2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
36 #include <linux/smp_lock.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
45 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 time_t nfsd4_lease = 90; /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static stateid_t zerostateid; /* bits all 0 */
55 static stateid_t onestateid; /* bits all 1 */
56 static u64 current_sessionid = 1;
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
73 * Currently used for the del_recall_lru and file hash table. In an
74 * effort to decrease the scope of the client_mutex, this spinlock may
75 * eventually cover more:
77 static DEFINE_SPINLOCK(recall_lock);
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
87 mutex_lock(&client_mutex);
91 nfs4_unlock_state(void)
93 mutex_unlock(&client_mutex);
97 opaque_hashval(const void *ptr, int nbytes)
99 unsigned char *cptr = (unsigned char *) ptr;
109 static struct list_head del_recall_lru;
112 put_nfs4_file(struct nfs4_file *fi)
114 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115 list_del(&fi->fi_hash);
116 spin_unlock(&recall_lock);
118 kmem_cache_free(file_slab, fi);
123 get_nfs4_file(struct nfs4_file *fi)
125 atomic_inc(&fi->fi_ref);
128 static int num_delegations;
129 unsigned int max_delegations;
132 * Open owner state (share locks)
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS 8
137 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
140 #define ownerid_hashval(id) \
141 ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS 8
150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS 10
154 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
157 #define file_hashval(x) \
158 hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id) \
160 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
167 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168 atomic_inc(&fp->fi_access[oflag]);
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
173 if (oflag == O_RDWR) {
174 __nfs4_file_get_access(fp, O_RDONLY);
175 __nfs4_file_get_access(fp, O_WRONLY);
177 __nfs4_file_get_access(fp, oflag);
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
182 if (fp->fi_fds[oflag]) {
183 fput(fp->fi_fds[oflag]);
184 fp->fi_fds[oflag] = NULL;
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
190 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191 nfs4_file_put_fd(fp, O_RDWR);
192 nfs4_file_put_fd(fp, oflag);
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
198 if (oflag == O_RDWR) {
199 __nfs4_file_put_access(fp, O_RDONLY);
200 __nfs4_file_put_access(fp, O_WRONLY);
202 __nfs4_file_put_access(fp, oflag);
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
208 struct nfs4_delegation *dp;
209 struct nfs4_file *fp = stp->st_file;
211 dprintk("NFSD alloc_init_deleg\n");
213 * Major work on the lease subsystem (for example, to support
214 * calbacks on stat) will be required before we can support
215 * write delegations properly.
217 if (type != NFS4_OPEN_DELEGATE_READ)
219 if (fp->fi_had_conflict)
221 if (num_delegations > max_delegations)
223 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
227 INIT_LIST_HEAD(&dp->dl_perfile);
228 INIT_LIST_HEAD(&dp->dl_perclnt);
229 INIT_LIST_HEAD(&dp->dl_recall_lru);
233 nfs4_file_get_access(fp, O_RDONLY);
236 dp->dl_stateid.si_boot = boot_time;
237 dp->dl_stateid.si_stateownerid = current_delegid++;
238 dp->dl_stateid.si_fileid = 0;
239 dp->dl_stateid.si_generation = 0;
240 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
242 atomic_set(&dp->dl_count, 1);
243 list_add(&dp->dl_perfile, &fp->fi_delegations);
244 list_add(&dp->dl_perclnt, &clp->cl_delegations);
245 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
250 nfs4_put_delegation(struct nfs4_delegation *dp)
252 if (atomic_dec_and_test(&dp->dl_count)) {
253 dprintk("NFSD: freeing dp %p\n",dp);
254 put_nfs4_file(dp->dl_file);
255 kmem_cache_free(deleg_slab, dp);
260 /* Remove the associated file_lock first, then remove the delegation.
261 * lease_modify() is called to remove the FS_LEASE file_lock from
262 * the i_flock list, eventually calling nfsd's lock_manager
263 * fl_release_callback.
266 nfs4_close_delegation(struct nfs4_delegation *dp)
268 struct file *filp = find_readable_file(dp->dl_file);
270 dprintk("NFSD: close_delegation dp %p\n",dp);
272 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
273 nfs4_file_put_access(dp->dl_file, O_RDONLY);
276 /* Called under the state lock. */
278 unhash_delegation(struct nfs4_delegation *dp)
280 list_del_init(&dp->dl_perfile);
281 list_del_init(&dp->dl_perclnt);
282 spin_lock(&recall_lock);
283 list_del_init(&dp->dl_recall_lru);
284 spin_unlock(&recall_lock);
285 nfs4_close_delegation(dp);
286 nfs4_put_delegation(dp);
293 /* client_lock protects the client lru list and session hash table */
294 static DEFINE_SPINLOCK(client_lock);
296 /* Hash tables for nfs4_clientid state */
297 #define CLIENT_HASH_BITS 4
298 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
299 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
301 #define clientid_hashval(id) \
302 ((id) & CLIENT_HASH_MASK)
303 #define clientstr_hashval(name) \
304 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
306 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
307 * used in reboot/reset lease grace period processing
309 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
310 * setclientid_confirmed info.
312 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
315 * client_lru holds client queue ordered by nfs4_client.cl_time
318 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
319 * for last close replay.
321 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
322 static int reclaim_str_hashtbl_size = 0;
323 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
324 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
325 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
326 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
327 static struct list_head client_lru;
328 static struct list_head close_lru;
330 static void unhash_generic_stateid(struct nfs4_stateid *stp)
332 list_del(&stp->st_hash);
333 list_del(&stp->st_perfile);
334 list_del(&stp->st_perstateowner);
337 static void free_generic_stateid(struct nfs4_stateid *stp)
339 put_nfs4_file(stp->st_file);
340 kmem_cache_free(stateid_slab, stp);
343 static void release_lock_stateid(struct nfs4_stateid *stp)
347 unhash_generic_stateid(stp);
348 file = find_any_file(stp->st_file);
350 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
351 free_generic_stateid(stp);
354 static void unhash_lockowner(struct nfs4_stateowner *sop)
356 struct nfs4_stateid *stp;
358 list_del(&sop->so_idhash);
359 list_del(&sop->so_strhash);
360 list_del(&sop->so_perstateid);
361 while (!list_empty(&sop->so_stateids)) {
362 stp = list_first_entry(&sop->so_stateids,
363 struct nfs4_stateid, st_perstateowner);
364 release_lock_stateid(stp);
368 static void release_lockowner(struct nfs4_stateowner *sop)
370 unhash_lockowner(sop);
371 nfs4_put_stateowner(sop);
375 release_stateid_lockowners(struct nfs4_stateid *open_stp)
377 struct nfs4_stateowner *lock_sop;
379 while (!list_empty(&open_stp->st_lockowners)) {
380 lock_sop = list_entry(open_stp->st_lockowners.next,
381 struct nfs4_stateowner, so_perstateid);
382 /* list_del(&open_stp->st_lockowners); */
383 BUG_ON(lock_sop->so_is_open_owner);
384 release_lockowner(lock_sop);
389 * We store the NONE, READ, WRITE, and BOTH bits separately in the
390 * st_{access,deny}_bmap field of the stateid, in order to track not
391 * only what share bits are currently in force, but also what
392 * combinations of share bits previous opens have used. This allows us
393 * to enforce the recommendation of rfc 3530 14.2.19 that the server
394 * return an error if the client attempt to downgrade to a combination
395 * of share bits not explicable by closing some of its previous opens.
397 * XXX: This enforcement is actually incomplete, since we don't keep
398 * track of access/deny bit combinations; so, e.g., we allow:
400 * OPEN allow read, deny write
401 * OPEN allow both, deny none
402 * DOWNGRADE allow read, deny none
404 * which we should reject.
407 set_access(unsigned int *access, unsigned long bmap) {
411 for (i = 1; i < 4; i++) {
412 if (test_bit(i, &bmap))
418 set_deny(unsigned int *deny, unsigned long bmap) {
422 for (i = 0; i < 4; i++) {
423 if (test_bit(i, &bmap))
429 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
430 unsigned int access, deny;
432 set_access(&access, stp->st_access_bmap);
433 set_deny(&deny, stp->st_deny_bmap);
434 if ((access & open->op_share_deny) || (deny & open->op_share_access))
439 static int nfs4_access_to_omode(u32 access)
441 switch (access & NFS4_SHARE_ACCESS_BOTH) {
442 case NFS4_SHARE_ACCESS_READ:
444 case NFS4_SHARE_ACCESS_WRITE:
446 case NFS4_SHARE_ACCESS_BOTH:
452 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
456 set_access(&access, stp->st_access_bmap);
457 return nfs4_access_to_omode(access);
460 static void release_open_stateid(struct nfs4_stateid *stp)
462 int oflag = nfs4_access_bmap_to_omode(stp);
464 unhash_generic_stateid(stp);
465 release_stateid_lockowners(stp);
466 nfs4_file_put_access(stp->st_file, oflag);
467 free_generic_stateid(stp);
470 static void unhash_openowner(struct nfs4_stateowner *sop)
472 struct nfs4_stateid *stp;
474 list_del(&sop->so_idhash);
475 list_del(&sop->so_strhash);
476 list_del(&sop->so_perclient);
477 list_del(&sop->so_perstateid); /* XXX: necessary? */
478 while (!list_empty(&sop->so_stateids)) {
479 stp = list_first_entry(&sop->so_stateids,
480 struct nfs4_stateid, st_perstateowner);
481 release_open_stateid(stp);
485 static void release_openowner(struct nfs4_stateowner *sop)
487 unhash_openowner(sop);
488 list_del(&sop->so_close_lru);
489 nfs4_put_stateowner(sop);
492 #define SESSION_HASH_SIZE 512
493 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
496 hash_sessionid(struct nfs4_sessionid *sessionid)
498 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
500 return sid->sequence % SESSION_HASH_SIZE;
504 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
506 u32 *ptr = (u32 *)(&sessionid->data[0]);
507 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
511 gen_sessionid(struct nfsd4_session *ses)
513 struct nfs4_client *clp = ses->se_client;
514 struct nfsd4_sessionid *sid;
516 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
517 sid->clientid = clp->cl_clientid;
518 sid->sequence = current_sessionid++;
523 * The protocol defines ca_maxresponssize_cached to include the size of
524 * the rpc header, but all we need to cache is the data starting after
525 * the end of the initial SEQUENCE operation--the rest we regenerate
526 * each time. Therefore we can advertise a ca_maxresponssize_cached
527 * value that is the number of bytes in our cache plus a few additional
528 * bytes. In order to stay on the safe side, and not promise more than
529 * we can cache, those additional bytes must be the minimum possible: 24
530 * bytes of rpc header (xid through accept state, with AUTH_NULL
531 * verifier), 12 for the compound header (with zero-length tag), and 44
532 * for the SEQUENCE op response:
534 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
537 free_session_slots(struct nfsd4_session *ses)
541 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
542 kfree(ses->se_slots[i]);
546 * We don't actually need to cache the rpc and session headers, so we
547 * can allocate a little less for each slot:
549 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
551 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
554 static int nfsd4_sanitize_slot_size(u32 size)
556 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
557 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
563 * XXX: If we run out of reserved DRC memory we could (up to a point)
564 * re-negotiate active sessions and reduce their slot usage to make
565 * rooom for new connections. For now we just fail the create session.
567 static int nfsd4_get_drc_mem(int slotsize, u32 num)
571 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
573 spin_lock(&nfsd_drc_lock);
574 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
575 nfsd_drc_max_mem - nfsd_drc_mem_used);
576 num = min_t(int, num, avail / slotsize);
577 nfsd_drc_mem_used += num * slotsize;
578 spin_unlock(&nfsd_drc_lock);
583 static void nfsd4_put_drc_mem(int slotsize, int num)
585 spin_lock(&nfsd_drc_lock);
586 nfsd_drc_mem_used -= slotsize * num;
587 spin_unlock(&nfsd_drc_lock);
590 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
592 struct nfsd4_session *new;
595 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
596 + sizeof(struct nfsd4_session) > PAGE_SIZE);
597 mem = numslots * sizeof(struct nfsd4_slot *);
599 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
602 /* allocate each struct nfsd4_slot and data cache in one piece */
603 for (i = 0; i < numslots; i++) {
604 mem = sizeof(struct nfsd4_slot) + slotsize;
605 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
606 if (!new->se_slots[i])
612 kfree(new->se_slots[i]);
617 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
619 u32 maxrpc = nfsd_serv->sv_max_mesg;
621 new->maxreqs = numslots;
622 new->maxresp_cached = slotsize + NFSD_MIN_HDR_SEQ_SZ;
623 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
624 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
625 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
628 static void free_conn(struct nfsd4_conn *c)
630 svc_xprt_put(c->cn_xprt);
634 static void nfsd4_conn_lost(struct svc_xpt_user *u)
636 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
637 struct nfs4_client *clp = c->cn_session->se_client;
639 spin_lock(&clp->cl_lock);
640 if (!list_empty(&c->cn_persession)) {
641 list_del(&c->cn_persession);
644 spin_unlock(&clp->cl_lock);
647 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp)
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 = NFS4_CDFC4_FORE;
657 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
661 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
663 conn->cn_session = ses;
664 list_add(&conn->cn_persession, &ses->se_conns);
667 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
669 struct nfs4_client *clp = ses->se_client;
671 spin_lock(&clp->cl_lock);
672 __nfsd4_hash_conn(conn, ses);
673 spin_unlock(&clp->cl_lock);
676 static void nfsd4_register_conn(struct nfsd4_conn *conn)
678 conn->cn_xpt_user.callback = nfsd4_conn_lost;
679 register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
682 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses)
684 struct nfsd4_conn *conn;
686 conn = alloc_conn(rqstp);
688 return nfserr_jukebox;
689 nfsd4_hash_conn(conn, ses);
690 nfsd4_register_conn(conn);
694 static void nfsd4_del_conns(struct nfsd4_session *s)
696 struct nfs4_client *clp = s->se_client;
697 struct nfsd4_conn *c;
699 spin_lock(&clp->cl_lock);
700 while (!list_empty(&s->se_conns)) {
701 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
702 list_del_init(&c->cn_persession);
703 spin_unlock(&clp->cl_lock);
705 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
708 spin_lock(&clp->cl_lock);
710 spin_unlock(&clp->cl_lock);
713 void free_session(struct kref *kref)
715 struct nfsd4_session *ses;
718 ses = container_of(kref, struct nfsd4_session, se_ref);
719 nfsd4_del_conns(ses);
720 spin_lock(&nfsd_drc_lock);
721 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
722 nfsd_drc_mem_used -= mem;
723 spin_unlock(&nfsd_drc_lock);
724 free_session_slots(ses);
729 static __be32 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
731 struct nfsd4_session *new;
732 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
733 int numslots, slotsize;
738 * Note decreasing slot size below client's request may
739 * make it difficult for client to function correctly, whereas
740 * decreasing the number of slots will (just?) affect
741 * performance. When short on memory we therefore prefer to
742 * decrease number of slots instead of their size.
744 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
745 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
747 new = alloc_session(slotsize, numslots);
749 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
750 return nfserr_jukebox;
752 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
754 new->se_client = clp;
756 memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
757 NFS4_MAX_SESSIONID_LEN);
759 INIT_LIST_HEAD(&new->se_conns);
761 new->se_flags = cses->flags;
762 kref_init(&new->se_ref);
763 idx = hash_sessionid(&new->se_sessionid);
764 spin_lock(&client_lock);
765 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
766 list_add(&new->se_perclnt, &clp->cl_sessions);
767 spin_unlock(&client_lock);
769 status = nfsd4_new_conn(rqstp, new);
771 free_session(&new->se_ref);
772 return nfserr_jukebox;
777 /* caller must hold client_lock */
778 static struct nfsd4_session *
779 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
781 struct nfsd4_session *elem;
784 dump_sessionid(__func__, sessionid);
785 idx = hash_sessionid(sessionid);
786 /* Search in the appropriate list */
787 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
788 if (!memcmp(elem->se_sessionid.data, sessionid->data,
789 NFS4_MAX_SESSIONID_LEN)) {
794 dprintk("%s: session not found\n", __func__);
798 /* caller must hold client_lock */
800 unhash_session(struct nfsd4_session *ses)
802 list_del(&ses->se_hash);
803 list_del(&ses->se_perclnt);
806 /* must be called under the client_lock */
808 renew_client_locked(struct nfs4_client *clp)
810 if (is_client_expired(clp)) {
811 dprintk("%s: client (clientid %08x/%08x) already expired\n",
813 clp->cl_clientid.cl_boot,
814 clp->cl_clientid.cl_id);
819 * Move client to the end to the LRU list.
821 dprintk("renewing client (clientid %08x/%08x)\n",
822 clp->cl_clientid.cl_boot,
823 clp->cl_clientid.cl_id);
824 list_move_tail(&clp->cl_lru, &client_lru);
825 clp->cl_time = get_seconds();
829 renew_client(struct nfs4_client *clp)
831 spin_lock(&client_lock);
832 renew_client_locked(clp);
833 spin_unlock(&client_lock);
836 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
838 STALE_CLIENTID(clientid_t *clid)
840 if (clid->cl_boot == boot_time)
842 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
843 clid->cl_boot, clid->cl_id, boot_time);
848 * XXX Should we use a slab cache ?
849 * This type of memory management is somewhat inefficient, but we use it
850 * anyway since SETCLIENTID is not a common operation.
852 static struct nfs4_client *alloc_client(struct xdr_netobj name)
854 struct nfs4_client *clp;
856 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
859 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
860 if (clp->cl_name.data == NULL) {
864 memcpy(clp->cl_name.data, name.data, name.len);
865 clp->cl_name.len = name.len;
870 free_client(struct nfs4_client *clp)
872 if (clp->cl_cred.cr_group_info)
873 put_group_info(clp->cl_cred.cr_group_info);
874 kfree(clp->cl_principal);
875 kfree(clp->cl_name.data);
880 release_session_client(struct nfsd4_session *session)
882 struct nfs4_client *clp = session->se_client;
884 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
886 if (is_client_expired(clp)) {
888 session->se_client = NULL;
890 renew_client_locked(clp);
891 spin_unlock(&client_lock);
894 /* must be called under the client_lock */
896 unhash_client_locked(struct nfs4_client *clp)
898 mark_client_expired(clp);
899 list_del(&clp->cl_lru);
900 while (!list_empty(&clp->cl_sessions)) {
901 struct nfsd4_session *ses;
902 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
905 nfsd4_put_session(ses);
910 expire_client(struct nfs4_client *clp)
912 struct nfs4_stateowner *sop;
913 struct nfs4_delegation *dp;
914 struct list_head reaplist;
916 INIT_LIST_HEAD(&reaplist);
917 spin_lock(&recall_lock);
918 while (!list_empty(&clp->cl_delegations)) {
919 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
920 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
922 list_del_init(&dp->dl_perclnt);
923 list_move(&dp->dl_recall_lru, &reaplist);
925 spin_unlock(&recall_lock);
926 while (!list_empty(&reaplist)) {
927 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
928 list_del_init(&dp->dl_recall_lru);
929 unhash_delegation(dp);
931 while (!list_empty(&clp->cl_openowners)) {
932 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
933 release_openowner(sop);
935 nfsd4_shutdown_callback(clp);
936 if (clp->cl_cb_conn.cb_xprt)
937 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
938 list_del(&clp->cl_idhash);
939 list_del(&clp->cl_strhash);
940 spin_lock(&client_lock);
941 unhash_client_locked(clp);
942 if (atomic_read(&clp->cl_refcount) == 0)
944 spin_unlock(&client_lock);
947 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
949 memcpy(target->cl_verifier.data, source->data,
950 sizeof(target->cl_verifier.data));
953 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
955 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
956 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
959 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
961 target->cr_uid = source->cr_uid;
962 target->cr_gid = source->cr_gid;
963 target->cr_group_info = source->cr_group_info;
964 get_group_info(target->cr_group_info);
967 static int same_name(const char *n1, const char *n2)
969 return 0 == memcmp(n1, n2, HEXDIR_LEN);
973 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
975 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
979 same_clid(clientid_t *cl1, clientid_t *cl2)
981 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
984 /* XXX what about NGROUP */
986 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
988 return cr1->cr_uid == cr2->cr_uid;
991 static void gen_clid(struct nfs4_client *clp)
993 static u32 current_clientid = 1;
995 clp->cl_clientid.cl_boot = boot_time;
996 clp->cl_clientid.cl_id = current_clientid++;
999 static void gen_confirm(struct nfs4_client *clp)
1004 p = (u32 *)clp->cl_confirm.data;
1005 *p++ = get_seconds();
1009 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1010 struct svc_rqst *rqstp, nfs4_verifier *verf)
1012 struct nfs4_client *clp;
1013 struct sockaddr *sa = svc_addr(rqstp);
1016 clp = alloc_client(name);
1020 princ = svc_gss_principal(rqstp);
1022 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1023 if (clp->cl_principal == NULL) {
1029 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1030 atomic_set(&clp->cl_refcount, 0);
1031 atomic_set(&clp->cl_cb_set, 0);
1032 INIT_LIST_HEAD(&clp->cl_idhash);
1033 INIT_LIST_HEAD(&clp->cl_strhash);
1034 INIT_LIST_HEAD(&clp->cl_openowners);
1035 INIT_LIST_HEAD(&clp->cl_delegations);
1036 INIT_LIST_HEAD(&clp->cl_sessions);
1037 INIT_LIST_HEAD(&clp->cl_lru);
1038 spin_lock_init(&clp->cl_lock);
1039 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1040 clp->cl_time = get_seconds();
1041 clear_bit(0, &clp->cl_cb_slot_busy);
1042 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1043 copy_verf(clp, verf);
1044 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1045 clp->cl_flavor = rqstp->rq_flavor;
1046 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1052 static int check_name(struct xdr_netobj name)
1056 if (name.len > NFS4_OPAQUE_LIMIT) {
1057 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1064 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1066 unsigned int idhashval;
1068 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1069 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1070 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1075 move_to_confirmed(struct nfs4_client *clp)
1077 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1078 unsigned int strhashval;
1080 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1081 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1082 strhashval = clientstr_hashval(clp->cl_recdir);
1083 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1087 static struct nfs4_client *
1088 find_confirmed_client(clientid_t *clid)
1090 struct nfs4_client *clp;
1091 unsigned int idhashval = clientid_hashval(clid->cl_id);
1093 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1094 if (same_clid(&clp->cl_clientid, clid))
1100 static struct nfs4_client *
1101 find_unconfirmed_client(clientid_t *clid)
1103 struct nfs4_client *clp;
1104 unsigned int idhashval = clientid_hashval(clid->cl_id);
1106 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1107 if (same_clid(&clp->cl_clientid, clid))
1114 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
1115 * parameter. Matching is based on the fact the at least one of the
1116 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
1118 * FIXME: we need to unify the clientid namespaces for nfsv4.x
1119 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
1120 * and SET_CLIENTID{,_CONFIRM}
1123 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
1125 bool has_exchange_flags = (clp->cl_exchange_flags != 0);
1126 return use_exchange_id == has_exchange_flags;
1129 static struct nfs4_client *
1130 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
1131 bool use_exchange_id)
1133 struct nfs4_client *clp;
1135 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1136 if (same_name(clp->cl_recdir, dname) &&
1137 match_clientid_establishment(clp, use_exchange_id))
1143 static struct nfs4_client *
1144 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
1145 bool use_exchange_id)
1147 struct nfs4_client *clp;
1149 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1150 if (same_name(clp->cl_recdir, dname) &&
1151 match_clientid_establishment(clp, use_exchange_id))
1158 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1160 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1161 unsigned short expected_family;
1163 /* Currently, we only support tcp and tcp6 for the callback channel */
1164 if (se->se_callback_netid_len == 3 &&
1165 !memcmp(se->se_callback_netid_val, "tcp", 3))
1166 expected_family = AF_INET;
1167 else if (se->se_callback_netid_len == 4 &&
1168 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1169 expected_family = AF_INET6;
1173 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1174 se->se_callback_addr_len,
1175 (struct sockaddr *)&conn->cb_addr,
1176 sizeof(conn->cb_addr));
1178 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1181 if (conn->cb_addr.ss_family == AF_INET6)
1182 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1184 conn->cb_minorversion = 0;
1185 conn->cb_prog = se->se_callback_prog;
1186 conn->cb_ident = se->se_callback_ident;
1189 conn->cb_addr.ss_family = AF_UNSPEC;
1190 conn->cb_addrlen = 0;
1191 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1192 "will not receive delegations\n",
1193 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1199 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1202 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1204 struct nfsd4_slot *slot = resp->cstate.slot;
1207 dprintk("--> %s slot %p\n", __func__, slot);
1209 slot->sl_opcnt = resp->opcnt;
1210 slot->sl_status = resp->cstate.status;
1212 if (nfsd4_not_cached(resp)) {
1213 slot->sl_datalen = 0;
1216 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1217 base = (char *)resp->cstate.datap -
1218 (char *)resp->xbuf->head[0].iov_base;
1219 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1221 WARN("%s: sessions DRC could not cache compound\n", __func__);
1226 * Encode the replay sequence operation from the slot values.
1227 * If cachethis is FALSE encode the uncached rep error on the next
1228 * operation which sets resp->p and increments resp->opcnt for
1229 * nfs4svc_encode_compoundres.
1233 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1234 struct nfsd4_compoundres *resp)
1236 struct nfsd4_op *op;
1237 struct nfsd4_slot *slot = resp->cstate.slot;
1239 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1240 resp->opcnt, resp->cstate.slot->sl_cachethis);
1242 /* Encode the replayed sequence operation */
1243 op = &args->ops[resp->opcnt - 1];
1244 nfsd4_encode_operation(resp, op);
1246 /* Return nfserr_retry_uncached_rep in next operation. */
1247 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1248 op = &args->ops[resp->opcnt++];
1249 op->status = nfserr_retry_uncached_rep;
1250 nfsd4_encode_operation(resp, op);
1256 * The sequence operation is not cached because we can use the slot and
1260 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1261 struct nfsd4_sequence *seq)
1263 struct nfsd4_slot *slot = resp->cstate.slot;
1266 dprintk("--> %s slot %p\n", __func__, slot);
1268 /* Either returns 0 or nfserr_retry_uncached */
1269 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1270 if (status == nfserr_retry_uncached_rep)
1273 /* The sequence operation has been encoded, cstate->datap set. */
1274 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1276 resp->opcnt = slot->sl_opcnt;
1277 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1278 status = slot->sl_status;
1284 * Set the exchange_id flags returned by the server.
1287 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1289 /* pNFS is not supported */
1290 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1292 /* Referrals are supported, Migration is not. */
1293 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1295 /* set the wire flags to return to client. */
1296 clid->flags = new->cl_exchange_flags;
1300 nfsd4_exchange_id(struct svc_rqst *rqstp,
1301 struct nfsd4_compound_state *cstate,
1302 struct nfsd4_exchange_id *exid)
1304 struct nfs4_client *unconf, *conf, *new;
1306 unsigned int strhashval;
1307 char dname[HEXDIR_LEN];
1308 char addr_str[INET6_ADDRSTRLEN];
1309 nfs4_verifier verf = exid->verifier;
1310 struct sockaddr *sa = svc_addr(rqstp);
1312 rpc_ntop(sa, addr_str, sizeof(addr_str));
1313 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1314 "ip_addr=%s flags %x, spa_how %d\n",
1315 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1316 addr_str, exid->flags, exid->spa_how);
1318 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1319 return nfserr_inval;
1321 /* Currently only support SP4_NONE */
1322 switch (exid->spa_how) {
1326 return nfserr_encr_alg_unsupp;
1328 BUG(); /* checked by xdr code */
1330 return nfserr_serverfault; /* no excuse :-/ */
1333 status = nfs4_make_rec_clidname(dname, &exid->clname);
1338 strhashval = clientstr_hashval(dname);
1343 conf = find_confirmed_client_by_str(dname, strhashval, true);
1345 if (!same_verf(&verf, &conf->cl_verifier)) {
1346 /* 18.35.4 case 8 */
1347 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1348 status = nfserr_not_same;
1351 /* Client reboot: destroy old state */
1352 expire_client(conf);
1355 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1356 /* 18.35.4 case 9 */
1357 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1358 status = nfserr_perm;
1361 expire_client(conf);
1365 * Set bit when the owner id and verifier map to an already
1366 * confirmed client id (18.35.3).
1368 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1371 * Falling into 18.35.4 case 2, possible router replay.
1372 * Leave confirmed record intact and return same result.
1374 copy_verf(conf, &verf);
1379 /* 18.35.4 case 7 */
1380 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1381 status = nfserr_noent;
1385 unconf = find_unconfirmed_client_by_str(dname, strhashval, true);
1388 * Possible retry or client restart. Per 18.35.4 case 4,
1389 * a new unconfirmed record should be generated regardless
1390 * of whether any properties have changed.
1392 expire_client(unconf);
1397 new = create_client(exid->clname, dname, rqstp, &verf);
1399 status = nfserr_jukebox;
1404 add_to_unconfirmed(new, strhashval);
1406 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1407 exid->clientid.cl_id = new->cl_clientid.cl_id;
1410 nfsd4_set_ex_flags(new, exid);
1412 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1413 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1417 nfs4_unlock_state();
1419 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1424 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1426 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1429 /* The slot is in use, and no response has been sent. */
1431 if (seqid == slot_seqid)
1432 return nfserr_jukebox;
1434 return nfserr_seq_misordered;
1437 if (likely(seqid == slot_seqid + 1))
1440 if (seqid == slot_seqid)
1441 return nfserr_replay_cache;
1443 if (seqid == 1 && (slot_seqid + 1) == 0)
1445 /* Misordered replay or misordered new request */
1446 return nfserr_seq_misordered;
1450 * Cache the create session result into the create session single DRC
1451 * slot cache by saving the xdr structure. sl_seqid has been set.
1452 * Do this for solo or embedded create session operations.
1455 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1456 struct nfsd4_clid_slot *slot, int nfserr)
1458 slot->sl_status = nfserr;
1459 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1463 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1464 struct nfsd4_clid_slot *slot)
1466 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1467 return slot->sl_status;
1471 nfsd4_create_session(struct svc_rqst *rqstp,
1472 struct nfsd4_compound_state *cstate,
1473 struct nfsd4_create_session *cr_ses)
1475 struct sockaddr *sa = svc_addr(rqstp);
1476 struct nfs4_client *conf, *unconf;
1477 struct nfsd4_clid_slot *cs_slot = NULL;
1481 unconf = find_unconfirmed_client(&cr_ses->clientid);
1482 conf = find_confirmed_client(&cr_ses->clientid);
1485 cs_slot = &conf->cl_cs_slot;
1486 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1487 if (status == nfserr_replay_cache) {
1488 dprintk("Got a create_session replay! seqid= %d\n",
1490 /* Return the cached reply status */
1491 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1493 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1494 status = nfserr_seq_misordered;
1495 dprintk("Sequence misordered!\n");
1496 dprintk("Expected seqid= %d but got seqid= %d\n",
1497 cs_slot->sl_seqid, cr_ses->seqid);
1500 cs_slot->sl_seqid++;
1501 } else if (unconf) {
1502 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1503 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1504 status = nfserr_clid_inuse;
1508 cs_slot = &unconf->cl_cs_slot;
1509 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1511 /* an unconfirmed replay returns misordered */
1512 status = nfserr_seq_misordered;
1516 cs_slot->sl_seqid++; /* from 0 to 1 */
1517 move_to_confirmed(unconf);
1519 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1520 unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1521 svc_xprt_get(rqstp->rq_xprt);
1523 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1525 unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1526 unconf->cl_cb_conn.cb_minorversion =
1527 cstate->minorversion;
1528 unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1529 unconf->cl_cb_seq_nr = 1;
1530 nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1534 status = nfserr_stale_clientid;
1539 * We do not support RDMA or persistent sessions
1541 cr_ses->flags &= ~SESSION4_PERSIST;
1542 cr_ses->flags &= ~SESSION4_RDMA;
1544 status = alloc_init_session(rqstp, conf, cr_ses);
1548 memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1549 NFS4_MAX_SESSIONID_LEN);
1550 cr_ses->seqid = cs_slot->sl_seqid;
1553 /* cache solo and embedded create sessions under the state lock */
1554 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1556 nfs4_unlock_state();
1557 dprintk("%s returns %d\n", __func__, ntohl(status));
1561 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1563 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1564 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1566 return argp->opcnt == resp->opcnt;
1569 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1573 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1577 nfsd4_destroy_session(struct svc_rqst *r,
1578 struct nfsd4_compound_state *cstate,
1579 struct nfsd4_destroy_session *sessionid)
1581 struct nfsd4_session *ses;
1582 u32 status = nfserr_badsession;
1585 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1586 * - Should we return nfserr_back_chan_busy if waiting for
1587 * callbacks on to-be-destroyed session?
1588 * - Do we need to clear any callback info from previous session?
1591 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1592 if (!nfsd4_last_compound_op(r))
1593 return nfserr_not_only_op;
1595 dump_sessionid(__func__, &sessionid->sessionid);
1596 spin_lock(&client_lock);
1597 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1599 spin_unlock(&client_lock);
1603 unhash_session(ses);
1604 spin_unlock(&client_lock);
1607 /* wait for callbacks */
1608 nfsd4_shutdown_callback(ses->se_client);
1609 nfs4_unlock_state();
1611 nfsd4_del_conns(ses);
1613 nfsd4_put_session(ses);
1616 dprintk("%s returns %d\n", __func__, ntohl(status));
1620 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_rqst *r, struct nfsd4_session *s)
1622 struct nfsd4_conn *c;
1624 list_for_each_entry(c, &s->se_conns, cn_persession) {
1625 if (c->cn_xprt == r->rq_xprt) {
1632 static void nfsd4_sequence_check_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses)
1634 struct nfs4_client *clp = ses->se_client;
1635 struct nfsd4_conn *c, *new = NULL;
1637 spin_lock(&clp->cl_lock);
1638 c = __nfsd4_find_conn(rqstp, ses);
1639 spin_unlock(&clp->cl_lock);
1643 new = alloc_conn(rqstp);
1645 spin_lock(&clp->cl_lock);
1646 c = __nfsd4_find_conn(rqstp, ses);
1648 spin_unlock(&clp->cl_lock);
1652 __nfsd4_hash_conn(new, ses);
1653 spin_unlock(&clp->cl_lock);
1654 nfsd4_register_conn(new);
1659 nfsd4_sequence(struct svc_rqst *rqstp,
1660 struct nfsd4_compound_state *cstate,
1661 struct nfsd4_sequence *seq)
1663 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1664 struct nfsd4_session *session;
1665 struct nfsd4_slot *slot;
1668 if (resp->opcnt != 1)
1669 return nfserr_sequence_pos;
1671 spin_lock(&client_lock);
1672 status = nfserr_badsession;
1673 session = find_in_sessionid_hashtbl(&seq->sessionid);
1677 status = nfserr_badslot;
1678 if (seq->slotid >= session->se_fchannel.maxreqs)
1681 slot = session->se_slots[seq->slotid];
1682 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1684 /* We do not negotiate the number of slots yet, so set the
1685 * maxslots to the session maxreqs which is used to encode
1686 * sr_highest_slotid and the sr_target_slot id to maxslots */
1687 seq->maxslots = session->se_fchannel.maxreqs;
1689 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1690 if (status == nfserr_replay_cache) {
1691 cstate->slot = slot;
1692 cstate->session = session;
1693 /* Return the cached reply status and set cstate->status
1694 * for nfsd4_proc_compound processing */
1695 status = nfsd4_replay_cache_entry(resp, seq);
1696 cstate->status = nfserr_replay_cache;
1702 nfsd4_sequence_check_conn(rqstp, session);
1704 /* Success! bump slot seqid */
1705 slot->sl_inuse = true;
1706 slot->sl_seqid = seq->seqid;
1707 slot->sl_cachethis = seq->cachethis;
1709 cstate->slot = slot;
1710 cstate->session = session;
1713 /* Hold a session reference until done processing the compound. */
1714 if (cstate->session) {
1715 nfsd4_get_session(cstate->session);
1716 atomic_inc(&session->se_client->cl_refcount);
1718 spin_unlock(&client_lock);
1719 dprintk("%s: return %d\n", __func__, ntohl(status));
1724 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1726 if (rc->rca_one_fs) {
1727 if (!cstate->current_fh.fh_dentry)
1728 return nfserr_nofilehandle;
1730 * We don't take advantage of the rca_one_fs case.
1731 * That's OK, it's optional, we can safely ignore it.
1736 if (is_client_expired(cstate->session->se_client)) {
1737 nfs4_unlock_state();
1739 * The following error isn't really legal.
1740 * But we only get here if the client just explicitly
1741 * destroyed the client. Surely it no longer cares what
1742 * error it gets back on an operation for the dead
1745 return nfserr_stale_clientid;
1747 nfsd4_create_clid_dir(cstate->session->se_client);
1748 nfs4_unlock_state();
1753 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1754 struct nfsd4_setclientid *setclid)
1756 struct sockaddr *sa = svc_addr(rqstp);
1757 struct xdr_netobj clname = {
1758 .len = setclid->se_namelen,
1759 .data = setclid->se_name,
1761 nfs4_verifier clverifier = setclid->se_verf;
1762 unsigned int strhashval;
1763 struct nfs4_client *conf, *unconf, *new;
1765 char dname[HEXDIR_LEN];
1767 if (!check_name(clname))
1768 return nfserr_inval;
1770 status = nfs4_make_rec_clidname(dname, &clname);
1775 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1776 * We get here on a DRC miss.
1779 strhashval = clientstr_hashval(dname);
1782 conf = find_confirmed_client_by_str(dname, strhashval, false);
1784 /* RFC 3530 14.2.33 CASE 0: */
1785 status = nfserr_clid_inuse;
1786 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1787 char addr_str[INET6_ADDRSTRLEN];
1788 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1790 dprintk("NFSD: setclientid: string in use by client "
1791 "at %s\n", addr_str);
1796 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1797 * has a description of SETCLIENTID request processing consisting
1798 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1800 unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1801 status = nfserr_resource;
1804 * RFC 3530 14.2.33 CASE 4:
1805 * placed first, because it is the normal case
1808 expire_client(unconf);
1809 new = create_client(clname, dname, rqstp, &clverifier);
1813 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1815 * RFC 3530 14.2.33 CASE 1:
1816 * probable callback update
1819 /* Note this is removing unconfirmed {*x***},
1820 * which is stronger than RFC recommended {vxc**}.
1821 * This has the advantage that there is at most
1822 * one {*x***} in either list at any time.
1824 expire_client(unconf);
1826 new = create_client(clname, dname, rqstp, &clverifier);
1829 copy_clid(new, conf);
1830 } else if (!unconf) {
1832 * RFC 3530 14.2.33 CASE 2:
1833 * probable client reboot; state will be removed if
1836 new = create_client(clname, dname, rqstp, &clverifier);
1842 * RFC 3530 14.2.33 CASE 3:
1843 * probable client reboot; state will be removed if
1846 expire_client(unconf);
1847 new = create_client(clname, dname, rqstp, &clverifier);
1852 gen_callback(new, setclid, rpc_get_scope_id(sa));
1853 add_to_unconfirmed(new, strhashval);
1854 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1855 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1856 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1859 nfs4_unlock_state();
1865 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1866 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1867 * bullets, labeled as CASE1 - CASE4 below.
1870 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1871 struct nfsd4_compound_state *cstate,
1872 struct nfsd4_setclientid_confirm *setclientid_confirm)
1874 struct sockaddr *sa = svc_addr(rqstp);
1875 struct nfs4_client *conf, *unconf;
1876 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1877 clientid_t * clid = &setclientid_confirm->sc_clientid;
1880 if (STALE_CLIENTID(clid))
1881 return nfserr_stale_clientid;
1883 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1884 * We get here on a DRC miss.
1889 conf = find_confirmed_client(clid);
1890 unconf = find_unconfirmed_client(clid);
1892 status = nfserr_clid_inuse;
1893 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1895 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1899 * section 14.2.34 of RFC 3530 has a description of
1900 * SETCLIENTID_CONFIRM request processing consisting
1901 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1903 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1905 * RFC 3530 14.2.34 CASE 1:
1908 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1909 status = nfserr_clid_inuse;
1911 atomic_set(&conf->cl_cb_set, 0);
1912 nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1913 expire_client(unconf);
1917 } else if (conf && !unconf) {
1919 * RFC 3530 14.2.34 CASE 2:
1920 * probable retransmitted request; play it safe and
1923 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1924 status = nfserr_clid_inuse;
1927 } else if (!conf && unconf
1928 && same_verf(&unconf->cl_confirm, &confirm)) {
1930 * RFC 3530 14.2.34 CASE 3:
1931 * Normal case; new or rebooted client:
1933 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1934 status = nfserr_clid_inuse;
1937 clientstr_hashval(unconf->cl_recdir);
1938 conf = find_confirmed_client_by_str(unconf->cl_recdir,
1941 nfsd4_remove_clid_dir(conf);
1942 expire_client(conf);
1944 move_to_confirmed(unconf);
1946 nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1949 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1950 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1953 * RFC 3530 14.2.34 CASE 4:
1954 * Client probably hasn't noticed that we rebooted yet.
1956 status = nfserr_stale_clientid;
1958 /* check that we have hit one of the cases...*/
1959 status = nfserr_clid_inuse;
1962 nfs4_unlock_state();
1966 /* OPEN Share state helper functions */
1967 static inline struct nfs4_file *
1968 alloc_init_file(struct inode *ino)
1970 struct nfs4_file *fp;
1971 unsigned int hashval = file_hashval(ino);
1973 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1975 atomic_set(&fp->fi_ref, 1);
1976 INIT_LIST_HEAD(&fp->fi_hash);
1977 INIT_LIST_HEAD(&fp->fi_stateids);
1978 INIT_LIST_HEAD(&fp->fi_delegations);
1979 fp->fi_inode = igrab(ino);
1980 fp->fi_id = current_fileid++;
1981 fp->fi_had_conflict = false;
1982 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
1983 memset(fp->fi_access, 0, sizeof(fp->fi_access));
1984 spin_lock(&recall_lock);
1985 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1986 spin_unlock(&recall_lock);
1993 nfsd4_free_slab(struct kmem_cache **slab)
1997 kmem_cache_destroy(*slab);
2002 nfsd4_free_slabs(void)
2004 nfsd4_free_slab(&stateowner_slab);
2005 nfsd4_free_slab(&file_slab);
2006 nfsd4_free_slab(&stateid_slab);
2007 nfsd4_free_slab(&deleg_slab);
2011 nfsd4_init_slabs(void)
2013 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2014 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2015 if (stateowner_slab == NULL)
2017 file_slab = kmem_cache_create("nfsd4_files",
2018 sizeof(struct nfs4_file), 0, 0, NULL);
2019 if (file_slab == NULL)
2021 stateid_slab = kmem_cache_create("nfsd4_stateids",
2022 sizeof(struct nfs4_stateid), 0, 0, NULL);
2023 if (stateid_slab == NULL)
2025 deleg_slab = kmem_cache_create("nfsd4_delegations",
2026 sizeof(struct nfs4_delegation), 0, 0, NULL);
2027 if (deleg_slab == NULL)
2032 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2037 nfs4_free_stateowner(struct kref *kref)
2039 struct nfs4_stateowner *sop =
2040 container_of(kref, struct nfs4_stateowner, so_ref);
2041 kfree(sop->so_owner.data);
2042 kmem_cache_free(stateowner_slab, sop);
2045 static inline struct nfs4_stateowner *
2046 alloc_stateowner(struct xdr_netobj *owner)
2048 struct nfs4_stateowner *sop;
2050 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2051 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2052 memcpy(sop->so_owner.data, owner->data, owner->len);
2053 sop->so_owner.len = owner->len;
2054 kref_init(&sop->so_ref);
2057 kmem_cache_free(stateowner_slab, sop);
2062 static struct nfs4_stateowner *
2063 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2064 struct nfs4_stateowner *sop;
2065 struct nfs4_replay *rp;
2066 unsigned int idhashval;
2068 if (!(sop = alloc_stateowner(&open->op_owner)))
2070 idhashval = ownerid_hashval(current_ownerid);
2071 INIT_LIST_HEAD(&sop->so_idhash);
2072 INIT_LIST_HEAD(&sop->so_strhash);
2073 INIT_LIST_HEAD(&sop->so_perclient);
2074 INIT_LIST_HEAD(&sop->so_stateids);
2075 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */
2076 INIT_LIST_HEAD(&sop->so_close_lru);
2078 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2079 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2080 list_add(&sop->so_perclient, &clp->cl_openowners);
2081 sop->so_is_open_owner = 1;
2082 sop->so_id = current_ownerid++;
2083 sop->so_client = clp;
2084 sop->so_seqid = open->op_seqid;
2085 sop->so_confirmed = 0;
2086 rp = &sop->so_replay;
2087 rp->rp_status = nfserr_serverfault;
2089 rp->rp_buf = rp->rp_ibuf;
2094 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2095 struct nfs4_stateowner *sop = open->op_stateowner;
2096 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2098 INIT_LIST_HEAD(&stp->st_hash);
2099 INIT_LIST_HEAD(&stp->st_perstateowner);
2100 INIT_LIST_HEAD(&stp->st_lockowners);
2101 INIT_LIST_HEAD(&stp->st_perfile);
2102 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2103 list_add(&stp->st_perstateowner, &sop->so_stateids);
2104 list_add(&stp->st_perfile, &fp->fi_stateids);
2105 stp->st_stateowner = sop;
2108 stp->st_stateid.si_boot = boot_time;
2109 stp->st_stateid.si_stateownerid = sop->so_id;
2110 stp->st_stateid.si_fileid = fp->fi_id;
2111 stp->st_stateid.si_generation = 0;
2112 stp->st_access_bmap = 0;
2113 stp->st_deny_bmap = 0;
2114 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2115 &stp->st_access_bmap);
2116 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2117 stp->st_openstp = NULL;
2121 move_to_close_lru(struct nfs4_stateowner *sop)
2123 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2125 list_move_tail(&sop->so_close_lru, &close_lru);
2126 sop->so_time = get_seconds();
2130 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2133 return (sop->so_owner.len == owner->len) &&
2134 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2135 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2138 static struct nfs4_stateowner *
2139 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2141 struct nfs4_stateowner *so = NULL;
2143 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2144 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2150 /* search file_hashtbl[] for file */
2151 static struct nfs4_file *
2152 find_file(struct inode *ino)
2154 unsigned int hashval = file_hashval(ino);
2155 struct nfs4_file *fp;
2157 spin_lock(&recall_lock);
2158 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2159 if (fp->fi_inode == ino) {
2161 spin_unlock(&recall_lock);
2165 spin_unlock(&recall_lock);
2169 static inline int access_valid(u32 x, u32 minorversion)
2171 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2173 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2175 x &= ~NFS4_SHARE_ACCESS_MASK;
2176 if (minorversion && x) {
2177 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2179 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2181 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2188 static inline int deny_valid(u32 x)
2190 /* Note: unlike access bits, deny bits may be zero. */
2191 return x <= NFS4_SHARE_DENY_BOTH;
2195 * Called to check deny when READ with all zero stateid or
2196 * WRITE with all zero or all one stateid
2199 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2201 struct inode *ino = current_fh->fh_dentry->d_inode;
2202 struct nfs4_file *fp;
2203 struct nfs4_stateid *stp;
2206 dprintk("NFSD: nfs4_share_conflict\n");
2208 fp = find_file(ino);
2211 ret = nfserr_locked;
2212 /* Search for conflicting share reservations */
2213 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2214 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2215 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2225 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2227 if (share_access & NFS4_SHARE_ACCESS_WRITE)
2228 nfs4_file_put_access(fp, O_WRONLY);
2229 if (share_access & NFS4_SHARE_ACCESS_READ)
2230 nfs4_file_put_access(fp, O_RDONLY);
2234 * Spawn a thread to perform a recall on the delegation represented
2235 * by the lease (file_lock)
2237 * Called from break_lease() with lock_kernel() held.
2238 * Note: we assume break_lease will only call this *once* for any given
2242 void nfsd_break_deleg_cb(struct file_lock *fl)
2244 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2246 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2250 /* We're assuming the state code never drops its reference
2251 * without first removing the lease. Since we're in this lease
2252 * callback (and since the lease code is serialized by the kernel
2253 * lock) we know the server hasn't removed the lease yet, we know
2254 * it's safe to take a reference: */
2255 atomic_inc(&dp->dl_count);
2257 spin_lock(&recall_lock);
2258 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2259 spin_unlock(&recall_lock);
2261 /* only place dl_time is set. protected by lock_kernel*/
2262 dp->dl_time = get_seconds();
2265 * We don't want the locks code to timeout the lease for us;
2266 * we'll remove it ourself if the delegation isn't returned
2269 fl->fl_break_time = 0;
2271 dp->dl_file->fi_had_conflict = true;
2272 nfsd4_cb_recall(dp);
2276 * The file_lock is being reapd.
2278 * Called by locks_free_lock() with lock_kernel() held.
2281 void nfsd_release_deleg_cb(struct file_lock *fl)
2283 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2285 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2287 if (!(fl->fl_flags & FL_LEASE) || !dp)
2289 dp->dl_flock = NULL;
2293 * Set the delegation file_lock back pointer.
2295 * Called from setlease() with lock_kernel() held.
2298 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2300 struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2302 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2309 * Called from setlease() with lock_kernel() held
2312 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2314 struct nfs4_delegation *onlistd =
2315 (struct nfs4_delegation *)onlist->fl_owner;
2316 struct nfs4_delegation *tryd =
2317 (struct nfs4_delegation *)try->fl_owner;
2319 if (onlist->fl_lmops != try->fl_lmops)
2322 return onlistd->dl_client == tryd->dl_client;
2327 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2330 return lease_modify(onlist, arg);
2335 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2336 .fl_break = nfsd_break_deleg_cb,
2337 .fl_release_private = nfsd_release_deleg_cb,
2338 .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2339 .fl_mylease = nfsd_same_client_deleg_cb,
2340 .fl_change = nfsd_change_deleg_cb,
2345 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2346 struct nfsd4_open *open)
2348 clientid_t *clientid = &open->op_clientid;
2349 struct nfs4_client *clp = NULL;
2350 unsigned int strhashval;
2351 struct nfs4_stateowner *sop = NULL;
2353 if (!check_name(open->op_owner))
2354 return nfserr_inval;
2356 if (STALE_CLIENTID(&open->op_clientid))
2357 return nfserr_stale_clientid;
2359 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2360 sop = find_openstateowner_str(strhashval, open);
2361 open->op_stateowner = sop;
2363 /* Make sure the client's lease hasn't expired. */
2364 clp = find_confirmed_client(clientid);
2366 return nfserr_expired;
2369 /* When sessions are used, skip open sequenceid processing */
2370 if (nfsd4_has_session(cstate))
2372 if (!sop->so_confirmed) {
2373 /* Replace unconfirmed owners without checking for replay. */
2374 clp = sop->so_client;
2375 release_openowner(sop);
2376 open->op_stateowner = NULL;
2379 if (open->op_seqid == sop->so_seqid - 1) {
2380 if (sop->so_replay.rp_buflen)
2381 return nfserr_replay_me;
2382 /* The original OPEN failed so spectacularly
2383 * that we don't even have replay data saved!
2384 * Therefore, we have no choice but to continue
2385 * processing this OPEN; presumably, we'll
2386 * fail again for the same reason.
2388 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2391 if (open->op_seqid != sop->so_seqid)
2392 return nfserr_bad_seqid;
2394 if (open->op_stateowner == NULL) {
2395 sop = alloc_init_open_stateowner(strhashval, clp, open);
2397 return nfserr_resource;
2398 open->op_stateowner = sop;
2400 list_del_init(&sop->so_close_lru);
2401 renew_client(sop->so_client);
2405 static inline __be32
2406 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2408 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2409 return nfserr_openmode;
2414 static struct nfs4_delegation *
2415 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2417 struct nfs4_delegation *dp;
2419 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2420 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2426 int share_access_to_flags(u32 share_access)
2428 share_access &= ~NFS4_SHARE_WANT_MASK;
2430 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2434 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2435 struct nfs4_delegation **dp)
2438 __be32 status = nfserr_bad_stateid;
2440 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2443 flags = share_access_to_flags(open->op_share_access);
2444 status = nfs4_check_delegmode(*dp, flags);
2448 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2452 open->op_stateowner->so_confirmed = 1;
2457 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2459 struct nfs4_stateid *local;
2460 __be32 status = nfserr_share_denied;
2461 struct nfs4_stateowner *sop = open->op_stateowner;
2463 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2464 /* ignore lock owners */
2465 if (local->st_stateowner->so_is_open_owner == 0)
2467 /* remember if we have seen this open owner */
2468 if (local->st_stateowner == sop)
2470 /* check for conflicting share reservations */
2471 if (!test_share(local, open))
2479 static inline struct nfs4_stateid *
2480 nfs4_alloc_stateid(void)
2482 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2485 static inline int nfs4_access_to_access(u32 nfs4_access)
2489 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2490 flags |= NFSD_MAY_READ;
2491 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2492 flags |= NFSD_MAY_WRITE;
2496 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2497 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2500 int oflag = nfs4_access_to_omode(nfs4_access);
2501 int access = nfs4_access_to_access(nfs4_access);
2503 if (!fp->fi_fds[oflag]) {
2504 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2505 &fp->fi_fds[oflag]);
2506 if (status == nfserr_dropit)
2507 status = nfserr_jukebox;
2511 nfs4_file_get_access(fp, oflag);
2517 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2518 struct nfs4_file *fp, struct svc_fh *cur_fh,
2519 struct nfsd4_open *open)
2521 struct nfs4_stateid *stp;
2524 stp = nfs4_alloc_stateid();
2526 return nfserr_resource;
2528 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2530 kmem_cache_free(stateid_slab, stp);
2537 static inline __be32
2538 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2539 struct nfsd4_open *open)
2541 struct iattr iattr = {
2542 .ia_valid = ATTR_SIZE,
2545 if (!open->op_truncate)
2547 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2548 return nfserr_inval;
2549 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2553 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2555 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2559 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2561 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2565 status = nfsd4_truncate(rqstp, cur_fh, open);
2568 int oflag = nfs4_access_to_omode(new_access);
2569 nfs4_file_put_access(fp, oflag);
2573 /* remember the open */
2574 __set_bit(op_share_access, &stp->st_access_bmap);
2575 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2582 nfs4_set_claim_prev(struct nfsd4_open *open)
2584 open->op_stateowner->so_confirmed = 1;
2585 open->op_stateowner->so_client->cl_firststate = 1;
2589 * Attempt to hand out a delegation.
2592 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2594 struct nfs4_delegation *dp;
2595 struct nfs4_stateowner *sop = stp->st_stateowner;
2596 int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2597 struct file_lock fl, *flp = &fl;
2598 int status, flag = 0;
2600 flag = NFS4_OPEN_DELEGATE_NONE;
2601 open->op_recall = 0;
2602 switch (open->op_claim_type) {
2603 case NFS4_OPEN_CLAIM_PREVIOUS:
2605 open->op_recall = 1;
2606 flag = open->op_delegate_type;
2607 if (flag == NFS4_OPEN_DELEGATE_NONE)
2610 case NFS4_OPEN_CLAIM_NULL:
2611 /* Let's not give out any delegations till everyone's
2612 * had the chance to reclaim theirs.... */
2613 if (locks_in_grace())
2615 if (!cb_up || !sop->so_confirmed)
2617 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2618 flag = NFS4_OPEN_DELEGATE_WRITE;
2620 flag = NFS4_OPEN_DELEGATE_READ;
2626 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2628 flag = NFS4_OPEN_DELEGATE_NONE;
2631 locks_init_lock(&fl);
2632 fl.fl_lmops = &nfsd_lease_mng_ops;
2633 fl.fl_flags = FL_LEASE;
2634 fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2635 fl.fl_end = OFFSET_MAX;
2636 fl.fl_owner = (fl_owner_t)dp;
2637 fl.fl_file = find_readable_file(stp->st_file);
2638 BUG_ON(!fl.fl_file);
2639 fl.fl_pid = current->tgid;
2641 /* vfs_setlease checks to see if delegation should be handed out.
2642 * the lock_manager callbacks fl_mylease and fl_change are used
2644 if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
2645 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2646 unhash_delegation(dp);
2647 flag = NFS4_OPEN_DELEGATE_NONE;
2651 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2653 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2654 STATEID_VAL(&dp->dl_stateid));
2656 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2657 && flag == NFS4_OPEN_DELEGATE_NONE
2658 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2659 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2660 open->op_delegate_type = flag;
2664 * called with nfs4_lock_state() held.
2667 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2669 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2670 struct nfs4_file *fp = NULL;
2671 struct inode *ino = current_fh->fh_dentry->d_inode;
2672 struct nfs4_stateid *stp = NULL;
2673 struct nfs4_delegation *dp = NULL;
2676 status = nfserr_inval;
2677 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2678 || !deny_valid(open->op_share_deny))
2681 * Lookup file; if found, lookup stateid and check open request,
2682 * and check for delegations in the process of being recalled.
2683 * If not found, create the nfs4_file struct
2685 fp = find_file(ino);
2687 if ((status = nfs4_check_open(fp, open, &stp)))
2689 status = nfs4_check_deleg(fp, open, &dp);
2693 status = nfserr_bad_stateid;
2694 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2696 status = nfserr_resource;
2697 fp = alloc_init_file(ino);
2703 * OPEN the file, or upgrade an existing OPEN.
2704 * If truncate fails, the OPEN fails.
2707 /* Stateid was found, this is an OPEN upgrade */
2708 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2711 update_stateid(&stp->st_stateid);
2713 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2716 init_stateid(stp, fp, open);
2717 status = nfsd4_truncate(rqstp, current_fh, open);
2719 release_open_stateid(stp);
2722 if (nfsd4_has_session(&resp->cstate))
2723 update_stateid(&stp->st_stateid);
2725 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2727 if (nfsd4_has_session(&resp->cstate))
2728 open->op_stateowner->so_confirmed = 1;
2731 * Attempt to hand out a delegation. No error return, because the
2732 * OPEN succeeds even if we fail.
2734 nfs4_open_delegation(current_fh, open, stp);
2738 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2739 STATEID_VAL(&stp->st_stateid));
2743 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2744 nfs4_set_claim_prev(open);
2746 * To finish the open response, we just need to set the rflags.
2748 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2749 if (!open->op_stateowner->so_confirmed &&
2750 !nfsd4_has_session(&resp->cstate))
2751 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2757 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2760 struct nfs4_client *clp;
2764 dprintk("process_renew(%08x/%08x): starting\n",
2765 clid->cl_boot, clid->cl_id);
2766 status = nfserr_stale_clientid;
2767 if (STALE_CLIENTID(clid))
2769 clp = find_confirmed_client(clid);
2770 status = nfserr_expired;
2772 /* We assume the client took too long to RENEW. */
2773 dprintk("nfsd4_renew: clientid not found!\n");
2777 status = nfserr_cb_path_down;
2778 if (!list_empty(&clp->cl_delegations)
2779 && !atomic_read(&clp->cl_cb_set))
2783 nfs4_unlock_state();
2787 struct lock_manager nfsd4_manager = {
2791 nfsd4_end_grace(void)
2793 dprintk("NFSD: end of grace period\n");
2794 nfsd4_recdir_purge_old();
2795 locks_end_grace(&nfsd4_manager);
2797 * Now that every NFSv4 client has had the chance to recover and
2798 * to see the (possibly new, possibly shorter) lease time, we
2799 * can safely set the next grace time to the current lease time:
2801 nfsd4_grace = nfsd4_lease;
2805 nfs4_laundromat(void)
2807 struct nfs4_client *clp;
2808 struct nfs4_stateowner *sop;
2809 struct nfs4_delegation *dp;
2810 struct list_head *pos, *next, reaplist;
2811 time_t cutoff = get_seconds() - nfsd4_lease;
2812 time_t t, clientid_val = nfsd4_lease;
2813 time_t u, test_val = nfsd4_lease;
2817 dprintk("NFSD: laundromat service - starting\n");
2818 if (locks_in_grace())
2820 INIT_LIST_HEAD(&reaplist);
2821 spin_lock(&client_lock);
2822 list_for_each_safe(pos, next, &client_lru) {
2823 clp = list_entry(pos, struct nfs4_client, cl_lru);
2824 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2825 t = clp->cl_time - cutoff;
2826 if (clientid_val > t)
2830 if (atomic_read(&clp->cl_refcount)) {
2831 dprintk("NFSD: client in use (clientid %08x)\n",
2832 clp->cl_clientid.cl_id);
2835 unhash_client_locked(clp);
2836 list_add(&clp->cl_lru, &reaplist);
2838 spin_unlock(&client_lock);
2839 list_for_each_safe(pos, next, &reaplist) {
2840 clp = list_entry(pos, struct nfs4_client, cl_lru);
2841 dprintk("NFSD: purging unused client (clientid %08x)\n",
2842 clp->cl_clientid.cl_id);
2843 nfsd4_remove_clid_dir(clp);
2846 spin_lock(&recall_lock);
2847 list_for_each_safe(pos, next, &del_recall_lru) {
2848 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2849 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2850 u = dp->dl_time - cutoff;
2855 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2857 list_move(&dp->dl_recall_lru, &reaplist);
2859 spin_unlock(&recall_lock);
2860 list_for_each_safe(pos, next, &reaplist) {
2861 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2862 list_del_init(&dp->dl_recall_lru);
2863 unhash_delegation(dp);
2865 test_val = nfsd4_lease;
2866 list_for_each_safe(pos, next, &close_lru) {
2867 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2868 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2869 u = sop->so_time - cutoff;
2874 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2876 release_openowner(sop);
2878 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2879 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2880 nfs4_unlock_state();
2881 return clientid_val;
2884 static struct workqueue_struct *laundry_wq;
2885 static void laundromat_main(struct work_struct *);
2886 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2889 laundromat_main(struct work_struct *not_used)
2893 t = nfs4_laundromat();
2894 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2895 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2898 static struct nfs4_stateowner *
2899 search_close_lru(u32 st_id, int flags)
2901 struct nfs4_stateowner *local = NULL;
2903 if (flags & CLOSE_STATE) {
2904 list_for_each_entry(local, &close_lru, so_close_lru) {
2905 if (local->so_id == st_id)
2913 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2915 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2919 STALE_STATEID(stateid_t *stateid)
2921 if (stateid->si_boot == boot_time)
2923 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2924 STATEID_VAL(stateid));
2929 access_permit_read(unsigned long access_bmap)
2931 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2932 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2933 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2937 access_permit_write(unsigned long access_bmap)
2939 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2940 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2944 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2946 __be32 status = nfserr_openmode;
2948 /* For lock stateid's, we test the parent open, not the lock: */
2949 if (stp->st_openstp)
2950 stp = stp->st_openstp;
2951 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2953 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2960 static inline __be32
2961 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2963 if (ONE_STATEID(stateid) && (flags & RD_STATE))
2965 else if (locks_in_grace()) {
2966 /* Answer in remaining cases depends on existance of
2967 * conflicting state; so we must wait out the grace period. */
2968 return nfserr_grace;
2969 } else if (flags & WR_STATE)
2970 return nfs4_share_conflict(current_fh,
2971 NFS4_SHARE_DENY_WRITE);
2972 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2973 return nfs4_share_conflict(current_fh,
2974 NFS4_SHARE_DENY_READ);
2978 * Allow READ/WRITE during grace period on recovered state only for files
2979 * that are not able to provide mandatory locking.
2982 grace_disallows_io(struct inode *inode)
2984 return locks_in_grace() && mandatory_lock(inode);
2987 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2990 * When sessions are used the stateid generation number is ignored
2993 if ((flags & HAS_SESSION) && in->si_generation == 0)
2996 /* If the client sends us a stateid from the future, it's buggy: */
2997 if (in->si_generation > ref->si_generation)
2998 return nfserr_bad_stateid;
3000 * The following, however, can happen. For example, if the
3001 * client sends an open and some IO at the same time, the open
3002 * may bump si_generation while the IO is still in flight.
3003 * Thanks to hard links and renames, the client never knows what
3004 * file an open will affect. So it could avoid that situation
3005 * only by serializing all opens and IO from the same open
3006 * owner. To recover from the old_stateid error, the client
3007 * will just have to retry the IO:
3009 if (in->si_generation < ref->si_generation)
3010 return nfserr_old_stateid;
3015 static int is_delegation_stateid(stateid_t *stateid)
3017 return stateid->si_fileid == 0;
3021 * Checks for stateid operations
3024 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3025 stateid_t *stateid, int flags, struct file **filpp)
3027 struct nfs4_stateid *stp = NULL;
3028 struct nfs4_delegation *dp = NULL;
3029 struct svc_fh *current_fh = &cstate->current_fh;
3030 struct inode *ino = current_fh->fh_dentry->d_inode;
3036 if (grace_disallows_io(ino))
3037 return nfserr_grace;
3039 if (nfsd4_has_session(cstate))
3040 flags |= HAS_SESSION;
3042 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3043 return check_special_stateids(current_fh, stateid, flags);
3045 status = nfserr_stale_stateid;
3046 if (STALE_STATEID(stateid))
3050 * We assume that any stateid that has the current boot time,
3051 * but that we can't find, is expired:
3053 status = nfserr_expired;
3054 if (is_delegation_stateid(stateid)) {
3055 dp = find_delegation_stateid(ino, stateid);
3058 status = check_stateid_generation(stateid, &dp->dl_stateid,
3062 status = nfs4_check_delegmode(dp, flags);
3065 renew_client(dp->dl_client);
3067 *filpp = find_readable_file(dp->dl_file);
3069 } else { /* open or lock stateid */
3070 stp = find_stateid(stateid, flags);
3073 status = nfserr_bad_stateid;
3074 if (nfs4_check_fh(current_fh, stp))
3076 if (!stp->st_stateowner->so_confirmed)
3078 status = check_stateid_generation(stateid, &stp->st_stateid,
3082 status = nfs4_check_openmode(stp, flags);
3085 renew_client(stp->st_stateowner->so_client);
3087 if (flags & RD_STATE)
3088 *filpp = find_readable_file(stp->st_file);
3090 *filpp = find_writeable_file(stp->st_file);
3101 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3102 RD_STATE : WR_STATE;
3106 * Checks for sequence id mutating operations.
3109 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3110 stateid_t *stateid, int flags,
3111 struct nfs4_stateowner **sopp,
3112 struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3114 struct nfs4_stateid *stp;
3115 struct nfs4_stateowner *sop;
3116 struct svc_fh *current_fh = &cstate->current_fh;
3119 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3120 seqid, STATEID_VAL(stateid));
3125 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3126 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3127 return nfserr_bad_stateid;
3130 if (STALE_STATEID(stateid))
3131 return nfserr_stale_stateid;
3133 if (nfsd4_has_session(cstate))
3134 flags |= HAS_SESSION;
3137 * We return BAD_STATEID if filehandle doesn't match stateid,
3138 * the confirmed flag is incorrecly set, or the generation
3139 * number is incorrect.
3141 stp = find_stateid(stateid, flags);
3144 * Also, we should make sure this isn't just the result of
3147 sop = search_close_lru(stateid->si_stateownerid, flags);
3148 /* It's not stale; let's assume it's expired: */
3150 return nfserr_expired;
3156 *sopp = sop = stp->st_stateowner;
3159 clientid_t *lockclid = &lock->v.new.clientid;
3160 struct nfs4_client *clp = sop->so_client;
3164 lkflg = setlkflg(lock->lk_type);
3166 if (lock->lk_is_new) {
3167 if (!sop->so_is_open_owner)
3168 return nfserr_bad_stateid;
3169 if (!(flags & HAS_SESSION) &&
3170 !same_clid(&clp->cl_clientid, lockclid))
3171 return nfserr_bad_stateid;
3172 /* stp is the open stateid */
3173 status = nfs4_check_openmode(stp, lkflg);
3177 /* stp is the lock stateid */
3178 status = nfs4_check_openmode(stp->st_openstp, lkflg);
3184 if (nfs4_check_fh(current_fh, stp)) {
3185 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3186 return nfserr_bad_stateid;
3190 * We now validate the seqid and stateid generation numbers.
3191 * For the moment, we ignore the possibility of
3192 * generation number wraparound.
3194 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3197 if (sop->so_confirmed && flags & CONFIRM) {
3198 dprintk("NFSD: preprocess_seqid_op: expected"
3199 " unconfirmed stateowner!\n");
3200 return nfserr_bad_stateid;
3202 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3203 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3204 " confirmed yet!\n");
3205 return nfserr_bad_stateid;
3207 status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3210 renew_client(sop->so_client);
3214 if (seqid == sop->so_seqid - 1) {
3215 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3216 /* indicate replay to calling function */
3217 return nfserr_replay_me;
3219 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3220 sop->so_seqid, seqid);
3222 return nfserr_bad_seqid;
3226 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3227 struct nfsd4_open_confirm *oc)
3230 struct nfs4_stateowner *sop;
3231 struct nfs4_stateid *stp;
3233 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3234 (int)cstate->current_fh.fh_dentry->d_name.len,
3235 cstate->current_fh.fh_dentry->d_name.name);
3237 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3243 if ((status = nfs4_preprocess_seqid_op(cstate,
3244 oc->oc_seqid, &oc->oc_req_stateid,
3245 CONFIRM | OPEN_STATE,
3246 &oc->oc_stateowner, &stp, NULL)))
3249 sop = oc->oc_stateowner;
3250 sop->so_confirmed = 1;
3251 update_stateid(&stp->st_stateid);
3252 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3253 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3254 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3256 nfsd4_create_clid_dir(sop->so_client);
3258 if (oc->oc_stateowner) {
3259 nfs4_get_stateowner(oc->oc_stateowner);
3260 cstate->replay_owner = oc->oc_stateowner;
3262 nfs4_unlock_state();
3268 * unset all bits in union bitmap (bmap) that
3269 * do not exist in share (from successful OPEN_DOWNGRADE)
3272 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3275 for (i = 1; i < 4; i++) {
3276 if ((i & access) != i)
3277 __clear_bit(i, bmap);
3282 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3285 for (i = 0; i < 4; i++) {
3286 if ((i & deny) != i)
3287 __clear_bit(i, bmap);
3292 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3293 struct nfsd4_compound_state *cstate,
3294 struct nfsd4_open_downgrade *od)
3297 struct nfs4_stateid *stp;
3298 unsigned int share_access;
3300 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3301 (int)cstate->current_fh.fh_dentry->d_name.len,
3302 cstate->current_fh.fh_dentry->d_name.name);
3304 if (!access_valid(od->od_share_access, cstate->minorversion)
3305 || !deny_valid(od->od_share_deny))
3306 return nfserr_inval;
3309 if ((status = nfs4_preprocess_seqid_op(cstate,
3313 &od->od_stateowner, &stp, NULL)))
3316 status = nfserr_inval;
3317 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3318 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3319 stp->st_access_bmap, od->od_share_access);
3322 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3323 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3324 stp->st_deny_bmap, od->od_share_deny);
3327 set_access(&share_access, stp->st_access_bmap);
3328 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3330 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3331 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3333 update_stateid(&stp->st_stateid);
3334 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3337 if (od->od_stateowner) {
3338 nfs4_get_stateowner(od->od_stateowner);
3339 cstate->replay_owner = od->od_stateowner;
3341 nfs4_unlock_state();
3346 * nfs4_unlock_state() called after encode
3349 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3350 struct nfsd4_close *close)
3353 struct nfs4_stateid *stp;
3355 dprintk("NFSD: nfsd4_close on file %.*s\n",
3356 (int)cstate->current_fh.fh_dentry->d_name.len,
3357 cstate->current_fh.fh_dentry->d_name.name);
3360 /* check close_lru for replay */
3361 if ((status = nfs4_preprocess_seqid_op(cstate,
3364 OPEN_STATE | CLOSE_STATE,
3365 &close->cl_stateowner, &stp, NULL)))
3368 update_stateid(&stp->st_stateid);
3369 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3371 /* release_stateid() calls nfsd_close() if needed */
3372 release_open_stateid(stp);
3374 /* place unused nfs4_stateowners on so_close_lru list to be
3375 * released by the laundromat service after the lease period
3376 * to enable us to handle CLOSE replay
3378 if (list_empty(&close->cl_stateowner->so_stateids))
3379 move_to_close_lru(close->cl_stateowner);
3381 if (close->cl_stateowner) {
3382 nfs4_get_stateowner(close->cl_stateowner);
3383 cstate->replay_owner = close->cl_stateowner;
3385 nfs4_unlock_state();
3390 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3391 struct nfsd4_delegreturn *dr)
3393 struct nfs4_delegation *dp;
3394 stateid_t *stateid = &dr->dr_stateid;
3395 struct inode *inode;
3399 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3401 inode = cstate->current_fh.fh_dentry->d_inode;
3403 if (nfsd4_has_session(cstate))
3404 flags |= HAS_SESSION;
3406 status = nfserr_bad_stateid;
3407 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3409 status = nfserr_stale_stateid;
3410 if (STALE_STATEID(stateid))
3412 status = nfserr_bad_stateid;
3413 if (!is_delegation_stateid(stateid))
3415 status = nfserr_expired;
3416 dp = find_delegation_stateid(inode, stateid);
3419 status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3422 renew_client(dp->dl_client);
3424 unhash_delegation(dp);
3426 nfs4_unlock_state();
3433 * Lock owner state (byte-range locks)
3435 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3436 #define LOCK_HASH_BITS 8
3437 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3438 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3441 end_offset(u64 start, u64 len)
3446 return end >= start ? end: NFS4_MAX_UINT64;
3449 /* last octet in a range */
3451 last_byte_offset(u64 start, u64 len)
3457 return end > start ? end - 1: NFS4_MAX_UINT64;
3460 #define lockownerid_hashval(id) \
3461 ((id) & LOCK_HASH_MASK)
3463 static inline unsigned int
3464 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3465 struct xdr_netobj *ownername)
3467 return (file_hashval(inode) + cl_id
3468 + opaque_hashval(ownername->data, ownername->len))
3472 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3473 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3474 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3476 static struct nfs4_stateid *
3477 find_stateid(stateid_t *stid, int flags)
3479 struct nfs4_stateid *local;
3480 u32 st_id = stid->si_stateownerid;
3481 u32 f_id = stid->si_fileid;
3482 unsigned int hashval;
3484 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3485 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3486 hashval = stateid_hashval(st_id, f_id);
3487 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3488 if ((local->st_stateid.si_stateownerid == st_id) &&
3489 (local->st_stateid.si_fileid == f_id))
3494 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3495 hashval = stateid_hashval(st_id, f_id);
3496 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3497 if ((local->st_stateid.si_stateownerid == st_id) &&
3498 (local->st_stateid.si_fileid == f_id))
3505 static struct nfs4_delegation *
3506 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3508 struct nfs4_file *fp;
3509 struct nfs4_delegation *dl;
3511 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3514 fp = find_file(ino);
3517 dl = find_delegation_file(fp, stid);
3523 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3524 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3525 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3526 * locking, this prevents us from being completely protocol-compliant. The
3527 * real solution to this problem is to start using unsigned file offsets in
3528 * the VFS, but this is a very deep change!
3531 nfs4_transform_lock_offset(struct file_lock *lock)
3533 if (lock->fl_start < 0)
3534 lock->fl_start = OFFSET_MAX;
3535 if (lock->fl_end < 0)
3536 lock->fl_end = OFFSET_MAX;
3539 /* Hack!: For now, we're defining this just so we can use a pointer to it
3540 * as a unique cookie to identify our (NFSv4's) posix locks. */
3541 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3545 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3547 struct nfs4_stateowner *sop;
3549 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3550 sop = (struct nfs4_stateowner *) fl->fl_owner;
3551 kref_get(&sop->so_ref);
3553 deny->ld_clientid = sop->so_client->cl_clientid;
3555 deny->ld_sop = NULL;
3556 deny->ld_clientid.cl_boot = 0;
3557 deny->ld_clientid.cl_id = 0;
3559 deny->ld_start = fl->fl_start;
3560 deny->ld_length = NFS4_MAX_UINT64;
3561 if (fl->fl_end != NFS4_MAX_UINT64)
3562 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3563 deny->ld_type = NFS4_READ_LT;
3564 if (fl->fl_type != F_RDLCK)
3565 deny->ld_type = NFS4_WRITE_LT;
3568 static struct nfs4_stateowner *
3569 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3570 struct xdr_netobj *owner)
3572 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3573 struct nfs4_stateowner *op;
3575 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3576 if (same_owner_str(op, owner, clid))
3583 * Alloc a lock owner structure.
3584 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3587 * strhashval = lock_ownerstr_hashval
3590 static struct nfs4_stateowner *
3591 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3592 struct nfs4_stateowner *sop;
3593 struct nfs4_replay *rp;
3594 unsigned int idhashval;
3596 if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3598 idhashval = lockownerid_hashval(current_ownerid);
3599 INIT_LIST_HEAD(&sop->so_idhash);
3600 INIT_LIST_HEAD(&sop->so_strhash);
3601 INIT_LIST_HEAD(&sop->so_perclient);
3602 INIT_LIST_HEAD(&sop->so_stateids);
3603 INIT_LIST_HEAD(&sop->so_perstateid);
3604 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3606 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3607 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3608 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3609 sop->so_is_open_owner = 0;
3610 sop->so_id = current_ownerid++;
3611 sop->so_client = clp;
3612 /* It is the openowner seqid that will be incremented in encode in the
3613 * case of new lockowners; so increment the lock seqid manually: */
3614 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3615 sop->so_confirmed = 1;
3616 rp = &sop->so_replay;
3617 rp->rp_status = nfserr_serverfault;
3619 rp->rp_buf = rp->rp_ibuf;
3623 static struct nfs4_stateid *
3624 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3626 struct nfs4_stateid *stp;
3627 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3629 stp = nfs4_alloc_stateid();
3632 INIT_LIST_HEAD(&stp->st_hash);
3633 INIT_LIST_HEAD(&stp->st_perfile);
3634 INIT_LIST_HEAD(&stp->st_perstateowner);
3635 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3636 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3637 list_add(&stp->st_perfile, &fp->fi_stateids);
3638 list_add(&stp->st_perstateowner, &sop->so_stateids);
3639 stp->st_stateowner = sop;
3642 stp->st_stateid.si_boot = boot_time;
3643 stp->st_stateid.si_stateownerid = sop->so_id;
3644 stp->st_stateid.si_fileid = fp->fi_id;
3645 stp->st_stateid.si_generation = 0;
3646 stp->st_deny_bmap = open_stp->st_deny_bmap;
3647 stp->st_openstp = open_stp;
3654 check_lock_length(u64 offset, u64 length)
3656 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3657 LOFF_OVERFLOW(offset, length)));
3664 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3665 struct nfsd4_lock *lock)
3667 struct nfs4_stateowner *open_sop = NULL;
3668 struct nfs4_stateowner *lock_sop = NULL;
3669 struct nfs4_stateid *lock_stp;
3670 struct nfs4_file *fp;
3671 struct file *filp = NULL;
3672 struct file_lock file_lock;
3673 struct file_lock conflock;
3675 unsigned int strhashval;
3679 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3680 (long long) lock->lk_offset,
3681 (long long) lock->lk_length);
3683 if (check_lock_length(lock->lk_offset, lock->lk_length))
3684 return nfserr_inval;
3686 if ((status = fh_verify(rqstp, &cstate->current_fh,
3687 S_IFREG, NFSD_MAY_LOCK))) {
3688 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3694 if (lock->lk_is_new) {
3696 * Client indicates that this is a new lockowner.
3697 * Use open owner and open stateid to create lock owner and
3700 struct nfs4_stateid *open_stp = NULL;
3702 status = nfserr_stale_clientid;
3703 if (!nfsd4_has_session(cstate) &&
3704 STALE_CLIENTID(&lock->lk_new_clientid))
3707 /* validate and update open stateid and open seqid */
3708 status = nfs4_preprocess_seqid_op(cstate,
3709 lock->lk_new_open_seqid,
3710 &lock->lk_new_open_stateid,
3712 &lock->lk_replay_owner, &open_stp,
3716 open_sop = lock->lk_replay_owner;
3717 /* create lockowner and lock stateid */
3718 fp = open_stp->st_file;
3719 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3720 open_sop->so_client->cl_clientid.cl_id,
3721 &lock->v.new.owner);
3722 /* XXX: Do we need to check for duplicate stateowners on
3723 * the same file, or should they just be allowed (and
3724 * create new stateids)? */
3725 status = nfserr_resource;
3726 lock_sop = alloc_init_lock_stateowner(strhashval,
3727 open_sop->so_client, open_stp, lock);
3728 if (lock_sop == NULL)
3730 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3731 if (lock_stp == NULL)
3734 /* lock (lock owner + lock stateid) already exists */
3735 status = nfs4_preprocess_seqid_op(cstate,
3736 lock->lk_old_lock_seqid,
3737 &lock->lk_old_lock_stateid,
3739 &lock->lk_replay_owner, &lock_stp, lock);
3742 lock_sop = lock->lk_replay_owner;
3743 fp = lock_stp->st_file;
3745 /* lock->lk_replay_owner and lock_stp have been created or found */
3747 status = nfserr_grace;
3748 if (locks_in_grace() && !lock->lk_reclaim)
3750 status = nfserr_no_grace;
3751 if (!locks_in_grace() && lock->lk_reclaim)
3754 locks_init_lock(&file_lock);
3755 switch (lock->lk_type) {
3758 if (find_readable_file(lock_stp->st_file)) {
3759 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3760 filp = find_readable_file(lock_stp->st_file);
3762 file_lock.fl_type = F_RDLCK;
3766 case NFS4_WRITEW_LT:
3767 if (find_writeable_file(lock_stp->st_file)) {
3768 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3769 filp = find_writeable_file(lock_stp->st_file);
3771 file_lock.fl_type = F_WRLCK;
3775 status = nfserr_inval;
3779 status = nfserr_openmode;
3782 file_lock.fl_owner = (fl_owner_t)lock_sop;
3783 file_lock.fl_pid = current->tgid;
3784 file_lock.fl_file = filp;
3785 file_lock.fl_flags = FL_POSIX;
3786 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3788 file_lock.fl_start = lock->lk_offset;
3789 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3790 nfs4_transform_lock_offset(&file_lock);
3793 * Try to lock the file in the VFS.
3794 * Note: locks.c uses the BKL to protect the inode's lock list.
3797 err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3799 case 0: /* success! */
3800 update_stateid(&lock_stp->st_stateid);
3801 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3805 case (EAGAIN): /* conflock holds conflicting lock */
3806 status = nfserr_denied;
3807 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3808 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3811 status = nfserr_deadlock;
3814 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3815 status = nfserr_resource;
3819 if (status && lock->lk_is_new && lock_sop)
3820 release_lockowner(lock_sop);
3821 if (lock->lk_replay_owner) {
3822 nfs4_get_stateowner(lock->lk_replay_owner);
3823 cstate->replay_owner = lock->lk_replay_owner;
3825 nfs4_unlock_state();
3830 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3831 * so we do a temporary open here just to get an open file to pass to
3832 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3835 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3840 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3843 err = vfs_test_lock(file, lock);
3852 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3853 struct nfsd4_lockt *lockt)
3855 struct inode *inode;
3856 struct file_lock file_lock;
3860 if (locks_in_grace())
3861 return nfserr_grace;
3863 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3864 return nfserr_inval;
3866 lockt->lt_stateowner = NULL;
3869 status = nfserr_stale_clientid;
3870 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3873 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3874 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3875 if (status == nfserr_symlink)
3876 status = nfserr_inval;
3880 inode = cstate->current_fh.fh_dentry->d_inode;
3881 locks_init_lock(&file_lock);
3882 switch (lockt->lt_type) {
3885 file_lock.fl_type = F_RDLCK;
3888 case NFS4_WRITEW_LT:
3889 file_lock.fl_type = F_WRLCK;
3892 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3893 status = nfserr_inval;
3897 lockt->lt_stateowner = find_lockstateowner_str(inode,
3898 &lockt->lt_clientid, &lockt->lt_owner);
3899 if (lockt->lt_stateowner)
3900 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3901 file_lock.fl_pid = current->tgid;
3902 file_lock.fl_flags = FL_POSIX;
3904 file_lock.fl_start = lockt->lt_offset;
3905 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3907 nfs4_transform_lock_offset(&file_lock);
3910 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3912 status = nfserrno(error);
3915 if (file_lock.fl_type != F_UNLCK) {
3916 status = nfserr_denied;
3917 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3920 nfs4_unlock_state();
3925 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3926 struct nfsd4_locku *locku)
3928 struct nfs4_stateid *stp;
3929 struct file *filp = NULL;
3930 struct file_lock file_lock;
3934 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3935 (long long) locku->lu_offset,
3936 (long long) locku->lu_length);
3938 if (check_lock_length(locku->lu_offset, locku->lu_length))
3939 return nfserr_inval;
3943 if ((status = nfs4_preprocess_seqid_op(cstate,
3947 &locku->lu_stateowner, &stp, NULL)))
3950 filp = find_any_file(stp->st_file);
3952 status = nfserr_lock_range;
3956 locks_init_lock(&file_lock);
3957 file_lock.fl_type = F_UNLCK;
3958 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3959 file_lock.fl_pid = current->tgid;
3960 file_lock.fl_file = filp;
3961 file_lock.fl_flags = FL_POSIX;
3962 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3963 file_lock.fl_start = locku->lu_offset;
3965 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3966 nfs4_transform_lock_offset(&file_lock);
3969 * Try to unlock the file in the VFS.
3971 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3973 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3977 * OK, unlock succeeded; the only thing left to do is update the stateid.
3979 update_stateid(&stp->st_stateid);
3980 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3983 if (locku->lu_stateowner) {
3984 nfs4_get_stateowner(locku->lu_stateowner);
3985 cstate->replay_owner = locku->lu_stateowner;
3987 nfs4_unlock_state();
3991 status = nfserrno(err);
3997 * 1: locks held by lockowner
3998 * 0: no locks held by lockowner
4001 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4003 struct file_lock **flpp;
4004 struct inode *inode = filp->fi_inode;
4008 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4009 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4020 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4021 struct nfsd4_compound_state *cstate,
4022 struct nfsd4_release_lockowner *rlockowner)
4024 clientid_t *clid = &rlockowner->rl_clientid;
4025 struct nfs4_stateowner *sop;
4026 struct nfs4_stateid *stp;
4027 struct xdr_netobj *owner = &rlockowner->rl_owner;
4028 struct list_head matches;
4032 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4033 clid->cl_boot, clid->cl_id);
4035 /* XXX check for lease expiration */
4037 status = nfserr_stale_clientid;
4038 if (STALE_CLIENTID(clid))
4043 status = nfserr_locks_held;
4044 /* XXX: we're doing a linear search through all the lockowners.
4045 * Yipes! For now we'll just hope clients aren't really using
4046 * release_lockowner much, but eventually we have to fix these
4047 * data structures. */
4048 INIT_LIST_HEAD(&matches);
4049 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4050 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4051 if (!same_owner_str(sop, owner, clid))
4053 list_for_each_entry(stp, &sop->so_stateids,
4055 if (check_for_locks(stp->st_file, sop))
4057 /* Note: so_perclient unused for lockowners,
4058 * so it's OK to fool with here. */
4059 list_add(&sop->so_perclient, &matches);
4063 /* Clients probably won't expect us to return with some (but not all)
4064 * of the lockowner state released; so don't release any until all
4065 * have been checked. */
4067 while (!list_empty(&matches)) {
4068 sop = list_entry(matches.next, struct nfs4_stateowner,
4070 /* unhash_stateowner deletes so_perclient only
4071 * for openowners. */
4072 list_del(&sop->so_perclient);
4073 release_lockowner(sop);
4076 nfs4_unlock_state();
4080 static inline struct nfs4_client_reclaim *
4083 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4087 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4089 unsigned int strhashval = clientstr_hashval(name);
4090 struct nfs4_client *clp;
4092 clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
4097 * failure => all reset bets are off, nfserr_no_grace...
4100 nfs4_client_to_reclaim(const char *name)
4102 unsigned int strhashval;
4103 struct nfs4_client_reclaim *crp = NULL;
4105 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4106 crp = alloc_reclaim();
4109 strhashval = clientstr_hashval(name);
4110 INIT_LIST_HEAD(&crp->cr_strhash);
4111 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4112 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4113 reclaim_str_hashtbl_size++;
4118 nfs4_release_reclaim(void)
4120 struct nfs4_client_reclaim *crp = NULL;
4123 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4124 while (!list_empty(&reclaim_str_hashtbl[i])) {
4125 crp = list_entry(reclaim_str_hashtbl[i].next,
4126 struct nfs4_client_reclaim, cr_strhash);
4127 list_del(&crp->cr_strhash);
4129 reclaim_str_hashtbl_size--;
4132 BUG_ON(reclaim_str_hashtbl_size);
4136 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4137 static struct nfs4_client_reclaim *
4138 nfs4_find_reclaim_client(clientid_t *clid)
4140 unsigned int strhashval;
4141 struct nfs4_client *clp;
4142 struct nfs4_client_reclaim *crp = NULL;
4145 /* find clientid in conf_id_hashtbl */
4146 clp = find_confirmed_client(clid);
4150 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4151 clp->cl_name.len, clp->cl_name.data,
4154 /* find clp->cl_name in reclaim_str_hashtbl */
4155 strhashval = clientstr_hashval(clp->cl_recdir);
4156 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4157 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4165 * Called from OPEN. Look for clientid in reclaim list.
4168 nfs4_check_open_reclaim(clientid_t *clid)
4170 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4173 /* initialization to perform at module load time: */
4176 nfs4_state_init(void)
4180 status = nfsd4_init_slabs();
4183 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4184 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4185 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4186 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4187 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4188 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4190 for (i = 0; i < SESSION_HASH_SIZE; i++)
4191 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4192 for (i = 0; i < FILE_HASH_SIZE; i++) {
4193 INIT_LIST_HEAD(&file_hashtbl[i]);
4195 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4196 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4197 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4199 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4200 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4201 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4203 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4204 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4205 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4207 memset(&onestateid, ~0, sizeof(stateid_t));
4208 INIT_LIST_HEAD(&close_lru);
4209 INIT_LIST_HEAD(&client_lru);
4210 INIT_LIST_HEAD(&del_recall_lru);
4211 reclaim_str_hashtbl_size = 0;
4216 nfsd4_load_reboot_recovery_data(void)
4221 nfsd4_init_recdir(user_recovery_dirname);
4222 status = nfsd4_recdir_load();
4223 nfs4_unlock_state();
4225 printk("NFSD: Failure reading reboot recovery data\n");
4229 * Since the lifetime of a delegation isn't limited to that of an open, a
4230 * client may quite reasonably hang on to a delegation as long as it has
4231 * the inode cached. This becomes an obvious problem the first time a
4232 * client's inode cache approaches the size of the server's total memory.
4234 * For now we avoid this problem by imposing a hard limit on the number
4235 * of delegations, which varies according to the server's memory size.
4238 set_max_delegations(void)
4241 * Allow at most 4 delegations per megabyte of RAM. Quick
4242 * estimates suggest that in the worst case (where every delegation
4243 * is for a different inode), a delegation could take about 1.5K,
4244 * giving a worst case usage of about 6% of memory.
4246 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4249 /* initialization to perform when the nfsd service is started: */
4252 __nfs4_state_start(void)
4256 boot_time = get_seconds();
4257 locks_start_grace(&nfsd4_manager);
4258 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4260 ret = set_callback_cred();
4263 laundry_wq = create_singlethread_workqueue("nfsd4");
4264 if (laundry_wq == NULL)
4266 ret = nfsd4_create_callback_queue();
4268 goto out_free_laundry;
4269 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4270 set_max_delegations();
4273 destroy_workqueue(laundry_wq);
4278 nfs4_state_start(void)
4280 nfsd4_load_reboot_recovery_data();
4281 return __nfs4_state_start();
4285 __nfs4_state_shutdown(void)
4288 struct nfs4_client *clp = NULL;
4289 struct nfs4_delegation *dp = NULL;
4290 struct list_head *pos, *next, reaplist;
4292 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4293 while (!list_empty(&conf_id_hashtbl[i])) {
4294 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4297 while (!list_empty(&unconf_str_hashtbl[i])) {
4298 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4302 INIT_LIST_HEAD(&reaplist);
4303 spin_lock(&recall_lock);
4304 list_for_each_safe(pos, next, &del_recall_lru) {
4305 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4306 list_move(&dp->dl_recall_lru, &reaplist);
4308 spin_unlock(&recall_lock);
4309 list_for_each_safe(pos, next, &reaplist) {
4310 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4311 list_del_init(&dp->dl_recall_lru);
4312 unhash_delegation(dp);
4315 nfsd4_shutdown_recdir();
4319 nfs4_state_shutdown(void)
4321 cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4322 destroy_workqueue(laundry_wq);
4323 locks_end_grace(&nfsd4_manager);
4325 nfs4_release_reclaim();
4326 __nfs4_state_shutdown();
4327 nfs4_unlock_state();
4328 nfsd4_destroy_callback_queue();
4332 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4333 * accessed when nfsd is starting.
4336 nfs4_set_recdir(char *recdir)
4338 strcpy(user_recovery_dirname, recdir);
4342 * Change the NFSv4 recovery directory to recdir.
4345 nfs4_reset_recoverydir(char *recdir)
4350 status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4354 if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4355 nfs4_set_recdir(recdir);
4363 nfs4_recoverydir(void)
4365 return user_recovery_dirname;