2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
45 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 time_t nfsd4_lease = 90; /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static stateid_t zerostateid; /* bits all 0 */
55 static stateid_t onestateid; /* bits all 1 */
56 static u64 current_sessionid = 1;
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
73 * Currently used for the del_recall_lru and file hash table. In an
74 * effort to decrease the scope of the client_mutex, this spinlock may
75 * eventually cover more:
77 static DEFINE_SPINLOCK(recall_lock);
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
87 mutex_lock(&client_mutex);
91 nfs4_unlock_state(void)
93 mutex_unlock(&client_mutex);
97 opaque_hashval(const void *ptr, int nbytes)
99 unsigned char *cptr = (unsigned char *) ptr;
109 static struct list_head del_recall_lru;
112 put_nfs4_file(struct nfs4_file *fi)
114 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115 list_del(&fi->fi_hash);
116 spin_unlock(&recall_lock);
118 kmem_cache_free(file_slab, fi);
123 get_nfs4_file(struct nfs4_file *fi)
125 atomic_inc(&fi->fi_ref);
128 static int num_delegations;
129 unsigned int max_delegations;
132 * Open owner state (share locks)
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS 8
137 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
140 #define ownerid_hashval(id) \
141 ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS 8
150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS 10
154 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
157 #define file_hashval(x) \
158 hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id) \
160 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
167 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168 atomic_inc(&fp->fi_access[oflag]);
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
173 if (oflag == O_RDWR) {
174 __nfs4_file_get_access(fp, O_RDONLY);
175 __nfs4_file_get_access(fp, O_WRONLY);
177 __nfs4_file_get_access(fp, oflag);
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
182 if (fp->fi_fds[oflag]) {
183 fput(fp->fi_fds[oflag]);
184 fp->fi_fds[oflag] = NULL;
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
190 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191 nfs4_file_put_fd(fp, O_RDWR);
192 nfs4_file_put_fd(fp, oflag);
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
198 if (oflag == O_RDWR) {
199 __nfs4_file_put_access(fp, O_RDONLY);
200 __nfs4_file_put_access(fp, O_WRONLY);
202 __nfs4_file_put_access(fp, oflag);
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
208 struct nfs4_delegation *dp;
209 struct nfs4_file *fp = stp->st_file;
211 dprintk("NFSD alloc_init_deleg\n");
213 * Major work on the lease subsystem (for example, to support
214 * calbacks on stat) will be required before we can support
215 * write delegations properly.
217 if (type != NFS4_OPEN_DELEGATE_READ)
219 if (fp->fi_had_conflict)
221 if (num_delegations > max_delegations)
223 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
227 INIT_LIST_HEAD(&dp->dl_perfile);
228 INIT_LIST_HEAD(&dp->dl_perclnt);
229 INIT_LIST_HEAD(&dp->dl_recall_lru);
233 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 int nfsd4_register_conn(struct nfsd4_conn *conn)
678 conn->cn_xpt_user.callback = nfsd4_conn_lost;
679 return 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;
688 if (ses->se_flags & SESSION4_BACK_CHAN)
689 flags |= NFS4_CDFC4_BACK;
690 conn = alloc_conn(rqstp, flags);
692 return nfserr_jukebox;
693 nfsd4_hash_conn(conn, ses);
694 ret = nfsd4_register_conn(conn);
696 /* oops; xprt is already down: */
697 nfsd4_conn_lost(&conn->cn_xpt_user);
701 static void nfsd4_del_conns(struct nfsd4_session *s)
703 struct nfs4_client *clp = s->se_client;
704 struct nfsd4_conn *c;
706 spin_lock(&clp->cl_lock);
707 while (!list_empty(&s->se_conns)) {
708 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
709 list_del_init(&c->cn_persession);
710 spin_unlock(&clp->cl_lock);
712 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
715 spin_lock(&clp->cl_lock);
717 spin_unlock(&clp->cl_lock);
720 void free_session(struct kref *kref)
722 struct nfsd4_session *ses;
725 ses = container_of(kref, struct nfsd4_session, se_ref);
726 nfsd4_del_conns(ses);
727 spin_lock(&nfsd_drc_lock);
728 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
729 nfsd_drc_mem_used -= mem;
730 spin_unlock(&nfsd_drc_lock);
731 free_session_slots(ses);
735 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
737 struct nfsd4_session *new;
738 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
739 int numslots, slotsize;
744 * Note decreasing slot size below client's request may
745 * make it difficult for client to function correctly, whereas
746 * decreasing the number of slots will (just?) affect
747 * performance. When short on memory we therefore prefer to
748 * decrease number of slots instead of their size.
750 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
751 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
755 new = alloc_session(slotsize, numslots);
757 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
760 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
762 new->se_client = clp;
765 INIT_LIST_HEAD(&new->se_conns);
767 new->se_cb_seq_nr = 1;
768 new->se_flags = cses->flags;
769 new->se_cb_prog = cses->callback_prog;
770 kref_init(&new->se_ref);
771 idx = hash_sessionid(&new->se_sessionid);
772 spin_lock(&client_lock);
773 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
774 list_add(&new->se_perclnt, &clp->cl_sessions);
775 spin_unlock(&client_lock);
777 status = nfsd4_new_conn(rqstp, new);
778 /* whoops: benny points out, status is ignored! (err, or bogus) */
780 free_session(&new->se_ref);
783 if (!clp->cl_cb_session && (cses->flags & SESSION4_BACK_CHAN)) {
784 struct sockaddr *sa = svc_addr(rqstp);
786 clp->cl_cb_session = new;
787 clp->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
788 svc_xprt_get(rqstp->rq_xprt);
789 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
790 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
791 nfsd4_probe_callback(clp);
796 /* caller must hold client_lock */
797 static struct nfsd4_session *
798 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
800 struct nfsd4_session *elem;
803 dump_sessionid(__func__, sessionid);
804 idx = hash_sessionid(sessionid);
805 /* Search in the appropriate list */
806 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
807 if (!memcmp(elem->se_sessionid.data, sessionid->data,
808 NFS4_MAX_SESSIONID_LEN)) {
813 dprintk("%s: session not found\n", __func__);
817 /* caller must hold client_lock */
819 unhash_session(struct nfsd4_session *ses)
821 list_del(&ses->se_hash);
822 list_del(&ses->se_perclnt);
825 /* must be called under the client_lock */
827 renew_client_locked(struct nfs4_client *clp)
829 if (is_client_expired(clp)) {
830 dprintk("%s: client (clientid %08x/%08x) already expired\n",
832 clp->cl_clientid.cl_boot,
833 clp->cl_clientid.cl_id);
838 * Move client to the end to the LRU list.
840 dprintk("renewing client (clientid %08x/%08x)\n",
841 clp->cl_clientid.cl_boot,
842 clp->cl_clientid.cl_id);
843 list_move_tail(&clp->cl_lru, &client_lru);
844 clp->cl_time = get_seconds();
848 renew_client(struct nfs4_client *clp)
850 spin_lock(&client_lock);
851 renew_client_locked(clp);
852 spin_unlock(&client_lock);
855 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
857 STALE_CLIENTID(clientid_t *clid)
859 if (clid->cl_boot == boot_time)
861 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
862 clid->cl_boot, clid->cl_id, boot_time);
867 * XXX Should we use a slab cache ?
868 * This type of memory management is somewhat inefficient, but we use it
869 * anyway since SETCLIENTID is not a common operation.
871 static struct nfs4_client *alloc_client(struct xdr_netobj name)
873 struct nfs4_client *clp;
875 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
878 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
879 if (clp->cl_name.data == NULL) {
883 memcpy(clp->cl_name.data, name.data, name.len);
884 clp->cl_name.len = name.len;
889 free_client(struct nfs4_client *clp)
891 while (!list_empty(&clp->cl_sessions)) {
892 struct nfsd4_session *ses;
893 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
895 list_del(&ses->se_perclnt);
896 nfsd4_put_session(ses);
898 if (clp->cl_cred.cr_group_info)
899 put_group_info(clp->cl_cred.cr_group_info);
900 kfree(clp->cl_principal);
901 kfree(clp->cl_name.data);
906 release_session_client(struct nfsd4_session *session)
908 struct nfs4_client *clp = session->se_client;
910 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
912 if (is_client_expired(clp)) {
914 session->se_client = NULL;
916 renew_client_locked(clp);
917 spin_unlock(&client_lock);
920 /* must be called under the client_lock */
922 unhash_client_locked(struct nfs4_client *clp)
924 struct nfsd4_session *ses;
926 mark_client_expired(clp);
927 list_del(&clp->cl_lru);
928 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
929 list_del_init(&ses->se_hash);
933 expire_client(struct nfs4_client *clp)
935 struct nfs4_stateowner *sop;
936 struct nfs4_delegation *dp;
937 struct list_head reaplist;
939 INIT_LIST_HEAD(&reaplist);
940 spin_lock(&recall_lock);
941 while (!list_empty(&clp->cl_delegations)) {
942 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
943 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
945 list_del_init(&dp->dl_perclnt);
946 list_move(&dp->dl_recall_lru, &reaplist);
948 spin_unlock(&recall_lock);
949 while (!list_empty(&reaplist)) {
950 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
951 list_del_init(&dp->dl_recall_lru);
952 unhash_delegation(dp);
954 while (!list_empty(&clp->cl_openowners)) {
955 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
956 release_openowner(sop);
958 nfsd4_shutdown_callback(clp);
959 if (clp->cl_cb_conn.cb_xprt)
960 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
961 list_del(&clp->cl_idhash);
962 list_del(&clp->cl_strhash);
963 spin_lock(&client_lock);
964 unhash_client_locked(clp);
965 if (atomic_read(&clp->cl_refcount) == 0)
967 spin_unlock(&client_lock);
970 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
972 memcpy(target->cl_verifier.data, source->data,
973 sizeof(target->cl_verifier.data));
976 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
978 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
979 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
982 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
984 target->cr_uid = source->cr_uid;
985 target->cr_gid = source->cr_gid;
986 target->cr_group_info = source->cr_group_info;
987 get_group_info(target->cr_group_info);
990 static int same_name(const char *n1, const char *n2)
992 return 0 == memcmp(n1, n2, HEXDIR_LEN);
996 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
998 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1002 same_clid(clientid_t *cl1, clientid_t *cl2)
1004 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1007 /* XXX what about NGROUP */
1009 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1011 return cr1->cr_uid == cr2->cr_uid;
1014 static void gen_clid(struct nfs4_client *clp)
1016 static u32 current_clientid = 1;
1018 clp->cl_clientid.cl_boot = boot_time;
1019 clp->cl_clientid.cl_id = current_clientid++;
1022 static void gen_confirm(struct nfs4_client *clp)
1027 p = (u32 *)clp->cl_confirm.data;
1028 *p++ = get_seconds();
1032 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1033 struct svc_rqst *rqstp, nfs4_verifier *verf)
1035 struct nfs4_client *clp;
1036 struct sockaddr *sa = svc_addr(rqstp);
1039 clp = alloc_client(name);
1043 INIT_LIST_HEAD(&clp->cl_sessions);
1045 princ = svc_gss_principal(rqstp);
1047 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1048 if (clp->cl_principal == NULL) {
1054 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1055 atomic_set(&clp->cl_refcount, 0);
1056 atomic_set(&clp->cl_cb_set, 0);
1057 INIT_LIST_HEAD(&clp->cl_idhash);
1058 INIT_LIST_HEAD(&clp->cl_strhash);
1059 INIT_LIST_HEAD(&clp->cl_openowners);
1060 INIT_LIST_HEAD(&clp->cl_delegations);
1061 INIT_LIST_HEAD(&clp->cl_lru);
1062 spin_lock_init(&clp->cl_lock);
1063 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1064 clp->cl_time = get_seconds();
1065 clear_bit(0, &clp->cl_cb_slot_busy);
1066 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1067 copy_verf(clp, verf);
1068 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1069 clp->cl_flavor = rqstp->rq_flavor;
1070 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1072 clp->cl_cb_session = NULL;
1076 static int check_name(struct xdr_netobj name)
1080 if (name.len > NFS4_OPAQUE_LIMIT) {
1081 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1088 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1090 unsigned int idhashval;
1092 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1093 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1094 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1099 move_to_confirmed(struct nfs4_client *clp)
1101 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1102 unsigned int strhashval;
1104 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1105 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1106 strhashval = clientstr_hashval(clp->cl_recdir);
1107 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1111 static struct nfs4_client *
1112 find_confirmed_client(clientid_t *clid)
1114 struct nfs4_client *clp;
1115 unsigned int idhashval = clientid_hashval(clid->cl_id);
1117 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1118 if (same_clid(&clp->cl_clientid, clid))
1124 static struct nfs4_client *
1125 find_unconfirmed_client(clientid_t *clid)
1127 struct nfs4_client *clp;
1128 unsigned int idhashval = clientid_hashval(clid->cl_id);
1130 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1131 if (same_clid(&clp->cl_clientid, clid))
1137 static bool clp_used_exchangeid(struct nfs4_client *clp)
1139 return clp->cl_exchange_flags != 0;
1142 static struct nfs4_client *
1143 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1145 struct nfs4_client *clp;
1147 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1148 if (same_name(clp->cl_recdir, dname))
1154 static struct nfs4_client *
1155 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1157 struct nfs4_client *clp;
1159 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1160 if (same_name(clp->cl_recdir, dname))
1167 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1169 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1170 unsigned short expected_family;
1172 /* Currently, we only support tcp and tcp6 for the callback channel */
1173 if (se->se_callback_netid_len == 3 &&
1174 !memcmp(se->se_callback_netid_val, "tcp", 3))
1175 expected_family = AF_INET;
1176 else if (se->se_callback_netid_len == 4 &&
1177 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1178 expected_family = AF_INET6;
1182 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1183 se->se_callback_addr_len,
1184 (struct sockaddr *)&conn->cb_addr,
1185 sizeof(conn->cb_addr));
1187 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1190 if (conn->cb_addr.ss_family == AF_INET6)
1191 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1193 conn->cb_prog = se->se_callback_prog;
1194 conn->cb_ident = se->se_callback_ident;
1197 conn->cb_addr.ss_family = AF_UNSPEC;
1198 conn->cb_addrlen = 0;
1199 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1200 "will not receive delegations\n",
1201 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1207 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1210 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1212 struct nfsd4_slot *slot = resp->cstate.slot;
1215 dprintk("--> %s slot %p\n", __func__, slot);
1217 slot->sl_opcnt = resp->opcnt;
1218 slot->sl_status = resp->cstate.status;
1220 if (nfsd4_not_cached(resp)) {
1221 slot->sl_datalen = 0;
1224 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1225 base = (char *)resp->cstate.datap -
1226 (char *)resp->xbuf->head[0].iov_base;
1227 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1229 WARN("%s: sessions DRC could not cache compound\n", __func__);
1234 * Encode the replay sequence operation from the slot values.
1235 * If cachethis is FALSE encode the uncached rep error on the next
1236 * operation which sets resp->p and increments resp->opcnt for
1237 * nfs4svc_encode_compoundres.
1241 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1242 struct nfsd4_compoundres *resp)
1244 struct nfsd4_op *op;
1245 struct nfsd4_slot *slot = resp->cstate.slot;
1247 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1248 resp->opcnt, resp->cstate.slot->sl_cachethis);
1250 /* Encode the replayed sequence operation */
1251 op = &args->ops[resp->opcnt - 1];
1252 nfsd4_encode_operation(resp, op);
1254 /* Return nfserr_retry_uncached_rep in next operation. */
1255 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1256 op = &args->ops[resp->opcnt++];
1257 op->status = nfserr_retry_uncached_rep;
1258 nfsd4_encode_operation(resp, op);
1264 * The sequence operation is not cached because we can use the slot and
1268 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1269 struct nfsd4_sequence *seq)
1271 struct nfsd4_slot *slot = resp->cstate.slot;
1274 dprintk("--> %s slot %p\n", __func__, slot);
1276 /* Either returns 0 or nfserr_retry_uncached */
1277 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1278 if (status == nfserr_retry_uncached_rep)
1281 /* The sequence operation has been encoded, cstate->datap set. */
1282 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1284 resp->opcnt = slot->sl_opcnt;
1285 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1286 status = slot->sl_status;
1292 * Set the exchange_id flags returned by the server.
1295 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1297 /* pNFS is not supported */
1298 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1300 /* Referrals are supported, Migration is not. */
1301 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1303 /* set the wire flags to return to client. */
1304 clid->flags = new->cl_exchange_flags;
1308 nfsd4_exchange_id(struct svc_rqst *rqstp,
1309 struct nfsd4_compound_state *cstate,
1310 struct nfsd4_exchange_id *exid)
1312 struct nfs4_client *unconf, *conf, *new;
1314 unsigned int strhashval;
1315 char dname[HEXDIR_LEN];
1316 char addr_str[INET6_ADDRSTRLEN];
1317 nfs4_verifier verf = exid->verifier;
1318 struct sockaddr *sa = svc_addr(rqstp);
1320 rpc_ntop(sa, addr_str, sizeof(addr_str));
1321 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1322 "ip_addr=%s flags %x, spa_how %d\n",
1323 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1324 addr_str, exid->flags, exid->spa_how);
1326 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1327 return nfserr_inval;
1329 /* Currently only support SP4_NONE */
1330 switch (exid->spa_how) {
1334 return nfserr_serverfault;
1336 BUG(); /* checked by xdr code */
1338 return nfserr_serverfault; /* no excuse :-/ */
1341 status = nfs4_make_rec_clidname(dname, &exid->clname);
1346 strhashval = clientstr_hashval(dname);
1351 conf = find_confirmed_client_by_str(dname, strhashval);
1353 if (!clp_used_exchangeid(conf)) {
1354 status = nfserr_clid_inuse; /* XXX: ? */
1357 if (!same_verf(&verf, &conf->cl_verifier)) {
1358 /* 18.35.4 case 8 */
1359 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1360 status = nfserr_not_same;
1363 /* Client reboot: destroy old state */
1364 expire_client(conf);
1367 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1368 /* 18.35.4 case 9 */
1369 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1370 status = nfserr_perm;
1373 expire_client(conf);
1377 * Set bit when the owner id and verifier map to an already
1378 * confirmed client id (18.35.3).
1380 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1383 * Falling into 18.35.4 case 2, possible router replay.
1384 * Leave confirmed record intact and return same result.
1386 copy_verf(conf, &verf);
1391 /* 18.35.4 case 7 */
1392 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1393 status = nfserr_noent;
1397 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1400 * Possible retry or client restart. Per 18.35.4 case 4,
1401 * a new unconfirmed record should be generated regardless
1402 * of whether any properties have changed.
1404 expire_client(unconf);
1409 new = create_client(exid->clname, dname, rqstp, &verf);
1411 status = nfserr_jukebox;
1416 add_to_unconfirmed(new, strhashval);
1418 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1419 exid->clientid.cl_id = new->cl_clientid.cl_id;
1422 nfsd4_set_ex_flags(new, exid);
1424 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1425 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1429 nfs4_unlock_state();
1431 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1436 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1438 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1441 /* The slot is in use, and no response has been sent. */
1443 if (seqid == slot_seqid)
1444 return nfserr_jukebox;
1446 return nfserr_seq_misordered;
1449 if (likely(seqid == slot_seqid + 1))
1452 if (seqid == slot_seqid)
1453 return nfserr_replay_cache;
1455 if (seqid == 1 && (slot_seqid + 1) == 0)
1457 /* Misordered replay or misordered new request */
1458 return nfserr_seq_misordered;
1462 * Cache the create session result into the create session single DRC
1463 * slot cache by saving the xdr structure. sl_seqid has been set.
1464 * Do this for solo or embedded create session operations.
1467 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1468 struct nfsd4_clid_slot *slot, int nfserr)
1470 slot->sl_status = nfserr;
1471 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1475 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1476 struct nfsd4_clid_slot *slot)
1478 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1479 return slot->sl_status;
1483 nfsd4_create_session(struct svc_rqst *rqstp,
1484 struct nfsd4_compound_state *cstate,
1485 struct nfsd4_create_session *cr_ses)
1487 struct sockaddr *sa = svc_addr(rqstp);
1488 struct nfs4_client *conf, *unconf;
1489 struct nfsd4_session *new;
1490 struct nfsd4_clid_slot *cs_slot = NULL;
1491 bool confirm_me = false;
1495 unconf = find_unconfirmed_client(&cr_ses->clientid);
1496 conf = find_confirmed_client(&cr_ses->clientid);
1499 cs_slot = &conf->cl_cs_slot;
1500 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1501 if (status == nfserr_replay_cache) {
1502 dprintk("Got a create_session replay! seqid= %d\n",
1504 /* Return the cached reply status */
1505 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1507 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1508 status = nfserr_seq_misordered;
1509 dprintk("Sequence misordered!\n");
1510 dprintk("Expected seqid= %d but got seqid= %d\n",
1511 cs_slot->sl_seqid, cr_ses->seqid);
1514 } else if (unconf) {
1515 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1516 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1517 status = nfserr_clid_inuse;
1521 cs_slot = &unconf->cl_cs_slot;
1522 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1524 /* an unconfirmed replay returns misordered */
1525 status = nfserr_seq_misordered;
1532 status = nfserr_stale_clientid;
1537 * XXX: we should probably set this at creation time, and check
1538 * for consistent minorversion use throughout:
1540 conf->cl_minorversion = 1;
1542 * We do not support RDMA or persistent sessions
1544 cr_ses->flags &= ~SESSION4_PERSIST;
1545 cr_ses->flags &= ~SESSION4_RDMA;
1547 status = nfserr_jukebox;
1548 new = alloc_init_session(rqstp, conf, cr_ses);
1552 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1553 NFS4_MAX_SESSIONID_LEN);
1554 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1555 sizeof(struct nfsd4_channel_attrs));
1556 cs_slot->sl_seqid++;
1557 cr_ses->seqid = cs_slot->sl_seqid;
1559 /* cache solo and embedded create sessions under the state lock */
1560 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1562 move_to_confirmed(conf);
1564 nfs4_unlock_state();
1565 dprintk("%s returns %d\n", __func__, ntohl(status));
1569 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1571 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1572 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1574 return argp->opcnt == resp->opcnt;
1577 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1581 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1585 nfsd4_destroy_session(struct svc_rqst *r,
1586 struct nfsd4_compound_state *cstate,
1587 struct nfsd4_destroy_session *sessionid)
1589 struct nfsd4_session *ses;
1590 u32 status = nfserr_badsession;
1593 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1594 * - Should we return nfserr_back_chan_busy if waiting for
1595 * callbacks on to-be-destroyed session?
1596 * - Do we need to clear any callback info from previous session?
1599 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1600 if (!nfsd4_last_compound_op(r))
1601 return nfserr_not_only_op;
1603 dump_sessionid(__func__, &sessionid->sessionid);
1604 spin_lock(&client_lock);
1605 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1607 spin_unlock(&client_lock);
1611 unhash_session(ses);
1612 spin_unlock(&client_lock);
1615 /* wait for callbacks */
1616 nfsd4_shutdown_callback(ses->se_client);
1617 nfs4_unlock_state();
1619 nfsd4_del_conns(ses);
1621 nfsd4_put_session(ses);
1624 dprintk("%s returns %d\n", __func__, ntohl(status));
1628 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1630 struct nfsd4_conn *c;
1632 list_for_each_entry(c, &s->se_conns, cn_persession) {
1633 if (c->cn_xprt == xpt) {
1640 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1642 struct nfs4_client *clp = ses->se_client;
1643 struct nfsd4_conn *c;
1646 spin_lock(&clp->cl_lock);
1647 c = __nfsd4_find_conn(new->cn_xprt, ses);
1649 spin_unlock(&clp->cl_lock);
1653 __nfsd4_hash_conn(new, ses);
1654 spin_unlock(&clp->cl_lock);
1655 ret = nfsd4_register_conn(new);
1657 /* oops; xprt is already down: */
1658 nfsd4_conn_lost(&new->cn_xpt_user);
1663 nfsd4_sequence(struct svc_rqst *rqstp,
1664 struct nfsd4_compound_state *cstate,
1665 struct nfsd4_sequence *seq)
1667 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1668 struct nfsd4_session *session;
1669 struct nfsd4_slot *slot;
1670 struct nfsd4_conn *conn;
1673 if (resp->opcnt != 1)
1674 return nfserr_sequence_pos;
1677 * Will be either used or freed by nfsd4_sequence_check_conn
1680 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1682 return nfserr_jukebox;
1684 spin_lock(&client_lock);
1685 status = nfserr_badsession;
1686 session = find_in_sessionid_hashtbl(&seq->sessionid);
1690 status = nfserr_badslot;
1691 if (seq->slotid >= session->se_fchannel.maxreqs)
1694 slot = session->se_slots[seq->slotid];
1695 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1697 /* We do not negotiate the number of slots yet, so set the
1698 * maxslots to the session maxreqs which is used to encode
1699 * sr_highest_slotid and the sr_target_slot id to maxslots */
1700 seq->maxslots = session->se_fchannel.maxreqs;
1702 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1703 if (status == nfserr_replay_cache) {
1704 cstate->slot = slot;
1705 cstate->session = session;
1706 /* Return the cached reply status and set cstate->status
1707 * for nfsd4_proc_compound processing */
1708 status = nfsd4_replay_cache_entry(resp, seq);
1709 cstate->status = nfserr_replay_cache;
1715 nfsd4_sequence_check_conn(conn, session);
1718 /* Success! bump slot seqid */
1719 slot->sl_inuse = true;
1720 slot->sl_seqid = seq->seqid;
1721 slot->sl_cachethis = seq->cachethis;
1723 cstate->slot = slot;
1724 cstate->session = session;
1727 /* Hold a session reference until done processing the compound. */
1728 if (cstate->session) {
1729 nfsd4_get_session(cstate->session);
1730 atomic_inc(&session->se_client->cl_refcount);
1733 spin_unlock(&client_lock);
1734 dprintk("%s: return %d\n", __func__, ntohl(status));
1739 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1741 if (rc->rca_one_fs) {
1742 if (!cstate->current_fh.fh_dentry)
1743 return nfserr_nofilehandle;
1745 * We don't take advantage of the rca_one_fs case.
1746 * That's OK, it's optional, we can safely ignore it.
1751 if (is_client_expired(cstate->session->se_client)) {
1752 nfs4_unlock_state();
1754 * The following error isn't really legal.
1755 * But we only get here if the client just explicitly
1756 * destroyed the client. Surely it no longer cares what
1757 * error it gets back on an operation for the dead
1760 return nfserr_stale_clientid;
1762 nfsd4_create_clid_dir(cstate->session->se_client);
1763 nfs4_unlock_state();
1768 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1769 struct nfsd4_setclientid *setclid)
1771 struct sockaddr *sa = svc_addr(rqstp);
1772 struct xdr_netobj clname = {
1773 .len = setclid->se_namelen,
1774 .data = setclid->se_name,
1776 nfs4_verifier clverifier = setclid->se_verf;
1777 unsigned int strhashval;
1778 struct nfs4_client *conf, *unconf, *new;
1780 char dname[HEXDIR_LEN];
1782 if (!check_name(clname))
1783 return nfserr_inval;
1785 status = nfs4_make_rec_clidname(dname, &clname);
1790 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1791 * We get here on a DRC miss.
1794 strhashval = clientstr_hashval(dname);
1797 conf = find_confirmed_client_by_str(dname, strhashval);
1799 /* RFC 3530 14.2.33 CASE 0: */
1800 status = nfserr_clid_inuse;
1801 if (clp_used_exchangeid(conf))
1803 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1804 char addr_str[INET6_ADDRSTRLEN];
1805 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1807 dprintk("NFSD: setclientid: string in use by client "
1808 "at %s\n", addr_str);
1813 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1814 * has a description of SETCLIENTID request processing consisting
1815 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1817 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1818 status = nfserr_resource;
1821 * RFC 3530 14.2.33 CASE 4:
1822 * placed first, because it is the normal case
1825 expire_client(unconf);
1826 new = create_client(clname, dname, rqstp, &clverifier);
1830 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1832 * RFC 3530 14.2.33 CASE 1:
1833 * probable callback update
1836 /* Note this is removing unconfirmed {*x***},
1837 * which is stronger than RFC recommended {vxc**}.
1838 * This has the advantage that there is at most
1839 * one {*x***} in either list at any time.
1841 expire_client(unconf);
1843 new = create_client(clname, dname, rqstp, &clverifier);
1846 copy_clid(new, conf);
1847 } else if (!unconf) {
1849 * RFC 3530 14.2.33 CASE 2:
1850 * probable client reboot; state will be removed if
1853 new = create_client(clname, dname, rqstp, &clverifier);
1859 * RFC 3530 14.2.33 CASE 3:
1860 * probable client reboot; state will be removed if
1863 expire_client(unconf);
1864 new = create_client(clname, dname, rqstp, &clverifier);
1870 * XXX: we should probably set this at creation time, and check
1871 * for consistent minorversion use throughout:
1873 new->cl_minorversion = 0;
1874 gen_callback(new, setclid, rpc_get_scope_id(sa));
1875 add_to_unconfirmed(new, strhashval);
1876 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1877 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1878 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1881 nfs4_unlock_state();
1887 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1888 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1889 * bullets, labeled as CASE1 - CASE4 below.
1892 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1893 struct nfsd4_compound_state *cstate,
1894 struct nfsd4_setclientid_confirm *setclientid_confirm)
1896 struct sockaddr *sa = svc_addr(rqstp);
1897 struct nfs4_client *conf, *unconf;
1898 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1899 clientid_t * clid = &setclientid_confirm->sc_clientid;
1902 if (STALE_CLIENTID(clid))
1903 return nfserr_stale_clientid;
1905 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1906 * We get here on a DRC miss.
1911 conf = find_confirmed_client(clid);
1912 unconf = find_unconfirmed_client(clid);
1914 status = nfserr_clid_inuse;
1915 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1917 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1921 * section 14.2.34 of RFC 3530 has a description of
1922 * SETCLIENTID_CONFIRM request processing consisting
1923 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1925 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1927 * RFC 3530 14.2.34 CASE 1:
1930 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1931 status = nfserr_clid_inuse;
1933 atomic_set(&conf->cl_cb_set, 0);
1934 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
1935 nfsd4_probe_callback(conf);
1936 expire_client(unconf);
1940 } else if (conf && !unconf) {
1942 * RFC 3530 14.2.34 CASE 2:
1943 * probable retransmitted request; play it safe and
1946 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1947 status = nfserr_clid_inuse;
1950 } else if (!conf && unconf
1951 && same_verf(&unconf->cl_confirm, &confirm)) {
1953 * RFC 3530 14.2.34 CASE 3:
1954 * Normal case; new or rebooted client:
1956 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1957 status = nfserr_clid_inuse;
1960 clientstr_hashval(unconf->cl_recdir);
1961 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1964 nfsd4_remove_clid_dir(conf);
1965 expire_client(conf);
1967 move_to_confirmed(unconf);
1969 nfsd4_probe_callback(conf);
1972 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1973 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1976 * RFC 3530 14.2.34 CASE 4:
1977 * Client probably hasn't noticed that we rebooted yet.
1979 status = nfserr_stale_clientid;
1981 /* check that we have hit one of the cases...*/
1982 status = nfserr_clid_inuse;
1985 nfs4_unlock_state();
1989 /* OPEN Share state helper functions */
1990 static inline struct nfs4_file *
1991 alloc_init_file(struct inode *ino)
1993 struct nfs4_file *fp;
1994 unsigned int hashval = file_hashval(ino);
1996 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1998 atomic_set(&fp->fi_ref, 1);
1999 INIT_LIST_HEAD(&fp->fi_hash);
2000 INIT_LIST_HEAD(&fp->fi_stateids);
2001 INIT_LIST_HEAD(&fp->fi_delegations);
2002 fp->fi_inode = igrab(ino);
2003 fp->fi_id = current_fileid++;
2004 fp->fi_had_conflict = false;
2005 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2006 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2007 spin_lock(&recall_lock);
2008 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2009 spin_unlock(&recall_lock);
2016 nfsd4_free_slab(struct kmem_cache **slab)
2020 kmem_cache_destroy(*slab);
2025 nfsd4_free_slabs(void)
2027 nfsd4_free_slab(&stateowner_slab);
2028 nfsd4_free_slab(&file_slab);
2029 nfsd4_free_slab(&stateid_slab);
2030 nfsd4_free_slab(&deleg_slab);
2034 nfsd4_init_slabs(void)
2036 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2037 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2038 if (stateowner_slab == NULL)
2040 file_slab = kmem_cache_create("nfsd4_files",
2041 sizeof(struct nfs4_file), 0, 0, NULL);
2042 if (file_slab == NULL)
2044 stateid_slab = kmem_cache_create("nfsd4_stateids",
2045 sizeof(struct nfs4_stateid), 0, 0, NULL);
2046 if (stateid_slab == NULL)
2048 deleg_slab = kmem_cache_create("nfsd4_delegations",
2049 sizeof(struct nfs4_delegation), 0, 0, NULL);
2050 if (deleg_slab == NULL)
2055 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2060 nfs4_free_stateowner(struct kref *kref)
2062 struct nfs4_stateowner *sop =
2063 container_of(kref, struct nfs4_stateowner, so_ref);
2064 kfree(sop->so_owner.data);
2065 kmem_cache_free(stateowner_slab, sop);
2068 static inline struct nfs4_stateowner *
2069 alloc_stateowner(struct xdr_netobj *owner)
2071 struct nfs4_stateowner *sop;
2073 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2074 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2075 memcpy(sop->so_owner.data, owner->data, owner->len);
2076 sop->so_owner.len = owner->len;
2077 kref_init(&sop->so_ref);
2080 kmem_cache_free(stateowner_slab, sop);
2085 static struct nfs4_stateowner *
2086 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2087 struct nfs4_stateowner *sop;
2088 struct nfs4_replay *rp;
2089 unsigned int idhashval;
2091 if (!(sop = alloc_stateowner(&open->op_owner)))
2093 idhashval = ownerid_hashval(current_ownerid);
2094 INIT_LIST_HEAD(&sop->so_idhash);
2095 INIT_LIST_HEAD(&sop->so_strhash);
2096 INIT_LIST_HEAD(&sop->so_perclient);
2097 INIT_LIST_HEAD(&sop->so_stateids);
2098 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2099 INIT_LIST_HEAD(&sop->so_close_lru);
2101 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2102 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2103 list_add(&sop->so_perclient, &clp->cl_openowners);
2104 sop->so_is_open_owner = 1;
2105 sop->so_id = current_ownerid++;
2106 sop->so_client = clp;
2107 sop->so_seqid = open->op_seqid;
2108 sop->so_confirmed = 0;
2109 rp = &sop->so_replay;
2110 rp->rp_status = nfserr_serverfault;
2112 rp->rp_buf = rp->rp_ibuf;
2117 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2118 struct nfs4_stateowner *sop = open->op_stateowner;
2119 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2121 INIT_LIST_HEAD(&stp->st_hash);
2122 INIT_LIST_HEAD(&stp->st_perstateowner);
2123 INIT_LIST_HEAD(&stp->st_lockowners);
2124 INIT_LIST_HEAD(&stp->st_perfile);
2125 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2126 list_add(&stp->st_perstateowner, &sop->so_stateids);
2127 list_add(&stp->st_perfile, &fp->fi_stateids);
2128 stp->st_stateowner = sop;
2131 stp->st_stateid.si_boot = boot_time;
2132 stp->st_stateid.si_stateownerid = sop->so_id;
2133 stp->st_stateid.si_fileid = fp->fi_id;
2134 stp->st_stateid.si_generation = 0;
2135 stp->st_access_bmap = 0;
2136 stp->st_deny_bmap = 0;
2137 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2138 &stp->st_access_bmap);
2139 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2140 stp->st_openstp = NULL;
2144 move_to_close_lru(struct nfs4_stateowner *sop)
2146 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2148 list_move_tail(&sop->so_close_lru, &close_lru);
2149 sop->so_time = get_seconds();
2153 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2156 return (sop->so_owner.len == owner->len) &&
2157 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2158 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2161 static struct nfs4_stateowner *
2162 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2164 struct nfs4_stateowner *so = NULL;
2166 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2167 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2173 /* search file_hashtbl[] for file */
2174 static struct nfs4_file *
2175 find_file(struct inode *ino)
2177 unsigned int hashval = file_hashval(ino);
2178 struct nfs4_file *fp;
2180 spin_lock(&recall_lock);
2181 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2182 if (fp->fi_inode == ino) {
2184 spin_unlock(&recall_lock);
2188 spin_unlock(&recall_lock);
2192 static inline int access_valid(u32 x, u32 minorversion)
2194 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2196 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2198 x &= ~NFS4_SHARE_ACCESS_MASK;
2199 if (minorversion && x) {
2200 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2202 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2204 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2211 static inline int deny_valid(u32 x)
2213 /* Note: unlike access bits, deny bits may be zero. */
2214 return x <= NFS4_SHARE_DENY_BOTH;
2218 * Called to check deny when READ with all zero stateid or
2219 * WRITE with all zero or all one stateid
2222 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2224 struct inode *ino = current_fh->fh_dentry->d_inode;
2225 struct nfs4_file *fp;
2226 struct nfs4_stateid *stp;
2229 dprintk("NFSD: nfs4_share_conflict\n");
2231 fp = find_file(ino);
2234 ret = nfserr_locked;
2235 /* Search for conflicting share reservations */
2236 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2237 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2238 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2248 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2250 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2251 nfs4_file_put_access(fp, O_WRONLY);
2252 if (share_access & NFS4_SHARE_ACCESS_READ)
2253 nfs4_file_put_access(fp, O_RDONLY);
2257 * Spawn a thread to perform a recall on the delegation represented
2258 * by the lease (file_lock)
2260 * Called from break_lease() with lock_flocks() held.
2261 * Note: we assume break_lease will only call this *once* for any given
2265 void nfsd_break_deleg_cb(struct file_lock *fl)
2267 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2269 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2273 /* We're assuming the state code never drops its reference
2274 * without first removing the lease. Since we're in this lease
2275 * callback (and since the lease code is serialized by the kernel
2276 * lock) we know the server hasn't removed the lease yet, we know
2277 * it's safe to take a reference: */
2278 atomic_inc(&dp->dl_count);
2280 spin_lock(&recall_lock);
2281 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2282 spin_unlock(&recall_lock);
2284 /* only place dl_time is set. protected by lock_flocks*/
2285 dp->dl_time = get_seconds();
2288 * We don't want the locks code to timeout the lease for us;
2289 * we'll remove it ourself if the delegation isn't returned
2292 fl->fl_break_time = 0;
2294 dp->dl_file->fi_had_conflict = true;
2295 nfsd4_cb_recall(dp);
2299 * The file_lock is being reapd.
2301 * Called by locks_free_lock() with lock_flocks() held.
2304 void nfsd_release_deleg_cb(struct file_lock *fl)
2306 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2308 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2310 if (!(fl->fl_flags & FL_LEASE) || !dp)
2312 dp->dl_flock = NULL;
2316 * Called from setlease() with lock_flocks() held
2319 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2321 struct nfs4_delegation *onlistd =
2322 (struct nfs4_delegation *)onlist->fl_owner;
2323 struct nfs4_delegation *tryd =
2324 (struct nfs4_delegation *)try->fl_owner;
2326 if (onlist->fl_lmops != try->fl_lmops)
2329 return onlistd->dl_client == tryd->dl_client;
2334 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2337 return lease_modify(onlist, arg);
2342 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2343 .fl_break = nfsd_break_deleg_cb,
2344 .fl_release_private = nfsd_release_deleg_cb,
2345 .fl_mylease = nfsd_same_client_deleg_cb,
2346 .fl_change = nfsd_change_deleg_cb,
2351 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2352 struct nfsd4_open *open)
2354 clientid_t *clientid = &open->op_clientid;
2355 struct nfs4_client *clp = NULL;
2356 unsigned int strhashval;
2357 struct nfs4_stateowner *sop = NULL;
2359 if (!check_name(open->op_owner))
2360 return nfserr_inval;
2362 if (STALE_CLIENTID(&open->op_clientid))
2363 return nfserr_stale_clientid;
2365 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2366 sop = find_openstateowner_str(strhashval, open);
2367 open->op_stateowner = sop;
2369 /* Make sure the client's lease hasn't expired. */
2370 clp = find_confirmed_client(clientid);
2372 return nfserr_expired;
2375 /* When sessions are used, skip open sequenceid processing */
2376 if (nfsd4_has_session(cstate))
2378 if (!sop->so_confirmed) {
2379 /* Replace unconfirmed owners without checking for replay. */
2380 clp = sop->so_client;
2381 release_openowner(sop);
2382 open->op_stateowner = NULL;
2385 if (open->op_seqid == sop->so_seqid - 1) {
2386 if (sop->so_replay.rp_buflen)
2387 return nfserr_replay_me;
2388 /* The original OPEN failed so spectacularly
2389 * that we don't even have replay data saved!
2390 * Therefore, we have no choice but to continue
2391 * processing this OPEN; presumably, we'll
2392 * fail again for the same reason.
2394 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2397 if (open->op_seqid != sop->so_seqid)
2398 return nfserr_bad_seqid;
2400 if (open->op_stateowner == NULL) {
2401 sop = alloc_init_open_stateowner(strhashval, clp, open);
2403 return nfserr_resource;
2404 open->op_stateowner = sop;
2406 list_del_init(&sop->so_close_lru);
2407 renew_client(sop->so_client);
2411 static inline __be32
2412 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2414 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2415 return nfserr_openmode;
2420 static struct nfs4_delegation *
2421 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2423 struct nfs4_delegation *dp;
2425 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2426 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2432 int share_access_to_flags(u32 share_access)
2434 share_access &= ~NFS4_SHARE_WANT_MASK;
2436 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2440 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2441 struct nfs4_delegation **dp)
2444 __be32 status = nfserr_bad_stateid;
2446 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2449 flags = share_access_to_flags(open->op_share_access);
2450 status = nfs4_check_delegmode(*dp, flags);
2454 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2458 open->op_stateowner->so_confirmed = 1;
2463 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2465 struct nfs4_stateid *local;
2466 __be32 status = nfserr_share_denied;
2467 struct nfs4_stateowner *sop = open->op_stateowner;
2469 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2470 /* ignore lock owners */
2471 if (local->st_stateowner->so_is_open_owner == 0)
2473 /* remember if we have seen this open owner */
2474 if (local->st_stateowner == sop)
2476 /* check for conflicting share reservations */
2477 if (!test_share(local, open))
2485 static inline struct nfs4_stateid *
2486 nfs4_alloc_stateid(void)
2488 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2491 static inline int nfs4_access_to_access(u32 nfs4_access)
2495 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2496 flags |= NFSD_MAY_READ;
2497 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2498 flags |= NFSD_MAY_WRITE;
2502 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2503 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2506 int oflag = nfs4_access_to_omode(nfs4_access);
2507 int access = nfs4_access_to_access(nfs4_access);
2509 if (!fp->fi_fds[oflag]) {
2510 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2511 &fp->fi_fds[oflag]);
2512 if (status == nfserr_dropit)
2513 status = nfserr_jukebox;
2517 nfs4_file_get_access(fp, oflag);
2523 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2524 struct nfs4_file *fp, struct svc_fh *cur_fh,
2525 struct nfsd4_open *open)
2527 struct nfs4_stateid *stp;
2530 stp = nfs4_alloc_stateid();
2532 return nfserr_resource;
2534 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2536 kmem_cache_free(stateid_slab, stp);
2543 static inline __be32
2544 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2545 struct nfsd4_open *open)
2547 struct iattr iattr = {
2548 .ia_valid = ATTR_SIZE,
2551 if (!open->op_truncate)
2553 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2554 return nfserr_inval;
2555 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2559 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2561 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2565 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2567 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2571 status = nfsd4_truncate(rqstp, cur_fh, open);
2574 int oflag = nfs4_access_to_omode(new_access);
2575 nfs4_file_put_access(fp, oflag);
2579 /* remember the open */
2580 __set_bit(op_share_access, &stp->st_access_bmap);
2581 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2588 nfs4_set_claim_prev(struct nfsd4_open *open)
2590 open->op_stateowner->so_confirmed = 1;
2591 open->op_stateowner->so_client->cl_firststate = 1;
2595 * Attempt to hand out a delegation.
2598 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2600 struct nfs4_delegation *dp;
2601 struct nfs4_stateowner *sop = stp->st_stateowner;
2602 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2603 struct file_lock *fl;
2604 int status, flag = 0;
2606 flag = NFS4_OPEN_DELEGATE_NONE;
2607 open->op_recall = 0;
2608 switch (open->op_claim_type) {
2609 case NFS4_OPEN_CLAIM_PREVIOUS:
2611 open->op_recall = 1;
2612 flag = open->op_delegate_type;
2613 if (flag == NFS4_OPEN_DELEGATE_NONE)
2616 case NFS4_OPEN_CLAIM_NULL:
2617 /* Let's not give out any delegations till everyone's
2618 * had the chance to reclaim theirs.... */
2619 if (locks_in_grace())
2621 if (!cb_up || !sop->so_confirmed)
2623 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2624 flag = NFS4_OPEN_DELEGATE_WRITE;
2626 flag = NFS4_OPEN_DELEGATE_READ;
2632 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2634 flag = NFS4_OPEN_DELEGATE_NONE;
2638 fl = locks_alloc_lock();
2641 locks_init_lock(fl);
2642 fl->fl_lmops = &nfsd_lease_mng_ops;
2643 fl->fl_flags = FL_LEASE;
2644 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2645 fl->fl_end = OFFSET_MAX;
2646 fl->fl_owner = (fl_owner_t)dp;
2647 fl->fl_file = find_readable_file(stp->st_file);
2648 BUG_ON(!fl->fl_file);
2649 fl->fl_pid = current->tgid;
2652 /* vfs_setlease checks to see if delegation should be handed out.
2653 * the lock_manager callbacks fl_mylease and fl_change are used
2655 if ((status = vfs_setlease(fl->fl_file, fl->fl_type, &fl))) {
2656 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2657 dp->dl_flock = NULL;
2658 locks_free_lock(fl);
2659 unhash_delegation(dp);
2660 flag = NFS4_OPEN_DELEGATE_NONE;
2664 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2666 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2667 STATEID_VAL(&dp->dl_stateid));
2669 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2670 && flag == NFS4_OPEN_DELEGATE_NONE
2671 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2672 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2673 open->op_delegate_type = flag;
2677 * called with nfs4_lock_state() held.
2680 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2682 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2683 struct nfs4_file *fp = NULL;
2684 struct inode *ino = current_fh->fh_dentry->d_inode;
2685 struct nfs4_stateid *stp = NULL;
2686 struct nfs4_delegation *dp = NULL;
2689 status = nfserr_inval;
2690 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2691 || !deny_valid(open->op_share_deny))
2694 * Lookup file; if found, lookup stateid and check open request,
2695 * and check for delegations in the process of being recalled.
2696 * If not found, create the nfs4_file struct
2698 fp = find_file(ino);
2700 if ((status = nfs4_check_open(fp, open, &stp)))
2702 status = nfs4_check_deleg(fp, open, &dp);
2706 status = nfserr_bad_stateid;
2707 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2709 status = nfserr_resource;
2710 fp = alloc_init_file(ino);
2716 * OPEN the file, or upgrade an existing OPEN.
2717 * If truncate fails, the OPEN fails.
2720 /* Stateid was found, this is an OPEN upgrade */
2721 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2724 update_stateid(&stp->st_stateid);
2726 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2729 init_stateid(stp, fp, open);
2730 status = nfsd4_truncate(rqstp, current_fh, open);
2732 release_open_stateid(stp);
2735 if (nfsd4_has_session(&resp->cstate))
2736 update_stateid(&stp->st_stateid);
2738 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2740 if (nfsd4_has_session(&resp->cstate))
2741 open->op_stateowner->so_confirmed = 1;
2744 * Attempt to hand out a delegation. No error return, because the
2745 * OPEN succeeds even if we fail.
2747 nfs4_open_delegation(current_fh, open, stp);
2751 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2752 STATEID_VAL(&stp->st_stateid));
2756 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2757 nfs4_set_claim_prev(open);
2759 * To finish the open response, we just need to set the rflags.
2761 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2762 if (!open->op_stateowner->so_confirmed &&
2763 !nfsd4_has_session(&resp->cstate))
2764 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2770 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2773 struct nfs4_client *clp;
2777 dprintk("process_renew(%08x/%08x): starting\n",
2778 clid->cl_boot, clid->cl_id);
2779 status = nfserr_stale_clientid;
2780 if (STALE_CLIENTID(clid))
2782 clp = find_confirmed_client(clid);
2783 status = nfserr_expired;
2785 /* We assume the client took too long to RENEW. */
2786 dprintk("nfsd4_renew: clientid not found!\n");
2790 status = nfserr_cb_path_down;
2791 if (!list_empty(&clp->cl_delegations)
2792 && !atomic_read(&clp->cl_cb_set))
2796 nfs4_unlock_state();
2800 struct lock_manager nfsd4_manager = {
2804 nfsd4_end_grace(void)
2806 dprintk("NFSD: end of grace period\n");
2807 nfsd4_recdir_purge_old();
2808 locks_end_grace(&nfsd4_manager);
2810 * Now that every NFSv4 client has had the chance to recover and
2811 * to see the (possibly new, possibly shorter) lease time, we
2812 * can safely set the next grace time to the current lease time:
2814 nfsd4_grace = nfsd4_lease;
2818 nfs4_laundromat(void)
2820 struct nfs4_client *clp;
2821 struct nfs4_stateowner *sop;
2822 struct nfs4_delegation *dp;
2823 struct list_head *pos, *next, reaplist;
2824 time_t cutoff = get_seconds() - nfsd4_lease;
2825 time_t t, clientid_val = nfsd4_lease;
2826 time_t u, test_val = nfsd4_lease;
2830 dprintk("NFSD: laundromat service - starting\n");
2831 if (locks_in_grace())
2833 INIT_LIST_HEAD(&reaplist);
2834 spin_lock(&client_lock);
2835 list_for_each_safe(pos, next, &client_lru) {
2836 clp = list_entry(pos, struct nfs4_client, cl_lru);
2837 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2838 t = clp->cl_time - cutoff;
2839 if (clientid_val > t)
2843 if (atomic_read(&clp->cl_refcount)) {
2844 dprintk("NFSD: client in use (clientid %08x)\n",
2845 clp->cl_clientid.cl_id);
2848 unhash_client_locked(clp);
2849 list_add(&clp->cl_lru, &reaplist);
2851 spin_unlock(&client_lock);
2852 list_for_each_safe(pos, next, &reaplist) {
2853 clp = list_entry(pos, struct nfs4_client, cl_lru);
2854 dprintk("NFSD: purging unused client (clientid %08x)\n",
2855 clp->cl_clientid.cl_id);
2856 nfsd4_remove_clid_dir(clp);
2859 spin_lock(&recall_lock);
2860 list_for_each_safe(pos, next, &del_recall_lru) {
2861 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2862 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2863 u = dp->dl_time - cutoff;
2868 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2870 list_move(&dp->dl_recall_lru, &reaplist);
2872 spin_unlock(&recall_lock);
2873 list_for_each_safe(pos, next, &reaplist) {
2874 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2875 list_del_init(&dp->dl_recall_lru);
2876 unhash_delegation(dp);
2878 test_val = nfsd4_lease;
2879 list_for_each_safe(pos, next, &close_lru) {
2880 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2881 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2882 u = sop->so_time - cutoff;
2887 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2889 release_openowner(sop);
2891 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2892 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2893 nfs4_unlock_state();
2894 return clientid_val;
2897 static struct workqueue_struct *laundry_wq;
2898 static void laundromat_main(struct work_struct *);
2899 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2902 laundromat_main(struct work_struct *not_used)
2906 t = nfs4_laundromat();
2907 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2908 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2911 static struct nfs4_stateowner *
2912 search_close_lru(u32 st_id, int flags)
2914 struct nfs4_stateowner *local = NULL;
2916 if (flags & CLOSE_STATE) {
2917 list_for_each_entry(local, &close_lru, so_close_lru) {
2918 if (local->so_id == st_id)
2926 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2928 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2932 STALE_STATEID(stateid_t *stateid)
2934 if (stateid->si_boot == boot_time)
2936 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2937 STATEID_VAL(stateid));
2942 access_permit_read(unsigned long access_bmap)
2944 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2945 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2946 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2950 access_permit_write(unsigned long access_bmap)
2952 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2953 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2957 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2959 __be32 status = nfserr_openmode;
2961 /* For lock stateid's, we test the parent open, not the lock: */
2962 if (stp->st_openstp)
2963 stp = stp->st_openstp;
2964 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2966 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2973 static inline __be32
2974 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2976 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2978 else if (locks_in_grace()) {
2979 /* Answer in remaining cases depends on existance of
2980 * conflicting state; so we must wait out the grace period. */
2981 return nfserr_grace;
2982 } else if (flags & WR_STATE)
2983 return nfs4_share_conflict(current_fh,
2984 NFS4_SHARE_DENY_WRITE);
2985 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2986 return nfs4_share_conflict(current_fh,
2987 NFS4_SHARE_DENY_READ);
2991 * Allow READ/WRITE during grace period on recovered state only for files
2992 * that are not able to provide mandatory locking.
2995 grace_disallows_io(struct inode *inode)
2997 return locks_in_grace() && mandatory_lock(inode);
3000 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3003 * When sessions are used the stateid generation number is ignored
3006 if ((flags & HAS_SESSION) && in->si_generation == 0)
3009 /* If the client sends us a stateid from the future, it's buggy: */
3010 if (in->si_generation > ref->si_generation)
3011 return nfserr_bad_stateid;
3013 * The following, however, can happen. For example, if the
3014 * client sends an open and some IO at the same time, the open
3015 * may bump si_generation while the IO is still in flight.
3016 * Thanks to hard links and renames, the client never knows what
3017 * file an open will affect. So it could avoid that situation
3018 * only by serializing all opens and IO from the same open
3019 * owner. To recover from the old_stateid error, the client
3020 * will just have to retry the IO:
3022 if (in->si_generation < ref->si_generation)
3023 return nfserr_old_stateid;
3028 static int is_delegation_stateid(stateid_t *stateid)
3030 return stateid->si_fileid == 0;
3034 * Checks for stateid operations
3037 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3038 stateid_t *stateid, int flags, struct file **filpp)
3040 struct nfs4_stateid *stp = NULL;
3041 struct nfs4_delegation *dp = NULL;
3042 struct svc_fh *current_fh = &cstate->current_fh;
3043 struct inode *ino = current_fh->fh_dentry->d_inode;
3049 if (grace_disallows_io(ino))
3050 return nfserr_grace;
3052 if (nfsd4_has_session(cstate))
3053 flags |= HAS_SESSION;
3055 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3056 return check_special_stateids(current_fh, stateid, flags);
3058 status = nfserr_stale_stateid;
3059 if (STALE_STATEID(stateid))
3063 * We assume that any stateid that has the current boot time,
3064 * but that we can't find, is expired:
3066 status = nfserr_expired;
3067 if (is_delegation_stateid(stateid)) {
3068 dp = find_delegation_stateid(ino, stateid);
3071 status = check_stateid_generation(stateid, &dp->dl_stateid,
3075 status = nfs4_check_delegmode(dp, flags);
3078 renew_client(dp->dl_client);
3080 *filpp = find_readable_file(dp->dl_file);
3083 } else { /* open or lock stateid */
3084 stp = find_stateid(stateid, flags);
3087 status = nfserr_bad_stateid;
3088 if (nfs4_check_fh(current_fh, stp))
3090 if (!stp->st_stateowner->so_confirmed)
3092 status = check_stateid_generation(stateid, &stp->st_stateid,
3096 status = nfs4_check_openmode(stp, flags);
3099 renew_client(stp->st_stateowner->so_client);
3101 if (flags & RD_STATE)
3102 *filpp = find_readable_file(stp->st_file);
3104 *filpp = find_writeable_file(stp->st_file);
3115 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3116 RD_STATE : WR_STATE;
3120 * Checks for sequence id mutating operations.
3123 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3124 stateid_t *stateid, int flags,
3125 struct nfs4_stateowner **sopp,
3126 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3128 struct nfs4_stateid *stp;
3129 struct nfs4_stateowner *sop;
3130 struct svc_fh *current_fh = &cstate->current_fh;
3133 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3134 seqid, STATEID_VAL(stateid));
3139 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3140 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3141 return nfserr_bad_stateid;
3144 if (STALE_STATEID(stateid))
3145 return nfserr_stale_stateid;
3147 if (nfsd4_has_session(cstate))
3148 flags |= HAS_SESSION;
3151 * We return BAD_STATEID if filehandle doesn't match stateid,
3152 * the confirmed flag is incorrecly set, or the generation
3153 * number is incorrect.
3155 stp = find_stateid(stateid, flags);
3158 * Also, we should make sure this isn't just the result of
3161 sop = search_close_lru(stateid->si_stateownerid, flags);
3162 /* It's not stale; let's assume it's expired: */
3164 return nfserr_expired;
3170 *sopp = sop = stp->st_stateowner;
3173 clientid_t *lockclid = &lock->v.new.clientid;
3174 struct nfs4_client *clp = sop->so_client;
3178 lkflg = setlkflg(lock->lk_type);
3180 if (lock->lk_is_new) {
3181 if (!sop->so_is_open_owner)
3182 return nfserr_bad_stateid;
3183 if (!(flags & HAS_SESSION) &&
3184 !same_clid(&clp->cl_clientid, lockclid))
3185 return nfserr_bad_stateid;
3186 /* stp is the open stateid */
3187 status = nfs4_check_openmode(stp, lkflg);
3191 /* stp is the lock stateid */
3192 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3198 if (nfs4_check_fh(current_fh, stp)) {
3199 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3200 return nfserr_bad_stateid;
3204 * We now validate the seqid and stateid generation numbers.
3205 * For the moment, we ignore the possibility of
3206 * generation number wraparound.
3208 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3211 if (sop->so_confirmed && flags & CONFIRM) {
3212 dprintk("NFSD: preprocess_seqid_op: expected"
3213 " unconfirmed stateowner!\n");
3214 return nfserr_bad_stateid;
3216 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3217 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3218 " confirmed yet!\n");
3219 return nfserr_bad_stateid;
3221 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3224 renew_client(sop->so_client);
3228 if (seqid == sop->so_seqid - 1) {
3229 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3230 /* indicate replay to calling function */
3231 return nfserr_replay_me;
3233 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3234 sop->so_seqid, seqid);
3236 return nfserr_bad_seqid;
3240 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3241 struct nfsd4_open_confirm *oc)
3244 struct nfs4_stateowner *sop;
3245 struct nfs4_stateid *stp;
3247 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3248 (int)cstate->current_fh.fh_dentry->d_name.len,
3249 cstate->current_fh.fh_dentry->d_name.name);
3251 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3257 if ((status = nfs4_preprocess_seqid_op(cstate,
3258 oc->oc_seqid, &oc->oc_req_stateid,
3259 CONFIRM | OPEN_STATE,
3260 &oc->oc_stateowner, &stp, NULL)))
3263 sop = oc->oc_stateowner;
3264 sop->so_confirmed = 1;
3265 update_stateid(&stp->st_stateid);
3266 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3267 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3268 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3270 nfsd4_create_clid_dir(sop->so_client);
3272 if (oc->oc_stateowner) {
3273 nfs4_get_stateowner(oc->oc_stateowner);
3274 cstate->replay_owner = oc->oc_stateowner;
3276 nfs4_unlock_state();
3282 * unset all bits in union bitmap (bmap) that
3283 * do not exist in share (from successful OPEN_DOWNGRADE)
3286 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3289 for (i = 1; i < 4; i++) {
3290 if ((i & access) != i)
3291 __clear_bit(i, bmap);
3296 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3299 for (i = 0; i < 4; i++) {
3300 if ((i & deny) != i)
3301 __clear_bit(i, bmap);
3306 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3307 struct nfsd4_compound_state *cstate,
3308 struct nfsd4_open_downgrade *od)
3311 struct nfs4_stateid *stp;
3312 unsigned int share_access;
3314 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3315 (int)cstate->current_fh.fh_dentry->d_name.len,
3316 cstate->current_fh.fh_dentry->d_name.name);
3318 if (!access_valid(od->od_share_access, cstate->minorversion)
3319 || !deny_valid(od->od_share_deny))
3320 return nfserr_inval;
3323 if ((status = nfs4_preprocess_seqid_op(cstate,
3327 &od->od_stateowner, &stp, NULL)))
3330 status = nfserr_inval;
3331 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3332 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3333 stp->st_access_bmap, od->od_share_access);
3336 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3337 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3338 stp->st_deny_bmap, od->od_share_deny);
3341 set_access(&share_access, stp->st_access_bmap);
3342 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3344 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3345 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3347 update_stateid(&stp->st_stateid);
3348 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3351 if (od->od_stateowner) {
3352 nfs4_get_stateowner(od->od_stateowner);
3353 cstate->replay_owner = od->od_stateowner;
3355 nfs4_unlock_state();
3360 * nfs4_unlock_state() called after encode
3363 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3364 struct nfsd4_close *close)
3367 struct nfs4_stateid *stp;
3369 dprintk("NFSD: nfsd4_close on file %.*s\n",
3370 (int)cstate->current_fh.fh_dentry->d_name.len,
3371 cstate->current_fh.fh_dentry->d_name.name);
3374 /* check close_lru for replay */
3375 if ((status = nfs4_preprocess_seqid_op(cstate,
3378 OPEN_STATE | CLOSE_STATE,
3379 &close->cl_stateowner, &stp, NULL)))
3382 update_stateid(&stp->st_stateid);
3383 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3385 /* release_stateid() calls nfsd_close() if needed */
3386 release_open_stateid(stp);
3388 /* place unused nfs4_stateowners on so_close_lru list to be
3389 * released by the laundromat service after the lease period
3390 * to enable us to handle CLOSE replay
3392 if (list_empty(&close->cl_stateowner->so_stateids))
3393 move_to_close_lru(close->cl_stateowner);
3395 if (close->cl_stateowner) {
3396 nfs4_get_stateowner(close->cl_stateowner);
3397 cstate->replay_owner = close->cl_stateowner;
3399 nfs4_unlock_state();
3404 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3405 struct nfsd4_delegreturn *dr)
3407 struct nfs4_delegation *dp;
3408 stateid_t *stateid = &dr->dr_stateid;
3409 struct inode *inode;
3413 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3415 inode = cstate->current_fh.fh_dentry->d_inode;
3417 if (nfsd4_has_session(cstate))
3418 flags |= HAS_SESSION;
3420 status = nfserr_bad_stateid;
3421 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3423 status = nfserr_stale_stateid;
3424 if (STALE_STATEID(stateid))
3426 status = nfserr_bad_stateid;
3427 if (!is_delegation_stateid(stateid))
3429 status = nfserr_expired;
3430 dp = find_delegation_stateid(inode, stateid);
3433 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3436 renew_client(dp->dl_client);
3438 unhash_delegation(dp);
3440 nfs4_unlock_state();
3447 * Lock owner state (byte-range locks)
3449 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3450 #define LOCK_HASH_BITS 8
3451 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3452 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3455 end_offset(u64 start, u64 len)
3460 return end >= start ? end: NFS4_MAX_UINT64;
3463 /* last octet in a range */
3465 last_byte_offset(u64 start, u64 len)
3471 return end > start ? end - 1: NFS4_MAX_UINT64;
3474 #define lockownerid_hashval(id) \
3475 ((id) & LOCK_HASH_MASK)
3477 static inline unsigned int
3478 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3479 struct xdr_netobj *ownername)
3481 return (file_hashval(inode) + cl_id
3482 + opaque_hashval(ownername->data, ownername->len))
3486 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3487 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3488 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3490 static struct nfs4_stateid *
3491 find_stateid(stateid_t *stid, int flags)
3493 struct nfs4_stateid *local;
3494 u32 st_id = stid->si_stateownerid;
3495 u32 f_id = stid->si_fileid;
3496 unsigned int hashval;
3498 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3499 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3500 hashval = stateid_hashval(st_id, f_id);
3501 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3502 if ((local->st_stateid.si_stateownerid == st_id) &&
3503 (local->st_stateid.si_fileid == f_id))
3508 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3509 hashval = stateid_hashval(st_id, f_id);
3510 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3511 if ((local->st_stateid.si_stateownerid == st_id) &&
3512 (local->st_stateid.si_fileid == f_id))
3519 static struct nfs4_delegation *
3520 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3522 struct nfs4_file *fp;
3523 struct nfs4_delegation *dl;
3525 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3528 fp = find_file(ino);
3531 dl = find_delegation_file(fp, stid);
3537 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3538 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3539 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3540 * locking, this prevents us from being completely protocol-compliant. The
3541 * real solution to this problem is to start using unsigned file offsets in
3542 * the VFS, but this is a very deep change!
3545 nfs4_transform_lock_offset(struct file_lock *lock)
3547 if (lock->fl_start < 0)
3548 lock->fl_start = OFFSET_MAX;
3549 if (lock->fl_end < 0)
3550 lock->fl_end = OFFSET_MAX;
3553 /* Hack!: For now, we're defining this just so we can use a pointer to it
3554 * as a unique cookie to identify our (NFSv4's) posix locks. */
3555 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3559 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3561 struct nfs4_stateowner *sop;
3563 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3564 sop = (struct nfs4_stateowner *) fl->fl_owner;
3565 kref_get(&sop->so_ref);
3567 deny->ld_clientid = sop->so_client->cl_clientid;
3569 deny->ld_sop = NULL;
3570 deny->ld_clientid.cl_boot = 0;
3571 deny->ld_clientid.cl_id = 0;
3573 deny->ld_start = fl->fl_start;
3574 deny->ld_length = NFS4_MAX_UINT64;
3575 if (fl->fl_end != NFS4_MAX_UINT64)
3576 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3577 deny->ld_type = NFS4_READ_LT;
3578 if (fl->fl_type != F_RDLCK)
3579 deny->ld_type = NFS4_WRITE_LT;
3582 static struct nfs4_stateowner *
3583 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3584 struct xdr_netobj *owner)
3586 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3587 struct nfs4_stateowner *op;
3589 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3590 if (same_owner_str(op, owner, clid))
3597 * Alloc a lock owner structure.
3598 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3601 * strhashval = lock_ownerstr_hashval
3604 static struct nfs4_stateowner *
3605 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3606 struct nfs4_stateowner *sop;
3607 struct nfs4_replay *rp;
3608 unsigned int idhashval;
3610 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3612 idhashval = lockownerid_hashval(current_ownerid);
3613 INIT_LIST_HEAD(&sop->so_idhash);
3614 INIT_LIST_HEAD(&sop->so_strhash);
3615 INIT_LIST_HEAD(&sop->so_perclient);
3616 INIT_LIST_HEAD(&sop->so_stateids);
3617 INIT_LIST_HEAD(&sop->so_perstateid);
3618 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3620 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3621 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3622 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3623 sop->so_is_open_owner = 0;
3624 sop->so_id = current_ownerid++;
3625 sop->so_client = clp;
3626 /* It is the openowner seqid that will be incremented in encode in the
3627 * case of new lockowners; so increment the lock seqid manually: */
3628 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3629 sop->so_confirmed = 1;
3630 rp = &sop->so_replay;
3631 rp->rp_status = nfserr_serverfault;
3633 rp->rp_buf = rp->rp_ibuf;
3637 static struct nfs4_stateid *
3638 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3640 struct nfs4_stateid *stp;
3641 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3643 stp = nfs4_alloc_stateid();
3646 INIT_LIST_HEAD(&stp->st_hash);
3647 INIT_LIST_HEAD(&stp->st_perfile);
3648 INIT_LIST_HEAD(&stp->st_perstateowner);
3649 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3650 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3651 list_add(&stp->st_perfile, &fp->fi_stateids);
3652 list_add(&stp->st_perstateowner, &sop->so_stateids);
3653 stp->st_stateowner = sop;
3656 stp->st_stateid.si_boot = boot_time;
3657 stp->st_stateid.si_stateownerid = sop->so_id;
3658 stp->st_stateid.si_fileid = fp->fi_id;
3659 stp->st_stateid.si_generation = 0;
3660 stp->st_deny_bmap = open_stp->st_deny_bmap;
3661 stp->st_openstp = open_stp;
3668 check_lock_length(u64 offset, u64 length)
3670 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3671 LOFF_OVERFLOW(offset, length)));
3678 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3679 struct nfsd4_lock *lock)
3681 struct nfs4_stateowner *open_sop = NULL;
3682 struct nfs4_stateowner *lock_sop = NULL;
3683 struct nfs4_stateid *lock_stp;
3684 struct nfs4_file *fp;
3685 struct file *filp = NULL;
3686 struct file_lock file_lock;
3687 struct file_lock conflock;
3689 unsigned int strhashval;
3693 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3694 (long long) lock->lk_offset,
3695 (long long) lock->lk_length);
3697 if (check_lock_length(lock->lk_offset, lock->lk_length))
3698 return nfserr_inval;
3700 if ((status = fh_verify(rqstp, &cstate->current_fh,
3701 S_IFREG, NFSD_MAY_LOCK))) {
3702 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3708 if (lock->lk_is_new) {
3710 * Client indicates that this is a new lockowner.
3711 * Use open owner and open stateid to create lock owner and
3714 struct nfs4_stateid *open_stp = NULL;
3716 status = nfserr_stale_clientid;
3717 if (!nfsd4_has_session(cstate) &&
3718 STALE_CLIENTID(&lock->lk_new_clientid))
3721 /* validate and update open stateid and open seqid */
3722 status = nfs4_preprocess_seqid_op(cstate,
3723 lock->lk_new_open_seqid,
3724 &lock->lk_new_open_stateid,
3726 &lock->lk_replay_owner, &open_stp,
3730 open_sop = lock->lk_replay_owner;
3731 /* create lockowner and lock stateid */
3732 fp = open_stp->st_file;
3733 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3734 open_sop->so_client->cl_clientid.cl_id,
3735 &lock->v.new.owner);
3736 /* XXX: Do we need to check for duplicate stateowners on
3737 * the same file, or should they just be allowed (and
3738 * create new stateids)? */
3739 status = nfserr_resource;
3740 lock_sop = alloc_init_lock_stateowner(strhashval,
3741 open_sop->so_client, open_stp, lock);
3742 if (lock_sop == NULL)
3744 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3745 if (lock_stp == NULL)
3748 /* lock (lock owner + lock stateid) already exists */
3749 status = nfs4_preprocess_seqid_op(cstate,
3750 lock->lk_old_lock_seqid,
3751 &lock->lk_old_lock_stateid,
3753 &lock->lk_replay_owner, &lock_stp, lock);
3756 lock_sop = lock->lk_replay_owner;
3757 fp = lock_stp->st_file;
3759 /* lock->lk_replay_owner and lock_stp have been created or found */
3761 status = nfserr_grace;
3762 if (locks_in_grace() && !lock->lk_reclaim)
3764 status = nfserr_no_grace;
3765 if (!locks_in_grace() && lock->lk_reclaim)
3768 locks_init_lock(&file_lock);
3769 switch (lock->lk_type) {
3772 if (find_readable_file(lock_stp->st_file)) {
3773 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3774 filp = find_readable_file(lock_stp->st_file);
3776 file_lock.fl_type = F_RDLCK;
3780 case NFS4_WRITEW_LT:
3781 if (find_writeable_file(lock_stp->st_file)) {
3782 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3783 filp = find_writeable_file(lock_stp->st_file);
3785 file_lock.fl_type = F_WRLCK;
3789 status = nfserr_inval;
3793 status = nfserr_openmode;
3796 file_lock.fl_owner = (fl_owner_t)lock_sop;
3797 file_lock.fl_pid = current->tgid;
3798 file_lock.fl_file = filp;
3799 file_lock.fl_flags = FL_POSIX;
3800 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3802 file_lock.fl_start = lock->lk_offset;
3803 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3804 nfs4_transform_lock_offset(&file_lock);
3807 * Try to lock the file in the VFS.
3808 * Note: locks.c uses the BKL to protect the inode's lock list.
3811 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3813 case 0: /* success! */
3814 update_stateid(&lock_stp->st_stateid);
3815 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3819 case (EAGAIN): /* conflock holds conflicting lock */
3820 status = nfserr_denied;
3821 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3822 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3825 status = nfserr_deadlock;
3828 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3829 status = nfserr_resource;
3833 if (status && lock->lk_is_new && lock_sop)
3834 release_lockowner(lock_sop);
3835 if (lock->lk_replay_owner) {
3836 nfs4_get_stateowner(lock->lk_replay_owner);
3837 cstate->replay_owner = lock->lk_replay_owner;
3839 nfs4_unlock_state();
3844 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3845 * so we do a temporary open here just to get an open file to pass to
3846 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3849 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3854 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3857 err = vfs_test_lock(file, lock);
3866 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3867 struct nfsd4_lockt *lockt)
3869 struct inode *inode;
3870 struct file_lock file_lock;
3874 if (locks_in_grace())
3875 return nfserr_grace;
3877 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3878 return nfserr_inval;
3880 lockt->lt_stateowner = NULL;
3883 status = nfserr_stale_clientid;
3884 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3887 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3888 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3889 if (status == nfserr_symlink)
3890 status = nfserr_inval;
3894 inode = cstate->current_fh.fh_dentry->d_inode;
3895 locks_init_lock(&file_lock);
3896 switch (lockt->lt_type) {
3899 file_lock.fl_type = F_RDLCK;
3902 case NFS4_WRITEW_LT:
3903 file_lock.fl_type = F_WRLCK;
3906 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3907 status = nfserr_inval;
3911 lockt->lt_stateowner = find_lockstateowner_str(inode,
3912 &lockt->lt_clientid, &lockt->lt_owner);
3913 if (lockt->lt_stateowner)
3914 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3915 file_lock.fl_pid = current->tgid;
3916 file_lock.fl_flags = FL_POSIX;
3918 file_lock.fl_start = lockt->lt_offset;
3919 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3921 nfs4_transform_lock_offset(&file_lock);
3924 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3926 status = nfserrno(error);
3929 if (file_lock.fl_type != F_UNLCK) {
3930 status = nfserr_denied;
3931 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3934 nfs4_unlock_state();
3939 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3940 struct nfsd4_locku *locku)
3942 struct nfs4_stateid *stp;
3943 struct file *filp = NULL;
3944 struct file_lock file_lock;
3948 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3949 (long long) locku->lu_offset,
3950 (long long) locku->lu_length);
3952 if (check_lock_length(locku->lu_offset, locku->lu_length))
3953 return nfserr_inval;
3957 if ((status = nfs4_preprocess_seqid_op(cstate,
3961 &locku->lu_stateowner, &stp, NULL)))
3964 filp = find_any_file(stp->st_file);
3966 status = nfserr_lock_range;
3970 locks_init_lock(&file_lock);
3971 file_lock.fl_type = F_UNLCK;
3972 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3973 file_lock.fl_pid = current->tgid;
3974 file_lock.fl_file = filp;
3975 file_lock.fl_flags = FL_POSIX;
3976 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3977 file_lock.fl_start = locku->lu_offset;
3979 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3980 nfs4_transform_lock_offset(&file_lock);
3983 * Try to unlock the file in the VFS.
3985 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3987 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3991 * OK, unlock succeeded; the only thing left to do is update the stateid.
3993 update_stateid(&stp->st_stateid);
3994 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3997 if (locku->lu_stateowner) {
3998 nfs4_get_stateowner(locku->lu_stateowner);
3999 cstate->replay_owner = locku->lu_stateowner;
4001 nfs4_unlock_state();
4005 status = nfserrno(err);
4011 * 1: locks held by lockowner
4012 * 0: no locks held by lockowner
4015 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4017 struct file_lock **flpp;
4018 struct inode *inode = filp->fi_inode;
4022 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4023 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4034 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4035 struct nfsd4_compound_state *cstate,
4036 struct nfsd4_release_lockowner *rlockowner)
4038 clientid_t *clid = &rlockowner->rl_clientid;
4039 struct nfs4_stateowner *sop;
4040 struct nfs4_stateid *stp;
4041 struct xdr_netobj *owner = &rlockowner->rl_owner;
4042 struct list_head matches;
4046 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4047 clid->cl_boot, clid->cl_id);
4049 /* XXX check for lease expiration */
4051 status = nfserr_stale_clientid;
4052 if (STALE_CLIENTID(clid))
4057 status = nfserr_locks_held;
4058 /* XXX: we're doing a linear search through all the lockowners.
4059 * Yipes! For now we'll just hope clients aren't really using
4060 * release_lockowner much, but eventually we have to fix these
4061 * data structures. */
4062 INIT_LIST_HEAD(&matches);
4063 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4064 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4065 if (!same_owner_str(sop, owner, clid))
4067 list_for_each_entry(stp, &sop->so_stateids,
4069 if (check_for_locks(stp->st_file, sop))
4071 /* Note: so_perclient unused for lockowners,
4072 * so it's OK to fool with here. */
4073 list_add(&sop->so_perclient, &matches);
4077 /* Clients probably won't expect us to return with some (but not all)
4078 * of the lockowner state released; so don't release any until all
4079 * have been checked. */
4081 while (!list_empty(&matches)) {
4082 sop = list_entry(matches.next, struct nfs4_stateowner,
4084 /* unhash_stateowner deletes so_perclient only
4085 * for openowners. */
4086 list_del(&sop->so_perclient);
4087 release_lockowner(sop);
4090 nfs4_unlock_state();
4094 static inline struct nfs4_client_reclaim *
4097 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4101 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4103 unsigned int strhashval = clientstr_hashval(name);
4104 struct nfs4_client *clp;
4106 clp = find_confirmed_client_by_str(name, strhashval);
4111 * failure => all reset bets are off, nfserr_no_grace...
4114 nfs4_client_to_reclaim(const char *name)
4116 unsigned int strhashval;
4117 struct nfs4_client_reclaim *crp = NULL;
4119 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4120 crp = alloc_reclaim();
4123 strhashval = clientstr_hashval(name);
4124 INIT_LIST_HEAD(&crp->cr_strhash);
4125 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4126 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4127 reclaim_str_hashtbl_size++;
4132 nfs4_release_reclaim(void)
4134 struct nfs4_client_reclaim *crp = NULL;
4137 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4138 while (!list_empty(&reclaim_str_hashtbl[i])) {
4139 crp = list_entry(reclaim_str_hashtbl[i].next,
4140 struct nfs4_client_reclaim, cr_strhash);
4141 list_del(&crp->cr_strhash);
4143 reclaim_str_hashtbl_size--;
4146 BUG_ON(reclaim_str_hashtbl_size);
4150 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4151 static struct nfs4_client_reclaim *
4152 nfs4_find_reclaim_client(clientid_t *clid)
4154 unsigned int strhashval;
4155 struct nfs4_client *clp;
4156 struct nfs4_client_reclaim *crp = NULL;
4159 /* find clientid in conf_id_hashtbl */
4160 clp = find_confirmed_client(clid);
4164 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4165 clp->cl_name.len, clp->cl_name.data,
4168 /* find clp->cl_name in reclaim_str_hashtbl */
4169 strhashval = clientstr_hashval(clp->cl_recdir);
4170 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4171 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4179 * Called from OPEN. Look for clientid in reclaim list.
4182 nfs4_check_open_reclaim(clientid_t *clid)
4184 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4187 /* initialization to perform at module load time: */
4190 nfs4_state_init(void)
4194 status = nfsd4_init_slabs();
4197 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4198 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4199 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4200 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4201 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4202 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4204 for (i = 0; i < SESSION_HASH_SIZE; i++)
4205 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4206 for (i = 0; i < FILE_HASH_SIZE; i++) {
4207 INIT_LIST_HEAD(&file_hashtbl[i]);
4209 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4210 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4211 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4213 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4214 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4215 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4217 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4218 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4219 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4221 memset(&onestateid, ~0, sizeof(stateid_t));
4222 INIT_LIST_HEAD(&close_lru);
4223 INIT_LIST_HEAD(&client_lru);
4224 INIT_LIST_HEAD(&del_recall_lru);
4225 reclaim_str_hashtbl_size = 0;
4230 nfsd4_load_reboot_recovery_data(void)
4235 nfsd4_init_recdir(user_recovery_dirname);
4236 status = nfsd4_recdir_load();
4237 nfs4_unlock_state();
4239 printk("NFSD: Failure reading reboot recovery data\n");
4243 * Since the lifetime of a delegation isn't limited to that of an open, a
4244 * client may quite reasonably hang on to a delegation as long as it has
4245 * the inode cached. This becomes an obvious problem the first time a
4246 * client's inode cache approaches the size of the server's total memory.
4248 * For now we avoid this problem by imposing a hard limit on the number
4249 * of delegations, which varies according to the server's memory size.
4252 set_max_delegations(void)
4255 * Allow at most 4 delegations per megabyte of RAM. Quick
4256 * estimates suggest that in the worst case (where every delegation
4257 * is for a different inode), a delegation could take about 1.5K,
4258 * giving a worst case usage of about 6% of memory.
4260 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4263 /* initialization to perform when the nfsd service is started: */
4266 __nfs4_state_start(void)
4270 boot_time = get_seconds();
4271 locks_start_grace(&nfsd4_manager);
4272 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4274 ret = set_callback_cred();
4277 laundry_wq = create_singlethread_workqueue("nfsd4");
4278 if (laundry_wq == NULL)
4280 ret = nfsd4_create_callback_queue();
4282 goto out_free_laundry;
4283 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4284 set_max_delegations();
4287 destroy_workqueue(laundry_wq);
4292 nfs4_state_start(void)
4294 nfsd4_load_reboot_recovery_data();
4295 return __nfs4_state_start();
4299 __nfs4_state_shutdown(void)
4302 struct nfs4_client *clp = NULL;
4303 struct nfs4_delegation *dp = NULL;
4304 struct list_head *pos, *next, reaplist;
4306 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4307 while (!list_empty(&conf_id_hashtbl[i])) {
4308 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4311 while (!list_empty(&unconf_str_hashtbl[i])) {
4312 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4316 INIT_LIST_HEAD(&reaplist);
4317 spin_lock(&recall_lock);
4318 list_for_each_safe(pos, next, &del_recall_lru) {
4319 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4320 list_move(&dp->dl_recall_lru, &reaplist);
4322 spin_unlock(&recall_lock);
4323 list_for_each_safe(pos, next, &reaplist) {
4324 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4325 list_del_init(&dp->dl_recall_lru);
4326 unhash_delegation(dp);
4329 nfsd4_shutdown_recdir();
4333 nfs4_state_shutdown(void)
4335 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4336 destroy_workqueue(laundry_wq);
4337 locks_end_grace(&nfsd4_manager);
4339 nfs4_release_reclaim();
4340 __nfs4_state_shutdown();
4341 nfs4_unlock_state();
4342 nfsd4_destroy_callback_queue();
4346 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4347 * accessed when nfsd is starting.
4350 nfs4_set_recdir(char *recdir)
4352 strcpy(user_recovery_dirname, recdir);
4356 * Change the NFSv4 recovery directory to recdir.
4359 nfs4_reset_recoverydir(char *recdir)
4364 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4368 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4369 nfs4_set_recdir(recdir);
4377 nfs4_recoverydir(void)
4379 return user_recovery_dirname;