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 kref_init(&new->se_ref);
764 idx = hash_sessionid(&new->se_sessionid);
765 spin_lock(&client_lock);
766 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
767 list_add(&new->se_perclnt, &clp->cl_sessions);
768 spin_unlock(&client_lock);
770 status = nfsd4_new_conn(rqstp, new);
771 /* whoops: benny points out, status is ignored! (err, or bogus) */
773 free_session(&new->se_ref);
776 if (!clp->cl_cb_session && (cses->flags & SESSION4_BACK_CHAN)) {
777 struct sockaddr *sa = svc_addr(rqstp);
779 clp->cl_cb_session = new;
780 clp->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
781 svc_xprt_get(rqstp->rq_xprt);
782 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
783 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
784 clp->cl_cb_conn.cb_minorversion = 1;
785 clp->cl_cb_conn.cb_prog = cses->callback_prog;
786 nfsd4_probe_callback(clp, &clp->cl_cb_conn);
791 /* caller must hold client_lock */
792 static struct nfsd4_session *
793 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
795 struct nfsd4_session *elem;
798 dump_sessionid(__func__, sessionid);
799 idx = hash_sessionid(sessionid);
800 /* Search in the appropriate list */
801 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
802 if (!memcmp(elem->se_sessionid.data, sessionid->data,
803 NFS4_MAX_SESSIONID_LEN)) {
808 dprintk("%s: session not found\n", __func__);
812 /* caller must hold client_lock */
814 unhash_session(struct nfsd4_session *ses)
816 list_del(&ses->se_hash);
817 list_del(&ses->se_perclnt);
820 /* must be called under the client_lock */
822 renew_client_locked(struct nfs4_client *clp)
824 if (is_client_expired(clp)) {
825 dprintk("%s: client (clientid %08x/%08x) already expired\n",
827 clp->cl_clientid.cl_boot,
828 clp->cl_clientid.cl_id);
833 * Move client to the end to the LRU list.
835 dprintk("renewing client (clientid %08x/%08x)\n",
836 clp->cl_clientid.cl_boot,
837 clp->cl_clientid.cl_id);
838 list_move_tail(&clp->cl_lru, &client_lru);
839 clp->cl_time = get_seconds();
843 renew_client(struct nfs4_client *clp)
845 spin_lock(&client_lock);
846 renew_client_locked(clp);
847 spin_unlock(&client_lock);
850 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
852 STALE_CLIENTID(clientid_t *clid)
854 if (clid->cl_boot == boot_time)
856 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
857 clid->cl_boot, clid->cl_id, boot_time);
862 * XXX Should we use a slab cache ?
863 * This type of memory management is somewhat inefficient, but we use it
864 * anyway since SETCLIENTID is not a common operation.
866 static struct nfs4_client *alloc_client(struct xdr_netobj name)
868 struct nfs4_client *clp;
870 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
873 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
874 if (clp->cl_name.data == NULL) {
878 memcpy(clp->cl_name.data, name.data, name.len);
879 clp->cl_name.len = name.len;
884 free_client(struct nfs4_client *clp)
886 if (clp->cl_cred.cr_group_info)
887 put_group_info(clp->cl_cred.cr_group_info);
888 kfree(clp->cl_principal);
889 kfree(clp->cl_name.data);
894 release_session_client(struct nfsd4_session *session)
896 struct nfs4_client *clp = session->se_client;
898 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
900 if (is_client_expired(clp)) {
902 session->se_client = NULL;
904 renew_client_locked(clp);
905 spin_unlock(&client_lock);
908 /* must be called under the client_lock */
910 unhash_client_locked(struct nfs4_client *clp)
912 mark_client_expired(clp);
913 list_del(&clp->cl_lru);
914 while (!list_empty(&clp->cl_sessions)) {
915 struct nfsd4_session *ses;
916 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
919 nfsd4_put_session(ses);
924 expire_client(struct nfs4_client *clp)
926 struct nfs4_stateowner *sop;
927 struct nfs4_delegation *dp;
928 struct list_head reaplist;
930 INIT_LIST_HEAD(&reaplist);
931 spin_lock(&recall_lock);
932 while (!list_empty(&clp->cl_delegations)) {
933 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
934 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
936 list_del_init(&dp->dl_perclnt);
937 list_move(&dp->dl_recall_lru, &reaplist);
939 spin_unlock(&recall_lock);
940 while (!list_empty(&reaplist)) {
941 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
942 list_del_init(&dp->dl_recall_lru);
943 unhash_delegation(dp);
945 while (!list_empty(&clp->cl_openowners)) {
946 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
947 release_openowner(sop);
949 nfsd4_shutdown_callback(clp);
950 if (clp->cl_cb_conn.cb_xprt)
951 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
952 list_del(&clp->cl_idhash);
953 list_del(&clp->cl_strhash);
954 spin_lock(&client_lock);
955 unhash_client_locked(clp);
956 if (atomic_read(&clp->cl_refcount) == 0)
958 spin_unlock(&client_lock);
961 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
963 memcpy(target->cl_verifier.data, source->data,
964 sizeof(target->cl_verifier.data));
967 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
969 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
970 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
973 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
975 target->cr_uid = source->cr_uid;
976 target->cr_gid = source->cr_gid;
977 target->cr_group_info = source->cr_group_info;
978 get_group_info(target->cr_group_info);
981 static int same_name(const char *n1, const char *n2)
983 return 0 == memcmp(n1, n2, HEXDIR_LEN);
987 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
989 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
993 same_clid(clientid_t *cl1, clientid_t *cl2)
995 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
998 /* XXX what about NGROUP */
1000 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1002 return cr1->cr_uid == cr2->cr_uid;
1005 static void gen_clid(struct nfs4_client *clp)
1007 static u32 current_clientid = 1;
1009 clp->cl_clientid.cl_boot = boot_time;
1010 clp->cl_clientid.cl_id = current_clientid++;
1013 static void gen_confirm(struct nfs4_client *clp)
1018 p = (u32 *)clp->cl_confirm.data;
1019 *p++ = get_seconds();
1023 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1024 struct svc_rqst *rqstp, nfs4_verifier *verf)
1026 struct nfs4_client *clp;
1027 struct sockaddr *sa = svc_addr(rqstp);
1030 clp = alloc_client(name);
1034 princ = svc_gss_principal(rqstp);
1036 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1037 if (clp->cl_principal == NULL) {
1043 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1044 atomic_set(&clp->cl_refcount, 0);
1045 atomic_set(&clp->cl_cb_set, 0);
1046 INIT_LIST_HEAD(&clp->cl_idhash);
1047 INIT_LIST_HEAD(&clp->cl_strhash);
1048 INIT_LIST_HEAD(&clp->cl_openowners);
1049 INIT_LIST_HEAD(&clp->cl_delegations);
1050 INIT_LIST_HEAD(&clp->cl_sessions);
1051 INIT_LIST_HEAD(&clp->cl_lru);
1052 spin_lock_init(&clp->cl_lock);
1053 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1054 clp->cl_time = get_seconds();
1055 clear_bit(0, &clp->cl_cb_slot_busy);
1056 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1057 copy_verf(clp, verf);
1058 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1059 clp->cl_flavor = rqstp->rq_flavor;
1060 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1062 clp->cl_cb_session = NULL;
1066 static int check_name(struct xdr_netobj name)
1070 if (name.len > NFS4_OPAQUE_LIMIT) {
1071 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1078 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1080 unsigned int idhashval;
1082 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1083 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1084 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1089 move_to_confirmed(struct nfs4_client *clp)
1091 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1092 unsigned int strhashval;
1094 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1095 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1096 strhashval = clientstr_hashval(clp->cl_recdir);
1097 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1101 static struct nfs4_client *
1102 find_confirmed_client(clientid_t *clid)
1104 struct nfs4_client *clp;
1105 unsigned int idhashval = clientid_hashval(clid->cl_id);
1107 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1108 if (same_clid(&clp->cl_clientid, clid))
1114 static struct nfs4_client *
1115 find_unconfirmed_client(clientid_t *clid)
1117 struct nfs4_client *clp;
1118 unsigned int idhashval = clientid_hashval(clid->cl_id);
1120 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1121 if (same_clid(&clp->cl_clientid, clid))
1128 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
1129 * parameter. Matching is based on the fact the at least one of the
1130 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
1132 * FIXME: we need to unify the clientid namespaces for nfsv4.x
1133 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
1134 * and SET_CLIENTID{,_CONFIRM}
1137 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
1139 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
1140 return use_exchange_id == has_exchange_flags;
1143 static struct nfs4_client *
1144 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
1145 bool use_exchange_id)
1147 struct nfs4_client *clp;
1149 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1150 if (same_name(clp->cl_recdir, dname) &&
1151 match_clientid_establishment(clp, use_exchange_id))
1157 static struct nfs4_client *
1158 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
1159 bool use_exchange_id)
1161 struct nfs4_client *clp;
1163 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1164 if (same_name(clp->cl_recdir, dname) &&
1165 match_clientid_establishment(clp, use_exchange_id))
1172 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1174 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1175 unsigned short expected_family;
1177 /* Currently, we only support tcp and tcp6 for the callback channel */
1178 if (se->se_callback_netid_len == 3 &&
1179 !memcmp(se->se_callback_netid_val, "tcp", 3))
1180 expected_family = AF_INET;
1181 else if (se->se_callback_netid_len == 4 &&
1182 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1183 expected_family = AF_INET6;
1187 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1188 se->se_callback_addr_len,
1189 (struct sockaddr *)&conn->cb_addr,
1190 sizeof(conn->cb_addr));
1192 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1195 if (conn->cb_addr.ss_family == AF_INET6)
1196 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1198 conn->cb_minorversion = 0;
1199 conn->cb_prog = se->se_callback_prog;
1200 conn->cb_ident = se->se_callback_ident;
1203 conn->cb_addr.ss_family = AF_UNSPEC;
1204 conn->cb_addrlen = 0;
1205 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1206 "will not receive delegations\n",
1207 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1213 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1216 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1218 struct nfsd4_slot *slot = resp->cstate.slot;
1221 dprintk("--> %s slot %p\n", __func__, slot);
1223 slot->sl_opcnt = resp->opcnt;
1224 slot->sl_status = resp->cstate.status;
1226 if (nfsd4_not_cached(resp)) {
1227 slot->sl_datalen = 0;
1230 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1231 base = (char *)resp->cstate.datap -
1232 (char *)resp->xbuf->head[0].iov_base;
1233 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1235 WARN("%s: sessions DRC could not cache compound\n", __func__);
1240 * Encode the replay sequence operation from the slot values.
1241 * If cachethis is FALSE encode the uncached rep error on the next
1242 * operation which sets resp->p and increments resp->opcnt for
1243 * nfs4svc_encode_compoundres.
1247 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1248 struct nfsd4_compoundres *resp)
1250 struct nfsd4_op *op;
1251 struct nfsd4_slot *slot = resp->cstate.slot;
1253 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1254 resp->opcnt, resp->cstate.slot->sl_cachethis);
1256 /* Encode the replayed sequence operation */
1257 op = &args->ops[resp->opcnt - 1];
1258 nfsd4_encode_operation(resp, op);
1260 /* Return nfserr_retry_uncached_rep in next operation. */
1261 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1262 op = &args->ops[resp->opcnt++];
1263 op->status = nfserr_retry_uncached_rep;
1264 nfsd4_encode_operation(resp, op);
1270 * The sequence operation is not cached because we can use the slot and
1274 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1275 struct nfsd4_sequence *seq)
1277 struct nfsd4_slot *slot = resp->cstate.slot;
1280 dprintk("--> %s slot %p\n", __func__, slot);
1282 /* Either returns 0 or nfserr_retry_uncached */
1283 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1284 if (status == nfserr_retry_uncached_rep)
1287 /* The sequence operation has been encoded, cstate->datap set. */
1288 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1290 resp->opcnt = slot->sl_opcnt;
1291 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1292 status = slot->sl_status;
1298 * Set the exchange_id flags returned by the server.
1301 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1303 /* pNFS is not supported */
1304 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1306 /* Referrals are supported, Migration is not. */
1307 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1309 /* set the wire flags to return to client. */
1310 clid->flags = new->cl_exchange_flags;
1314 nfsd4_exchange_id(struct svc_rqst *rqstp,
1315 struct nfsd4_compound_state *cstate,
1316 struct nfsd4_exchange_id *exid)
1318 struct nfs4_client *unconf, *conf, *new;
1320 unsigned int strhashval;
1321 char dname[HEXDIR_LEN];
1322 char addr_str[INET6_ADDRSTRLEN];
1323 nfs4_verifier verf = exid->verifier;
1324 struct sockaddr *sa = svc_addr(rqstp);
1326 rpc_ntop(sa, addr_str, sizeof(addr_str));
1327 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1328 "ip_addr=%s flags %x, spa_how %d\n",
1329 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1330 addr_str, exid->flags, exid->spa_how);
1332 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1333 return nfserr_inval;
1335 /* Currently only support SP4_NONE */
1336 switch (exid->spa_how) {
1340 return nfserr_encr_alg_unsupp;
1342 BUG(); /* checked by xdr code */
1344 return nfserr_serverfault; /* no excuse :-/ */
1347 status = nfs4_make_rec_clidname(dname, &exid->clname);
1352 strhashval = clientstr_hashval(dname);
1357 conf = find_confirmed_client_by_str(dname, strhashval, true);
1359 if (!same_verf(&verf, &conf->cl_verifier)) {
1360 /* 18.35.4 case 8 */
1361 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1362 status = nfserr_not_same;
1365 /* Client reboot: destroy old state */
1366 expire_client(conf);
1369 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1370 /* 18.35.4 case 9 */
1371 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1372 status = nfserr_perm;
1375 expire_client(conf);
1379 * Set bit when the owner id and verifier map to an already
1380 * confirmed client id (18.35.3).
1382 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1385 * Falling into 18.35.4 case 2, possible router replay.
1386 * Leave confirmed record intact and return same result.
1388 copy_verf(conf, &verf);
1393 /* 18.35.4 case 7 */
1394 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1395 status = nfserr_noent;
1399 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1402 * Possible retry or client restart. Per 18.35.4 case 4,
1403 * a new unconfirmed record should be generated regardless
1404 * of whether any properties have changed.
1406 expire_client(unconf);
1411 new = create_client(exid->clname, dname, rqstp, &verf);
1413 status = nfserr_jukebox;
1418 add_to_unconfirmed(new, strhashval);
1420 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1421 exid->clientid.cl_id = new->cl_clientid.cl_id;
1424 nfsd4_set_ex_flags(new, exid);
1426 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1427 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1431 nfs4_unlock_state();
1433 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1438 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1440 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1443 /* The slot is in use, and no response has been sent. */
1445 if (seqid == slot_seqid)
1446 return nfserr_jukebox;
1448 return nfserr_seq_misordered;
1451 if (likely(seqid == slot_seqid + 1))
1454 if (seqid == slot_seqid)
1455 return nfserr_replay_cache;
1457 if (seqid == 1 && (slot_seqid + 1) == 0)
1459 /* Misordered replay or misordered new request */
1460 return nfserr_seq_misordered;
1464 * Cache the create session result into the create session single DRC
1465 * slot cache by saving the xdr structure. sl_seqid has been set.
1466 * Do this for solo or embedded create session operations.
1469 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1470 struct nfsd4_clid_slot *slot, int nfserr)
1472 slot->sl_status = nfserr;
1473 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1477 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1478 struct nfsd4_clid_slot *slot)
1480 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1481 return slot->sl_status;
1485 nfsd4_create_session(struct svc_rqst *rqstp,
1486 struct nfsd4_compound_state *cstate,
1487 struct nfsd4_create_session *cr_ses)
1489 struct sockaddr *sa = svc_addr(rqstp);
1490 struct nfs4_client *conf, *unconf;
1491 struct nfsd4_session *new;
1492 struct nfsd4_clid_slot *cs_slot = NULL;
1493 bool confirm_me = false;
1497 unconf = find_unconfirmed_client(&cr_ses->clientid);
1498 conf = find_confirmed_client(&cr_ses->clientid);
1501 cs_slot = &conf->cl_cs_slot;
1502 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1503 if (status == nfserr_replay_cache) {
1504 dprintk("Got a create_session replay! seqid= %d\n",
1506 /* Return the cached reply status */
1507 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1509 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1510 status = nfserr_seq_misordered;
1511 dprintk("Sequence misordered!\n");
1512 dprintk("Expected seqid= %d but got seqid= %d\n",
1513 cs_slot->sl_seqid, cr_ses->seqid);
1516 } else if (unconf) {
1517 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1518 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1519 status = nfserr_clid_inuse;
1523 cs_slot = &unconf->cl_cs_slot;
1524 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1526 /* an unconfirmed replay returns misordered */
1527 status = nfserr_seq_misordered;
1534 status = nfserr_stale_clientid;
1539 * We do not support RDMA or persistent sessions
1541 cr_ses->flags &= ~SESSION4_PERSIST;
1542 cr_ses->flags &= ~SESSION4_RDMA;
1544 status = nfserr_jukebox;
1545 new = alloc_init_session(rqstp, conf, cr_ses);
1549 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1550 NFS4_MAX_SESSIONID_LEN);
1551 cs_slot->sl_seqid++;
1552 cr_ses->seqid = cs_slot->sl_seqid;
1554 /* cache solo and embedded create sessions under the state lock */
1555 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1557 move_to_confirmed(conf);
1559 nfs4_unlock_state();
1560 dprintk("%s returns %d\n", __func__, ntohl(status));
1564 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1566 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1567 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1569 return argp->opcnt == resp->opcnt;
1572 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1576 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1580 nfsd4_destroy_session(struct svc_rqst *r,
1581 struct nfsd4_compound_state *cstate,
1582 struct nfsd4_destroy_session *sessionid)
1584 struct nfsd4_session *ses;
1585 u32 status = nfserr_badsession;
1588 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1589 * - Should we return nfserr_back_chan_busy if waiting for
1590 * callbacks on to-be-destroyed session?
1591 * - Do we need to clear any callback info from previous session?
1594 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1595 if (!nfsd4_last_compound_op(r))
1596 return nfserr_not_only_op;
1598 dump_sessionid(__func__, &sessionid->sessionid);
1599 spin_lock(&client_lock);
1600 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1602 spin_unlock(&client_lock);
1606 unhash_session(ses);
1607 spin_unlock(&client_lock);
1610 /* wait for callbacks */
1611 nfsd4_shutdown_callback(ses->se_client);
1612 nfs4_unlock_state();
1614 nfsd4_del_conns(ses);
1616 nfsd4_put_session(ses);
1619 dprintk("%s returns %d\n", __func__, ntohl(status));
1623 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_rqst *r, struct nfsd4_session *s)
1625 struct nfsd4_conn *c;
1627 list_for_each_entry(c, &s->se_conns, cn_persession) {
1628 if (c->cn_xprt == r->rq_xprt) {
1635 static void nfsd4_sequence_check_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses)
1637 struct nfs4_client *clp = ses->se_client;
1638 struct nfsd4_conn *c, *new = NULL;
1640 spin_lock(&clp->cl_lock);
1641 c = __nfsd4_find_conn(rqstp, ses);
1642 spin_unlock(&clp->cl_lock);
1646 new = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1648 spin_lock(&clp->cl_lock);
1649 c = __nfsd4_find_conn(rqstp, ses);
1651 spin_unlock(&clp->cl_lock);
1655 __nfsd4_hash_conn(new, ses);
1656 spin_unlock(&clp->cl_lock);
1657 nfsd4_register_conn(new);
1662 nfsd4_sequence(struct svc_rqst *rqstp,
1663 struct nfsd4_compound_state *cstate,
1664 struct nfsd4_sequence *seq)
1666 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1667 struct nfsd4_session *session;
1668 struct nfsd4_slot *slot;
1671 if (resp->opcnt != 1)
1672 return nfserr_sequence_pos;
1674 spin_lock(&client_lock);
1675 status = nfserr_badsession;
1676 session = find_in_sessionid_hashtbl(&seq->sessionid);
1680 status = nfserr_badslot;
1681 if (seq->slotid >= session->se_fchannel.maxreqs)
1684 slot = session->se_slots[seq->slotid];
1685 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1687 /* We do not negotiate the number of slots yet, so set the
1688 * maxslots to the session maxreqs which is used to encode
1689 * sr_highest_slotid and the sr_target_slot id to maxslots */
1690 seq->maxslots = session->se_fchannel.maxreqs;
1692 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1693 if (status == nfserr_replay_cache) {
1694 cstate->slot = slot;
1695 cstate->session = session;
1696 /* Return the cached reply status and set cstate->status
1697 * for nfsd4_proc_compound processing */
1698 status = nfsd4_replay_cache_entry(resp, seq);
1699 cstate->status = nfserr_replay_cache;
1705 nfsd4_sequence_check_conn(rqstp, session);
1707 /* Success! bump slot seqid */
1708 slot->sl_inuse = true;
1709 slot->sl_seqid = seq->seqid;
1710 slot->sl_cachethis = seq->cachethis;
1712 cstate->slot = slot;
1713 cstate->session = session;
1716 /* Hold a session reference until done processing the compound. */
1717 if (cstate->session) {
1718 nfsd4_get_session(cstate->session);
1719 atomic_inc(&session->se_client->cl_refcount);
1721 spin_unlock(&client_lock);
1722 dprintk("%s: return %d\n", __func__, ntohl(status));
1727 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1729 if (rc->rca_one_fs) {
1730 if (!cstate->current_fh.fh_dentry)
1731 return nfserr_nofilehandle;
1733 * We don't take advantage of the rca_one_fs case.
1734 * That's OK, it's optional, we can safely ignore it.
1739 if (is_client_expired(cstate->session->se_client)) {
1740 nfs4_unlock_state();
1742 * The following error isn't really legal.
1743 * But we only get here if the client just explicitly
1744 * destroyed the client. Surely it no longer cares what
1745 * error it gets back on an operation for the dead
1748 return nfserr_stale_clientid;
1750 nfsd4_create_clid_dir(cstate->session->se_client);
1751 nfs4_unlock_state();
1756 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1757 struct nfsd4_setclientid *setclid)
1759 struct sockaddr *sa = svc_addr(rqstp);
1760 struct xdr_netobj clname = {
1761 .len = setclid->se_namelen,
1762 .data = setclid->se_name,
1764 nfs4_verifier clverifier = setclid->se_verf;
1765 unsigned int strhashval;
1766 struct nfs4_client *conf, *unconf, *new;
1768 char dname[HEXDIR_LEN];
1770 if (!check_name(clname))
1771 return nfserr_inval;
1773 status = nfs4_make_rec_clidname(dname, &clname);
1778 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1779 * We get here on a DRC miss.
1782 strhashval = clientstr_hashval(dname);
1785 conf = find_confirmed_client_by_str(dname, strhashval, false);
1787 /* RFC 3530 14.2.33 CASE 0: */
1788 status = nfserr_clid_inuse;
1789 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1790 char addr_str[INET6_ADDRSTRLEN];
1791 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1793 dprintk("NFSD: setclientid: string in use by client "
1794 "at %s\n", addr_str);
1799 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1800 * has a description of SETCLIENTID request processing consisting
1801 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1803 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1804 status = nfserr_resource;
1807 * RFC 3530 14.2.33 CASE 4:
1808 * placed first, because it is the normal case
1811 expire_client(unconf);
1812 new = create_client(clname, dname, rqstp, &clverifier);
1816 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1818 * RFC 3530 14.2.33 CASE 1:
1819 * probable callback update
1822 /* Note this is removing unconfirmed {*x***},
1823 * which is stronger than RFC recommended {vxc**}.
1824 * This has the advantage that there is at most
1825 * one {*x***} in either list at any time.
1827 expire_client(unconf);
1829 new = create_client(clname, dname, rqstp, &clverifier);
1832 copy_clid(new, conf);
1833 } else if (!unconf) {
1835 * RFC 3530 14.2.33 CASE 2:
1836 * probable client reboot; state will be removed if
1839 new = create_client(clname, dname, rqstp, &clverifier);
1845 * RFC 3530 14.2.33 CASE 3:
1846 * probable client reboot; state will be removed if
1849 expire_client(unconf);
1850 new = create_client(clname, dname, rqstp, &clverifier);
1855 gen_callback(new, setclid, rpc_get_scope_id(sa));
1856 add_to_unconfirmed(new, strhashval);
1857 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1858 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1859 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1862 nfs4_unlock_state();
1868 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1869 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1870 * bullets, labeled as CASE1 - CASE4 below.
1873 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1874 struct nfsd4_compound_state *cstate,
1875 struct nfsd4_setclientid_confirm *setclientid_confirm)
1877 struct sockaddr *sa = svc_addr(rqstp);
1878 struct nfs4_client *conf, *unconf;
1879 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1880 clientid_t * clid = &setclientid_confirm->sc_clientid;
1883 if (STALE_CLIENTID(clid))
1884 return nfserr_stale_clientid;
1886 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1887 * We get here on a DRC miss.
1892 conf = find_confirmed_client(clid);
1893 unconf = find_unconfirmed_client(clid);
1895 status = nfserr_clid_inuse;
1896 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1898 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1902 * section 14.2.34 of RFC 3530 has a description of
1903 * SETCLIENTID_CONFIRM request processing consisting
1904 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1906 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1908 * RFC 3530 14.2.34 CASE 1:
1911 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1912 status = nfserr_clid_inuse;
1914 atomic_set(&conf->cl_cb_set, 0);
1915 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1916 expire_client(unconf);
1920 } else if (conf && !unconf) {
1922 * RFC 3530 14.2.34 CASE 2:
1923 * probable retransmitted request; play it safe and
1926 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1927 status = nfserr_clid_inuse;
1930 } else if (!conf && unconf
1931 && same_verf(&unconf->cl_confirm, &confirm)) {
1933 * RFC 3530 14.2.34 CASE 3:
1934 * Normal case; new or rebooted client:
1936 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1937 status = nfserr_clid_inuse;
1940 clientstr_hashval(unconf->cl_recdir);
1941 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1944 nfsd4_remove_clid_dir(conf);
1945 expire_client(conf);
1947 move_to_confirmed(unconf);
1949 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1952 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1953 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1956 * RFC 3530 14.2.34 CASE 4:
1957 * Client probably hasn't noticed that we rebooted yet.
1959 status = nfserr_stale_clientid;
1961 /* check that we have hit one of the cases...*/
1962 status = nfserr_clid_inuse;
1965 nfs4_unlock_state();
1969 /* OPEN Share state helper functions */
1970 static inline struct nfs4_file *
1971 alloc_init_file(struct inode *ino)
1973 struct nfs4_file *fp;
1974 unsigned int hashval = file_hashval(ino);
1976 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1978 atomic_set(&fp->fi_ref, 1);
1979 INIT_LIST_HEAD(&fp->fi_hash);
1980 INIT_LIST_HEAD(&fp->fi_stateids);
1981 INIT_LIST_HEAD(&fp->fi_delegations);
1982 fp->fi_inode = igrab(ino);
1983 fp->fi_id = current_fileid++;
1984 fp->fi_had_conflict = false;
1985 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
1986 memset(fp->fi_access, 0, sizeof(fp->fi_access));
1987 spin_lock(&recall_lock);
1988 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1989 spin_unlock(&recall_lock);
1996 nfsd4_free_slab(struct kmem_cache **slab)
2000 kmem_cache_destroy(*slab);
2005 nfsd4_free_slabs(void)
2007 nfsd4_free_slab(&stateowner_slab);
2008 nfsd4_free_slab(&file_slab);
2009 nfsd4_free_slab(&stateid_slab);
2010 nfsd4_free_slab(&deleg_slab);
2014 nfsd4_init_slabs(void)
2016 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2017 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2018 if (stateowner_slab == NULL)
2020 file_slab = kmem_cache_create("nfsd4_files",
2021 sizeof(struct nfs4_file), 0, 0, NULL);
2022 if (file_slab == NULL)
2024 stateid_slab = kmem_cache_create("nfsd4_stateids",
2025 sizeof(struct nfs4_stateid), 0, 0, NULL);
2026 if (stateid_slab == NULL)
2028 deleg_slab = kmem_cache_create("nfsd4_delegations",
2029 sizeof(struct nfs4_delegation), 0, 0, NULL);
2030 if (deleg_slab == NULL)
2035 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2040 nfs4_free_stateowner(struct kref *kref)
2042 struct nfs4_stateowner *sop =
2043 container_of(kref, struct nfs4_stateowner, so_ref);
2044 kfree(sop->so_owner.data);
2045 kmem_cache_free(stateowner_slab, sop);
2048 static inline struct nfs4_stateowner *
2049 alloc_stateowner(struct xdr_netobj *owner)
2051 struct nfs4_stateowner *sop;
2053 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2054 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2055 memcpy(sop->so_owner.data, owner->data, owner->len);
2056 sop->so_owner.len = owner->len;
2057 kref_init(&sop->so_ref);
2060 kmem_cache_free(stateowner_slab, sop);
2065 static struct nfs4_stateowner *
2066 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2067 struct nfs4_stateowner *sop;
2068 struct nfs4_replay *rp;
2069 unsigned int idhashval;
2071 if (!(sop = alloc_stateowner(&open->op_owner)))
2073 idhashval = ownerid_hashval(current_ownerid);
2074 INIT_LIST_HEAD(&sop->so_idhash);
2075 INIT_LIST_HEAD(&sop->so_strhash);
2076 INIT_LIST_HEAD(&sop->so_perclient);
2077 INIT_LIST_HEAD(&sop->so_stateids);
2078 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2079 INIT_LIST_HEAD(&sop->so_close_lru);
2081 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2082 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2083 list_add(&sop->so_perclient, &clp->cl_openowners);
2084 sop->so_is_open_owner = 1;
2085 sop->so_id = current_ownerid++;
2086 sop->so_client = clp;
2087 sop->so_seqid = open->op_seqid;
2088 sop->so_confirmed = 0;
2089 rp = &sop->so_replay;
2090 rp->rp_status = nfserr_serverfault;
2092 rp->rp_buf = rp->rp_ibuf;
2097 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2098 struct nfs4_stateowner *sop = open->op_stateowner;
2099 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2101 INIT_LIST_HEAD(&stp->st_hash);
2102 INIT_LIST_HEAD(&stp->st_perstateowner);
2103 INIT_LIST_HEAD(&stp->st_lockowners);
2104 INIT_LIST_HEAD(&stp->st_perfile);
2105 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2106 list_add(&stp->st_perstateowner, &sop->so_stateids);
2107 list_add(&stp->st_perfile, &fp->fi_stateids);
2108 stp->st_stateowner = sop;
2111 stp->st_stateid.si_boot = boot_time;
2112 stp->st_stateid.si_stateownerid = sop->so_id;
2113 stp->st_stateid.si_fileid = fp->fi_id;
2114 stp->st_stateid.si_generation = 0;
2115 stp->st_access_bmap = 0;
2116 stp->st_deny_bmap = 0;
2117 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2118 &stp->st_access_bmap);
2119 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2120 stp->st_openstp = NULL;
2124 move_to_close_lru(struct nfs4_stateowner *sop)
2126 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2128 list_move_tail(&sop->so_close_lru, &close_lru);
2129 sop->so_time = get_seconds();
2133 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2136 return (sop->so_owner.len == owner->len) &&
2137 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2138 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2141 static struct nfs4_stateowner *
2142 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2144 struct nfs4_stateowner *so = NULL;
2146 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2147 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2153 /* search file_hashtbl[] for file */
2154 static struct nfs4_file *
2155 find_file(struct inode *ino)
2157 unsigned int hashval = file_hashval(ino);
2158 struct nfs4_file *fp;
2160 spin_lock(&recall_lock);
2161 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2162 if (fp->fi_inode == ino) {
2164 spin_unlock(&recall_lock);
2168 spin_unlock(&recall_lock);
2172 static inline int access_valid(u32 x, u32 minorversion)
2174 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2176 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2178 x &= ~NFS4_SHARE_ACCESS_MASK;
2179 if (minorversion && x) {
2180 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2182 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2184 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2191 static inline int deny_valid(u32 x)
2193 /* Note: unlike access bits, deny bits may be zero. */
2194 return x <= NFS4_SHARE_DENY_BOTH;
2198 * Called to check deny when READ with all zero stateid or
2199 * WRITE with all zero or all one stateid
2202 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2204 struct inode *ino = current_fh->fh_dentry->d_inode;
2205 struct nfs4_file *fp;
2206 struct nfs4_stateid *stp;
2209 dprintk("NFSD: nfs4_share_conflict\n");
2211 fp = find_file(ino);
2214 ret = nfserr_locked;
2215 /* Search for conflicting share reservations */
2216 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2217 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2218 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2228 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2230 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2231 nfs4_file_put_access(fp, O_WRONLY);
2232 if (share_access & NFS4_SHARE_ACCESS_READ)
2233 nfs4_file_put_access(fp, O_RDONLY);
2237 * Spawn a thread to perform a recall on the delegation represented
2238 * by the lease (file_lock)
2240 * Called from break_lease() with lock_kernel() held.
2241 * Note: we assume break_lease will only call this *once* for any given
2245 void nfsd_break_deleg_cb(struct file_lock *fl)
2247 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2249 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2253 /* We're assuming the state code never drops its reference
2254 * without first removing the lease. Since we're in this lease
2255 * callback (and since the lease code is serialized by the kernel
2256 * lock) we know the server hasn't removed the lease yet, we know
2257 * it's safe to take a reference: */
2258 atomic_inc(&dp->dl_count);
2260 spin_lock(&recall_lock);
2261 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2262 spin_unlock(&recall_lock);
2264 /* only place dl_time is set. protected by lock_kernel*/
2265 dp->dl_time = get_seconds();
2268 * We don't want the locks code to timeout the lease for us;
2269 * we'll remove it ourself if the delegation isn't returned
2272 fl->fl_break_time = 0;
2274 dp->dl_file->fi_had_conflict = true;
2275 nfsd4_cb_recall(dp);
2279 * The file_lock is being reapd.
2281 * Called by locks_free_lock() with lock_kernel() held.
2284 void nfsd_release_deleg_cb(struct file_lock *fl)
2286 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2288 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2290 if (!(fl->fl_flags & FL_LEASE) || !dp)
2292 dp->dl_flock = NULL;
2296 * Set the delegation file_lock back pointer.
2298 * Called from setlease() with lock_kernel() held.
2301 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2303 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2305 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2312 * Called from setlease() with lock_kernel() held
2315 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2317 struct nfs4_delegation *onlistd =
2318 (struct nfs4_delegation *)onlist->fl_owner;
2319 struct nfs4_delegation *tryd =
2320 (struct nfs4_delegation *)try->fl_owner;
2322 if (onlist->fl_lmops != try->fl_lmops)
2325 return onlistd->dl_client == tryd->dl_client;
2330 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2333 return lease_modify(onlist, arg);
2338 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2339 .fl_break = nfsd_break_deleg_cb,
2340 .fl_release_private = nfsd_release_deleg_cb,
2341 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2342 .fl_mylease = nfsd_same_client_deleg_cb,
2343 .fl_change = nfsd_change_deleg_cb,
2348 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2349 struct nfsd4_open *open)
2351 clientid_t *clientid = &open->op_clientid;
2352 struct nfs4_client *clp = NULL;
2353 unsigned int strhashval;
2354 struct nfs4_stateowner *sop = NULL;
2356 if (!check_name(open->op_owner))
2357 return nfserr_inval;
2359 if (STALE_CLIENTID(&open->op_clientid))
2360 return nfserr_stale_clientid;
2362 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2363 sop = find_openstateowner_str(strhashval, open);
2364 open->op_stateowner = sop;
2366 /* Make sure the client's lease hasn't expired. */
2367 clp = find_confirmed_client(clientid);
2369 return nfserr_expired;
2372 /* When sessions are used, skip open sequenceid processing */
2373 if (nfsd4_has_session(cstate))
2375 if (!sop->so_confirmed) {
2376 /* Replace unconfirmed owners without checking for replay. */
2377 clp = sop->so_client;
2378 release_openowner(sop);
2379 open->op_stateowner = NULL;
2382 if (open->op_seqid == sop->so_seqid - 1) {
2383 if (sop->so_replay.rp_buflen)
2384 return nfserr_replay_me;
2385 /* The original OPEN failed so spectacularly
2386 * that we don't even have replay data saved!
2387 * Therefore, we have no choice but to continue
2388 * processing this OPEN; presumably, we'll
2389 * fail again for the same reason.
2391 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2394 if (open->op_seqid != sop->so_seqid)
2395 return nfserr_bad_seqid;
2397 if (open->op_stateowner == NULL) {
2398 sop = alloc_init_open_stateowner(strhashval, clp, open);
2400 return nfserr_resource;
2401 open->op_stateowner = sop;
2403 list_del_init(&sop->so_close_lru);
2404 renew_client(sop->so_client);
2408 static inline __be32
2409 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2411 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2412 return nfserr_openmode;
2417 static struct nfs4_delegation *
2418 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2420 struct nfs4_delegation *dp;
2422 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2423 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2429 int share_access_to_flags(u32 share_access)
2431 share_access &= ~NFS4_SHARE_WANT_MASK;
2433 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2437 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2438 struct nfs4_delegation **dp)
2441 __be32 status = nfserr_bad_stateid;
2443 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2446 flags = share_access_to_flags(open->op_share_access);
2447 status = nfs4_check_delegmode(*dp, flags);
2451 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2455 open->op_stateowner->so_confirmed = 1;
2460 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2462 struct nfs4_stateid *local;
2463 __be32 status = nfserr_share_denied;
2464 struct nfs4_stateowner *sop = open->op_stateowner;
2466 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2467 /* ignore lock owners */
2468 if (local->st_stateowner->so_is_open_owner == 0)
2470 /* remember if we have seen this open owner */
2471 if (local->st_stateowner == sop)
2473 /* check for conflicting share reservations */
2474 if (!test_share(local, open))
2482 static inline struct nfs4_stateid *
2483 nfs4_alloc_stateid(void)
2485 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2488 static inline int nfs4_access_to_access(u32 nfs4_access)
2492 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2493 flags |= NFSD_MAY_READ;
2494 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2495 flags |= NFSD_MAY_WRITE;
2499 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2500 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2503 int oflag = nfs4_access_to_omode(nfs4_access);
2504 int access = nfs4_access_to_access(nfs4_access);
2506 if (!fp->fi_fds[oflag]) {
2507 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2508 &fp->fi_fds[oflag]);
2509 if (status == nfserr_dropit)
2510 status = nfserr_jukebox;
2514 nfs4_file_get_access(fp, oflag);
2520 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2521 struct nfs4_file *fp, struct svc_fh *cur_fh,
2522 struct nfsd4_open *open)
2524 struct nfs4_stateid *stp;
2527 stp = nfs4_alloc_stateid();
2529 return nfserr_resource;
2531 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2533 kmem_cache_free(stateid_slab, stp);
2540 static inline __be32
2541 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2542 struct nfsd4_open *open)
2544 struct iattr iattr = {
2545 .ia_valid = ATTR_SIZE,
2548 if (!open->op_truncate)
2550 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2551 return nfserr_inval;
2552 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2556 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2558 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2562 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2564 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2568 status = nfsd4_truncate(rqstp, cur_fh, open);
2571 int oflag = nfs4_access_to_omode(new_access);
2572 nfs4_file_put_access(fp, oflag);
2576 /* remember the open */
2577 __set_bit(op_share_access, &stp->st_access_bmap);
2578 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2585 nfs4_set_claim_prev(struct nfsd4_open *open)
2587 open->op_stateowner->so_confirmed = 1;
2588 open->op_stateowner->so_client->cl_firststate = 1;
2592 * Attempt to hand out a delegation.
2595 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2597 struct nfs4_delegation *dp;
2598 struct nfs4_stateowner *sop = stp->st_stateowner;
2599 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2600 struct file_lock fl, *flp = &fl;
2601 int status, flag = 0;
2603 flag = NFS4_OPEN_DELEGATE_NONE;
2604 open->op_recall = 0;
2605 switch (open->op_claim_type) {
2606 case NFS4_OPEN_CLAIM_PREVIOUS:
2608 open->op_recall = 1;
2609 flag = open->op_delegate_type;
2610 if (flag == NFS4_OPEN_DELEGATE_NONE)
2613 case NFS4_OPEN_CLAIM_NULL:
2614 /* Let's not give out any delegations till everyone's
2615 * had the chance to reclaim theirs.... */
2616 if (locks_in_grace())
2618 if (!cb_up || !sop->so_confirmed)
2620 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2621 flag = NFS4_OPEN_DELEGATE_WRITE;
2623 flag = NFS4_OPEN_DELEGATE_READ;
2629 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2631 flag = NFS4_OPEN_DELEGATE_NONE;
2634 locks_init_lock(&fl);
2635 fl.fl_lmops = &nfsd_lease_mng_ops;
2636 fl.fl_flags = FL_LEASE;
2637 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2638 fl.fl_end = OFFSET_MAX;
2639 fl.fl_owner = (fl_owner_t)dp;
2640 fl.fl_file = find_readable_file(stp->st_file);
2641 BUG_ON(!fl.fl_file);
2642 fl.fl_pid = current->tgid;
2644 /* vfs_setlease checks to see if delegation should be handed out.
2645 * the lock_manager callbacks fl_mylease and fl_change are used
2647 if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
2648 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2649 unhash_delegation(dp);
2650 flag = NFS4_OPEN_DELEGATE_NONE;
2654 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2656 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2657 STATEID_VAL(&dp->dl_stateid));
2659 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2660 && flag == NFS4_OPEN_DELEGATE_NONE
2661 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2662 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2663 open->op_delegate_type = flag;
2667 * called with nfs4_lock_state() held.
2670 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2672 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2673 struct nfs4_file *fp = NULL;
2674 struct inode *ino = current_fh->fh_dentry->d_inode;
2675 struct nfs4_stateid *stp = NULL;
2676 struct nfs4_delegation *dp = NULL;
2679 status = nfserr_inval;
2680 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2681 || !deny_valid(open->op_share_deny))
2684 * Lookup file; if found, lookup stateid and check open request,
2685 * and check for delegations in the process of being recalled.
2686 * If not found, create the nfs4_file struct
2688 fp = find_file(ino);
2690 if ((status = nfs4_check_open(fp, open, &stp)))
2692 status = nfs4_check_deleg(fp, open, &dp);
2696 status = nfserr_bad_stateid;
2697 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2699 status = nfserr_resource;
2700 fp = alloc_init_file(ino);
2706 * OPEN the file, or upgrade an existing OPEN.
2707 * If truncate fails, the OPEN fails.
2710 /* Stateid was found, this is an OPEN upgrade */
2711 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2714 update_stateid(&stp->st_stateid);
2716 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2719 init_stateid(stp, fp, open);
2720 status = nfsd4_truncate(rqstp, current_fh, open);
2722 release_open_stateid(stp);
2725 if (nfsd4_has_session(&resp->cstate))
2726 update_stateid(&stp->st_stateid);
2728 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2730 if (nfsd4_has_session(&resp->cstate))
2731 open->op_stateowner->so_confirmed = 1;
2734 * Attempt to hand out a delegation. No error return, because the
2735 * OPEN succeeds even if we fail.
2737 nfs4_open_delegation(current_fh, open, stp);
2741 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2742 STATEID_VAL(&stp->st_stateid));
2746 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2747 nfs4_set_claim_prev(open);
2749 * To finish the open response, we just need to set the rflags.
2751 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2752 if (!open->op_stateowner->so_confirmed &&
2753 !nfsd4_has_session(&resp->cstate))
2754 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2760 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2763 struct nfs4_client *clp;
2767 dprintk("process_renew(%08x/%08x): starting\n",
2768 clid->cl_boot, clid->cl_id);
2769 status = nfserr_stale_clientid;
2770 if (STALE_CLIENTID(clid))
2772 clp = find_confirmed_client(clid);
2773 status = nfserr_expired;
2775 /* We assume the client took too long to RENEW. */
2776 dprintk("nfsd4_renew: clientid not found!\n");
2780 status = nfserr_cb_path_down;
2781 if (!list_empty(&clp->cl_delegations)
2782 && !atomic_read(&clp->cl_cb_set))
2786 nfs4_unlock_state();
2790 struct lock_manager nfsd4_manager = {
2794 nfsd4_end_grace(void)
2796 dprintk("NFSD: end of grace period\n");
2797 nfsd4_recdir_purge_old();
2798 locks_end_grace(&nfsd4_manager);
2800 * Now that every NFSv4 client has had the chance to recover and
2801 * to see the (possibly new, possibly shorter) lease time, we
2802 * can safely set the next grace time to the current lease time:
2804 nfsd4_grace = nfsd4_lease;
2808 nfs4_laundromat(void)
2810 struct nfs4_client *clp;
2811 struct nfs4_stateowner *sop;
2812 struct nfs4_delegation *dp;
2813 struct list_head *pos, *next, reaplist;
2814 time_t cutoff = get_seconds() - nfsd4_lease;
2815 time_t t, clientid_val = nfsd4_lease;
2816 time_t u, test_val = nfsd4_lease;
2820 dprintk("NFSD: laundromat service - starting\n");
2821 if (locks_in_grace())
2823 INIT_LIST_HEAD(&reaplist);
2824 spin_lock(&client_lock);
2825 list_for_each_safe(pos, next, &client_lru) {
2826 clp = list_entry(pos, struct nfs4_client, cl_lru);
2827 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2828 t = clp->cl_time - cutoff;
2829 if (clientid_val > t)
2833 if (atomic_read(&clp->cl_refcount)) {
2834 dprintk("NFSD: client in use (clientid %08x)\n",
2835 clp->cl_clientid.cl_id);
2838 unhash_client_locked(clp);
2839 list_add(&clp->cl_lru, &reaplist);
2841 spin_unlock(&client_lock);
2842 list_for_each_safe(pos, next, &reaplist) {
2843 clp = list_entry(pos, struct nfs4_client, cl_lru);
2844 dprintk("NFSD: purging unused client (clientid %08x)\n",
2845 clp->cl_clientid.cl_id);
2846 nfsd4_remove_clid_dir(clp);
2849 spin_lock(&recall_lock);
2850 list_for_each_safe(pos, next, &del_recall_lru) {
2851 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2852 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2853 u = dp->dl_time - cutoff;
2858 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2860 list_move(&dp->dl_recall_lru, &reaplist);
2862 spin_unlock(&recall_lock);
2863 list_for_each_safe(pos, next, &reaplist) {
2864 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2865 list_del_init(&dp->dl_recall_lru);
2866 unhash_delegation(dp);
2868 test_val = nfsd4_lease;
2869 list_for_each_safe(pos, next, &close_lru) {
2870 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2871 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2872 u = sop->so_time - cutoff;
2877 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2879 release_openowner(sop);
2881 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2882 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2883 nfs4_unlock_state();
2884 return clientid_val;
2887 static struct workqueue_struct *laundry_wq;
2888 static void laundromat_main(struct work_struct *);
2889 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2892 laundromat_main(struct work_struct *not_used)
2896 t = nfs4_laundromat();
2897 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2898 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2901 static struct nfs4_stateowner *
2902 search_close_lru(u32 st_id, int flags)
2904 struct nfs4_stateowner *local = NULL;
2906 if (flags & CLOSE_STATE) {
2907 list_for_each_entry(local, &close_lru, so_close_lru) {
2908 if (local->so_id == st_id)
2916 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2918 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2922 STALE_STATEID(stateid_t *stateid)
2924 if (stateid->si_boot == boot_time)
2926 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2927 STATEID_VAL(stateid));
2932 access_permit_read(unsigned long access_bmap)
2934 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2935 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2936 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2940 access_permit_write(unsigned long access_bmap)
2942 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2943 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2947 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2949 __be32 status = nfserr_openmode;
2951 /* For lock stateid's, we test the parent open, not the lock: */
2952 if (stp->st_openstp)
2953 stp = stp->st_openstp;
2954 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2956 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2963 static inline __be32
2964 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2966 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2968 else if (locks_in_grace()) {
2969 /* Answer in remaining cases depends on existance of
2970 * conflicting state; so we must wait out the grace period. */
2971 return nfserr_grace;
2972 } else if (flags & WR_STATE)
2973 return nfs4_share_conflict(current_fh,
2974 NFS4_SHARE_DENY_WRITE);
2975 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2976 return nfs4_share_conflict(current_fh,
2977 NFS4_SHARE_DENY_READ);
2981 * Allow READ/WRITE during grace period on recovered state only for files
2982 * that are not able to provide mandatory locking.
2985 grace_disallows_io(struct inode *inode)
2987 return locks_in_grace() && mandatory_lock(inode);
2990 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2993 * When sessions are used the stateid generation number is ignored
2996 if ((flags & HAS_SESSION) && in->si_generation == 0)
2999 /* If the client sends us a stateid from the future, it's buggy: */
3000 if (in->si_generation > ref->si_generation)
3001 return nfserr_bad_stateid;
3003 * The following, however, can happen. For example, if the
3004 * client sends an open and some IO at the same time, the open
3005 * may bump si_generation while the IO is still in flight.
3006 * Thanks to hard links and renames, the client never knows what
3007 * file an open will affect. So it could avoid that situation
3008 * only by serializing all opens and IO from the same open
3009 * owner. To recover from the old_stateid error, the client
3010 * will just have to retry the IO:
3012 if (in->si_generation < ref->si_generation)
3013 return nfserr_old_stateid;
3018 static int is_delegation_stateid(stateid_t *stateid)
3020 return stateid->si_fileid == 0;
3024 * Checks for stateid operations
3027 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3028 stateid_t *stateid, int flags, struct file **filpp)
3030 struct nfs4_stateid *stp = NULL;
3031 struct nfs4_delegation *dp = NULL;
3032 struct svc_fh *current_fh = &cstate->current_fh;
3033 struct inode *ino = current_fh->fh_dentry->d_inode;
3039 if (grace_disallows_io(ino))
3040 return nfserr_grace;
3042 if (nfsd4_has_session(cstate))
3043 flags |= HAS_SESSION;
3045 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3046 return check_special_stateids(current_fh, stateid, flags);
3048 status = nfserr_stale_stateid;
3049 if (STALE_STATEID(stateid))
3053 * We assume that any stateid that has the current boot time,
3054 * but that we can't find, is expired:
3056 status = nfserr_expired;
3057 if (is_delegation_stateid(stateid)) {
3058 dp = find_delegation_stateid(ino, stateid);
3061 status = check_stateid_generation(stateid, &dp->dl_stateid,
3065 status = nfs4_check_delegmode(dp, flags);
3068 renew_client(dp->dl_client);
3070 *filpp = find_readable_file(dp->dl_file);
3072 } else { /* open or lock stateid */
3073 stp = find_stateid(stateid, flags);
3076 status = nfserr_bad_stateid;
3077 if (nfs4_check_fh(current_fh, stp))
3079 if (!stp->st_stateowner->so_confirmed)
3081 status = check_stateid_generation(stateid, &stp->st_stateid,
3085 status = nfs4_check_openmode(stp, flags);
3088 renew_client(stp->st_stateowner->so_client);
3090 if (flags & RD_STATE)
3091 *filpp = find_readable_file(stp->st_file);
3093 *filpp = find_writeable_file(stp->st_file);
3104 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3105 RD_STATE : WR_STATE;
3109 * Checks for sequence id mutating operations.
3112 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3113 stateid_t *stateid, int flags,
3114 struct nfs4_stateowner **sopp,
3115 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3117 struct nfs4_stateid *stp;
3118 struct nfs4_stateowner *sop;
3119 struct svc_fh *current_fh = &cstate->current_fh;
3122 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3123 seqid, STATEID_VAL(stateid));
3128 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3129 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3130 return nfserr_bad_stateid;
3133 if (STALE_STATEID(stateid))
3134 return nfserr_stale_stateid;
3136 if (nfsd4_has_session(cstate))
3137 flags |= HAS_SESSION;
3140 * We return BAD_STATEID if filehandle doesn't match stateid,
3141 * the confirmed flag is incorrecly set, or the generation
3142 * number is incorrect.
3144 stp = find_stateid(stateid, flags);
3147 * Also, we should make sure this isn't just the result of
3150 sop = search_close_lru(stateid->si_stateownerid, flags);
3151 /* It's not stale; let's assume it's expired: */
3153 return nfserr_expired;
3159 *sopp = sop = stp->st_stateowner;
3162 clientid_t *lockclid = &lock->v.new.clientid;
3163 struct nfs4_client *clp = sop->so_client;
3167 lkflg = setlkflg(lock->lk_type);
3169 if (lock->lk_is_new) {
3170 if (!sop->so_is_open_owner)
3171 return nfserr_bad_stateid;
3172 if (!(flags & HAS_SESSION) &&
3173 !same_clid(&clp->cl_clientid, lockclid))
3174 return nfserr_bad_stateid;
3175 /* stp is the open stateid */
3176 status = nfs4_check_openmode(stp, lkflg);
3180 /* stp is the lock stateid */
3181 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3187 if (nfs4_check_fh(current_fh, stp)) {
3188 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3189 return nfserr_bad_stateid;
3193 * We now validate the seqid and stateid generation numbers.
3194 * For the moment, we ignore the possibility of
3195 * generation number wraparound.
3197 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3200 if (sop->so_confirmed && flags & CONFIRM) {
3201 dprintk("NFSD: preprocess_seqid_op: expected"
3202 " unconfirmed stateowner!\n");
3203 return nfserr_bad_stateid;
3205 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3206 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3207 " confirmed yet!\n");
3208 return nfserr_bad_stateid;
3210 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3213 renew_client(sop->so_client);
3217 if (seqid == sop->so_seqid - 1) {
3218 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3219 /* indicate replay to calling function */
3220 return nfserr_replay_me;
3222 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3223 sop->so_seqid, seqid);
3225 return nfserr_bad_seqid;
3229 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3230 struct nfsd4_open_confirm *oc)
3233 struct nfs4_stateowner *sop;
3234 struct nfs4_stateid *stp;
3236 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3237 (int)cstate->current_fh.fh_dentry->d_name.len,
3238 cstate->current_fh.fh_dentry->d_name.name);
3240 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3246 if ((status = nfs4_preprocess_seqid_op(cstate,
3247 oc->oc_seqid, &oc->oc_req_stateid,
3248 CONFIRM | OPEN_STATE,
3249 &oc->oc_stateowner, &stp, NULL)))
3252 sop = oc->oc_stateowner;
3253 sop->so_confirmed = 1;
3254 update_stateid(&stp->st_stateid);
3255 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3256 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3257 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3259 nfsd4_create_clid_dir(sop->so_client);
3261 if (oc->oc_stateowner) {
3262 nfs4_get_stateowner(oc->oc_stateowner);
3263 cstate->replay_owner = oc->oc_stateowner;
3265 nfs4_unlock_state();
3271 * unset all bits in union bitmap (bmap) that
3272 * do not exist in share (from successful OPEN_DOWNGRADE)
3275 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3278 for (i = 1; i < 4; i++) {
3279 if ((i & access) != i)
3280 __clear_bit(i, bmap);
3285 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3288 for (i = 0; i < 4; i++) {
3289 if ((i & deny) != i)
3290 __clear_bit(i, bmap);
3295 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3296 struct nfsd4_compound_state *cstate,
3297 struct nfsd4_open_downgrade *od)
3300 struct nfs4_stateid *stp;
3301 unsigned int share_access;
3303 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3304 (int)cstate->current_fh.fh_dentry->d_name.len,
3305 cstate->current_fh.fh_dentry->d_name.name);
3307 if (!access_valid(od->od_share_access, cstate->minorversion)
3308 || !deny_valid(od->od_share_deny))
3309 return nfserr_inval;
3312 if ((status = nfs4_preprocess_seqid_op(cstate,
3316 &od->od_stateowner, &stp, NULL)))
3319 status = nfserr_inval;
3320 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3321 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3322 stp->st_access_bmap, od->od_share_access);
3325 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3326 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3327 stp->st_deny_bmap, od->od_share_deny);
3330 set_access(&share_access, stp->st_access_bmap);
3331 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3333 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3334 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3336 update_stateid(&stp->st_stateid);
3337 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3340 if (od->od_stateowner) {
3341 nfs4_get_stateowner(od->od_stateowner);
3342 cstate->replay_owner = od->od_stateowner;
3344 nfs4_unlock_state();
3349 * nfs4_unlock_state() called after encode
3352 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3353 struct nfsd4_close *close)
3356 struct nfs4_stateid *stp;
3358 dprintk("NFSD: nfsd4_close on file %.*s\n",
3359 (int)cstate->current_fh.fh_dentry->d_name.len,
3360 cstate->current_fh.fh_dentry->d_name.name);
3363 /* check close_lru for replay */
3364 if ((status = nfs4_preprocess_seqid_op(cstate,
3367 OPEN_STATE | CLOSE_STATE,
3368 &close->cl_stateowner, &stp, NULL)))
3371 update_stateid(&stp->st_stateid);
3372 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3374 /* release_stateid() calls nfsd_close() if needed */
3375 release_open_stateid(stp);
3377 /* place unused nfs4_stateowners on so_close_lru list to be
3378 * released by the laundromat service after the lease period
3379 * to enable us to handle CLOSE replay
3381 if (list_empty(&close->cl_stateowner->so_stateids))
3382 move_to_close_lru(close->cl_stateowner);
3384 if (close->cl_stateowner) {
3385 nfs4_get_stateowner(close->cl_stateowner);
3386 cstate->replay_owner = close->cl_stateowner;
3388 nfs4_unlock_state();
3393 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3394 struct nfsd4_delegreturn *dr)
3396 struct nfs4_delegation *dp;
3397 stateid_t *stateid = &dr->dr_stateid;
3398 struct inode *inode;
3402 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3404 inode = cstate->current_fh.fh_dentry->d_inode;
3406 if (nfsd4_has_session(cstate))
3407 flags |= HAS_SESSION;
3409 status = nfserr_bad_stateid;
3410 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3412 status = nfserr_stale_stateid;
3413 if (STALE_STATEID(stateid))
3415 status = nfserr_bad_stateid;
3416 if (!is_delegation_stateid(stateid))
3418 status = nfserr_expired;
3419 dp = find_delegation_stateid(inode, stateid);
3422 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3425 renew_client(dp->dl_client);
3427 unhash_delegation(dp);
3429 nfs4_unlock_state();
3436 * Lock owner state (byte-range locks)
3438 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3439 #define LOCK_HASH_BITS 8
3440 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3441 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3444 end_offset(u64 start, u64 len)
3449 return end >= start ? end: NFS4_MAX_UINT64;
3452 /* last octet in a range */
3454 last_byte_offset(u64 start, u64 len)
3460 return end > start ? end - 1: NFS4_MAX_UINT64;
3463 #define lockownerid_hashval(id) \
3464 ((id) & LOCK_HASH_MASK)
3466 static inline unsigned int
3467 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3468 struct xdr_netobj *ownername)
3470 return (file_hashval(inode) + cl_id
3471 + opaque_hashval(ownername->data, ownername->len))
3475 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3476 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3477 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3479 static struct nfs4_stateid *
3480 find_stateid(stateid_t *stid, int flags)
3482 struct nfs4_stateid *local;
3483 u32 st_id = stid->si_stateownerid;
3484 u32 f_id = stid->si_fileid;
3485 unsigned int hashval;
3487 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3488 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3489 hashval = stateid_hashval(st_id, f_id);
3490 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3491 if ((local->st_stateid.si_stateownerid == st_id) &&
3492 (local->st_stateid.si_fileid == f_id))
3497 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3498 hashval = stateid_hashval(st_id, f_id);
3499 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3500 if ((local->st_stateid.si_stateownerid == st_id) &&
3501 (local->st_stateid.si_fileid == f_id))
3508 static struct nfs4_delegation *
3509 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3511 struct nfs4_file *fp;
3512 struct nfs4_delegation *dl;
3514 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3517 fp = find_file(ino);
3520 dl = find_delegation_file(fp, stid);
3526 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3527 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3528 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3529 * locking, this prevents us from being completely protocol-compliant. The
3530 * real solution to this problem is to start using unsigned file offsets in
3531 * the VFS, but this is a very deep change!
3534 nfs4_transform_lock_offset(struct file_lock *lock)
3536 if (lock->fl_start < 0)
3537 lock->fl_start = OFFSET_MAX;
3538 if (lock->fl_end < 0)
3539 lock->fl_end = OFFSET_MAX;
3542 /* Hack!: For now, we're defining this just so we can use a pointer to it
3543 * as a unique cookie to identify our (NFSv4's) posix locks. */
3544 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3548 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3550 struct nfs4_stateowner *sop;
3552 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3553 sop = (struct nfs4_stateowner *) fl->fl_owner;
3554 kref_get(&sop->so_ref);
3556 deny->ld_clientid = sop->so_client->cl_clientid;
3558 deny->ld_sop = NULL;
3559 deny->ld_clientid.cl_boot = 0;
3560 deny->ld_clientid.cl_id = 0;
3562 deny->ld_start = fl->fl_start;
3563 deny->ld_length = NFS4_MAX_UINT64;
3564 if (fl->fl_end != NFS4_MAX_UINT64)
3565 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3566 deny->ld_type = NFS4_READ_LT;
3567 if (fl->fl_type != F_RDLCK)
3568 deny->ld_type = NFS4_WRITE_LT;
3571 static struct nfs4_stateowner *
3572 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3573 struct xdr_netobj *owner)
3575 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3576 struct nfs4_stateowner *op;
3578 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3579 if (same_owner_str(op, owner, clid))
3586 * Alloc a lock owner structure.
3587 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3590 * strhashval = lock_ownerstr_hashval
3593 static struct nfs4_stateowner *
3594 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3595 struct nfs4_stateowner *sop;
3596 struct nfs4_replay *rp;
3597 unsigned int idhashval;
3599 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3601 idhashval = lockownerid_hashval(current_ownerid);
3602 INIT_LIST_HEAD(&sop->so_idhash);
3603 INIT_LIST_HEAD(&sop->so_strhash);
3604 INIT_LIST_HEAD(&sop->so_perclient);
3605 INIT_LIST_HEAD(&sop->so_stateids);
3606 INIT_LIST_HEAD(&sop->so_perstateid);
3607 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3609 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3610 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3611 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3612 sop->so_is_open_owner = 0;
3613 sop->so_id = current_ownerid++;
3614 sop->so_client = clp;
3615 /* It is the openowner seqid that will be incremented in encode in the
3616 * case of new lockowners; so increment the lock seqid manually: */
3617 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3618 sop->so_confirmed = 1;
3619 rp = &sop->so_replay;
3620 rp->rp_status = nfserr_serverfault;
3622 rp->rp_buf = rp->rp_ibuf;
3626 static struct nfs4_stateid *
3627 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3629 struct nfs4_stateid *stp;
3630 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3632 stp = nfs4_alloc_stateid();
3635 INIT_LIST_HEAD(&stp->st_hash);
3636 INIT_LIST_HEAD(&stp->st_perfile);
3637 INIT_LIST_HEAD(&stp->st_perstateowner);
3638 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3639 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3640 list_add(&stp->st_perfile, &fp->fi_stateids);
3641 list_add(&stp->st_perstateowner, &sop->so_stateids);
3642 stp->st_stateowner = sop;
3645 stp->st_stateid.si_boot = boot_time;
3646 stp->st_stateid.si_stateownerid = sop->so_id;
3647 stp->st_stateid.si_fileid = fp->fi_id;
3648 stp->st_stateid.si_generation = 0;
3649 stp->st_deny_bmap = open_stp->st_deny_bmap;
3650 stp->st_openstp = open_stp;
3657 check_lock_length(u64 offset, u64 length)
3659 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3660 LOFF_OVERFLOW(offset, length)));
3667 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3668 struct nfsd4_lock *lock)
3670 struct nfs4_stateowner *open_sop = NULL;
3671 struct nfs4_stateowner *lock_sop = NULL;
3672 struct nfs4_stateid *lock_stp;
3673 struct nfs4_file *fp;
3674 struct file *filp = NULL;
3675 struct file_lock file_lock;
3676 struct file_lock conflock;
3678 unsigned int strhashval;
3682 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3683 (long long) lock->lk_offset,
3684 (long long) lock->lk_length);
3686 if (check_lock_length(lock->lk_offset, lock->lk_length))
3687 return nfserr_inval;
3689 if ((status = fh_verify(rqstp, &cstate->current_fh,
3690 S_IFREG, NFSD_MAY_LOCK))) {
3691 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3697 if (lock->lk_is_new) {
3699 * Client indicates that this is a new lockowner.
3700 * Use open owner and open stateid to create lock owner and
3703 struct nfs4_stateid *open_stp = NULL;
3705 status = nfserr_stale_clientid;
3706 if (!nfsd4_has_session(cstate) &&
3707 STALE_CLIENTID(&lock->lk_new_clientid))
3710 /* validate and update open stateid and open seqid */
3711 status = nfs4_preprocess_seqid_op(cstate,
3712 lock->lk_new_open_seqid,
3713 &lock->lk_new_open_stateid,
3715 &lock->lk_replay_owner, &open_stp,
3719 open_sop = lock->lk_replay_owner;
3720 /* create lockowner and lock stateid */
3721 fp = open_stp->st_file;
3722 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3723 open_sop->so_client->cl_clientid.cl_id,
3724 &lock->v.new.owner);
3725 /* XXX: Do we need to check for duplicate stateowners on
3726 * the same file, or should they just be allowed (and
3727 * create new stateids)? */
3728 status = nfserr_resource;
3729 lock_sop = alloc_init_lock_stateowner(strhashval,
3730 open_sop->so_client, open_stp, lock);
3731 if (lock_sop == NULL)
3733 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3734 if (lock_stp == NULL)
3737 /* lock (lock owner + lock stateid) already exists */
3738 status = nfs4_preprocess_seqid_op(cstate,
3739 lock->lk_old_lock_seqid,
3740 &lock->lk_old_lock_stateid,
3742 &lock->lk_replay_owner, &lock_stp, lock);
3745 lock_sop = lock->lk_replay_owner;
3746 fp = lock_stp->st_file;
3748 /* lock->lk_replay_owner and lock_stp have been created or found */
3750 status = nfserr_grace;
3751 if (locks_in_grace() && !lock->lk_reclaim)
3753 status = nfserr_no_grace;
3754 if (!locks_in_grace() && lock->lk_reclaim)
3757 locks_init_lock(&file_lock);
3758 switch (lock->lk_type) {
3761 if (find_readable_file(lock_stp->st_file)) {
3762 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3763 filp = find_readable_file(lock_stp->st_file);
3765 file_lock.fl_type = F_RDLCK;
3769 case NFS4_WRITEW_LT:
3770 if (find_writeable_file(lock_stp->st_file)) {
3771 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3772 filp = find_writeable_file(lock_stp->st_file);
3774 file_lock.fl_type = F_WRLCK;
3778 status = nfserr_inval;
3782 status = nfserr_openmode;
3785 file_lock.fl_owner = (fl_owner_t)lock_sop;
3786 file_lock.fl_pid = current->tgid;
3787 file_lock.fl_file = filp;
3788 file_lock.fl_flags = FL_POSIX;
3789 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3791 file_lock.fl_start = lock->lk_offset;
3792 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3793 nfs4_transform_lock_offset(&file_lock);
3796 * Try to lock the file in the VFS.
3797 * Note: locks.c uses the BKL to protect the inode's lock list.
3800 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3802 case 0: /* success! */
3803 update_stateid(&lock_stp->st_stateid);
3804 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3808 case (EAGAIN): /* conflock holds conflicting lock */
3809 status = nfserr_denied;
3810 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3811 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3814 status = nfserr_deadlock;
3817 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3818 status = nfserr_resource;
3822 if (status && lock->lk_is_new && lock_sop)
3823 release_lockowner(lock_sop);
3824 if (lock->lk_replay_owner) {
3825 nfs4_get_stateowner(lock->lk_replay_owner);
3826 cstate->replay_owner = lock->lk_replay_owner;
3828 nfs4_unlock_state();
3833 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3834 * so we do a temporary open here just to get an open file to pass to
3835 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3838 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3843 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3846 err = vfs_test_lock(file, lock);
3855 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3856 struct nfsd4_lockt *lockt)
3858 struct inode *inode;
3859 struct file_lock file_lock;
3863 if (locks_in_grace())
3864 return nfserr_grace;
3866 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3867 return nfserr_inval;
3869 lockt->lt_stateowner = NULL;
3872 status = nfserr_stale_clientid;
3873 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3876 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3877 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3878 if (status == nfserr_symlink)
3879 status = nfserr_inval;
3883 inode = cstate->current_fh.fh_dentry->d_inode;
3884 locks_init_lock(&file_lock);
3885 switch (lockt->lt_type) {
3888 file_lock.fl_type = F_RDLCK;
3891 case NFS4_WRITEW_LT:
3892 file_lock.fl_type = F_WRLCK;
3895 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3896 status = nfserr_inval;
3900 lockt->lt_stateowner = find_lockstateowner_str(inode,
3901 &lockt->lt_clientid, &lockt->lt_owner);
3902 if (lockt->lt_stateowner)
3903 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3904 file_lock.fl_pid = current->tgid;
3905 file_lock.fl_flags = FL_POSIX;
3907 file_lock.fl_start = lockt->lt_offset;
3908 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3910 nfs4_transform_lock_offset(&file_lock);
3913 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3915 status = nfserrno(error);
3918 if (file_lock.fl_type != F_UNLCK) {
3919 status = nfserr_denied;
3920 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3923 nfs4_unlock_state();
3928 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3929 struct nfsd4_locku *locku)
3931 struct nfs4_stateid *stp;
3932 struct file *filp = NULL;
3933 struct file_lock file_lock;
3937 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3938 (long long) locku->lu_offset,
3939 (long long) locku->lu_length);
3941 if (check_lock_length(locku->lu_offset, locku->lu_length))
3942 return nfserr_inval;
3946 if ((status = nfs4_preprocess_seqid_op(cstate,
3950 &locku->lu_stateowner, &stp, NULL)))
3953 filp = find_any_file(stp->st_file);
3955 status = nfserr_lock_range;
3959 locks_init_lock(&file_lock);
3960 file_lock.fl_type = F_UNLCK;
3961 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3962 file_lock.fl_pid = current->tgid;
3963 file_lock.fl_file = filp;
3964 file_lock.fl_flags = FL_POSIX;
3965 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3966 file_lock.fl_start = locku->lu_offset;
3968 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3969 nfs4_transform_lock_offset(&file_lock);
3972 * Try to unlock the file in the VFS.
3974 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3976 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3980 * OK, unlock succeeded; the only thing left to do is update the stateid.
3982 update_stateid(&stp->st_stateid);
3983 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3986 if (locku->lu_stateowner) {
3987 nfs4_get_stateowner(locku->lu_stateowner);
3988 cstate->replay_owner = locku->lu_stateowner;
3990 nfs4_unlock_state();
3994 status = nfserrno(err);
4000 * 1: locks held by lockowner
4001 * 0: no locks held by lockowner
4004 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4006 struct file_lock **flpp;
4007 struct inode *inode = filp->fi_inode;
4011 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4012 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4023 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4024 struct nfsd4_compound_state *cstate,
4025 struct nfsd4_release_lockowner *rlockowner)
4027 clientid_t *clid = &rlockowner->rl_clientid;
4028 struct nfs4_stateowner *sop;
4029 struct nfs4_stateid *stp;
4030 struct xdr_netobj *owner = &rlockowner->rl_owner;
4031 struct list_head matches;
4035 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4036 clid->cl_boot, clid->cl_id);
4038 /* XXX check for lease expiration */
4040 status = nfserr_stale_clientid;
4041 if (STALE_CLIENTID(clid))
4046 status = nfserr_locks_held;
4047 /* XXX: we're doing a linear search through all the lockowners.
4048 * Yipes! For now we'll just hope clients aren't really using
4049 * release_lockowner much, but eventually we have to fix these
4050 * data structures. */
4051 INIT_LIST_HEAD(&matches);
4052 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4053 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4054 if (!same_owner_str(sop, owner, clid))
4056 list_for_each_entry(stp, &sop->so_stateids,
4058 if (check_for_locks(stp->st_file, sop))
4060 /* Note: so_perclient unused for lockowners,
4061 * so it's OK to fool with here. */
4062 list_add(&sop->so_perclient, &matches);
4066 /* Clients probably won't expect us to return with some (but not all)
4067 * of the lockowner state released; so don't release any until all
4068 * have been checked. */
4070 while (!list_empty(&matches)) {
4071 sop = list_entry(matches.next, struct nfs4_stateowner,
4073 /* unhash_stateowner deletes so_perclient only
4074 * for openowners. */
4075 list_del(&sop->so_perclient);
4076 release_lockowner(sop);
4079 nfs4_unlock_state();
4083 static inline struct nfs4_client_reclaim *
4086 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4090 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4092 unsigned int strhashval = clientstr_hashval(name);
4093 struct nfs4_client *clp;
4095 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
4100 * failure => all reset bets are off, nfserr_no_grace...
4103 nfs4_client_to_reclaim(const char *name)
4105 unsigned int strhashval;
4106 struct nfs4_client_reclaim *crp = NULL;
4108 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4109 crp = alloc_reclaim();
4112 strhashval = clientstr_hashval(name);
4113 INIT_LIST_HEAD(&crp->cr_strhash);
4114 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4115 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4116 reclaim_str_hashtbl_size++;
4121 nfs4_release_reclaim(void)
4123 struct nfs4_client_reclaim *crp = NULL;
4126 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4127 while (!list_empty(&reclaim_str_hashtbl[i])) {
4128 crp = list_entry(reclaim_str_hashtbl[i].next,
4129 struct nfs4_client_reclaim, cr_strhash);
4130 list_del(&crp->cr_strhash);
4132 reclaim_str_hashtbl_size--;
4135 BUG_ON(reclaim_str_hashtbl_size);
4139 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4140 static struct nfs4_client_reclaim *
4141 nfs4_find_reclaim_client(clientid_t *clid)
4143 unsigned int strhashval;
4144 struct nfs4_client *clp;
4145 struct nfs4_client_reclaim *crp = NULL;
4148 /* find clientid in conf_id_hashtbl */
4149 clp = find_confirmed_client(clid);
4153 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4154 clp->cl_name.len, clp->cl_name.data,
4157 /* find clp->cl_name in reclaim_str_hashtbl */
4158 strhashval = clientstr_hashval(clp->cl_recdir);
4159 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4160 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4168 * Called from OPEN. Look for clientid in reclaim list.
4171 nfs4_check_open_reclaim(clientid_t *clid)
4173 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4176 /* initialization to perform at module load time: */
4179 nfs4_state_init(void)
4183 status = nfsd4_init_slabs();
4186 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4187 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4188 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4189 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4190 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4191 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4193 for (i = 0; i < SESSION_HASH_SIZE; i++)
4194 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4195 for (i = 0; i < FILE_HASH_SIZE; i++) {
4196 INIT_LIST_HEAD(&file_hashtbl[i]);
4198 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4199 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4200 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4202 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4203 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4204 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4206 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4207 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4208 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4210 memset(&onestateid, ~0, sizeof(stateid_t));
4211 INIT_LIST_HEAD(&close_lru);
4212 INIT_LIST_HEAD(&client_lru);
4213 INIT_LIST_HEAD(&del_recall_lru);
4214 reclaim_str_hashtbl_size = 0;
4219 nfsd4_load_reboot_recovery_data(void)
4224 nfsd4_init_recdir(user_recovery_dirname);
4225 status = nfsd4_recdir_load();
4226 nfs4_unlock_state();
4228 printk("NFSD: Failure reading reboot recovery data\n");
4232 * Since the lifetime of a delegation isn't limited to that of an open, a
4233 * client may quite reasonably hang on to a delegation as long as it has
4234 * the inode cached. This becomes an obvious problem the first time a
4235 * client's inode cache approaches the size of the server's total memory.
4237 * For now we avoid this problem by imposing a hard limit on the number
4238 * of delegations, which varies according to the server's memory size.
4241 set_max_delegations(void)
4244 * Allow at most 4 delegations per megabyte of RAM. Quick
4245 * estimates suggest that in the worst case (where every delegation
4246 * is for a different inode), a delegation could take about 1.5K,
4247 * giving a worst case usage of about 6% of memory.
4249 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4252 /* initialization to perform when the nfsd service is started: */
4255 __nfs4_state_start(void)
4259 boot_time = get_seconds();
4260 locks_start_grace(&nfsd4_manager);
4261 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4263 ret = set_callback_cred();
4266 laundry_wq = create_singlethread_workqueue("nfsd4");
4267 if (laundry_wq == NULL)
4269 ret = nfsd4_create_callback_queue();
4271 goto out_free_laundry;
4272 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4273 set_max_delegations();
4276 destroy_workqueue(laundry_wq);
4281 nfs4_state_start(void)
4283 nfsd4_load_reboot_recovery_data();
4284 return __nfs4_state_start();
4288 __nfs4_state_shutdown(void)
4291 struct nfs4_client *clp = NULL;
4292 struct nfs4_delegation *dp = NULL;
4293 struct list_head *pos, *next, reaplist;
4295 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4296 while (!list_empty(&conf_id_hashtbl[i])) {
4297 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4300 while (!list_empty(&unconf_str_hashtbl[i])) {
4301 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4305 INIT_LIST_HEAD(&reaplist);
4306 spin_lock(&recall_lock);
4307 list_for_each_safe(pos, next, &del_recall_lru) {
4308 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4309 list_move(&dp->dl_recall_lru, &reaplist);
4311 spin_unlock(&recall_lock);
4312 list_for_each_safe(pos, next, &reaplist) {
4313 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4314 list_del_init(&dp->dl_recall_lru);
4315 unhash_delegation(dp);
4318 nfsd4_shutdown_recdir();
4322 nfs4_state_shutdown(void)
4324 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4325 destroy_workqueue(laundry_wq);
4326 locks_end_grace(&nfsd4_manager);
4328 nfs4_release_reclaim();
4329 __nfs4_state_shutdown();
4330 nfs4_unlock_state();
4331 nfsd4_destroy_callback_queue();
4335 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4336 * accessed when nfsd is starting.
4339 nfs4_set_recdir(char *recdir)
4341 strcpy(user_recovery_dirname, recdir);
4345 * Change the NFSv4 recovery directory to recdir.
4348 nfs4_reset_recoverydir(char *recdir)
4353 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4357 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4358 nfs4_set_recdir(recdir);
4366 nfs4_recoverydir(void)
4368 return user_recovery_dirname;