaque_hashval(const void *ptr, int nbytes)
99 unsigned char *cptr = (unsigned char *) ptr;
109 static struct list_head del_recall_lru;
112 put_nfs4_file(struct nfs4_file *fi)
114 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115 list_del(&fi->fi_hash);
116 spin_unlock(&recall_lock);
118 kmem_cache_free(file_slab, fi);
123 get_nfs4_file(struct nfs4_file *fi)
125 atomic_inc(&fi->fi_ref);
128 static int num_delegations;
129 unsigned int max_delegations;
132 * Open owner state (share locks)
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS 8
137 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
140 #define ownerid_hashval(id) \
141 ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS 8
150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS 10
154 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
157 #define file_hashval(x) \
158 hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id) \
160 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
167 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168 atomic_inc(&fp->fi_access[oflag]);
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
173 if (oflag == O_RDWR) {
174 __nfs4_file_get_access(fp, O_RDONLY);
175 __nfs4_file_get_access(fp, O_WRONLY);
177 __nfs4_file_get_access(fp, oflag);
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
182 if (fp->fi_fds[oflag]) {
183 fput(fp->fi_fds[oflag]);
184 fp->fi_fds[oflag] = NULL;
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
190 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191 nfs4_file_put_fd(fp, O_RDWR);
192 nfs4_file_put_fd(fp, oflag);
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
198 if (oflag == O_RDWR) {
199 __nfs4_file_put_access(fp, O_RDONLY);
200 __nfs4_file_put_access(fp, O_WRONLY);
202 __nfs4_file_put_access(fp, oflag);
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
208 struct nfs4_delegation *dp;
209 struct nfs4_file *fp = stp->st_file;
211 dprintk("NFSD alloc_init_deleg\n");
213 * Major work on the lease subsystem (for example, to support
214 * calbacks on stat) will be required before we can support
215 * write delegations properly.
217 if (type != NFS4_OPEN_DELEGATE_READ)
219 if (fp->fi_had_conflict)
221 if (num_delegations > max_delegations)
223 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
227 INIT_LIST_HEAD(&dp->dl_perfile);
228 INIT_LIST_HEAD(&dp->dl_perclnt);
229 INIT_LIST_HEAD(&dp->dl_recall_lru);
233 nfs4_file_get_access(fp, O_RDONLY);
236 dp->dl_stateid.si_boot = boot_time;
237 dp->dl_stateid.si_stateownerid = current_delegid++;
238 dp->dl_stateid.si_fileid = 0;
239 dp->dl_stateid.si_generation = 0;
240 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
242 atomic_set(&dp->dl_count, 1);
243 list_add(&dp->dl_perfile, &fp->fi_delegations);
244 list_add(&dp->dl_perclnt, &clp->cl_delegations);
245 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
250 nfs4_put_delegation(struct nfs4_delegation *dp)
252 if (atomic_dec_and_test(&dp->dl_count)) {
253 dprintk("NFSD: freeing dp %p\n",dp);
254 put_nfs4_file(dp->dl_file);
255 kmem_cache_free(deleg_slab, dp);
260 /* Remove the associated file_lock first, then remove the delegation.
261 * lease_modify() is called to remove the FS_LEASE file_lock from
262 * the i_flock list, eventually calling nfsd's lock_manager
263 * fl_release_callback.
266 nfs4_close_delegation(struct nfs4_delegation *dp)
268 struct file *filp = find_readable_file(dp->dl_file);
270 dprintk("NFSD: close_delegation dp %p\n",dp);
272 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
273 nfs4_file_put_access(dp->dl_file, O_RDONLY);
276 /* Called under the state lock. */
278 unhash_delegation(struct nfs4_delegation *dp)
280 list_del_init(&dp->dl_perfile);
281 list_del_init(&dp->dl_perclnt);
282 spin_lock(&recall_lock);
283 list_del_init(&dp->dl_recall_lru);
284 spin_unlock(&recall_lock);
285 nfs4_close_delegation(dp);
286 nfs4_put_delegation(dp);
293 /* client_lock protects the client lru list and session hash table */
294 static DEFINE_SPINLOCK(client_lock);
296 /* Hash tables for nfs4_clientid state */
297 #define CLIENT_HASH_BITS 4
298 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
299 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
301 #define clientid_hashval(id) \
302 ((id) & CLIENT_HASH_MASK)
303 #define clientstr_hashval(name) \
304 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
306 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
307 * used in reboot/reset lease grace period processing
309 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
310 * setclientid_confirmed info.
312 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
315 * client_lru holds client queue ordered by nfs4_client.cl_time
318 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
319 * for last close replay.
321 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
322 static int reclaim_str_hashtbl_size = 0;
323 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
324 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
325 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
326 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
327 static struct list_head client_lru;
328 static struct list_head close_lru;
330 static void unhash_generic_stateid(struct nfs4_stateid *stp)
332 list_del(&stp->st_hash);
333 list_del(&stp->st_perfile);
334 list_del(&stp->st_perstateowner);
337 static void free_generic_stateid(struct nfs4_stateid *stp)
339 put_nfs4_file(stp->st_file);
340 kmem_cache_free(stateid_slab, stp);
343 static void release_lock_stateid(struct nfs4_stateid *stp)
347 unhash_generic_stateid(stp);
348 file = find_any_file(stp->st_file);
350 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
351 free_generic_stateid(stp);
354 static void unhash_lockowner(struct nfs4_stateowner *sop)
356 struct nfs4_stateid *stp;
358 list_del(&sop->so_idhash);
359 list_del(&sop->so_strhash);
360 list_del(&sop->so_perstateid);
361 while (!list_empty(&sop->so_stateids)) {
362 stp = list_first_entry(&sop->so_stateids,
363 struct nfs4_stateid, st_perstateowner);
364 release_lock_stateid(stp);
368 static void release_lockowner(struct nfs4_stateowner *sop)
370 unhash_lockowner(sop);
371 nfs4_put_stateowner(sop);
375 release_stateid_lockowners(struct nfs4_stateid *open_stp)
377 struct nfs4_stateowner *lock_sop;
379 while (!list_empty(&open_stp->st_lockowners)) {
380 lock_sop = list_entry(open_stp->st_lockowners.next,
381 struct nfs4_stateowner, so_perstateid);
382 /* list_del(&open_stp->st_lockowners); */
383 BUG_ON(lock_sop->so_is_open_owner);
384 release_lockowner(lock_sop);
389 * We store the NONE, READ, WRITE, and BOTH bits separately in the
390 * st_{access,deny}_bmap field of the stateid, in order to track not
391 * only what share bits are currently in force, but also what
392 * combinations of share bits previous opens have used. This allows us
393 * to enforce the recommendation of rfc 3530 14.2.19 that the server
394 * return an error if the client attempt to downgrade to a combination
395 * of share bits not explicable by closing some of its previous opens.
397 * XXX: This enforcement is actually incomplete, since we don't keep
398 * track of access/deny bit combinations; so, e.g., we allow:
400 * OPEN allow read, deny write
401 * OPEN allow both, deny none
402 * DOWNGRADE allow read, deny none
404 * which we should reject.
407 set_access(unsigned int *access, unsigned long bmap) {
411 for (i = 1; i < 4; i++) {
412 if (test_bit(i, &bmap))
418 set_deny(unsigned int *deny, unsigned long bmap) {
422 for (i = 0; i < 4; i++) {
423 if (test_bit(i, &bmap))
429 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
430 unsigned int access, deny;
432 set_access(&access, stp->st_access_bmap);
433 set_deny(&deny, stp->st_deny_bmap);
434 if ((access & open->op_share_deny) || (deny & open->op_share_access))
439 static int nfs4_access_to_omode(u32 access)
441 switch (access & NFS4_SHARE_ACCESS_BOTH) {
442 case NFS4_SHARE_ACCESS_READ:
444 case NFS4_SHARE_ACCESS_WRITE:
446 case NFS4_SHARE_ACCESS_BOTH:
452 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
456 set_access(&access, stp->st_access_bmap);
457 return nfs4_access_to_omode(access);
460 static void release_open_stateid(struct nfs4_stateid *stp)
462 int oflag = nfs4_access_bmap_to_omode(stp);
464 unhash_generic_stateid(stp);
465 release_stateid_lockowners(stp);
466 nfs4_file_put_access(stp->st_file, oflag);
467 free_generic_stateid(stp);
470 static void unhash_openowner(struct nfs4_stateowner *sop)
472 struct nfs4_stateid *stp;
474 list_del(&sop->so_idhash);
475 list_del(&sop->so_strhash);
476 list_del(&sop->so_perclient);
477 list_del(&sop->so_perstateid); /* XXX: necessary? */
478 while (!list_empty(&sop->so_stateids)) {
479 stp = list_first_entry(&sop->so_stateids,
480 struct nfs4_stateid, st_perstateowner);
481 release_open_stateid(stp);
485 static void release_openowner(struct nfs4_stateowner *sop)
487 unhash_openowner(sop);
488 list_del(&sop->so_close_lru);
489 nfs4_put_stateowner(sop);
492 #define SESSION_HASH_SIZE 512
493 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
496 hash_sessionid(struct nfs4_sessionid *sessionid)
498 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
500 return sid->sequence % SESSION_HASH_SIZE;
504 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
506 u32 *ptr = (u32 *)(&sessionid->data[0]);
507 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
511 gen_sessionid(struct nfsd4_session *ses)
513 struct nfs4_client *clp = ses->se_client;
514 struct nfsd4_sessionid *sid;
516 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
517 sid->clientid = clp->cl_clientid;
518 sid->sequence = current_sessionid++;
523 * The protocol defines ca_maxresponssize_cached to include the size of
524 * the rpc header, but all we need to cache is the data starting after
525 * the end of the initial SEQUENCE operation--the rest we regenerate
526 * each time. Therefore we can advertise a ca_maxresponssize_cached
527 * value that is the number of bytes in our cache plus a few additional
528 * bytes. In order to stay on the safe side, and not promise more than
529 * we can cache, those additional bytes must be the minimum possible: 24
530 * bytes of rpc header (xid through accept state, with AUTH_NULL
531 * verifier), 12 for the compound header (with zero-length tag), and 44
532 * for the SEQUENCE op response:
534 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
537 free_session_slots(struct nfsd4_session *ses)
541 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
542 kfree(ses->se_slots[i]);
546 * We don't actually need to cache the rpc and session headers, so we
547 * can allocate a little less for each slot:
549 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
551 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
554 static int nfsd4_sanitize_slot_size(u32 size)
556 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
557 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
563 * XXX: If we run out of reserved DRC memory we could (up to a point)
564 * re-negotiate active sessions and reduce their slot usage to make
565 * rooom for new connections. For now we just fail the create session.
567 static int nfsd4_get_drc_mem(int slotsize, u32 num)
571 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
573 spin_lock(&nfsd_drc_lock);
574 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
575 nfsd_drc_max_mem - nfsd_drc_mem_used);
576 num = min_t(int, num, avail / slotsize);
577 nfsd_drc_mem_used += num * slotsize;
578 spin_unlock(&nfsd_drc_lock);
583 static void nfsd4_put_drc_mem(int slotsize, int num)
585 spin_lock(&nfsd_drc_lock);
586 nfsd_drc_mem_used -= slotsize * num;
587 spin_unlock(&nfsd_drc_lock);
590 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
592 struct nfsd4_session *new;
595 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
596 + sizeof(struct nfsd4_session) > PAGE_SIZE);
597 mem = numslots * sizeof(struct nfsd4_slot *);
599 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
602 /* allocate each struct nfsd4_slot and data cache in one piece */
603 for (i = 0; i < numslots; i++) {
604 mem = sizeof(struct nfsd4_slot) + slotsize;
605 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
606 if (!new->se_slots[i])
612 kfree(new->se_slots[i]);
617 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
619 u32 maxrpc = nfsd_serv->sv_max_mesg;
621 new->maxreqs = numslots;
622 new->maxresp_cached = slotsize + NFSD_MIN_HDR_SEQ_SZ;
623 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
624 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
625 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
628 static void free_conn(struct nfsd4_conn *c)
630 svc_xprt_put(c->cn_xprt);
634 static void nfsd4_conn_lost(struct svc_xpt_user *u)
636 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
637 struct nfs4_client *clp = c->cn_session->se_client;
639 spin_lock(&clp->cl_lock);
640 if (!list_empty(&c->cn_persession)) {
641 list_del(&c->cn_persession);
644 spin_unlock(&clp->cl_lock);
647 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
649 struct nfsd4_conn *conn;
651 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
654 svc_xprt_get(rqstp->rq_xprt);
655 conn->cn_xprt = rqstp->rq_xprt;
656 conn->cn_flags = flags;
657 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
661 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
663 conn->cn_session = ses;
664 list_add(&conn->cn_persession, &ses->se_conns);
667 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
669 struct nfs4_client *clp = ses->se_client;
671 spin_lock(&clp->cl_lock);
672 __nfsd4_hash_conn(conn, ses);
673 spin_unlock(&clp->cl_lock);
676 static void nfsd4_register_conn(struct nfsd4_conn *conn)
678 conn->cn_xpt_user.callback = nfsd4_conn_lost;
679 register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
682 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses)
684 struct nfsd4_conn *conn;
685 u32 flags = NFS4_CDFC4_FORE;
687 if (ses->se_flags & SESSION4_BACK_CHAN)
688 flags |= NFS4_CDFC4_BACK;
689 conn = alloc_conn(rqstp, flags);
691 return nfserr_jukebox;
692 nfsd4_hash_conn(conn, ses);
693 nfsd4_register_conn(conn);
697 static void nfsd4_del_conns(struct nfsd4_session *s)
699 struct nfs4_client *clp = s->se_client;
700 struct nfsd4_conn *c;
702 spin_lock(&clp->cl_lock);
703 while (!list_empty(&s->se_conns)) {
704 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
705 list_del_init(&c->cn_persession);
706 spin_unlock(&clp->cl_lock);
708 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
711 spin_lock(&clp->cl_lock);
713 spin_unlock(&clp->cl_lock);
716 void free_session(struct kref *kref)
718 struct nfsd4_session *ses;
721 ses = container_of(kref, struct nfsd4_session, se_ref);
722 nfsd4_del_conns(ses);
723 spin_lock(&nfsd_drc_lock);
724 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
725 nfsd_drc_mem_used -= mem;
726 spin_unlock(&nfsd_drc_lock);
727 free_session_slots(ses);
731 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
733 struct nfsd4_session *new;
734 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
735 int numslots, slotsize;
740 * Note decreasing slot size below client's request may
741 * make it difficult for client to function correctly, whereas
742 * decreasing the number of slots will (just?) affect
743 * performance. When short on memory we therefore prefer to
744 * decrease number of slots instead of their size.
746 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
747 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
749 new = alloc_session(slotsize, numslots);
751 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
754 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
756 new->se_client = clp;
759 INIT_LIST_HEAD(&new->se_conns);
761 new->se_cb_seq_nr = 1;
762 new->se_flags = cses->flags;
763 new->se_cb_prog = cses->callback_prog;
764 kref_init(&new->se_ref);
765 idx = hash_sessionid(&new->se_sessionid);
766 spin_lock(&client_lock);
767 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
768 list_add(&new->se_perclnt, &clp->cl_sessions);
769 spin_unlock(&client_lock);
771 status = nfsd4_new_conn(rqstp, new);
772 /* whoops: benny points out, status is ignored! (err, or bogus) */
774 free_session(&new->se_ref);
777 if (!clp->cl_cb_session && (cses->flags & SESSION4_BACK_CHAN)) {
778 struct sockaddr *sa = svc_addr(rqstp);
780 clp->cl_cb_session = new;
781 clp->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
782 svc_xprt_get(rqstp->rq_xprt);
783 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
784 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
785 nfsd4_probe_callback(clp);
790 /* caller must hold client_lock */
791 static struct nfsd4_session *
792 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
794 struct nfsd4_session *elem;
797 dump_sessionid(__func__, sessionid);
798 idx = hash_sessionid(sessionid);
799 /* Search in the appropriate list */
800 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
801 if (!memcmp(elem->se_sessionid.data, sessionid->data,
802 NFS4_MAX_SESSIONID_LEN)) {
807 dprintk("%s: session not found\n", __func__);
811 /* caller must hold client_lock */
813 unhash_session(struct nfsd4_session *ses)
815 list_del(&ses->se_hash);
816 list_del(&ses->se_perclnt);
819 /* must be called under the client_lock */
821 renew_client_locked(struct nfs4_client *clp)
823 if (is_client_expired(clp)) {
824 dprintk("%s: client (clientid %08x/%08x) already expired\n",
826 clp->cl_clientid.cl_boot,
827 clp->cl_clientid.cl_id);
832 * Move client to the end to the LRU list.
834 dprintk("renewing client (clientid %08x/%08x)\n",
835 clp->cl_clientid.cl_boot,
836 clp->cl_clientid.cl_id);
837 list_move_tail(&clp->cl_lru, &client_lru);
838 clp->cl_time = get_seconds();
842 renew_client(struct nfs4_client *clp)
844 spin_lock(&client_lock);
845 renew_client_locked(clp);
846 spin_unlock(&client_lock);
849 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
851 STALE_CLIENTID(clientid_t *clid)
853 if (clid->cl_boot == boot_time)
855 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
856 clid->cl_boot, clid->cl_id, boot_time);
861 * XXX Should we use a slab cache ?
862 * This type of memory management is somewhat inefficient, but we use it
863 * anyway since SETCLIENTID is not a common operation.
865 static struct nfs4_client *alloc_client(struct xdr_netobj name)
867 struct nfs4_client *clp;
869 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
872 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
873 if (clp->cl_name.data == NULL) {
877 memcpy(clp->cl_name.data, name.data, name.len);
878 clp->cl_name.len = name.len;
883 free_client(struct nfs4_client *clp)
885 while (!list_empty(&clp->cl_sessions)) {
886 struct nfsd4_session *ses;
887 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
889 list_del(&ses->se_perclnt);
890 nfsd4_put_session(ses);
892 if (clp->cl_cred.cr_group_info)
893 put_group_info(clp->cl_cred.cr_group_info);
894 kfree(clp->cl_principal);
895 kfree(clp->cl_name.data);
900 release_session_client(struct nfsd4_session *session)
902 struct nfs4_client *clp = session->se_client;
904 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
906 if (is_client_expired(clp)) {
908 session->se_client = NULL;
910 renew_client_locked(clp);
911 spin_unlock(&client_lock);
914 /* must be called under the client_lock */
916 unhash_client_locked(struct nfs4_client *clp)
918 struct nfsd4_session *ses;
920 mark_client_expired(clp);
921 list_del(&clp->cl_lru);
922 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
923 list_del_init(&ses->se_hash);
927 expire_client(struct nfs4_client *clp)
929 struct nfs4_stateowner *sop;
930 struct nfs4_delegation *dp;
931 struct list_head reaplist;
933 INIT_LIST_HEAD(&reaplist);
934 spin_lock(&recall_lock);
935 while (!list_empty(&clp->cl_delegations)) {
936 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
937 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
939 list_del_init(&dp->dl_perclnt);
940 list_move(&dp->dl_recall_lru, &reaplist);
942 spin_unlock(&recall_lock);
943 while (!list_empty(&reaplist)) {
944 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
945 list_del_init(&dp->dl_recall_lru);
946 unhash_delegation(dp);
948 while (!list_empty(&clp->cl_openowners)) {
949 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
950 release_openowner(sop);
952 nfsd4_shutdown_callback(clp);
953 if (clp->cl_cb_conn.cb_xprt)
954 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
955 list_del(&clp->cl_idhash);
956 list_del(&clp->cl_strhash);
957 spin_lock(&client_lock);
958 unhash_client_locked(clp);
959 if (atomic_read(&clp->cl_refcount) == 0)
961 spin_unlock(&client_lock);
964 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
966 memcpy(target->cl_verifier.data, source->data,
967 sizeof(target->cl_verifier.data));
970 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
972 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
973 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
976 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
978 target->cr_uid = source->cr_uid;
979 target->cr_gid = source->cr_gid;
980 target->cr_group_info = source->cr_group_info;
981 get_group_info(target->cr_group_info);
984 static int same_name(const char *n1, const char *n2)
986 return 0 == memcmp(n1, n2, HEXDIR_LEN);
990 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
992 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
996 same_clid(clientid_t *cl1, clientid_t *cl2)
998 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1001 /* XXX what about NGROUP */
1003 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1005 return cr1->cr_uid == cr2->cr_uid;
1008 static void gen_clid(struct nfs4_client *clp)
1010 static u32 current_clientid = 1;
1012 clp->cl_clientid.cl_boot = boot_time;
1013 clp->cl_clientid.cl_id = current_clientid++;
1016 static void gen_confirm(struct nfs4_client *clp)
1021 p = (u32 *)clp->cl_confirm.data;
1022 *p++ = get_seconds();
1026 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1027 struct svc_rqst *rqstp, nfs4_verifier *verf)
1029 struct nfs4_client *clp;
1030 struct sockaddr *sa = svc_addr(rqstp);
1033 clp = alloc_client(name);
1037 INIT_LIST_HEAD(&clp->cl_sessions);
1039 princ = svc_gss_principal(rqstp);
1041 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1042 if (clp->cl_principal == NULL) {
1048 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1049 atomic_set(&clp->cl_refcount, 0);
1050 atomic_set(&clp->cl_cb_set, 0);
1051 INIT_LIST_HEAD(&clp->cl_idhash);
1052 INIT_LIST_HEAD(&clp->cl_strhash);
1053 INIT_LIST_HEAD(&clp->cl_openowners);
1054 INIT_LIST_HEAD(&clp->cl_delegations);
1055 INIT_LIST_HEAD(&clp->cl_lru);
1056 spin_lock_init(&clp->cl_lock);
1057 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1058 clp->cl_time = get_seconds();
1059 clear_bit(0, &clp->cl_cb_slot_busy);
1060 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1061 copy_verf(clp, verf);
1062 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1063 clp->cl_flavor = rqstp->rq_flavor;
1064 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1066 clp->cl_cb_session = NULL;
1070 static int check_name(struct xdr_netobj name)
1074 if (name.len > NFS4_OPAQUE_LIMIT) {
1075 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1082 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1084 unsigned int idhashval;
1086 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1087 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1088 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1093 move_to_confirmed(struct nfs4_client *clp)
1095 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1096 unsigned int strhashval;
1098 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1099 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1100 strhashval = clientstr_hashval(clp->cl_recdir);
1101 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1105 static struct nfs4_client *
1106 find_confirmed_client(clientid_t *clid)
1108 struct nfs4_client *clp;
1109 unsigned int idhashval = clientid_hashval(clid->cl_id);
1111 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1112 if (same_clid(&clp->cl_clientid, clid))
1118 static struct nfs4_client *
1119 find_unconfirmed_client(clientid_t *clid)
1121 struct nfs4_client *clp;
1122 unsigned int idhashval = clientid_hashval(clid->cl_id);
1124 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1125 if (same_clid(&clp->cl_clientid, clid))
1132 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
1133 * parameter. Matching is based on the fact the at least one of the
1134 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
1136 * FIXME: we need to unify the clientid namespaces for nfsv4.x
1137 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
1138 * and SET_CLIENTID{,_CONFIRM}
1141 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
1143 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
1144 return use_exchange_id == has_exchange_flags;
1147 static struct nfs4_client *
1148 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
1149 bool use_exchange_id)
1151 struct nfs4_client *clp;
1153 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1154 if (same_name(clp->cl_recdir, dname) &&
1155 match_clientid_establishment(clp, use_exchange_id))
1161 static struct nfs4_client *
1162 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
1163 bool use_exchange_id)
1165 struct nfs4_client *clp;
1167 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1168 if (same_name(clp->cl_recdir, dname) &&
1169 match_clientid_establishment(clp, use_exchange_id))
1176 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1178 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1179 unsigned short expected_family;
1181 /* Currently, we only support tcp and tcp6 for the callback channel */
1182 if (se->se_callback_netid_len == 3 &&
1183 !memcmp(se->se_callback_netid_val, "tcp", 3))
1184 expected_family = AF_INET;
1185 else if (se->se_callback_netid_len == 4 &&
1186 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1187 expected_family = AF_INET6;
1191 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1192 se->se_callback_addr_len,
1193 (struct sockaddr *)&conn->cb_addr,
1194 sizeof(conn->cb_addr));
1196 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1199 if (conn->cb_addr.ss_family == AF_INET6)
1200 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1202 conn->cb_prog = se->se_callback_prog;
1203 conn->cb_ident = se->se_callback_ident;
1206 conn->cb_addr.ss_family = AF_UNSPEC;
1207 conn->cb_addrlen = 0;
1208 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1209 "will not receive delegations\n",
1210 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1216 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1219 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1221 struct nfsd4_slot *slot = resp->cstate.slot;
1224 dprintk("--> %s slot %p\n", __func__, slot);
1226 slot->sl_opcnt = resp->opcnt;
1227 slot->sl_status = resp->cstate.status;
1229 if (nfsd4_not_cached(resp)) {
1230 slot->sl_datalen = 0;
1233 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1234 base = (char *)resp->cstate.datap -
1235 (char *)resp->xbuf->head[0].iov_base;
1236 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1238 WARN("%s: sessions DRC could not cache compound\n", __func__);
1243 * Encode the replay sequence operation from the slot values.
1244 * If cachethis is FALSE encode the uncached rep error on the next
1245 * operation which sets resp->p and increments resp->opcnt for
1246 * nfs4svc_encode_compoundres.
1250 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1251 struct nfsd4_compoundres *resp)
1253 struct nfsd4_op *op;
1254 struct nfsd4_slot *slot = resp->cstate.slot;
1256 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1257 resp->opcnt, resp->cstate.slot->sl_cachethis);
1259 /* Encode the replayed sequence operation */
1260 op = &args->ops[resp->opcnt - 1];
1261 nfsd4_encode_operation(resp, op);
1263 /* Return nfserr_retry_uncached_rep in next operation. */
1264 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1265 op = &args->ops[resp->opcnt++];
1266 op->status = nfserr_retry_uncached_rep;
1267 nfsd4_encode_operation(resp, op);
1273 * The sequence operation is not cached because we can use the slot and
1277 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1278 struct nfsd4_sequence *seq)
1280 struct nfsd4_slot *slot = resp->cstate.slot;
1283 dprintk("--> %s slot %p\n", __func__, slot);
1285 /* Either returns 0 or nfserr_retry_uncached */
1286 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1287 if (status == nfserr_retry_uncached_rep)
1290 /* The sequence operation has been encoded, cstate->datap set. */
1291 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1293 resp->opcnt = slot->sl_opcnt;
1294 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1295 status = slot->sl_status;
1301 * Set the exchange_id flags returned by the server.
1304 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1306 /* pNFS is not supported */
1307 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1309 /* Referrals are supported, Migration is not. */
1310 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1312 /* set the wire flags to return to client. */
1313 clid->flags = new->cl_exchange_flags;
1317 nfsd4_exchange_id(struct svc_rqst *rqstp,
1318 struct nfsd4_compound_state *cstate,
1319 struct nfsd4_exchange_id *exid)
1321 struct nfs4_client *unconf, *conf, *new;
1323 unsigned int strhashval;
1324 char dname[HEXDIR_LEN];
1325 char addr_str[INET6_ADDRSTRLEN];
1326 nfs4_verifier verf = exid->verifier;
1327 struct sockaddr *sa = svc_addr(rqstp);
1329 rpc_ntop(sa, addr_str, sizeof(addr_str));
1330 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1331 "ip_addr=%s flags %x, spa_how %d\n",
1332 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1333 addr_str, exid->flags, exid->spa_how);
1335 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1336 return nfserr_inval;
1338 /* Currently only support SP4_NONE */
1339 switch (exid->spa_how) {
1343 return nfserr_encr_alg_unsupp;
1345 BUG(); /* checked by xdr code */
1347 return nfserr_serverfault; /* no excuse :-/ */
1350 status = nfs4_make_rec_clidname(dname, &exid->clname);
1355 strhashval = clientstr_hashval(dname);
1360 conf = find_confirmed_client_by_str(dname, strhashval, true);
1362 if (!same_verf(&verf, &conf->cl_verifier)) {
1363 /* 18.35.4 case 8 */
1364 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1365 status = nfserr_not_same;
1368 /* Client reboot: destroy old state */
1369 expire_client(conf);
1372 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1373 /* 18.35.4 case 9 */
1374 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1375 status = nfserr_perm;
1378 expire_client(conf);
1382 * Set bit when the owner id and verifier map to an already
1383 * confirmed client id (18.35.3).
1385 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1388 * Falling into 18.35.4 case 2, possible router replay.
1389 * Leave confirmed record intact and return same result.
1391 copy_verf(conf, &verf);
1396 /* 18.35.4 case 7 */
1397 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1398 status = nfserr_noent;
1402 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1405 * Possible retry or client restart. Per 18.35.4 case 4,
1406 * a new unconfirmed record should be generated regardless
1407 * of whether any properties have changed.
1409 expire_client(unconf);
1414 new = create_client(exid->clname, dname, rqstp, &verf);
1416 status = nfserr_jukebox;
1421 add_to_unconfirmed(new, strhashval);
1423 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1424 exid->clientid.cl_id = new->cl_clientid.cl_id;
1427 nfsd4_set_ex_flags(new, exid);
1429 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1430 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1434 nfs4_unlock_state();
1436 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1441 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1443 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1446 /* The slot is in use, and no response has been sent. */
1448 if (seqid == slot_seqid)
1449 return nfserr_jukebox;
1451 return nfserr_seq_misordered;
1454 if (likely(seqid == slot_seqid + 1))
1457 if (seqid == slot_seqid)
1458 return nfserr_replay_cache;
1460 if (seqid == 1 && (slot_seqid + 1) == 0)
1462 /* Misordered replay or misordered new request */
1463 return nfserr_seq_misordered;
1467 * Cache the create session result into the create session single DRC
1468 * slot cache by saving the xdr structure. sl_seqid has been set.
1469 * Do this for solo or embedded create session operations.
1472 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1473 struct nfsd4_clid_slot *slot, int nfserr)
1475 slot->sl_status = nfserr;
1476 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1480 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1481 struct nfsd4_clid_slot *slot)
1483 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1484 return slot->sl_status;
1488 nfsd4_create_session(struct svc_rqst *rqstp,
1489 struct nfsd4_compound_state *cstate,
1490 struct nfsd4_create_session *cr_ses)
1492 struct sockaddr *sa = svc_addr(rqstp);
1493 struct nfs4_client *conf, *unconf;
1494 struct nfsd4_session *new;
1495 struct nfsd4_clid_slot *cs_slot = NULL;
1496 bool confirm_me = false;
1500 unconf = find_unconfirmed_client(&cr_ses->clientid);
1501 conf = find_confirmed_client(&cr_ses->clientid);
1504 cs_slot = &conf->cl_cs_slot;
1505 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1506 if (status == nfserr_replay_cache) {
1507 dprintk("Got a create_session replay! seqid= %d\n",
1509 /* Return the cached reply status */
1510 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1512 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1513 status = nfserr_seq_misordered;
1514 dprintk("Sequence misordered!\n");
1515 dprintk("Expected seqid= %d but got seqid= %d\n",
1516 cs_slot->sl_seqid, cr_ses->seqid);
1519 } else if (unconf) {
1520 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1521 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1522 status = nfserr_clid_inuse;
1526 cs_slot = &unconf->cl_cs_slot;
1527 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1529 /* an unconfirmed replay returns misordered */
1530 status = nfserr_seq_misordered;
1537 status = nfserr_stale_clientid;
1542 * XXX: we should probably set this at creation time, and check
1543 * for consistent minorversion use throughout:
1545 conf->cl_minorversion = 1;
1547 * We do not support RDMA or persistent sessions
1549 cr_ses->flags &= ~SESSION4_PERSIST;
1550 cr_ses->flags &= ~SESSION4_RDMA;
1552 status = nfserr_jukebox;
1553 new = alloc_init_session(rqstp, conf, cr_ses);
1557 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1558 NFS4_MAX_SESSIONID_LEN);
1559 cs_slot->sl_seqid++;
1560 cr_ses->seqid = cs_slot->sl_seqid;
1562 /* cache solo and embedded create sessions under the state lock */
1563 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1565 move_to_confirmed(conf);
1567 nfs4_unlock_state();
1568 dprintk("%s returns %d\n", __func__, ntohl(status));
1572 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1574 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1575 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1577 return argp->opcnt == resp->opcnt;
1580 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1584 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1588 nfsd4_destroy_session(struct svc_rqst *r,
1589 struct nfsd4_compound_state *cstate,
1590 struct nfsd4_destroy_session *sessionid)
1592 struct nfsd4_session *ses;
1593 u32 status = nfserr_badsession;
1596 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1597 * - Should we return nfserr_back_chan_busy if waiting for
1598 * callbacks on to-be-destroyed session?
1599 * - Do we need to clear any callback info from previous session?
1602 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1603 if (!nfsd4_last_compound_op(r))
1604 return nfserr_not_only_op;
1606 dump_sessionid(__func__, &sessionid->sessionid);
1607 spin_lock(&client_lock);
1608 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1610 spin_unlock(&client_lock);
1614 unhash_session(ses);
1615 spin_unlock(&client_lock);
1618 /* wait for callbacks */
1619 nfsd4_shutdown_callback(ses->se_client);
1620 nfs4_unlock_state();
1622 nfsd4_del_conns(ses);
1624 nfsd4_put_session(ses);
1627 dprintk("%s returns %d\n", __func__, ntohl(status));
1631 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_rqst *r, struct nfsd4_session *s)
1633 struct nfsd4_conn *c;
1635 list_for_each_entry(c, &s->se_conns, cn_persession) {
1636 if (c->cn_xprt == r->rq_xprt) {
1643 static void nfsd4_sequence_check_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses)
1645 struct nfs4_client *clp = ses->se_client;
1646 struct nfsd4_conn *c, *new = NULL;
1648 spin_lock(&clp->cl_lock);
1649 c = __nfsd4_find_conn(rqstp, ses);
1650 spin_unlock(&clp->cl_lock);
1654 new = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1656 spin_lock(&clp->cl_lock);
1657 c = __nfsd4_find_conn(rqstp, ses);
1659 spin_unlock(&clp->cl_lock);
1663 __nfsd4_hash_conn(new, ses);
1664 spin_unlock(&clp->cl_lock);
1665 nfsd4_register_conn(new);
1670 nfsd4_sequence(struct svc_rqst *rqstp,
1671 struct nfsd4_compound_state *cstate,
1672 struct nfsd4_sequence *seq)
1674 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1675 struct nfsd4_session *session;
1676 struct nfsd4_slot *slot;
1679 if (resp->opcnt != 1)
1680 return nfserr_sequence_pos;
1682 spin_lock(&client_lock);
1683 status = nfserr_badsession;
1684 session = find_in_sessionid_hashtbl(&seq->sessionid);
1688 status = nfserr_badslot;
1689 if (seq->slotid >= session->se_fchannel.maxreqs)
1692 slot = session->se_slots[seq->slotid];
1693 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1695 /* We do not negotiate the number of slots yet, so set the
1696 * maxslots to the session maxreqs which is used to encode
1697 * sr_highest_slotid and the sr_target_slot id to maxslots */
1698 seq->maxslots = session->se_fchannel.maxreqs;
1700 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1701 if (status == nfserr_replay_cache) {
1702 cstate->slot = slot;
1703 cstate->session = session;
1704 /* Return the cached reply status and set cstate->status
1705 * for nfsd4_proc_compound processing */
1706 status = nfsd4_replay_cache_entry(resp, seq);
1707 cstate->status = nfserr_replay_cache;
1713 nfsd4_sequence_check_conn(rqstp, session);
1715 /* Success! bump slot seqid */
1716 slot->sl_inuse = true;
1717 slot->sl_seqid = seq->seqid;
1718 slot->sl_cachethis = seq->cachethis;
1720 cstate->slot = slot;
1721 cstate->session = session;
1724 /* Hold a session reference until done processing the compound. */
1725 if (cstate->session) {
1726 nfsd4_get_session(cstate->session);
1727 atomic_inc(&session->se_client->cl_refcount);
1729 spin_unlock(&client_lock);
1730 dprintk("%s: return %d\n", __func__, ntohl(status));
1735 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1737 if (rc->rca_one_fs) {
1738 if (!cstate->current_fh.fh_dentry)
1739 return nfserr_nofilehandle;
1741 * We don't take advantage of the rca_one_fs case.
1742 * That's OK, it's optional, we can safely ignore it.
1747 if (is_client_expired(cstate->session->se_client)) {
1748 nfs4_unlock_state();
1750 * The following error isn't really legal.
1751 * But we only get here if the client just explicitly
1752 * destroyed the client. Surely it no longer cares what
1753 * error it gets back on an operation for the dead
1756 return nfserr_stale_clientid;
1758 nfsd4_create_clid_dir(cstate->session->se_client);
1759 nfs4_unlock_state();
1764 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1765 struct nfsd4_setclientid *setclid)
1767 struct sockaddr *sa = svc_addr(rqstp);
1768 struct xdr_netobj clname = {
1769 .len = setclid->se_namelen,
1770 .data = setclid->se_name,
1772 nfs4_verifier clverifier = setclid->se_verf;
1773 unsigned int strhashval;
1774 struct nfs4_client *conf, *unconf, *new;
1776 char dname[HEXDIR_LEN];
1778 if (!check_name(clname))
1779 return nfserr_inval;
1781 status = nfs4_make_rec_clidname(dname, &clname);
1786 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1787 * We get here on a DRC miss.
1790 strhashval = clientstr_hashval(dname);
1793 conf = find_confirmed_client_by_str(dname, strhashval, false);
1795 /* RFC 3530 14.2.33 CASE 0: */
1796 status = nfserr_clid_inuse;
1797 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1798 char addr_str[INET6_ADDRSTRLEN];
1799 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1801 dprintk("NFSD: setclientid: string in use by client "
1802 "at %s\n", addr_str);
1807 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1808 * has a description of SETCLIENTID request processing consisting
1809 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1811 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1812 status = nfserr_resource;
1815 * RFC 3530 14.2.33 CASE 4:
1816 * placed first, because it is the normal case
1819 expire_client(unconf);
1820 new = create_client(clname, dname, rqstp, &clverifier);
1824 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1826 * RFC 3530 14.2.33 CASE 1:
1827 * probable callback update
1830 /* Note this is removing unconfirmed {*x***},
1831 * which is stronger than RFC recommended {vxc**}.
1832 * This has the advantage that there is at most
1833 * one {*x***} in either list at any time.
1835 expire_client(unconf);
1837 new = create_client(clname, dname, rqstp, &clverifier);
1840 copy_clid(new, conf);
1841 } else if (!unconf) {
1843 * RFC 3530 14.2.33 CASE 2:
1844 * probable client reboot; state will be removed if
1847 new = create_client(clname, dname, rqstp, &clverifier);
1853 * RFC 3530 14.2.33 CASE 3:
1854 * probable client reboot; state will be removed if
1857 expire_client(unconf);
1858 new = create_client(clname, dname, rqstp, &clverifier);
1864 * XXX: we should probably set this at creation time, and check
1865 * for consistent minorversion use throughout:
1867 new->cl_minorversion = 0;
1868 gen_callback(new, setclid, rpc_get_scope_id(sa));
1869 add_to_unconfirmed(new, strhashval);
1870 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1871 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1872 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1875 nfs4_unlock_state();
1881 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1882 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1883 * bullets, labeled as CASE1 - CASE4 below.
1886 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1887 struct nfsd4_compound_state *cstate,
1888 struct nfsd4_setclientid_confirm *setclientid_confirm)
1890 struct sockaddr *sa = svc_addr(rqstp);
1891 struct nfs4_client *conf, *unconf;
1892 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1893 clientid_t * clid = &setclientid_confirm->sc_clientid;
1896 if (STALE_CLIENTID(clid))
1897 return nfserr_stale_clientid;
1899 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1900 * We get here on a DRC miss.
1905 conf = find_confirmed_client(clid);
1906 unconf = find_unconfirmed_client(clid);
1908 status = nfserr_clid_inuse;
1909 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1911 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1915 * section 14.2.34 of RFC 3530 has a description of
1916 * SETCLIENTID_CONFIRM request processing consisting
1917 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1919 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1921 * RFC 3530 14.2.34 CASE 1:
1924 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1925 status = nfserr_clid_inuse;
1927 atomic_set(&conf->cl_cb_set, 0);
1928 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
1929 nfsd4_probe_callback(conf);
1930 expire_client(unconf);
1934 } else if (conf && !unconf) {
1936 * RFC 3530 14.2.34 CASE 2:
1937 * probable retransmitted request; play it safe and
1940 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1941 status = nfserr_clid_inuse;
1944 } else if (!conf && unconf
1945 && same_verf(&unconf->cl_confirm, &confirm)) {
1947 * RFC 3530 14.2.34 CASE 3:
1948 * Normal case; new or rebooted client:
1950 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1951 status = nfserr_clid_inuse;
1954 clientstr_hashval(unconf->cl_recdir);
1955 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1958 nfsd4_remove_clid_dir(conf);
1959 expire_client(conf);
1961 move_to_confirmed(unconf);
1963 nfsd4_probe_callback(conf);
1966 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1967 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1970 * RFC 3530 14.2.34 CASE 4:
1971 * Client probably hasn't noticed that we rebooted yet.
1973 status = nfserr_stale_clientid;
1975 /* check that we have hit one of the cases...*/
1976 status = nfserr_clid_inuse;
1979 nfs4_unlock_state();
1983 /* OPEN Share state helper functions */
1984 static inline struct nfs4_file *
1985 alloc_init_file(struct inode *ino)
1987 struct nfs4_file *fp;
1988 unsigned int hashval = file_hashval(ino);
1990 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1992 atomic_set(&fp->fi_ref, 1);
1993 INIT_LIST_HEAD(&fp->fi_hash);
1994 INIT_LIST_HEAD(&fp->fi_stateids);
1995 INIT_LIST_HEAD(&fp->fi_delegations);
1996 fp->fi_inode = igrab(ino);
1997 fp->fi_id = current_fileid++;
1998 fp->fi_had_conflict = false;
1999 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2000 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2001 spin_lock(&recall_lock);
2002 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2003 spin_unlock(&recall_lock);
2010 nfsd4_free_slab(struct kmem_cache **slab)
2014 kmem_cache_destroy(*slab);
2019 nfsd4_free_slabs(void)
2021 nfsd4_free_slab(&stateowner_slab);
2022 nfsd4_free_slab(&file_slab);
2023 nfsd4_free_slab(&stateid_slab);
2024 nfsd4_free_slab(&deleg_slab);
2028 nfsd4_init_slabs(void)
2030 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2031 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2032 if (stateowner_slab == NULL)
2034 file_slab = kmem_cache_create("nfsd4_files",
2035 sizeof(struct nfs4_file), 0, 0, NULL);
2036 if (file_slab == NULL)
2038 stateid_slab = kmem_cache_create("nfsd4_stateids",
2039 sizeof(struct nfs4_stateid), 0, 0, NULL);
2040 if (stateid_slab == NULL)
2042 deleg_slab = kmem_cache_create("nfsd4_delegations",
2043 sizeof(struct nfs4_delegation), 0, 0, NULL);
2044 if (deleg_slab == NULL)
2049 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2054 nfs4_free_stateowner(struct kref *kref)
2056 struct nfs4_stateowner *sop =
2057 container_of(kref, struct nfs4_stateowner, so_ref);
2058 kfree(sop->so_owner.data);
2059 kmem_cache_free(stateowner_slab, sop);
2062 static inline struct nfs4_stateowner *
2063 alloc_stateowner(struct xdr_netobj *owner)
2065 struct nfs4_stateowner *sop;
2067 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2068 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2069 memcpy(sop->so_owner.data, owner->data, owner->len);
2070 sop->so_owner.len = owner->len;
2071 kref_init(&sop->so_ref);
2074 kmem_cache_free(stateowner_slab, sop);
2079 static struct nfs4_stateowner *
2080 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2081 struct nfs4_stateowner *sop;
2082 struct nfs4_replay *rp;
2083 unsigned int idhashval;
2085 if (!(sop = alloc_stateowner(&open->op_owner)))
2087 idhashval = ownerid_hashval(current_ownerid);
2088 INIT_LIST_HEAD(&sop->so_idhash);
2089 INIT_LIST_HEAD(&sop->so_strhash);
2090 INIT_LIST_HEAD(&sop->so_perclient);
2091 INIT_LIST_HEAD(&sop->so_stateids);
2092 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2093 INIT_LIST_HEAD(&sop->so_close_lru);
2095 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2096 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2097 list_add(&sop->so_perclient, &clp->cl_openowners);
2098 sop->so_is_open_owner = 1;
2099 sop->so_id = current_ownerid++;
2100 sop->so_client = clp;
2101 sop->so_seqid = open->op_seqid;
2102 sop->so_confirmed = 0;
2103 rp = &sop->so_replay;
2104 rp->rp_status = nfserr_serverfault;
2106 rp->rp_buf = rp->rp_ibuf;
2111 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2112 struct nfs4_stateowner *sop = open->op_stateowner;
2113 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2115 INIT_LIST_HEAD(&stp->st_hash);
2116 INIT_LIST_HEAD(&stp->st_perstateowner);
2117 INIT_LIST_HEAD(&stp->st_lockowners);
2118 INIT_LIST_HEAD(&stp->st_perfile);
2119 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2120 list_add(&stp->st_perstateowner, &sop->so_stateids);
2121 list_add(&stp->st_perfile, &fp->fi_stateids);
2122 stp->st_stateowner = sop;
2125 stp->st_stateid.si_boot = boot_time;
2126 stp->st_stateid.si_stateownerid = sop->so_id;
2127 stp->st_stateid.si_fileid = fp->fi_id;
2128 stp->st_stateid.si_generation = 0;
2129 stp->st_access_bmap = 0;
2130 stp->st_deny_bmap = 0;
2131 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2132 &stp->st_access_bmap);
2133 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2134 stp->st_openstp = NULL;
2138 move_to_close_lru(struct nfs4_stateowner *sop)
2140 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2142 list_move_tail(&sop->so_close_lru, &close_lru);
2143 sop->so_time = get_seconds();
2147 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2150 return (sop->so_owner.len == owner->len) &&
2151 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2152 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2155 static struct nfs4_stateowner *
2156 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2158 struct nfs4_stateowner *so = NULL;
2160 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2161 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2167 /* search file_hashtbl[] for file */
2168 static struct nfs4_file *
2169 find_file(struct inode *ino)
2171 unsigned int hashval = file_hashval(ino);
2172 struct nfs4_file *fp;
2174 spin_lock(&recall_lock);
2175 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2176 if (fp->fi_inode == ino) {
2178 spin_unlock(&recall_lock);
2182 spin_unlock(&recall_lock);
2186 static inline int access_valid(u32 x, u32 minorversion)
2188 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2190 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2192 x &= ~NFS4_SHARE_ACCESS_MASK;
2193 if (minorversion && x) {
2194 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2196 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2198 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2205 static inline int deny_valid(u32 x)
2207 /* Note: unlike access bits, deny bits may be zero. */
2208 return x <= NFS4_SHARE_DENY_BOTH;
2212 * Called to check deny when READ with all zero stateid or
2213 * WRITE with all zero or all one stateid
2216 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2218 struct inode *ino = current_fh->fh_dentry->d_inode;
2219 struct nfs4_file *fp;
2220 struct nfs4_stateid *stp;
2223 dprintk("NFSD: nfs4_share_conflict\n");
2225 fp = find_file(ino);
2228 ret = nfserr_locked;
2229 /* Search for conflicting share reservations */
2230 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2231 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2232 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2242 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2244 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2245 nfs4_file_put_access(fp, O_WRONLY);
2246 if (share_access & NFS4_SHARE_ACCESS_READ)
2247 nfs4_file_put_access(fp, O_RDONLY);
2251 * Spawn a thread to perform a recall on the delegation represented
2252 * by the lease (file_lock)
2254 * Called from break_lease() with lock_kernel() held.
2255 * Note: we assume break_lease will only call this *once* for any given
2259 void nfsd_break_deleg_cb(struct file_lock *fl)
2261 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2263 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2267 /* We're assuming the state code never drops its reference
2268 * without first removing the lease. Since we're in this lease
2269 * callback (and since the lease code is serialized by the kernel
2270 * lock) we know the server hasn't removed the lease yet, we know
2271 * it's safe to take a reference: */
2272 atomic_inc(&dp->dl_count);
2274 spin_lock(&recall_lock);
2275 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2276 spin_unlock(&recall_lock);
2278 /* only place dl_time is set. protected by lock_kernel*/
2279 dp->dl_time = get_seconds();
2282 * We don't want the locks code to timeout the lease for us;
2283 * we'll remove it ourself if the delegation isn't returned
2286 fl->fl_break_time = 0;
2288 dp->dl_file->fi_had_conflict = true;
2289 nfsd4_cb_recall(dp);
2293 * The file_lock is being reapd.
2295 * Called by locks_free_lock() with lock_kernel() held.
2298 void nfsd_release_deleg_cb(struct file_lock *fl)
2300 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2302 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2304 if (!(fl->fl_flags & FL_LEASE) || !dp)
2306 dp->dl_flock = NULL;
2310 * Set the delegation file_lock back pointer.
2312 * Called from setlease() with lock_kernel() held.
2315 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2317 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2319 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2326 * Called from setlease() with lock_kernel() held
2329 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2331 struct nfs4_delegation *onlistd =
2332 (struct nfs4_delegation *)onlist->fl_owner;
2333 struct nfs4_delegation *tryd =
2334 (struct nfs4_delegation *)try->fl_owner;
2336 if (onlist->fl_lmops != try->fl_lmops)
2339 return onlistd->dl_client == tryd->dl_client;
2344 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2347 return lease_modify(onlist, arg);
2352 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2353 .fl_break = nfsd_break_deleg_cb,
2354 .fl_release_private = nfsd_release_deleg_cb,
2355 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2356 .fl_mylease = nfsd_same_client_deleg_cb,
2357 .fl_change = nfsd_change_deleg_cb,
2362 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2363 struct nfsd4_open *open)
2365 clientid_t *clientid = &open->op_clientid;
2366 struct nfs4_client *clp = NULL;
2367 unsigned int strhashval;
2368 struct nfs4_stateowner *sop = NULL;
2370 if (!check_name(open->op_owner))
2371 return nfserr_inval;
2373 if (STALE_CLIENTID(&open->op_clientid))
2374 return nfserr_stale_clientid;
2376 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2377 sop = find_openstateowner_str(strhashval, open);
2378 open->op_stateowner = sop;
2380 /* Make sure the client's lease hasn't expired. */
2381 clp = find_confirmed_client(clientid);
2383 return nfserr_expired;
2386 /* When sessions are used, skip open sequenceid processing */
2387 if (nfsd4_has_session(cstate))
2389 if (!sop->so_confirmed) {
2390 /* Replace unconfirmed owners without checking for replay. */
2391 clp = sop->so_client;
2392 release_openowner(sop);
2393 open->op_stateowner = NULL;
2396 if (open->op_seqid == sop->so_seqid - 1) {
2397 if (sop->so_replay.rp_buflen)
2398 return nfserr_replay_me;
2399 /* The original OPEN failed so spectacularly
2400 * that we don't even have replay data saved!
2401 * Therefore, we have no choice but to continue
2402 * processing this OPEN; presumably, we'll
2403 * fail again for the same reason.
2405 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2408 if (open->op_seqid != sop->so_seqid)
2409 return nfserr_bad_seqid;
2411 if (open->op_stateowner == NULL) {
2412 sop = alloc_init_open_stateowner(strhashval, clp, open);
2414 return nfserr_resource;
2415 open->op_stateowner = sop;
2417 list_del_init(&sop->so_close_lru);
2418 renew_client(sop->so_client);
2422 static inline __be32
2423 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2425 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2426 return nfserr_openmode;
2431 static struct nfs4_delegation *
2432 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2434 struct nfs4_delegation *dp;
2436 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2437 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2443 int share_access_to_flags(u32 share_access)
2445 share_access &= ~NFS4_SHARE_WANT_MASK;
2447 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2451 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2452 struct nfs4_delegation **dp)
2455 __be32 status = nfserr_bad_stateid;
2457 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2460 flags = share_access_to_flags(open->op_share_access);
2461 status = nfs4_check_delegmode(*dp, flags);
2465 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2469 open->op_stateowner->so_confirmed = 1;
2474 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2476 struct nfs4_stateid *local;
2477 __be32 status = nfserr_share_denied;
2478 struct nfs4_stateowner *sop = open->op_stateowner;
2480 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2481 /* ignore lock owners */
2482 if (local->st_stateowner->so_is_open_owner == 0)
2484 /* remember if we have seen this open owner */
2485 if (local->st_stateowner == sop)
2487 /* check for conflicting share reservations */
2488 if (!test_share(local, open))
2496 static inline struct nfs4_stateid *
2497 nfs4_alloc_stateid(void)
2499 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2502 static inline int nfs4_access_to_access(u32 nfs4_access)
2506 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2507 flags |= NFSD_MAY_READ;
2508 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2509 flags |= NFSD_MAY_WRITE;
2513 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2514 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2517 int oflag = nfs4_access_to_omode(nfs4_access);
2518 int access = nfs4_access_to_access(nfs4_access);
2520 if (!fp->fi_fds[oflag]) {
2521 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2522 &fp->fi_fds[oflag]);
2523 if (status == nfserr_dropit)
2524 status = nfserr_jukebox;
2528 nfs4_file_get_access(fp, oflag);
2534 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2535 struct nfs4_file *fp, struct svc_fh *cur_fh,
2536 struct nfsd4_open *open)
2538 struct nfs4_stateid *stp;
2541 stp = nfs4_alloc_stateid();
2543 return nfserr_resource;
2545 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2547 kmem_cache_free(stateid_slab, stp);
2554 static inline __be32
2555 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2556 struct nfsd4_open *open)
2558 struct iattr iattr = {
2559 .ia_valid = ATTR_SIZE,
2562 if (!open->op_truncate)
2564 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2565 return nfserr_inval;
2566 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2570 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2572 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2576 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2578 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2582 status = nfsd4_truncate(rqstp, cur_fh, open);
2585 int oflag = nfs4_access_to_omode(new_access);
2586 nfs4_file_put_access(fp, oflag);
2590 /* remember the open */
2591 __set_bit(op_share_access, &stp->st_access_bmap);
2592 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2599 nfs4_set_claim_prev(struct nfsd4_open *open)
2601 open->op_stateowner->so_confirmed = 1;
2602 open->op_stateowner->so_client->cl_firststate = 1;
2606 * Attempt to hand out a delegation.
2609 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2611 struct nfs4_delegation *dp;
2612 struct nfs4_stateowner *sop = stp->st_stateowner;
2613 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2614 struct file_lock fl, *flp = &fl;
2615 int status, flag = 0;
2617 flag = NFS4_OPEN_DELEGATE_NONE;
2618 open->op_recall = 0;
2619 switch (open->op_claim_type) {
2620 case NFS4_OPEN_CLAIM_PREVIOUS:
2622 open->op_recall = 1;
2623 flag = open->op_delegate_type;
2624 if (flag == NFS4_OPEN_DELEGATE_NONE)
2627 case NFS4_OPEN_CLAIM_NULL:
2628 /* Let's not give out any delegations till everyone's
2629 * had the chance to reclaim theirs.... */
2630 if (locks_in_grace())
2632 if (!cb_up || !sop->so_confirmed)
2634 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2635 flag = NFS4_OPEN_DELEGATE_WRITE;
2637 flag = NFS4_OPEN_DELEGATE_READ;
2643 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2645 flag = NFS4_OPEN_DELEGATE_NONE;
2648 locks_init_lock(&fl);
2649 fl.fl_lmops = &nfsd_lease_mng_ops;
2650 fl.fl_flags = FL_LEASE;
2651 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2652 fl.fl_end = OFFSET_MAX;
2653 fl.fl_owner = (fl_owner_t)dp;
2654 fl.fl_file = find_readable_file(stp->st_file);
2655 BUG_ON(!fl.fl_file);
2656 fl.fl_pid = current->tgid;
2658 /* vfs_setlease checks to see if delegation should be handed out.
2659 * the lock_manager callbacks fl_mylease and fl_change are used
2661 if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
2662 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2663 unhash_delegation(dp);
2664 flag = NFS4_OPEN_DELEGATE_NONE;
2668 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2670 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2671 STATEID_VAL(&dp->dl_stateid));
2673 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2674 && flag == NFS4_OPEN_DELEGATE_NONE
2675 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2676 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2677 open->op_delegate_type = flag;
2681 * called with nfs4_lock_state() held.
2684 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2686 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2687 struct nfs4_file *fp = NULL;
2688 struct inode *ino = current_fh->fh_dentry->d_inode;
2689 struct nfs4_stateid *stp = NULL;
2690 struct nfs4_delegation *dp = NULL;
2693 status = nfserr_inval;
2694 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2695 || !deny_valid(open->op_share_deny))
2698 * Lookup file; if found, lookup stateid and check open request,
2699 * and check for delegations in the process of being recalled.
2700 * If not found, create the nfs4_file struct
2702 fp = find_file(ino);
2704 if ((status = nfs4_check_open(fp, open, &stp)))
2706 status = nfs4_check_deleg(fp, open, &dp);
2710 status = nfserr_bad_stateid;
2711 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2713 status = nfserr_resource;
2714 fp = alloc_init_file(ino);
2720 * OPEN the file, or upgrade an existing OPEN.
2721 * If truncate fails, the OPEN fails.
2724 /* Stateid was found, this is an OPEN upgrade */
2725 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2728 update_stateid(&stp->st_stateid);
2730 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2733 init_stateid(stp, fp, open);
2734 status = nfsd4_truncate(rqstp, current_fh, open);
2736 release_open_stateid(stp);
2739 if (nfsd4_has_session(&resp->cstate))
2740 update_stateid(&stp->st_stateid);
2742 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2744 if (nfsd4_has_session(&resp->cstate))
2745 open->op_stateowner->so_confirmed = 1;
2748 * Attempt to hand out a delegation. No error return, because the
2749 * OPEN succeeds even if we fail.
2751 nfs4_open_delegation(current_fh, open, stp);
2755 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2756 STATEID_VAL(&stp->st_stateid));
2760 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2761 nfs4_set_claim_prev(open);
2763 * To finish the open response, we just need to set the rflags.
2765 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2766 if (!open->op_stateowner->so_confirmed &&
2767 !nfsd4_has_session(&resp->cstate))
2768 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2774 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2777 struct nfs4_client *clp;
2781 dprintk("process_renew(%08x/%08x): starting\n",
2782 clid->cl_boot, clid->cl_id);
2783 status = nfserr_stale_clientid;
2784 if (STALE_CLIENTID(clid))
2786 clp = find_confirmed_client(clid);
2787 status = nfserr_expired;
2789 /* We assume the client took too long to RENEW. */
2790 dprintk("nfsd4_renew: clientid not found!\n");
2794 status = nfserr_cb_path_down;
2795 if (!list_empty(&clp->cl_delegations)
2796 && !atomic_read(&clp->cl_cb_set))
2800 nfs4_unlock_state();
2804 struct lock_manager nfsd4_manager = {
2808 nfsd4_end_grace(void)
2810 dprintk("NFSD: end of grace period\n");
2811 nfsd4_recdir_purge_old();
2812 locks_end_grace(&nfsd4_manager);
2814 * Now that every NFSv4 client has had the chance to recover and
2815 * to see the (possibly new, possibly shorter) lease time, we
2816 * can safely set the next grace time to the current lease time:
2818 nfsd4_grace = nfsd4_lease;
2822 nfs4_laundromat(void)
2824 struct nfs4_client *clp;
2825 struct nfs4_stateowner *sop;
2826 struct nfs4_delegation *dp;
2827 struct list_head *pos, *next, reaplist;
2828 time_t cutoff = get_seconds() - nfsd4_lease;
2829 time_t t, clientid_val = nfsd4_lease;
2830 time_t u, test_val = nfsd4_lease;
2834 dprintk("NFSD: laundromat service - starting\n");
2835 if (locks_in_grace())
2837 INIT_LIST_HEAD(&reaplist);
2838 spin_lock(&client_lock);
2839 list_for_each_safe(pos, next, &client_lru) {
2840 clp = list_entry(pos, struct nfs4_client, cl_lru);
2841 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2842 t = clp->cl_time - cutoff;
2843 if (clientid_val > t)
2847 if (atomic_read(&clp->cl_refcount)) {
2848 dprintk("NFSD: client in use (clientid %08x)\n",
2849 clp->cl_clientid.cl_id);
2852 unhash_client_locked(clp);
2853 list_add(&clp->cl_lru, &reaplist);
2855 spin_unlock(&client_lock);
2856 list_for_each_safe(pos, next, &reaplist) {
2857 clp = list_entry(pos, struct nfs4_client, cl_lru);
2858 dprintk("NFSD: purging unused client (clientid %08x)\n",
2859 clp->cl_clientid.cl_id);
2860 nfsd4_remove_clid_dir(clp);
2863 spin_lock(&recall_lock);
2864 list_for_each_safe(pos, next, &del_recall_lru) {
2865 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2866 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2867 u = dp->dl_time - cutoff;
2872 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2874 list_move(&dp->dl_recall_lru, &reaplist);
2876 spin_unlock(&recall_lock);
2877 list_for_each_safe(pos, next, &reaplist) {
2878 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2879 list_del_init(&dp->dl_recall_lru);
2880 unhash_delegation(dp);
2882 test_val = nfsd4_lease;
2883 list_for_each_safe(pos, next, &close_lru) {
2884 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2885 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2886 u = sop->so_time - cutoff;
2891 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2893 release_openowner(sop);
2895 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2896 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2897 nfs4_unlock_state();
2898 return clientid_val;
2901 static struct workqueue_struct *laundry_wq;
2902 static void laundromat_main(struct work_struct *);
2903 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2906 laundromat_main(struct work_struct *not_used)
2910 t = nfs4_laundromat();
2911 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2912 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2915 static struct nfs4_stateowner *
2916 search_close_lru(u32 st_id, int flags)
2918 struct nfs4_stateowner *local = NULL;
2920 if (flags & CLOSE_STATE) {
2921 list_for_each_entry(local, &close_lru, so_close_lru) {
2922 if (local->so_id == st_id)
2930 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2932 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2936 STALE_STATEID(stateid_t *stateid)
2938 if (stateid->si_boot == boot_time)
2940 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2941 STATEID_VAL(stateid));
2946 access_permit_read(unsigned long access_bmap)
2948 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2949 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2950 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2954 access_permit_write(unsigned long access_bmap)
2956 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2957 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2961 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2963 __be32 status = nfserr_openmode;
2965 /* For lock stateid's, we test the parent open, not the lock: */
2966 if (stp->st_openstp)
2967 stp = stp->st_openstp;
2968 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2970 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2977 static inline __be32
2978 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2980 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2982 else if (locks_in_grace()) {
2983 /* Answer in remaining cases depends on existance of
2984 * conflicting state; so we must wait out the grace period. */
2985 return nfserr_grace;
2986 } else if (flags & WR_STATE)
2987 return nfs4_share_conflict(current_fh,
2988 NFS4_SHARE_DENY_WRITE);
2989 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2990 return nfs4_share_conflict(current_fh,
2991 NFS4_SHARE_DENY_READ);
2995 * Allow READ/WRITE during grace period on recovered state only for files
2996 * that are not able to provide mandatory locking.
2999 grace_disallows_io(struct inode *inode)
3001 return locks_in_grace() && mandatory_lock(inode);
3004 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3007 * When sessions are used the stateid generation number is ignored
3010 if ((flags & HAS_SESSION) && in->si_generation == 0)
3013 /* If the client sends us a stateid from the future, it's buggy: */
3014 if (in->si_generation > ref->si_generation)
3015 return nfserr_bad_stateid;
3017 * The following, however, can happen. For example, if the
3018 * client sends an open and some IO at the same time, the open
3019 * may bump si_generation while the IO is still in flight.
3020 * Thanks to hard links and renames, the client never knows what
3021 * file an open will affect. So it could avoid that situation
3022 * only by serializing all opens and IO from the same open
3023 * owner. To recover from the old_stateid error, the client
3024 * will just have to retry the IO:
3026 if (in->si_generation < ref->si_generation)
3027 return nfserr_old_stateid;
3032 static int is_delegation_stateid(stateid_t *stateid)
3034 return stateid->si_fileid == 0;
3038 * Checks for stateid operations
3041 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3042 stateid_t *stateid, int flags, struct file **filpp)
3044 struct nfs4_stateid *stp = NULL;
3045 struct nfs4_delegation *dp = NULL;
3046 struct svc_fh *current_fh = &cstate->current_fh;
3047 struct inode *ino = current_fh->fh_dentry->d_inode;
3053 if (grace_disallows_io(ino))
3054 return nfserr_grace;
3056 if (nfsd4_has_session(cstate))
3057 flags |= HAS_SESSION;
3059 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3060 return check_special_stateids(current_fh, stateid, flags);
3062 status = nfserr_stale_stateid;
3063 if (STALE_STATEID(stateid))
3067 * We assume that any stateid that has the current boot time,
3068 * but that we can't find, is expired:
3070 status = nfserr_expired;
3071 if (is_delegation_stateid(stateid)) {
3072 dp = find_delegation_stateid(ino, stateid);
3075 status = check_stateid_generation(stateid, &dp->dl_stateid,
3079 status = nfs4_check_delegmode(dp, flags);
3082 renew_client(dp->dl_client);
3084 *filpp = find_readable_file(dp->dl_file);
3086 } else { /* open or lock stateid */
3087 stp = find_stateid(stateid, flags);
3090 status = nfserr_bad_stateid;
3091 if (nfs4_check_fh(current_fh, stp))
3093 if (!stp->st_stateowner->so_confirmed)
3095 status = check_stateid_generation(stateid, &stp->st_stateid,
3099 status = nfs4_check_openmode(stp, flags);
3102 renew_client(stp->st_stateowner->so_client);
3104 if (flags & RD_STATE)
3105 *filpp = find_readable_file(stp->st_file);
3107 *filpp = find_writeable_file(stp->st_file);
3118 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3119 RD_STATE : WR_STATE;
3123 * Checks for sequence id mutating operations.
3126 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3127 stateid_t *stateid, int flags,
3128 struct nfs4_stateowner **sopp,
3129 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3131 struct nfs4_stateid *stp;
3132 struct nfs4_stateowner *sop;
3133 struct svc_fh *current_fh = &cstate->current_fh;
3136 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3137 seqid, STATEID_VAL(stateid));
3142 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3143 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3144 return nfserr_bad_stateid;
3147 if (STALE_STATEID(stateid))
3148 return nfserr_stale_stateid;
3150 if (nfsd4_has_session(cstate))
3151 flags |= HAS_SESSION;
3154 * We return BAD_STATEID if filehandle doesn't match stateid,
3155 * the confirmed flag is incorrecly set, or the generation
3156 * number is incorrect.
3158 stp = find_stateid(stateid, flags);
3161 * Also, we should make sure this isn't just the result of
3164 sop = search_close_lru(stateid->si_stateownerid, flags);
3165 /* It's not stale; let's assume it's expired: */
3167 return nfserr_expired;
3173 *sopp = sop = stp->st_stateowner;
3176 clientid_t *lockclid = &lock->v.new.clientid;
3177 struct nfs4_client *clp = sop->so_client;
3181 lkflg = setlkflg(lock->lk_type);
3183 if (lock->lk_is_new) {
3184 if (!sop->so_is_open_owner)
3185 return nfserr_bad_stateid;
3186 if (!(flags & HAS_SESSION) &&
3187 !same_clid(&clp->cl_clientid, lockclid))
3188 return nfserr_bad_stateid;
3189 /* stp is the open stateid */
3190 status = nfs4_check_openmode(stp, lkflg);
3194 /* stp is the lock stateid */
3195 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3201 if (nfs4_check_fh(current_fh, stp)) {
3202 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3203 return nfserr_bad_stateid;
3207 * We now validate the seqid and stateid generation numbers.
3208 * For the moment, we ignore the possibility of
3209 * generation number wraparound.
3211 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3214 if (sop->so_confirmed && flags & CONFIRM) {
3215 dprintk("NFSD: preprocess_seqid_op: expected"
3216 " unconfirmed stateowner!\n");
3217 return nfserr_bad_stateid;
3219 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3220 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3221 " confirmed yet!\n");
3222 return nfserr_bad_stateid;
3224 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3227 renew_client(sop->so_client);
3231 if (seqid == sop->so_seqid - 1) {
3232 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3233 /* indicate replay to calling function */
3234 return nfserr_replay_me;
3236 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3237 sop->so_seqid, seqid);
3239 return nfserr_bad_seqid;
3243 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3244 struct nfsd4_open_confirm *oc)
3247 struct nfs4_stateowner *sop;
3248 struct nfs4_stateid *stp;
3250 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3251 (int)cstate->current_fh.fh_dentry->d_name.len,
3252 cstate->current_fh.fh_dentry->d_name.name);
3254 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3260 if ((status = nfs4_preprocess_seqid_op(cstate,
3261 oc->oc_seqid, &oc->oc_req_stateid,
3262 CONFIRM | OPEN_STATE,
3263 &oc->oc_stateowner, &stp, NULL)))
3266 sop = oc->oc_stateowner;
3267 sop->so_confirmed = 1;
3268 update_stateid(&stp->st_stateid);
3269 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3270 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3271 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3273 nfsd4_create_clid_dir(sop->so_client);
3275 if (oc->oc_stateowner) {
3276 nfs4_get_stateowner(oc->oc_stateowner);
3277 cstate->replay_owner = oc->oc_stateowner;
3279 nfs4_unlock_state();
3285 * unset all bits in union bitmap (bmap) that
3286 * do not exist in share (from successful OPEN_DOWNGRADE)
3289 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3292 for (i = 1; i < 4; i++) {
3293 if ((i & access) != i)
3294 __clear_bit(i, bmap);
3299 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3302 for (i = 0; i < 4; i++) {
3303 if ((i & deny) != i)
3304 __clear_bit(i, bmap);
3309 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3310 struct nfsd4_compound_state *cstate,
3311 struct nfsd4_open_downgrade *od)
3314 struct nfs4_stateid *stp;
3315 unsigned int share_access;
3317 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3318 (int)cstate->current_fh.fh_dentry->d_name.len,
3319 cstate->current_fh.fh_dentry->d_name.name);
3321 if (!access_valid(od->od_share_access, cstate->minorversion)
3322 || !deny_valid(od->od_share_deny))
3323 return nfserr_inval;
3326 if ((status = nfs4_preprocess_seqid_op(cstate,
3330 &od->od_stateowner, &stp, NULL)))
3333 status = nfserr_inval;
3334 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3335 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3336 stp->st_access_bmap, od->od_share_access);
3339 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3340 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3341 stp->st_deny_bmap, od->od_share_deny);
3344 set_access(&share_access, stp->st_access_bmap);
3345 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3347 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3348 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3350 update_stateid(&stp->st_stateid);
3351 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3354 if (od->od_stateowner) {
3355 nfs4_get_stateowner(od->od_stateowner);
3356 cstate->replay_owner = od->od_stateowner;
3358 nfs4_unlock_state();
3363 * nfs4_unlock_state() called after encode
3366 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3367 struct nfsd4_close *close)
3370 struct nfs4_stateid *stp;
3372 dprintk("NFSD: nfsd4_close on file %.*s\n",
3373 (int)cstate->current_fh.fh_dentry->d_name.len,
3374 cstate->current_fh.fh_dentry->d_name.name);
3377 /* check close_lru for replay */
3378 if ((status = nfs4_preprocess_seqid_op(cstate,
3381 OPEN_STATE | CLOSE_STATE,
3382 &close->cl_stateowner, &stp, NULL)))
3385 update_stateid(&stp->st_stateid);
3386 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3388 /* release_stateid() calls nfsd_close() if needed */
3389 release_open_stateid(stp);
3391 /* place unused nfs4_stateowners on so_close_lru list to be
3392 * released by the laundromat service after the lease period
3393 * to enable us to handle CLOSE replay
3395 if (list_empty(&close->cl_stateowner->so_stateids))
3396 move_to_close_lru(close->cl_stateowner);
3398 if (close->cl_stateowner) {
3399 nfs4_get_stateowner(close->cl_stateowner);
3400 cstate->replay_owner = close->cl_stateowner;
3402 nfs4_unlock_state();
3407 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3408 struct nfsd4_delegreturn *dr)
3410 struct nfs4_delegation *dp;
3411 stateid_t *stateid = &dr->dr_stateid;
3412 struct inode *inode;
3416 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3418 inode = cstate->current_fh.fh_dentry->d_inode;
3420 if (nfsd4_has_session(cstate))
3421 flags |= HAS_SESSION;
3423 status = nfserr_bad_stateid;
3424 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3426 status = nfserr_stale_stateid;
3427 if (STALE_STATEID(stateid))
3429 status = nfserr_bad_stateid;
3430 if (!is_delegation_stateid(stateid))
3432 status = nfserr_expired;
3433 dp = find_delegation_stateid(inode, stateid);
3436 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3439 renew_client(dp->dl_client);
3441 unhash_delegation(dp);
3443 nfs4_unlock_state();
3450 * Lock owner state (byte-range locks)
3452 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3453 #define LOCK_HASH_BITS 8
3454 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3455 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3458 end_offset(u64 start, u64 len)
3463 return end >= start ? end: NFS4_MAX_UINT64;
3466 /* last octet in a range */
3468 last_byte_offset(u64 start, u64 len)
3474 return end > start ? end - 1: NFS4_MAX_UINT64;
3477 #define lockownerid_hashval(id) \
3478 ((id) & LOCK_HASH_MASK)
3480 static inline unsigned int
3481 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3482 struct xdr_netobj *ownername)
3484 return (file_hashval(inode) + cl_id
3485 + opaque_hashval(ownername->data, ownername->len))
3489 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3490 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3491 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3493 static struct nfs4_stateid *
3494 find_stateid(stateid_t *stid, int flags)
3496 struct nfs4_stateid *local;
3497 u32 st_id = stid->si_stateownerid;
3498 u32 f_id = stid->si_fileid;
3499 unsigned int hashval;
3501 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3502 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3503 hashval = stateid_hashval(st_id, f_id);
3504 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3505 if ((local->st_stateid.si_stateownerid == st_id) &&
3506 (local->st_stateid.si_fileid == f_id))
3511 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3512 hashval = stateid_hashval(st_id, f_id);
3513 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3514 if ((local->st_stateid.si_stateownerid == st_id) &&
3515 (local->st_stateid.si_fileid == f_id))
3522 static struct nfs4_delegation *
3523 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3525 struct nfs4_file *fp;
3526 struct nfs4_delegation *dl;
3528 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3531 fp = find_file(ino);
3534 dl = find_delegation_file(fp, stid);
3540 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3541 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3542 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3543 * locking, this prevents us from being completely protocol-compliant. The
3544 * real solution to this problem is to start using unsigned file offsets in
3545 * the VFS, but this is a very deep change!
3548 nfs4_transform_lock_offset(struct file_lock *lock)
3550 if (lock->fl_start < 0)
3551 lock->fl_start = OFFSET_MAX;
3552 if (lock->fl_end < 0)
3553 lock->fl_end = OFFSET_MAX;
3556 /* Hack!: For now, we're defining this just so we can use a pointer to it
3557 * as a unique cookie to identify our (NFSv4's) posix locks. */
3558 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3562 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3564 struct nfs4_stateowner *sop;
3566 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3567 sop = (struct nfs4_stateowner *) fl->fl_owner;
3568 kref_get(&sop->so_ref);
3570 deny->ld_clientid = sop->so_client->cl_clientid;
3572 deny->ld_sop = NULL;
3573 deny->ld_clientid.cl_boot = 0;
3574 deny->ld_clientid.cl_id = 0;
3576 deny->ld_start = fl->fl_start;
3577 deny->ld_length = NFS4_MAX_UINT64;
3578 if (fl->fl_end != NFS4_MAX_UINT64)
3579 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3580 deny->ld_type = NFS4_READ_LT;
3581 if (fl->fl_type != F_RDLCK)
3582 deny->ld_type = NFS4_WRITE_LT;
3585 static struct nfs4_stateowner *
3586 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3587 struct xdr_netobj *owner)
3589 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3590 struct nfs4_stateowner *op;
3592 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3593 if (same_owner_str(op, owner, clid))
3600 * Alloc a lock owner structure.
3601 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3604 * strhashval = lock_ownerstr_hashval
3607 static struct nfs4_stateowner *
3608 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3609 struct nfs4_stateowner *sop;
3610 struct nfs4_replay *rp;
3611 unsigned int idhashval;
3613 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3615 idhashval = lockownerid_hashval(current_ownerid);
3616 INIT_LIST_HEAD(&sop->so_idhash);
3617 INIT_LIST_HEAD(&sop->so_strhash);
3618 INIT_LIST_HEAD(&sop->so_perclient);
3619 INIT_LIST_HEAD(&sop->so_stateids);
3620 INIT_LIST_HEAD(&sop->so_perstateid);
3621 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3623 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3624 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3625 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3626 sop->so_is_open_owner = 0;
3627 sop->so_id = current_ownerid++;
3628 sop->so_client = clp;
3629 /* It is the openowner seqid that will be incremented in encode in the
3630 * case of new lockowners; so increment the lock seqid manually: */
3631 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3632 sop->so_confirmed = 1;
3633 rp = &sop->so_replay;
3634 rp->rp_status = nfserr_serverfault;
3636 rp->rp_buf = rp->rp_ibuf;
3640 static struct nfs4_stateid *
3641 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3643 struct nfs4_stateid *stp;
3644 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3646 stp = nfs4_alloc_stateid();
3649 INIT_LIST_HEAD(&stp->st_hash);
3650 INIT_LIST_HEAD(&stp->st_perfile);
3651 INIT_LIST_HEAD(&stp->st_perstateowner);
3652 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3653 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3654 list_add(&stp->st_perfile, &fp->fi_stateids);
3655 list_add(&stp->st_perstateowner, &sop->so_stateids);
3656 stp->st_stateowner = sop;
3659 stp->st_stateid.si_boot = boot_time;
3660 stp->st_stateid.si_stateownerid = sop->so_id;
3661 stp->st_stateid.si_fileid = fp->fi_id;
3662 stp->st_stateid.si_generation = 0;
3663 stp->st_deny_bmap = open_stp->st_deny_bmap;
3664 stp->st_openstp = open_stp;
3671 check_lock_length(u64 offset, u64 length)
3673 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3674 LOFF_OVERFLOW(offset, length)));
3681 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3682 struct nfsd4_lock *lock)
3684 struct nfs4_stateowner *open_sop = NULL;
3685 struct nfs4_stateowner *lock_sop = NULL;
3686 struct nfs4_stateid *lock_stp;
3687 struct nfs4_file *fp;
3688 struct file *filp = NULL;
3689 struct file_lock file_lock;
3690 struct file_lock conflock;
3692 unsigned int strhashval;
3696 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3697 (long long) lock->lk_offset,
3698 (long long) lock->lk_length);
3700 if (check_lock_length(lock->lk_offset, lock->lk_length))
3701 return nfserr_inval;
3703 if ((status = fh_verify(rqstp, &cstate->current_fh,
3704 S_IFREG, NFSD_MAY_LOCK))) {
3705 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3711 if (lock->lk_is_new) {
3713 * Client indicates that this is a new lockowner.
3714 * Use open owner and open stateid to create lock owner and
3717 struct nfs4_stateid *open_stp = NULL;
3719 status = nfserr_stale_clientid;
3720 if (!nfsd4_has_session(cstate) &&
3721 STALE_CLIENTID(&lock->lk_new_clientid))
3724 /* validate and update open stateid and open seqid */
3725 status = nfs4_preprocess_seqid_op(cstate,
3726 lock->lk_new_open_seqid,
3727 &lock->lk_new_open_stateid,
3729 &lock->lk_replay_owner, &open_stp,
3733 open_sop = lock->lk_replay_owner;
3734 /* create lockowner and lock stateid */
3735 fp = open_stp->st_file;
3736 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3737 open_sop->so_client->cl_clientid.cl_id,
3738 &lock->v.new.owner);
3739 /* XXX: Do we need to check for duplicate stateowners on
3740 * the same file, or should they just be allowed (and
3741 * create new stateids)? */
3742 status = nfserr_resource;
3743 lock_sop = alloc_init_lock_stateowner(strhashval,
3744 open_sop->so_client, open_stp, lock);
3745 if (lock_sop == NULL)
3747 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3748 if (lock_stp == NULL)
3751 /* lock (lock owner + lock stateid) already exists */
3752 status = nfs4_preprocess_seqid_op(cstate,
3753 lock->lk_old_lock_seqid,
3754 &lock->lk_old_lock_stateid,
3756 &lock->lk_replay_owner, &lock_stp, lock);
3759 lock_sop = lock->lk_replay_owner;
3760 fp = lock_stp->st_file;
3762 /* lock->lk_replay_owner and lock_stp have been created or found */
3764 status = nfserr_grace;
3765 if (locks_in_grace() && !lock->lk_reclaim)
3767 status = nfserr_no_grace;
3768 if (!locks_in_grace() && lock->lk_reclaim)
3771 locks_init_lock(&file_lock);
3772 switch (lock->lk_type) {
3775 if (find_readable_file(lock_stp->st_file)) {
3776 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3777 filp = find_readable_file(lock_stp->st_file);
3779 file_lock.fl_type = F_RDLCK;
3783 case NFS4_WRITEW_LT:
3784 if (find_writeable_file(lock_stp->st_file)) {
3785 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3786 filp = find_writeable_file(lock_stp->st_file);
3788 file_lock.fl_type = F_WRLCK;
3792 status = nfserr_inval;
3796 status = nfserr_openmode;
3799 file_lock.fl_owner = (fl_owner_t)lock_sop;
3800 file_lock.fl_pid = current->tgid;
3801 file_lock.fl_file = filp;
3802 file_lock.fl_flags = FL_POSIX;
3803 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3805 file_lock.fl_start = lock->lk_offset;
3806 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3807 nfs4_transform_lock_offset(&file_lock);
3810 * Try to lock the file in the VFS.
3811 * Note: locks.c uses the BKL to protect the inode's lock list.
3814 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3816 case 0: /* success! */
3817 update_stateid(&lock_stp->st_stateid);
3818 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3822 case (EAGAIN): /* conflock holds conflicting lock */
3823 status = nfserr_denied;
3824 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3825 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3828 status = nfserr_deadlock;
3831 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3832 status = nfserr_resource;
3836 if (status && lock->lk_is_new && lock_sop)
3837 release_lockowner(lock_sop);
3838 if (lock->lk_replay_owner) {
3839 nfs4_get_stateowner(lock->lk_replay_owner);
3840 cstate->replay_owner = lock->lk_replay_owner;
3842 nfs4_unlock_state();
3847 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3848 * so we do a temporary open here just to get an open file to pass to
3849 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3852 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3857 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3860 err = vfs_test_lock(file, lock);
3869 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3870 struct nfsd4_lockt *lockt)
3872 struct inode *inode;
3873 struct file_lock file_lock;
3877 if (locks_in_grace())
3878 return nfserr_grace;
3880 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3881 return nfserr_inval;
3883 lockt->lt_stateowner = NULL;
3886 status = nfserr_stale_clientid;
3887 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3890 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3891 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3892 if (status == nfserr_symlink)
3893 status = nfserr_inval;
3897 inode = cstate->current_fh.fh_dentry->d_inode;
3898 locks_init_lock(&file_lock);
3899 switch (lockt->lt_type) {
3902 file_lock.fl_type = F_RDLCK;
3905 case NFS4_WRITEW_LT:
3906 file_lock.fl_type = F_WRLCK;
3909 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3910 status = nfserr_inval;
3914 lockt->lt_stateowner = find_lockstateowner_str(inode,
3915 &lockt->lt_clientid, &lockt->lt_owner);
3916 if (lockt->lt_stateowner)
3917 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3918 file_lock.fl_pid = current->tgid;
3919 file_lock.fl_flags = FL_POSIX;
3921 file_lock.fl_start = lockt->lt_offset;
3922 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3924 nfs4_transform_lock_offset(&file_lock);
3927 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3929 status = nfserrno(error);
3932 if (file_lock.fl_type != F_UNLCK) {
3933 status = nfserr_denied;
3934 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3937 nfs4_unlock_state();
3942 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3943 struct nfsd4_locku *locku)
3945 struct nfs4_stateid *stp;
3946 struct file *filp = NULL;
3947 struct file_lock file_lock;
3951 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3952 (long long) locku->lu_offset,
3953 (long long) locku->lu_length);
3955 if (check_lock_length(locku->lu_offset, locku->lu_length))
3956 return nfserr_inval;
3960 if ((status = nfs4_preprocess_seqid_op(cstate,
3964 &locku->lu_stateowner, &stp, NULL)))
3967 filp = find_any_file(stp->st_file);
3969 status = nfserr_lock_range;
3973 locks_init_lock(&file_lock);
3974 file_lock.fl_type = F_UNLCK;
3975 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3976 file_lock.fl_pid = current->tgid;
3977 file_lock.fl_file = filp;
3978 file_lock.fl_flags = FL_POSIX;
3979 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3980 file_lock.fl_start = locku->lu_offset;
3982 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3983 nfs4_transform_lock_offset(&file_lock);
3986 * Try to unlock the file in the VFS.
3988 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3990 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3994 * OK, unlock succeeded; the only thing left to do is update the stateid.
3996 update_stateid(&stp->st_stateid);
3997 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4000 if (locku->lu_stateowner) {
4001 nfs4_get_stateowner(locku->lu_stateowner);
4002 cstate->replay_owner = locku->lu_stateowner;
4004 nfs4_unlock_state();
4008 status = nfserrno(err);
4014 * 1: locks held by lockowner
4015 * 0: no locks held by lockowner
4018 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4020 struct file_lock **flpp;
4021 struct inode *inode = filp->fi_inode;
4025 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4026 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4037 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4038 struct nfsd4_compound_state *cstate,
4039 struct nfsd4_release_lockowner *rlockowner)
4041 clientid_t *clid = &rlockowner->rl_clientid;
4042 struct nfs4_stateowner *sop;
4043 struct nfs4_stateid *stp;
4044 struct xdr_netobj *owner = &rlockowner->rl_owner;
4045 struct list_head matches;
4049 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4050 clid->cl_boot, clid->cl_id);
4052 /* XXX check for lease expiration */
4054 status = nfserr_stale_clientid;
4055 if (STALE_CLIENTID(clid))
4060 status = nfserr_locks_held;
4061 /* XXX: we're doing a linear search through all the lockowners.
4062 * Yipes! For now we'll just hope clients aren't really using
4063 * release_lockowner much, but eventually we have to fix these
4064 * data structures. */
4065 INIT_LIST_HEAD(&matches);
4066 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4067 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4068 if (!same_owner_str(sop, owner, clid))
4070 list_for_each_entry(stp, &sop->so_stateids,
4072 if (check_for_locks(stp->st_file, sop))
4074 /* Note: so_perclient unused for lockowners,
4075 * so it's OK to fool with here. */
4076 list_add(&sop->so_perclient, &matches);
4080 /* Clients probably won't expect us to return with some (but not all)
4081 * of the lockowner state released; so don't release any until all
4082 * have been checked. */
4084 while (!list_empty(&matches)) {
4085 sop = list_entry(matches.next, struct nfs4_stateowner,
4087 /* unhash_stateowner deletes so_perclient only
4088 * for openowners. */
4089 list_del(&sop->so_perclient);
4090 release_lockowner(sop);
4093 nfs4_unlock_state();
4097 static inline struct nfs4_client_reclaim *
4100 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4104 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4106 unsigned int strhashval = clientstr_hashval(name);
4107 struct nfs4_client *clp;
4109 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
4114 * failure => all reset bets are off, nfserr_no_grace...
4117 nfs4_client_to_reclaim(const char *name)
4119 unsigned int strhashval;
4120 struct nfs4_client_reclaim *crp = NULL;
4122 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4123 crp = alloc_reclaim();
4126 strhashval = clientstr_hashval(name);
4127 INIT_LIST_HEAD(&crp->cr_strhash);
4128 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4129 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4130 reclaim_str_hashtbl_size++;
4135 nfs4_release_reclaim(void)
4137 struct nfs4_client_reclaim *crp = NULL;
4140 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4141 while (!list_empty(&reclaim_str_hashtbl[i])) {
4142 crp = list_entry(reclaim_str_hashtbl[i].next,
4143 struct nfs4_client_reclaim, cr_strhash);
4144 list_del(&crp->cr_strhash);
4146 reclaim_str_hashtbl_size--;
4149 BUG_ON(reclaim_str_hashtbl_size);
4153 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4154 static struct nfs4_client_reclaim *
4155 nfs4_find_reclaim_client(clientid_t *clid)
4157 unsigned int strhashval;
4158 struct nfs4_client *clp;
4159 struct nfs4_client_reclaim *crp = NULL;
4162 /* find clientid in conf_id_hashtbl */
4163 clp = find_confirmed_client(clid);
4167 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4168 clp->cl_name.len, clp->cl_name.data,
4171 /* find clp->cl_name in reclaim_str_hashtbl */
4172 strhashval = clientstr_hashval(clp->cl_recdir);
4173 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4174 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4182 * Called from OPEN. Look for clientid in reclaim list.
4185 nfs4_check_open_reclaim(clientid_t *clid)
4187 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4190 /* initialization to perform at module load time: */
4193 nfs4_state_init(void)
4197 status = nfsd4_init_slabs();
4200 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4201 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4202 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4203 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4204 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4205 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4207 for (i = 0; i < SESSION_HASH_SIZE; i++)
4208 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4209 for (i = 0; i < FILE_HASH_SIZE; i++) {
4210 INIT_LIST_HEAD(&file_hashtbl[i]);
4212 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4213 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4214 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4216 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4217 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4218 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4220 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4221 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4222 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4224 memset(&onestateid, ~0, sizeof(stateid_t));
4225 INIT_LIST_HEAD(&close_lru);
4226 INIT_LIST_HEAD(&client_lru);
4227 INIT_LIST_HEAD(&del_recall_lru);
4228 reclaim_str_hashtbl_size = 0;
4233 nfsd4_load_reboot_recovery_data(void)
4238 nfsd4_init_recdir(user_recovery_dirname);
4239 status = nfsd4_recdir_load();
4240 nfs4_unlock_state();
4242 printk("NFSD: Failure reading reboot recovery data\n");
4246 * Since the lifetime of a delegation isn't limited to that of an open, a
4247 * client may quite reasonably hang on to a delegation as long as it has
4248 * the inode cached. This becomes an obvious problem the first time a
4249 * client's inode cache approaches the size of the server's total memory.
4251 * For now we avoid this problem by imposing a hard limit on the number
4252 * of delegations, which varies according to the server's memory size.
4255 set_max_delegations(void)
4258 * Allow at most 4 delegations per megabyte of RAM. Quick
4259 * estimates suggest that in the worst case (where every delegation
4260 * is for a different inode), a delegation could take about 1.5K,
4261 * giving a worst case usage of about 6% of memory.
4263 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4266 /* initialization to perform when the nfsd service is started: */
4269 __nfs4_state_start(void)
4273 boot_time = get_seconds();
4274 locks_start_grace(&nfsd4_manager);
4275 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4277 ret = set_callback_cred();
4280 laundry_wq = create_singlethread_workqueue("nfsd4");
4281 if (laundry_wq == NULL)
4283 ret = nfsd4_create_callback_queue();
4285 goto out_free_laundry;
4286 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4287 set_max_delegations();
4290 destroy_workqueue(laundry_wq);
4295 nfs4_state_start(void)
4297 nfsd4_load_reboot_recovery_data();
4298 return __nfs4_state_start();
4302 __nfs4_state_shutdown(void)
4305 struct nfs4_client *clp = NULL;
4306 struct nfs4_delegation *dp = NULL;
4307 struct list_head *pos, *next, reaplist;
4309 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4310 while (!list_empty(&conf_id_hashtbl[i])) {
4311 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4314 while (!list_empty(&unconf_str_hashtbl[i])) {
4315 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4319 INIT_LIST_HEAD(&reaplist);
4320 spin_lock(&recall_lock);
4321 list_for_each_safe(pos, next, &del_recall_lru) {
4322 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4323 list_move(&dp->dl_recall_lru, &reaplist);
4325 spin_unlock(&recall_lock);
4326 list_for_each_safe(pos, next, &reaplist) {
4327 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4328 list_del_init(&dp->dl_recall_lru);
4329 unhash_delegation(dp);
4332 nfsd4_shutdown_recdir();
4336 nfs4_state_shutdown(void)
4338 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4339 destroy_workqueue(laundry_wq);
4340 locks_end_grace(&nfsd4_manager);
4342 nfs4_release_reclaim();
4343 __nfs4_state_shutdown();
4344 nfs4_unlock_state();
4345 nfsd4_destroy_callback_queue();
4349 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4350 * accessed when nfsd is starting.
4353 nfs4_set_recdir(char *recdir)
4355 strcpy(user_recovery_dirname, recdir);
4359 * Change the NFSv4 recovery directory to recdir.
4362 nfs4_reset_recoverydir(char *recdir)
4367 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4371 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4372 nfs4_set_recdir(recdir);
4380 nfs4_recoverydir(void)
4382 return user_recovery_dirname;