2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/sunrpc/svcauth_gss.h>
42 #include <linux/sunrpc/clnt.h>
46 #define NFSDDBG_FACILITY NFSDDBG_PROC
49 time_t nfsd4_lease = 90; /* default lease time */
50 time_t nfsd4_grace = 90;
51 static time_t boot_time;
52 static stateid_t zerostateid; /* bits all 0 */
53 static stateid_t onestateid; /* bits all 1 */
54 static u64 current_sessionid = 1;
56 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
57 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
59 /* forward declarations */
60 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
64 /* Currently used for almost all code touching nfsv4 state: */
65 static DEFINE_MUTEX(client_mutex);
68 * Currently used for the del_recall_lru and file hash table. In an
69 * effort to decrease the scope of the client_mutex, this spinlock may
70 * eventually cover more:
72 static DEFINE_SPINLOCK(recall_lock);
74 static struct kmem_cache *openowner_slab = NULL;
75 static struct kmem_cache *lockowner_slab = NULL;
76 static struct kmem_cache *file_slab = NULL;
77 static struct kmem_cache *stateid_slab = NULL;
78 static struct kmem_cache *deleg_slab = NULL;
83 mutex_lock(&client_mutex);
87 nfs4_unlock_state(void)
89 mutex_unlock(&client_mutex);
93 opaque_hashval(const void *ptr, int nbytes)
95 unsigned char *cptr = (unsigned char *) ptr;
105 static struct list_head del_recall_lru;
108 put_nfs4_file(struct nfs4_file *fi)
110 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
111 list_del(&fi->fi_hash);
112 spin_unlock(&recall_lock);
114 kmem_cache_free(file_slab, fi);
119 get_nfs4_file(struct nfs4_file *fi)
121 atomic_inc(&fi->fi_ref);
124 static int num_delegations;
125 unsigned int max_delegations;
128 * Open owner state (share locks)
131 /* hash tables for open owners */
132 #define OPEN_OWNER_HASH_BITS 8
133 #define OPEN_OWNER_HASH_SIZE (1 << OPEN_OWNER_HASH_BITS)
134 #define OPEN_OWNER_HASH_MASK (OPEN_OWNER_HASH_SIZE - 1)
136 static unsigned int open_ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
140 ret = opaque_hashval(ownername->data, ownername->len);
142 return ret & OPEN_OWNER_HASH_MASK;
145 static struct list_head open_ownerstr_hashtbl[OPEN_OWNER_HASH_SIZE];
147 /* hash table for nfs4_file */
148 #define FILE_HASH_BITS 8
149 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
151 static unsigned int file_hashval(struct inode *ino)
153 /* XXX: why are we hashing on inode pointer, anyway? */
154 return hash_ptr(ino, FILE_HASH_BITS);
157 static struct list_head file_hashtbl[FILE_HASH_SIZE];
159 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
161 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
162 atomic_inc(&fp->fi_access[oflag]);
165 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
167 if (oflag == O_RDWR) {
168 __nfs4_file_get_access(fp, O_RDONLY);
169 __nfs4_file_get_access(fp, O_WRONLY);
171 __nfs4_file_get_access(fp, oflag);
174 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
176 if (fp->fi_fds[oflag]) {
177 fput(fp->fi_fds[oflag]);
178 fp->fi_fds[oflag] = NULL;
182 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
184 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
185 nfs4_file_put_fd(fp, oflag);
187 * It's also safe to get rid of the RDWR open *if*
188 * we no longer have need of the other kind of access
189 * or if we already have the other kind of open:
191 if (fp->fi_fds[1-oflag]
192 || atomic_read(&fp->fi_access[1 - oflag]) == 0)
193 nfs4_file_put_fd(fp, O_RDWR);
197 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
199 if (oflag == O_RDWR) {
200 __nfs4_file_put_access(fp, O_RDONLY);
201 __nfs4_file_put_access(fp, O_WRONLY);
203 __nfs4_file_put_access(fp, oflag);
206 static inline int get_new_stid(struct nfs4_stid *stid)
208 static int min_stateid = 0;
209 struct idr *stateids = &stid->sc_client->cl_stateids;
213 if (!idr_pre_get(stateids, GFP_KERNEL))
216 error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
218 * All this code is currently serialized; the preallocation
219 * above should still be ours:
223 * It shouldn't be a problem to reuse an opaque stateid value.
224 * I don't think it is for 4.1. But with 4.0 I worry that, for
225 * example, a stray write retransmission could be accepted by
226 * the server when it should have been rejected. Therefore,
227 * adopt a trick from the sctp code to attempt to maximize the
228 * amount of time until an id is reused, by ensuring they always
229 * "increase" (mod INT_MAX):
232 min_stateid = new_stid+1;
233 if (min_stateid == INT_MAX)
238 static inline __be32 init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
240 stateid_t *s = &stid->sc_stateid;
243 stid->sc_type = type;
244 stid->sc_client = cl;
245 s->si_opaque.so_clid = cl->cl_clientid;
246 new_id = get_new_stid(stid);
248 return nfserr_jukebox;
249 s->si_opaque.so_id = (u32)new_id;
250 /* Will be incremented before return to client: */
251 s->si_generation = 0;
255 static struct nfs4_delegation *
256 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
258 struct nfs4_delegation *dp;
259 struct nfs4_file *fp = stp->st_file;
262 dprintk("NFSD alloc_init_deleg\n");
264 * Major work on the lease subsystem (for example, to support
265 * calbacks on stat) will be required before we can support
266 * write delegations properly.
268 if (type != NFS4_OPEN_DELEGATE_READ)
270 if (fp->fi_had_conflict)
272 if (num_delegations > max_delegations)
274 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
277 status = init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
279 kmem_cache_free(deleg_slab, dp);
283 * delegation seqid's are never incremented. The 4.1 special
284 * meaning of seqid 0 isn't meaningful, really, but let's avoid
285 * 0 anyway just for consistency and use 1:
287 dp->dl_stid.sc_stateid.si_generation = 1;
289 INIT_LIST_HEAD(&dp->dl_perfile);
290 INIT_LIST_HEAD(&dp->dl_perclnt);
291 INIT_LIST_HEAD(&dp->dl_recall_lru);
295 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
297 atomic_set(&dp->dl_count, 1);
298 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
303 nfs4_put_delegation(struct nfs4_delegation *dp)
305 if (atomic_dec_and_test(&dp->dl_count)) {
306 dprintk("NFSD: freeing dp %p\n",dp);
307 put_nfs4_file(dp->dl_file);
308 kmem_cache_free(deleg_slab, dp);
313 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
315 if (atomic_dec_and_test(&fp->fi_delegees)) {
316 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
318 fput(fp->fi_deleg_file);
319 fp->fi_deleg_file = NULL;
323 static void unhash_stid(struct nfs4_stid *s)
325 struct idr *stateids = &s->sc_client->cl_stateids;
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);
934 dprintk("renewing client (clientid %08x/%08x)\n",
935 clp->cl_clientid.cl_boot,
936 clp->cl_clientid.cl_id);
937 list_move_tail(&clp->cl_lru, &client_lru);
938 clp->cl_time = get_seconds();
942 renew_client(struct nfs4_client *clp)
944 spin_lock(&client_lock);
945 renew_client_locked(clp);
946 spin_unlock(&client_lock);
949 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
951 STALE_CLIENTID(clientid_t *clid)
953 if (clid->cl_boot == boot_time)
955 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
956 clid->cl_boot, clid->cl_id, boot_time);
961 * XXX Should we use a slab cache ?
962 * This type of memory management is somewhat inefficient, but we use it
963 * anyway since SETCLIENTID is not a common operation.
965 static struct nfs4_client *alloc_client(struct xdr_netobj name)
967 struct nfs4_client *clp;
969 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
972 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
973 if (clp->cl_name.data == NULL) {
977 memcpy(clp->cl_name.data, name.data, name.len);
978 clp->cl_name.len = name.len;
983 free_client(struct nfs4_client *clp)
985 while (!list_empty(&clp->cl_sessions)) {
986 struct nfsd4_session *ses;
987 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
989 list_del(&ses->se_perclnt);
990 nfsd4_put_session(ses);
992 if (clp->cl_cred.cr_group_info)
993 put_group_info(clp->cl_cred.cr_group_info);
994 kfree(clp->cl_principal);
995 kfree(clp->cl_name.data);
1000 release_session_client(struct nfsd4_session *session)
1002 struct nfs4_client *clp = session->se_client;
1004 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1006 if (is_client_expired(clp)) {
1008 session->se_client = NULL;
1010 renew_client_locked(clp);
1011 spin_unlock(&client_lock);
1014 /* must be called under the client_lock */
1016 unhash_client_locked(struct nfs4_client *clp)
1018 struct nfsd4_session *ses;
1020 mark_client_expired(clp);
1021 list_del(&clp->cl_lru);
1022 spin_lock(&clp->cl_lock);
1023 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1024 list_del_init(&ses->se_hash);
1025 spin_unlock(&clp->cl_lock);
1029 expire_client(struct nfs4_client *clp)
1031 struct nfs4_openowner *oo;
1032 struct nfs4_delegation *dp;
1033 struct list_head reaplist;
1035 INIT_LIST_HEAD(&reaplist);
1036 spin_lock(&recall_lock);
1037 while (!list_empty(&clp->cl_delegations)) {
1038 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1039 list_del_init(&dp->dl_perclnt);
1040 list_move(&dp->dl_recall_lru, &reaplist);
1042 spin_unlock(&recall_lock);
1043 while (!list_empty(&reaplist)) {
1044 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1045 list_del_init(&dp->dl_recall_lru);
1046 unhash_delegation(dp);
1048 while (!list_empty(&clp->cl_openowners)) {
1049 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1050 release_openowner(oo);
1052 nfsd4_shutdown_callback(clp);
1053 if (clp->cl_cb_conn.cb_xprt)
1054 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1055 list_del(&clp->cl_idhash);
1056 list_del(&clp->cl_strhash);
1057 spin_lock(&client_lock);
1058 unhash_client_locked(clp);
1059 if (atomic_read(&clp->cl_refcount) == 0)
1061 spin_unlock(&client_lock);
1064 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1066 memcpy(target->cl_verifier.data, source->data,
1067 sizeof(target->cl_verifier.data));
1070 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1072 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1073 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1076 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1078 target->cr_uid = source->cr_uid;
1079 target->cr_gid = source->cr_gid;
1080 target->cr_group_info = source->cr_group_info;
1081 get_group_info(target->cr_group_info);
1084 static int same_name(const char *n1, const char *n2)
1086 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1090 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1092 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1096 same_clid(clientid_t *cl1, clientid_t *cl2)
1098 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1101 /* XXX what about NGROUP */
1103 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1105 return cr1->cr_uid == cr2->cr_uid;
1108 static void gen_clid(struct nfs4_client *clp)
1110 static u32 current_clientid = 1;
1112 clp->cl_clientid.cl_boot = boot_time;
1113 clp->cl_clientid.cl_id = current_clientid++;
1116 static void gen_confirm(struct nfs4_client *clp)
1121 p = (u32 *)clp->cl_confirm.data;
1122 *p++ = get_seconds();
1126 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1128 return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1131 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1133 struct nfs4_stid *s;
1135 s = find_stateid(cl, t);
1138 if (typemask & s->sc_type)
1143 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1144 struct svc_rqst *rqstp, nfs4_verifier *verf)
1146 struct nfs4_client *clp;
1147 struct sockaddr *sa = svc_addr(rqstp);
1150 clp = alloc_client(name);
1154 INIT_LIST_HEAD(&clp->cl_sessions);
1156 princ = svc_gss_principal(rqstp);
1158 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1159 if (clp->cl_principal == NULL) {
1165 idr_init(&clp->cl_stateids);
1166 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1167 atomic_set(&clp->cl_refcount, 0);
1168 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1169 INIT_LIST_HEAD(&clp->cl_idhash);
1170 INIT_LIST_HEAD(&clp->cl_strhash);
1171 INIT_LIST_HEAD(&clp->cl_openowners);
1172 INIT_LIST_HEAD(&clp->cl_delegations);
1173 INIT_LIST_HEAD(&clp->cl_lru);
1174 INIT_LIST_HEAD(&clp->cl_callbacks);
1175 spin_lock_init(&clp->cl_lock);
1176 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1177 clp->cl_time = get_seconds();
1178 clear_bit(0, &clp->cl_cb_slot_busy);
1179 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1180 copy_verf(clp, verf);
1181 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1182 clp->cl_flavor = rqstp->rq_flavor;
1183 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1185 clp->cl_cb_session = NULL;
1190 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1192 unsigned int idhashval;
1194 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1195 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1196 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1201 move_to_confirmed(struct nfs4_client *clp)
1203 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1204 unsigned int strhashval;
1206 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1207 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1208 strhashval = clientstr_hashval(clp->cl_recdir);
1209 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1213 static struct nfs4_client *
1214 find_confirmed_client(clientid_t *clid)
1216 struct nfs4_client *clp;
1217 unsigned int idhashval = clientid_hashval(clid->cl_id);
1219 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1220 if (same_clid(&clp->cl_clientid, clid)) {
1228 static struct nfs4_client *
1229 find_unconfirmed_client(clientid_t *clid)
1231 struct nfs4_client *clp;
1232 unsigned int idhashval = clientid_hashval(clid->cl_id);
1234 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1235 if (same_clid(&clp->cl_clientid, clid))
1241 static bool clp_used_exchangeid(struct nfs4_client *clp)
1243 return clp->cl_exchange_flags != 0;
1246 static struct nfs4_client *
1247 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1249 struct nfs4_client *clp;
1251 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1252 if (same_name(clp->cl_recdir, dname))
1258 static struct nfs4_client *
1259 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1261 struct nfs4_client *clp;
1263 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1264 if (same_name(clp->cl_recdir, dname))
1271 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1273 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1274 struct sockaddr *sa = svc_addr(rqstp);
1275 u32 scopeid = rpc_get_scope_id(sa);
1276 unsigned short expected_family;
1278 /* Currently, we only support tcp and tcp6 for the callback channel */
1279 if (se->se_callback_netid_len == 3 &&
1280 !memcmp(se->se_callback_netid_val, "tcp", 3))
1281 expected_family = AF_INET;
1282 else if (se->se_callback_netid_len == 4 &&
1283 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1284 expected_family = AF_INET6;
1288 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1289 se->se_callback_addr_len,
1290 (struct sockaddr *)&conn->cb_addr,
1291 sizeof(conn->cb_addr));
1293 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1296 if (conn->cb_addr.ss_family == AF_INET6)
1297 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1299 conn->cb_prog = se->se_callback_prog;
1300 conn->cb_ident = se->se_callback_ident;
1301 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1304 conn->cb_addr.ss_family = AF_UNSPEC;
1305 conn->cb_addrlen = 0;
1306 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1307 "will not receive delegations\n",
1308 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1314 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1317 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1319 struct nfsd4_slot *slot = resp->cstate.slot;
1322 dprintk("--> %s slot %p\n", __func__, slot);
1324 slot->sl_opcnt = resp->opcnt;
1325 slot->sl_status = resp->cstate.status;
1327 if (nfsd4_not_cached(resp)) {
1328 slot->sl_datalen = 0;
1331 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1332 base = (char *)resp->cstate.datap -
1333 (char *)resp->xbuf->head[0].iov_base;
1334 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1336 WARN("%s: sessions DRC could not cache compound\n", __func__);
1341 * Encode the replay sequence operation from the slot values.
1342 * If cachethis is FALSE encode the uncached rep error on the next
1343 * operation which sets resp->p and increments resp->opcnt for
1344 * nfs4svc_encode_compoundres.
1348 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1349 struct nfsd4_compoundres *resp)
1351 struct nfsd4_op *op;
1352 struct nfsd4_slot *slot = resp->cstate.slot;
1354 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1355 resp->opcnt, resp->cstate.slot->sl_cachethis);
1357 /* Encode the replayed sequence operation */
1358 op = &args->ops[resp->opcnt - 1];
1359 nfsd4_encode_operation(resp, op);
1361 /* Return nfserr_retry_uncached_rep in next operation. */
1362 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1363 op = &args->ops[resp->opcnt++];
1364 op->status = nfserr_retry_uncached_rep;
1365 nfsd4_encode_operation(resp, op);
1371 * The sequence operation is not cached because we can use the slot and
1375 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1376 struct nfsd4_sequence *seq)
1378 struct nfsd4_slot *slot = resp->cstate.slot;
1381 dprintk("--> %s slot %p\n", __func__, slot);
1383 /* Either returns 0 or nfserr_retry_uncached */
1384 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1385 if (status == nfserr_retry_uncached_rep)
1388 /* The sequence operation has been encoded, cstate->datap set. */
1389 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1391 resp->opcnt = slot->sl_opcnt;
1392 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1393 status = slot->sl_status;
1399 * Set the exchange_id flags returned by the server.
1402 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1404 /* pNFS is not supported */
1405 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1407 /* Referrals are supported, Migration is not. */
1408 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1410 /* set the wire flags to return to client. */
1411 clid->flags = new->cl_exchange_flags;
1415 nfsd4_exchange_id(struct svc_rqst *rqstp,
1416 struct nfsd4_compound_state *cstate,
1417 struct nfsd4_exchange_id *exid)
1419 struct nfs4_client *unconf, *conf, *new;
1421 unsigned int strhashval;
1422 char dname[HEXDIR_LEN];
1423 char addr_str[INET6_ADDRSTRLEN];
1424 nfs4_verifier verf = exid->verifier;
1425 struct sockaddr *sa = svc_addr(rqstp);
1427 rpc_ntop(sa, addr_str, sizeof(addr_str));
1428 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1429 "ip_addr=%s flags %x, spa_how %d\n",
1430 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1431 addr_str, exid->flags, exid->spa_how);
1433 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1434 return nfserr_inval;
1436 /* Currently only support SP4_NONE */
1437 switch (exid->spa_how) {
1441 return nfserr_serverfault;
1443 BUG(); /* checked by xdr code */
1445 return nfserr_serverfault; /* no excuse :-/ */
1448 status = nfs4_make_rec_clidname(dname, &exid->clname);
1453 strhashval = clientstr_hashval(dname);
1458 conf = find_confirmed_client_by_str(dname, strhashval);
1460 if (!clp_used_exchangeid(conf)) {
1461 status = nfserr_clid_inuse; /* XXX: ? */
1464 if (!same_verf(&verf, &conf->cl_verifier)) {
1465 /* 18.35.4 case 8 */
1466 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1467 status = nfserr_not_same;
1470 /* Client reboot: destroy old state */
1471 expire_client(conf);
1474 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1475 /* 18.35.4 case 9 */
1476 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1477 status = nfserr_perm;
1480 expire_client(conf);
1484 * Set bit when the owner id and verifier map to an already
1485 * confirmed client id (18.35.3).
1487 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1490 * Falling into 18.35.4 case 2, possible router replay.
1491 * Leave confirmed record intact and return same result.
1493 copy_verf(conf, &verf);
1498 /* 18.35.4 case 7 */
1499 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1500 status = nfserr_noent;
1504 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1507 * Possible retry or client restart. Per 18.35.4 case 4,
1508 * a new unconfirmed record should be generated regardless
1509 * of whether any properties have changed.
1511 expire_client(unconf);
1516 new = create_client(exid->clname, dname, rqstp, &verf);
1518 status = nfserr_jukebox;
1523 add_to_unconfirmed(new, strhashval);
1525 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1526 exid->clientid.cl_id = new->cl_clientid.cl_id;
1529 nfsd4_set_ex_flags(new, exid);
1531 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1532 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1536 nfs4_unlock_state();
1538 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1543 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1545 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1548 /* The slot is in use, and no response has been sent. */
1550 if (seqid == slot_seqid)
1551 return nfserr_jukebox;
1553 return nfserr_seq_misordered;
1556 if (likely(seqid == slot_seqid + 1))
1559 if (seqid == slot_seqid)
1560 return nfserr_replay_cache;
1562 if (seqid == 1 && (slot_seqid + 1) == 0)
1564 /* Misordered replay or misordered new request */
1565 return nfserr_seq_misordered;
1569 * Cache the create session result into the create session single DRC
1570 * slot cache by saving the xdr structure. sl_seqid has been set.
1571 * Do this for solo or embedded create session operations.
1574 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1575 struct nfsd4_clid_slot *slot, int nfserr)
1577 slot->sl_status = nfserr;
1578 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1582 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1583 struct nfsd4_clid_slot *slot)
1585 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1586 return slot->sl_status;
1589 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1590 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1591 1 + /* MIN tag is length with zero, only length */ \
1592 3 + /* version, opcount, opcode */ \
1593 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1594 /* seqid, slotID, slotID, cache */ \
1595 4 ) * sizeof(__be32))
1597 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1598 2 + /* verifier: AUTH_NULL, length 0 */\
1600 1 + /* MIN tag is length with zero, only length */ \
1601 3 + /* opcount, opcode, opstatus*/ \
1602 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1603 /* seqid, slotID, slotID, slotID, status */ \
1604 5 ) * sizeof(__be32))
1606 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1608 return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1609 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1613 nfsd4_create_session(struct svc_rqst *rqstp,
1614 struct nfsd4_compound_state *cstate,
1615 struct nfsd4_create_session *cr_ses)
1617 struct sockaddr *sa = svc_addr(rqstp);
1618 struct nfs4_client *conf, *unconf;
1619 struct nfsd4_session *new;
1620 struct nfsd4_clid_slot *cs_slot = NULL;
1621 bool confirm_me = false;
1624 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1625 return nfserr_inval;
1628 unconf = find_unconfirmed_client(&cr_ses->clientid);
1629 conf = find_confirmed_client(&cr_ses->clientid);
1632 cs_slot = &conf->cl_cs_slot;
1633 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1634 if (status == nfserr_replay_cache) {
1635 dprintk("Got a create_session replay! seqid= %d\n",
1637 /* Return the cached reply status */
1638 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1640 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1641 status = nfserr_seq_misordered;
1642 dprintk("Sequence misordered!\n");
1643 dprintk("Expected seqid= %d but got seqid= %d\n",
1644 cs_slot->sl_seqid, cr_ses->seqid);
1647 } else if (unconf) {
1648 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1649 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1650 status = nfserr_clid_inuse;
1654 cs_slot = &unconf->cl_cs_slot;
1655 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1657 /* an unconfirmed replay returns misordered */
1658 status = nfserr_seq_misordered;
1665 status = nfserr_stale_clientid;
1670 * XXX: we should probably set this at creation time, and check
1671 * for consistent minorversion use throughout:
1673 conf->cl_minorversion = 1;
1675 * We do not support RDMA or persistent sessions
1677 cr_ses->flags &= ~SESSION4_PERSIST;
1678 cr_ses->flags &= ~SESSION4_RDMA;
1680 status = nfserr_toosmall;
1681 if (check_forechannel_attrs(cr_ses->fore_channel))
1684 status = nfserr_jukebox;
1685 new = alloc_init_session(rqstp, conf, cr_ses);
1689 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1690 NFS4_MAX_SESSIONID_LEN);
1691 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1692 sizeof(struct nfsd4_channel_attrs));
1693 cs_slot->sl_seqid++;
1694 cr_ses->seqid = cs_slot->sl_seqid;
1696 /* cache solo and embedded create sessions under the state lock */
1697 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1699 move_to_confirmed(conf);
1701 nfs4_unlock_state();
1702 dprintk("%s returns %d\n", __func__, ntohl(status));
1706 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1708 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1709 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1711 return argp->opcnt == resp->opcnt;
1714 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1717 case NFS4_CDFC4_FORE:
1718 case NFS4_CDFC4_BACK:
1720 case NFS4_CDFC4_FORE_OR_BOTH:
1721 case NFS4_CDFC4_BACK_OR_BOTH:
1722 *dir = NFS4_CDFC4_BOTH;
1725 return nfserr_inval;
1728 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1729 struct nfsd4_compound_state *cstate,
1730 struct nfsd4_bind_conn_to_session *bcts)
1734 if (!nfsd4_last_compound_op(rqstp))
1735 return nfserr_not_only_op;
1736 spin_lock(&client_lock);
1737 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1738 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1739 * client_lock iself: */
1740 if (cstate->session) {
1741 nfsd4_get_session(cstate->session);
1742 atomic_inc(&cstate->session->se_client->cl_refcount);
1744 spin_unlock(&client_lock);
1745 if (!cstate->session)
1746 return nfserr_badsession;
1748 status = nfsd4_map_bcts_dir(&bcts->dir);
1750 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1754 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1758 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1762 nfsd4_destroy_session(struct svc_rqst *r,
1763 struct nfsd4_compound_state *cstate,
1764 struct nfsd4_destroy_session *sessionid)
1766 struct nfsd4_session *ses;
1767 u32 status = nfserr_badsession;
1770 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1771 * - Should we return nfserr_back_chan_busy if waiting for
1772 * callbacks on to-be-destroyed session?
1773 * - Do we need to clear any callback info from previous session?
1776 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1777 if (!nfsd4_last_compound_op(r))
1778 return nfserr_not_only_op;
1780 dump_sessionid(__func__, &sessionid->sessionid);
1781 spin_lock(&client_lock);
1782 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1784 spin_unlock(&client_lock);
1788 unhash_session(ses);
1789 spin_unlock(&client_lock);
1792 nfsd4_probe_callback_sync(ses->se_client);
1793 nfs4_unlock_state();
1795 nfsd4_del_conns(ses);
1797 nfsd4_put_session(ses);
1800 dprintk("%s returns %d\n", __func__, ntohl(status));
1804 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1806 struct nfsd4_conn *c;
1808 list_for_each_entry(c, &s->se_conns, cn_persession) {
1809 if (c->cn_xprt == xpt) {
1816 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1818 struct nfs4_client *clp = ses->se_client;
1819 struct nfsd4_conn *c;
1822 spin_lock(&clp->cl_lock);
1823 c = __nfsd4_find_conn(new->cn_xprt, ses);
1825 spin_unlock(&clp->cl_lock);
1829 __nfsd4_hash_conn(new, ses);
1830 spin_unlock(&clp->cl_lock);
1831 ret = nfsd4_register_conn(new);
1833 /* oops; xprt is already down: */
1834 nfsd4_conn_lost(&new->cn_xpt_user);
1838 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1840 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1842 return args->opcnt > session->se_fchannel.maxops;
1845 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1846 struct nfsd4_session *session)
1848 struct xdr_buf *xb = &rqstp->rq_arg;
1850 return xb->len > session->se_fchannel.maxreq_sz;
1854 nfsd4_sequence(struct svc_rqst *rqstp,
1855 struct nfsd4_compound_state *cstate,
1856 struct nfsd4_sequence *seq)
1858 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1859 struct nfsd4_session *session;
1860 struct nfsd4_slot *slot;
1861 struct nfsd4_conn *conn;
1864 if (resp->opcnt != 1)
1865 return nfserr_sequence_pos;
1868 * Will be either used or freed by nfsd4_sequence_check_conn
1871 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1873 return nfserr_jukebox;
1875 spin_lock(&client_lock);
1876 status = nfserr_badsession;
1877 session = find_in_sessionid_hashtbl(&seq->sessionid);
1881 status = nfserr_too_many_ops;
1882 if (nfsd4_session_too_many_ops(rqstp, session))
1885 status = nfserr_req_too_big;
1886 if (nfsd4_request_too_big(rqstp, session))
1889 status = nfserr_badslot;
1890 if (seq->slotid >= session->se_fchannel.maxreqs)
1893 slot = session->se_slots[seq->slotid];
1894 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1896 /* We do not negotiate the number of slots yet, so set the
1897 * maxslots to the session maxreqs which is used to encode
1898 * sr_highest_slotid and the sr_target_slot id to maxslots */
1899 seq->maxslots = session->se_fchannel.maxreqs;
1901 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1902 if (status == nfserr_replay_cache) {
1903 cstate->slot = slot;
1904 cstate->session = session;
1905 /* Return the cached reply status and set cstate->status
1906 * for nfsd4_proc_compound processing */
1907 status = nfsd4_replay_cache_entry(resp, seq);
1908 cstate->status = nfserr_replay_cache;
1914 nfsd4_sequence_check_conn(conn, session);
1917 /* Success! bump slot seqid */
1918 slot->sl_inuse = true;
1919 slot->sl_seqid = seq->seqid;
1920 slot->sl_cachethis = seq->cachethis;
1922 cstate->slot = slot;
1923 cstate->session = session;
1926 /* Hold a session reference until done processing the compound. */
1927 if (cstate->session) {
1928 struct nfs4_client *clp = session->se_client;
1930 nfsd4_get_session(cstate->session);
1931 atomic_inc(&clp->cl_refcount);
1932 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1933 seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1936 spin_unlock(&client_lock);
1937 dprintk("%s: return %d\n", __func__, ntohl(status));
1942 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1946 if (rc->rca_one_fs) {
1947 if (!cstate->current_fh.fh_dentry)
1948 return nfserr_nofilehandle;
1950 * We don't take advantage of the rca_one_fs case.
1951 * That's OK, it's optional, we can safely ignore it.
1957 status = nfserr_complete_already;
1958 if (cstate->session->se_client->cl_firststate)
1961 status = nfserr_stale_clientid;
1962 if (is_client_expired(cstate->session->se_client))
1964 * The following error isn't really legal.
1965 * But we only get here if the client just explicitly
1966 * destroyed the client. Surely it no longer cares what
1967 * error it gets back on an operation for the dead
1973 nfsd4_create_clid_dir(cstate->session->se_client);
1975 nfs4_unlock_state();
1980 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1981 struct nfsd4_setclientid *setclid)
1983 struct xdr_netobj clname = setclid->se_name;
1984 nfs4_verifier clverifier = setclid->se_verf;
1985 unsigned int strhashval;
1986 struct nfs4_client *conf, *unconf, *new;
1988 char dname[HEXDIR_LEN];
1990 status = nfs4_make_rec_clidname(dname, &clname);
1995 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1996 * We get here on a DRC miss.
1999 strhashval = clientstr_hashval(dname);
2002 conf = find_confirmed_client_by_str(dname, strhashval);
2004 /* RFC 3530 14.2.33 CASE 0: */
2005 status = nfserr_clid_inuse;
2006 if (clp_used_exchangeid(conf))
2008 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2009 char addr_str[INET6_ADDRSTRLEN];
2010 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2012 dprintk("NFSD: setclientid: string in use by client "
2013 "at %s\n", addr_str);
2018 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
2019 * has a description of SETCLIENTID request processing consisting
2020 * of 5 bullet points, labeled as CASE0 - CASE4 below.
2022 unconf = find_unconfirmed_client_by_str(dname, strhashval);
2023 status = nfserr_jukebox;
2026 * RFC 3530 14.2.33 CASE 4:
2027 * placed first, because it is the normal case
2030 expire_client(unconf);
2031 new = create_client(clname, dname, rqstp, &clverifier);
2035 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
2037 * RFC 3530 14.2.33 CASE 1:
2038 * probable callback update
2041 /* Note this is removing unconfirmed {*x***},
2042 * which is stronger than RFC recommended {vxc**}.
2043 * This has the advantage that there is at most
2044 * one {*x***} in either list at any time.
2046 expire_client(unconf);
2048 new = create_client(clname, dname, rqstp, &clverifier);
2051 copy_clid(new, conf);
2052 } else if (!unconf) {
2054 * RFC 3530 14.2.33 CASE 2:
2055 * probable client reboot; state will be removed if
2058 new = create_client(clname, dname, rqstp, &clverifier);
2064 * RFC 3530 14.2.33 CASE 3:
2065 * probable client reboot; state will be removed if
2068 expire_client(unconf);
2069 new = create_client(clname, dname, rqstp, &clverifier);
2075 * XXX: we should probably set this at creation time, and check
2076 * for consistent minorversion use throughout:
2078 new->cl_minorversion = 0;
2079 gen_callback(new, setclid, rqstp);
2080 add_to_unconfirmed(new, strhashval);
2081 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2082 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2083 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2086 nfs4_unlock_state();
2092 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2093 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2094 * bullets, labeled as CASE1 - CASE4 below.
2097 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2098 struct nfsd4_compound_state *cstate,
2099 struct nfsd4_setclientid_confirm *setclientid_confirm)
2101 struct sockaddr *sa = svc_addr(rqstp);
2102 struct nfs4_client *conf, *unconf;
2103 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2104 clientid_t * clid = &setclientid_confirm->sc_clientid;
2107 if (STALE_CLIENTID(clid))
2108 return nfserr_stale_clientid;
2110 * XXX The Duplicate Request Cache (DRC) has been checked (??)
2111 * We get here on a DRC miss.
2116 conf = find_confirmed_client(clid);
2117 unconf = find_unconfirmed_client(clid);
2119 status = nfserr_clid_inuse;
2120 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2122 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2126 * section 14.2.34 of RFC 3530 has a description of
2127 * SETCLIENTID_CONFIRM request processing consisting
2128 * of 4 bullet points, labeled as CASE1 - CASE4 below.
2130 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2132 * RFC 3530 14.2.34 CASE 1:
2135 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2136 status = nfserr_clid_inuse;
2138 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2139 nfsd4_probe_callback(conf);
2140 expire_client(unconf);
2144 } else if (conf && !unconf) {
2146 * RFC 3530 14.2.34 CASE 2:
2147 * probable retransmitted request; play it safe and
2150 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2151 status = nfserr_clid_inuse;
2154 } else if (!conf && unconf
2155 && same_verf(&unconf->cl_confirm, &confirm)) {
2157 * RFC 3530 14.2.34 CASE 3:
2158 * Normal case; new or rebooted client:
2160 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2161 status = nfserr_clid_inuse;
2164 clientstr_hashval(unconf->cl_recdir);
2165 conf = find_confirmed_client_by_str(unconf->cl_recdir,
2168 nfsd4_remove_clid_dir(conf);
2169 expire_client(conf);
2171 move_to_confirmed(unconf);
2173 nfsd4_probe_callback(conf);
2176 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2177 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2180 * RFC 3530 14.2.34 CASE 4:
2181 * Client probably hasn't noticed that we rebooted yet.
2183 status = nfserr_stale_clientid;
2185 /* check that we have hit one of the cases...*/
2186 status = nfserr_clid_inuse;
2189 nfs4_unlock_state();
2193 /* OPEN Share state helper functions */
2194 static inline struct nfs4_file *
2195 alloc_init_file(struct inode *ino)
2197 struct nfs4_file *fp;
2198 unsigned int hashval = file_hashval(ino);
2200 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2202 atomic_set(&fp->fi_ref, 1);
2203 INIT_LIST_HEAD(&fp->fi_hash);
2204 INIT_LIST_HEAD(&fp->fi_stateids);
2205 INIT_LIST_HEAD(&fp->fi_delegations);
2206 fp->fi_inode = igrab(ino);
2207 fp->fi_had_conflict = false;
2208 fp->fi_lease = NULL;
2209 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2210 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2211 spin_lock(&recall_lock);
2212 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2213 spin_unlock(&recall_lock);
2220 nfsd4_free_slab(struct kmem_cache **slab)
2224 kmem_cache_destroy(*slab);
2229 nfsd4_free_slabs(void)
2231 nfsd4_free_slab(&openowner_slab);
2232 nfsd4_free_slab(&lockowner_slab);
2233 nfsd4_free_slab(&file_slab);
2234 nfsd4_free_slab(&stateid_slab);
2235 nfsd4_free_slab(&deleg_slab);
2239 nfsd4_init_slabs(void)
2241 openowner_slab = kmem_cache_create("nfsd4_openowners",
2242 sizeof(struct nfs4_openowner), 0, 0, NULL);
2243 if (openowner_slab == NULL)
2245 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2246 sizeof(struct nfs4_openowner), 0, 0, NULL);
2247 if (lockowner_slab == NULL)
2249 file_slab = kmem_cache_create("nfsd4_files",
2250 sizeof(struct nfs4_file), 0, 0, NULL);
2251 if (file_slab == NULL)
2253 stateid_slab = kmem_cache_create("nfsd4_stateids",
2254 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2255 if (stateid_slab == NULL)
2257 deleg_slab = kmem_cache_create("nfsd4_delegations",
2258 sizeof(struct nfs4_delegation), 0, 0, NULL);
2259 if (deleg_slab == NULL)
2264 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2268 void nfs4_free_openowner(struct nfs4_openowner *oo)
2270 kfree(oo->oo_owner.so_owner.data);
2271 kmem_cache_free(openowner_slab, oo);
2274 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2276 kfree(lo->lo_owner.so_owner.data);
2277 kmem_cache_free(lockowner_slab, lo);
2280 static void init_nfs4_replay(struct nfs4_replay *rp)
2282 rp->rp_status = nfserr_serverfault;
2284 rp->rp_buf = rp->rp_ibuf;
2287 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2289 struct nfs4_stateowner *sop;
2291 sop = kmem_cache_alloc(slab, GFP_KERNEL);
2295 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2296 if (!sop->so_owner.data) {
2297 kmem_cache_free(slab, sop);
2300 sop->so_owner.len = owner->len;
2302 INIT_LIST_HEAD(&sop->so_stateids);
2303 sop->so_client = clp;
2304 init_nfs4_replay(&sop->so_replay);
2308 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2310 list_add(&oo->oo_owner.so_strhash, &open_ownerstr_hashtbl[strhashval]);
2311 list_add(&oo->oo_perclient, &clp->cl_openowners);
2314 static struct nfs4_openowner *
2315 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2316 struct nfs4_openowner *oo;
2318 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2321 oo->oo_owner.so_is_open_owner = 1;
2322 oo->oo_owner.so_seqid = open->op_seqid;
2323 oo->oo_flags = NFS4_OO_NEW;
2325 oo->oo_last_closed_stid = NULL;
2326 INIT_LIST_HEAD(&oo->oo_close_lru);
2327 hash_openowner(oo, clp, strhashval);
2331 static inline __be32 init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2332 struct nfs4_openowner *oo = open->op_openowner;
2333 struct nfs4_client *clp = oo->oo_owner.so_client;
2336 status = init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2339 INIT_LIST_HEAD(&stp->st_lockowners);
2340 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2341 list_add(&stp->st_perfile, &fp->fi_stateids);
2342 stp->st_stateowner = &oo->oo_owner;
2345 stp->st_access_bmap = 0;
2346 stp->st_deny_bmap = 0;
2347 __set_bit(open->op_share_access, &stp->st_access_bmap);
2348 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2349 stp->st_openstp = NULL;
2354 move_to_close_lru(struct nfs4_openowner *oo)
2356 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2358 list_move_tail(&oo->oo_close_lru, &close_lru);
2359 oo->oo_time = get_seconds();
2363 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2366 return (sop->so_owner.len == owner->len) &&
2367 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2368 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2371 static struct nfs4_openowner *
2372 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2374 struct nfs4_stateowner *so;
2375 struct nfs4_openowner *oo;
2377 list_for_each_entry(so, &open_ownerstr_hashtbl[hashval], so_strhash) {
2378 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2380 renew_client(oo->oo_owner.so_client);
2387 /* search file_hashtbl[] for file */
2388 static struct nfs4_file *
2389 find_file(struct inode *ino)
2391 unsigned int hashval = file_hashval(ino);
2392 struct nfs4_file *fp;
2394 spin_lock(&recall_lock);
2395 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2396 if (fp->fi_inode == ino) {
2398 spin_unlock(&recall_lock);
2402 spin_unlock(&recall_lock);
2407 * Called to check deny when READ with all zero stateid or
2408 * WRITE with all zero or all one stateid
2411 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2413 struct inode *ino = current_fh->fh_dentry->d_inode;
2414 struct nfs4_file *fp;
2415 struct nfs4_ol_stateid *stp;
2418 dprintk("NFSD: nfs4_share_conflict\n");
2420 fp = find_file(ino);
2423 ret = nfserr_locked;
2424 /* Search for conflicting share reservations */
2425 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2426 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2427 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2436 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2438 /* We're assuming the state code never drops its reference
2439 * without first removing the lease. Since we're in this lease
2440 * callback (and since the lease code is serialized by the kernel
2441 * lock) we know the server hasn't removed the lease yet, we know
2442 * it's safe to take a reference: */
2443 atomic_inc(&dp->dl_count);
2445 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2447 /* only place dl_time is set. protected by lock_flocks*/
2448 dp->dl_time = get_seconds();
2450 nfsd4_cb_recall(dp);
2453 /* Called from break_lease() with lock_flocks() held. */
2454 static void nfsd_break_deleg_cb(struct file_lock *fl)
2456 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2457 struct nfs4_delegation *dp;
2460 /* We assume break_lease is only called once per lease: */
2461 BUG_ON(fp->fi_had_conflict);
2463 * We don't want the locks code to timeout the lease for us;
2464 * we'll remove it ourself if a delegation isn't returned
2467 fl->fl_break_time = 0;
2469 spin_lock(&recall_lock);
2470 fp->fi_had_conflict = true;
2471 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2472 nfsd_break_one_deleg(dp);
2473 spin_unlock(&recall_lock);
2477 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2480 return lease_modify(onlist, arg);
2485 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2486 .lm_break = nfsd_break_deleg_cb,
2487 .lm_change = nfsd_change_deleg_cb,
2490 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2492 if (nfsd4_has_session(cstate))
2494 if (seqid == so->so_seqid - 1)
2495 return nfserr_replay_me;
2496 if (seqid == so->so_seqid)
2498 return nfserr_bad_seqid;
2502 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2503 struct nfsd4_open *open)
2505 clientid_t *clientid = &open->op_clientid;
2506 struct nfs4_client *clp = NULL;
2507 unsigned int strhashval;
2508 struct nfs4_openowner *oo = NULL;
2510 if (STALE_CLIENTID(&open->op_clientid))
2511 return nfserr_stale_clientid;
2513 strhashval = open_ownerstr_hashval(clientid->cl_id, &open->op_owner);
2514 oo = find_openstateowner_str(strhashval, open);
2515 open->op_openowner = oo;
2517 clp = find_confirmed_client(clientid);
2519 return nfserr_expired;
2522 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2523 /* Replace unconfirmed owners without checking for replay. */
2524 clp = oo->oo_owner.so_client;
2525 release_openowner(oo);
2526 open->op_openowner = NULL;
2529 return nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2531 oo = alloc_init_open_stateowner(strhashval, clp, open);
2533 return nfserr_jukebox;
2534 open->op_openowner = oo;
2538 static inline __be32
2539 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2541 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2542 return nfserr_openmode;
2547 static int share_access_to_flags(u32 share_access)
2549 share_access &= ~NFS4_SHARE_WANT_MASK;
2551 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2554 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2556 struct nfs4_stid *ret;
2558 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2561 return delegstateid(ret);
2565 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2566 struct nfs4_delegation **dp)
2569 __be32 status = nfserr_bad_stateid;
2571 *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2574 flags = share_access_to_flags(open->op_share_access);
2575 status = nfs4_check_delegmode(*dp, flags);
2579 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2583 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2588 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2590 struct nfs4_ol_stateid *local;
2591 struct nfs4_openowner *oo = open->op_openowner;
2593 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2594 /* ignore lock owners */
2595 if (local->st_stateowner->so_is_open_owner == 0)
2597 /* remember if we have seen this open owner */
2598 if (local->st_stateowner == &oo->oo_owner)
2600 /* check for conflicting share reservations */
2601 if (!test_share(local, open))
2602 return nfserr_share_denied;
2607 static inline struct nfs4_ol_stateid *
2608 nfs4_alloc_stateid(void)
2610 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2613 static inline int nfs4_access_to_access(u32 nfs4_access)
2617 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2618 flags |= NFSD_MAY_READ;
2619 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2620 flags |= NFSD_MAY_WRITE;
2624 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2625 struct svc_fh *cur_fh, struct nfsd4_open *open)
2628 int oflag = nfs4_access_to_omode(open->op_share_access);
2629 int access = nfs4_access_to_access(open->op_share_access);
2631 /* CLAIM_DELEGATE_CUR is used in response to a broken lease;
2632 * allowing it to break the lease and return EAGAIN leaves the
2633 * client unable to make progress in returning the delegation */
2634 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2635 access |= NFSD_MAY_NOT_BREAK_LEASE;
2637 if (!fp->fi_fds[oflag]) {
2638 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2639 &fp->fi_fds[oflag]);
2643 nfs4_file_get_access(fp, oflag);
2649 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_ol_stateid **stpp,
2650 struct nfs4_file *fp, struct svc_fh *cur_fh,
2651 struct nfsd4_open *open)
2653 struct nfs4_ol_stateid *stp;
2656 stp = nfs4_alloc_stateid();
2658 return nfserr_jukebox;
2660 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2662 kmem_cache_free(stateid_slab, stp);
2669 static inline __be32
2670 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2671 struct nfsd4_open *open)
2673 struct iattr iattr = {
2674 .ia_valid = ATTR_SIZE,
2677 if (!open->op_truncate)
2679 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2680 return nfserr_inval;
2681 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2685 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)
2687 u32 op_share_access = open->op_share_access;
2691 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2693 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2697 status = nfsd4_truncate(rqstp, cur_fh, open);
2700 int oflag = nfs4_access_to_omode(op_share_access);
2701 nfs4_file_put_access(fp, oflag);
2705 /* remember the open */
2706 __set_bit(op_share_access, &stp->st_access_bmap);
2707 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2714 nfs4_set_claim_prev(struct nfsd4_open *open)
2716 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2717 open->op_openowner->oo_owner.so_client->cl_firststate = 1;
2720 /* Should we give out recallable state?: */
2721 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2723 if (clp->cl_cb_state == NFSD4_CB_UP)
2726 * In the sessions case, since we don't have to establish a
2727 * separate connection for callbacks, we assume it's OK
2728 * until we hear otherwise:
2730 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2733 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2735 struct file_lock *fl;
2737 fl = locks_alloc_lock();
2740 locks_init_lock(fl);
2741 fl->fl_lmops = &nfsd_lease_mng_ops;
2742 fl->fl_flags = FL_LEASE;
2743 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2744 fl->fl_end = OFFSET_MAX;
2745 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2746 fl->fl_pid = current->tgid;
2750 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2752 struct nfs4_file *fp = dp->dl_file;
2753 struct file_lock *fl;
2756 fl = nfs4_alloc_init_lease(dp, flag);
2759 fl->fl_file = find_readable_file(fp);
2760 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2761 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2763 list_del_init(&dp->dl_perclnt);
2764 locks_free_lock(fl);
2768 fp->fi_deleg_file = fl->fl_file;
2769 get_file(fp->fi_deleg_file);
2770 atomic_set(&fp->fi_delegees, 1);
2771 list_add(&dp->dl_perfile, &fp->fi_delegations);
2775 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2777 struct nfs4_file *fp = dp->dl_file;
2780 return nfs4_setlease(dp, flag);
2781 spin_lock(&recall_lock);
2782 if (fp->fi_had_conflict) {
2783 spin_unlock(&recall_lock);
2786 atomic_inc(&fp->fi_delegees);
2787 list_add(&dp->dl_perfile, &fp->fi_delegations);
2788 spin_unlock(&recall_lock);
2789 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2794 * Attempt to hand out a delegation.
2797 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2799 struct nfs4_delegation *dp;
2800 struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2802 int status, flag = 0;
2804 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2805 flag = NFS4_OPEN_DELEGATE_NONE;
2806 open->op_recall = 0;
2807 switch (open->op_claim_type) {
2808 case NFS4_OPEN_CLAIM_PREVIOUS:
2810 open->op_recall = 1;
2811 flag = open->op_delegate_type;
2812 if (flag == NFS4_OPEN_DELEGATE_NONE)
2815 case NFS4_OPEN_CLAIM_NULL:
2816 /* Let's not give out any delegations till everyone's
2817 * had the chance to reclaim theirs.... */
2818 if (locks_in_grace())
2820 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2822 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2823 flag = NFS4_OPEN_DELEGATE_WRITE;
2825 flag = NFS4_OPEN_DELEGATE_READ;
2831 dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2834 status = nfs4_set_delegation(dp, flag);
2838 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2840 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2841 STATEID_VAL(&dp->dl_stid.sc_stateid));
2843 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2844 && flag == NFS4_OPEN_DELEGATE_NONE
2845 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2846 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2847 open->op_delegate_type = flag;
2850 nfs4_put_delegation(dp);
2852 flag = NFS4_OPEN_DELEGATE_NONE;
2857 * called with nfs4_lock_state() held.
2860 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2862 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2863 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
2864 struct nfs4_file *fp = NULL;
2865 struct inode *ino = current_fh->fh_dentry->d_inode;
2866 struct nfs4_ol_stateid *stp = NULL;
2867 struct nfs4_delegation *dp = NULL;
2871 * Lookup file; if found, lookup stateid and check open request,
2872 * and check for delegations in the process of being recalled.
2873 * If not found, create the nfs4_file struct
2875 fp = find_file(ino);
2877 if ((status = nfs4_check_open(fp, open, &stp)))
2879 status = nfs4_check_deleg(cl, fp, open, &dp);
2883 status = nfserr_bad_stateid;
2884 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2886 status = nfserr_jukebox;
2887 fp = alloc_init_file(ino);
2893 * OPEN the file, or upgrade an existing OPEN.
2894 * If truncate fails, the OPEN fails.
2897 /* Stateid was found, this is an OPEN upgrade */
2898 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2902 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2905 status = init_open_stateid(stp, fp, open);
2907 release_open_stateid(stp);
2910 status = nfsd4_truncate(rqstp, current_fh, open);
2912 release_open_stateid(stp);
2916 update_stateid(&stp->st_stid.sc_stateid);
2917 memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
2919 if (nfsd4_has_session(&resp->cstate))
2920 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2923 * Attempt to hand out a delegation. No error return, because the
2924 * OPEN succeeds even if we fail.
2926 nfs4_open_delegation(current_fh, open, stp);
2930 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2931 STATEID_VAL(&stp->st_stid.sc_stateid));
2935 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2936 nfs4_set_claim_prev(open);
2938 * To finish the open response, we just need to set the rflags.
2940 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2941 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
2942 !nfsd4_has_session(&resp->cstate))
2943 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2948 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
2950 if (open->op_openowner) {
2951 struct nfs4_openowner *oo = open->op_openowner;
2953 if (!list_empty(&oo->oo_owner.so_stateids))
2954 list_del_init(&oo->oo_close_lru);
2955 if (oo->oo_flags & NFS4_OO_NEW) {
2957 release_openowner(oo);
2958 open->op_openowner = NULL;
2960 oo->oo_flags &= ~NFS4_OO_NEW;
2966 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2969 struct nfs4_client *clp;
2973 dprintk("process_renew(%08x/%08x): starting\n",
2974 clid->cl_boot, clid->cl_id);
2975 status = nfserr_stale_clientid;
2976 if (STALE_CLIENTID(clid))
2978 clp = find_confirmed_client(clid);
2979 status = nfserr_expired;
2981 /* We assume the client took too long to RENEW. */
2982 dprintk("nfsd4_renew: clientid not found!\n");
2985 status = nfserr_cb_path_down;
2986 if (!list_empty(&clp->cl_delegations)
2987 && clp->cl_cb_state != NFSD4_CB_UP)
2991 nfs4_unlock_state();
2995 static struct lock_manager nfsd4_manager = {
2999 nfsd4_end_grace(void)
3001 dprintk("NFSD: end of grace period\n");
3002 nfsd4_recdir_purge_old();
3003 locks_end_grace(&nfsd4_manager);
3005 * Now that every NFSv4 client has had the chance to recover and
3006 * to see the (possibly new, possibly shorter) lease time, we
3007 * can safely set the next grace time to the current lease time:
3009 nfsd4_grace = nfsd4_lease;
3013 nfs4_laundromat(void)
3015 struct nfs4_client *clp;
3016 struct nfs4_openowner *oo;
3017 struct nfs4_delegation *dp;
3018 struct list_head *pos, *next, reaplist;
3019 time_t cutoff = get_seconds() - nfsd4_lease;
3020 time_t t, clientid_val = nfsd4_lease;
3021 time_t u, test_val = nfsd4_lease;
3025 dprintk("NFSD: laundromat service - starting\n");
3026 if (locks_in_grace())
3028 INIT_LIST_HEAD(&reaplist);
3029 spin_lock(&client_lock);
3030 list_for_each_safe(pos, next, &client_lru) {
3031 clp = list_entry(pos, struct nfs4_client, cl_lru);
3032 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3033 t = clp->cl_time - cutoff;
3034 if (clientid_val > t)
3038 if (atomic_read(&clp->cl_refcount)) {
3039 dprintk("NFSD: client in use (clientid %08x)\n",
3040 clp->cl_clientid.cl_id);
3043 unhash_client_locked(clp);
3044 list_add(&clp->cl_lru, &reaplist);
3046 spin_unlock(&client_lock);
3047 list_for_each_safe(pos, next, &reaplist) {
3048 clp = list_entry(pos, struct nfs4_client, cl_lru);
3049 dprintk("NFSD: purging unused client (clientid %08x)\n",
3050 clp->cl_clientid.cl_id);
3051 nfsd4_remove_clid_dir(clp);
3054 spin_lock(&recall_lock);
3055 list_for_each_safe(pos, next, &del_recall_lru) {
3056 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3057 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3058 u = dp->dl_time - cutoff;
3063 list_move(&dp->dl_recall_lru, &reaplist);
3065 spin_unlock(&recall_lock);
3066 list_for_each_safe(pos, next, &reaplist) {
3067 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3068 list_del_init(&dp->dl_recall_lru);
3069 unhash_delegation(dp);
3071 test_val = nfsd4_lease;
3072 list_for_each_safe(pos, next, &close_lru) {
3073 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3074 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3075 u = oo->oo_time - cutoff;
3080 release_openowner(oo);
3082 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3083 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3084 nfs4_unlock_state();
3085 return clientid_val;
3088 static struct workqueue_struct *laundry_wq;
3089 static void laundromat_main(struct work_struct *);
3090 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3093 laundromat_main(struct work_struct *not_used)
3097 t = nfs4_laundromat();
3098 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3099 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3102 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3104 if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3105 return nfserr_bad_stateid;
3110 STALE_STATEID(stateid_t *stateid)
3112 if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3114 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3115 STATEID_VAL(stateid));
3120 access_permit_read(unsigned long access_bmap)
3122 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3123 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3124 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3128 access_permit_write(unsigned long access_bmap)
3130 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3131 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3135 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3137 __be32 status = nfserr_openmode;
3139 /* For lock stateid's, we test the parent open, not the lock: */
3140 if (stp->st_openstp)
3141 stp = stp->st_openstp;
3142 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3144 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3151 static inline __be32
3152 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3154 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3156 else if (locks_in_grace()) {
3157 /* Answer in remaining cases depends on existence of
3158 * conflicting state; so we must wait out the grace period. */
3159 return nfserr_grace;
3160 } else if (flags & WR_STATE)
3161 return nfs4_share_conflict(current_fh,
3162 NFS4_SHARE_DENY_WRITE);
3163 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3164 return nfs4_share_conflict(current_fh,
3165 NFS4_SHARE_DENY_READ);
3169 * Allow READ/WRITE during grace period on recovered state only for files
3170 * that are not able to provide mandatory locking.
3173 grace_disallows_io(struct inode *inode)
3175 return locks_in_grace() && mandatory_lock(inode);
3178 /* Returns true iff a is later than b: */
3179 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3181 return (s32)a->si_generation - (s32)b->si_generation > 0;
3184 static int check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3187 * When sessions are used the stateid generation number is ignored
3190 if (has_session && in->si_generation == 0)
3193 if (in->si_generation == ref->si_generation)
3196 /* If the client sends us a stateid from the future, it's buggy: */
3197 if (stateid_generation_after(in, ref))
3198 return nfserr_bad_stateid;
3200 * However, we could see a stateid from the past, even from a
3201 * non-buggy client. For example, if the client sends a lock
3202 * while some IO is outstanding, the lock may bump si_generation
3203 * while the IO is still in flight. The client could avoid that
3204 * situation by waiting for responses on all the IO requests,
3205 * but better performance may result in retrying IO that
3206 * receives an old_stateid error if requests are rarely
3207 * reordered in flight:
3209 return nfserr_old_stateid;
3212 __be32 nfs4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3214 struct nfs4_stid *s;
3215 struct nfs4_ol_stateid *ols;
3218 if (STALE_STATEID(stateid))
3219 return nfserr_stale_stateid;
3221 s = find_stateid(cl, stateid);
3223 return nfserr_stale_stateid;
3224 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3227 if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3229 ols = openlockstateid(s);
3230 if (ols->st_stateowner->so_is_open_owner
3231 && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3232 return nfserr_bad_stateid;
3236 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3238 struct nfs4_client *cl;
3240 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3241 return nfserr_bad_stateid;
3242 if (STALE_STATEID(stateid))
3243 return nfserr_stale_stateid;
3244 cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3246 return nfserr_expired;
3247 *s = find_stateid_by_type(cl, stateid, typemask);
3249 return nfserr_bad_stateid;
3255 * Checks for stateid operations
3258 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3259 stateid_t *stateid, int flags, struct file **filpp)
3261 struct nfs4_stid *s;
3262 struct nfs4_ol_stateid *stp = NULL;
3263 struct nfs4_delegation *dp = NULL;
3264 struct svc_fh *current_fh = &cstate->current_fh;
3265 struct inode *ino = current_fh->fh_dentry->d_inode;
3271 if (grace_disallows_io(ino))
3272 return nfserr_grace;
3274 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3275 return check_special_stateids(current_fh, stateid, flags);
3277 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s);
3280 status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3283 switch (s->sc_type) {
3284 case NFS4_DELEG_STID:
3285 dp = delegstateid(s);
3286 status = nfs4_check_delegmode(dp, flags);
3290 *filpp = dp->dl_file->fi_deleg_file;
3294 case NFS4_OPEN_STID:
3295 case NFS4_LOCK_STID:
3296 stp = openlockstateid(s);
3297 status = nfs4_check_fh(current_fh, stp);
3300 if (stp->st_stateowner->so_is_open_owner
3301 && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3303 status = nfs4_check_openmode(stp, flags);
3307 if (flags & RD_STATE)
3308 *filpp = find_readable_file(stp->st_file);
3310 *filpp = find_writeable_file(stp->st_file);
3314 return nfserr_bad_stateid;
3322 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3324 if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3325 return nfserr_locks_held;
3326 release_lock_stateid(stp);
3331 * Test if the stateid is valid
3334 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3335 struct nfsd4_test_stateid *test_stateid)
3337 /* real work is done during encoding */
3342 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3343 struct nfsd4_free_stateid *free_stateid)
3345 stateid_t *stateid = &free_stateid->fr_stateid;
3346 struct nfs4_stid *s;
3347 struct nfs4_client *cl = cstate->session->se_client;
3348 __be32 ret = nfserr_bad_stateid;
3351 s = find_stateid(cl, stateid);
3354 switch (s->sc_type) {
3355 case NFS4_DELEG_STID:
3356 ret = nfserr_locks_held;
3358 case NFS4_OPEN_STID:
3359 case NFS4_LOCK_STID:
3360 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3363 if (s->sc_type == NFS4_LOCK_STID)
3364 ret = nfsd4_free_lock_stateid(openlockstateid(s));
3366 ret = nfserr_locks_held;
3369 ret = nfserr_bad_stateid;
3372 nfs4_unlock_state();
3379 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3380 RD_STATE : WR_STATE;
3383 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3385 struct svc_fh *current_fh = &cstate->current_fh;
3386 struct nfs4_stateowner *sop = stp->st_stateowner;
3389 status = nfsd4_check_seqid(cstate, sop, seqid);
3392 if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3394 * "Closed" stateid's exist *only* to return
3395 * nfserr_replay_me from the previous step.
3397 return nfserr_bad_stateid;
3398 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3401 return nfs4_check_fh(current_fh, stp);
3405 * Checks for sequence id mutating operations.
3408 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3409 stateid_t *stateid, char typemask,
3410 struct nfs4_ol_stateid **stpp)
3413 struct nfs4_stid *s;
3415 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3416 seqid, STATEID_VAL(stateid));
3419 status = nfsd4_lookup_stateid(stateid, typemask, &s);
3422 *stpp = openlockstateid(s);
3423 cstate->replay_owner = (*stpp)->st_stateowner;
3425 return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3428 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3431 struct nfs4_openowner *oo;
3433 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3434 NFS4_OPEN_STID, stpp);
3437 oo = openowner((*stpp)->st_stateowner);
3438 if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3439 return nfserr_bad_stateid;
3444 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3445 struct nfsd4_open_confirm *oc)
3448 struct nfs4_openowner *oo;
3449 struct nfs4_ol_stateid *stp;
3451 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3452 (int)cstate->current_fh.fh_dentry->d_name.len,
3453 cstate->current_fh.fh_dentry->d_name.name);
3455 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3461 status = nfs4_preprocess_seqid_op(cstate,
3462 oc->oc_seqid, &oc->oc_req_stateid,
3463 NFS4_OPEN_STID, &stp);
3466 oo = openowner(stp->st_stateowner);
3467 status = nfserr_bad_stateid;
3468 if (oo->oo_flags & NFS4_OO_CONFIRMED)
3470 oo->oo_flags |= NFS4_OO_CONFIRMED;
3471 update_stateid(&stp->st_stid.sc_stateid);
3472 memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3473 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3474 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3476 nfsd4_create_clid_dir(oo->oo_owner.so_client);
3479 if (!cstate->replay_owner)
3480 nfs4_unlock_state();
3484 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3486 if (!test_bit(access, &stp->st_access_bmap))
3488 nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3489 __clear_bit(access, &stp->st_access_bmap);
3492 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3494 switch (to_access) {
3495 case NFS4_SHARE_ACCESS_READ:
3496 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3497 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3499 case NFS4_SHARE_ACCESS_WRITE:
3500 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3501 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3503 case NFS4_SHARE_ACCESS_BOTH:
3511 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3514 for (i = 0; i < 4; i++) {
3515 if ((i & deny) != i)
3516 __clear_bit(i, bmap);
3521 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3522 struct nfsd4_compound_state *cstate,
3523 struct nfsd4_open_downgrade *od)
3526 struct nfs4_ol_stateid *stp;
3528 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3529 (int)cstate->current_fh.fh_dentry->d_name.len,
3530 cstate->current_fh.fh_dentry->d_name.name);
3532 /* We don't yet support WANT bits: */
3533 od->od_share_access &= NFS4_SHARE_ACCESS_MASK;
3536 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3537 &od->od_stateid, &stp);
3540 status = nfserr_inval;
3541 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3542 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3543 stp->st_access_bmap, od->od_share_access);
3546 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3547 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3548 stp->st_deny_bmap, od->od_share_deny);
3551 nfs4_stateid_downgrade(stp, od->od_share_access);
3553 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3555 update_stateid(&stp->st_stid.sc_stateid);
3556 memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3559 if (!cstate->replay_owner)
3560 nfs4_unlock_state();
3564 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3566 struct nfs4_openowner *oo;
3567 struct nfs4_ol_stateid *s;
3569 if (!so->so_is_open_owner)
3572 s = oo->oo_last_closed_stid;
3575 if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3576 /* Release the last_closed_stid on the next seqid bump: */
3577 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3580 oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3581 release_last_closed_stateid(oo);
3584 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3586 unhash_open_stateid(s);
3587 s->st_stid.sc_type = NFS4_CLOSED_STID;
3591 * nfs4_unlock_state() called after encode
3594 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3595 struct nfsd4_close *close)
3598 struct nfs4_openowner *oo;
3599 struct nfs4_ol_stateid *stp;
3601 dprintk("NFSD: nfsd4_close on file %.*s\n",
3602 (int)cstate->current_fh.fh_dentry->d_name.len,
3603 cstate->current_fh.fh_dentry->d_name.name);
3606 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3608 NFS4_OPEN_STID|NFS4_CLOSED_STID,
3612 oo = openowner(stp->st_stateowner);
3614 update_stateid(&stp->st_stid.sc_stateid);
3615 memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3617 nfsd4_close_open_stateid(stp);
3618 oo->oo_last_closed_stid = stp;
3620 /* place unused nfs4_stateowners on so_close_lru list to be
3621 * released by the laundromat service after the lease period
3622 * to enable us to handle CLOSE replay
3624 if (list_empty(&oo->oo_owner.so_stateids))
3625 move_to_close_lru(oo);
3627 if (!cstate->replay_owner)
3628 nfs4_unlock_state();
3633 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3634 struct nfsd4_delegreturn *dr)
3636 struct nfs4_delegation *dp;
3637 stateid_t *stateid = &dr->dr_stateid;
3638 struct nfs4_stid *s;
3639 struct inode *inode;
3642 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3644 inode = cstate->current_fh.fh_dentry->d_inode;
3647 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s);
3650 dp = delegstateid(s);
3651 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3655 unhash_delegation(dp);
3657 nfs4_unlock_state();
3664 * Lock owner state (byte-range locks)
3666 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3667 #define LOCK_HASH_BITS 8
3668 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3669 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3672 end_offset(u64 start, u64 len)
3677 return end >= start ? end: NFS4_MAX_UINT64;
3680 /* last octet in a range */
3682 last_byte_offset(u64 start, u64 len)
3688 return end > start ? end - 1: NFS4_MAX_UINT64;
3691 static inline unsigned int
3692 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3693 struct xdr_netobj *ownername)
3695 return (file_hashval(inode) + cl_id
3696 + opaque_hashval(ownername->data, ownername->len))
3700 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3703 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3704 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3705 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3706 * locking, this prevents us from being completely protocol-compliant. The
3707 * real solution to this problem is to start using unsigned file offsets in
3708 * the VFS, but this is a very deep change!
3711 nfs4_transform_lock_offset(struct file_lock *lock)
3713 if (lock->fl_start < 0)
3714 lock->fl_start = OFFSET_MAX;
3715 if (lock->fl_end < 0)
3716 lock->fl_end = OFFSET_MAX;
3719 /* Hack!: For now, we're defining this just so we can use a pointer to it
3720 * as a unique cookie to identify our (NFSv4's) posix locks. */
3721 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3725 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3727 struct nfs4_lockowner *lo;
3729 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3730 lo = (struct nfs4_lockowner *) fl->fl_owner;
3731 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3732 lo->lo_owner.so_owner.len, GFP_KERNEL);
3733 if (!deny->ld_owner.data)
3734 /* We just don't care that much */
3736 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3737 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3740 deny->ld_owner.len = 0;
3741 deny->ld_owner.data = NULL;
3742 deny->ld_clientid.cl_boot = 0;
3743 deny->ld_clientid.cl_id = 0;
3745 deny->ld_start = fl->fl_start;
3746 deny->ld_length = NFS4_MAX_UINT64;
3747 if (fl->fl_end != NFS4_MAX_UINT64)
3748 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3749 deny->ld_type = NFS4_READ_LT;
3750 if (fl->fl_type != F_RDLCK)
3751 deny->ld_type = NFS4_WRITE_LT;
3754 static struct nfs4_lockowner *
3755 find_lockowner_str(struct inode *inode, clientid_t *clid,
3756 struct xdr_netobj *owner)
3758 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3759 struct nfs4_stateowner *op;
3761 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3762 if (same_owner_str(op, owner, clid))
3763 return lockowner(op);
3768 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3770 list_add(&lo->lo_owner.so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3771 list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3775 * Alloc a lock owner structure.
3776 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3779 * strhashval = lock_ownerstr_hashval
3782 static struct nfs4_lockowner *
3783 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3784 struct nfs4_lockowner *lo;
3786 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
3789 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
3790 lo->lo_owner.so_is_open_owner = 0;
3791 /* It is the openowner seqid that will be incremented in encode in the
3792 * case of new lockowners; so increment the lock seqid manually: */
3793 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
3794 hash_lockowner(lo, strhashval, clp, open_stp);
3798 static struct nfs4_ol_stateid *
3799 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
3801 struct nfs4_ol_stateid *stp;
3802 struct nfs4_client *clp = lo->lo_owner.so_client;
3805 stp = nfs4_alloc_stateid();
3808 status = init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
3810 free_generic_stateid(stp);
3813 list_add(&stp->st_perfile, &fp->fi_stateids);
3814 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
3815 stp->st_stateowner = &lo->lo_owner;
3818 stp->st_access_bmap = 0;
3819 stp->st_deny_bmap = open_stp->st_deny_bmap;
3820 stp->st_openstp = open_stp;
3825 check_lock_length(u64 offset, u64 length)
3827 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3828 LOFF_OVERFLOW(offset, length)));
3831 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
3833 struct nfs4_file *fp = lock_stp->st_file;
3834 int oflag = nfs4_access_to_omode(access);
3836 if (test_bit(access, &lock_stp->st_access_bmap))
3838 nfs4_file_get_access(fp, oflag);
3839 __set_bit(access, &lock_stp->st_access_bmap);
3846 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3847 struct nfsd4_lock *lock)
3849 struct nfs4_openowner *open_sop = NULL;
3850 struct nfs4_lockowner *lock_sop = NULL;
3851 struct nfs4_ol_stateid *lock_stp;
3852 struct nfs4_file *fp;
3853 struct file *filp = NULL;
3854 struct file_lock file_lock;
3855 struct file_lock conflock;
3857 unsigned int strhashval;
3861 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3862 (long long) lock->lk_offset,
3863 (long long) lock->lk_length);
3865 if (check_lock_length(lock->lk_offset, lock->lk_length))
3866 return nfserr_inval;
3868 if ((status = fh_verify(rqstp, &cstate->current_fh,
3869 S_IFREG, NFSD_MAY_LOCK))) {
3870 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3876 if (lock->lk_is_new) {
3878 * Client indicates that this is a new lockowner.
3879 * Use open owner and open stateid to create lock owner and
3882 struct nfs4_ol_stateid *open_stp = NULL;
3884 status = nfserr_stale_clientid;
3885 if (!nfsd4_has_session(cstate) &&
3886 STALE_CLIENTID(&lock->lk_new_clientid))
3889 /* validate and update open stateid and open seqid */
3890 status = nfs4_preprocess_confirmed_seqid_op(cstate,
3891 lock->lk_new_open_seqid,
3892 &lock->lk_new_open_stateid,
3896 open_sop = openowner(open_stp->st_stateowner);
3897 status = nfserr_bad_stateid;
3898 if (!nfsd4_has_session(cstate) &&
3899 !same_clid(&open_sop->oo_owner.so_client->cl_clientid,
3900 &lock->v.new.clientid))
3902 /* create lockowner and lock stateid */
3903 fp = open_stp->st_file;
3904 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3905 open_sop->oo_owner.so_client->cl_clientid.cl_id,
3906 &lock->v.new.owner);
3907 /* XXX: Do we need to check for duplicate stateowners on
3908 * the same file, or should they just be allowed (and
3909 * create new stateids)? */
3910 status = nfserr_jukebox;
3911 lock_sop = alloc_init_lock_stateowner(strhashval,
3912 open_sop->oo_owner.so_client, open_stp, lock);
3913 if (lock_sop == NULL)
3915 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3916 if (lock_stp == NULL)
3919 /* lock (lock owner + lock stateid) already exists */
3920 status = nfs4_preprocess_seqid_op(cstate,
3921 lock->lk_old_lock_seqid,
3922 &lock->lk_old_lock_stateid,
3923 NFS4_LOCK_STID, &lock_stp);
3926 lock_sop = lockowner(lock_stp->st_stateowner);
3927 fp = lock_stp->st_file;
3929 /* lock_sop and lock_stp have been created or found */
3931 lkflg = setlkflg(lock->lk_type);
3932 status = nfs4_check_openmode(lock_stp, lkflg);
3936 status = nfserr_grace;
3937 if (locks_in_grace() && !lock->lk_reclaim)
3939 status = nfserr_no_grace;
3940 if (!locks_in_grace() && lock->lk_reclaim)
3943 locks_init_lock(&file_lock);
3944 switch (lock->lk_type) {
3947 filp = find_readable_file(lock_stp->st_file);
3949 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
3950 file_lock.fl_type = F_RDLCK;
3953 case NFS4_WRITEW_LT:
3954 filp = find_writeable_file(lock_stp->st_file);
3956 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
3957 file_lock.fl_type = F_WRLCK;
3960 status = nfserr_inval;
3964 status = nfserr_openmode;
3967 file_lock.fl_owner = (fl_owner_t)lock_sop;
3968 file_lock.fl_pid = current->tgid;
3969 file_lock.fl_file = filp;
3970 file_lock.fl_flags = FL_POSIX;
3971 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3973 file_lock.fl_start = lock->lk_offset;
3974 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3975 nfs4_transform_lock_offset(&file_lock);
3978 * Try to lock the file in the VFS.
3979 * Note: locks.c uses the BKL to protect the inode's lock list.
3982 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
3984 case 0: /* success! */
3985 update_stateid(&lock_stp->st_stid.sc_stateid);
3986 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
3990 case (EAGAIN): /* conflock holds conflicting lock */
3991 status = nfserr_denied;
3992 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3993 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3996 status = nfserr_deadlock;
3999 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4000 status = nfserrno(err);
4004 if (status && lock->lk_is_new && lock_sop)
4005 release_lockowner(lock_sop);
4006 if (!cstate->replay_owner)
4007 nfs4_unlock_state();
4012 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4013 * so we do a temporary open here just to get an open file to pass to
4014 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
4017 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4022 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4025 err = vfs_test_lock(file, lock);
4034 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4035 struct nfsd4_lockt *lockt)
4037 struct inode *inode;
4038 struct file_lock file_lock;
4039 struct nfs4_lockowner *lo;
4043 if (locks_in_grace())
4044 return nfserr_grace;
4046 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4047 return nfserr_inval;
4051 status = nfserr_stale_clientid;
4052 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4055 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4058 inode = cstate->current_fh.fh_dentry->d_inode;
4059 locks_init_lock(&file_lock);
4060 switch (lockt->lt_type) {
4063 file_lock.fl_type = F_RDLCK;
4066 case NFS4_WRITEW_LT:
4067 file_lock.fl_type = F_WRLCK;
4070 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4071 status = nfserr_inval;
4075 lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4077 file_lock.fl_owner = (fl_owner_t)lo;
4078 file_lock.fl_pid = current->tgid;
4079 file_lock.fl_flags = FL_POSIX;
4081 file_lock.fl_start = lockt->lt_offset;
4082 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4084 nfs4_transform_lock_offset(&file_lock);
4087 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4089 status = nfserrno(error);
4092 if (file_lock.fl_type != F_UNLCK) {
4093 status = nfserr_denied;
4094 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4097 nfs4_unlock_state();
4102 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4103 struct nfsd4_locku *locku)
4105 struct nfs4_ol_stateid *stp;
4106 struct file *filp = NULL;
4107 struct file_lock file_lock;
4111 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4112 (long long) locku->lu_offset,
4113 (long long) locku->lu_length);
4115 if (check_lock_length(locku->lu_offset, locku->lu_length))
4116 return nfserr_inval;
4120 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4121 &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4124 filp = find_any_file(stp->st_file);
4126 status = nfserr_lock_range;
4130 locks_init_lock(&file_lock);
4131 file_lock.fl_type = F_UNLCK;
4132 file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4133 file_lock.fl_pid = current->tgid;
4134 file_lock.fl_file = filp;
4135 file_lock.fl_flags = FL_POSIX;
4136 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4137 file_lock.fl_start = locku->lu_offset;
4139 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4140 nfs4_transform_lock_offset(&file_lock);
4143 * Try to unlock the file in the VFS.
4145 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4147 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4151 * OK, unlock succeeded; the only thing left to do is update the stateid.
4153 update_stateid(&stp->st_stid.sc_stateid);
4154 memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4157 if (!cstate->replay_owner)
4158 nfs4_unlock_state();
4162 status = nfserrno(err);
4168 * 1: locks held by lockowner
4169 * 0: no locks held by lockowner
4172 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4174 struct file_lock **flpp;
4175 struct inode *inode = filp->fi_inode;
4179 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4180 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4191 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4192 struct nfsd4_compound_state *cstate,
4193 struct nfsd4_release_lockowner *rlockowner)
4195 clientid_t *clid = &rlockowner->rl_clientid;
4196 struct nfs4_stateowner *sop;
4197 struct nfs4_lockowner *lo;
4198 struct nfs4_ol_stateid *stp;
4199 struct xdr_netobj *owner = &rlockowner->rl_owner;
4200 struct list_head matches;
4204 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4205 clid->cl_boot, clid->cl_id);
4207 /* XXX check for lease expiration */
4209 status = nfserr_stale_clientid;
4210 if (STALE_CLIENTID(clid))
4215 status = nfserr_locks_held;
4216 /* XXX: we're doing a linear search through all the lockowners.
4217 * Yipes! For now we'll just hope clients aren't really using
4218 * release_lockowner much, but eventually we have to fix these
4219 * data structures. */
4220 INIT_LIST_HEAD(&matches);
4221 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4222 list_for_each_entry(sop, &lock_ownerstr_hashtbl[i], so_strhash) {
4223 if (!same_owner_str(sop, owner, clid))
4225 list_for_each_entry(stp, &sop->so_stateids,
4227 lo = lockowner(sop);
4228 if (check_for_locks(stp->st_file, lo))
4230 list_add(&lo->lo_list, &matches);
4234 /* Clients probably won't expect us to return with some (but not all)
4235 * of the lockowner state released; so don't release any until all
4236 * have been checked. */
4238 while (!list_empty(&matches)) {
4239 lo = list_entry(matches.next, struct nfs4_lockowner,
4241 /* unhash_stateowner deletes so_perclient only
4242 * for openowners. */
4243 list_del(&lo->lo_list);
4244 release_lockowner(lo);
4247 nfs4_unlock_state();
4251 static inline struct nfs4_client_reclaim *
4254 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4258 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4260 unsigned int strhashval = clientstr_hashval(name);
4261 struct nfs4_client *clp;
4263 clp = find_confirmed_client_by_str(name, strhashval);
4268 * failure => all reset bets are off, nfserr_no_grace...
4271 nfs4_client_to_reclaim(const char *name)
4273 unsigned int strhashval;
4274 struct nfs4_client_reclaim *crp = NULL;
4276 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4277 crp = alloc_reclaim();
4280 strhashval = clientstr_hashval(name);
4281 INIT_LIST_HEAD(&crp->cr_strhash);
4282 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4283 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4284 reclaim_str_hashtbl_size++;
4289 nfs4_release_reclaim(void)
4291 struct nfs4_client_reclaim *crp = NULL;
4294 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4295 while (!list_empty(&reclaim_str_hashtbl[i])) {
4296 crp = list_entry(reclaim_str_hashtbl[i].next,
4297 struct nfs4_client_reclaim, cr_strhash);
4298 list_del(&crp->cr_strhash);
4300 reclaim_str_hashtbl_size--;
4303 BUG_ON(reclaim_str_hashtbl_size);
4307 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4308 static struct nfs4_client_reclaim *
4309 nfs4_find_reclaim_client(clientid_t *clid)
4311 unsigned int strhashval;
4312 struct nfs4_client *clp;
4313 struct nfs4_client_reclaim *crp = NULL;
4316 /* find clientid in conf_id_hashtbl */
4317 clp = find_confirmed_client(clid);
4321 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4322 clp->cl_name.len, clp->cl_name.data,
4325 /* find clp->cl_name in reclaim_str_hashtbl */
4326 strhashval = clientstr_hashval(clp->cl_recdir);
4327 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4328 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4336 * Called from OPEN. Look for clientid in reclaim list.
4339 nfs4_check_open_reclaim(clientid_t *clid)
4341 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4344 /* initialization to perform at module load time: */
4347 nfs4_state_init(void)
4351 status = nfsd4_init_slabs();
4354 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4355 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4356 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4357 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4358 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4359 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4361 for (i = 0; i < SESSION_HASH_SIZE; i++)
4362 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4363 for (i = 0; i < FILE_HASH_SIZE; i++) {
4364 INIT_LIST_HEAD(&file_hashtbl[i]);
4366 for (i = 0; i < OPEN_OWNER_HASH_SIZE; i++) {
4367 INIT_LIST_HEAD(&open_ownerstr_hashtbl[i]);
4369 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4370 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4372 memset(&onestateid, ~0, sizeof(stateid_t));
4373 INIT_LIST_HEAD(&close_lru);
4374 INIT_LIST_HEAD(&client_lru);
4375 INIT_LIST_HEAD(&del_recall_lru);
4376 reclaim_str_hashtbl_size = 0;
4381 nfsd4_load_reboot_recovery_data(void)
4386 nfsd4_init_recdir();
4387 status = nfsd4_recdir_load();
4388 nfs4_unlock_state();
4390 printk("NFSD: Failure reading reboot recovery data\n");
4394 * Since the lifetime of a delegation isn't limited to that of an open, a
4395 * client may quite reasonably hang on to a delegation as long as it has
4396 * the inode cached. This becomes an obvious problem the first time a
4397 * client's inode cache approaches the size of the server's total memory.
4399 * For now we avoid this problem by imposing a hard limit on the number
4400 * of delegations, which varies according to the server's memory size.
4403 set_max_delegations(void)
4406 * Allow at most 4 delegations per megabyte of RAM. Quick
4407 * estimates suggest that in the worst case (where every delegation
4408 * is for a different inode), a delegation could take about 1.5K,
4409 * giving a worst case usage of about 6% of memory.
4411 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4414 /* initialization to perform when the nfsd service is started: */
4417 __nfs4_state_start(void)
4421 boot_time = get_seconds();
4422 locks_start_grace(&nfsd4_manager);
4423 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4425 ret = set_callback_cred();
4428 laundry_wq = create_singlethread_workqueue("nfsd4");
4429 if (laundry_wq == NULL)
4431 ret = nfsd4_create_callback_queue();
4433 goto out_free_laundry;
4434 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4435 set_max_delegations();
4438 destroy_workqueue(laundry_wq);
4443 nfs4_state_start(void)
4445 nfsd4_load_reboot_recovery_data();
4446 return __nfs4_state_start();
4450 __nfs4_state_shutdown(void)
4453 struct nfs4_client *clp = NULL;
4454 struct nfs4_delegation *dp = NULL;
4455 struct list_head *pos, *next, reaplist;
4457 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4458 while (!list_empty(&conf_id_hashtbl[i])) {
4459 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4462 while (!list_empty(&unconf_str_hashtbl[i])) {
4463 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4467 INIT_LIST_HEAD(&reaplist);
4468 spin_lock(&recall_lock);
4469 list_for_each_safe(pos, next, &del_recall_lru) {
4470 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4471 list_move(&dp->dl_recall_lru, &reaplist);
4473 spin_unlock(&recall_lock);
4474 list_for_each_safe(pos, next, &reaplist) {
4475 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4476 list_del_init(&dp->dl_recall_lru);
4477 unhash_delegation(dp);
4480 nfsd4_shutdown_recdir();
4484 nfs4_state_shutdown(void)
4486 cancel_delayed_work_sync(&laundromat_work);
4487 destroy_workqueue(laundry_wq);
4488 locks_end_grace(&nfsd4_manager);
4490 nfs4_release_reclaim();
4491 __nfs4_state_shutdown();
4492 nfs4_unlock_state();
4493 nfsd4_destroy_callback_queue();