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/idr.h>
36 #include <linux/file.h>
38 #include <linux/slab.h>
39 #include <linux/namei.h>
40 #include <linux/swap.h>
41 #include <linux/pagemap.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/clnt.h>
47 #define NFSDDBG_FACILITY NFSDDBG_PROC
50 time_t nfsd4_lease = 90; /* default lease time */
51 time_t nfsd4_grace = 90;
52 static time_t boot_time;
53 static stateid_t zerostateid; /* bits all 0 */
54 static stateid_t onestateid; /* bits all 1 */
55 static u64 current_sessionid = 1;
57 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
58 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60 /* forward declarations */
61 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
65 /* Currently used for almost all code touching nfsv4 state: */
66 static DEFINE_MUTEX(client_mutex);
69 * Currently used for the del_recall_lru and file hash table. In an
70 * effort to decrease the scope of the client_mutex, this spinlock may
71 * eventually cover more:
73 static DEFINE_SPINLOCK(recall_lock);
75 static struct kmem_cache *openowner_slab = NULL;
76 static struct kmem_cache *lockowner_slab = NULL;
77 static struct kmem_cache *file_slab = NULL;
78 static struct kmem_cache *stateid_slab = NULL;
79 static struct kmem_cache *deleg_slab = NULL;
84 mutex_lock(&client_mutex);
88 nfs4_unlock_state(void)
90 mutex_unlock(&client_mutex);
94 opaque_hashval(const void *ptr, int nbytes)
96 unsigned char *cptr = (unsigned char *) ptr;
106 static struct list_head del_recall_lru;
109 put_nfs4_file(struct nfs4_file *fi)
111 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
112 list_del(&fi->fi_hash);
113 spin_unlock(&recall_lock);
115 kmem_cache_free(file_slab, fi);
120 get_nfs4_file(struct nfs4_file *fi)
122 atomic_inc(&fi->fi_ref);
125 static int num_delegations;
126 unsigned int max_delegations;
129 * Open owner state (share locks)
132 /* hash tables for open owners */
133 #define OPEN_OWNER_HASH_BITS 8
134 #define OPEN_OWNER_HASH_SIZE (1 << OPEN_OWNER_HASH_BITS)
135 #define OPEN_OWNER_HASH_MASK (OPEN_OWNER_HASH_SIZE - 1)
137 static unsigned int open_ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
141 ret = opaque_hashval(ownername->data, ownername->len);
143 return ret & OPEN_OWNER_HASH_MASK;
146 static struct list_head open_ownerstr_hashtbl[OPEN_OWNER_HASH_SIZE];
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS 8
150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
154 static unsigned int file_hashval(struct inode *ino)
156 /* XXX: why are we hashing on inode pointer, anyway? */
157 return hash_ptr(ino, FILE_HASH_BITS);
160 static struct list_head file_hashtbl[FILE_HASH_SIZE];
162 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
164 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
165 atomic_inc(&fp->fi_access[oflag]);
168 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
170 if (oflag == O_RDWR) {
171 __nfs4_file_get_access(fp, O_RDONLY);
172 __nfs4_file_get_access(fp, O_WRONLY);
174 __nfs4_file_get_access(fp, oflag);
177 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
179 if (fp->fi_fds[oflag]) {
180 fput(fp->fi_fds[oflag]);
181 fp->fi_fds[oflag] = NULL;
185 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
187 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
188 nfs4_file_put_fd(fp, oflag);
190 * It's also safe to get rid of the RDWR open *if*
191 * we no longer have need of the other kind of access
192 * or if we already have the other kind of open:
194 if (fp->fi_fds[1-oflag]
195 || atomic_read(&fp->fi_access[1 - oflag]) == 0)
196 nfs4_file_put_fd(fp, O_RDWR);
200 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
202 if (oflag == O_RDWR) {
203 __nfs4_file_put_access(fp, O_RDONLY);
204 __nfs4_file_put_access(fp, O_WRONLY);
206 __nfs4_file_put_access(fp, oflag);
209 static inline int get_new_stid(struct nfs4_stid *stid)
211 static int min_stateid = 0;
215 if (!idr_pre_get(&stateids, GFP_KERNEL))
218 error = idr_get_new_above(&stateids, stid, min_stateid, &new_stid);
220 * All this code is currently serialized; the preallocation
221 * above should still be ours:
225 * It shouldn't be a problem to reuse an opaque stateid value.
226 * I don't think it is for 4.1. But with 4.0 I worry that, for
227 * example, a stray write retransmission could be accepted by
228 * the server when it should have been rejected. Therefore,
229 * adopt a trick from the sctp code to attempt to maximize the
230 * amount of time until an id is reused, by ensuring they always
231 * "increase" (mod INT_MAX):
234 min_stateid = new_stid+1;
235 if (min_stateid == INT_MAX)
240 static inline __be32 init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
242 stateid_t *s = &stid->sc_stateid;
245 stid->sc_type = type;
246 stid->sc_client = cl;
247 s->si_opaque.so_clid = cl->cl_clientid;
248 new_id = get_new_stid(stid);
250 return nfserr_jukebox;
251 s->si_opaque.so_id = (u32)new_id;
252 /* Will be incremented before return to client: */
253 s->si_generation = 0;
257 static struct nfs4_delegation *
258 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
260 struct nfs4_delegation *dp;
261 struct nfs4_file *fp = stp->st_file;
264 dprintk("NFSD alloc_init_deleg\n");
266 * Major work on the lease subsystem (for example, to support
267 * calbacks on stat) will be required before we can support
268 * write delegations properly.
270 if (type != NFS4_OPEN_DELEGATE_READ)
272 if (fp->fi_had_conflict)
274 if (num_delegations > max_delegations)
276 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
279 status = init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
281 kmem_cache_free(deleg_slab, dp);
285 * delegation seqid's are never incremented. The 4.1 special
286 * meaning of seqid 0 isn't meaningful, really, but let's avoid
287 * 0 anyway just for consistency and use 1:
289 dp->dl_stid.sc_stateid.si_generation = 1;
291 INIT_LIST_HEAD(&dp->dl_perfile);
292 INIT_LIST_HEAD(&dp->dl_perclnt);
293 INIT_LIST_HEAD(&dp->dl_recall_lru);
297 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
299 atomic_set(&dp->dl_count, 1);
300 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
305 nfs4_put_delegation(struct nfs4_delegation *dp)
307 if (atomic_dec_and_test(&dp->dl_count)) {
308 dprintk("NFSD: freeing dp %p\n",dp);
309 put_nfs4_file(dp->dl_file);
310 kmem_cache_free(deleg_slab, dp);
315 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
317 if (atomic_dec_and_test(&fp->fi_delegees)) {
318 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
320 fput(fp->fi_deleg_file);
321 fp->fi_deleg_file = NULL;
325 static void unhash_stid(struct nfs4_stid *s)
327 idr_remove(&stateids, s->sc_stateid.si_opaque.so_id);
330 /* Called under the state lock. */
332 unhash_delegation(struct nfs4_delegation *dp)
334 unhash_stid(&dp->dl_stid);
335 list_del_init(&dp->dl_perclnt);
336 spin_lock(&recall_lock);
337 list_del_init(&dp->dl_perfile);
338 list_del_init(&dp->dl_recall_lru);
339 spin_unlock(&recall_lock);
340 nfs4_put_deleg_lease(dp->dl_file);
341 nfs4_put_delegation(dp);
348 /* client_lock protects the client lru list and session hash table */
349 static DEFINE_SPINLOCK(client_lock);
351 /* Hash tables for nfs4_clientid state */
352 #define CLIENT_HASH_BITS 4
353 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
354 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
356 static unsigned int clientid_hashval(u32 id)
358 return id & CLIENT_HASH_MASK;
361 static unsigned int clientstr_hashval(const char *name)
363 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
367 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
368 * used in reboot/reset lease grace period processing
370 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
371 * setclientid_confirmed info.
373 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
376 * client_lru holds client queue ordered by nfs4_client.cl_time
379 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
380 * for last close replay.
382 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
383 static int reclaim_str_hashtbl_size = 0;
384 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
385 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
386 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
387 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
388 static struct list_head client_lru;
389 static struct list_head close_lru;
392 * We store the NONE, READ, WRITE, and BOTH bits separately in the
393 * st_{access,deny}_bmap field of the stateid, in order to track not
394 * only what share bits are currently in force, but also what
395 * combinations of share bits previous opens have used. This allows us
396 * to enforce the recommendation of rfc 3530 14.2.19 that the server
397 * return an error if the client attempt to downgrade to a combination
398 * of share bits not explicable by closing some of its previous opens.
400 * XXX: This enforcement is actually incomplete, since we don't keep
401 * track of access/deny bit combinations; so, e.g., we allow:
403 * OPEN allow read, deny write
404 * OPEN allow both, deny none
405 * DOWNGRADE allow read, deny none
407 * which we should reject.
410 set_access(unsigned int *access, unsigned long bmap) {
414 for (i = 1; i < 4; i++) {
415 if (test_bit(i, &bmap))
421 set_deny(unsigned int *deny, unsigned long bmap) {
425 for (i = 0; i < 4; i++) {
426 if (test_bit(i, &bmap))
432 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
433 unsigned int access, deny;
435 set_access(&access, stp->st_access_bmap);
436 set_deny(&deny, stp->st_deny_bmap);
437 if ((access & open->op_share_deny) || (deny & open->op_share_access))
442 static int nfs4_access_to_omode(u32 access)
444 switch (access & NFS4_SHARE_ACCESS_BOTH) {
445 case NFS4_SHARE_ACCESS_READ:
447 case NFS4_SHARE_ACCESS_WRITE:
449 case NFS4_SHARE_ACCESS_BOTH:
455 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
457 list_del(&stp->st_perfile);
458 list_del(&stp->st_perstateowner);
461 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
465 if (stp->st_access_bmap) {
466 for (i = 1; i < 4; i++) {
467 if (test_bit(i, &stp->st_access_bmap))
468 nfs4_file_put_access(stp->st_file,
469 nfs4_access_to_omode(i));
470 __clear_bit(i, &stp->st_access_bmap);
473 put_nfs4_file(stp->st_file);
477 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
479 kmem_cache_free(stateid_slab, stp);
482 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
486 unhash_generic_stateid(stp);
487 unhash_stid(&stp->st_stid);
488 file = find_any_file(stp->st_file);
490 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
491 close_generic_stateid(stp);
492 free_generic_stateid(stp);
495 static void unhash_lockowner(struct nfs4_lockowner *lo)
497 struct nfs4_ol_stateid *stp;
499 list_del(&lo->lo_owner.so_strhash);
500 list_del(&lo->lo_perstateid);
501 while (!list_empty(&lo->lo_owner.so_stateids)) {
502 stp = list_first_entry(&lo->lo_owner.so_stateids,
503 struct nfs4_ol_stateid, st_perstateowner);
504 release_lock_stateid(stp);
508 static void release_lockowner(struct nfs4_lockowner *lo)
510 unhash_lockowner(lo);
511 nfs4_free_lockowner(lo);
515 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
517 struct nfs4_lockowner *lo;
519 while (!list_empty(&open_stp->st_lockowners)) {
520 lo = list_entry(open_stp->st_lockowners.next,
521 struct nfs4_lockowner, lo_perstateid);
522 release_lockowner(lo);
526 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
528 unhash_generic_stateid(stp);
529 release_stateid_lockowners(stp);
530 close_generic_stateid(stp);
533 static void release_open_stateid(struct nfs4_ol_stateid *stp)
535 unhash_open_stateid(stp);
536 unhash_stid(&stp->st_stid);
537 free_generic_stateid(stp);
540 static void unhash_openowner(struct nfs4_openowner *oo)
542 struct nfs4_ol_stateid *stp;
544 list_del(&oo->oo_owner.so_strhash);
545 list_del(&oo->oo_perclient);
546 while (!list_empty(&oo->oo_owner.so_stateids)) {
547 stp = list_first_entry(&oo->oo_owner.so_stateids,
548 struct nfs4_ol_stateid, st_perstateowner);
549 release_open_stateid(stp);
553 static void release_last_closed_stateid(struct nfs4_openowner *oo)
555 struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
558 unhash_stid(&s->st_stid);
559 free_generic_stateid(s);
560 oo->oo_last_closed_stid = NULL;
564 static void release_openowner(struct nfs4_openowner *oo)
566 unhash_openowner(oo);
567 list_del(&oo->oo_close_lru);
568 release_last_closed_stateid(oo);
569 nfs4_free_openowner(oo);
572 #define SESSION_HASH_SIZE 512
573 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
576 hash_sessionid(struct nfs4_sessionid *sessionid)
578 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
580 return sid->sequence % SESSION_HASH_SIZE;
584 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
586 u32 *ptr = (u32 *)(&sessionid->data[0]);
587 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
591 gen_sessionid(struct nfsd4_session *ses)
593 struct nfs4_client *clp = ses->se_client;
594 struct nfsd4_sessionid *sid;
596 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
597 sid->clientid = clp->cl_clientid;
598 sid->sequence = current_sessionid++;
603 * The protocol defines ca_maxresponssize_cached to include the size of
604 * the rpc header, but all we need to cache is the data starting after
605 * the end of the initial SEQUENCE operation--the rest we regenerate
606 * each time. Therefore we can advertise a ca_maxresponssize_cached
607 * value that is the number of bytes in our cache plus a few additional
608 * bytes. In order to stay on the safe side, and not promise more than
609 * we can cache, those additional bytes must be the minimum possible: 24
610 * bytes of rpc header (xid through accept state, with AUTH_NULL
611 * verifier), 12 for the compound header (with zero-length tag), and 44
612 * for the SEQUENCE op response:
614 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
617 free_session_slots(struct nfsd4_session *ses)
621 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
622 kfree(ses->se_slots[i]);
626 * We don't actually need to cache the rpc and session headers, so we
627 * can allocate a little less for each slot:
629 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
631 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
634 static int nfsd4_sanitize_slot_size(u32 size)
636 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
637 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
643 * XXX: If we run out of reserved DRC memory we could (up to a point)
644 * re-negotiate active sessions and reduce their slot usage to make
645 * rooom for new connections. For now we just fail the create session.
647 static int nfsd4_get_drc_mem(int slotsize, u32 num)
651 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
653 spin_lock(&nfsd_drc_lock);
654 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
655 nfsd_drc_max_mem - nfsd_drc_mem_used);
656 num = min_t(int, num, avail / slotsize);
657 nfsd_drc_mem_used += num * slotsize;
658 spin_unlock(&nfsd_drc_lock);
663 static void nfsd4_put_drc_mem(int slotsize, int num)
665 spin_lock(&nfsd_drc_lock);
666 nfsd_drc_mem_used -= slotsize * num;
667 spin_unlock(&nfsd_drc_lock);
670 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
672 struct nfsd4_session *new;
675 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
676 + sizeof(struct nfsd4_session) > PAGE_SIZE);
677 mem = numslots * sizeof(struct nfsd4_slot *);
679 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
682 /* allocate each struct nfsd4_slot and data cache in one piece */
683 for (i = 0; i < numslots; i++) {
684 mem = sizeof(struct nfsd4_slot) + slotsize;
685 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
686 if (!new->se_slots[i])
692 kfree(new->se_slots[i]);
697 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
699 u32 maxrpc = nfsd_serv->sv_max_mesg;
701 new->maxreqs = numslots;
702 new->maxresp_cached = min_t(u32, req->maxresp_cached,
703 slotsize + NFSD_MIN_HDR_SEQ_SZ);
704 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
705 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
706 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
709 static void free_conn(struct nfsd4_conn *c)
711 svc_xprt_put(c->cn_xprt);
715 static void nfsd4_conn_lost(struct svc_xpt_user *u)
717 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
718 struct nfs4_client *clp = c->cn_session->se_client;
720 spin_lock(&clp->cl_lock);
721 if (!list_empty(&c->cn_persession)) {
722 list_del(&c->cn_persession);
725 spin_unlock(&clp->cl_lock);
726 nfsd4_probe_callback(clp);
729 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
731 struct nfsd4_conn *conn;
733 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
736 svc_xprt_get(rqstp->rq_xprt);
737 conn->cn_xprt = rqstp->rq_xprt;
738 conn->cn_flags = flags;
739 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
743 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
745 conn->cn_session = ses;
746 list_add(&conn->cn_persession, &ses->se_conns);
749 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
751 struct nfs4_client *clp = ses->se_client;
753 spin_lock(&clp->cl_lock);
754 __nfsd4_hash_conn(conn, ses);
755 spin_unlock(&clp->cl_lock);
758 static int nfsd4_register_conn(struct nfsd4_conn *conn)
760 conn->cn_xpt_user.callback = nfsd4_conn_lost;
761 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
764 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
766 struct nfsd4_conn *conn;
769 conn = alloc_conn(rqstp, dir);
771 return nfserr_jukebox;
772 nfsd4_hash_conn(conn, ses);
773 ret = nfsd4_register_conn(conn);
775 /* oops; xprt is already down: */
776 nfsd4_conn_lost(&conn->cn_xpt_user);
780 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
782 u32 dir = NFS4_CDFC4_FORE;
784 if (ses->se_flags & SESSION4_BACK_CHAN)
785 dir |= NFS4_CDFC4_BACK;
787 return nfsd4_new_conn(rqstp, ses, dir);
790 /* must be called under client_lock */
791 static void nfsd4_del_conns(struct nfsd4_session *s)
793 struct nfs4_client *clp = s->se_client;
794 struct nfsd4_conn *c;
796 spin_lock(&clp->cl_lock);
797 while (!list_empty(&s->se_conns)) {
798 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
799 list_del_init(&c->cn_persession);
800 spin_unlock(&clp->cl_lock);
802 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
805 spin_lock(&clp->cl_lock);
807 spin_unlock(&clp->cl_lock);
810 void free_session(struct kref *kref)
812 struct nfsd4_session *ses;
815 ses = container_of(kref, struct nfsd4_session, se_ref);
816 nfsd4_del_conns(ses);
817 spin_lock(&nfsd_drc_lock);
818 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
819 nfsd_drc_mem_used -= mem;
820 spin_unlock(&nfsd_drc_lock);
821 free_session_slots(ses);
825 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
827 struct nfsd4_session *new;
828 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
829 int numslots, slotsize;
834 * Note decreasing slot size below client's request may
835 * make it difficult for client to function correctly, whereas
836 * decreasing the number of slots will (just?) affect
837 * performance. When short on memory we therefore prefer to
838 * decrease number of slots instead of their size.
840 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
841 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
845 new = alloc_session(slotsize, numslots);
847 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
850 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
852 new->se_client = clp;
855 INIT_LIST_HEAD(&new->se_conns);
857 new->se_cb_seq_nr = 1;
858 new->se_flags = cses->flags;
859 new->se_cb_prog = cses->callback_prog;
860 kref_init(&new->se_ref);
861 idx = hash_sessionid(&new->se_sessionid);
862 spin_lock(&client_lock);
863 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
864 spin_lock(&clp->cl_lock);
865 list_add(&new->se_perclnt, &clp->cl_sessions);
866 spin_unlock(&clp->cl_lock);
867 spin_unlock(&client_lock);
869 status = nfsd4_new_conn_from_crses(rqstp, new);
870 /* whoops: benny points out, status is ignored! (err, or bogus) */
872 free_session(&new->se_ref);
875 if (cses->flags & SESSION4_BACK_CHAN) {
876 struct sockaddr *sa = svc_addr(rqstp);
878 * This is a little silly; with sessions there's no real
879 * use for the callback address. Use the peer address
880 * as a reasonable default for now, but consider fixing
881 * the rpc client not to require an address in the
884 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
885 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
887 nfsd4_probe_callback(clp);
891 /* caller must hold client_lock */
892 static struct nfsd4_session *
893 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
895 struct nfsd4_session *elem;
898 dump_sessionid(__func__, sessionid);
899 idx = hash_sessionid(sessionid);
900 /* Search in the appropriate list */
901 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
902 if (!memcmp(elem->se_sessionid.data, sessionid->data,
903 NFS4_MAX_SESSIONID_LEN)) {
908 dprintk("%s: session not found\n", __func__);
912 /* caller must hold client_lock */
914 unhash_session(struct nfsd4_session *ses)
916 list_del(&ses->se_hash);
917 spin_lock(&ses->se_client->cl_lock);
918 list_del(&ses->se_perclnt);
919 spin_unlock(&ses->se_client->cl_lock);
922 /* must be called under the client_lock */
924 renew_client_locked(struct nfs4_client *clp)
926 if (is_client_expired(clp)) {
927 dprintk("%s: client (clientid %08x/%08x) already expired\n",
929 clp->cl_clientid.cl_boot,
930 clp->cl_clientid.cl_id);
935 * Move client to the end to the LRU list.
937 dprintk("renewing client (clientid %08x/%08x)\n",
938 clp->cl_clientid.cl_boot,
939 clp->cl_clientid.cl_id);
940 list_move_tail(&clp->cl_lru, &client_lru);
941 clp->cl_time = get_seconds();
945 renew_client(struct nfs4_client *clp)
947 spin_lock(&client_lock);
948 renew_client_locked(clp);
949 spin_unlock(&client_lock);
952 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
954 STALE_CLIENTID(clientid_t *clid)
956 if (clid->cl_boot == boot_time)
958 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
959 clid->cl_boot, clid->cl_id, boot_time);
964 * XXX Should we use a slab cache ?
965 * This type of memory management is somewhat inefficient, but we use it
966 * anyway since SETCLIENTID is not a common operation.
968 static struct nfs4_client *alloc_client(struct xdr_netobj name)
970 struct nfs4_client *clp;
972 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
975 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
976 if (clp->cl_name.data == NULL) {
980 memcpy(clp->cl_name.data, name.data, name.len);
981 clp->cl_name.len = name.len;
986 free_client(struct nfs4_client *clp)
988 while (!list_empty(&clp->cl_sessions)) {
989 struct nfsd4_session *ses;
990 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
992 list_del(&ses->se_perclnt);
993 nfsd4_put_session(ses);
995 if (clp->cl_cred.cr_group_info)
996 put_group_info(clp->cl_cred.cr_group_info);
997 kfree(clp->cl_principal);
998 kfree(clp->cl_name.data);
1003 release_session_client(struct nfsd4_session *session)
1005 struct nfs4_client *clp = session->se_client;
1007 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1009 if (is_client_expired(clp)) {
1011 session->se_client = NULL;
1013 renew_client_locked(clp);
1014 spin_unlock(&client_lock);
1017 /* must be called under the client_lock */
1019 unhash_client_locked(struct nfs4_client *clp)
1021 struct nfsd4_session *ses;
1023 mark_client_expired(clp);
1024 list_del(&clp->cl_lru);
1025 spin_lock(&clp->cl_lock);
1026 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1027 list_del_init(&ses->se_hash);
1028 spin_unlock(&clp->cl_lock);
1032 expire_client(struct nfs4_client *clp)
1034 struct nfs4_openowner *oo;
1035 struct nfs4_delegation *dp;
1036 struct list_head reaplist;
1038 INIT_LIST_HEAD(&reaplist);
1039 spin_lock(&recall_lock);
1040 while (!list_empty(&clp->cl_delegations)) {
1041 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1042 list_del_init(&dp->dl_perclnt);
1043 list_move(&dp->dl_recall_lru, &reaplist);
1045 spin_unlock(&recall_lock);
1046 while (!list_empty(&reaplist)) {
1047 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1048 list_del_init(&dp->dl_recall_lru);
1049 unhash_delegation(dp);
1051 while (!list_empty(&clp->cl_openowners)) {
1052 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1053 release_openowner(oo);
1055 nfsd4_shutdown_callback(clp);
1056 if (clp->cl_cb_conn.cb_xprt)
1057 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1058 list_del(&clp->cl_idhash);
1059 list_del(&clp->cl_strhash);
1060 spin_lock(&client_lock);
1061 unhash_client_locked(clp);
1062 if (atomic_read(&clp->cl_refcount) == 0)
1064 spin_unlock(&client_lock);
1067 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1069 memcpy(target->cl_verifier.data, source->data,
1070 sizeof(target->cl_verifier.data));
1073 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1075 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1076 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1079 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1081 target->cr_uid = source->cr_uid;
1082 target->cr_gid = source->cr_gid;
1083 target->cr_group_info = source->cr_group_info;
1084 get_group_info(target->cr_group_info);
1087 static int same_name(const char *n1, const char *n2)
1089 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1093 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1095 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1099 same_clid(clientid_t *cl1, clientid_t *cl2)
1101 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1104 /* XXX what about NGROUP */
1106 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1108 return cr1->cr_uid == cr2->cr_uid;
1111 static void gen_clid(struct nfs4_client *clp)
1113 static u32 current_clientid = 1;
1115 clp->cl_clientid.cl_boot = boot_time;
1116 clp->cl_clientid.cl_id = current_clientid++;
1119 static void gen_confirm(struct nfs4_client *clp)
1124 p = (u32 *)clp->cl_confirm.data;
1125 *p++ = get_seconds();
1129 static struct nfs4_stid *find_stateid(stateid_t *t)
1131 return idr_find(&stateids, t->si_opaque.so_id);
1134 static struct nfs4_stid *find_stateid_by_type(stateid_t *t, char typemask)
1136 struct nfs4_stid *s;
1138 s = find_stateid(t);
1141 if (typemask & s->sc_type)
1146 static struct nfs4_ol_stateid *find_ol_stateid_by_type(stateid_t *t, char typemask)
1148 struct nfs4_stid *s;
1150 s = find_stateid_by_type(t, typemask);
1153 return openlockstateid(s);
1156 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1157 struct svc_rqst *rqstp, nfs4_verifier *verf)
1159 struct nfs4_client *clp;
1160 struct sockaddr *sa = svc_addr(rqstp);
1163 clp = alloc_client(name);
1167 INIT_LIST_HEAD(&clp->cl_sessions);
1169 princ = svc_gss_principal(rqstp);
1171 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1172 if (clp->cl_principal == NULL) {
1178 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1179 atomic_set(&clp->cl_refcount, 0);
1180 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1181 INIT_LIST_HEAD(&clp->cl_idhash);
1182 INIT_LIST_HEAD(&clp->cl_strhash);
1183 INIT_LIST_HEAD(&clp->cl_openowners);
1184 INIT_LIST_HEAD(&clp->cl_delegations);
1185 INIT_LIST_HEAD(&clp->cl_lru);
1186 INIT_LIST_HEAD(&clp->cl_callbacks);
1187 spin_lock_init(&clp->cl_lock);
1188 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1189 clp->cl_time = get_seconds();
1190 clear_bit(0, &clp->cl_cb_slot_busy);
1191 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1192 copy_verf(clp, verf);
1193 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1194 clp->cl_flavor = rqstp->rq_flavor;
1195 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1197 clp->cl_cb_session = NULL;
1201 static int check_name(struct xdr_netobj name)
1205 if (name.len > NFS4_OPAQUE_LIMIT) {
1206 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1213 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1215 unsigned int idhashval;
1217 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1218 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1219 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1224 move_to_confirmed(struct nfs4_client *clp)
1226 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1227 unsigned int strhashval;
1229 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1230 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1231 strhashval = clientstr_hashval(clp->cl_recdir);
1232 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1236 static struct nfs4_client *
1237 find_confirmed_client(clientid_t *clid)
1239 struct nfs4_client *clp;
1240 unsigned int idhashval = clientid_hashval(clid->cl_id);
1242 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1243 if (same_clid(&clp->cl_clientid, clid))
1249 static struct nfs4_client *
1250 find_unconfirmed_client(clientid_t *clid)
1252 struct nfs4_client *clp;
1253 unsigned int idhashval = clientid_hashval(clid->cl_id);
1255 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1256 if (same_clid(&clp->cl_clientid, clid))
1262 static bool clp_used_exchangeid(struct nfs4_client *clp)
1264 return clp->cl_exchange_flags != 0;
1267 static struct nfs4_client *
1268 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1270 struct nfs4_client *clp;
1272 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1273 if (same_name(clp->cl_recdir, dname))
1279 static struct nfs4_client *
1280 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1282 struct nfs4_client *clp;
1284 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1285 if (same_name(clp->cl_recdir, dname))
1292 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1294 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1295 struct sockaddr *sa = svc_addr(rqstp);
1296 u32 scopeid = rpc_get_scope_id(sa);
1297 unsigned short expected_family;
1299 /* Currently, we only support tcp and tcp6 for the callback channel */
1300 if (se->se_callback_netid_len == 3 &&
1301 !memcmp(se->se_callback_netid_val, "tcp", 3))
1302 expected_family = AF_INET;
1303 else if (se->se_callback_netid_len == 4 &&
1304 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1305 expected_family = AF_INET6;
1309 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1310 se->se_callback_addr_len,
1311 (struct sockaddr *)&conn->cb_addr,
1312 sizeof(conn->cb_addr));
1314 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1317 if (conn->cb_addr.ss_family == AF_INET6)
1318 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1320 conn->cb_prog = se->se_callback_prog;
1321 conn->cb_ident = se->se_callback_ident;
1322 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1325 conn->cb_addr.ss_family = AF_UNSPEC;
1326 conn->cb_addrlen = 0;
1327 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1328 "will not receive delegations\n",
1329 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1335 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1338 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1340 struct nfsd4_slot *slot = resp->cstate.slot;
1343 dprintk("--> %s slot %p\n", __func__, slot);
1345 slot->sl_opcnt = resp->opcnt;
1346 slot->sl_status = resp->cstate.status;
1348 if (nfsd4_not_cached(resp)) {
1349 slot->sl_datalen = 0;
1352 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1353 base = (char *)resp->cstate.datap -
1354 (char *)resp->xbuf->head[0].iov_base;
1355 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1357 WARN("%s: sessions DRC could not cache compound\n", __func__);
1362 * Encode the replay sequence operation from the slot values.
1363 * If cachethis is FALSE encode the uncached rep error on the next
1364 * operation which sets resp->p and increments resp->opcnt for
1365 * nfs4svc_encode_compoundres.
1369 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1370 struct nfsd4_compoundres *resp)
1372 struct nfsd4_op *op;
1373 struct nfsd4_slot *slot = resp->cstate.slot;
1375 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1376 resp->opcnt, resp->cstate.slot->sl_cachethis);
1378 /* Encode the replayed sequence operation */
1379 op = &args->ops[resp->opcnt - 1];
1380 nfsd4_encode_operation(resp, op);
1382 /* Return nfserr_retry_uncached_rep in next operation. */
1383 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1384 op = &args->ops[resp->opcnt++];
1385 op->status = nfserr_retry_uncached_rep;
1386 nfsd4_encode_operation(resp, op);
1392 * The sequence operation is not cached because we can use the slot and
1396 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1397 struct nfsd4_sequence *seq)
1399 struct nfsd4_slot *slot = resp->cstate.slot;
1402 dprintk("--> %s slot %p\n", __func__, slot);
1404 /* Either returns 0 or nfserr_retry_uncached */
1405 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1406 if (status == nfserr_retry_uncached_rep)
1409 /* The sequence operation has been encoded, cstate->datap set. */
1410 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1412 resp->opcnt = slot->sl_opcnt;
1413 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1414 status = slot->sl_status;
1420 * Set the exchange_id flags returned by the server.
1423 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1425 /* pNFS is not supported */
1426 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1428 /* Referrals are supported, Migration is not. */
1429 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1431 /* set the wire flags to return to client. */
1432 clid->flags = new->cl_exchange_flags;
1436 nfsd4_exchange_id(struct svc_rqst *rqstp,
1437 struct nfsd4_compound_state *cstate,
1438 struct nfsd4_exchange_id *exid)
1440 struct nfs4_client *unconf, *conf, *new;
1442 unsigned int strhashval;
1443 char dname[HEXDIR_LEN];
1444 char addr_str[INET6_ADDRSTRLEN];
1445 nfs4_verifier verf = exid->verifier;
1446 struct sockaddr *sa = svc_addr(rqstp);
1448 rpc_ntop(sa, addr_str, sizeof(addr_str));
1449 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1450 "ip_addr=%s flags %x, spa_how %d\n",
1451 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1452 addr_str, exid->flags, exid->spa_how);
1454 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1455 return nfserr_inval;
1457 /* Currently only support SP4_NONE */
1458 switch (exid->spa_how) {
1462 return nfserr_serverfault;
1464 BUG(); /* checked by xdr code */
1466 return nfserr_serverfault; /* no excuse :-/ */
1469 status = nfs4_make_rec_clidname(dname, &exid->clname);
1474 strhashval = clientstr_hashval(dname);
1479 conf = find_confirmed_client_by_str(dname, strhashval);
1481 if (!clp_used_exchangeid(conf)) {
1482 status = nfserr_clid_inuse; /* XXX: ? */
1485 if (!same_verf(&verf, &conf->cl_verifier)) {
1486 /* 18.35.4 case 8 */
1487 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1488 status = nfserr_not_same;
1491 /* Client reboot: destroy old state */
1492 expire_client(conf);
1495 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1496 /* 18.35.4 case 9 */
1497 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1498 status = nfserr_perm;
1501 expire_client(conf);
1505 * Set bit when the owner id and verifier map to an already
1506 * confirmed client id (18.35.3).
1508 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1511 * Falling into 18.35.4 case 2, possible router replay.
1512 * Leave confirmed record intact and return same result.
1514 copy_verf(conf, &verf);
1519 /* 18.35.4 case 7 */
1520 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1521 status = nfserr_noent;
1525 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1528 * Possible retry or client restart. Per 18.35.4 case 4,
1529 * a new unconfirmed record should be generated regardless
1530 * of whether any properties have changed.
1532 expire_client(unconf);
1537 new = create_client(exid->clname, dname, rqstp, &verf);
1539 status = nfserr_jukebox;
1544 add_to_unconfirmed(new, strhashval);
1546 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1547 exid->clientid.cl_id = new->cl_clientid.cl_id;
1550 nfsd4_set_ex_flags(new, exid);
1552 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1553 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1557 nfs4_unlock_state();
1559 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1564 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1566 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1569 /* The slot is in use, and no response has been sent. */
1571 if (seqid == slot_seqid)
1572 return nfserr_jukebox;
1574 return nfserr_seq_misordered;
1577 if (likely(seqid == slot_seqid + 1))
1580 if (seqid == slot_seqid)
1581 return nfserr_replay_cache;
1583 if (seqid == 1 && (slot_seqid + 1) == 0)
1585 /* Misordered replay or misordered new request */
1586 return nfserr_seq_misordered;
1590 * Cache the create session result into the create session single DRC
1591 * slot cache by saving the xdr structure. sl_seqid has been set.
1592 * Do this for solo or embedded create session operations.
1595 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1596 struct nfsd4_clid_slot *slot, int nfserr)
1598 slot->sl_status = nfserr;
1599 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1603 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1604 struct nfsd4_clid_slot *slot)
1606 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1607 return slot->sl_status;
1610 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1611 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1612 1 + /* MIN tag is length with zero, only length */ \
1613 3 + /* version, opcount, opcode */ \
1614 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1615 /* seqid, slotID, slotID, cache */ \
1616 4 ) * sizeof(__be32))
1618 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1619 2 + /* verifier: AUTH_NULL, length 0 */\
1621 1 + /* MIN tag is length with zero, only length */ \
1622 3 + /* opcount, opcode, opstatus*/ \
1623 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1624 /* seqid, slotID, slotID, slotID, status */ \
1625 5 ) * sizeof(__be32))
1627 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1629 return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1630 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1634 nfsd4_create_session(struct svc_rqst *rqstp,
1635 struct nfsd4_compound_state *cstate,
1636 struct nfsd4_create_session *cr_ses)
1638 struct sockaddr *sa = svc_addr(rqstp);
1639 struct nfs4_client *conf, *unconf;
1640 struct nfsd4_session *new;
1641 struct nfsd4_clid_slot *cs_slot = NULL;
1642 bool confirm_me = false;
1645 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1646 return nfserr_inval;
1649 unconf = find_unconfirmed_client(&cr_ses->clientid);
1650 conf = find_confirmed_client(&cr_ses->clientid);
1653 cs_slot = &conf->cl_cs_slot;
1654 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1655 if (status == nfserr_replay_cache) {
1656 dprintk("Got a create_session replay! seqid= %d\n",
1658 /* Return the cached reply status */
1659 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1661 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1662 status = nfserr_seq_misordered;
1663 dprintk("Sequence misordered!\n");
1664 dprintk("Expected seqid= %d but got seqid= %d\n",
1665 cs_slot->sl_seqid, cr_ses->seqid);
1668 } else if (unconf) {
1669 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1670 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1671 status = nfserr_clid_inuse;
1675 cs_slot = &unconf->cl_cs_slot;
1676 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1678 /* an unconfirmed replay returns misordered */
1679 status = nfserr_seq_misordered;
1686 status = nfserr_stale_clientid;
1691 * XXX: we should probably set this at creation time, and check
1692 * for consistent minorversion use throughout:
1694 conf->cl_minorversion = 1;
1696 * We do not support RDMA or persistent sessions
1698 cr_ses->flags &= ~SESSION4_PERSIST;
1699 cr_ses->flags &= ~SESSION4_RDMA;
1701 status = nfserr_toosmall;
1702 if (check_forechannel_attrs(cr_ses->fore_channel))
1705 status = nfserr_jukebox;
1706 new = alloc_init_session(rqstp, conf, cr_ses);
1710 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1711 NFS4_MAX_SESSIONID_LEN);
1712 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1713 sizeof(struct nfsd4_channel_attrs));
1714 cs_slot->sl_seqid++;
1715 cr_ses->seqid = cs_slot->sl_seqid;
1717 /* cache solo and embedded create sessions under the state lock */
1718 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1720 move_to_confirmed(conf);
1722 nfs4_unlock_state();
1723 dprintk("%s returns %d\n", __func__, ntohl(status));
1727 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1729 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1730 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1732 return argp->opcnt == resp->opcnt;
1735 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1738 case NFS4_CDFC4_FORE:
1739 case NFS4_CDFC4_BACK:
1741 case NFS4_CDFC4_FORE_OR_BOTH:
1742 case NFS4_CDFC4_BACK_OR_BOTH:
1743 *dir = NFS4_CDFC4_BOTH;
1746 return nfserr_inval;
1749 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1750 struct nfsd4_compound_state *cstate,
1751 struct nfsd4_bind_conn_to_session *bcts)
1755 if (!nfsd4_last_compound_op(rqstp))
1756 return nfserr_not_only_op;
1757 spin_lock(&client_lock);
1758 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1759 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1760 * client_lock iself: */
1761 if (cstate->session) {
1762 nfsd4_get_session(cstate->session);
1763 atomic_inc(&cstate->session->se_client->cl_refcount);
1765 spin_unlock(&client_lock);
1766 if (!cstate->session)
1767 return nfserr_badsession;
1769 status = nfsd4_map_bcts_dir(&bcts->dir);
1771 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1775 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1779 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1783 nfsd4_destroy_session(struct svc_rqst *r,
1784 struct nfsd4_compound_state *cstate,
1785 struct nfsd4_destroy_session *sessionid)
1787 struct nfsd4_session *ses;
1788 u32 status = nfserr_badsession;
1791 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1792 * - Should we return nfserr_back_chan_busy if waiting for
1793 * callbacks on to-be-destroyed session?
1794 * - Do we need to clear any callback info from previous session?
1797 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1798 if (!nfsd4_last_compound_op(r))
1799 return nfserr_not_only_op;
1801 dump_sessionid(__func__, &sessionid->sessionid);
1802 spin_lock(&client_lock);
1803 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1805 spin_unlock(&client_lock);
1809 unhash_session(ses);
1810 spin_unlock(&client_lock);
1813 nfsd4_probe_callback_sync(ses->se_client);
1814 nfs4_unlock_state();
1816 nfsd4_del_conns(ses);
1818 nfsd4_put_session(ses);
1821 dprintk("%s returns %d\n", __func__, ntohl(status));
1825 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1827 struct nfsd4_conn *c;
1829 list_for_each_entry(c, &s->se_conns, cn_persession) {
1830 if (c->cn_xprt == xpt) {
1837 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1839 struct nfs4_client *clp = ses->se_client;
1840 struct nfsd4_conn *c;
1843 spin_lock(&clp->cl_lock);
1844 c = __nfsd4_find_conn(new->cn_xprt, ses);
1846 spin_unlock(&clp->cl_lock);
1850 __nfsd4_hash_conn(new, ses);
1851 spin_unlock(&clp->cl_lock);
1852 ret = nfsd4_register_conn(new);
1854 /* oops; xprt is already down: */
1855 nfsd4_conn_lost(&new->cn_xpt_user);
1859 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1861 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1863 return args->opcnt > session->se_fchannel.maxops;
1866 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1867 struct nfsd4_session *session)
1869 struct xdr_buf *xb = &rqstp->rq_arg;
1871 return xb->len > session->se_fchannel.maxreq_sz;
1875 nfsd4_sequence(struct svc_rqst *rqstp,
1876 struct nfsd4_compound_state *cstate,
1877 struct nfsd4_sequence *seq)
1879 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1880 struct nfsd4_session *session;
1881 struct nfsd4_slot *slot;
1882 struct nfsd4_conn *conn;
1885 if (resp->opcnt != 1)
1886 return nfserr_sequence_pos;
1889 * Will be either used or freed by nfsd4_sequence_check_conn
1892 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1894 return nfserr_jukebox;
1896 spin_lock(&client_lock);
1897 status = nfserr_badsession;
1898 session = find_in_sessionid_hashtbl(&seq->sessionid);
1902 status = nfserr_too_many_ops;
1903 if (nfsd4_session_too_many_ops(rqstp, session))
1906 status = nfserr_req_too_big;
1907 if (nfsd4_request_too_big(rqstp, session))
1910 status = nfserr_badslot;
1911 if (seq->slotid >= session->se_fchannel.maxreqs)
1914 slot = session->se_slots[seq->slotid];
1915 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1917 /* We do not negotiate the number of slots yet, so set the
1918 * maxslots to the session maxreqs which is used to encode
1919 * sr_highest_slotid and the sr_target_slot id to maxslots */
1920 seq->maxslots = session->se_fchannel.maxreqs;
1922 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1923 if (status == nfserr_replay_cache) {
1924 cstate->slot = slot;
1925 cstate->session = session;
1926 /* Return the cached reply status and set cstate->status
1927 * for nfsd4_proc_compound processing */
1928 status = nfsd4_replay_cache_entry(resp, seq);
1929 cstate->status = nfserr_replay_cache;
1935 nfsd4_sequence_check_conn(conn, session);
1938 /* Success! bump slot seqid */
1939 slot->sl_inuse = true;
1940 slot->sl_seqid = seq->seqid;
1941 slot->sl_cachethis = seq->cachethis;
1943 cstate->slot = slot;
1944 cstate->session = session;
1947 /* Hold a session reference until done processing the compound. */
1948 if (cstate->session) {
1949 struct nfs4_client *clp = session->se_client;
1951 nfsd4_get_session(cstate->session);
1952 atomic_inc(&clp->cl_refcount);
1953 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1954 seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1957 spin_unlock(&client_lock);
1958 dprintk("%s: return %d\n", __func__, ntohl(status));
1963 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1967 if (rc->rca_one_fs) {
1968 if (!cstate->current_fh.fh_dentry)
1969 return nfserr_nofilehandle;
1971 * We don't take advantage of the rca_one_fs case.
1972 * That's OK, it's optional, we can safely ignore it.
1978 status = nfserr_complete_already;
1979 if (cstate->session->se_client->cl_firststate)
1982 status = nfserr_stale_clientid;
1983 if (is_client_expired(cstate->session->se_client))
1985 * The following error isn't really legal.
1986 * But we only get here if the client just explicitly
1987 * destroyed the client. Surely it no longer cares what
1988 * error it gets back on an operation for the dead
1994 nfsd4_create_clid_dir(cstate->session->se_client);
1996 nfs4_unlock_state();
2001 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2002 struct nfsd4_setclientid *setclid)
2004 struct xdr_netobj clname = {
2005 .len = setclid->se_namelen,
2006 .data = setclid->se_name,
2008 nfs4_verifier clverifier = setclid->se_verf;
2009 unsigned int strhashval;
2010 struct nfs4_client *conf, *unconf, *new;
2012 char dname[HEXDIR_LEN];
2014 if (!check_name(clname))
2015 return nfserr_inval;
2017 status = nfs4_make_rec_clidname(dname, &clname);
2022 * XXX The Duplicate Request Cache (DRC) has been checked (??)
2023 * We get here on a DRC miss.
2026 strhashval = clientstr_hashval(dname);
2029 conf = find_confirmed_client_by_str(dname, strhashval);
2031 /* RFC 3530 14.2.33 CASE 0: */
2032 status = nfserr_clid_inuse;
2033 if (clp_used_exchangeid(conf))
2035 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2036 char addr_str[INET6_ADDRSTRLEN];
2037 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2039 dprintk("NFSD: setclientid: string in use by client "
2040 "at %s\n", addr_str);
2045 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
2046 * has a description of SETCLIENTID request processing consisting
2047 * of 5 bullet points, labeled as CASE0 - CASE4 below.
2049 unconf = find_unconfirmed_client_by_str(dname, strhashval);
2050 status = nfserr_jukebox;
2053 * RFC 3530 14.2.33 CASE 4:
2054 * placed first, because it is the normal case
2057 expire_client(unconf);
2058 new = create_client(clname, dname, rqstp, &clverifier);
2062 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
2064 * RFC 3530 14.2.33 CASE 1:
2065 * probable callback update
2068 /* Note this is removing unconfirmed {*x***},
2069 * which is stronger than RFC recommended {vxc**}.
2070 * This has the advantage that there is at most
2071 * one {*x***} in either list at any time.
2073 expire_client(unconf);
2075 new = create_client(clname, dname, rqstp, &clverifier);
2078 copy_clid(new, conf);
2079 } else if (!unconf) {
2081 * RFC 3530 14.2.33 CASE 2:
2082 * probable client reboot; state will be removed if
2085 new = create_client(clname, dname, rqstp, &clverifier);
2091 * RFC 3530 14.2.33 CASE 3:
2092 * probable client reboot; state will be removed if
2095 expire_client(unconf);
2096 new = create_client(clname, dname, rqstp, &clverifier);
2102 * XXX: we should probably set this at creation time, and check
2103 * for consistent minorversion use throughout:
2105 new->cl_minorversion = 0;
2106 gen_callback(new, setclid, rqstp);
2107 add_to_unconfirmed(new, strhashval);
2108 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2109 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2110 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2113 nfs4_unlock_state();
2119 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2120 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2121 * bullets, labeled as CASE1 - CASE4 below.
2124 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2125 struct nfsd4_compound_state *cstate,
2126 struct nfsd4_setclientid_confirm *setclientid_confirm)
2128 struct sockaddr *sa = svc_addr(rqstp);
2129 struct nfs4_client *conf, *unconf;
2130 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2131 clientid_t * clid = &setclientid_confirm->sc_clientid;
2134 if (STALE_CLIENTID(clid))
2135 return nfserr_stale_clientid;
2137 * XXX The Duplicate Request Cache (DRC) has been checked (??)
2138 * We get here on a DRC miss.
2143 conf = find_confirmed_client(clid);
2144 unconf = find_unconfirmed_client(clid);
2146 status = nfserr_clid_inuse;
2147 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2149 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2153 * section 14.2.34 of RFC 3530 has a description of
2154 * SETCLIENTID_CONFIRM request processing consisting
2155 * of 4 bullet points, labeled as CASE1 - CASE4 below.
2157 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2159 * RFC 3530 14.2.34 CASE 1:
2162 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2163 status = nfserr_clid_inuse;
2165 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2166 nfsd4_probe_callback(conf);
2167 expire_client(unconf);
2171 } else if (conf && !unconf) {
2173 * RFC 3530 14.2.34 CASE 2:
2174 * probable retransmitted request; play it safe and
2177 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2178 status = nfserr_clid_inuse;
2181 } else if (!conf && unconf
2182 && same_verf(&unconf->cl_confirm, &confirm)) {
2184 * RFC 3530 14.2.34 CASE 3:
2185 * Normal case; new or rebooted client:
2187 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2188 status = nfserr_clid_inuse;
2191 clientstr_hashval(unconf->cl_recdir);
2192 conf = find_confirmed_client_by_str(unconf->cl_recdir,
2195 nfsd4_remove_clid_dir(conf);
2196 expire_client(conf);
2198 move_to_confirmed(unconf);
2200 nfsd4_probe_callback(conf);
2203 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2204 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2207 * RFC 3530 14.2.34 CASE 4:
2208 * Client probably hasn't noticed that we rebooted yet.
2210 status = nfserr_stale_clientid;
2212 /* check that we have hit one of the cases...*/
2213 status = nfserr_clid_inuse;
2216 nfs4_unlock_state();
2220 /* OPEN Share state helper functions */
2221 static inline struct nfs4_file *
2222 alloc_init_file(struct inode *ino)
2224 struct nfs4_file *fp;
2225 unsigned int hashval = file_hashval(ino);
2227 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2229 atomic_set(&fp->fi_ref, 1);
2230 INIT_LIST_HEAD(&fp->fi_hash);
2231 INIT_LIST_HEAD(&fp->fi_stateids);
2232 INIT_LIST_HEAD(&fp->fi_delegations);
2233 fp->fi_inode = igrab(ino);
2234 fp->fi_had_conflict = false;
2235 fp->fi_lease = NULL;
2236 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2237 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2238 spin_lock(&recall_lock);
2239 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2240 spin_unlock(&recall_lock);
2247 nfsd4_free_slab(struct kmem_cache **slab)
2251 kmem_cache_destroy(*slab);
2256 nfsd4_free_slabs(void)
2258 nfsd4_free_slab(&openowner_slab);
2259 nfsd4_free_slab(&lockowner_slab);
2260 nfsd4_free_slab(&file_slab);
2261 nfsd4_free_slab(&stateid_slab);
2262 nfsd4_free_slab(&deleg_slab);
2266 nfsd4_init_slabs(void)
2268 openowner_slab = kmem_cache_create("nfsd4_openowners",
2269 sizeof(struct nfs4_openowner), 0, 0, NULL);
2270 if (openowner_slab == NULL)
2272 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2273 sizeof(struct nfs4_openowner), 0, 0, NULL);
2274 if (lockowner_slab == NULL)
2276 file_slab = kmem_cache_create("nfsd4_files",
2277 sizeof(struct nfs4_file), 0, 0, NULL);
2278 if (file_slab == NULL)
2280 stateid_slab = kmem_cache_create("nfsd4_stateids",
2281 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2282 if (stateid_slab == NULL)
2284 deleg_slab = kmem_cache_create("nfsd4_delegations",
2285 sizeof(struct nfs4_delegation), 0, 0, NULL);
2286 if (deleg_slab == NULL)
2291 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2295 void nfs4_free_openowner(struct nfs4_openowner *oo)
2297 kfree(oo->oo_owner.so_owner.data);
2298 kmem_cache_free(openowner_slab, oo);
2301 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2303 kfree(lo->lo_owner.so_owner.data);
2304 kmem_cache_free(lockowner_slab, lo);
2307 static void init_nfs4_replay(struct nfs4_replay *rp)
2309 rp->rp_status = nfserr_serverfault;
2311 rp->rp_buf = rp->rp_ibuf;
2314 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2316 struct nfs4_stateowner *sop;
2318 sop = kmem_cache_alloc(slab, GFP_KERNEL);
2322 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2323 if (!sop->so_owner.data) {
2324 kmem_cache_free(slab, sop);
2327 sop->so_owner.len = owner->len;
2329 INIT_LIST_HEAD(&sop->so_stateids);
2330 sop->so_client = clp;
2331 init_nfs4_replay(&sop->so_replay);
2335 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2337 list_add(&oo->oo_owner.so_strhash, &open_ownerstr_hashtbl[strhashval]);
2338 list_add(&oo->oo_perclient, &clp->cl_openowners);
2341 static struct nfs4_openowner *
2342 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2343 struct nfs4_openowner *oo;
2345 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2348 oo->oo_owner.so_is_open_owner = 1;
2349 oo->oo_owner.so_seqid = open->op_seqid;
2352 oo->oo_last_closed_stid = NULL;
2353 INIT_LIST_HEAD(&oo->oo_close_lru);
2354 hash_openowner(oo, clp, strhashval);
2358 static inline __be32 init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2359 struct nfs4_openowner *oo = open->op_openowner;
2360 struct nfs4_client *clp = oo->oo_owner.so_client;
2363 status = init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2366 INIT_LIST_HEAD(&stp->st_lockowners);
2367 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2368 list_add(&stp->st_perfile, &fp->fi_stateids);
2369 stp->st_stateowner = &oo->oo_owner;
2372 stp->st_access_bmap = 0;
2373 stp->st_deny_bmap = 0;
2374 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2375 &stp->st_access_bmap);
2376 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2377 stp->st_openstp = NULL;
2382 move_to_close_lru(struct nfs4_openowner *oo)
2384 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2386 list_move_tail(&oo->oo_close_lru, &close_lru);
2387 oo->oo_time = get_seconds();
2391 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2394 return (sop->so_owner.len == owner->len) &&
2395 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2396 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2399 static struct nfs4_openowner *
2400 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2402 struct nfs4_stateowner *so = NULL;
2404 list_for_each_entry(so, &open_ownerstr_hashtbl[hashval], so_strhash) {
2405 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2406 return container_of(so, struct nfs4_openowner, oo_owner);
2411 /* search file_hashtbl[] for file */
2412 static struct nfs4_file *
2413 find_file(struct inode *ino)
2415 unsigned int hashval = file_hashval(ino);
2416 struct nfs4_file *fp;
2418 spin_lock(&recall_lock);
2419 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2420 if (fp->fi_inode == ino) {
2422 spin_unlock(&recall_lock);
2426 spin_unlock(&recall_lock);
2430 static inline int access_valid(u32 x, u32 minorversion)
2432 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2434 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2436 x &= ~NFS4_SHARE_ACCESS_MASK;
2437 if (minorversion && x) {
2438 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2440 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2442 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2449 static inline int deny_valid(u32 x)
2451 /* Note: unlike access bits, deny bits may be zero. */
2452 return x <= NFS4_SHARE_DENY_BOTH;
2456 * Called to check deny when READ with all zero stateid or
2457 * WRITE with all zero or all one stateid
2460 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2462 struct inode *ino = current_fh->fh_dentry->d_inode;
2463 struct nfs4_file *fp;
2464 struct nfs4_ol_stateid *stp;
2467 dprintk("NFSD: nfs4_share_conflict\n");
2469 fp = find_file(ino);
2472 ret = nfserr_locked;
2473 /* Search for conflicting share reservations */
2474 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2475 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2476 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2485 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2487 /* We're assuming the state code never drops its reference
2488 * without first removing the lease. Since we're in this lease
2489 * callback (and since the lease code is serialized by the kernel
2490 * lock) we know the server hasn't removed the lease yet, we know
2491 * it's safe to take a reference: */
2492 atomic_inc(&dp->dl_count);
2494 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2496 /* only place dl_time is set. protected by lock_flocks*/
2497 dp->dl_time = get_seconds();
2499 nfsd4_cb_recall(dp);
2502 /* Called from break_lease() with lock_flocks() held. */
2503 static void nfsd_break_deleg_cb(struct file_lock *fl)
2505 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2506 struct nfs4_delegation *dp;
2509 /* We assume break_lease is only called once per lease: */
2510 BUG_ON(fp->fi_had_conflict);
2512 * We don't want the locks code to timeout the lease for us;
2513 * we'll remove it ourself if a delegation isn't returned
2516 fl->fl_break_time = 0;
2518 spin_lock(&recall_lock);
2519 fp->fi_had_conflict = true;
2520 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2521 nfsd_break_one_deleg(dp);
2522 spin_unlock(&recall_lock);
2526 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2529 return lease_modify(onlist, arg);
2534 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2535 .lm_break = nfsd_break_deleg_cb,
2536 .lm_change = nfsd_change_deleg_cb,
2539 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2541 if (nfsd4_has_session(cstate))
2543 if (seqid == so->so_seqid - 1)
2544 return nfserr_replay_me;
2545 if (seqid == so->so_seqid)
2547 return nfserr_bad_seqid;
2551 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2552 struct nfsd4_open *open)
2554 clientid_t *clientid = &open->op_clientid;
2555 struct nfs4_client *clp = NULL;
2556 unsigned int strhashval;
2557 struct nfs4_openowner *oo = NULL;
2560 if (!check_name(open->op_owner))
2561 return nfserr_inval;
2563 if (STALE_CLIENTID(&open->op_clientid))
2564 return nfserr_stale_clientid;
2566 strhashval = open_ownerstr_hashval(clientid->cl_id, &open->op_owner);
2567 oo = find_openstateowner_str(strhashval, open);
2568 open->op_openowner = oo;
2570 /* Make sure the client's lease hasn't expired. */
2571 clp = find_confirmed_client(clientid);
2573 return nfserr_expired;
2576 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2577 /* Replace unconfirmed owners without checking for replay. */
2578 clp = oo->oo_owner.so_client;
2579 release_openowner(oo);
2580 open->op_openowner = NULL;
2583 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2587 if (open->op_openowner == NULL) {
2588 oo = alloc_init_open_stateowner(strhashval, clp, open);
2590 return nfserr_jukebox;
2591 open->op_openowner = oo;
2593 list_del_init(&oo->oo_close_lru);
2594 renew_client(oo->oo_owner.so_client);
2598 static inline __be32
2599 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2601 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2602 return nfserr_openmode;
2607 static int share_access_to_flags(u32 share_access)
2609 share_access &= ~NFS4_SHARE_WANT_MASK;
2611 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2614 static struct nfs4_delegation *find_deleg_stateid(stateid_t *s)
2616 struct nfs4_stid *ret;
2618 ret = find_stateid_by_type(s, NFS4_DELEG_STID);
2621 return delegstateid(ret);
2625 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2626 struct nfs4_delegation **dp)
2629 __be32 status = nfserr_bad_stateid;
2631 *dp = find_deleg_stateid(&open->op_delegate_stateid);
2634 flags = share_access_to_flags(open->op_share_access);
2635 status = nfs4_check_delegmode(*dp, flags);
2639 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2643 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2648 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2650 struct nfs4_ol_stateid *local;
2651 struct nfs4_openowner *oo = open->op_openowner;
2653 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2654 /* ignore lock owners */
2655 if (local->st_stateowner->so_is_open_owner == 0)
2657 /* remember if we have seen this open owner */
2658 if (local->st_stateowner == &oo->oo_owner)
2660 /* check for conflicting share reservations */
2661 if (!test_share(local, open))
2662 return nfserr_share_denied;
2667 static inline struct nfs4_ol_stateid *
2668 nfs4_alloc_stateid(void)
2670 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2673 static inline int nfs4_access_to_access(u32 nfs4_access)
2677 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2678 flags |= NFSD_MAY_READ;
2679 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2680 flags |= NFSD_MAY_WRITE;
2684 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2685 struct svc_fh *cur_fh, struct nfsd4_open *open)
2688 int oflag = nfs4_access_to_omode(open->op_share_access);
2689 int access = nfs4_access_to_access(open->op_share_access);
2691 /* CLAIM_DELEGATE_CUR is used in response to a broken lease;
2692 * allowing it to break the lease and return EAGAIN leaves the
2693 * client unable to make progress in returning the delegation */
2694 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2695 access |= NFSD_MAY_NOT_BREAK_LEASE;
2697 if (!fp->fi_fds[oflag]) {
2698 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2699 &fp->fi_fds[oflag]);
2703 nfs4_file_get_access(fp, oflag);
2709 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_ol_stateid **stpp,
2710 struct nfs4_file *fp, struct svc_fh *cur_fh,
2711 struct nfsd4_open *open)
2713 struct nfs4_ol_stateid *stp;
2716 stp = nfs4_alloc_stateid();
2718 return nfserr_jukebox;
2720 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2722 kmem_cache_free(stateid_slab, stp);
2729 static inline __be32
2730 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2731 struct nfsd4_open *open)
2733 struct iattr iattr = {
2734 .ia_valid = ATTR_SIZE,
2737 if (!open->op_truncate)
2739 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2740 return nfserr_inval;
2741 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2745 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
2747 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2751 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2753 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2757 status = nfsd4_truncate(rqstp, cur_fh, open);
2760 int oflag = nfs4_access_to_omode(op_share_access);
2761 nfs4_file_put_access(fp, oflag);
2765 /* remember the open */
2766 __set_bit(op_share_access, &stp->st_access_bmap);
2767 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2774 nfs4_set_claim_prev(struct nfsd4_open *open)
2776 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2777 open->op_openowner->oo_owner.so_client->cl_firststate = 1;
2780 /* Should we give out recallable state?: */
2781 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2783 if (clp->cl_cb_state == NFSD4_CB_UP)
2786 * In the sessions case, since we don't have to establish a
2787 * separate connection for callbacks, we assume it's OK
2788 * until we hear otherwise:
2790 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2793 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2795 struct file_lock *fl;
2797 fl = locks_alloc_lock();
2800 locks_init_lock(fl);
2801 fl->fl_lmops = &nfsd_lease_mng_ops;
2802 fl->fl_flags = FL_LEASE;
2803 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2804 fl->fl_end = OFFSET_MAX;
2805 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2806 fl->fl_pid = current->tgid;
2810 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2812 struct nfs4_file *fp = dp->dl_file;
2813 struct file_lock *fl;
2816 fl = nfs4_alloc_init_lease(dp, flag);
2819 fl->fl_file = find_readable_file(fp);
2820 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2821 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2823 list_del_init(&dp->dl_perclnt);
2824 locks_free_lock(fl);
2828 fp->fi_deleg_file = fl->fl_file;
2829 get_file(fp->fi_deleg_file);
2830 atomic_set(&fp->fi_delegees, 1);
2831 list_add(&dp->dl_perfile, &fp->fi_delegations);
2835 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2837 struct nfs4_file *fp = dp->dl_file;
2840 return nfs4_setlease(dp, flag);
2841 spin_lock(&recall_lock);
2842 if (fp->fi_had_conflict) {
2843 spin_unlock(&recall_lock);
2846 atomic_inc(&fp->fi_delegees);
2847 list_add(&dp->dl_perfile, &fp->fi_delegations);
2848 spin_unlock(&recall_lock);
2849 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2854 * Attempt to hand out a delegation.
2857 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2859 struct nfs4_delegation *dp;
2860 struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2862 int status, flag = 0;
2864 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2865 flag = NFS4_OPEN_DELEGATE_NONE;
2866 open->op_recall = 0;
2867 switch (open->op_claim_type) {
2868 case NFS4_OPEN_CLAIM_PREVIOUS:
2870 open->op_recall = 1;
2871 flag = open->op_delegate_type;
2872 if (flag == NFS4_OPEN_DELEGATE_NONE)
2875 case NFS4_OPEN_CLAIM_NULL:
2876 /* Let's not give out any delegations till everyone's
2877 * had the chance to reclaim theirs.... */
2878 if (locks_in_grace())
2880 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2882 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2883 flag = NFS4_OPEN_DELEGATE_WRITE;
2885 flag = NFS4_OPEN_DELEGATE_READ;
2891 dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2894 status = nfs4_set_delegation(dp, flag);
2898 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2900 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2901 STATEID_VAL(&dp->dl_stid.sc_stateid));
2903 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2904 && flag == NFS4_OPEN_DELEGATE_NONE
2905 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2906 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2907 open->op_delegate_type = flag;
2910 nfs4_put_delegation(dp);
2912 flag = NFS4_OPEN_DELEGATE_NONE;
2917 * called with nfs4_lock_state() held.
2920 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2922 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2923 struct nfs4_file *fp = NULL;
2924 struct inode *ino = current_fh->fh_dentry->d_inode;
2925 struct nfs4_ol_stateid *stp = NULL;
2926 struct nfs4_delegation *dp = NULL;
2929 status = nfserr_inval;
2930 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2931 || !deny_valid(open->op_share_deny))
2934 * Lookup file; if found, lookup stateid and check open request,
2935 * and check for delegations in the process of being recalled.
2936 * If not found, create the nfs4_file struct
2938 fp = find_file(ino);
2940 if ((status = nfs4_check_open(fp, open, &stp)))
2942 status = nfs4_check_deleg(fp, open, &dp);
2946 status = nfserr_bad_stateid;
2947 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2949 status = nfserr_jukebox;
2950 fp = alloc_init_file(ino);
2956 * OPEN the file, or upgrade an existing OPEN.
2957 * If truncate fails, the OPEN fails.
2960 /* Stateid was found, this is an OPEN upgrade */
2961 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2965 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2968 status = init_open_stateid(stp, fp, open);
2970 release_open_stateid(stp);
2973 status = nfsd4_truncate(rqstp, current_fh, open);
2975 release_open_stateid(stp);
2979 update_stateid(&stp->st_stid.sc_stateid);
2980 memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
2982 if (nfsd4_has_session(&resp->cstate))
2983 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2986 * Attempt to hand out a delegation. No error return, because the
2987 * OPEN succeeds even if we fail.
2989 nfs4_open_delegation(current_fh, open, stp);
2993 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2994 STATEID_VAL(&stp->st_stid.sc_stateid));
2998 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2999 nfs4_set_claim_prev(open);
3001 * To finish the open response, we just need to set the rflags.
3003 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3004 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3005 !nfsd4_has_session(&resp->cstate))
3006 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3012 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3015 struct nfs4_client *clp;
3019 dprintk("process_renew(%08x/%08x): starting\n",
3020 clid->cl_boot, clid->cl_id);
3021 status = nfserr_stale_clientid;
3022 if (STALE_CLIENTID(clid))
3024 clp = find_confirmed_client(clid);
3025 status = nfserr_expired;
3027 /* We assume the client took too long to RENEW. */
3028 dprintk("nfsd4_renew: clientid not found!\n");
3032 status = nfserr_cb_path_down;
3033 if (!list_empty(&clp->cl_delegations)
3034 && clp->cl_cb_state != NFSD4_CB_UP)
3038 nfs4_unlock_state();
3042 static struct lock_manager nfsd4_manager = {
3046 nfsd4_end_grace(void)
3048 dprintk("NFSD: end of grace period\n");
3049 nfsd4_recdir_purge_old();
3050 locks_end_grace(&nfsd4_manager);
3052 * Now that every NFSv4 client has had the chance to recover and
3053 * to see the (possibly new, possibly shorter) lease time, we
3054 * can safely set the next grace time to the current lease time:
3056 nfsd4_grace = nfsd4_lease;
3060 nfs4_laundromat(void)
3062 struct nfs4_client *clp;
3063 struct nfs4_openowner *oo;
3064 struct nfs4_delegation *dp;
3065 struct list_head *pos, *next, reaplist;
3066 time_t cutoff = get_seconds() - nfsd4_lease;
3067 time_t t, clientid_val = nfsd4_lease;
3068 time_t u, test_val = nfsd4_lease;
3072 dprintk("NFSD: laundromat service - starting\n");
3073 if (locks_in_grace())
3075 INIT_LIST_HEAD(&reaplist);
3076 spin_lock(&client_lock);
3077 list_for_each_safe(pos, next, &client_lru) {
3078 clp = list_entry(pos, struct nfs4_client, cl_lru);
3079 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3080 t = clp->cl_time - cutoff;
3081 if (clientid_val > t)
3085 if (atomic_read(&clp->cl_refcount)) {
3086 dprintk("NFSD: client in use (clientid %08x)\n",
3087 clp->cl_clientid.cl_id);
3090 unhash_client_locked(clp);
3091 list_add(&clp->cl_lru, &reaplist);
3093 spin_unlock(&client_lock);
3094 list_for_each_safe(pos, next, &reaplist) {
3095 clp = list_entry(pos, struct nfs4_client, cl_lru);
3096 dprintk("NFSD: purging unused client (clientid %08x)\n",
3097 clp->cl_clientid.cl_id);
3098 nfsd4_remove_clid_dir(clp);
3101 spin_lock(&recall_lock);
3102 list_for_each_safe(pos, next, &del_recall_lru) {
3103 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3104 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3105 u = dp->dl_time - cutoff;
3110 list_move(&dp->dl_recall_lru, &reaplist);
3112 spin_unlock(&recall_lock);
3113 list_for_each_safe(pos, next, &reaplist) {
3114 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3115 list_del_init(&dp->dl_recall_lru);
3116 unhash_delegation(dp);
3118 test_val = nfsd4_lease;
3119 list_for_each_safe(pos, next, &close_lru) {
3120 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3121 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3122 u = oo->oo_time - cutoff;
3127 release_openowner(oo);
3129 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3130 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3131 nfs4_unlock_state();
3132 return clientid_val;
3135 static struct workqueue_struct *laundry_wq;
3136 static void laundromat_main(struct work_struct *);
3137 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3140 laundromat_main(struct work_struct *not_used)
3144 t = nfs4_laundromat();
3145 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3146 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3149 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3151 if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3152 return nfserr_bad_stateid;
3157 STALE_STATEID(stateid_t *stateid)
3159 if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3161 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3162 STATEID_VAL(stateid));
3167 access_permit_read(unsigned long access_bmap)
3169 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3170 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3171 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3175 access_permit_write(unsigned long access_bmap)
3177 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3178 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3182 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3184 __be32 status = nfserr_openmode;
3186 /* For lock stateid's, we test the parent open, not the lock: */
3187 if (stp->st_openstp)
3188 stp = stp->st_openstp;
3189 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3191 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3198 static inline __be32
3199 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3201 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3203 else if (locks_in_grace()) {
3204 /* Answer in remaining cases depends on existence of
3205 * conflicting state; so we must wait out the grace period. */
3206 return nfserr_grace;
3207 } else if (flags & WR_STATE)
3208 return nfs4_share_conflict(current_fh,
3209 NFS4_SHARE_DENY_WRITE);
3210 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3211 return nfs4_share_conflict(current_fh,
3212 NFS4_SHARE_DENY_READ);
3216 * Allow READ/WRITE during grace period on recovered state only for files
3217 * that are not able to provide mandatory locking.
3220 grace_disallows_io(struct inode *inode)
3222 return locks_in_grace() && mandatory_lock(inode);
3225 /* Returns true iff a is later than b: */
3226 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3228 return (s32)a->si_generation - (s32)b->si_generation > 0;
3231 static int check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3234 * When sessions are used the stateid generation number is ignored
3237 if (has_session && in->si_generation == 0)
3240 if (in->si_generation == ref->si_generation)
3243 /* If the client sends us a stateid from the future, it's buggy: */
3244 if (stateid_generation_after(in, ref))
3245 return nfserr_bad_stateid;
3247 * However, we could see a stateid from the past, even from a
3248 * non-buggy client. For example, if the client sends a lock
3249 * while some IO is outstanding, the lock may bump si_generation
3250 * while the IO is still in flight. The client could avoid that
3251 * situation by waiting for responses on all the IO requests,
3252 * but better performance may result in retrying IO that
3253 * receives an old_stateid error if requests are rarely
3254 * reordered in flight:
3256 return nfserr_old_stateid;
3259 __be32 nfs4_validate_stateid(stateid_t *stateid)
3261 struct nfs4_stid *s;
3262 struct nfs4_ol_stateid *ols;
3265 if (STALE_STATEID(stateid))
3266 return nfserr_stale_stateid;
3268 s = find_stateid(stateid);
3270 return nfserr_stale_stateid;
3271 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3274 if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3276 ols = openlockstateid(s);
3277 if (ols->st_stateowner->so_is_open_owner
3278 && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3279 return nfserr_bad_stateid;
3284 * Checks for stateid operations
3287 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3288 stateid_t *stateid, int flags, struct file **filpp)
3290 struct nfs4_stid *s;
3291 struct nfs4_ol_stateid *stp = NULL;
3292 struct nfs4_delegation *dp = NULL;
3293 struct svc_fh *current_fh = &cstate->current_fh;
3294 struct inode *ino = current_fh->fh_dentry->d_inode;
3300 if (grace_disallows_io(ino))
3301 return nfserr_grace;
3303 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3304 return check_special_stateids(current_fh, stateid, flags);
3306 status = nfserr_stale_stateid;
3307 if (STALE_STATEID(stateid))
3311 * We assume that any stateid that has the current boot time,
3312 * but that we can't find, is expired:
3314 status = nfserr_expired;
3315 s = find_stateid(stateid);
3318 status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3321 switch (s->sc_type) {
3322 case NFS4_DELEG_STID:
3323 dp = delegstateid(s);
3324 status = nfs4_check_delegmode(dp, flags);
3327 renew_client(dp->dl_stid.sc_client);
3329 *filpp = dp->dl_file->fi_deleg_file;
3333 case NFS4_OPEN_STID:
3334 case NFS4_LOCK_STID:
3335 stp = openlockstateid(s);
3336 status = nfs4_check_fh(current_fh, stp);
3339 if (stp->st_stateowner->so_is_open_owner
3340 && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3342 status = nfs4_check_openmode(stp, flags);
3345 renew_client(stp->st_stateowner->so_client);
3347 if (flags & RD_STATE)
3348 *filpp = find_readable_file(stp->st_file);
3350 *filpp = find_writeable_file(stp->st_file);
3354 return nfserr_bad_stateid;
3362 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3364 if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3365 return nfserr_locks_held;
3366 release_lock_stateid(stp);
3371 * Test if the stateid is valid
3374 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3375 struct nfsd4_test_stateid *test_stateid)
3377 /* real work is done during encoding */
3382 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3383 struct nfsd4_free_stateid *free_stateid)
3385 stateid_t *stateid = &free_stateid->fr_stateid;
3386 struct nfs4_stid *s;
3387 __be32 ret = nfserr_bad_stateid;
3390 s = find_stateid(stateid);
3393 switch (s->sc_type) {
3394 case NFS4_DELEG_STID:
3395 ret = nfserr_locks_held;
3397 case NFS4_OPEN_STID:
3398 case NFS4_LOCK_STID:
3399 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3402 if (s->sc_type == NFS4_LOCK_STID)
3403 ret = nfsd4_free_lock_stateid(openlockstateid(s));
3405 ret = nfserr_locks_held;
3408 ret = nfserr_bad_stateid;
3411 nfs4_unlock_state();
3418 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3419 RD_STATE : WR_STATE;
3422 static __be32 nfs4_nospecial_stateid_checks(stateid_t *stateid)
3424 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3425 return nfserr_bad_stateid;
3426 if (STALE_STATEID(stateid))
3427 return nfserr_stale_stateid;
3431 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3433 struct svc_fh *current_fh = &cstate->current_fh;
3434 struct nfs4_stateowner *sop = stp->st_stateowner;
3437 status = nfsd4_check_seqid(cstate, sop, seqid);
3440 if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3442 * "Closed" stateid's exist *only* to return
3443 * nfserr_replay_me from the previous step.
3445 return nfserr_bad_stateid;
3446 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3449 return nfs4_check_fh(current_fh, stp);
3453 * Checks for sequence id mutating operations.
3456 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3457 stateid_t *stateid, char typemask,
3458 struct nfs4_ol_stateid **stpp)
3462 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3463 seqid, STATEID_VAL(stateid));
3466 status = nfs4_nospecial_stateid_checks(stateid);
3469 *stpp = find_ol_stateid_by_type(stateid, typemask);
3471 return nfserr_expired;
3472 cstate->replay_owner = (*stpp)->st_stateowner;
3473 renew_client((*stpp)->st_stateowner->so_client);
3475 return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3478 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3481 struct nfs4_openowner *oo;
3483 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3484 NFS4_OPEN_STID, stpp);
3487 oo = openowner((*stpp)->st_stateowner);
3488 if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3489 return nfserr_bad_stateid;
3494 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3495 struct nfsd4_open_confirm *oc)
3498 struct nfs4_openowner *oo;
3499 struct nfs4_ol_stateid *stp;
3501 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3502 (int)cstate->current_fh.fh_dentry->d_name.len,
3503 cstate->current_fh.fh_dentry->d_name.name);
3505 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3511 status = nfs4_preprocess_seqid_op(cstate,
3512 oc->oc_seqid, &oc->oc_req_stateid,
3513 NFS4_OPEN_STID, &stp);
3516 oo = openowner(stp->st_stateowner);
3517 status = nfserr_bad_stateid;
3518 if (oo->oo_flags & NFS4_OO_CONFIRMED)
3520 oo->oo_flags |= NFS4_OO_CONFIRMED;
3521 update_stateid(&stp->st_stid.sc_stateid);
3522 memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3523 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3524 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3526 nfsd4_create_clid_dir(oo->oo_owner.so_client);
3529 if (!cstate->replay_owner)
3530 nfs4_unlock_state();
3534 static inline void nfs4_file_downgrade(struct nfs4_ol_stateid *stp, unsigned int to_access)
3538 for (i = 1; i < 4; i++) {
3539 if (test_bit(i, &stp->st_access_bmap)
3540 && ((i & to_access) != i)) {
3541 nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(i));
3542 __clear_bit(i, &stp->st_access_bmap);
3548 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3551 for (i = 0; i < 4; i++) {
3552 if ((i & deny) != i)
3553 __clear_bit(i, bmap);
3558 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3559 struct nfsd4_compound_state *cstate,
3560 struct nfsd4_open_downgrade *od)
3563 struct nfs4_ol_stateid *stp;
3565 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3566 (int)cstate->current_fh.fh_dentry->d_name.len,
3567 cstate->current_fh.fh_dentry->d_name.name);
3569 if (!access_valid(od->od_share_access, cstate->minorversion)
3570 || !deny_valid(od->od_share_deny))
3571 return nfserr_inval;
3574 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3575 &od->od_stateid, &stp);
3578 status = nfserr_inval;
3579 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3580 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3581 stp->st_access_bmap, od->od_share_access);
3584 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3585 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3586 stp->st_deny_bmap, od->od_share_deny);
3589 nfs4_file_downgrade(stp, od->od_share_access);
3591 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3593 update_stateid(&stp->st_stid.sc_stateid);
3594 memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3597 if (!cstate->replay_owner)
3598 nfs4_unlock_state();
3602 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3604 struct nfs4_openowner *oo;
3605 struct nfs4_ol_stateid *s;
3607 if (!so->so_is_open_owner)
3610 s = oo->oo_last_closed_stid;
3613 if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3614 /* Release the last_closed_stid on the next seqid bump: */
3615 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3618 oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3619 release_last_closed_stateid(oo);
3622 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3624 unhash_open_stateid(s);
3625 s->st_stid.sc_type = NFS4_CLOSED_STID;
3629 * nfs4_unlock_state() called after encode
3632 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3633 struct nfsd4_close *close)
3636 struct nfs4_openowner *oo;
3637 struct nfs4_ol_stateid *stp;
3639 dprintk("NFSD: nfsd4_close on file %.*s\n",
3640 (int)cstate->current_fh.fh_dentry->d_name.len,
3641 cstate->current_fh.fh_dentry->d_name.name);
3644 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3646 NFS4_OPEN_STID|NFS4_CLOSED_STID,
3650 oo = openowner(stp->st_stateowner);
3652 update_stateid(&stp->st_stid.sc_stateid);
3653 memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3655 nfsd4_close_open_stateid(stp);
3656 oo->oo_last_closed_stid = stp;
3658 /* place unused nfs4_stateowners on so_close_lru list to be
3659 * released by the laundromat service after the lease period
3660 * to enable us to handle CLOSE replay
3662 if (list_empty(&oo->oo_owner.so_stateids))
3663 move_to_close_lru(oo);
3665 if (!cstate->replay_owner)
3666 nfs4_unlock_state();
3671 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3672 struct nfsd4_delegreturn *dr)
3674 struct nfs4_delegation *dp;
3675 stateid_t *stateid = &dr->dr_stateid;
3676 struct inode *inode;
3679 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3681 inode = cstate->current_fh.fh_dentry->d_inode;
3684 status = nfserr_bad_stateid;
3685 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3687 status = nfserr_stale_stateid;
3688 if (STALE_STATEID(stateid))
3690 status = nfserr_expired;
3691 dp = find_deleg_stateid(stateid);
3694 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3697 renew_client(dp->dl_stid.sc_client);
3699 unhash_delegation(dp);
3701 nfs4_unlock_state();
3708 * Lock owner state (byte-range locks)
3710 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3711 #define LOCK_HASH_BITS 8
3712 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3713 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3716 end_offset(u64 start, u64 len)
3721 return end >= start ? end: NFS4_MAX_UINT64;
3724 /* last octet in a range */
3726 last_byte_offset(u64 start, u64 len)
3732 return end > start ? end - 1: NFS4_MAX_UINT64;
3735 static inline unsigned int
3736 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3737 struct xdr_netobj *ownername)
3739 return (file_hashval(inode) + cl_id
3740 + opaque_hashval(ownername->data, ownername->len))
3744 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3747 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3748 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3749 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3750 * locking, this prevents us from being completely protocol-compliant. The
3751 * real solution to this problem is to start using unsigned file offsets in
3752 * the VFS, but this is a very deep change!
3755 nfs4_transform_lock_offset(struct file_lock *lock)
3757 if (lock->fl_start < 0)
3758 lock->fl_start = OFFSET_MAX;
3759 if (lock->fl_end < 0)
3760 lock->fl_end = OFFSET_MAX;
3763 /* Hack!: For now, we're defining this just so we can use a pointer to it
3764 * as a unique cookie to identify our (NFSv4's) posix locks. */
3765 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3769 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3771 struct nfs4_lockowner *lo;
3773 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3774 lo = (struct nfs4_lockowner *) fl->fl_owner;
3775 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3776 lo->lo_owner.so_owner.len, GFP_KERNEL);
3777 if (!deny->ld_owner.data)
3778 /* We just don't care that much */
3780 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3781 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3784 deny->ld_owner.len = 0;
3785 deny->ld_owner.data = NULL;
3786 deny->ld_clientid.cl_boot = 0;
3787 deny->ld_clientid.cl_id = 0;
3789 deny->ld_start = fl->fl_start;
3790 deny->ld_length = NFS4_MAX_UINT64;
3791 if (fl->fl_end != NFS4_MAX_UINT64)
3792 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3793 deny->ld_type = NFS4_READ_LT;
3794 if (fl->fl_type != F_RDLCK)
3795 deny->ld_type = NFS4_WRITE_LT;
3798 static struct nfs4_lockowner *
3799 find_lockowner_str(struct inode *inode, clientid_t *clid,
3800 struct xdr_netobj *owner)
3802 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3803 struct nfs4_stateowner *op;
3805 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3806 if (same_owner_str(op, owner, clid))
3807 return lockowner(op);
3812 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3814 list_add(&lo->lo_owner.so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3815 list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3819 * Alloc a lock owner structure.
3820 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3823 * strhashval = lock_ownerstr_hashval
3826 static struct nfs4_lockowner *
3827 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3828 struct nfs4_lockowner *lo;
3830 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
3833 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
3834 lo->lo_owner.so_is_open_owner = 0;
3835 /* It is the openowner seqid that will be incremented in encode in the
3836 * case of new lockowners; so increment the lock seqid manually: */
3837 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
3838 hash_lockowner(lo, strhashval, clp, open_stp);
3842 static struct nfs4_ol_stateid *
3843 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
3845 struct nfs4_ol_stateid *stp;
3846 struct nfs4_client *clp = lo->lo_owner.so_client;
3849 stp = nfs4_alloc_stateid();
3852 status = init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
3854 free_generic_stateid(stp);
3857 list_add(&stp->st_perfile, &fp->fi_stateids);
3858 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
3859 stp->st_stateowner = &lo->lo_owner;
3862 stp->st_access_bmap = 0;
3863 stp->st_deny_bmap = open_stp->st_deny_bmap;
3864 stp->st_openstp = open_stp;
3869 check_lock_length(u64 offset, u64 length)
3871 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3872 LOFF_OVERFLOW(offset, length)));
3875 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
3877 struct nfs4_file *fp = lock_stp->st_file;
3878 int oflag = nfs4_access_to_omode(access);
3880 if (test_bit(access, &lock_stp->st_access_bmap))
3882 nfs4_file_get_access(fp, oflag);
3883 __set_bit(access, &lock_stp->st_access_bmap);
3890 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3891 struct nfsd4_lock *lock)
3893 struct nfs4_openowner *open_sop = NULL;
3894 struct nfs4_lockowner *lock_sop = NULL;
3895 struct nfs4_ol_stateid *lock_stp;
3896 struct nfs4_file *fp;
3897 struct file *filp = NULL;
3898 struct file_lock file_lock;
3899 struct file_lock conflock;
3901 unsigned int strhashval;
3905 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3906 (long long) lock->lk_offset,
3907 (long long) lock->lk_length);
3909 if (check_lock_length(lock->lk_offset, lock->lk_length))
3910 return nfserr_inval;
3912 if ((status = fh_verify(rqstp, &cstate->current_fh,
3913 S_IFREG, NFSD_MAY_LOCK))) {
3914 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3920 if (lock->lk_is_new) {
3922 * Client indicates that this is a new lockowner.
3923 * Use open owner and open stateid to create lock owner and
3926 struct nfs4_ol_stateid *open_stp = NULL;
3928 status = nfserr_stale_clientid;
3929 if (!nfsd4_has_session(cstate) &&
3930 STALE_CLIENTID(&lock->lk_new_clientid))
3933 /* validate and update open stateid and open seqid */
3934 status = nfs4_preprocess_confirmed_seqid_op(cstate,
3935 lock->lk_new_open_seqid,
3936 &lock->lk_new_open_stateid,
3940 open_sop = openowner(open_stp->st_stateowner);
3941 status = nfserr_bad_stateid;
3942 if (!nfsd4_has_session(cstate) &&
3943 !same_clid(&open_sop->oo_owner.so_client->cl_clientid,
3944 &lock->v.new.clientid))
3946 /* create lockowner and lock stateid */
3947 fp = open_stp->st_file;
3948 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3949 open_sop->oo_owner.so_client->cl_clientid.cl_id,
3950 &lock->v.new.owner);
3951 /* XXX: Do we need to check for duplicate stateowners on
3952 * the same file, or should they just be allowed (and
3953 * create new stateids)? */
3954 status = nfserr_jukebox;
3955 lock_sop = alloc_init_lock_stateowner(strhashval,
3956 open_sop->oo_owner.so_client, open_stp, lock);
3957 if (lock_sop == NULL)
3959 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3960 if (lock_stp == NULL)
3963 /* lock (lock owner + lock stateid) already exists */
3964 status = nfs4_preprocess_seqid_op(cstate,
3965 lock->lk_old_lock_seqid,
3966 &lock->lk_old_lock_stateid,
3967 NFS4_LOCK_STID, &lock_stp);
3970 lock_sop = lockowner(lock_stp->st_stateowner);
3971 fp = lock_stp->st_file;
3973 /* lock_sop and lock_stp have been created or found */
3975 lkflg = setlkflg(lock->lk_type);
3976 status = nfs4_check_openmode(lock_stp, lkflg);
3980 status = nfserr_grace;
3981 if (locks_in_grace() && !lock->lk_reclaim)
3983 status = nfserr_no_grace;
3984 if (!locks_in_grace() && lock->lk_reclaim)
3987 locks_init_lock(&file_lock);
3988 switch (lock->lk_type) {
3991 filp = find_readable_file(lock_stp->st_file);
3993 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
3994 file_lock.fl_type = F_RDLCK;
3997 case NFS4_WRITEW_LT:
3998 filp = find_writeable_file(lock_stp->st_file);
4000 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4001 file_lock.fl_type = F_WRLCK;
4004 status = nfserr_inval;
4008 status = nfserr_openmode;
4011 file_lock.fl_owner = (fl_owner_t)lock_sop;
4012 file_lock.fl_pid = current->tgid;
4013 file_lock.fl_file = filp;
4014 file_lock.fl_flags = FL_POSIX;
4015 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4017 file_lock.fl_start = lock->lk_offset;
4018 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4019 nfs4_transform_lock_offset(&file_lock);
4022 * Try to lock the file in the VFS.
4023 * Note: locks.c uses the BKL to protect the inode's lock list.
4026 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4028 case 0: /* success! */
4029 update_stateid(&lock_stp->st_stid.sc_stateid);
4030 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4034 case (EAGAIN): /* conflock holds conflicting lock */
4035 status = nfserr_denied;
4036 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4037 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4040 status = nfserr_deadlock;
4043 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4044 status = nfserrno(err);
4048 if (status && lock->lk_is_new && lock_sop)
4049 release_lockowner(lock_sop);
4050 if (!cstate->replay_owner)
4051 nfs4_unlock_state();
4056 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4057 * so we do a temporary open here just to get an open file to pass to
4058 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
4061 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4066 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4069 err = vfs_test_lock(file, lock);
4078 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4079 struct nfsd4_lockt *lockt)
4081 struct inode *inode;
4082 struct file_lock file_lock;
4083 struct nfs4_lockowner *lo;
4087 if (locks_in_grace())
4088 return nfserr_grace;
4090 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4091 return nfserr_inval;
4095 status = nfserr_stale_clientid;
4096 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4099 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4102 inode = cstate->current_fh.fh_dentry->d_inode;
4103 locks_init_lock(&file_lock);
4104 switch (lockt->lt_type) {
4107 file_lock.fl_type = F_RDLCK;
4110 case NFS4_WRITEW_LT:
4111 file_lock.fl_type = F_WRLCK;
4114 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4115 status = nfserr_inval;
4119 lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4121 file_lock.fl_owner = (fl_owner_t)lo;
4122 file_lock.fl_pid = current->tgid;
4123 file_lock.fl_flags = FL_POSIX;
4125 file_lock.fl_start = lockt->lt_offset;
4126 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4128 nfs4_transform_lock_offset(&file_lock);
4131 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4133 status = nfserrno(error);
4136 if (file_lock.fl_type != F_UNLCK) {
4137 status = nfserr_denied;
4138 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4141 nfs4_unlock_state();
4146 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4147 struct nfsd4_locku *locku)
4149 struct nfs4_ol_stateid *stp;
4150 struct file *filp = NULL;
4151 struct file_lock file_lock;
4155 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4156 (long long) locku->lu_offset,
4157 (long long) locku->lu_length);
4159 if (check_lock_length(locku->lu_offset, locku->lu_length))
4160 return nfserr_inval;
4164 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4165 &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4168 filp = find_any_file(stp->st_file);
4170 status = nfserr_lock_range;
4174 locks_init_lock(&file_lock);
4175 file_lock.fl_type = F_UNLCK;
4176 file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4177 file_lock.fl_pid = current->tgid;
4178 file_lock.fl_file = filp;
4179 file_lock.fl_flags = FL_POSIX;
4180 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4181 file_lock.fl_start = locku->lu_offset;
4183 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4184 nfs4_transform_lock_offset(&file_lock);
4187 * Try to unlock the file in the VFS.
4189 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4191 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4195 * OK, unlock succeeded; the only thing left to do is update the stateid.
4197 update_stateid(&stp->st_stid.sc_stateid);
4198 memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4201 nfs4_unlock_state();
4205 status = nfserrno(err);
4211 * 1: locks held by lockowner
4212 * 0: no locks held by lockowner
4215 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4217 struct file_lock **flpp;
4218 struct inode *inode = filp->fi_inode;
4222 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4223 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4234 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4235 struct nfsd4_compound_state *cstate,
4236 struct nfsd4_release_lockowner *rlockowner)
4238 clientid_t *clid = &rlockowner->rl_clientid;
4239 struct nfs4_stateowner *sop;
4240 struct nfs4_lockowner *lo;
4241 struct nfs4_ol_stateid *stp;
4242 struct xdr_netobj *owner = &rlockowner->rl_owner;
4243 struct list_head matches;
4247 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4248 clid->cl_boot, clid->cl_id);
4250 /* XXX check for lease expiration */
4252 status = nfserr_stale_clientid;
4253 if (STALE_CLIENTID(clid))
4258 status = nfserr_locks_held;
4259 /* XXX: we're doing a linear search through all the lockowners.
4260 * Yipes! For now we'll just hope clients aren't really using
4261 * release_lockowner much, but eventually we have to fix these
4262 * data structures. */
4263 INIT_LIST_HEAD(&matches);
4264 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4265 list_for_each_entry(sop, &lock_ownerstr_hashtbl[i], so_strhash) {
4266 if (!same_owner_str(sop, owner, clid))
4268 list_for_each_entry(stp, &sop->so_stateids,
4270 lo = lockowner(sop);
4271 if (check_for_locks(stp->st_file, lo))
4273 list_add(&lo->lo_list, &matches);
4277 /* Clients probably won't expect us to return with some (but not all)
4278 * of the lockowner state released; so don't release any until all
4279 * have been checked. */
4281 while (!list_empty(&matches)) {
4282 lo = list_entry(matches.next, struct nfs4_lockowner,
4284 /* unhash_stateowner deletes so_perclient only
4285 * for openowners. */
4286 list_del(&lo->lo_list);
4287 release_lockowner(lo);
4290 nfs4_unlock_state();
4294 static inline struct nfs4_client_reclaim *
4297 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4301 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4303 unsigned int strhashval = clientstr_hashval(name);
4304 struct nfs4_client *clp;
4306 clp = find_confirmed_client_by_str(name, strhashval);
4311 * failure => all reset bets are off, nfserr_no_grace...
4314 nfs4_client_to_reclaim(const char *name)
4316 unsigned int strhashval;
4317 struct nfs4_client_reclaim *crp = NULL;
4319 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4320 crp = alloc_reclaim();
4323 strhashval = clientstr_hashval(name);
4324 INIT_LIST_HEAD(&crp->cr_strhash);
4325 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4326 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4327 reclaim_str_hashtbl_size++;
4332 nfs4_release_reclaim(void)
4334 struct nfs4_client_reclaim *crp = NULL;
4337 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4338 while (!list_empty(&reclaim_str_hashtbl[i])) {
4339 crp = list_entry(reclaim_str_hashtbl[i].next,
4340 struct nfs4_client_reclaim, cr_strhash);
4341 list_del(&crp->cr_strhash);
4343 reclaim_str_hashtbl_size--;
4346 BUG_ON(reclaim_str_hashtbl_size);
4350 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4351 static struct nfs4_client_reclaim *
4352 nfs4_find_reclaim_client(clientid_t *clid)
4354 unsigned int strhashval;
4355 struct nfs4_client *clp;
4356 struct nfs4_client_reclaim *crp = NULL;
4359 /* find clientid in conf_id_hashtbl */
4360 clp = find_confirmed_client(clid);
4364 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4365 clp->cl_name.len, clp->cl_name.data,
4368 /* find clp->cl_name in reclaim_str_hashtbl */
4369 strhashval = clientstr_hashval(clp->cl_recdir);
4370 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4371 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4379 * Called from OPEN. Look for clientid in reclaim list.
4382 nfs4_check_open_reclaim(clientid_t *clid)
4384 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4387 /* initialization to perform at module load time: */
4390 nfs4_state_init(void)
4394 status = nfsd4_init_slabs();
4397 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4398 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4399 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4400 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4401 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4402 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4404 for (i = 0; i < SESSION_HASH_SIZE; i++)
4405 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4406 for (i = 0; i < FILE_HASH_SIZE; i++) {
4407 INIT_LIST_HEAD(&file_hashtbl[i]);
4409 for (i = 0; i < OPEN_OWNER_HASH_SIZE; i++) {
4410 INIT_LIST_HEAD(&open_ownerstr_hashtbl[i]);
4412 idr_init(&stateids);
4413 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4414 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4416 memset(&onestateid, ~0, sizeof(stateid_t));
4417 INIT_LIST_HEAD(&close_lru);
4418 INIT_LIST_HEAD(&client_lru);
4419 INIT_LIST_HEAD(&del_recall_lru);
4420 reclaim_str_hashtbl_size = 0;
4425 nfsd4_load_reboot_recovery_data(void)
4430 nfsd4_init_recdir();
4431 status = nfsd4_recdir_load();
4432 nfs4_unlock_state();
4434 printk("NFSD: Failure reading reboot recovery data\n");
4438 * Since the lifetime of a delegation isn't limited to that of an open, a
4439 * client may quite reasonably hang on to a delegation as long as it has
4440 * the inode cached. This becomes an obvious problem the first time a
4441 * client's inode cache approaches the size of the server's total memory.
4443 * For now we avoid this problem by imposing a hard limit on the number
4444 * of delegations, which varies according to the server's memory size.
4447 set_max_delegations(void)
4450 * Allow at most 4 delegations per megabyte of RAM. Quick
4451 * estimates suggest that in the worst case (where every delegation
4452 * is for a different inode), a delegation could take about 1.5K,
4453 * giving a worst case usage of about 6% of memory.
4455 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4458 /* initialization to perform when the nfsd service is started: */
4461 __nfs4_state_start(void)
4465 boot_time = get_seconds();
4466 locks_start_grace(&nfsd4_manager);
4467 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4469 ret = set_callback_cred();
4472 laundry_wq = create_singlethread_workqueue("nfsd4");
4473 if (laundry_wq == NULL)
4475 ret = nfsd4_create_callback_queue();
4477 goto out_free_laundry;
4478 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4479 set_max_delegations();
4482 destroy_workqueue(laundry_wq);
4487 nfs4_state_start(void)
4489 nfsd4_load_reboot_recovery_data();
4490 return __nfs4_state_start();
4494 __nfs4_state_shutdown(void)
4497 struct nfs4_client *clp = NULL;
4498 struct nfs4_delegation *dp = NULL;
4499 struct list_head *pos, *next, reaplist;
4501 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4502 while (!list_empty(&conf_id_hashtbl[i])) {
4503 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4506 while (!list_empty(&unconf_str_hashtbl[i])) {
4507 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4511 INIT_LIST_HEAD(&reaplist);
4512 spin_lock(&recall_lock);
4513 list_for_each_safe(pos, next, &del_recall_lru) {
4514 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4515 list_move(&dp->dl_recall_lru, &reaplist);
4517 spin_unlock(&recall_lock);
4518 list_for_each_safe(pos, next, &reaplist) {
4519 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4520 list_del_init(&dp->dl_recall_lru);
4521 unhash_delegation(dp);
4524 nfsd4_shutdown_recdir();
4528 nfs4_state_shutdown(void)
4530 cancel_delayed_work_sync(&laundromat_work);
4531 destroy_workqueue(laundry_wq);
4532 locks_end_grace(&nfsd4_manager);
4534 nfs4_release_reclaim();
4535 __nfs4_state_shutdown();
4536 nfs4_unlock_state();
4537 nfsd4_destroy_callback_queue();