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;
107 static void nfsd4_free_file(struct nfs4_file *f)
109 kmem_cache_free(file_slab, f);
113 put_nfs4_file(struct nfs4_file *fi)
115 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
116 list_del(&fi->fi_hash);
117 spin_unlock(&recall_lock);
124 get_nfs4_file(struct nfs4_file *fi)
126 atomic_inc(&fi->fi_ref);
129 static int num_delegations;
130 unsigned int max_delegations;
133 * Open owner state (share locks)
136 /* hash tables for open owners */
137 #define OPEN_OWNER_HASH_BITS 8
138 #define OPEN_OWNER_HASH_SIZE (1 << OPEN_OWNER_HASH_BITS)
139 #define OPEN_OWNER_HASH_MASK (OPEN_OWNER_HASH_SIZE - 1)
141 static unsigned int open_ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
145 ret = opaque_hashval(ownername->data, ownername->len);
147 return ret & OPEN_OWNER_HASH_MASK;
150 static struct list_head open_ownerstr_hashtbl[OPEN_OWNER_HASH_SIZE];
152 /* hash table for nfs4_file */
153 #define FILE_HASH_BITS 8
154 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
156 static unsigned int file_hashval(struct inode *ino)
158 /* XXX: why are we hashing on inode pointer, anyway? */
159 return hash_ptr(ino, FILE_HASH_BITS);
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
164 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
166 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
167 atomic_inc(&fp->fi_access[oflag]);
170 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
172 if (oflag == O_RDWR) {
173 __nfs4_file_get_access(fp, O_RDONLY);
174 __nfs4_file_get_access(fp, O_WRONLY);
176 __nfs4_file_get_access(fp, oflag);
179 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
181 if (fp->fi_fds[oflag]) {
182 fput(fp->fi_fds[oflag]);
183 fp->fi_fds[oflag] = NULL;
187 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
189 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
190 nfs4_file_put_fd(fp, oflag);
192 * It's also safe to get rid of the RDWR open *if*
193 * we no longer have need of the other kind of access
194 * or if we already have the other kind of open:
196 if (fp->fi_fds[1-oflag]
197 || atomic_read(&fp->fi_access[1 - oflag]) == 0)
198 nfs4_file_put_fd(fp, O_RDWR);
202 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
204 if (oflag == O_RDWR) {
205 __nfs4_file_put_access(fp, O_RDONLY);
206 __nfs4_file_put_access(fp, O_WRONLY);
208 __nfs4_file_put_access(fp, oflag);
211 static inline int get_new_stid(struct nfs4_stid *stid)
213 static int min_stateid = 0;
214 struct idr *stateids = &stid->sc_client->cl_stateids;
218 error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
220 * Note: the necessary preallocation was done in
221 * nfs4_alloc_stateid(). The idr code caps the number of
222 * preallocations that can exist at a time, but the state lock
223 * prevents anyone from using ours before we get here:
227 * It shouldn't be a problem to reuse an opaque stateid value.
228 * I don't think it is for 4.1. But with 4.0 I worry that, for
229 * example, a stray write retransmission could be accepted by
230 * the server when it should have been rejected. Therefore,
231 * adopt a trick from the sctp code to attempt to maximize the
232 * amount of time until an id is reused, by ensuring they always
233 * "increase" (mod INT_MAX):
236 min_stateid = new_stid+1;
237 if (min_stateid == INT_MAX)
242 static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
244 stateid_t *s = &stid->sc_stateid;
247 stid->sc_type = type;
248 stid->sc_client = cl;
249 s->si_opaque.so_clid = cl->cl_clientid;
250 new_id = get_new_stid(stid);
251 s->si_opaque.so_id = (u32)new_id;
252 /* Will be incremented before return to client: */
253 s->si_generation = 0;
256 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
258 struct idr *stateids = &cl->cl_stateids;
260 if (!idr_pre_get(stateids, GFP_KERNEL))
263 * Note: if we fail here (or any time between now and the time
264 * we actually get the new idr), we won't need to undo the idr
265 * preallocation, since the idr code caps the number of
266 * preallocated entries.
268 return kmem_cache_alloc(slab, GFP_KERNEL);
271 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
273 return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
276 static struct nfs4_delegation *
277 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
279 struct nfs4_delegation *dp;
280 struct nfs4_file *fp = stp->st_file;
282 dprintk("NFSD alloc_init_deleg\n");
284 * Major work on the lease subsystem (for example, to support
285 * calbacks on stat) will be required before we can support
286 * write delegations properly.
288 if (type != NFS4_OPEN_DELEGATE_READ)
290 if (fp->fi_had_conflict)
292 if (num_delegations > max_delegations)
294 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
297 init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
299 * delegation seqid's are never incremented. The 4.1 special
300 * meaning of seqid 0 isn't meaningful, really, but let's avoid
301 * 0 anyway just for consistency and use 1:
303 dp->dl_stid.sc_stateid.si_generation = 1;
305 INIT_LIST_HEAD(&dp->dl_perfile);
306 INIT_LIST_HEAD(&dp->dl_perclnt);
307 INIT_LIST_HEAD(&dp->dl_recall_lru);
311 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle);
313 atomic_set(&dp->dl_count, 1);
314 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
319 nfs4_put_delegation(struct nfs4_delegation *dp)
321 if (atomic_dec_and_test(&dp->dl_count)) {
322 dprintk("NFSD: freeing dp %p\n",dp);
323 put_nfs4_file(dp->dl_file);
324 kmem_cache_free(deleg_slab, dp);
329 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
331 if (atomic_dec_and_test(&fp->fi_delegees)) {
332 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
334 fput(fp->fi_deleg_file);
335 fp->fi_deleg_file = NULL;
339 static void unhash_stid(struct nfs4_stid *s)
341 struct idr *stateids = &s->sc_client->cl_stateids;
343 idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
346 /* Called under the state lock. */
348 unhash_delegation(struct nfs4_delegation *dp)
350 unhash_stid(&dp->dl_stid);
351 list_del_init(&dp->dl_perclnt);
352 spin_lock(&recall_lock);
353 list_del_init(&dp->dl_perfile);
354 list_del_init(&dp->dl_recall_lru);
355 spin_unlock(&recall_lock);
356 nfs4_put_deleg_lease(dp->dl_file);
357 nfs4_put_delegation(dp);
364 /* client_lock protects the client lru list and session hash table */
365 static DEFINE_SPINLOCK(client_lock);
367 /* Hash tables for nfs4_clientid state */
368 #define CLIENT_HASH_BITS 4
369 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
370 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
372 static unsigned int clientid_hashval(u32 id)
374 return id & CLIENT_HASH_MASK;
377 static unsigned int clientstr_hashval(const char *name)
379 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
383 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
384 * used in reboot/reset lease grace period processing
386 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
387 * setclientid_confirmed info.
389 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
392 * client_lru holds client queue ordered by nfs4_client.cl_time
395 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
396 * for last close replay.
398 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
399 static int reclaim_str_hashtbl_size = 0;
400 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
401 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
402 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
403 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
404 static struct list_head client_lru;
405 static struct list_head close_lru;
408 * We store the NONE, READ, WRITE, and BOTH bits separately in the
409 * st_{access,deny}_bmap field of the stateid, in order to track not
410 * only what share bits are currently in force, but also what
411 * combinations of share bits previous opens have used. This allows us
412 * to enforce the recommendation of rfc 3530 14.2.19 that the server
413 * return an error if the client attempt to downgrade to a combination
414 * of share bits not explicable by closing some of its previous opens.
416 * XXX: This enforcement is actually incomplete, since we don't keep
417 * track of access/deny bit combinations; so, e.g., we allow:
419 * OPEN allow read, deny write
420 * OPEN allow both, deny none
421 * DOWNGRADE allow read, deny none
423 * which we should reject.
426 set_access(unsigned int *access, unsigned long bmap) {
430 for (i = 1; i < 4; i++) {
431 if (test_bit(i, &bmap))
437 set_deny(unsigned int *deny, unsigned long bmap) {
441 for (i = 0; i < 4; i++) {
442 if (test_bit(i, &bmap))
448 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
449 unsigned int access, deny;
451 set_access(&access, stp->st_access_bmap);
452 set_deny(&deny, stp->st_deny_bmap);
453 if ((access & open->op_share_deny) || (deny & open->op_share_access))
458 static int nfs4_access_to_omode(u32 access)
460 switch (access & NFS4_SHARE_ACCESS_BOTH) {
461 case NFS4_SHARE_ACCESS_READ:
463 case NFS4_SHARE_ACCESS_WRITE:
465 case NFS4_SHARE_ACCESS_BOTH:
471 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
473 list_del(&stp->st_perfile);
474 list_del(&stp->st_perstateowner);
477 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
481 if (stp->st_access_bmap) {
482 for (i = 1; i < 4; i++) {
483 if (test_bit(i, &stp->st_access_bmap))
484 nfs4_file_put_access(stp->st_file,
485 nfs4_access_to_omode(i));
486 __clear_bit(i, &stp->st_access_bmap);
489 put_nfs4_file(stp->st_file);
493 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
495 kmem_cache_free(stateid_slab, stp);
498 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
502 unhash_generic_stateid(stp);
503 unhash_stid(&stp->st_stid);
504 file = find_any_file(stp->st_file);
506 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
507 close_generic_stateid(stp);
508 free_generic_stateid(stp);
511 static void unhash_lockowner(struct nfs4_lockowner *lo)
513 struct nfs4_ol_stateid *stp;
515 list_del(&lo->lo_owner.so_strhash);
516 list_del(&lo->lo_perstateid);
517 while (!list_empty(&lo->lo_owner.so_stateids)) {
518 stp = list_first_entry(&lo->lo_owner.so_stateids,
519 struct nfs4_ol_stateid, st_perstateowner);
520 release_lock_stateid(stp);
524 static void release_lockowner(struct nfs4_lockowner *lo)
526 unhash_lockowner(lo);
527 nfs4_free_lockowner(lo);
531 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
533 struct nfs4_lockowner *lo;
535 while (!list_empty(&open_stp->st_lockowners)) {
536 lo = list_entry(open_stp->st_lockowners.next,
537 struct nfs4_lockowner, lo_perstateid);
538 release_lockowner(lo);
542 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
544 unhash_generic_stateid(stp);
545 release_stateid_lockowners(stp);
546 close_generic_stateid(stp);
549 static void release_open_stateid(struct nfs4_ol_stateid *stp)
551 unhash_open_stateid(stp);
552 unhash_stid(&stp->st_stid);
553 free_generic_stateid(stp);
556 static void unhash_openowner(struct nfs4_openowner *oo)
558 struct nfs4_ol_stateid *stp;
560 list_del(&oo->oo_owner.so_strhash);
561 list_del(&oo->oo_perclient);
562 while (!list_empty(&oo->oo_owner.so_stateids)) {
563 stp = list_first_entry(&oo->oo_owner.so_stateids,
564 struct nfs4_ol_stateid, st_perstateowner);
565 release_open_stateid(stp);
569 static void release_last_closed_stateid(struct nfs4_openowner *oo)
571 struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
574 unhash_stid(&s->st_stid);
575 free_generic_stateid(s);
576 oo->oo_last_closed_stid = NULL;
580 static void release_openowner(struct nfs4_openowner *oo)
582 unhash_openowner(oo);
583 list_del(&oo->oo_close_lru);
584 release_last_closed_stateid(oo);
585 nfs4_free_openowner(oo);
588 #define SESSION_HASH_SIZE 512
589 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
592 hash_sessionid(struct nfs4_sessionid *sessionid)
594 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
596 return sid->sequence % SESSION_HASH_SIZE;
600 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
602 u32 *ptr = (u32 *)(&sessionid->data[0]);
603 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
607 gen_sessionid(struct nfsd4_session *ses)
609 struct nfs4_client *clp = ses->se_client;
610 struct nfsd4_sessionid *sid;
612 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
613 sid->clientid = clp->cl_clientid;
614 sid->sequence = current_sessionid++;
619 * The protocol defines ca_maxresponssize_cached to include the size of
620 * the rpc header, but all we need to cache is the data starting after
621 * the end of the initial SEQUENCE operation--the rest we regenerate
622 * each time. Therefore we can advertise a ca_maxresponssize_cached
623 * value that is the number of bytes in our cache plus a few additional
624 * bytes. In order to stay on the safe side, and not promise more than
625 * we can cache, those additional bytes must be the minimum possible: 24
626 * bytes of rpc header (xid through accept state, with AUTH_NULL
627 * verifier), 12 for the compound header (with zero-length tag), and 44
628 * for the SEQUENCE op response:
630 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
633 free_session_slots(struct nfsd4_session *ses)
637 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
638 kfree(ses->se_slots[i]);
642 * We don't actually need to cache the rpc and session headers, so we
643 * can allocate a little less for each slot:
645 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
647 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
650 static int nfsd4_sanitize_slot_size(u32 size)
652 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
653 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
659 * XXX: If we run out of reserved DRC memory we could (up to a point)
660 * re-negotiate active sessions and reduce their slot usage to make
661 * rooom for new connections. For now we just fail the create session.
663 static int nfsd4_get_drc_mem(int slotsize, u32 num)
667 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
669 spin_lock(&nfsd_drc_lock);
670 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
671 nfsd_drc_max_mem - nfsd_drc_mem_used);
672 num = min_t(int, num, avail / slotsize);
673 nfsd_drc_mem_used += num * slotsize;
674 spin_unlock(&nfsd_drc_lock);
679 static void nfsd4_put_drc_mem(int slotsize, int num)
681 spin_lock(&nfsd_drc_lock);
682 nfsd_drc_mem_used -= slotsize * num;
683 spin_unlock(&nfsd_drc_lock);
686 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
688 struct nfsd4_session *new;
691 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
692 + sizeof(struct nfsd4_session) > PAGE_SIZE);
693 mem = numslots * sizeof(struct nfsd4_slot *);
695 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
698 /* allocate each struct nfsd4_slot and data cache in one piece */
699 for (i = 0; i < numslots; i++) {
700 mem = sizeof(struct nfsd4_slot) + slotsize;
701 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
702 if (!new->se_slots[i])
708 kfree(new->se_slots[i]);
713 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
715 u32 maxrpc = nfsd_serv->sv_max_mesg;
717 new->maxreqs = numslots;
718 new->maxresp_cached = min_t(u32, req->maxresp_cached,
719 slotsize + NFSD_MIN_HDR_SEQ_SZ);
720 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
721 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
722 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
725 static void free_conn(struct nfsd4_conn *c)
727 svc_xprt_put(c->cn_xprt);
731 static void nfsd4_conn_lost(struct svc_xpt_user *u)
733 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
734 struct nfs4_client *clp = c->cn_session->se_client;
736 spin_lock(&clp->cl_lock);
737 if (!list_empty(&c->cn_persession)) {
738 list_del(&c->cn_persession);
741 spin_unlock(&clp->cl_lock);
742 nfsd4_probe_callback(clp);
745 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
747 struct nfsd4_conn *conn;
749 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
752 svc_xprt_get(rqstp->rq_xprt);
753 conn->cn_xprt = rqstp->rq_xprt;
754 conn->cn_flags = flags;
755 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
759 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
761 conn->cn_session = ses;
762 list_add(&conn->cn_persession, &ses->se_conns);
765 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
767 struct nfs4_client *clp = ses->se_client;
769 spin_lock(&clp->cl_lock);
770 __nfsd4_hash_conn(conn, ses);
771 spin_unlock(&clp->cl_lock);
774 static int nfsd4_register_conn(struct nfsd4_conn *conn)
776 conn->cn_xpt_user.callback = nfsd4_conn_lost;
777 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
780 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
782 struct nfsd4_conn *conn;
785 conn = alloc_conn(rqstp, dir);
787 return nfserr_jukebox;
788 nfsd4_hash_conn(conn, ses);
789 ret = nfsd4_register_conn(conn);
791 /* oops; xprt is already down: */
792 nfsd4_conn_lost(&conn->cn_xpt_user);
796 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
798 u32 dir = NFS4_CDFC4_FORE;
800 if (ses->se_flags & SESSION4_BACK_CHAN)
801 dir |= NFS4_CDFC4_BACK;
803 return nfsd4_new_conn(rqstp, ses, dir);
806 /* must be called under client_lock */
807 static void nfsd4_del_conns(struct nfsd4_session *s)
809 struct nfs4_client *clp = s->se_client;
810 struct nfsd4_conn *c;
812 spin_lock(&clp->cl_lock);
813 while (!list_empty(&s->se_conns)) {
814 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
815 list_del_init(&c->cn_persession);
816 spin_unlock(&clp->cl_lock);
818 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
821 spin_lock(&clp->cl_lock);
823 spin_unlock(&clp->cl_lock);
826 void free_session(struct kref *kref)
828 struct nfsd4_session *ses;
831 ses = container_of(kref, struct nfsd4_session, se_ref);
832 nfsd4_del_conns(ses);
833 spin_lock(&nfsd_drc_lock);
834 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
835 nfsd_drc_mem_used -= mem;
836 spin_unlock(&nfsd_drc_lock);
837 free_session_slots(ses);
841 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
843 struct nfsd4_session *new;
844 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
845 int numslots, slotsize;
850 * Note decreasing slot size below client's request may
851 * make it difficult for client to function correctly, whereas
852 * decreasing the number of slots will (just?) affect
853 * performance. When short on memory we therefore prefer to
854 * decrease number of slots instead of their size.
856 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
857 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
861 new = alloc_session(slotsize, numslots);
863 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
866 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
868 new->se_client = clp;
871 INIT_LIST_HEAD(&new->se_conns);
873 new->se_cb_seq_nr = 1;
874 new->se_flags = cses->flags;
875 new->se_cb_prog = cses->callback_prog;
876 kref_init(&new->se_ref);
877 idx = hash_sessionid(&new->se_sessionid);
878 spin_lock(&client_lock);
879 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
880 spin_lock(&clp->cl_lock);
881 list_add(&new->se_perclnt, &clp->cl_sessions);
882 spin_unlock(&clp->cl_lock);
883 spin_unlock(&client_lock);
885 status = nfsd4_new_conn_from_crses(rqstp, new);
886 /* whoops: benny points out, status is ignored! (err, or bogus) */
888 free_session(&new->se_ref);
891 if (cses->flags & SESSION4_BACK_CHAN) {
892 struct sockaddr *sa = svc_addr(rqstp);
894 * This is a little silly; with sessions there's no real
895 * use for the callback address. Use the peer address
896 * as a reasonable default for now, but consider fixing
897 * the rpc client not to require an address in the
900 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
901 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
903 nfsd4_probe_callback(clp);
907 /* caller must hold client_lock */
908 static struct nfsd4_session *
909 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
911 struct nfsd4_session *elem;
914 dump_sessionid(__func__, sessionid);
915 idx = hash_sessionid(sessionid);
916 /* Search in the appropriate list */
917 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
918 if (!memcmp(elem->se_sessionid.data, sessionid->data,
919 NFS4_MAX_SESSIONID_LEN)) {
924 dprintk("%s: session not found\n", __func__);
928 /* caller must hold client_lock */
930 unhash_session(struct nfsd4_session *ses)
932 list_del(&ses->se_hash);
933 spin_lock(&ses->se_client->cl_lock);
934 list_del(&ses->se_perclnt);
935 spin_unlock(&ses->se_client->cl_lock);
938 /* must be called under the client_lock */
940 renew_client_locked(struct nfs4_client *clp)
942 if (is_client_expired(clp)) {
943 dprintk("%s: client (clientid %08x/%08x) already expired\n",
945 clp->cl_clientid.cl_boot,
946 clp->cl_clientid.cl_id);
950 dprintk("renewing client (clientid %08x/%08x)\n",
951 clp->cl_clientid.cl_boot,
952 clp->cl_clientid.cl_id);
953 list_move_tail(&clp->cl_lru, &client_lru);
954 clp->cl_time = get_seconds();
958 renew_client(struct nfs4_client *clp)
960 spin_lock(&client_lock);
961 renew_client_locked(clp);
962 spin_unlock(&client_lock);
965 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
967 STALE_CLIENTID(clientid_t *clid)
969 if (clid->cl_boot == boot_time)
971 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
972 clid->cl_boot, clid->cl_id, boot_time);
977 * XXX Should we use a slab cache ?
978 * This type of memory management is somewhat inefficient, but we use it
979 * anyway since SETCLIENTID is not a common operation.
981 static struct nfs4_client *alloc_client(struct xdr_netobj name)
983 struct nfs4_client *clp;
985 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
988 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
989 if (clp->cl_name.data == NULL) {
993 memcpy(clp->cl_name.data, name.data, name.len);
994 clp->cl_name.len = name.len;
999 free_client(struct nfs4_client *clp)
1001 while (!list_empty(&clp->cl_sessions)) {
1002 struct nfsd4_session *ses;
1003 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1005 list_del(&ses->se_perclnt);
1006 nfsd4_put_session(ses);
1008 if (clp->cl_cred.cr_group_info)
1009 put_group_info(clp->cl_cred.cr_group_info);
1010 kfree(clp->cl_principal);
1011 kfree(clp->cl_name.data);
1016 release_session_client(struct nfsd4_session *session)
1018 struct nfs4_client *clp = session->se_client;
1020 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1022 if (is_client_expired(clp)) {
1024 session->se_client = NULL;
1026 renew_client_locked(clp);
1027 spin_unlock(&client_lock);
1030 /* must be called under the client_lock */
1032 unhash_client_locked(struct nfs4_client *clp)
1034 struct nfsd4_session *ses;
1036 mark_client_expired(clp);
1037 list_del(&clp->cl_lru);
1038 spin_lock(&clp->cl_lock);
1039 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1040 list_del_init(&ses->se_hash);
1041 spin_unlock(&clp->cl_lock);
1045 expire_client(struct nfs4_client *clp)
1047 struct nfs4_openowner *oo;
1048 struct nfs4_delegation *dp;
1049 struct list_head reaplist;
1051 INIT_LIST_HEAD(&reaplist);
1052 spin_lock(&recall_lock);
1053 while (!list_empty(&clp->cl_delegations)) {
1054 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1055 list_del_init(&dp->dl_perclnt);
1056 list_move(&dp->dl_recall_lru, &reaplist);
1058 spin_unlock(&recall_lock);
1059 while (!list_empty(&reaplist)) {
1060 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1061 list_del_init(&dp->dl_recall_lru);
1062 unhash_delegation(dp);
1064 while (!list_empty(&clp->cl_openowners)) {
1065 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1066 release_openowner(oo);
1068 nfsd4_shutdown_callback(clp);
1069 if (clp->cl_cb_conn.cb_xprt)
1070 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1071 list_del(&clp->cl_idhash);
1072 list_del(&clp->cl_strhash);
1073 spin_lock(&client_lock);
1074 unhash_client_locked(clp);
1075 if (atomic_read(&clp->cl_refcount) == 0)
1077 spin_unlock(&client_lock);
1080 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1082 memcpy(target->cl_verifier.data, source->data,
1083 sizeof(target->cl_verifier.data));
1086 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1088 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1089 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1092 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1094 target->cr_uid = source->cr_uid;
1095 target->cr_gid = source->cr_gid;
1096 target->cr_group_info = source->cr_group_info;
1097 get_group_info(target->cr_group_info);
1100 static int same_name(const char *n1, const char *n2)
1102 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1106 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1108 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1112 same_clid(clientid_t *cl1, clientid_t *cl2)
1114 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1117 /* XXX what about NGROUP */
1119 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1121 return cr1->cr_uid == cr2->cr_uid;
1124 static void gen_clid(struct nfs4_client *clp)
1126 static u32 current_clientid = 1;
1128 clp->cl_clientid.cl_boot = boot_time;
1129 clp->cl_clientid.cl_id = current_clientid++;
1132 static void gen_confirm(struct nfs4_client *clp)
1137 p = (u32 *)clp->cl_confirm.data;
1138 *p++ = get_seconds();
1142 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1144 return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1147 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1149 struct nfs4_stid *s;
1151 s = find_stateid(cl, t);
1154 if (typemask & s->sc_type)
1159 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1160 struct svc_rqst *rqstp, nfs4_verifier *verf)
1162 struct nfs4_client *clp;
1163 struct sockaddr *sa = svc_addr(rqstp);
1166 clp = alloc_client(name);
1170 INIT_LIST_HEAD(&clp->cl_sessions);
1172 princ = svc_gss_principal(rqstp);
1174 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1175 if (clp->cl_principal == NULL) {
1181 idr_init(&clp->cl_stateids);
1182 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1183 atomic_set(&clp->cl_refcount, 0);
1184 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1185 INIT_LIST_HEAD(&clp->cl_idhash);
1186 INIT_LIST_HEAD(&clp->cl_strhash);
1187 INIT_LIST_HEAD(&clp->cl_openowners);
1188 INIT_LIST_HEAD(&clp->cl_delegations);
1189 INIT_LIST_HEAD(&clp->cl_lru);
1190 INIT_LIST_HEAD(&clp->cl_callbacks);
1191 spin_lock_init(&clp->cl_lock);
1192 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1193 clp->cl_time = get_seconds();
1194 clear_bit(0, &clp->cl_cb_slot_busy);
1195 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1196 copy_verf(clp, verf);
1197 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1198 clp->cl_flavor = rqstp->rq_flavor;
1199 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1201 clp->cl_cb_session = NULL;
1206 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1208 unsigned int idhashval;
1210 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1211 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1212 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1217 move_to_confirmed(struct nfs4_client *clp)
1219 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1220 unsigned int strhashval;
1222 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1223 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1224 strhashval = clientstr_hashval(clp->cl_recdir);
1225 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1229 static struct nfs4_client *
1230 find_confirmed_client(clientid_t *clid)
1232 struct nfs4_client *clp;
1233 unsigned int idhashval = clientid_hashval(clid->cl_id);
1235 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1236 if (same_clid(&clp->cl_clientid, clid)) {
1244 static struct nfs4_client *
1245 find_unconfirmed_client(clientid_t *clid)
1247 struct nfs4_client *clp;
1248 unsigned int idhashval = clientid_hashval(clid->cl_id);
1250 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1251 if (same_clid(&clp->cl_clientid, clid))
1257 static bool clp_used_exchangeid(struct nfs4_client *clp)
1259 return clp->cl_exchange_flags != 0;
1262 static struct nfs4_client *
1263 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1265 struct nfs4_client *clp;
1267 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1268 if (same_name(clp->cl_recdir, dname))
1274 static struct nfs4_client *
1275 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1277 struct nfs4_client *clp;
1279 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1280 if (same_name(clp->cl_recdir, dname))
1287 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1289 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1290 struct sockaddr *sa = svc_addr(rqstp);
1291 u32 scopeid = rpc_get_scope_id(sa);
1292 unsigned short expected_family;
1294 /* Currently, we only support tcp and tcp6 for the callback channel */
1295 if (se->se_callback_netid_len == 3 &&
1296 !memcmp(se->se_callback_netid_val, "tcp", 3))
1297 expected_family = AF_INET;
1298 else if (se->se_callback_netid_len == 4 &&
1299 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1300 expected_family = AF_INET6;
1304 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1305 se->se_callback_addr_len,
1306 (struct sockaddr *)&conn->cb_addr,
1307 sizeof(conn->cb_addr));
1309 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1312 if (conn->cb_addr.ss_family == AF_INET6)
1313 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1315 conn->cb_prog = se->se_callback_prog;
1316 conn->cb_ident = se->se_callback_ident;
1317 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1320 conn->cb_addr.ss_family = AF_UNSPEC;
1321 conn->cb_addrlen = 0;
1322 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1323 "will not receive delegations\n",
1324 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1330 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1333 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1335 struct nfsd4_slot *slot = resp->cstate.slot;
1338 dprintk("--> %s slot %p\n", __func__, slot);
1340 slot->sl_opcnt = resp->opcnt;
1341 slot->sl_status = resp->cstate.status;
1343 if (nfsd4_not_cached(resp)) {
1344 slot->sl_datalen = 0;
1347 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1348 base = (char *)resp->cstate.datap -
1349 (char *)resp->xbuf->head[0].iov_base;
1350 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1352 WARN("%s: sessions DRC could not cache compound\n", __func__);
1357 * Encode the replay sequence operation from the slot values.
1358 * If cachethis is FALSE encode the uncached rep error on the next
1359 * operation which sets resp->p and increments resp->opcnt for
1360 * nfs4svc_encode_compoundres.
1364 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1365 struct nfsd4_compoundres *resp)
1367 struct nfsd4_op *op;
1368 struct nfsd4_slot *slot = resp->cstate.slot;
1370 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1371 resp->opcnt, resp->cstate.slot->sl_cachethis);
1373 /* Encode the replayed sequence operation */
1374 op = &args->ops[resp->opcnt - 1];
1375 nfsd4_encode_operation(resp, op);
1377 /* Return nfserr_retry_uncached_rep in next operation. */
1378 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1379 op = &args->ops[resp->opcnt++];
1380 op->status = nfserr_retry_uncached_rep;
1381 nfsd4_encode_operation(resp, op);
1387 * The sequence operation is not cached because we can use the slot and
1391 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1392 struct nfsd4_sequence *seq)
1394 struct nfsd4_slot *slot = resp->cstate.slot;
1397 dprintk("--> %s slot %p\n", __func__, slot);
1399 /* Either returns 0 or nfserr_retry_uncached */
1400 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1401 if (status == nfserr_retry_uncached_rep)
1404 /* The sequence operation has been encoded, cstate->datap set. */
1405 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1407 resp->opcnt = slot->sl_opcnt;
1408 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1409 status = slot->sl_status;
1415 * Set the exchange_id flags returned by the server.
1418 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1420 /* pNFS is not supported */
1421 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1423 /* Referrals are supported, Migration is not. */
1424 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1426 /* set the wire flags to return to client. */
1427 clid->flags = new->cl_exchange_flags;
1431 nfsd4_exchange_id(struct svc_rqst *rqstp,
1432 struct nfsd4_compound_state *cstate,
1433 struct nfsd4_exchange_id *exid)
1435 struct nfs4_client *unconf, *conf, *new;
1437 unsigned int strhashval;
1438 char dname[HEXDIR_LEN];
1439 char addr_str[INET6_ADDRSTRLEN];
1440 nfs4_verifier verf = exid->verifier;
1441 struct sockaddr *sa = svc_addr(rqstp);
1443 rpc_ntop(sa, addr_str, sizeof(addr_str));
1444 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1445 "ip_addr=%s flags %x, spa_how %d\n",
1446 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1447 addr_str, exid->flags, exid->spa_how);
1449 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1450 return nfserr_inval;
1452 /* Currently only support SP4_NONE */
1453 switch (exid->spa_how) {
1457 return nfserr_serverfault;
1459 BUG(); /* checked by xdr code */
1461 return nfserr_serverfault; /* no excuse :-/ */
1464 status = nfs4_make_rec_clidname(dname, &exid->clname);
1469 strhashval = clientstr_hashval(dname);
1474 conf = find_confirmed_client_by_str(dname, strhashval);
1476 if (!clp_used_exchangeid(conf)) {
1477 status = nfserr_clid_inuse; /* XXX: ? */
1480 if (!same_verf(&verf, &conf->cl_verifier)) {
1481 /* 18.35.4 case 8 */
1482 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1483 status = nfserr_not_same;
1486 /* Client reboot: destroy old state */
1487 expire_client(conf);
1490 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1491 /* 18.35.4 case 9 */
1492 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1493 status = nfserr_perm;
1496 expire_client(conf);
1500 * Set bit when the owner id and verifier map to an already
1501 * confirmed client id (18.35.3).
1503 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1506 * Falling into 18.35.4 case 2, possible router replay.
1507 * Leave confirmed record intact and return same result.
1509 copy_verf(conf, &verf);
1514 /* 18.35.4 case 7 */
1515 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1516 status = nfserr_noent;
1520 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1523 * Possible retry or client restart. Per 18.35.4 case 4,
1524 * a new unconfirmed record should be generated regardless
1525 * of whether any properties have changed.
1527 expire_client(unconf);
1532 new = create_client(exid->clname, dname, rqstp, &verf);
1534 status = nfserr_jukebox;
1539 add_to_unconfirmed(new, strhashval);
1541 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1542 exid->clientid.cl_id = new->cl_clientid.cl_id;
1545 nfsd4_set_ex_flags(new, exid);
1547 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1548 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1552 nfs4_unlock_state();
1554 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1559 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1561 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1564 /* The slot is in use, and no response has been sent. */
1566 if (seqid == slot_seqid)
1567 return nfserr_jukebox;
1569 return nfserr_seq_misordered;
1572 if (likely(seqid == slot_seqid + 1))
1575 if (seqid == slot_seqid)
1576 return nfserr_replay_cache;
1578 if (seqid == 1 && (slot_seqid + 1) == 0)
1580 /* Misordered replay or misordered new request */
1581 return nfserr_seq_misordered;
1585 * Cache the create session result into the create session single DRC
1586 * slot cache by saving the xdr structure. sl_seqid has been set.
1587 * Do this for solo or embedded create session operations.
1590 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1591 struct nfsd4_clid_slot *slot, int nfserr)
1593 slot->sl_status = nfserr;
1594 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1598 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1599 struct nfsd4_clid_slot *slot)
1601 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1602 return slot->sl_status;
1605 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1606 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1607 1 + /* MIN tag is length with zero, only length */ \
1608 3 + /* version, opcount, opcode */ \
1609 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1610 /* seqid, slotID, slotID, cache */ \
1611 4 ) * sizeof(__be32))
1613 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1614 2 + /* verifier: AUTH_NULL, length 0 */\
1616 1 + /* MIN tag is length with zero, only length */ \
1617 3 + /* opcount, opcode, opstatus*/ \
1618 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1619 /* seqid, slotID, slotID, slotID, status */ \
1620 5 ) * sizeof(__be32))
1622 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1624 return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1625 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1629 nfsd4_create_session(struct svc_rqst *rqstp,
1630 struct nfsd4_compound_state *cstate,
1631 struct nfsd4_create_session *cr_ses)
1633 struct sockaddr *sa = svc_addr(rqstp);
1634 struct nfs4_client *conf, *unconf;
1635 struct nfsd4_session *new;
1636 struct nfsd4_clid_slot *cs_slot = NULL;
1637 bool confirm_me = false;
1640 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1641 return nfserr_inval;
1644 unconf = find_unconfirmed_client(&cr_ses->clientid);
1645 conf = find_confirmed_client(&cr_ses->clientid);
1648 cs_slot = &conf->cl_cs_slot;
1649 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1650 if (status == nfserr_replay_cache) {
1651 dprintk("Got a create_session replay! seqid= %d\n",
1653 /* Return the cached reply status */
1654 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1656 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1657 status = nfserr_seq_misordered;
1658 dprintk("Sequence misordered!\n");
1659 dprintk("Expected seqid= %d but got seqid= %d\n",
1660 cs_slot->sl_seqid, cr_ses->seqid);
1663 } else if (unconf) {
1664 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1665 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1666 status = nfserr_clid_inuse;
1670 cs_slot = &unconf->cl_cs_slot;
1671 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1673 /* an unconfirmed replay returns misordered */
1674 status = nfserr_seq_misordered;
1681 status = nfserr_stale_clientid;
1686 * XXX: we should probably set this at creation time, and check
1687 * for consistent minorversion use throughout:
1689 conf->cl_minorversion = 1;
1691 * We do not support RDMA or persistent sessions
1693 cr_ses->flags &= ~SESSION4_PERSIST;
1694 cr_ses->flags &= ~SESSION4_RDMA;
1696 status = nfserr_toosmall;
1697 if (check_forechannel_attrs(cr_ses->fore_channel))
1700 status = nfserr_jukebox;
1701 new = alloc_init_session(rqstp, conf, cr_ses);
1705 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1706 NFS4_MAX_SESSIONID_LEN);
1707 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1708 sizeof(struct nfsd4_channel_attrs));
1709 cs_slot->sl_seqid++;
1710 cr_ses->seqid = cs_slot->sl_seqid;
1712 /* cache solo and embedded create sessions under the state lock */
1713 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1715 move_to_confirmed(conf);
1717 nfs4_unlock_state();
1718 dprintk("%s returns %d\n", __func__, ntohl(status));
1722 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1724 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1725 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1727 return argp->opcnt == resp->opcnt;
1730 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1733 case NFS4_CDFC4_FORE:
1734 case NFS4_CDFC4_BACK:
1736 case NFS4_CDFC4_FORE_OR_BOTH:
1737 case NFS4_CDFC4_BACK_OR_BOTH:
1738 *dir = NFS4_CDFC4_BOTH;
1741 return nfserr_inval;
1744 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1745 struct nfsd4_compound_state *cstate,
1746 struct nfsd4_bind_conn_to_session *bcts)
1750 if (!nfsd4_last_compound_op(rqstp))
1751 return nfserr_not_only_op;
1752 spin_lock(&client_lock);
1753 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1754 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1755 * client_lock iself: */
1756 if (cstate->session) {
1757 nfsd4_get_session(cstate->session);
1758 atomic_inc(&cstate->session->se_client->cl_refcount);
1760 spin_unlock(&client_lock);
1761 if (!cstate->session)
1762 return nfserr_badsession;
1764 status = nfsd4_map_bcts_dir(&bcts->dir);
1766 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1770 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1774 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1778 nfsd4_destroy_session(struct svc_rqst *r,
1779 struct nfsd4_compound_state *cstate,
1780 struct nfsd4_destroy_session *sessionid)
1782 struct nfsd4_session *ses;
1783 u32 status = nfserr_badsession;
1786 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1787 * - Should we return nfserr_back_chan_busy if waiting for
1788 * callbacks on to-be-destroyed session?
1789 * - Do we need to clear any callback info from previous session?
1792 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1793 if (!nfsd4_last_compound_op(r))
1794 return nfserr_not_only_op;
1796 dump_sessionid(__func__, &sessionid->sessionid);
1797 spin_lock(&client_lock);
1798 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1800 spin_unlock(&client_lock);
1804 unhash_session(ses);
1805 spin_unlock(&client_lock);
1808 nfsd4_probe_callback_sync(ses->se_client);
1809 nfs4_unlock_state();
1811 nfsd4_del_conns(ses);
1813 nfsd4_put_session(ses);
1816 dprintk("%s returns %d\n", __func__, ntohl(status));
1820 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1822 struct nfsd4_conn *c;
1824 list_for_each_entry(c, &s->se_conns, cn_persession) {
1825 if (c->cn_xprt == xpt) {
1832 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1834 struct nfs4_client *clp = ses->se_client;
1835 struct nfsd4_conn *c;
1838 spin_lock(&clp->cl_lock);
1839 c = __nfsd4_find_conn(new->cn_xprt, ses);
1841 spin_unlock(&clp->cl_lock);
1845 __nfsd4_hash_conn(new, ses);
1846 spin_unlock(&clp->cl_lock);
1847 ret = nfsd4_register_conn(new);
1849 /* oops; xprt is already down: */
1850 nfsd4_conn_lost(&new->cn_xpt_user);
1854 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1856 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1858 return args->opcnt > session->se_fchannel.maxops;
1861 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1862 struct nfsd4_session *session)
1864 struct xdr_buf *xb = &rqstp->rq_arg;
1866 return xb->len > session->se_fchannel.maxreq_sz;
1870 nfsd4_sequence(struct svc_rqst *rqstp,
1871 struct nfsd4_compound_state *cstate,
1872 struct nfsd4_sequence *seq)
1874 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1875 struct nfsd4_session *session;
1876 struct nfsd4_slot *slot;
1877 struct nfsd4_conn *conn;
1880 if (resp->opcnt != 1)
1881 return nfserr_sequence_pos;
1884 * Will be either used or freed by nfsd4_sequence_check_conn
1887 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1889 return nfserr_jukebox;
1891 spin_lock(&client_lock);
1892 status = nfserr_badsession;
1893 session = find_in_sessionid_hashtbl(&seq->sessionid);
1897 status = nfserr_too_many_ops;
1898 if (nfsd4_session_too_many_ops(rqstp, session))
1901 status = nfserr_req_too_big;
1902 if (nfsd4_request_too_big(rqstp, session))
1905 status = nfserr_badslot;
1906 if (seq->slotid >= session->se_fchannel.maxreqs)
1909 slot = session->se_slots[seq->slotid];
1910 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1912 /* We do not negotiate the number of slots yet, so set the
1913 * maxslots to the session maxreqs which is used to encode
1914 * sr_highest_slotid and the sr_target_slot id to maxslots */
1915 seq->maxslots = session->se_fchannel.maxreqs;
1917 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1918 if (status == nfserr_replay_cache) {
1919 cstate->slot = slot;
1920 cstate->session = session;
1921 /* Return the cached reply status and set cstate->status
1922 * for nfsd4_proc_compound processing */
1923 status = nfsd4_replay_cache_entry(resp, seq);
1924 cstate->status = nfserr_replay_cache;
1930 nfsd4_sequence_check_conn(conn, session);
1933 /* Success! bump slot seqid */
1934 slot->sl_inuse = true;
1935 slot->sl_seqid = seq->seqid;
1936 slot->sl_cachethis = seq->cachethis;
1938 cstate->slot = slot;
1939 cstate->session = session;
1942 /* Hold a session reference until done processing the compound. */
1943 if (cstate->session) {
1944 struct nfs4_client *clp = session->se_client;
1946 nfsd4_get_session(cstate->session);
1947 atomic_inc(&clp->cl_refcount);
1948 switch (clp->cl_cb_state) {
1950 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
1952 case NFSD4_CB_FAULT:
1953 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
1956 seq->status_flags = 0;
1960 spin_unlock(&client_lock);
1961 dprintk("%s: return %d\n", __func__, ntohl(status));
1966 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1970 if (rc->rca_one_fs) {
1971 if (!cstate->current_fh.fh_dentry)
1972 return nfserr_nofilehandle;
1974 * We don't take advantage of the rca_one_fs case.
1975 * That's OK, it's optional, we can safely ignore it.
1981 status = nfserr_complete_already;
1982 if (cstate->session->se_client->cl_firststate)
1985 status = nfserr_stale_clientid;
1986 if (is_client_expired(cstate->session->se_client))
1988 * The following error isn't really legal.
1989 * But we only get here if the client just explicitly
1990 * destroyed the client. Surely it no longer cares what
1991 * error it gets back on an operation for the dead
1997 nfsd4_create_clid_dir(cstate->session->se_client);
1999 nfs4_unlock_state();
2004 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2005 struct nfsd4_setclientid *setclid)
2007 struct xdr_netobj clname = setclid->se_name;
2008 nfs4_verifier clverifier = setclid->se_verf;
2009 unsigned int strhashval;
2010 struct nfs4_client *conf, *unconf, *new;
2012 char dname[HEXDIR_LEN];
2014 status = nfs4_make_rec_clidname(dname, &clname);
2019 * XXX The Duplicate Request Cache (DRC) has been checked (??)
2020 * We get here on a DRC miss.
2023 strhashval = clientstr_hashval(dname);
2026 conf = find_confirmed_client_by_str(dname, strhashval);
2028 /* RFC 3530 14.2.33 CASE 0: */
2029 status = nfserr_clid_inuse;
2030 if (clp_used_exchangeid(conf))
2032 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2033 char addr_str[INET6_ADDRSTRLEN];
2034 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2036 dprintk("NFSD: setclientid: string in use by client "
2037 "at %s\n", addr_str);
2042 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
2043 * has a description of SETCLIENTID request processing consisting
2044 * of 5 bullet points, labeled as CASE0 - CASE4 below.
2046 unconf = find_unconfirmed_client_by_str(dname, strhashval);
2047 status = nfserr_jukebox;
2050 * RFC 3530 14.2.33 CASE 4:
2051 * placed first, because it is the normal case
2054 expire_client(unconf);
2055 new = create_client(clname, dname, rqstp, &clverifier);
2059 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
2061 * RFC 3530 14.2.33 CASE 1:
2062 * probable callback update
2065 /* Note this is removing unconfirmed {*x***},
2066 * which is stronger than RFC recommended {vxc**}.
2067 * This has the advantage that there is at most
2068 * one {*x***} in either list at any time.
2070 expire_client(unconf);
2072 new = create_client(clname, dname, rqstp, &clverifier);
2075 copy_clid(new, conf);
2076 } else if (!unconf) {
2078 * RFC 3530 14.2.33 CASE 2:
2079 * probable client reboot; state will be removed if
2082 new = create_client(clname, dname, rqstp, &clverifier);
2088 * RFC 3530 14.2.33 CASE 3:
2089 * probable client reboot; state will be removed if
2092 expire_client(unconf);
2093 new = create_client(clname, dname, rqstp, &clverifier);
2099 * XXX: we should probably set this at creation time, and check
2100 * for consistent minorversion use throughout:
2102 new->cl_minorversion = 0;
2103 gen_callback(new, setclid, rqstp);
2104 add_to_unconfirmed(new, strhashval);
2105 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2106 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2107 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2110 nfs4_unlock_state();
2116 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2117 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2118 * bullets, labeled as CASE1 - CASE4 below.
2121 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2122 struct nfsd4_compound_state *cstate,
2123 struct nfsd4_setclientid_confirm *setclientid_confirm)
2125 struct sockaddr *sa = svc_addr(rqstp);
2126 struct nfs4_client *conf, *unconf;
2127 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2128 clientid_t * clid = &setclientid_confirm->sc_clientid;
2131 if (STALE_CLIENTID(clid))
2132 return nfserr_stale_clientid;
2134 * XXX The Duplicate Request Cache (DRC) has been checked (??)
2135 * We get here on a DRC miss.
2140 conf = find_confirmed_client(clid);
2141 unconf = find_unconfirmed_client(clid);
2143 status = nfserr_clid_inuse;
2144 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2146 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2150 * section 14.2.34 of RFC 3530 has a description of
2151 * SETCLIENTID_CONFIRM request processing consisting
2152 * of 4 bullet points, labeled as CASE1 - CASE4 below.
2154 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2156 * RFC 3530 14.2.34 CASE 1:
2159 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2160 status = nfserr_clid_inuse;
2162 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2163 nfsd4_probe_callback(conf);
2164 expire_client(unconf);
2168 } else if (conf && !unconf) {
2170 * RFC 3530 14.2.34 CASE 2:
2171 * probable retransmitted request; play it safe and
2174 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2175 status = nfserr_clid_inuse;
2178 } else if (!conf && unconf
2179 && same_verf(&unconf->cl_confirm, &confirm)) {
2181 * RFC 3530 14.2.34 CASE 3:
2182 * Normal case; new or rebooted client:
2184 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2185 status = nfserr_clid_inuse;
2188 clientstr_hashval(unconf->cl_recdir);
2189 conf = find_confirmed_client_by_str(unconf->cl_recdir,
2192 nfsd4_remove_clid_dir(conf);
2193 expire_client(conf);
2195 move_to_confirmed(unconf);
2197 nfsd4_probe_callback(conf);
2200 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2201 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2204 * RFC 3530 14.2.34 CASE 4:
2205 * Client probably hasn't noticed that we rebooted yet.
2207 status = nfserr_stale_clientid;
2209 /* check that we have hit one of the cases...*/
2210 status = nfserr_clid_inuse;
2213 nfs4_unlock_state();
2217 static struct nfs4_file *nfsd4_alloc_file(void)
2219 return kmem_cache_alloc(file_slab, GFP_KERNEL);
2222 /* OPEN Share state helper functions */
2223 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2225 unsigned int hashval = file_hashval(ino);
2227 atomic_set(&fp->fi_ref, 1);
2228 INIT_LIST_HEAD(&fp->fi_hash);
2229 INIT_LIST_HEAD(&fp->fi_stateids);
2230 INIT_LIST_HEAD(&fp->fi_delegations);
2231 fp->fi_inode = igrab(ino);
2232 fp->fi_had_conflict = false;
2233 fp->fi_lease = NULL;
2234 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2235 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2236 spin_lock(&recall_lock);
2237 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2238 spin_unlock(&recall_lock);
2242 nfsd4_free_slab(struct kmem_cache **slab)
2246 kmem_cache_destroy(*slab);
2251 nfsd4_free_slabs(void)
2253 nfsd4_free_slab(&openowner_slab);
2254 nfsd4_free_slab(&lockowner_slab);
2255 nfsd4_free_slab(&file_slab);
2256 nfsd4_free_slab(&stateid_slab);
2257 nfsd4_free_slab(&deleg_slab);
2261 nfsd4_init_slabs(void)
2263 openowner_slab = kmem_cache_create("nfsd4_openowners",
2264 sizeof(struct nfs4_openowner), 0, 0, NULL);
2265 if (openowner_slab == NULL)
2267 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2268 sizeof(struct nfs4_openowner), 0, 0, NULL);
2269 if (lockowner_slab == NULL)
2271 file_slab = kmem_cache_create("nfsd4_files",
2272 sizeof(struct nfs4_file), 0, 0, NULL);
2273 if (file_slab == NULL)
2275 stateid_slab = kmem_cache_create("nfsd4_stateids",
2276 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2277 if (stateid_slab == NULL)
2279 deleg_slab = kmem_cache_create("nfsd4_delegations",
2280 sizeof(struct nfs4_delegation), 0, 0, NULL);
2281 if (deleg_slab == NULL)
2286 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2290 void nfs4_free_openowner(struct nfs4_openowner *oo)
2292 kfree(oo->oo_owner.so_owner.data);
2293 kmem_cache_free(openowner_slab, oo);
2296 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2298 kfree(lo->lo_owner.so_owner.data);
2299 kmem_cache_free(lockowner_slab, lo);
2302 static void init_nfs4_replay(struct nfs4_replay *rp)
2304 rp->rp_status = nfserr_serverfault;
2306 rp->rp_buf = rp->rp_ibuf;
2309 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2311 struct nfs4_stateowner *sop;
2313 sop = kmem_cache_alloc(slab, GFP_KERNEL);
2317 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2318 if (!sop->so_owner.data) {
2319 kmem_cache_free(slab, sop);
2322 sop->so_owner.len = owner->len;
2324 INIT_LIST_HEAD(&sop->so_stateids);
2325 sop->so_client = clp;
2326 init_nfs4_replay(&sop->so_replay);
2330 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2332 list_add(&oo->oo_owner.so_strhash, &open_ownerstr_hashtbl[strhashval]);
2333 list_add(&oo->oo_perclient, &clp->cl_openowners);
2336 static struct nfs4_openowner *
2337 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2338 struct nfs4_openowner *oo;
2340 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2343 oo->oo_owner.so_is_open_owner = 1;
2344 oo->oo_owner.so_seqid = open->op_seqid;
2345 oo->oo_flags = NFS4_OO_NEW;
2347 oo->oo_last_closed_stid = NULL;
2348 INIT_LIST_HEAD(&oo->oo_close_lru);
2349 hash_openowner(oo, clp, strhashval);
2353 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2354 struct nfs4_openowner *oo = open->op_openowner;
2355 struct nfs4_client *clp = oo->oo_owner.so_client;
2357 init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2358 INIT_LIST_HEAD(&stp->st_lockowners);
2359 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2360 list_add(&stp->st_perfile, &fp->fi_stateids);
2361 stp->st_stateowner = &oo->oo_owner;
2364 stp->st_access_bmap = 0;
2365 stp->st_deny_bmap = 0;
2366 __set_bit(open->op_share_access, &stp->st_access_bmap);
2367 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2368 stp->st_openstp = NULL;
2372 move_to_close_lru(struct nfs4_openowner *oo)
2374 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2376 list_move_tail(&oo->oo_close_lru, &close_lru);
2377 oo->oo_time = get_seconds();
2381 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2384 return (sop->so_owner.len == owner->len) &&
2385 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2386 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2389 static struct nfs4_openowner *
2390 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2392 struct nfs4_stateowner *so;
2393 struct nfs4_openowner *oo;
2395 list_for_each_entry(so, &open_ownerstr_hashtbl[hashval], so_strhash) {
2396 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2398 renew_client(oo->oo_owner.so_client);
2405 /* search file_hashtbl[] for file */
2406 static struct nfs4_file *
2407 find_file(struct inode *ino)
2409 unsigned int hashval = file_hashval(ino);
2410 struct nfs4_file *fp;
2412 spin_lock(&recall_lock);
2413 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2414 if (fp->fi_inode == ino) {
2416 spin_unlock(&recall_lock);
2420 spin_unlock(&recall_lock);
2425 * Called to check deny when READ with all zero stateid or
2426 * WRITE with all zero or all one stateid
2429 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2431 struct inode *ino = current_fh->fh_dentry->d_inode;
2432 struct nfs4_file *fp;
2433 struct nfs4_ol_stateid *stp;
2436 dprintk("NFSD: nfs4_share_conflict\n");
2438 fp = find_file(ino);
2441 ret = nfserr_locked;
2442 /* Search for conflicting share reservations */
2443 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2444 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2445 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2454 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2456 /* We're assuming the state code never drops its reference
2457 * without first removing the lease. Since we're in this lease
2458 * callback (and since the lease code is serialized by the kernel
2459 * lock) we know the server hasn't removed the lease yet, we know
2460 * it's safe to take a reference: */
2461 atomic_inc(&dp->dl_count);
2463 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2465 /* only place dl_time is set. protected by lock_flocks*/
2466 dp->dl_time = get_seconds();
2468 nfsd4_cb_recall(dp);
2471 /* Called from break_lease() with lock_flocks() held. */
2472 static void nfsd_break_deleg_cb(struct file_lock *fl)
2474 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2475 struct nfs4_delegation *dp;
2478 /* We assume break_lease is only called once per lease: */
2479 BUG_ON(fp->fi_had_conflict);
2481 * We don't want the locks code to timeout the lease for us;
2482 * we'll remove it ourself if a delegation isn't returned
2485 fl->fl_break_time = 0;
2487 spin_lock(&recall_lock);
2488 fp->fi_had_conflict = true;
2489 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2490 nfsd_break_one_deleg(dp);
2491 spin_unlock(&recall_lock);
2495 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2498 return lease_modify(onlist, arg);
2503 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2504 .lm_break = nfsd_break_deleg_cb,
2505 .lm_change = nfsd_change_deleg_cb,
2508 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2510 if (nfsd4_has_session(cstate))
2512 if (seqid == so->so_seqid - 1)
2513 return nfserr_replay_me;
2514 if (seqid == so->so_seqid)
2516 return nfserr_bad_seqid;
2520 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2521 struct nfsd4_open *open)
2523 clientid_t *clientid = &open->op_clientid;
2524 struct nfs4_client *clp = NULL;
2525 unsigned int strhashval;
2526 struct nfs4_openowner *oo = NULL;
2529 if (STALE_CLIENTID(&open->op_clientid))
2530 return nfserr_stale_clientid;
2532 * In case we need it later, after we've already created the
2533 * file and don't want to risk a further failure:
2535 open->op_file = nfsd4_alloc_file();
2536 if (open->op_file == NULL)
2537 return nfserr_jukebox;
2539 strhashval = open_ownerstr_hashval(clientid->cl_id, &open->op_owner);
2540 oo = find_openstateowner_str(strhashval, open);
2541 open->op_openowner = oo;
2543 clp = find_confirmed_client(clientid);
2545 return nfserr_expired;
2548 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2549 /* Replace unconfirmed owners without checking for replay. */
2550 clp = oo->oo_owner.so_client;
2551 release_openowner(oo);
2552 open->op_openowner = NULL;
2555 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2558 clp = oo->oo_owner.so_client;
2561 oo = alloc_init_open_stateowner(strhashval, clp, open);
2563 return nfserr_jukebox;
2564 open->op_openowner = oo;
2566 open->op_stp = nfs4_alloc_stateid(clp);
2568 return nfserr_jukebox;
2572 static inline __be32
2573 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2575 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2576 return nfserr_openmode;
2581 static int share_access_to_flags(u32 share_access)
2583 share_access &= ~NFS4_SHARE_WANT_MASK;
2585 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2588 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2590 struct nfs4_stid *ret;
2592 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2595 return delegstateid(ret);
2598 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2600 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2601 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2605 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2606 struct nfs4_delegation **dp)
2609 __be32 status = nfserr_bad_stateid;
2611 *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2614 flags = share_access_to_flags(open->op_share_access);
2615 status = nfs4_check_delegmode(*dp, flags);
2619 if (!nfsd4_is_deleg_cur(open))
2623 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2628 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2630 struct nfs4_ol_stateid *local;
2631 struct nfs4_openowner *oo = open->op_openowner;
2633 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2634 /* ignore lock owners */
2635 if (local->st_stateowner->so_is_open_owner == 0)
2637 /* remember if we have seen this open owner */
2638 if (local->st_stateowner == &oo->oo_owner)
2640 /* check for conflicting share reservations */
2641 if (!test_share(local, open))
2642 return nfserr_share_denied;
2647 static void nfs4_free_stateid(struct nfs4_ol_stateid *s)
2649 kmem_cache_free(stateid_slab, s);
2652 static inline int nfs4_access_to_access(u32 nfs4_access)
2656 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2657 flags |= NFSD_MAY_READ;
2658 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2659 flags |= NFSD_MAY_WRITE;
2663 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2664 struct svc_fh *cur_fh, struct nfsd4_open *open)
2667 int oflag = nfs4_access_to_omode(open->op_share_access);
2668 int access = nfs4_access_to_access(open->op_share_access);
2670 if (!fp->fi_fds[oflag]) {
2671 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2672 &fp->fi_fds[oflag]);
2676 nfs4_file_get_access(fp, oflag);
2681 static inline __be32
2682 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2683 struct nfsd4_open *open)
2685 struct iattr iattr = {
2686 .ia_valid = ATTR_SIZE,
2689 if (!open->op_truncate)
2691 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2692 return nfserr_inval;
2693 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2697 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)
2699 u32 op_share_access = open->op_share_access;
2703 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2705 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2709 status = nfsd4_truncate(rqstp, cur_fh, open);
2712 int oflag = nfs4_access_to_omode(op_share_access);
2713 nfs4_file_put_access(fp, oflag);
2717 /* remember the open */
2718 __set_bit(op_share_access, &stp->st_access_bmap);
2719 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2726 nfs4_set_claim_prev(struct nfsd4_open *open)
2728 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2729 open->op_openowner->oo_owner.so_client->cl_firststate = 1;
2732 /* Should we give out recallable state?: */
2733 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2735 if (clp->cl_cb_state == NFSD4_CB_UP)
2738 * In the sessions case, since we don't have to establish a
2739 * separate connection for callbacks, we assume it's OK
2740 * until we hear otherwise:
2742 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2745 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2747 struct file_lock *fl;
2749 fl = locks_alloc_lock();
2752 locks_init_lock(fl);
2753 fl->fl_lmops = &nfsd_lease_mng_ops;
2754 fl->fl_flags = FL_LEASE;
2755 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2756 fl->fl_end = OFFSET_MAX;
2757 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2758 fl->fl_pid = current->tgid;
2762 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2764 struct nfs4_file *fp = dp->dl_file;
2765 struct file_lock *fl;
2768 fl = nfs4_alloc_init_lease(dp, flag);
2771 fl->fl_file = find_readable_file(fp);
2772 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2773 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2775 list_del_init(&dp->dl_perclnt);
2776 locks_free_lock(fl);
2780 fp->fi_deleg_file = fl->fl_file;
2781 get_file(fp->fi_deleg_file);
2782 atomic_set(&fp->fi_delegees, 1);
2783 list_add(&dp->dl_perfile, &fp->fi_delegations);
2787 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2789 struct nfs4_file *fp = dp->dl_file;
2792 return nfs4_setlease(dp, flag);
2793 spin_lock(&recall_lock);
2794 if (fp->fi_had_conflict) {
2795 spin_unlock(&recall_lock);
2798 atomic_inc(&fp->fi_delegees);
2799 list_add(&dp->dl_perfile, &fp->fi_delegations);
2800 spin_unlock(&recall_lock);
2801 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2806 * Attempt to hand out a delegation.
2809 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2811 struct nfs4_delegation *dp;
2812 struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2814 int status, flag = 0;
2816 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2817 flag = NFS4_OPEN_DELEGATE_NONE;
2818 open->op_recall = 0;
2819 switch (open->op_claim_type) {
2820 case NFS4_OPEN_CLAIM_PREVIOUS:
2822 open->op_recall = 1;
2823 flag = open->op_delegate_type;
2824 if (flag == NFS4_OPEN_DELEGATE_NONE)
2827 case NFS4_OPEN_CLAIM_NULL:
2828 /* Let's not give out any delegations till everyone's
2829 * had the chance to reclaim theirs.... */
2830 if (locks_in_grace())
2832 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2834 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2835 flag = NFS4_OPEN_DELEGATE_WRITE;
2837 flag = NFS4_OPEN_DELEGATE_READ;
2843 dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2846 status = nfs4_set_delegation(dp, flag);
2850 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2852 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2853 STATEID_VAL(&dp->dl_stid.sc_stateid));
2855 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2856 && flag == NFS4_OPEN_DELEGATE_NONE
2857 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2858 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2859 open->op_delegate_type = flag;
2862 nfs4_put_delegation(dp);
2864 flag = NFS4_OPEN_DELEGATE_NONE;
2869 * called with nfs4_lock_state() held.
2872 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2874 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2875 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
2876 struct nfs4_file *fp = NULL;
2877 struct inode *ino = current_fh->fh_dentry->d_inode;
2878 struct nfs4_ol_stateid *stp = NULL;
2879 struct nfs4_delegation *dp = NULL;
2883 * Lookup file; if found, lookup stateid and check open request,
2884 * and check for delegations in the process of being recalled.
2885 * If not found, create the nfs4_file struct
2887 fp = find_file(ino);
2889 if ((status = nfs4_check_open(fp, open, &stp)))
2891 status = nfs4_check_deleg(cl, fp, open, &dp);
2895 status = nfserr_bad_stateid;
2896 if (nfsd4_is_deleg_cur(open))
2898 status = nfserr_jukebox;
2900 open->op_file = NULL;
2901 nfsd4_init_file(fp, ino);
2905 * OPEN the file, or upgrade an existing OPEN.
2906 * If truncate fails, the OPEN fails.
2909 /* Stateid was found, this is an OPEN upgrade */
2910 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2914 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
2918 open->op_stp = NULL;
2919 init_open_stateid(stp, fp, open);
2920 status = nfsd4_truncate(rqstp, current_fh, open);
2922 release_open_stateid(stp);
2926 update_stateid(&stp->st_stid.sc_stateid);
2927 memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
2929 if (nfsd4_has_session(&resp->cstate))
2930 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2933 * Attempt to hand out a delegation. No error return, because the
2934 * OPEN succeeds even if we fail.
2936 nfs4_open_delegation(current_fh, open, stp);
2940 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2941 STATEID_VAL(&stp->st_stid.sc_stateid));
2945 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2946 nfs4_set_claim_prev(open);
2948 * To finish the open response, we just need to set the rflags.
2950 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2951 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
2952 !nfsd4_has_session(&resp->cstate))
2953 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2958 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
2960 if (open->op_openowner) {
2961 struct nfs4_openowner *oo = open->op_openowner;
2963 if (!list_empty(&oo->oo_owner.so_stateids))
2964 list_del_init(&oo->oo_close_lru);
2965 if (oo->oo_flags & NFS4_OO_NEW) {
2967 release_openowner(oo);
2968 open->op_openowner = NULL;
2970 oo->oo_flags &= ~NFS4_OO_NEW;
2974 nfsd4_free_file(open->op_file);
2976 nfs4_free_stateid(open->op_stp);
2980 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2983 struct nfs4_client *clp;
2987 dprintk("process_renew(%08x/%08x): starting\n",
2988 clid->cl_boot, clid->cl_id);
2989 status = nfserr_stale_clientid;
2990 if (STALE_CLIENTID(clid))
2992 clp = find_confirmed_client(clid);
2993 status = nfserr_expired;
2995 /* We assume the client took too long to RENEW. */
2996 dprintk("nfsd4_renew: clientid not found!\n");
2999 status = nfserr_cb_path_down;
3000 if (!list_empty(&clp->cl_delegations)
3001 && clp->cl_cb_state != NFSD4_CB_UP)
3005 nfs4_unlock_state();
3009 static struct lock_manager nfsd4_manager = {
3013 nfsd4_end_grace(void)
3015 dprintk("NFSD: end of grace period\n");
3016 nfsd4_recdir_purge_old();
3017 locks_end_grace(&nfsd4_manager);
3019 * Now that every NFSv4 client has had the chance to recover and
3020 * to see the (possibly new, possibly shorter) lease time, we
3021 * can safely set the next grace time to the current lease time:
3023 nfsd4_grace = nfsd4_lease;
3027 nfs4_laundromat(void)
3029 struct nfs4_client *clp;
3030 struct nfs4_openowner *oo;
3031 struct nfs4_delegation *dp;
3032 struct list_head *pos, *next, reaplist;
3033 time_t cutoff = get_seconds() - nfsd4_lease;
3034 time_t t, clientid_val = nfsd4_lease;
3035 time_t u, test_val = nfsd4_lease;
3039 dprintk("NFSD: laundromat service - starting\n");
3040 if (locks_in_grace())
3042 INIT_LIST_HEAD(&reaplist);
3043 spin_lock(&client_lock);
3044 list_for_each_safe(pos, next, &client_lru) {
3045 clp = list_entry(pos, struct nfs4_client, cl_lru);
3046 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3047 t = clp->cl_time - cutoff;
3048 if (clientid_val > t)
3052 if (atomic_read(&clp->cl_refcount)) {
3053 dprintk("NFSD: client in use (clientid %08x)\n",
3054 clp->cl_clientid.cl_id);
3057 unhash_client_locked(clp);
3058 list_add(&clp->cl_lru, &reaplist);
3060 spin_unlock(&client_lock);
3061 list_for_each_safe(pos, next, &reaplist) {
3062 clp = list_entry(pos, struct nfs4_client, cl_lru);
3063 dprintk("NFSD: purging unused client (clientid %08x)\n",
3064 clp->cl_clientid.cl_id);
3065 nfsd4_remove_clid_dir(clp);
3068 spin_lock(&recall_lock);
3069 list_for_each_safe(pos, next, &del_recall_lru) {
3070 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3071 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3072 u = dp->dl_time - cutoff;
3077 list_move(&dp->dl_recall_lru, &reaplist);
3079 spin_unlock(&recall_lock);
3080 list_for_each_safe(pos, next, &reaplist) {
3081 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3082 list_del_init(&dp->dl_recall_lru);
3083 unhash_delegation(dp);
3085 test_val = nfsd4_lease;
3086 list_for_each_safe(pos, next, &close_lru) {
3087 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3088 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3089 u = oo->oo_time - cutoff;
3094 release_openowner(oo);
3096 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3097 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3098 nfs4_unlock_state();
3099 return clientid_val;
3102 static struct workqueue_struct *laundry_wq;
3103 static void laundromat_main(struct work_struct *);
3104 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3107 laundromat_main(struct work_struct *not_used)
3111 t = nfs4_laundromat();
3112 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3113 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3116 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3118 if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3119 return nfserr_bad_stateid;
3124 STALE_STATEID(stateid_t *stateid)
3126 if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3128 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3129 STATEID_VAL(stateid));
3134 access_permit_read(unsigned long access_bmap)
3136 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3137 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3138 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3142 access_permit_write(unsigned long access_bmap)
3144 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3145 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3149 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3151 __be32 status = nfserr_openmode;
3153 /* For lock stateid's, we test the parent open, not the lock: */
3154 if (stp->st_openstp)
3155 stp = stp->st_openstp;
3156 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3158 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3165 static inline __be32
3166 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3168 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3170 else if (locks_in_grace()) {
3171 /* Answer in remaining cases depends on existence of
3172 * conflicting state; so we must wait out the grace period. */
3173 return nfserr_grace;
3174 } else if (flags & WR_STATE)
3175 return nfs4_share_conflict(current_fh,
3176 NFS4_SHARE_DENY_WRITE);
3177 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3178 return nfs4_share_conflict(current_fh,
3179 NFS4_SHARE_DENY_READ);
3183 * Allow READ/WRITE during grace period on recovered state only for files
3184 * that are not able to provide mandatory locking.
3187 grace_disallows_io(struct inode *inode)
3189 return locks_in_grace() && mandatory_lock(inode);
3192 /* Returns true iff a is later than b: */
3193 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3195 return (s32)a->si_generation - (s32)b->si_generation > 0;
3198 static int check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3201 * When sessions are used the stateid generation number is ignored
3204 if (has_session && in->si_generation == 0)
3207 if (in->si_generation == ref->si_generation)
3210 /* If the client sends us a stateid from the future, it's buggy: */
3211 if (stateid_generation_after(in, ref))
3212 return nfserr_bad_stateid;
3214 * However, we could see a stateid from the past, even from a
3215 * non-buggy client. For example, if the client sends a lock
3216 * while some IO is outstanding, the lock may bump si_generation
3217 * while the IO is still in flight. The client could avoid that
3218 * situation by waiting for responses on all the IO requests,
3219 * but better performance may result in retrying IO that
3220 * receives an old_stateid error if requests are rarely
3221 * reordered in flight:
3223 return nfserr_old_stateid;
3226 __be32 nfs4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3228 struct nfs4_stid *s;
3229 struct nfs4_ol_stateid *ols;
3232 if (STALE_STATEID(stateid))
3233 return nfserr_stale_stateid;
3235 s = find_stateid(cl, stateid);
3237 return nfserr_stale_stateid;
3238 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3241 if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3243 ols = openlockstateid(s);
3244 if (ols->st_stateowner->so_is_open_owner
3245 && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3246 return nfserr_bad_stateid;
3250 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3252 struct nfs4_client *cl;
3254 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3255 return nfserr_bad_stateid;
3256 if (STALE_STATEID(stateid))
3257 return nfserr_stale_stateid;
3258 cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3260 return nfserr_expired;
3261 *s = find_stateid_by_type(cl, stateid, typemask);
3263 return nfserr_bad_stateid;
3269 * Checks for stateid operations
3272 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3273 stateid_t *stateid, int flags, struct file **filpp)
3275 struct nfs4_stid *s;
3276 struct nfs4_ol_stateid *stp = NULL;
3277 struct nfs4_delegation *dp = NULL;
3278 struct svc_fh *current_fh = &cstate->current_fh;
3279 struct inode *ino = current_fh->fh_dentry->d_inode;
3285 if (grace_disallows_io(ino))
3286 return nfserr_grace;
3288 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3289 return check_special_stateids(current_fh, stateid, flags);
3291 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s);
3294 status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3297 switch (s->sc_type) {
3298 case NFS4_DELEG_STID:
3299 dp = delegstateid(s);
3300 status = nfs4_check_delegmode(dp, flags);
3304 *filpp = dp->dl_file->fi_deleg_file;
3308 case NFS4_OPEN_STID:
3309 case NFS4_LOCK_STID:
3310 stp = openlockstateid(s);
3311 status = nfs4_check_fh(current_fh, stp);
3314 if (stp->st_stateowner->so_is_open_owner
3315 && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3317 status = nfs4_check_openmode(stp, flags);
3321 if (flags & RD_STATE)
3322 *filpp = find_readable_file(stp->st_file);
3324 *filpp = find_writeable_file(stp->st_file);
3328 return nfserr_bad_stateid;
3336 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3338 if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3339 return nfserr_locks_held;
3340 release_lock_stateid(stp);
3345 * Test if the stateid is valid
3348 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3349 struct nfsd4_test_stateid *test_stateid)
3351 /* real work is done during encoding */
3356 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3357 struct nfsd4_free_stateid *free_stateid)
3359 stateid_t *stateid = &free_stateid->fr_stateid;
3360 struct nfs4_stid *s;
3361 struct nfs4_client *cl = cstate->session->se_client;
3362 __be32 ret = nfserr_bad_stateid;
3365 s = find_stateid(cl, stateid);
3368 switch (s->sc_type) {
3369 case NFS4_DELEG_STID:
3370 ret = nfserr_locks_held;
3372 case NFS4_OPEN_STID:
3373 case NFS4_LOCK_STID:
3374 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3377 if (s->sc_type == NFS4_LOCK_STID)
3378 ret = nfsd4_free_lock_stateid(openlockstateid(s));
3380 ret = nfserr_locks_held;
3383 ret = nfserr_bad_stateid;
3386 nfs4_unlock_state();
3393 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3394 RD_STATE : WR_STATE;
3397 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3399 struct svc_fh *current_fh = &cstate->current_fh;
3400 struct nfs4_stateowner *sop = stp->st_stateowner;
3403 status = nfsd4_check_seqid(cstate, sop, seqid);
3406 if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3408 * "Closed" stateid's exist *only* to return
3409 * nfserr_replay_me from the previous step.
3411 return nfserr_bad_stateid;
3412 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3415 return nfs4_check_fh(current_fh, stp);
3419 * Checks for sequence id mutating operations.
3422 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3423 stateid_t *stateid, char typemask,
3424 struct nfs4_ol_stateid **stpp)
3427 struct nfs4_stid *s;
3429 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3430 seqid, STATEID_VAL(stateid));
3433 status = nfsd4_lookup_stateid(stateid, typemask, &s);
3436 *stpp = openlockstateid(s);
3437 cstate->replay_owner = (*stpp)->st_stateowner;
3439 return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3442 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3445 struct nfs4_openowner *oo;
3447 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3448 NFS4_OPEN_STID, stpp);
3451 oo = openowner((*stpp)->st_stateowner);
3452 if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3453 return nfserr_bad_stateid;
3458 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3459 struct nfsd4_open_confirm *oc)
3462 struct nfs4_openowner *oo;
3463 struct nfs4_ol_stateid *stp;
3465 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3466 (int)cstate->current_fh.fh_dentry->d_name.len,
3467 cstate->current_fh.fh_dentry->d_name.name);
3469 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3475 status = nfs4_preprocess_seqid_op(cstate,
3476 oc->oc_seqid, &oc->oc_req_stateid,
3477 NFS4_OPEN_STID, &stp);
3480 oo = openowner(stp->st_stateowner);
3481 status = nfserr_bad_stateid;
3482 if (oo->oo_flags & NFS4_OO_CONFIRMED)
3484 oo->oo_flags |= NFS4_OO_CONFIRMED;
3485 update_stateid(&stp->st_stid.sc_stateid);
3486 memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3487 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3488 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3490 nfsd4_create_clid_dir(oo->oo_owner.so_client);
3493 if (!cstate->replay_owner)
3494 nfs4_unlock_state();
3498 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3500 if (!test_bit(access, &stp->st_access_bmap))
3502 nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3503 __clear_bit(access, &stp->st_access_bmap);
3506 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3508 switch (to_access) {
3509 case NFS4_SHARE_ACCESS_READ:
3510 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3511 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3513 case NFS4_SHARE_ACCESS_WRITE:
3514 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3515 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3517 case NFS4_SHARE_ACCESS_BOTH:
3525 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3528 for (i = 0; i < 4; i++) {
3529 if ((i & deny) != i)
3530 __clear_bit(i, bmap);
3535 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3536 struct nfsd4_compound_state *cstate,
3537 struct nfsd4_open_downgrade *od)
3540 struct nfs4_ol_stateid *stp;
3542 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3543 (int)cstate->current_fh.fh_dentry->d_name.len,
3544 cstate->current_fh.fh_dentry->d_name.name);
3546 /* We don't yet support WANT bits: */
3547 od->od_share_access &= NFS4_SHARE_ACCESS_MASK;
3550 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3551 &od->od_stateid, &stp);
3554 status = nfserr_inval;
3555 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3556 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3557 stp->st_access_bmap, od->od_share_access);
3560 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3561 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3562 stp->st_deny_bmap, od->od_share_deny);
3565 nfs4_stateid_downgrade(stp, od->od_share_access);
3567 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3569 update_stateid(&stp->st_stid.sc_stateid);
3570 memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3573 if (!cstate->replay_owner)
3574 nfs4_unlock_state();
3578 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3580 struct nfs4_openowner *oo;
3581 struct nfs4_ol_stateid *s;
3583 if (!so->so_is_open_owner)
3586 s = oo->oo_last_closed_stid;
3589 if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3590 /* Release the last_closed_stid on the next seqid bump: */
3591 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3594 oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3595 release_last_closed_stateid(oo);
3598 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3600 unhash_open_stateid(s);
3601 s->st_stid.sc_type = NFS4_CLOSED_STID;
3605 * nfs4_unlock_state() called after encode
3608 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3609 struct nfsd4_close *close)
3612 struct nfs4_openowner *oo;
3613 struct nfs4_ol_stateid *stp;
3615 dprintk("NFSD: nfsd4_close on file %.*s\n",
3616 (int)cstate->current_fh.fh_dentry->d_name.len,
3617 cstate->current_fh.fh_dentry->d_name.name);
3620 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3622 NFS4_OPEN_STID|NFS4_CLOSED_STID,
3626 oo = openowner(stp->st_stateowner);
3628 update_stateid(&stp->st_stid.sc_stateid);
3629 memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3631 nfsd4_close_open_stateid(stp);
3632 oo->oo_last_closed_stid = stp;
3634 /* place unused nfs4_stateowners on so_close_lru list to be
3635 * released by the laundromat service after the lease period
3636 * to enable us to handle CLOSE replay
3638 if (list_empty(&oo->oo_owner.so_stateids))
3639 move_to_close_lru(oo);
3641 if (!cstate->replay_owner)
3642 nfs4_unlock_state();
3647 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3648 struct nfsd4_delegreturn *dr)
3650 struct nfs4_delegation *dp;
3651 stateid_t *stateid = &dr->dr_stateid;
3652 struct nfs4_stid *s;
3653 struct inode *inode;
3656 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3658 inode = cstate->current_fh.fh_dentry->d_inode;
3661 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s);
3664 dp = delegstateid(s);
3665 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3669 unhash_delegation(dp);
3671 nfs4_unlock_state();
3678 * Lock owner state (byte-range locks)
3680 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3681 #define LOCK_HASH_BITS 8
3682 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3683 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3686 end_offset(u64 start, u64 len)
3691 return end >= start ? end: NFS4_MAX_UINT64;
3694 /* last octet in a range */
3696 last_byte_offset(u64 start, u64 len)
3702 return end > start ? end - 1: NFS4_MAX_UINT64;
3705 static inline unsigned int
3706 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3707 struct xdr_netobj *ownername)
3709 return (file_hashval(inode) + cl_id
3710 + opaque_hashval(ownername->data, ownername->len))
3714 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3717 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3718 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3719 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3720 * locking, this prevents us from being completely protocol-compliant. The
3721 * real solution to this problem is to start using unsigned file offsets in
3722 * the VFS, but this is a very deep change!
3725 nfs4_transform_lock_offset(struct file_lock *lock)
3727 if (lock->fl_start < 0)
3728 lock->fl_start = OFFSET_MAX;
3729 if (lock->fl_end < 0)
3730 lock->fl_end = OFFSET_MAX;
3733 /* Hack!: For now, we're defining this just so we can use a pointer to it
3734 * as a unique cookie to identify our (NFSv4's) posix locks. */
3735 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3739 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3741 struct nfs4_lockowner *lo;
3743 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3744 lo = (struct nfs4_lockowner *) fl->fl_owner;
3745 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3746 lo->lo_owner.so_owner.len, GFP_KERNEL);
3747 if (!deny->ld_owner.data)
3748 /* We just don't care that much */
3750 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3751 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3754 deny->ld_owner.len = 0;
3755 deny->ld_owner.data = NULL;
3756 deny->ld_clientid.cl_boot = 0;
3757 deny->ld_clientid.cl_id = 0;
3759 deny->ld_start = fl->fl_start;
3760 deny->ld_length = NFS4_MAX_UINT64;
3761 if (fl->fl_end != NFS4_MAX_UINT64)
3762 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3763 deny->ld_type = NFS4_READ_LT;
3764 if (fl->fl_type != F_RDLCK)
3765 deny->ld_type = NFS4_WRITE_LT;
3768 static struct nfs4_lockowner *
3769 find_lockowner_str(struct inode *inode, clientid_t *clid,
3770 struct xdr_netobj *owner)
3772 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3773 struct nfs4_stateowner *op;
3775 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3776 if (same_owner_str(op, owner, clid))
3777 return lockowner(op);
3782 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3784 list_add(&lo->lo_owner.so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3785 list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3789 * Alloc a lock owner structure.
3790 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3793 * strhashval = lock_ownerstr_hashval
3796 static struct nfs4_lockowner *
3797 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3798 struct nfs4_lockowner *lo;
3800 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
3803 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
3804 lo->lo_owner.so_is_open_owner = 0;
3805 /* It is the openowner seqid that will be incremented in encode in the
3806 * case of new lockowners; so increment the lock seqid manually: */
3807 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
3808 hash_lockowner(lo, strhashval, clp, open_stp);
3812 static struct nfs4_ol_stateid *
3813 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
3815 struct nfs4_ol_stateid *stp;
3816 struct nfs4_client *clp = lo->lo_owner.so_client;
3818 stp = nfs4_alloc_stateid(clp);
3821 init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
3822 list_add(&stp->st_perfile, &fp->fi_stateids);
3823 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
3824 stp->st_stateowner = &lo->lo_owner;
3827 stp->st_access_bmap = 0;
3828 stp->st_deny_bmap = open_stp->st_deny_bmap;
3829 stp->st_openstp = open_stp;
3834 check_lock_length(u64 offset, u64 length)
3836 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3837 LOFF_OVERFLOW(offset, length)));
3840 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
3842 struct nfs4_file *fp = lock_stp->st_file;
3843 int oflag = nfs4_access_to_omode(access);
3845 if (test_bit(access, &lock_stp->st_access_bmap))
3847 nfs4_file_get_access(fp, oflag);
3848 __set_bit(access, &lock_stp->st_access_bmap);
3855 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3856 struct nfsd4_lock *lock)
3858 struct nfs4_openowner *open_sop = NULL;
3859 struct nfs4_lockowner *lock_sop = NULL;
3860 struct nfs4_ol_stateid *lock_stp;
3861 struct nfs4_file *fp;
3862 struct file *filp = NULL;
3863 struct file_lock file_lock;
3864 struct file_lock conflock;
3866 unsigned int strhashval;
3870 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3871 (long long) lock->lk_offset,
3872 (long long) lock->lk_length);
3874 if (check_lock_length(lock->lk_offset, lock->lk_length))
3875 return nfserr_inval;
3877 if ((status = fh_verify(rqstp, &cstate->current_fh,
3878 S_IFREG, NFSD_MAY_LOCK))) {
3879 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3885 if (lock->lk_is_new) {
3887 * Client indicates that this is a new lockowner.
3888 * Use open owner and open stateid to create lock owner and
3891 struct nfs4_ol_stateid *open_stp = NULL;
3893 status = nfserr_stale_clientid;
3894 if (!nfsd4_has_session(cstate) &&
3895 STALE_CLIENTID(&lock->lk_new_clientid))
3898 /* validate and update open stateid and open seqid */
3899 status = nfs4_preprocess_confirmed_seqid_op(cstate,
3900 lock->lk_new_open_seqid,
3901 &lock->lk_new_open_stateid,
3905 open_sop = openowner(open_stp->st_stateowner);
3906 status = nfserr_bad_stateid;
3907 if (!nfsd4_has_session(cstate) &&
3908 !same_clid(&open_sop->oo_owner.so_client->cl_clientid,
3909 &lock->v.new.clientid))
3911 /* create lockowner and lock stateid */
3912 fp = open_stp->st_file;
3913 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3914 open_sop->oo_owner.so_client->cl_clientid.cl_id,
3915 &lock->v.new.owner);
3916 /* XXX: Do we need to check for duplicate stateowners on
3917 * the same file, or should they just be allowed (and
3918 * create new stateids)? */
3919 status = nfserr_jukebox;
3920 lock_sop = alloc_init_lock_stateowner(strhashval,
3921 open_sop->oo_owner.so_client, open_stp, lock);
3922 if (lock_sop == NULL)
3924 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3925 if (lock_stp == NULL)
3928 /* lock (lock owner + lock stateid) already exists */
3929 status = nfs4_preprocess_seqid_op(cstate,
3930 lock->lk_old_lock_seqid,
3931 &lock->lk_old_lock_stateid,
3932 NFS4_LOCK_STID, &lock_stp);
3935 lock_sop = lockowner(lock_stp->st_stateowner);
3936 fp = lock_stp->st_file;
3938 /* lock_sop and lock_stp have been created or found */
3940 lkflg = setlkflg(lock->lk_type);
3941 status = nfs4_check_openmode(lock_stp, lkflg);
3945 status = nfserr_grace;
3946 if (locks_in_grace() && !lock->lk_reclaim)
3948 status = nfserr_no_grace;
3949 if (!locks_in_grace() && lock->lk_reclaim)
3952 locks_init_lock(&file_lock);
3953 switch (lock->lk_type) {
3956 filp = find_readable_file(lock_stp->st_file);
3958 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
3959 file_lock.fl_type = F_RDLCK;
3962 case NFS4_WRITEW_LT:
3963 filp = find_writeable_file(lock_stp->st_file);
3965 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
3966 file_lock.fl_type = F_WRLCK;
3969 status = nfserr_inval;
3973 status = nfserr_openmode;
3976 file_lock.fl_owner = (fl_owner_t)lock_sop;
3977 file_lock.fl_pid = current->tgid;
3978 file_lock.fl_file = filp;
3979 file_lock.fl_flags = FL_POSIX;
3980 file_lock.fl_lmops = &nfsd_posix_mng_ops;
3982 file_lock.fl_start = lock->lk_offset;
3983 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3984 nfs4_transform_lock_offset(&file_lock);
3987 * Try to lock the file in the VFS.
3988 * Note: locks.c uses the BKL to protect the inode's lock list.
3991 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
3993 case 0: /* success! */
3994 update_stateid(&lock_stp->st_stid.sc_stateid);
3995 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
3999 case (EAGAIN): /* conflock holds conflicting lock */
4000 status = nfserr_denied;
4001 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4002 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4005 status = nfserr_deadlock;
4008 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4009 status = nfserrno(err);
4013 if (status && lock->lk_is_new && lock_sop)
4014 release_lockowner(lock_sop);
4015 if (!cstate->replay_owner)
4016 nfs4_unlock_state();
4021 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4022 * so we do a temporary open here just to get an open file to pass to
4023 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
4026 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4031 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4034 err = vfs_test_lock(file, lock);
4043 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4044 struct nfsd4_lockt *lockt)
4046 struct inode *inode;
4047 struct file_lock file_lock;
4048 struct nfs4_lockowner *lo;
4052 if (locks_in_grace())
4053 return nfserr_grace;
4055 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4056 return nfserr_inval;
4060 status = nfserr_stale_clientid;
4061 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4064 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4067 inode = cstate->current_fh.fh_dentry->d_inode;
4068 locks_init_lock(&file_lock);
4069 switch (lockt->lt_type) {
4072 file_lock.fl_type = F_RDLCK;
4075 case NFS4_WRITEW_LT:
4076 file_lock.fl_type = F_WRLCK;
4079 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4080 status = nfserr_inval;
4084 lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4086 file_lock.fl_owner = (fl_owner_t)lo;
4087 file_lock.fl_pid = current->tgid;
4088 file_lock.fl_flags = FL_POSIX;
4090 file_lock.fl_start = lockt->lt_offset;
4091 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4093 nfs4_transform_lock_offset(&file_lock);
4096 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4098 status = nfserrno(error);
4101 if (file_lock.fl_type != F_UNLCK) {
4102 status = nfserr_denied;
4103 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4106 nfs4_unlock_state();
4111 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4112 struct nfsd4_locku *locku)
4114 struct nfs4_ol_stateid *stp;
4115 struct file *filp = NULL;
4116 struct file_lock file_lock;
4120 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4121 (long long) locku->lu_offset,
4122 (long long) locku->lu_length);
4124 if (check_lock_length(locku->lu_offset, locku->lu_length))
4125 return nfserr_inval;
4129 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4130 &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4133 filp = find_any_file(stp->st_file);
4135 status = nfserr_lock_range;
4139 locks_init_lock(&file_lock);
4140 file_lock.fl_type = F_UNLCK;
4141 file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4142 file_lock.fl_pid = current->tgid;
4143 file_lock.fl_file = filp;
4144 file_lock.fl_flags = FL_POSIX;
4145 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4146 file_lock.fl_start = locku->lu_offset;
4148 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4149 nfs4_transform_lock_offset(&file_lock);
4152 * Try to unlock the file in the VFS.
4154 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4156 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4160 * OK, unlock succeeded; the only thing left to do is update the stateid.
4162 update_stateid(&stp->st_stid.sc_stateid);
4163 memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4166 if (!cstate->replay_owner)
4167 nfs4_unlock_state();
4171 status = nfserrno(err);
4177 * 1: locks held by lockowner
4178 * 0: no locks held by lockowner
4181 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4183 struct file_lock **flpp;
4184 struct inode *inode = filp->fi_inode;
4188 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4189 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4200 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4201 struct nfsd4_compound_state *cstate,
4202 struct nfsd4_release_lockowner *rlockowner)
4204 clientid_t *clid = &rlockowner->rl_clientid;
4205 struct nfs4_stateowner *sop;
4206 struct nfs4_lockowner *lo;
4207 struct nfs4_ol_stateid *stp;
4208 struct xdr_netobj *owner = &rlockowner->rl_owner;
4209 struct list_head matches;
4213 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4214 clid->cl_boot, clid->cl_id);
4216 /* XXX check for lease expiration */
4218 status = nfserr_stale_clientid;
4219 if (STALE_CLIENTID(clid))
4224 status = nfserr_locks_held;
4225 /* XXX: we're doing a linear search through all the lockowners.
4226 * Yipes! For now we'll just hope clients aren't really using
4227 * release_lockowner much, but eventually we have to fix these
4228 * data structures. */
4229 INIT_LIST_HEAD(&matches);
4230 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4231 list_for_each_entry(sop, &lock_ownerstr_hashtbl[i], so_strhash) {
4232 if (!same_owner_str(sop, owner, clid))
4234 list_for_each_entry(stp, &sop->so_stateids,
4236 lo = lockowner(sop);
4237 if (check_for_locks(stp->st_file, lo))
4239 list_add(&lo->lo_list, &matches);
4243 /* Clients probably won't expect us to return with some (but not all)
4244 * of the lockowner state released; so don't release any until all
4245 * have been checked. */
4247 while (!list_empty(&matches)) {
4248 lo = list_entry(matches.next, struct nfs4_lockowner,
4250 /* unhash_stateowner deletes so_perclient only
4251 * for openowners. */
4252 list_del(&lo->lo_list);
4253 release_lockowner(lo);
4256 nfs4_unlock_state();
4260 static inline struct nfs4_client_reclaim *
4263 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4267 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4269 unsigned int strhashval = clientstr_hashval(name);
4270 struct nfs4_client *clp;
4272 clp = find_confirmed_client_by_str(name, strhashval);
4277 * failure => all reset bets are off, nfserr_no_grace...
4280 nfs4_client_to_reclaim(const char *name)
4282 unsigned int strhashval;
4283 struct nfs4_client_reclaim *crp = NULL;
4285 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4286 crp = alloc_reclaim();
4289 strhashval = clientstr_hashval(name);
4290 INIT_LIST_HEAD(&crp->cr_strhash);
4291 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4292 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4293 reclaim_str_hashtbl_size++;
4298 nfs4_release_reclaim(void)
4300 struct nfs4_client_reclaim *crp = NULL;
4303 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4304 while (!list_empty(&reclaim_str_hashtbl[i])) {
4305 crp = list_entry(reclaim_str_hashtbl[i].next,
4306 struct nfs4_client_reclaim, cr_strhash);
4307 list_del(&crp->cr_strhash);
4309 reclaim_str_hashtbl_size--;
4312 BUG_ON(reclaim_str_hashtbl_size);
4316 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4317 static struct nfs4_client_reclaim *
4318 nfs4_find_reclaim_client(clientid_t *clid)
4320 unsigned int strhashval;
4321 struct nfs4_client *clp;
4322 struct nfs4_client_reclaim *crp = NULL;
4325 /* find clientid in conf_id_hashtbl */
4326 clp = find_confirmed_client(clid);
4330 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4331 clp->cl_name.len, clp->cl_name.data,
4334 /* find clp->cl_name in reclaim_str_hashtbl */
4335 strhashval = clientstr_hashval(clp->cl_recdir);
4336 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4337 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4345 * Called from OPEN. Look for clientid in reclaim list.
4348 nfs4_check_open_reclaim(clientid_t *clid)
4350 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4353 /* initialization to perform at module load time: */
4356 nfs4_state_init(void)
4360 status = nfsd4_init_slabs();
4363 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4364 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4365 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4366 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4367 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4368 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4370 for (i = 0; i < SESSION_HASH_SIZE; i++)
4371 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4372 for (i = 0; i < FILE_HASH_SIZE; i++) {
4373 INIT_LIST_HEAD(&file_hashtbl[i]);
4375 for (i = 0; i < OPEN_OWNER_HASH_SIZE; i++) {
4376 INIT_LIST_HEAD(&open_ownerstr_hashtbl[i]);
4378 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4379 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4381 memset(&onestateid, ~0, sizeof(stateid_t));
4382 INIT_LIST_HEAD(&close_lru);
4383 INIT_LIST_HEAD(&client_lru);
4384 INIT_LIST_HEAD(&del_recall_lru);
4385 reclaim_str_hashtbl_size = 0;
4390 nfsd4_load_reboot_recovery_data(void)
4395 nfsd4_init_recdir();
4396 status = nfsd4_recdir_load();
4397 nfs4_unlock_state();
4399 printk("NFSD: Failure reading reboot recovery data\n");
4403 * Since the lifetime of a delegation isn't limited to that of an open, a
4404 * client may quite reasonably hang on to a delegation as long as it has
4405 * the inode cached. This becomes an obvious problem the first time a
4406 * client's inode cache approaches the size of the server's total memory.
4408 * For now we avoid this problem by imposing a hard limit on the number
4409 * of delegations, which varies according to the server's memory size.
4412 set_max_delegations(void)
4415 * Allow at most 4 delegations per megabyte of RAM. Quick
4416 * estimates suggest that in the worst case (where every delegation
4417 * is for a different inode), a delegation could take about 1.5K,
4418 * giving a worst case usage of about 6% of memory.
4420 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4423 /* initialization to perform when the nfsd service is started: */
4426 __nfs4_state_start(void)
4430 boot_time = get_seconds();
4431 locks_start_grace(&nfsd4_manager);
4432 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4434 ret = set_callback_cred();
4437 laundry_wq = create_singlethread_workqueue("nfsd4");
4438 if (laundry_wq == NULL)
4440 ret = nfsd4_create_callback_queue();
4442 goto out_free_laundry;
4443 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4444 set_max_delegations();
4447 destroy_workqueue(laundry_wq);
4452 nfs4_state_start(void)
4454 nfsd4_load_reboot_recovery_data();
4455 return __nfs4_state_start();
4459 __nfs4_state_shutdown(void)
4462 struct nfs4_client *clp = NULL;
4463 struct nfs4_delegation *dp = NULL;
4464 struct list_head *pos, *next, reaplist;
4466 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4467 while (!list_empty(&conf_id_hashtbl[i])) {
4468 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4471 while (!list_empty(&unconf_str_hashtbl[i])) {
4472 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4476 INIT_LIST_HEAD(&reaplist);
4477 spin_lock(&recall_lock);
4478 list_for_each_safe(pos, next, &del_recall_lru) {
4479 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4480 list_move(&dp->dl_recall_lru, &reaplist);
4482 spin_unlock(&recall_lock);
4483 list_for_each_safe(pos, next, &reaplist) {
4484 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4485 list_del_init(&dp->dl_recall_lru);
4486 unhash_delegation(dp);
4489 nfsd4_shutdown_recdir();
4493 nfs4_state_shutdown(void)
4495 cancel_delayed_work_sync(&laundromat_work);
4496 destroy_workqueue(laundry_wq);
4497 locks_end_grace(&nfsd4_manager);
4499 nfs4_release_reclaim();
4500 __nfs4_state_shutdown();
4501 nfs4_unlock_state();
4502 nfsd4_destroy_callback_queue();