2 * linux/fs/nfsd/nfs4state.c
4 * Copyright (c) 2001 The Regents of the University of Michigan.
7 * Kendrick Smith <kmsmith@umich.edu>
8 * Andy Adamson <kandros@umich.edu>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
59 #define NFSDDBG_FACILITY NFSDDBG_PROC
62 static time_t lease_time
= 90; /* default lease time */
63 static time_t user_lease_time
= 90;
64 static time_t boot_time
;
65 static u32 current_ownerid
= 1;
66 static u32 current_fileid
= 1;
67 static u32 current_delegid
= 1;
69 static stateid_t zerostateid
; /* bits all 0 */
70 static stateid_t onestateid
; /* bits all 1 */
71 static u64 current_sessionid
= 1;
73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
74 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
76 /* forward declarations */
77 static struct nfs4_stateid
* find_stateid(stateid_t
*stid
, int flags
);
78 static struct nfs4_delegation
* find_delegation_stateid(struct inode
*ino
, stateid_t
*stid
);
79 static char user_recovery_dirname
[PATH_MAX
] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir
);
84 /* Currently used for almost all code touching nfsv4 state: */
85 static DEFINE_MUTEX(client_mutex
);
88 * Currently used for the del_recall_lru and file hash table. In an
89 * effort to decrease the scope of the client_mutex, this spinlock may
90 * eventually cover more:
92 static DEFINE_SPINLOCK(recall_lock
);
94 static struct kmem_cache
*stateowner_slab
= NULL
;
95 static struct kmem_cache
*file_slab
= NULL
;
96 static struct kmem_cache
*stateid_slab
= NULL
;
97 static struct kmem_cache
*deleg_slab
= NULL
;
100 nfs4_lock_state(void)
102 mutex_lock(&client_mutex
);
106 nfs4_unlock_state(void)
108 mutex_unlock(&client_mutex
);
112 opaque_hashval(const void *ptr
, int nbytes
)
114 unsigned char *cptr
= (unsigned char *) ptr
;
124 static struct list_head del_recall_lru
;
127 put_nfs4_file(struct nfs4_file
*fi
)
129 if (atomic_dec_and_lock(&fi
->fi_ref
, &recall_lock
)) {
130 list_del(&fi
->fi_hash
);
131 spin_unlock(&recall_lock
);
133 kmem_cache_free(file_slab
, fi
);
138 get_nfs4_file(struct nfs4_file
*fi
)
140 atomic_inc(&fi
->fi_ref
);
143 static int num_delegations
;
144 unsigned int max_delegations
;
147 * Open owner state (share locks)
150 /* hash tables for nfs4_stateowner */
151 #define OWNER_HASH_BITS 8
152 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
153 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
155 #define ownerid_hashval(id) \
156 ((id) & OWNER_HASH_MASK)
157 #define ownerstr_hashval(clientid, ownername) \
158 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
160 static struct list_head ownerid_hashtbl
[OWNER_HASH_SIZE
];
161 static struct list_head ownerstr_hashtbl
[OWNER_HASH_SIZE
];
163 /* hash table for nfs4_file */
164 #define FILE_HASH_BITS 8
165 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
166 #define FILE_HASH_MASK (FILE_HASH_SIZE - 1)
167 /* hash table for (open)nfs4_stateid */
168 #define STATEID_HASH_BITS 10
169 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
170 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
172 #define file_hashval(x) \
173 hash_ptr(x, FILE_HASH_BITS)
174 #define stateid_hashval(owner_id, file_id) \
175 (((owner_id) + (file_id)) & STATEID_HASH_MASK)
177 static struct list_head file_hashtbl
[FILE_HASH_SIZE
];
178 static struct list_head stateid_hashtbl
[STATEID_HASH_SIZE
];
180 static struct nfs4_delegation
*
181 alloc_init_deleg(struct nfs4_client
*clp
, struct nfs4_stateid
*stp
, struct svc_fh
*current_fh
, u32 type
)
183 struct nfs4_delegation
*dp
;
184 struct nfs4_file
*fp
= stp
->st_file
;
185 struct nfs4_cb_conn
*cb
= &stp
->st_stateowner
->so_client
->cl_cb_conn
;
187 dprintk("NFSD alloc_init_deleg\n");
188 if (fp
->fi_had_conflict
)
190 if (num_delegations
> max_delegations
)
192 dp
= kmem_cache_alloc(deleg_slab
, GFP_KERNEL
);
196 INIT_LIST_HEAD(&dp
->dl_perfile
);
197 INIT_LIST_HEAD(&dp
->dl_perclnt
);
198 INIT_LIST_HEAD(&dp
->dl_recall_lru
);
203 get_file(stp
->st_vfs_file
);
204 dp
->dl_vfs_file
= stp
->st_vfs_file
;
206 dp
->dl_ident
= cb
->cb_ident
;
207 dp
->dl_stateid
.si_boot
= get_seconds();
208 dp
->dl_stateid
.si_stateownerid
= current_delegid
++;
209 dp
->dl_stateid
.si_fileid
= 0;
210 dp
->dl_stateid
.si_generation
= 0;
211 fh_copy_shallow(&dp
->dl_fh
, ¤t_fh
->fh_handle
);
213 atomic_set(&dp
->dl_count
, 1);
214 list_add(&dp
->dl_perfile
, &fp
->fi_delegations
);
215 list_add(&dp
->dl_perclnt
, &clp
->cl_delegations
);
220 nfs4_put_delegation(struct nfs4_delegation
*dp
)
222 if (atomic_dec_and_test(&dp
->dl_count
)) {
223 dprintk("NFSD: freeing dp %p\n",dp
);
224 put_nfs4_file(dp
->dl_file
);
225 kmem_cache_free(deleg_slab
, dp
);
230 /* Remove the associated file_lock first, then remove the delegation.
231 * lease_modify() is called to remove the FS_LEASE file_lock from
232 * the i_flock list, eventually calling nfsd's lock_manager
233 * fl_release_callback.
236 nfs4_close_delegation(struct nfs4_delegation
*dp
)
238 struct file
*filp
= dp
->dl_vfs_file
;
240 dprintk("NFSD: close_delegation dp %p\n",dp
);
241 dp
->dl_vfs_file
= NULL
;
242 /* The following nfsd_close may not actually close the file,
243 * but we want to remove the lease in any case. */
245 vfs_setlease(filp
, F_UNLCK
, &dp
->dl_flock
);
249 /* Called under the state lock. */
251 unhash_delegation(struct nfs4_delegation
*dp
)
253 list_del_init(&dp
->dl_perfile
);
254 list_del_init(&dp
->dl_perclnt
);
255 spin_lock(&recall_lock
);
256 list_del_init(&dp
->dl_recall_lru
);
257 spin_unlock(&recall_lock
);
258 nfs4_close_delegation(dp
);
259 nfs4_put_delegation(dp
);
266 /* Hash tables for nfs4_clientid state */
267 #define CLIENT_HASH_BITS 4
268 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
269 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
271 #define clientid_hashval(id) \
272 ((id) & CLIENT_HASH_MASK)
273 #define clientstr_hashval(name) \
274 (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
276 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
277 * used in reboot/reset lease grace period processing
279 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
280 * setclientid_confirmed info.
282 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
285 * client_lru holds client queue ordered by nfs4_client.cl_time
288 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
289 * for last close replay.
291 static struct list_head reclaim_str_hashtbl
[CLIENT_HASH_SIZE
];
292 static int reclaim_str_hashtbl_size
= 0;
293 static struct list_head conf_id_hashtbl
[CLIENT_HASH_SIZE
];
294 static struct list_head conf_str_hashtbl
[CLIENT_HASH_SIZE
];
295 static struct list_head unconf_str_hashtbl
[CLIENT_HASH_SIZE
];
296 static struct list_head unconf_id_hashtbl
[CLIENT_HASH_SIZE
];
297 static struct list_head client_lru
;
298 static struct list_head close_lru
;
300 static void unhash_generic_stateid(struct nfs4_stateid
*stp
)
302 list_del(&stp
->st_hash
);
303 list_del(&stp
->st_perfile
);
304 list_del(&stp
->st_perstateowner
);
307 static void free_generic_stateid(struct nfs4_stateid
*stp
)
309 put_nfs4_file(stp
->st_file
);
310 kmem_cache_free(stateid_slab
, stp
);
313 static void release_lock_stateid(struct nfs4_stateid
*stp
)
315 unhash_generic_stateid(stp
);
316 locks_remove_posix(stp
->st_vfs_file
, (fl_owner_t
)stp
->st_stateowner
);
317 free_generic_stateid(stp
);
320 static void unhash_lockowner(struct nfs4_stateowner
*sop
)
322 struct nfs4_stateid
*stp
;
324 list_del(&sop
->so_idhash
);
325 list_del(&sop
->so_strhash
);
326 list_del(&sop
->so_perstateid
);
327 while (!list_empty(&sop
->so_stateids
)) {
328 stp
= list_first_entry(&sop
->so_stateids
,
329 struct nfs4_stateid
, st_perstateowner
);
330 release_lock_stateid(stp
);
334 static void release_lockowner(struct nfs4_stateowner
*sop
)
336 unhash_lockowner(sop
);
337 nfs4_put_stateowner(sop
);
341 release_stateid_lockowners(struct nfs4_stateid
*open_stp
)
343 struct nfs4_stateowner
*lock_sop
;
345 while (!list_empty(&open_stp
->st_lockowners
)) {
346 lock_sop
= list_entry(open_stp
->st_lockowners
.next
,
347 struct nfs4_stateowner
, so_perstateid
);
348 /* list_del(&open_stp->st_lockowners); */
349 BUG_ON(lock_sop
->so_is_open_owner
);
350 release_lockowner(lock_sop
);
354 static void release_open_stateid(struct nfs4_stateid
*stp
)
356 unhash_generic_stateid(stp
);
357 release_stateid_lockowners(stp
);
358 nfsd_close(stp
->st_vfs_file
);
359 free_generic_stateid(stp
);
362 static void unhash_openowner(struct nfs4_stateowner
*sop
)
364 struct nfs4_stateid
*stp
;
366 list_del(&sop
->so_idhash
);
367 list_del(&sop
->so_strhash
);
368 list_del(&sop
->so_perclient
);
369 list_del(&sop
->so_perstateid
); /* XXX: necessary? */
370 while (!list_empty(&sop
->so_stateids
)) {
371 stp
= list_first_entry(&sop
->so_stateids
,
372 struct nfs4_stateid
, st_perstateowner
);
373 release_open_stateid(stp
);
377 static void release_openowner(struct nfs4_stateowner
*sop
)
379 unhash_openowner(sop
);
380 list_del(&sop
->so_close_lru
);
381 nfs4_put_stateowner(sop
);
384 static DEFINE_SPINLOCK(sessionid_lock
);
385 #define SESSION_HASH_SIZE 512
386 static struct list_head sessionid_hashtbl
[SESSION_HASH_SIZE
];
389 hash_sessionid(struct nfs4_sessionid
*sessionid
)
391 struct nfsd4_sessionid
*sid
= (struct nfsd4_sessionid
*)sessionid
;
393 return sid
->sequence
% SESSION_HASH_SIZE
;
397 dump_sessionid(const char *fn
, struct nfs4_sessionid
*sessionid
)
399 u32
*ptr
= (u32
*)(&sessionid
->data
[0]);
400 dprintk("%s: %u:%u:%u:%u\n", fn
, ptr
[0], ptr
[1], ptr
[2], ptr
[3]);
404 gen_sessionid(struct nfsd4_session
*ses
)
406 struct nfs4_client
*clp
= ses
->se_client
;
407 struct nfsd4_sessionid
*sid
;
409 sid
= (struct nfsd4_sessionid
*)ses
->se_sessionid
.data
;
410 sid
->clientid
= clp
->cl_clientid
;
411 sid
->sequence
= current_sessionid
++;
416 * Give the client the number of slots it requests bound by
417 * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
419 * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
420 * should (up to a point) re-negotiate active sessions and reduce their
421 * slot usage to make rooom for new connections. For now we just fail the
424 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs
*fchan
)
426 int status
= 0, np
= fchan
->maxreqs
* NFSD_PAGES_PER_SLOT
;
428 if (fchan
->maxreqs
< 1)
430 else if (fchan
->maxreqs
> NFSD_MAX_SLOTS_PER_SESSION
)
431 fchan
->maxreqs
= NFSD_MAX_SLOTS_PER_SESSION
;
433 spin_lock(&nfsd_serv
->sv_lock
);
434 if (np
+ nfsd_serv
->sv_drc_pages_used
> nfsd_serv
->sv_drc_max_pages
)
435 np
= nfsd_serv
->sv_drc_max_pages
- nfsd_serv
->sv_drc_pages_used
;
436 nfsd_serv
->sv_drc_pages_used
+= np
;
437 spin_unlock(&nfsd_serv
->sv_lock
);
440 status
= nfserr_resource
;
443 fchan
->maxreqs
= np
/ NFSD_PAGES_PER_SLOT
;
449 * fchan holds the client values on input, and the server values on output
451 static int init_forechannel_attrs(struct svc_rqst
*rqstp
,
452 struct nfsd4_channel_attrs
*session_fchan
,
453 struct nfsd4_channel_attrs
*fchan
)
456 __u32 maxcount
= svc_max_payload(rqstp
);
458 /* headerpadsz set to zero in encode routine */
460 /* Use the client's max request and max response size if possible */
461 if (fchan
->maxreq_sz
> maxcount
)
462 fchan
->maxreq_sz
= maxcount
;
463 session_fchan
->maxreq_sz
= fchan
->maxreq_sz
;
465 if (fchan
->maxresp_sz
> maxcount
)
466 fchan
->maxresp_sz
= maxcount
;
467 session_fchan
->maxresp_sz
= fchan
->maxresp_sz
;
469 /* Set the max response cached size our default which is
470 * a multiple of PAGE_SIZE and small */
471 session_fchan
->maxresp_cached
= NFSD_PAGES_PER_SLOT
* PAGE_SIZE
;
472 fchan
->maxresp_cached
= session_fchan
->maxresp_cached
;
474 /* Use the client's maxops if possible */
475 if (fchan
->maxops
> NFSD_MAX_OPS_PER_COMPOUND
)
476 fchan
->maxops
= NFSD_MAX_OPS_PER_COMPOUND
;
477 session_fchan
->maxops
= fchan
->maxops
;
479 /* try to use the client requested number of slots */
480 if (fchan
->maxreqs
> NFSD_MAX_SLOTS_PER_SESSION
)
481 fchan
->maxreqs
= NFSD_MAX_SLOTS_PER_SESSION
;
483 /* FIXME: Error means no more DRC pages so the server should
484 * recover pages from existing sessions. For now fail session
487 status
= set_forechannel_maxreqs(fchan
);
489 session_fchan
->maxreqs
= fchan
->maxreqs
;
494 alloc_init_session(struct svc_rqst
*rqstp
, struct nfs4_client
*clp
,
495 struct nfsd4_create_session
*cses
)
497 struct nfsd4_session
*new, tmp
;
498 int idx
, status
= nfserr_resource
, slotsize
;
500 memset(&tmp
, 0, sizeof(tmp
));
502 /* FIXME: For now, we just accept the client back channel attributes. */
503 tmp
.se_bchannel
= cses
->back_channel
;
504 status
= init_forechannel_attrs(rqstp
, &tmp
.se_fchannel
,
505 &cses
->fore_channel
);
509 /* allocate struct nfsd4_session and slot table in one piece */
510 slotsize
= tmp
.se_fchannel
.maxreqs
* sizeof(struct nfsd4_slot
);
511 new = kzalloc(sizeof(*new) + slotsize
, GFP_KERNEL
);
515 memcpy(new, &tmp
, sizeof(*new));
517 new->se_client
= clp
;
519 idx
= hash_sessionid(&new->se_sessionid
);
520 memcpy(clp
->cl_sessionid
.data
, new->se_sessionid
.data
,
521 NFS4_MAX_SESSIONID_LEN
);
523 new->se_flags
= cses
->flags
;
524 kref_init(&new->se_ref
);
525 spin_lock(&sessionid_lock
);
526 list_add(&new->se_hash
, &sessionid_hashtbl
[idx
]);
527 list_add(&new->se_perclnt
, &clp
->cl_sessions
);
528 spin_unlock(&sessionid_lock
);
535 /* caller must hold sessionid_lock */
536 static struct nfsd4_session
*
537 find_in_sessionid_hashtbl(struct nfs4_sessionid
*sessionid
)
539 struct nfsd4_session
*elem
;
542 dump_sessionid(__func__
, sessionid
);
543 idx
= hash_sessionid(sessionid
);
544 dprintk("%s: idx is %d\n", __func__
, idx
);
545 /* Search in the appropriate list */
546 list_for_each_entry(elem
, &sessionid_hashtbl
[idx
], se_hash
) {
547 dump_sessionid("list traversal", &elem
->se_sessionid
);
548 if (!memcmp(elem
->se_sessionid
.data
, sessionid
->data
,
549 NFS4_MAX_SESSIONID_LEN
)) {
554 dprintk("%s: session not found\n", __func__
);
558 /* caller must hold sessionid_lock */
560 unhash_session(struct nfsd4_session
*ses
)
562 list_del(&ses
->se_hash
);
563 list_del(&ses
->se_perclnt
);
567 release_session(struct nfsd4_session
*ses
)
569 spin_lock(&sessionid_lock
);
571 spin_unlock(&sessionid_lock
);
572 nfsd4_put_session(ses
);
575 static void nfsd4_release_respages(struct page
**respages
, short resused
);
578 free_session(struct kref
*kref
)
580 struct nfsd4_session
*ses
;
583 ses
= container_of(kref
, struct nfsd4_session
, se_ref
);
584 for (i
= 0; i
< ses
->se_fchannel
.maxreqs
; i
++) {
585 struct nfsd4_cache_entry
*e
= &ses
->se_slots
[i
].sl_cache_entry
;
586 nfsd4_release_respages(e
->ce_respages
, e
->ce_resused
);
592 renew_client(struct nfs4_client
*clp
)
595 * Move client to the end to the LRU list.
597 dprintk("renewing client (clientid %08x/%08x)\n",
598 clp
->cl_clientid
.cl_boot
,
599 clp
->cl_clientid
.cl_id
);
600 list_move_tail(&clp
->cl_lru
, &client_lru
);
601 clp
->cl_time
= get_seconds();
604 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
606 STALE_CLIENTID(clientid_t
*clid
)
608 if (clid
->cl_boot
== boot_time
)
610 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
611 clid
->cl_boot
, clid
->cl_id
, boot_time
);
616 * XXX Should we use a slab cache ?
617 * This type of memory management is somewhat inefficient, but we use it
618 * anyway since SETCLIENTID is not a common operation.
620 static struct nfs4_client
*alloc_client(struct xdr_netobj name
)
622 struct nfs4_client
*clp
;
624 clp
= kzalloc(sizeof(struct nfs4_client
), GFP_KERNEL
);
627 clp
->cl_name
.data
= kmalloc(name
.len
, GFP_KERNEL
);
628 if (clp
->cl_name
.data
== NULL
) {
632 memcpy(clp
->cl_name
.data
, name
.data
, name
.len
);
633 clp
->cl_name
.len
= name
.len
;
638 shutdown_callback_client(struct nfs4_client
*clp
)
640 struct rpc_clnt
*clnt
= clp
->cl_cb_conn
.cb_client
;
644 * Callback threads take a reference on the client, so there
645 * should be no outstanding callbacks at this point.
647 clp
->cl_cb_conn
.cb_client
= NULL
;
648 rpc_shutdown_client(clnt
);
650 if (clp
->cl_cb_conn
.cb_cred
) {
651 put_rpccred(clp
->cl_cb_conn
.cb_cred
);
652 clp
->cl_cb_conn
.cb_cred
= NULL
;
657 free_client(struct nfs4_client
*clp
)
659 shutdown_callback_client(clp
);
660 nfsd4_release_respages(clp
->cl_slot
.sl_cache_entry
.ce_respages
,
661 clp
->cl_slot
.sl_cache_entry
.ce_resused
);
662 if (clp
->cl_cred
.cr_group_info
)
663 put_group_info(clp
->cl_cred
.cr_group_info
);
664 kfree(clp
->cl_principal
);
665 kfree(clp
->cl_name
.data
);
670 put_nfs4_client(struct nfs4_client
*clp
)
672 if (atomic_dec_and_test(&clp
->cl_count
))
677 expire_client(struct nfs4_client
*clp
)
679 struct nfs4_stateowner
*sop
;
680 struct nfs4_delegation
*dp
;
681 struct list_head reaplist
;
683 dprintk("NFSD: expire_client cl_count %d\n",
684 atomic_read(&clp
->cl_count
));
686 INIT_LIST_HEAD(&reaplist
);
687 spin_lock(&recall_lock
);
688 while (!list_empty(&clp
->cl_delegations
)) {
689 dp
= list_entry(clp
->cl_delegations
.next
, struct nfs4_delegation
, dl_perclnt
);
690 dprintk("NFSD: expire client. dp %p, fp %p\n", dp
,
692 list_del_init(&dp
->dl_perclnt
);
693 list_move(&dp
->dl_recall_lru
, &reaplist
);
695 spin_unlock(&recall_lock
);
696 while (!list_empty(&reaplist
)) {
697 dp
= list_entry(reaplist
.next
, struct nfs4_delegation
, dl_recall_lru
);
698 list_del_init(&dp
->dl_recall_lru
);
699 unhash_delegation(dp
);
701 list_del(&clp
->cl_idhash
);
702 list_del(&clp
->cl_strhash
);
703 list_del(&clp
->cl_lru
);
704 while (!list_empty(&clp
->cl_openowners
)) {
705 sop
= list_entry(clp
->cl_openowners
.next
, struct nfs4_stateowner
, so_perclient
);
706 release_openowner(sop
);
708 while (!list_empty(&clp
->cl_sessions
)) {
709 struct nfsd4_session
*ses
;
710 ses
= list_entry(clp
->cl_sessions
.next
, struct nfsd4_session
,
712 release_session(ses
);
714 put_nfs4_client(clp
);
717 static struct nfs4_client
*create_client(struct xdr_netobj name
, char *recdir
)
719 struct nfs4_client
*clp
;
721 clp
= alloc_client(name
);
724 memcpy(clp
->cl_recdir
, recdir
, HEXDIR_LEN
);
725 atomic_set(&clp
->cl_count
, 1);
726 atomic_set(&clp
->cl_cb_conn
.cb_set
, 0);
727 INIT_LIST_HEAD(&clp
->cl_idhash
);
728 INIT_LIST_HEAD(&clp
->cl_strhash
);
729 INIT_LIST_HEAD(&clp
->cl_openowners
);
730 INIT_LIST_HEAD(&clp
->cl_delegations
);
731 INIT_LIST_HEAD(&clp
->cl_sessions
);
732 INIT_LIST_HEAD(&clp
->cl_lru
);
736 static void copy_verf(struct nfs4_client
*target
, nfs4_verifier
*source
)
738 memcpy(target
->cl_verifier
.data
, source
->data
,
739 sizeof(target
->cl_verifier
.data
));
742 static void copy_clid(struct nfs4_client
*target
, struct nfs4_client
*source
)
744 target
->cl_clientid
.cl_boot
= source
->cl_clientid
.cl_boot
;
745 target
->cl_clientid
.cl_id
= source
->cl_clientid
.cl_id
;
748 static void copy_cred(struct svc_cred
*target
, struct svc_cred
*source
)
750 target
->cr_uid
= source
->cr_uid
;
751 target
->cr_gid
= source
->cr_gid
;
752 target
->cr_group_info
= source
->cr_group_info
;
753 get_group_info(target
->cr_group_info
);
756 static int same_name(const char *n1
, const char *n2
)
758 return 0 == memcmp(n1
, n2
, HEXDIR_LEN
);
762 same_verf(nfs4_verifier
*v1
, nfs4_verifier
*v2
)
764 return 0 == memcmp(v1
->data
, v2
->data
, sizeof(v1
->data
));
768 same_clid(clientid_t
*cl1
, clientid_t
*cl2
)
770 return (cl1
->cl_boot
== cl2
->cl_boot
) && (cl1
->cl_id
== cl2
->cl_id
);
773 /* XXX what about NGROUP */
775 same_creds(struct svc_cred
*cr1
, struct svc_cred
*cr2
)
777 return cr1
->cr_uid
== cr2
->cr_uid
;
780 static void gen_clid(struct nfs4_client
*clp
)
782 static u32 current_clientid
= 1;
784 clp
->cl_clientid
.cl_boot
= boot_time
;
785 clp
->cl_clientid
.cl_id
= current_clientid
++;
788 static void gen_confirm(struct nfs4_client
*clp
)
793 p
= (u32
*)clp
->cl_confirm
.data
;
794 *p
++ = get_seconds();
798 static int check_name(struct xdr_netobj name
)
802 if (name
.len
> NFS4_OPAQUE_LIMIT
) {
803 dprintk("NFSD: check_name: name too long(%d)!\n", name
.len
);
810 add_to_unconfirmed(struct nfs4_client
*clp
, unsigned int strhashval
)
812 unsigned int idhashval
;
814 list_add(&clp
->cl_strhash
, &unconf_str_hashtbl
[strhashval
]);
815 idhashval
= clientid_hashval(clp
->cl_clientid
.cl_id
);
816 list_add(&clp
->cl_idhash
, &unconf_id_hashtbl
[idhashval
]);
817 list_add_tail(&clp
->cl_lru
, &client_lru
);
818 clp
->cl_time
= get_seconds();
822 move_to_confirmed(struct nfs4_client
*clp
)
824 unsigned int idhashval
= clientid_hashval(clp
->cl_clientid
.cl_id
);
825 unsigned int strhashval
;
827 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp
);
828 list_del_init(&clp
->cl_strhash
);
829 list_move(&clp
->cl_idhash
, &conf_id_hashtbl
[idhashval
]);
830 strhashval
= clientstr_hashval(clp
->cl_recdir
);
831 list_add(&clp
->cl_strhash
, &conf_str_hashtbl
[strhashval
]);
835 static struct nfs4_client
*
836 find_confirmed_client(clientid_t
*clid
)
838 struct nfs4_client
*clp
;
839 unsigned int idhashval
= clientid_hashval(clid
->cl_id
);
841 list_for_each_entry(clp
, &conf_id_hashtbl
[idhashval
], cl_idhash
) {
842 if (same_clid(&clp
->cl_clientid
, clid
))
848 static struct nfs4_client
*
849 find_unconfirmed_client(clientid_t
*clid
)
851 struct nfs4_client
*clp
;
852 unsigned int idhashval
= clientid_hashval(clid
->cl_id
);
854 list_for_each_entry(clp
, &unconf_id_hashtbl
[idhashval
], cl_idhash
) {
855 if (same_clid(&clp
->cl_clientid
, clid
))
862 * Return 1 iff clp's clientid establishment method matches the use_exchange_id
863 * parameter. Matching is based on the fact the at least one of the
864 * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
866 * FIXME: we need to unify the clientid namespaces for nfsv4.x
867 * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
868 * and SET_CLIENTID{,_CONFIRM}
871 match_clientid_establishment(struct nfs4_client
*clp
, bool use_exchange_id
)
873 bool has_exchange_flags
= (clp
->cl_exchange_flags
!= 0);
874 return use_exchange_id
== has_exchange_flags
;
877 static struct nfs4_client
*
878 find_confirmed_client_by_str(const char *dname
, unsigned int hashval
,
879 bool use_exchange_id
)
881 struct nfs4_client
*clp
;
883 list_for_each_entry(clp
, &conf_str_hashtbl
[hashval
], cl_strhash
) {
884 if (same_name(clp
->cl_recdir
, dname
) &&
885 match_clientid_establishment(clp
, use_exchange_id
))
891 static struct nfs4_client
*
892 find_unconfirmed_client_by_str(const char *dname
, unsigned int hashval
,
893 bool use_exchange_id
)
895 struct nfs4_client
*clp
;
897 list_for_each_entry(clp
, &unconf_str_hashtbl
[hashval
], cl_strhash
) {
898 if (same_name(clp
->cl_recdir
, dname
) &&
899 match_clientid_establishment(clp
, use_exchange_id
))
905 /* a helper function for parse_callback */
907 parse_octet(unsigned int *lenp
, char **addrp
)
909 unsigned int len
= *lenp
;
921 if ((c
< '0') || (c
> '9')) {
927 n
= (n
* 10) + (c
- '0');
938 /* parse and set the setclientid ipv4 callback address */
940 parse_ipv4(unsigned int addr_len
, char *addr_val
, unsigned int *cbaddrp
, unsigned short *cbportp
)
945 u32 addrlen
= addr_len
;
946 char *addr
= addr_val
;
951 for(i
= 4; i
> 0 ; i
--) {
952 if ((temp
= parse_octet(&addrlen
, &addr
)) < 0) {
955 cbaddr
|= (temp
<< shift
);
963 for(i
= 2; i
> 0 ; i
--) {
964 if ((temp
= parse_octet(&addrlen
, &addr
)) < 0) {
967 cbport
|= (temp
<< shift
);
976 gen_callback(struct nfs4_client
*clp
, struct nfsd4_setclientid
*se
)
978 struct nfs4_cb_conn
*cb
= &clp
->cl_cb_conn
;
980 /* Currently, we only support tcp for the callback channel */
981 if ((se
->se_callback_netid_len
!= 3) || memcmp((char *)se
->se_callback_netid_val
, "tcp", 3))
984 if ( !(parse_ipv4(se
->se_callback_addr_len
, se
->se_callback_addr_val
,
985 &cb
->cb_addr
, &cb
->cb_port
)))
987 cb
->cb_prog
= se
->se_callback_prog
;
988 cb
->cb_ident
= se
->se_callback_ident
;
991 dprintk(KERN_INFO
"NFSD: this client (clientid %08x/%08x) "
992 "will not receive delegations\n",
993 clp
->cl_clientid
.cl_boot
, clp
->cl_clientid
.cl_id
);
999 nfsd4_set_statp(struct svc_rqst
*rqstp
, __be32
*statp
)
1001 struct nfsd4_compoundres
*resp
= rqstp
->rq_resp
;
1003 resp
->cstate
.statp
= statp
;
1007 * Dereference the result pages.
1010 nfsd4_release_respages(struct page
**respages
, short resused
)
1014 dprintk("--> %s\n", __func__
);
1015 for (i
= 0; i
< resused
; i
++) {
1018 put_page(respages
[i
]);
1024 nfsd4_copy_pages(struct page
**topages
, struct page
**frompages
, short count
)
1028 for (i
= 0; i
< count
; i
++) {
1029 topages
[i
] = frompages
[i
];
1032 get_page(topages
[i
]);
1037 * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1038 * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1039 * length of the XDR response is less than se_fmaxresp_cached
1040 * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1041 * of the reply (e.g. readdir).
1043 * Store the base and length of the rq_req.head[0] page
1044 * of the NFSv4.1 data, just past the rpc header.
1047 nfsd4_store_cache_entry(struct nfsd4_compoundres
*resp
)
1049 struct nfsd4_cache_entry
*entry
= &resp
->cstate
.slot
->sl_cache_entry
;
1050 struct svc_rqst
*rqstp
= resp
->rqstp
;
1051 struct nfsd4_compoundargs
*args
= rqstp
->rq_argp
;
1052 struct nfsd4_op
*op
= &args
->ops
[resp
->opcnt
];
1053 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1055 dprintk("--> %s entry %p\n", __func__
, entry
);
1057 /* Don't cache a failed OP_SEQUENCE. */
1058 if (resp
->opcnt
== 1 && op
->opnum
== OP_SEQUENCE
&& resp
->cstate
.status
)
1061 nfsd4_release_respages(entry
->ce_respages
, entry
->ce_resused
);
1062 entry
->ce_opcnt
= resp
->opcnt
;
1063 entry
->ce_status
= resp
->cstate
.status
;
1066 * Don't need a page to cache just the sequence operation - the slot
1070 if (nfsd4_not_cached(resp
)) {
1071 entry
->ce_resused
= 0;
1072 entry
->ce_rpchdrlen
= 0;
1073 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__
,
1074 resp
->cstate
.slot
->sl_cache_entry
.ce_cachethis
);
1077 entry
->ce_resused
= rqstp
->rq_resused
;
1078 if (entry
->ce_resused
> NFSD_PAGES_PER_SLOT
+ 1)
1079 entry
->ce_resused
= NFSD_PAGES_PER_SLOT
+ 1;
1080 nfsd4_copy_pages(entry
->ce_respages
, rqstp
->rq_respages
,
1082 entry
->ce_datav
.iov_base
= resp
->cstate
.statp
;
1083 entry
->ce_datav
.iov_len
= resv
->iov_len
- ((char *)resp
->cstate
.statp
-
1084 (char *)page_address(rqstp
->rq_respages
[0]));
1085 /* Current request rpc header length*/
1086 entry
->ce_rpchdrlen
= (char *)resp
->cstate
.statp
-
1087 (char *)page_address(rqstp
->rq_respages
[0]);
1091 * We keep the rpc header, but take the nfs reply from the replycache.
1094 nfsd41_copy_replay_data(struct nfsd4_compoundres
*resp
,
1095 struct nfsd4_cache_entry
*entry
)
1097 struct svc_rqst
*rqstp
= resp
->rqstp
;
1098 struct kvec
*resv
= &resp
->rqstp
->rq_res
.head
[0];
1101 /* Current request rpc header length*/
1102 len
= (char *)resp
->cstate
.statp
-
1103 (char *)page_address(rqstp
->rq_respages
[0]);
1104 if (entry
->ce_datav
.iov_len
+ len
> PAGE_SIZE
) {
1105 dprintk("%s v41 cached reply too large (%Zd).\n", __func__
,
1106 entry
->ce_datav
.iov_len
);
1109 /* copy the cached reply nfsd data past the current rpc header */
1110 memcpy((char *)resv
->iov_base
+ len
, entry
->ce_datav
.iov_base
,
1111 entry
->ce_datav
.iov_len
);
1112 resv
->iov_len
= len
+ entry
->ce_datav
.iov_len
;
1117 * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1118 * cached page. Replace any futher replay pages from the cache.
1121 nfsd4_replay_cache_entry(struct nfsd4_compoundres
*resp
,
1122 struct nfsd4_sequence
*seq
)
1124 struct nfsd4_cache_entry
*entry
= &resp
->cstate
.slot
->sl_cache_entry
;
1127 dprintk("--> %s entry %p\n", __func__
, entry
);
1130 * If this is just the sequence operation, we did not keep
1131 * a page in the cache entry because we can just use the
1132 * slot info stored in struct nfsd4_sequence that was checked
1133 * against the slot in nfsd4_sequence().
1135 * This occurs when seq->cachethis is FALSE, or when the client
1136 * session inactivity timer fires and a solo sequence operation
1137 * is sent (lease renewal).
1139 if (seq
&& nfsd4_not_cached(resp
)) {
1140 seq
->maxslots
= resp
->cstate
.session
->se_fchannel
.maxreqs
;
1144 if (!nfsd41_copy_replay_data(resp
, entry
)) {
1146 * Not enough room to use the replay rpc header, send the
1147 * cached header. Release all the allocated result pages.
1149 svc_free_res_pages(resp
->rqstp
);
1150 nfsd4_copy_pages(resp
->rqstp
->rq_respages
, entry
->ce_respages
,
1153 /* Release all but the first allocated result page */
1155 resp
->rqstp
->rq_resused
--;
1156 svc_free_res_pages(resp
->rqstp
);
1158 nfsd4_copy_pages(&resp
->rqstp
->rq_respages
[1],
1159 &entry
->ce_respages
[1],
1160 entry
->ce_resused
- 1);
1163 resp
->rqstp
->rq_resused
= entry
->ce_resused
;
1164 resp
->opcnt
= entry
->ce_opcnt
;
1165 resp
->cstate
.iovlen
= entry
->ce_datav
.iov_len
+ entry
->ce_rpchdrlen
;
1166 status
= entry
->ce_status
;
1172 * Set the exchange_id flags returned by the server.
1175 nfsd4_set_ex_flags(struct nfs4_client
*new, struct nfsd4_exchange_id
*clid
)
1177 /* pNFS is not supported */
1178 new->cl_exchange_flags
|= EXCHGID4_FLAG_USE_NON_PNFS
;
1180 /* Referrals are supported, Migration is not. */
1181 new->cl_exchange_flags
|= EXCHGID4_FLAG_SUPP_MOVED_REFER
;
1183 /* set the wire flags to return to client. */
1184 clid
->flags
= new->cl_exchange_flags
;
1188 nfsd4_exchange_id(struct svc_rqst
*rqstp
,
1189 struct nfsd4_compound_state
*cstate
,
1190 struct nfsd4_exchange_id
*exid
)
1192 struct nfs4_client
*unconf
, *conf
, *new;
1194 unsigned int strhashval
;
1195 char dname
[HEXDIR_LEN
];
1196 nfs4_verifier verf
= exid
->verifier
;
1197 u32 ip_addr
= svc_addr_in(rqstp
)->sin_addr
.s_addr
;
1199 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1200 " ip_addr=%u flags %x, spa_how %d\n",
1201 __func__
, rqstp
, exid
, exid
->clname
.len
, exid
->clname
.data
,
1202 ip_addr
, exid
->flags
, exid
->spa_how
);
1204 if (!check_name(exid
->clname
) || (exid
->flags
& ~EXCHGID4_FLAG_MASK_A
))
1205 return nfserr_inval
;
1207 /* Currently only support SP4_NONE */
1208 switch (exid
->spa_how
) {
1212 return nfserr_encr_alg_unsupp
;
1214 BUG(); /* checked by xdr code */
1216 return nfserr_serverfault
; /* no excuse :-/ */
1219 status
= nfs4_make_rec_clidname(dname
, &exid
->clname
);
1224 strhashval
= clientstr_hashval(dname
);
1229 conf
= find_confirmed_client_by_str(dname
, strhashval
, true);
1231 if (!same_verf(&verf
, &conf
->cl_verifier
)) {
1232 /* 18.35.4 case 8 */
1233 if (exid
->flags
& EXCHGID4_FLAG_UPD_CONFIRMED_REC_A
) {
1234 status
= nfserr_not_same
;
1237 /* Client reboot: destroy old state */
1238 expire_client(conf
);
1241 if (!same_creds(&conf
->cl_cred
, &rqstp
->rq_cred
)) {
1242 /* 18.35.4 case 9 */
1243 if (exid
->flags
& EXCHGID4_FLAG_UPD_CONFIRMED_REC_A
) {
1244 status
= nfserr_perm
;
1247 expire_client(conf
);
1250 if (ip_addr
!= conf
->cl_addr
&&
1251 !(exid
->flags
& EXCHGID4_FLAG_UPD_CONFIRMED_REC_A
)) {
1252 /* Client collision. 18.35.4 case 3 */
1253 status
= nfserr_clid_inuse
;
1257 * Set bit when the owner id and verifier map to an already
1258 * confirmed client id (18.35.3).
1260 exid
->flags
|= EXCHGID4_FLAG_CONFIRMED_R
;
1263 * Falling into 18.35.4 case 2, possible router replay.
1264 * Leave confirmed record intact and return same result.
1266 copy_verf(conf
, &verf
);
1270 /* 18.35.4 case 7 */
1271 if (exid
->flags
& EXCHGID4_FLAG_UPD_CONFIRMED_REC_A
) {
1272 status
= nfserr_noent
;
1277 unconf
= find_unconfirmed_client_by_str(dname
, strhashval
, true);
1280 * Possible retry or client restart. Per 18.35.4 case 4,
1281 * a new unconfirmed record should be generated regardless
1282 * of whether any properties have changed.
1284 expire_client(unconf
);
1289 new = create_client(exid
->clname
, dname
);
1291 status
= nfserr_resource
;
1295 copy_verf(new, &verf
);
1296 copy_cred(&new->cl_cred
, &rqstp
->rq_cred
);
1297 new->cl_addr
= ip_addr
;
1300 add_to_unconfirmed(new, strhashval
);
1302 exid
->clientid
.cl_boot
= new->cl_clientid
.cl_boot
;
1303 exid
->clientid
.cl_id
= new->cl_clientid
.cl_id
;
1305 new->cl_slot
.sl_seqid
= 0;
1307 nfsd4_set_ex_flags(new, exid
);
1309 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1310 new->cl_slot
.sl_seqid
, new->cl_exchange_flags
);
1314 nfs4_unlock_state();
1316 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status
));
1321 check_slot_seqid(u32 seqid
, struct nfsd4_slot
*slot
)
1323 dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__
, seqid
,
1326 /* The slot is in use, and no response has been sent. */
1327 if (slot
->sl_inuse
) {
1328 if (seqid
== slot
->sl_seqid
)
1329 return nfserr_jukebox
;
1331 return nfserr_seq_misordered
;
1334 if (likely(seqid
== slot
->sl_seqid
+ 1))
1337 if (seqid
== slot
->sl_seqid
)
1338 return nfserr_replay_cache
;
1340 if (seqid
== 1 && (slot
->sl_seqid
+ 1) == 0)
1342 /* Misordered replay or misordered new request */
1343 return nfserr_seq_misordered
;
1347 nfsd4_create_session(struct svc_rqst
*rqstp
,
1348 struct nfsd4_compound_state
*cstate
,
1349 struct nfsd4_create_session
*cr_ses
)
1351 u32 ip_addr
= svc_addr_in(rqstp
)->sin_addr
.s_addr
;
1352 struct nfsd4_compoundres
*resp
= rqstp
->rq_resp
;
1353 struct nfs4_client
*conf
, *unconf
;
1354 struct nfsd4_slot
*slot
= NULL
;
1358 unconf
= find_unconfirmed_client(&cr_ses
->clientid
);
1359 conf
= find_confirmed_client(&cr_ses
->clientid
);
1362 slot
= &conf
->cl_slot
;
1363 status
= check_slot_seqid(cr_ses
->seqid
, slot
);
1364 if (status
== nfserr_replay_cache
) {
1365 dprintk("Got a create_session replay! seqid= %d\n",
1367 cstate
->slot
= slot
;
1368 cstate
->status
= status
;
1369 /* Return the cached reply status */
1370 status
= nfsd4_replay_cache_entry(resp
, NULL
);
1372 } else if (cr_ses
->seqid
!= conf
->cl_slot
.sl_seqid
+ 1) {
1373 status
= nfserr_seq_misordered
;
1374 dprintk("Sequence misordered!\n");
1375 dprintk("Expected seqid= %d but got seqid= %d\n",
1376 slot
->sl_seqid
, cr_ses
->seqid
);
1379 conf
->cl_slot
.sl_seqid
++;
1380 } else if (unconf
) {
1381 if (!same_creds(&unconf
->cl_cred
, &rqstp
->rq_cred
) ||
1382 (ip_addr
!= unconf
->cl_addr
)) {
1383 status
= nfserr_clid_inuse
;
1387 slot
= &unconf
->cl_slot
;
1388 status
= check_slot_seqid(cr_ses
->seqid
, slot
);
1390 /* an unconfirmed replay returns misordered */
1391 status
= nfserr_seq_misordered
;
1395 slot
->sl_seqid
++; /* from 0 to 1 */
1396 move_to_confirmed(unconf
);
1399 * We do not support RDMA or persistent sessions
1401 cr_ses
->flags
&= ~SESSION4_PERSIST
;
1402 cr_ses
->flags
&= ~SESSION4_RDMA
;
1406 status
= nfserr_stale_clientid
;
1410 status
= alloc_init_session(rqstp
, conf
, cr_ses
);
1414 memcpy(cr_ses
->sessionid
.data
, conf
->cl_sessionid
.data
,
1415 NFS4_MAX_SESSIONID_LEN
);
1416 cr_ses
->seqid
= slot
->sl_seqid
;
1418 slot
->sl_inuse
= true;
1419 cstate
->slot
= slot
;
1420 /* Ensure a page is used for the cache */
1421 slot
->sl_cache_entry
.ce_cachethis
= 1;
1423 nfs4_unlock_state();
1424 dprintk("%s returns %d\n", __func__
, ntohl(status
));
1429 nfsd4_destroy_session(struct svc_rqst
*r
,
1430 struct nfsd4_compound_state
*cstate
,
1431 struct nfsd4_destroy_session
*sessionid
)
1433 struct nfsd4_session
*ses
;
1434 u32 status
= nfserr_badsession
;
1437 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1438 * - Should we return nfserr_back_chan_busy if waiting for
1439 * callbacks on to-be-destroyed session?
1440 * - Do we need to clear any callback info from previous session?
1443 dump_sessionid(__func__
, &sessionid
->sessionid
);
1444 spin_lock(&sessionid_lock
);
1445 ses
= find_in_sessionid_hashtbl(&sessionid
->sessionid
);
1447 spin_unlock(&sessionid_lock
);
1451 unhash_session(ses
);
1452 spin_unlock(&sessionid_lock
);
1454 /* wait for callbacks */
1455 shutdown_callback_client(ses
->se_client
);
1456 nfsd4_put_session(ses
);
1459 dprintk("%s returns %d\n", __func__
, ntohl(status
));
1464 nfsd4_sequence(struct svc_rqst
*rqstp
,
1465 struct nfsd4_compound_state
*cstate
,
1466 struct nfsd4_sequence
*seq
)
1468 struct nfsd4_compoundres
*resp
= rqstp
->rq_resp
;
1469 struct nfsd4_session
*session
;
1470 struct nfsd4_slot
*slot
;
1473 if (resp
->opcnt
!= 1)
1474 return nfserr_sequence_pos
;
1476 spin_lock(&sessionid_lock
);
1477 status
= nfserr_badsession
;
1478 session
= find_in_sessionid_hashtbl(&seq
->sessionid
);
1482 status
= nfserr_badslot
;
1483 if (seq
->slotid
>= session
->se_fchannel
.maxreqs
)
1486 slot
= &session
->se_slots
[seq
->slotid
];
1487 dprintk("%s: slotid %d\n", __func__
, seq
->slotid
);
1489 status
= check_slot_seqid(seq
->seqid
, slot
);
1490 if (status
== nfserr_replay_cache
) {
1491 cstate
->slot
= slot
;
1492 cstate
->session
= session
;
1493 /* Return the cached reply status and set cstate->status
1494 * for nfsd4_svc_encode_compoundres processing */
1495 status
= nfsd4_replay_cache_entry(resp
, seq
);
1496 cstate
->status
= nfserr_replay_cache
;
1502 /* Success! bump slot seqid */
1503 slot
->sl_inuse
= true;
1504 slot
->sl_seqid
= seq
->seqid
;
1505 slot
->sl_cache_entry
.ce_cachethis
= seq
->cachethis
;
1506 /* Always set the cache entry cachethis for solo sequence */
1507 if (nfsd4_is_solo_sequence(resp
))
1508 slot
->sl_cache_entry
.ce_cachethis
= 1;
1510 cstate
->slot
= slot
;
1511 cstate
->session
= session
;
1514 /* Renew the clientid on success and on replay.
1515 * Hold a session reference until done processing the compound:
1516 * nfsd4_put_session called only if the cstate slot is set.
1518 renew_client(session
->se_client
);
1519 nfsd4_get_session(session
);
1521 spin_unlock(&sessionid_lock
);
1522 dprintk("%s: return %d\n", __func__
, ntohl(status
));
1527 nfsd4_setclientid(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
1528 struct nfsd4_setclientid
*setclid
)
1530 struct sockaddr_in
*sin
= svc_addr_in(rqstp
);
1531 struct xdr_netobj clname
= {
1532 .len
= setclid
->se_namelen
,
1533 .data
= setclid
->se_name
,
1535 nfs4_verifier clverifier
= setclid
->se_verf
;
1536 unsigned int strhashval
;
1537 struct nfs4_client
*conf
, *unconf
, *new;
1540 char dname
[HEXDIR_LEN
];
1542 if (!check_name(clname
))
1543 return nfserr_inval
;
1545 status
= nfs4_make_rec_clidname(dname
, &clname
);
1550 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1551 * We get here on a DRC miss.
1554 strhashval
= clientstr_hashval(dname
);
1557 conf
= find_confirmed_client_by_str(dname
, strhashval
, false);
1559 /* RFC 3530 14.2.33 CASE 0: */
1560 status
= nfserr_clid_inuse
;
1561 if (!same_creds(&conf
->cl_cred
, &rqstp
->rq_cred
)) {
1562 dprintk("NFSD: setclientid: string in use by client"
1563 " at %pI4\n", &conf
->cl_addr
);
1568 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1569 * has a description of SETCLIENTID request processing consisting
1570 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1572 unconf
= find_unconfirmed_client_by_str(dname
, strhashval
, false);
1573 status
= nfserr_resource
;
1576 * RFC 3530 14.2.33 CASE 4:
1577 * placed first, because it is the normal case
1580 expire_client(unconf
);
1581 new = create_client(clname
, dname
);
1585 } else if (same_verf(&conf
->cl_verifier
, &clverifier
)) {
1587 * RFC 3530 14.2.33 CASE 1:
1588 * probable callback update
1591 /* Note this is removing unconfirmed {*x***},
1592 * which is stronger than RFC recommended {vxc**}.
1593 * This has the advantage that there is at most
1594 * one {*x***} in either list at any time.
1596 expire_client(unconf
);
1598 new = create_client(clname
, dname
);
1601 copy_clid(new, conf
);
1602 } else if (!unconf
) {
1604 * RFC 3530 14.2.33 CASE 2:
1605 * probable client reboot; state will be removed if
1608 new = create_client(clname
, dname
);
1614 * RFC 3530 14.2.33 CASE 3:
1615 * probable client reboot; state will be removed if
1618 expire_client(unconf
);
1619 new = create_client(clname
, dname
);
1624 copy_verf(new, &clverifier
);
1625 new->cl_addr
= sin
->sin_addr
.s_addr
;
1626 new->cl_flavor
= rqstp
->rq_flavor
;
1627 princ
= svc_gss_principal(rqstp
);
1629 new->cl_principal
= kstrdup(princ
, GFP_KERNEL
);
1630 if (new->cl_principal
== NULL
) {
1635 copy_cred(&new->cl_cred
, &rqstp
->rq_cred
);
1637 gen_callback(new, setclid
);
1638 add_to_unconfirmed(new, strhashval
);
1639 setclid
->se_clientid
.cl_boot
= new->cl_clientid
.cl_boot
;
1640 setclid
->se_clientid
.cl_id
= new->cl_clientid
.cl_id
;
1641 memcpy(setclid
->se_confirm
.data
, new->cl_confirm
.data
, sizeof(setclid
->se_confirm
.data
));
1644 nfs4_unlock_state();
1650 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1651 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1652 * bullets, labeled as CASE1 - CASE4 below.
1655 nfsd4_setclientid_confirm(struct svc_rqst
*rqstp
,
1656 struct nfsd4_compound_state
*cstate
,
1657 struct nfsd4_setclientid_confirm
*setclientid_confirm
)
1659 struct sockaddr_in
*sin
= svc_addr_in(rqstp
);
1660 struct nfs4_client
*conf
, *unconf
;
1661 nfs4_verifier confirm
= setclientid_confirm
->sc_confirm
;
1662 clientid_t
* clid
= &setclientid_confirm
->sc_clientid
;
1665 if (STALE_CLIENTID(clid
))
1666 return nfserr_stale_clientid
;
1668 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1669 * We get here on a DRC miss.
1674 conf
= find_confirmed_client(clid
);
1675 unconf
= find_unconfirmed_client(clid
);
1677 status
= nfserr_clid_inuse
;
1678 if (conf
&& conf
->cl_addr
!= sin
->sin_addr
.s_addr
)
1680 if (unconf
&& unconf
->cl_addr
!= sin
->sin_addr
.s_addr
)
1684 * section 14.2.34 of RFC 3530 has a description of
1685 * SETCLIENTID_CONFIRM request processing consisting
1686 * of 4 bullet points, labeled as CASE1 - CASE4 below.
1688 if (conf
&& unconf
&& same_verf(&confirm
, &unconf
->cl_confirm
)) {
1690 * RFC 3530 14.2.34 CASE 1:
1693 if (!same_creds(&conf
->cl_cred
, &unconf
->cl_cred
))
1694 status
= nfserr_clid_inuse
;
1696 /* XXX: We just turn off callbacks until we can handle
1697 * change request correctly. */
1698 atomic_set(&conf
->cl_cb_conn
.cb_set
, 0);
1699 expire_client(unconf
);
1703 } else if (conf
&& !unconf
) {
1705 * RFC 3530 14.2.34 CASE 2:
1706 * probable retransmitted request; play it safe and
1709 if (!same_creds(&conf
->cl_cred
, &rqstp
->rq_cred
))
1710 status
= nfserr_clid_inuse
;
1713 } else if (!conf
&& unconf
1714 && same_verf(&unconf
->cl_confirm
, &confirm
)) {
1716 * RFC 3530 14.2.34 CASE 3:
1717 * Normal case; new or rebooted client:
1719 if (!same_creds(&unconf
->cl_cred
, &rqstp
->rq_cred
)) {
1720 status
= nfserr_clid_inuse
;
1723 clientstr_hashval(unconf
->cl_recdir
);
1724 conf
= find_confirmed_client_by_str(unconf
->cl_recdir
,
1727 nfsd4_remove_clid_dir(conf
);
1728 expire_client(conf
);
1730 move_to_confirmed(unconf
);
1732 nfsd4_probe_callback(conf
);
1735 } else if ((!conf
|| (conf
&& !same_verf(&conf
->cl_confirm
, &confirm
)))
1736 && (!unconf
|| (unconf
&& !same_verf(&unconf
->cl_confirm
,
1739 * RFC 3530 14.2.34 CASE 4:
1740 * Client probably hasn't noticed that we rebooted yet.
1742 status
= nfserr_stale_clientid
;
1744 /* check that we have hit one of the cases...*/
1745 status
= nfserr_clid_inuse
;
1748 nfs4_unlock_state();
1752 /* OPEN Share state helper functions */
1753 static inline struct nfs4_file
*
1754 alloc_init_file(struct inode
*ino
)
1756 struct nfs4_file
*fp
;
1757 unsigned int hashval
= file_hashval(ino
);
1759 fp
= kmem_cache_alloc(file_slab
, GFP_KERNEL
);
1761 atomic_set(&fp
->fi_ref
, 1);
1762 INIT_LIST_HEAD(&fp
->fi_hash
);
1763 INIT_LIST_HEAD(&fp
->fi_stateids
);
1764 INIT_LIST_HEAD(&fp
->fi_delegations
);
1765 spin_lock(&recall_lock
);
1766 list_add(&fp
->fi_hash
, &file_hashtbl
[hashval
]);
1767 spin_unlock(&recall_lock
);
1768 fp
->fi_inode
= igrab(ino
);
1769 fp
->fi_id
= current_fileid
++;
1770 fp
->fi_had_conflict
= false;
1777 nfsd4_free_slab(struct kmem_cache
**slab
)
1781 kmem_cache_destroy(*slab
);
1786 nfsd4_free_slabs(void)
1788 nfsd4_free_slab(&stateowner_slab
);
1789 nfsd4_free_slab(&file_slab
);
1790 nfsd4_free_slab(&stateid_slab
);
1791 nfsd4_free_slab(&deleg_slab
);
1795 nfsd4_init_slabs(void)
1797 stateowner_slab
= kmem_cache_create("nfsd4_stateowners",
1798 sizeof(struct nfs4_stateowner
), 0, 0, NULL
);
1799 if (stateowner_slab
== NULL
)
1801 file_slab
= kmem_cache_create("nfsd4_files",
1802 sizeof(struct nfs4_file
), 0, 0, NULL
);
1803 if (file_slab
== NULL
)
1805 stateid_slab
= kmem_cache_create("nfsd4_stateids",
1806 sizeof(struct nfs4_stateid
), 0, 0, NULL
);
1807 if (stateid_slab
== NULL
)
1809 deleg_slab
= kmem_cache_create("nfsd4_delegations",
1810 sizeof(struct nfs4_delegation
), 0, 0, NULL
);
1811 if (deleg_slab
== NULL
)
1816 dprintk("nfsd4: out of memory while initializing nfsv4\n");
1821 nfs4_free_stateowner(struct kref
*kref
)
1823 struct nfs4_stateowner
*sop
=
1824 container_of(kref
, struct nfs4_stateowner
, so_ref
);
1825 kfree(sop
->so_owner
.data
);
1826 kmem_cache_free(stateowner_slab
, sop
);
1829 static inline struct nfs4_stateowner
*
1830 alloc_stateowner(struct xdr_netobj
*owner
)
1832 struct nfs4_stateowner
*sop
;
1834 if ((sop
= kmem_cache_alloc(stateowner_slab
, GFP_KERNEL
))) {
1835 if ((sop
->so_owner
.data
= kmalloc(owner
->len
, GFP_KERNEL
))) {
1836 memcpy(sop
->so_owner
.data
, owner
->data
, owner
->len
);
1837 sop
->so_owner
.len
= owner
->len
;
1838 kref_init(&sop
->so_ref
);
1841 kmem_cache_free(stateowner_slab
, sop
);
1846 static struct nfs4_stateowner
*
1847 alloc_init_open_stateowner(unsigned int strhashval
, struct nfs4_client
*clp
, struct nfsd4_open
*open
) {
1848 struct nfs4_stateowner
*sop
;
1849 struct nfs4_replay
*rp
;
1850 unsigned int idhashval
;
1852 if (!(sop
= alloc_stateowner(&open
->op_owner
)))
1854 idhashval
= ownerid_hashval(current_ownerid
);
1855 INIT_LIST_HEAD(&sop
->so_idhash
);
1856 INIT_LIST_HEAD(&sop
->so_strhash
);
1857 INIT_LIST_HEAD(&sop
->so_perclient
);
1858 INIT_LIST_HEAD(&sop
->so_stateids
);
1859 INIT_LIST_HEAD(&sop
->so_perstateid
); /* not used */
1860 INIT_LIST_HEAD(&sop
->so_close_lru
);
1862 list_add(&sop
->so_idhash
, &ownerid_hashtbl
[idhashval
]);
1863 list_add(&sop
->so_strhash
, &ownerstr_hashtbl
[strhashval
]);
1864 list_add(&sop
->so_perclient
, &clp
->cl_openowners
);
1865 sop
->so_is_open_owner
= 1;
1866 sop
->so_id
= current_ownerid
++;
1867 sop
->so_client
= clp
;
1868 sop
->so_seqid
= open
->op_seqid
;
1869 sop
->so_confirmed
= 0;
1870 rp
= &sop
->so_replay
;
1871 rp
->rp_status
= nfserr_serverfault
;
1873 rp
->rp_buf
= rp
->rp_ibuf
;
1878 init_stateid(struct nfs4_stateid
*stp
, struct nfs4_file
*fp
, struct nfsd4_open
*open
) {
1879 struct nfs4_stateowner
*sop
= open
->op_stateowner
;
1880 unsigned int hashval
= stateid_hashval(sop
->so_id
, fp
->fi_id
);
1882 INIT_LIST_HEAD(&stp
->st_hash
);
1883 INIT_LIST_HEAD(&stp
->st_perstateowner
);
1884 INIT_LIST_HEAD(&stp
->st_lockowners
);
1885 INIT_LIST_HEAD(&stp
->st_perfile
);
1886 list_add(&stp
->st_hash
, &stateid_hashtbl
[hashval
]);
1887 list_add(&stp
->st_perstateowner
, &sop
->so_stateids
);
1888 list_add(&stp
->st_perfile
, &fp
->fi_stateids
);
1889 stp
->st_stateowner
= sop
;
1892 stp
->st_stateid
.si_boot
= get_seconds();
1893 stp
->st_stateid
.si_stateownerid
= sop
->so_id
;
1894 stp
->st_stateid
.si_fileid
= fp
->fi_id
;
1895 stp
->st_stateid
.si_generation
= 0;
1896 stp
->st_access_bmap
= 0;
1897 stp
->st_deny_bmap
= 0;
1898 __set_bit(open
->op_share_access
& ~NFS4_SHARE_WANT_MASK
,
1899 &stp
->st_access_bmap
);
1900 __set_bit(open
->op_share_deny
, &stp
->st_deny_bmap
);
1901 stp
->st_openstp
= NULL
;
1905 move_to_close_lru(struct nfs4_stateowner
*sop
)
1907 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop
);
1909 list_move_tail(&sop
->so_close_lru
, &close_lru
);
1910 sop
->so_time
= get_seconds();
1914 same_owner_str(struct nfs4_stateowner
*sop
, struct xdr_netobj
*owner
,
1917 return (sop
->so_owner
.len
== owner
->len
) &&
1918 0 == memcmp(sop
->so_owner
.data
, owner
->data
, owner
->len
) &&
1919 (sop
->so_client
->cl_clientid
.cl_id
== clid
->cl_id
);
1922 static struct nfs4_stateowner
*
1923 find_openstateowner_str(unsigned int hashval
, struct nfsd4_open
*open
)
1925 struct nfs4_stateowner
*so
= NULL
;
1927 list_for_each_entry(so
, &ownerstr_hashtbl
[hashval
], so_strhash
) {
1928 if (same_owner_str(so
, &open
->op_owner
, &open
->op_clientid
))
1934 /* search file_hashtbl[] for file */
1935 static struct nfs4_file
*
1936 find_file(struct inode
*ino
)
1938 unsigned int hashval
= file_hashval(ino
);
1939 struct nfs4_file
*fp
;
1941 spin_lock(&recall_lock
);
1942 list_for_each_entry(fp
, &file_hashtbl
[hashval
], fi_hash
) {
1943 if (fp
->fi_inode
== ino
) {
1945 spin_unlock(&recall_lock
);
1949 spin_unlock(&recall_lock
);
1953 static inline int access_valid(u32 x
, u32 minorversion
)
1955 if ((x
& NFS4_SHARE_ACCESS_MASK
) < NFS4_SHARE_ACCESS_READ
)
1957 if ((x
& NFS4_SHARE_ACCESS_MASK
) > NFS4_SHARE_ACCESS_BOTH
)
1959 x
&= ~NFS4_SHARE_ACCESS_MASK
;
1960 if (minorversion
&& x
) {
1961 if ((x
& NFS4_SHARE_WANT_MASK
) > NFS4_SHARE_WANT_CANCEL
)
1963 if ((x
& NFS4_SHARE_WHEN_MASK
) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED
)
1965 x
&= ~(NFS4_SHARE_WANT_MASK
| NFS4_SHARE_WHEN_MASK
);
1972 static inline int deny_valid(u32 x
)
1974 /* Note: unlike access bits, deny bits may be zero. */
1975 return x
<= NFS4_SHARE_DENY_BOTH
;
1979 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1980 * st_{access,deny}_bmap field of the stateid, in order to track not
1981 * only what share bits are currently in force, but also what
1982 * combinations of share bits previous opens have used. This allows us
1983 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1984 * return an error if the client attempt to downgrade to a combination
1985 * of share bits not explicable by closing some of its previous opens.
1987 * XXX: This enforcement is actually incomplete, since we don't keep
1988 * track of access/deny bit combinations; so, e.g., we allow:
1990 * OPEN allow read, deny write
1991 * OPEN allow both, deny none
1992 * DOWNGRADE allow read, deny none
1994 * which we should reject.
1997 set_access(unsigned int *access
, unsigned long bmap
) {
2001 for (i
= 1; i
< 4; i
++) {
2002 if (test_bit(i
, &bmap
))
2008 set_deny(unsigned int *deny
, unsigned long bmap
) {
2012 for (i
= 0; i
< 4; i
++) {
2013 if (test_bit(i
, &bmap
))
2019 test_share(struct nfs4_stateid
*stp
, struct nfsd4_open
*open
) {
2020 unsigned int access
, deny
;
2022 set_access(&access
, stp
->st_access_bmap
);
2023 set_deny(&deny
, stp
->st_deny_bmap
);
2024 if ((access
& open
->op_share_deny
) || (deny
& open
->op_share_access
))
2030 * Called to check deny when READ with all zero stateid or
2031 * WRITE with all zero or all one stateid
2034 nfs4_share_conflict(struct svc_fh
*current_fh
, unsigned int deny_type
)
2036 struct inode
*ino
= current_fh
->fh_dentry
->d_inode
;
2037 struct nfs4_file
*fp
;
2038 struct nfs4_stateid
*stp
;
2041 dprintk("NFSD: nfs4_share_conflict\n");
2043 fp
= find_file(ino
);
2046 ret
= nfserr_locked
;
2047 /* Search for conflicting share reservations */
2048 list_for_each_entry(stp
, &fp
->fi_stateids
, st_perfile
) {
2049 if (test_bit(deny_type
, &stp
->st_deny_bmap
) ||
2050 test_bit(NFS4_SHARE_DENY_BOTH
, &stp
->st_deny_bmap
))
2060 nfs4_file_downgrade(struct file
*filp
, unsigned int share_access
)
2062 if (share_access
& NFS4_SHARE_ACCESS_WRITE
) {
2063 drop_file_write_access(filp
);
2064 filp
->f_mode
= (filp
->f_mode
| FMODE_READ
) & ~FMODE_WRITE
;
2069 * Spawn a thread to perform a recall on the delegation represented
2070 * by the lease (file_lock)
2072 * Called from break_lease() with lock_kernel() held.
2073 * Note: we assume break_lease will only call this *once* for any given
2077 void nfsd_break_deleg_cb(struct file_lock
*fl
)
2079 struct nfs4_delegation
*dp
= (struct nfs4_delegation
*)fl
->fl_owner
;
2081 dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp
,fl
);
2085 /* We're assuming the state code never drops its reference
2086 * without first removing the lease. Since we're in this lease
2087 * callback (and since the lease code is serialized by the kernel
2088 * lock) we know the server hasn't removed the lease yet, we know
2089 * it's safe to take a reference: */
2090 atomic_inc(&dp
->dl_count
);
2091 atomic_inc(&dp
->dl_client
->cl_count
);
2093 spin_lock(&recall_lock
);
2094 list_add_tail(&dp
->dl_recall_lru
, &del_recall_lru
);
2095 spin_unlock(&recall_lock
);
2097 /* only place dl_time is set. protected by lock_kernel*/
2098 dp
->dl_time
= get_seconds();
2101 * We don't want the locks code to timeout the lease for us;
2102 * we'll remove it ourself if the delegation isn't returned
2105 fl
->fl_break_time
= 0;
2107 dp
->dl_file
->fi_had_conflict
= true;
2108 nfsd4_cb_recall(dp
);
2112 * The file_lock is being reapd.
2114 * Called by locks_free_lock() with lock_kernel() held.
2117 void nfsd_release_deleg_cb(struct file_lock
*fl
)
2119 struct nfs4_delegation
*dp
= (struct nfs4_delegation
*)fl
->fl_owner
;
2121 dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl
,dp
, atomic_read(&dp
->dl_count
));
2123 if (!(fl
->fl_flags
& FL_LEASE
) || !dp
)
2125 dp
->dl_flock
= NULL
;
2129 * Set the delegation file_lock back pointer.
2131 * Called from setlease() with lock_kernel() held.
2134 void nfsd_copy_lock_deleg_cb(struct file_lock
*new, struct file_lock
*fl
)
2136 struct nfs4_delegation
*dp
= (struct nfs4_delegation
*)new->fl_owner
;
2138 dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp
);
2145 * Called from setlease() with lock_kernel() held
2148 int nfsd_same_client_deleg_cb(struct file_lock
*onlist
, struct file_lock
*try)
2150 struct nfs4_delegation
*onlistd
=
2151 (struct nfs4_delegation
*)onlist
->fl_owner
;
2152 struct nfs4_delegation
*tryd
=
2153 (struct nfs4_delegation
*)try->fl_owner
;
2155 if (onlist
->fl_lmops
!= try->fl_lmops
)
2158 return onlistd
->dl_client
== tryd
->dl_client
;
2163 int nfsd_change_deleg_cb(struct file_lock
**onlist
, int arg
)
2166 return lease_modify(onlist
, arg
);
2171 static struct lock_manager_operations nfsd_lease_mng_ops
= {
2172 .fl_break
= nfsd_break_deleg_cb
,
2173 .fl_release_private
= nfsd_release_deleg_cb
,
2174 .fl_copy_lock
= nfsd_copy_lock_deleg_cb
,
2175 .fl_mylease
= nfsd_same_client_deleg_cb
,
2176 .fl_change
= nfsd_change_deleg_cb
,
2181 nfsd4_process_open1(struct nfsd4_compound_state
*cstate
,
2182 struct nfsd4_open
*open
)
2184 clientid_t
*clientid
= &open
->op_clientid
;
2185 struct nfs4_client
*clp
= NULL
;
2186 unsigned int strhashval
;
2187 struct nfs4_stateowner
*sop
= NULL
;
2189 if (!check_name(open
->op_owner
))
2190 return nfserr_inval
;
2192 if (STALE_CLIENTID(&open
->op_clientid
))
2193 return nfserr_stale_clientid
;
2195 strhashval
= ownerstr_hashval(clientid
->cl_id
, open
->op_owner
);
2196 sop
= find_openstateowner_str(strhashval
, open
);
2197 open
->op_stateowner
= sop
;
2199 /* Make sure the client's lease hasn't expired. */
2200 clp
= find_confirmed_client(clientid
);
2202 return nfserr_expired
;
2205 /* When sessions are used, skip open sequenceid processing */
2206 if (nfsd4_has_session(cstate
))
2208 if (!sop
->so_confirmed
) {
2209 /* Replace unconfirmed owners without checking for replay. */
2210 clp
= sop
->so_client
;
2211 release_openowner(sop
);
2212 open
->op_stateowner
= NULL
;
2215 if (open
->op_seqid
== sop
->so_seqid
- 1) {
2216 if (sop
->so_replay
.rp_buflen
)
2217 return nfserr_replay_me
;
2218 /* The original OPEN failed so spectacularly
2219 * that we don't even have replay data saved!
2220 * Therefore, we have no choice but to continue
2221 * processing this OPEN; presumably, we'll
2222 * fail again for the same reason.
2224 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2227 if (open
->op_seqid
!= sop
->so_seqid
)
2228 return nfserr_bad_seqid
;
2230 if (open
->op_stateowner
== NULL
) {
2231 sop
= alloc_init_open_stateowner(strhashval
, clp
, open
);
2233 return nfserr_resource
;
2234 open
->op_stateowner
= sop
;
2236 list_del_init(&sop
->so_close_lru
);
2237 renew_client(sop
->so_client
);
2241 static inline __be32
2242 nfs4_check_delegmode(struct nfs4_delegation
*dp
, int flags
)
2244 if ((flags
& WR_STATE
) && (dp
->dl_type
== NFS4_OPEN_DELEGATE_READ
))
2245 return nfserr_openmode
;
2250 static struct nfs4_delegation
*
2251 find_delegation_file(struct nfs4_file
*fp
, stateid_t
*stid
)
2253 struct nfs4_delegation
*dp
;
2255 list_for_each_entry(dp
, &fp
->fi_delegations
, dl_perfile
) {
2256 if (dp
->dl_stateid
.si_stateownerid
== stid
->si_stateownerid
)
2263 nfs4_check_deleg(struct nfs4_file
*fp
, struct nfsd4_open
*open
,
2264 struct nfs4_delegation
**dp
)
2267 __be32 status
= nfserr_bad_stateid
;
2269 *dp
= find_delegation_file(fp
, &open
->op_delegate_stateid
);
2272 flags
= open
->op_share_access
== NFS4_SHARE_ACCESS_READ
?
2273 RD_STATE
: WR_STATE
;
2274 status
= nfs4_check_delegmode(*dp
, flags
);
2278 if (open
->op_claim_type
!= NFS4_OPEN_CLAIM_DELEGATE_CUR
)
2282 open
->op_stateowner
->so_confirmed
= 1;
2287 nfs4_check_open(struct nfs4_file
*fp
, struct nfsd4_open
*open
, struct nfs4_stateid
**stpp
)
2289 struct nfs4_stateid
*local
;
2290 __be32 status
= nfserr_share_denied
;
2291 struct nfs4_stateowner
*sop
= open
->op_stateowner
;
2293 list_for_each_entry(local
, &fp
->fi_stateids
, st_perfile
) {
2294 /* ignore lock owners */
2295 if (local
->st_stateowner
->so_is_open_owner
== 0)
2297 /* remember if we have seen this open owner */
2298 if (local
->st_stateowner
== sop
)
2300 /* check for conflicting share reservations */
2301 if (!test_share(local
, open
))
2309 static inline struct nfs4_stateid
*
2310 nfs4_alloc_stateid(void)
2312 return kmem_cache_alloc(stateid_slab
, GFP_KERNEL
);
2316 nfs4_new_open(struct svc_rqst
*rqstp
, struct nfs4_stateid
**stpp
,
2317 struct nfs4_delegation
*dp
,
2318 struct svc_fh
*cur_fh
, int flags
)
2320 struct nfs4_stateid
*stp
;
2322 stp
= nfs4_alloc_stateid();
2324 return nfserr_resource
;
2327 get_file(dp
->dl_vfs_file
);
2328 stp
->st_vfs_file
= dp
->dl_vfs_file
;
2331 status
= nfsd_open(rqstp
, cur_fh
, S_IFREG
, flags
,
2334 if (status
== nfserr_dropit
)
2335 status
= nfserr_jukebox
;
2336 kmem_cache_free(stateid_slab
, stp
);
2344 static inline __be32
2345 nfsd4_truncate(struct svc_rqst
*rqstp
, struct svc_fh
*fh
,
2346 struct nfsd4_open
*open
)
2348 struct iattr iattr
= {
2349 .ia_valid
= ATTR_SIZE
,
2352 if (!open
->op_truncate
)
2354 if (!(open
->op_share_access
& NFS4_SHARE_ACCESS_WRITE
))
2355 return nfserr_inval
;
2356 return nfsd_setattr(rqstp
, fh
, &iattr
, 0, (time_t)0);
2360 nfs4_upgrade_open(struct svc_rqst
*rqstp
, struct svc_fh
*cur_fh
, struct nfs4_stateid
*stp
, struct nfsd4_open
*open
)
2362 struct file
*filp
= stp
->st_vfs_file
;
2363 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
2364 unsigned int share_access
, new_writer
;
2367 set_access(&share_access
, stp
->st_access_bmap
);
2368 new_writer
= (~share_access
) & open
->op_share_access
2369 & NFS4_SHARE_ACCESS_WRITE
;
2372 int err
= get_write_access(inode
);
2374 return nfserrno(err
);
2375 err
= mnt_want_write(cur_fh
->fh_export
->ex_path
.mnt
);
2377 return nfserrno(err
);
2378 file_take_write(filp
);
2380 status
= nfsd4_truncate(rqstp
, cur_fh
, open
);
2383 put_write_access(inode
);
2386 /* remember the open */
2387 filp
->f_mode
|= open
->op_share_access
;
2388 __set_bit(open
->op_share_access
, &stp
->st_access_bmap
);
2389 __set_bit(open
->op_share_deny
, &stp
->st_deny_bmap
);
2396 nfs4_set_claim_prev(struct nfsd4_open
*open
)
2398 open
->op_stateowner
->so_confirmed
= 1;
2399 open
->op_stateowner
->so_client
->cl_firststate
= 1;
2403 * Attempt to hand out a delegation.
2406 nfs4_open_delegation(struct svc_fh
*fh
, struct nfsd4_open
*open
, struct nfs4_stateid
*stp
)
2408 struct nfs4_delegation
*dp
;
2409 struct nfs4_stateowner
*sop
= stp
->st_stateowner
;
2410 struct nfs4_cb_conn
*cb
= &sop
->so_client
->cl_cb_conn
;
2411 struct file_lock fl
, *flp
= &fl
;
2412 int status
, flag
= 0;
2414 flag
= NFS4_OPEN_DELEGATE_NONE
;
2415 open
->op_recall
= 0;
2416 switch (open
->op_claim_type
) {
2417 case NFS4_OPEN_CLAIM_PREVIOUS
:
2418 if (!atomic_read(&cb
->cb_set
))
2419 open
->op_recall
= 1;
2420 flag
= open
->op_delegate_type
;
2421 if (flag
== NFS4_OPEN_DELEGATE_NONE
)
2424 case NFS4_OPEN_CLAIM_NULL
:
2425 /* Let's not give out any delegations till everyone's
2426 * had the chance to reclaim theirs.... */
2427 if (locks_in_grace())
2429 if (!atomic_read(&cb
->cb_set
) || !sop
->so_confirmed
)
2431 if (open
->op_share_access
& NFS4_SHARE_ACCESS_WRITE
)
2432 flag
= NFS4_OPEN_DELEGATE_WRITE
;
2434 flag
= NFS4_OPEN_DELEGATE_READ
;
2440 dp
= alloc_init_deleg(sop
->so_client
, stp
, fh
, flag
);
2442 flag
= NFS4_OPEN_DELEGATE_NONE
;
2445 locks_init_lock(&fl
);
2446 fl
.fl_lmops
= &nfsd_lease_mng_ops
;
2447 fl
.fl_flags
= FL_LEASE
;
2448 fl
.fl_type
= flag
== NFS4_OPEN_DELEGATE_READ
? F_RDLCK
: F_WRLCK
;
2449 fl
.fl_end
= OFFSET_MAX
;
2450 fl
.fl_owner
= (fl_owner_t
)dp
;
2451 fl
.fl_file
= stp
->st_vfs_file
;
2452 fl
.fl_pid
= current
->tgid
;
2454 /* vfs_setlease checks to see if delegation should be handed out.
2455 * the lock_manager callbacks fl_mylease and fl_change are used
2457 if ((status
= vfs_setlease(stp
->st_vfs_file
, fl
.fl_type
, &flp
))) {
2458 dprintk("NFSD: setlease failed [%d], no delegation\n", status
);
2459 unhash_delegation(dp
);
2460 flag
= NFS4_OPEN_DELEGATE_NONE
;
2464 memcpy(&open
->op_delegate_stateid
, &dp
->dl_stateid
, sizeof(dp
->dl_stateid
));
2466 dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2467 dp
->dl_stateid
.si_boot
,
2468 dp
->dl_stateid
.si_stateownerid
,
2469 dp
->dl_stateid
.si_fileid
,
2470 dp
->dl_stateid
.si_generation
);
2472 if (open
->op_claim_type
== NFS4_OPEN_CLAIM_PREVIOUS
2473 && flag
== NFS4_OPEN_DELEGATE_NONE
2474 && open
->op_delegate_type
!= NFS4_OPEN_DELEGATE_NONE
)
2475 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2476 open
->op_delegate_type
= flag
;
2480 * called with nfs4_lock_state() held.
2483 nfsd4_process_open2(struct svc_rqst
*rqstp
, struct svc_fh
*current_fh
, struct nfsd4_open
*open
)
2485 struct nfsd4_compoundres
*resp
= rqstp
->rq_resp
;
2486 struct nfs4_file
*fp
= NULL
;
2487 struct inode
*ino
= current_fh
->fh_dentry
->d_inode
;
2488 struct nfs4_stateid
*stp
= NULL
;
2489 struct nfs4_delegation
*dp
= NULL
;
2492 status
= nfserr_inval
;
2493 if (!access_valid(open
->op_share_access
, resp
->cstate
.minorversion
)
2494 || !deny_valid(open
->op_share_deny
))
2497 * Lookup file; if found, lookup stateid and check open request,
2498 * and check for delegations in the process of being recalled.
2499 * If not found, create the nfs4_file struct
2501 fp
= find_file(ino
);
2503 if ((status
= nfs4_check_open(fp
, open
, &stp
)))
2505 status
= nfs4_check_deleg(fp
, open
, &dp
);
2509 status
= nfserr_bad_stateid
;
2510 if (open
->op_claim_type
== NFS4_OPEN_CLAIM_DELEGATE_CUR
)
2512 status
= nfserr_resource
;
2513 fp
= alloc_init_file(ino
);
2519 * OPEN the file, or upgrade an existing OPEN.
2520 * If truncate fails, the OPEN fails.
2523 /* Stateid was found, this is an OPEN upgrade */
2524 status
= nfs4_upgrade_open(rqstp
, current_fh
, stp
, open
);
2527 update_stateid(&stp
->st_stateid
);
2529 /* Stateid was not found, this is a new OPEN */
2531 if (open
->op_share_access
& NFS4_SHARE_ACCESS_READ
)
2532 flags
|= NFSD_MAY_READ
;
2533 if (open
->op_share_access
& NFS4_SHARE_ACCESS_WRITE
)
2534 flags
|= NFSD_MAY_WRITE
;
2535 status
= nfs4_new_open(rqstp
, &stp
, dp
, current_fh
, flags
);
2538 init_stateid(stp
, fp
, open
);
2539 status
= nfsd4_truncate(rqstp
, current_fh
, open
);
2541 release_open_stateid(stp
);
2544 if (nfsd4_has_session(&resp
->cstate
))
2545 update_stateid(&stp
->st_stateid
);
2547 memcpy(&open
->op_stateid
, &stp
->st_stateid
, sizeof(stateid_t
));
2549 if (nfsd4_has_session(&resp
->cstate
))
2550 open
->op_stateowner
->so_confirmed
= 1;
2553 * Attempt to hand out a delegation. No error return, because the
2554 * OPEN succeeds even if we fail.
2556 nfs4_open_delegation(current_fh
, open
, stp
);
2560 dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2561 stp
->st_stateid
.si_boot
, stp
->st_stateid
.si_stateownerid
,
2562 stp
->st_stateid
.si_fileid
, stp
->st_stateid
.si_generation
);
2566 if (status
== 0 && open
->op_claim_type
== NFS4_OPEN_CLAIM_PREVIOUS
)
2567 nfs4_set_claim_prev(open
);
2569 * To finish the open response, we just need to set the rflags.
2571 open
->op_rflags
= NFS4_OPEN_RESULT_LOCKTYPE_POSIX
;
2572 if (!open
->op_stateowner
->so_confirmed
&&
2573 !nfsd4_has_session(&resp
->cstate
))
2574 open
->op_rflags
|= NFS4_OPEN_RESULT_CONFIRM
;
2580 nfsd4_renew(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
2583 struct nfs4_client
*clp
;
2587 dprintk("process_renew(%08x/%08x): starting\n",
2588 clid
->cl_boot
, clid
->cl_id
);
2589 status
= nfserr_stale_clientid
;
2590 if (STALE_CLIENTID(clid
))
2592 clp
= find_confirmed_client(clid
);
2593 status
= nfserr_expired
;
2595 /* We assume the client took too long to RENEW. */
2596 dprintk("nfsd4_renew: clientid not found!\n");
2600 status
= nfserr_cb_path_down
;
2601 if (!list_empty(&clp
->cl_delegations
)
2602 && !atomic_read(&clp
->cl_cb_conn
.cb_set
))
2606 nfs4_unlock_state();
2610 struct lock_manager nfsd4_manager
= {
2614 nfsd4_end_grace(void)
2616 dprintk("NFSD: end of grace period\n");
2617 nfsd4_recdir_purge_old();
2618 locks_end_grace(&nfsd4_manager
);
2622 nfs4_laundromat(void)
2624 struct nfs4_client
*clp
;
2625 struct nfs4_stateowner
*sop
;
2626 struct nfs4_delegation
*dp
;
2627 struct list_head
*pos
, *next
, reaplist
;
2628 time_t cutoff
= get_seconds() - NFSD_LEASE_TIME
;
2629 time_t t
, clientid_val
= NFSD_LEASE_TIME
;
2630 time_t u
, test_val
= NFSD_LEASE_TIME
;
2634 dprintk("NFSD: laundromat service - starting\n");
2635 if (locks_in_grace())
2637 list_for_each_safe(pos
, next
, &client_lru
) {
2638 clp
= list_entry(pos
, struct nfs4_client
, cl_lru
);
2639 if (time_after((unsigned long)clp
->cl_time
, (unsigned long)cutoff
)) {
2640 t
= clp
->cl_time
- cutoff
;
2641 if (clientid_val
> t
)
2645 dprintk("NFSD: purging unused client (clientid %08x)\n",
2646 clp
->cl_clientid
.cl_id
);
2647 nfsd4_remove_clid_dir(clp
);
2650 INIT_LIST_HEAD(&reaplist
);
2651 spin_lock(&recall_lock
);
2652 list_for_each_safe(pos
, next
, &del_recall_lru
) {
2653 dp
= list_entry (pos
, struct nfs4_delegation
, dl_recall_lru
);
2654 if (time_after((unsigned long)dp
->dl_time
, (unsigned long)cutoff
)) {
2655 u
= dp
->dl_time
- cutoff
;
2660 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2662 list_move(&dp
->dl_recall_lru
, &reaplist
);
2664 spin_unlock(&recall_lock
);
2665 list_for_each_safe(pos
, next
, &reaplist
) {
2666 dp
= list_entry (pos
, struct nfs4_delegation
, dl_recall_lru
);
2667 list_del_init(&dp
->dl_recall_lru
);
2668 unhash_delegation(dp
);
2670 test_val
= NFSD_LEASE_TIME
;
2671 list_for_each_safe(pos
, next
, &close_lru
) {
2672 sop
= list_entry(pos
, struct nfs4_stateowner
, so_close_lru
);
2673 if (time_after((unsigned long)sop
->so_time
, (unsigned long)cutoff
)) {
2674 u
= sop
->so_time
- cutoff
;
2679 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2681 release_openowner(sop
);
2683 if (clientid_val
< NFSD_LAUNDROMAT_MINTIMEOUT
)
2684 clientid_val
= NFSD_LAUNDROMAT_MINTIMEOUT
;
2685 nfs4_unlock_state();
2686 return clientid_val
;
2689 static struct workqueue_struct
*laundry_wq
;
2690 static void laundromat_main(struct work_struct
*);
2691 static DECLARE_DELAYED_WORK(laundromat_work
, laundromat_main
);
2694 laundromat_main(struct work_struct
*not_used
)
2698 t
= nfs4_laundromat();
2699 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t
);
2700 queue_delayed_work(laundry_wq
, &laundromat_work
, t
*HZ
);
2703 static struct nfs4_stateowner
*
2704 search_close_lru(u32 st_id
, int flags
)
2706 struct nfs4_stateowner
*local
= NULL
;
2708 if (flags
& CLOSE_STATE
) {
2709 list_for_each_entry(local
, &close_lru
, so_close_lru
) {
2710 if (local
->so_id
== st_id
)
2718 nfs4_check_fh(struct svc_fh
*fhp
, struct nfs4_stateid
*stp
)
2720 return fhp
->fh_dentry
->d_inode
!= stp
->st_vfs_file
->f_path
.dentry
->d_inode
;
2724 STALE_STATEID(stateid_t
*stateid
)
2726 if (time_after((unsigned long)boot_time
,
2727 (unsigned long)stateid
->si_boot
)) {
2728 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2729 stateid
->si_boot
, stateid
->si_stateownerid
,
2730 stateid
->si_fileid
, stateid
->si_generation
);
2737 EXPIRED_STATEID(stateid_t
*stateid
)
2739 if (time_before((unsigned long)boot_time
,
2740 ((unsigned long)stateid
->si_boot
)) &&
2741 time_before((unsigned long)(stateid
->si_boot
+ lease_time
), get_seconds())) {
2742 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2743 stateid
->si_boot
, stateid
->si_stateownerid
,
2744 stateid
->si_fileid
, stateid
->si_generation
);
2751 stateid_error_map(stateid_t
*stateid
)
2753 if (STALE_STATEID(stateid
))
2754 return nfserr_stale_stateid
;
2755 if (EXPIRED_STATEID(stateid
))
2756 return nfserr_expired
;
2758 dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2759 stateid
->si_boot
, stateid
->si_stateownerid
,
2760 stateid
->si_fileid
, stateid
->si_generation
);
2761 return nfserr_bad_stateid
;
2765 access_permit_read(unsigned long access_bmap
)
2767 return test_bit(NFS4_SHARE_ACCESS_READ
, &access_bmap
) ||
2768 test_bit(NFS4_SHARE_ACCESS_BOTH
, &access_bmap
) ||
2769 test_bit(NFS4_SHARE_ACCESS_WRITE
, &access_bmap
);
2773 access_permit_write(unsigned long access_bmap
)
2775 return test_bit(NFS4_SHARE_ACCESS_WRITE
, &access_bmap
) ||
2776 test_bit(NFS4_SHARE_ACCESS_BOTH
, &access_bmap
);
2780 __be32
nfs4_check_openmode(struct nfs4_stateid
*stp
, int flags
)
2782 __be32 status
= nfserr_openmode
;
2784 if ((flags
& WR_STATE
) && (!access_permit_write(stp
->st_access_bmap
)))
2786 if ((flags
& RD_STATE
) && (!access_permit_read(stp
->st_access_bmap
)))
2793 static inline __be32
2794 check_special_stateids(svc_fh
*current_fh
, stateid_t
*stateid
, int flags
)
2796 if (ONE_STATEID(stateid
) && (flags
& RD_STATE
))
2798 else if (locks_in_grace()) {
2799 /* Answer in remaining cases depends on existance of
2800 * conflicting state; so we must wait out the grace period. */
2801 return nfserr_grace
;
2802 } else if (flags
& WR_STATE
)
2803 return nfs4_share_conflict(current_fh
,
2804 NFS4_SHARE_DENY_WRITE
);
2805 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2806 return nfs4_share_conflict(current_fh
,
2807 NFS4_SHARE_DENY_READ
);
2811 * Allow READ/WRITE during grace period on recovered state only for files
2812 * that are not able to provide mandatory locking.
2815 grace_disallows_io(struct inode
*inode
)
2817 return locks_in_grace() && mandatory_lock(inode
);
2820 static int check_stateid_generation(stateid_t
*in
, stateid_t
*ref
, int flags
)
2823 * When sessions are used the stateid generation number is ignored
2826 if ((flags
& HAS_SESSION
) && in
->si_generation
== 0)
2829 /* If the client sends us a stateid from the future, it's buggy: */
2830 if (in
->si_generation
> ref
->si_generation
)
2831 return nfserr_bad_stateid
;
2833 * The following, however, can happen. For example, if the
2834 * client sends an open and some IO at the same time, the open
2835 * may bump si_generation while the IO is still in flight.
2836 * Thanks to hard links and renames, the client never knows what
2837 * file an open will affect. So it could avoid that situation
2838 * only by serializing all opens and IO from the same open
2839 * owner. To recover from the old_stateid error, the client
2840 * will just have to retry the IO:
2842 if (in
->si_generation
< ref
->si_generation
)
2843 return nfserr_old_stateid
;
2848 static int is_delegation_stateid(stateid_t
*stateid
)
2850 return stateid
->si_fileid
== 0;
2854 * Checks for stateid operations
2857 nfs4_preprocess_stateid_op(struct nfsd4_compound_state
*cstate
,
2858 stateid_t
*stateid
, int flags
, struct file
**filpp
)
2860 struct nfs4_stateid
*stp
= NULL
;
2861 struct nfs4_delegation
*dp
= NULL
;
2862 struct svc_fh
*current_fh
= &cstate
->current_fh
;
2863 struct inode
*ino
= current_fh
->fh_dentry
->d_inode
;
2869 if (grace_disallows_io(ino
))
2870 return nfserr_grace
;
2872 if (nfsd4_has_session(cstate
))
2873 flags
|= HAS_SESSION
;
2875 if (ZERO_STATEID(stateid
) || ONE_STATEID(stateid
))
2876 return check_special_stateids(current_fh
, stateid
, flags
);
2878 status
= nfserr_stale_stateid
;
2879 if (STALE_STATEID(stateid
))
2882 status
= nfserr_bad_stateid
;
2883 if (is_delegation_stateid(stateid
)) {
2884 dp
= find_delegation_stateid(ino
, stateid
);
2886 status
= stateid_error_map(stateid
);
2889 status
= check_stateid_generation(stateid
, &dp
->dl_stateid
,
2893 status
= nfs4_check_delegmode(dp
, flags
);
2896 renew_client(dp
->dl_client
);
2898 *filpp
= dp
->dl_vfs_file
;
2899 } else { /* open or lock stateid */
2900 stp
= find_stateid(stateid
, flags
);
2902 status
= stateid_error_map(stateid
);
2905 if (nfs4_check_fh(current_fh
, stp
))
2907 if (!stp
->st_stateowner
->so_confirmed
)
2909 status
= check_stateid_generation(stateid
, &stp
->st_stateid
,
2913 status
= nfs4_check_openmode(stp
, flags
);
2916 renew_client(stp
->st_stateowner
->so_client
);
2918 *filpp
= stp
->st_vfs_file
;
2928 return (type
== NFS4_READW_LT
|| type
== NFS4_READ_LT
) ?
2929 RD_STATE
: WR_STATE
;
2933 * Checks for sequence id mutating operations.
2936 nfs4_preprocess_seqid_op(struct nfsd4_compound_state
*cstate
, u32 seqid
,
2937 stateid_t
*stateid
, int flags
,
2938 struct nfs4_stateowner
**sopp
,
2939 struct nfs4_stateid
**stpp
, struct nfsd4_lock
*lock
)
2941 struct nfs4_stateid
*stp
;
2942 struct nfs4_stateowner
*sop
;
2943 struct svc_fh
*current_fh
= &cstate
->current_fh
;
2946 dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2947 "stateid = (%08x/%08x/%08x/%08x)\n", seqid
,
2948 stateid
->si_boot
, stateid
->si_stateownerid
, stateid
->si_fileid
,
2949 stateid
->si_generation
);
2954 if (ZERO_STATEID(stateid
) || ONE_STATEID(stateid
)) {
2955 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2956 return nfserr_bad_stateid
;
2959 if (STALE_STATEID(stateid
))
2960 return nfserr_stale_stateid
;
2962 if (nfsd4_has_session(cstate
))
2963 flags
|= HAS_SESSION
;
2966 * We return BAD_STATEID if filehandle doesn't match stateid,
2967 * the confirmed flag is incorrecly set, or the generation
2968 * number is incorrect.
2970 stp
= find_stateid(stateid
, flags
);
2973 * Also, we should make sure this isn't just the result of
2976 sop
= search_close_lru(stateid
->si_stateownerid
, flags
);
2978 return stateid_error_map(stateid
);
2984 *sopp
= sop
= stp
->st_stateowner
;
2987 clientid_t
*lockclid
= &lock
->v
.new.clientid
;
2988 struct nfs4_client
*clp
= sop
->so_client
;
2992 lkflg
= setlkflg(lock
->lk_type
);
2994 if (lock
->lk_is_new
) {
2995 if (!sop
->so_is_open_owner
)
2996 return nfserr_bad_stateid
;
2997 if (!(flags
& HAS_SESSION
) &&
2998 !same_clid(&clp
->cl_clientid
, lockclid
))
2999 return nfserr_bad_stateid
;
3000 /* stp is the open stateid */
3001 status
= nfs4_check_openmode(stp
, lkflg
);
3005 /* stp is the lock stateid */
3006 status
= nfs4_check_openmode(stp
->st_openstp
, lkflg
);
3012 if (nfs4_check_fh(current_fh
, stp
)) {
3013 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3014 return nfserr_bad_stateid
;
3018 * We now validate the seqid and stateid generation numbers.
3019 * For the moment, we ignore the possibility of
3020 * generation number wraparound.
3022 if (!(flags
& HAS_SESSION
) && seqid
!= sop
->so_seqid
)
3025 if (sop
->so_confirmed
&& flags
& CONFIRM
) {
3026 dprintk("NFSD: preprocess_seqid_op: expected"
3027 " unconfirmed stateowner!\n");
3028 return nfserr_bad_stateid
;
3030 if (!sop
->so_confirmed
&& !(flags
& CONFIRM
)) {
3031 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3032 " confirmed yet!\n");
3033 return nfserr_bad_stateid
;
3035 status
= check_stateid_generation(stateid
, &stp
->st_stateid
, flags
);
3038 renew_client(sop
->so_client
);
3042 if (seqid
== sop
->so_seqid
- 1) {
3043 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3044 /* indicate replay to calling function */
3045 return nfserr_replay_me
;
3047 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3048 sop
->so_seqid
, seqid
);
3050 return nfserr_bad_seqid
;
3054 nfsd4_open_confirm(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
3055 struct nfsd4_open_confirm
*oc
)
3058 struct nfs4_stateowner
*sop
;
3059 struct nfs4_stateid
*stp
;
3061 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3062 (int)cstate
->current_fh
.fh_dentry
->d_name
.len
,
3063 cstate
->current_fh
.fh_dentry
->d_name
.name
);
3065 status
= fh_verify(rqstp
, &cstate
->current_fh
, S_IFREG
, 0);
3071 if ((status
= nfs4_preprocess_seqid_op(cstate
,
3072 oc
->oc_seqid
, &oc
->oc_req_stateid
,
3073 CONFIRM
| OPEN_STATE
,
3074 &oc
->oc_stateowner
, &stp
, NULL
)))
3077 sop
= oc
->oc_stateowner
;
3078 sop
->so_confirmed
= 1;
3079 update_stateid(&stp
->st_stateid
);
3080 memcpy(&oc
->oc_resp_stateid
, &stp
->st_stateid
, sizeof(stateid_t
));
3081 dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3082 "stateid=(%08x/%08x/%08x/%08x)\n", oc
->oc_seqid
,
3083 stp
->st_stateid
.si_boot
,
3084 stp
->st_stateid
.si_stateownerid
,
3085 stp
->st_stateid
.si_fileid
,
3086 stp
->st_stateid
.si_generation
);
3088 nfsd4_create_clid_dir(sop
->so_client
);
3090 if (oc
->oc_stateowner
) {
3091 nfs4_get_stateowner(oc
->oc_stateowner
);
3092 cstate
->replay_owner
= oc
->oc_stateowner
;
3094 nfs4_unlock_state();
3100 * unset all bits in union bitmap (bmap) that
3101 * do not exist in share (from successful OPEN_DOWNGRADE)
3104 reset_union_bmap_access(unsigned long access
, unsigned long *bmap
)
3107 for (i
= 1; i
< 4; i
++) {
3108 if ((i
& access
) != i
)
3109 __clear_bit(i
, bmap
);
3114 reset_union_bmap_deny(unsigned long deny
, unsigned long *bmap
)
3117 for (i
= 0; i
< 4; i
++) {
3118 if ((i
& deny
) != i
)
3119 __clear_bit(i
, bmap
);
3124 nfsd4_open_downgrade(struct svc_rqst
*rqstp
,
3125 struct nfsd4_compound_state
*cstate
,
3126 struct nfsd4_open_downgrade
*od
)
3129 struct nfs4_stateid
*stp
;
3130 unsigned int share_access
;
3132 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3133 (int)cstate
->current_fh
.fh_dentry
->d_name
.len
,
3134 cstate
->current_fh
.fh_dentry
->d_name
.name
);
3136 if (!access_valid(od
->od_share_access
, cstate
->minorversion
)
3137 || !deny_valid(od
->od_share_deny
))
3138 return nfserr_inval
;
3141 if ((status
= nfs4_preprocess_seqid_op(cstate
,
3145 &od
->od_stateowner
, &stp
, NULL
)))
3148 status
= nfserr_inval
;
3149 if (!test_bit(od
->od_share_access
, &stp
->st_access_bmap
)) {
3150 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3151 stp
->st_access_bmap
, od
->od_share_access
);
3154 if (!test_bit(od
->od_share_deny
, &stp
->st_deny_bmap
)) {
3155 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3156 stp
->st_deny_bmap
, od
->od_share_deny
);
3159 set_access(&share_access
, stp
->st_access_bmap
);
3160 nfs4_file_downgrade(stp
->st_vfs_file
,
3161 share_access
& ~od
->od_share_access
);
3163 reset_union_bmap_access(od
->od_share_access
, &stp
->st_access_bmap
);
3164 reset_union_bmap_deny(od
->od_share_deny
, &stp
->st_deny_bmap
);
3166 update_stateid(&stp
->st_stateid
);
3167 memcpy(&od
->od_stateid
, &stp
->st_stateid
, sizeof(stateid_t
));
3170 if (od
->od_stateowner
) {
3171 nfs4_get_stateowner(od
->od_stateowner
);
3172 cstate
->replay_owner
= od
->od_stateowner
;
3174 nfs4_unlock_state();
3179 * nfs4_unlock_state() called after encode
3182 nfsd4_close(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
3183 struct nfsd4_close
*close
)
3186 struct nfs4_stateid
*stp
;
3188 dprintk("NFSD: nfsd4_close on file %.*s\n",
3189 (int)cstate
->current_fh
.fh_dentry
->d_name
.len
,
3190 cstate
->current_fh
.fh_dentry
->d_name
.name
);
3193 /* check close_lru for replay */
3194 if ((status
= nfs4_preprocess_seqid_op(cstate
,
3197 OPEN_STATE
| CLOSE_STATE
,
3198 &close
->cl_stateowner
, &stp
, NULL
)))
3201 update_stateid(&stp
->st_stateid
);
3202 memcpy(&close
->cl_stateid
, &stp
->st_stateid
, sizeof(stateid_t
));
3204 /* release_stateid() calls nfsd_close() if needed */
3205 release_open_stateid(stp
);
3207 /* place unused nfs4_stateowners on so_close_lru list to be
3208 * released by the laundromat service after the lease period
3209 * to enable us to handle CLOSE replay
3211 if (list_empty(&close
->cl_stateowner
->so_stateids
))
3212 move_to_close_lru(close
->cl_stateowner
);
3214 if (close
->cl_stateowner
) {
3215 nfs4_get_stateowner(close
->cl_stateowner
);
3216 cstate
->replay_owner
= close
->cl_stateowner
;
3218 nfs4_unlock_state();
3223 nfsd4_delegreturn(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
3224 struct nfsd4_delegreturn
*dr
)
3226 struct nfs4_delegation
*dp
;
3227 stateid_t
*stateid
= &dr
->dr_stateid
;
3228 struct inode
*inode
;
3232 if ((status
= fh_verify(rqstp
, &cstate
->current_fh
, S_IFREG
, 0)))
3234 inode
= cstate
->current_fh
.fh_dentry
->d_inode
;
3236 if (nfsd4_has_session(cstate
))
3237 flags
|= HAS_SESSION
;
3239 status
= nfserr_bad_stateid
;
3240 if (ZERO_STATEID(stateid
) || ONE_STATEID(stateid
))
3242 status
= nfserr_stale_stateid
;
3243 if (STALE_STATEID(stateid
))
3245 status
= nfserr_bad_stateid
;
3246 if (!is_delegation_stateid(stateid
))
3248 dp
= find_delegation_stateid(inode
, stateid
);
3250 status
= stateid_error_map(stateid
);
3253 status
= check_stateid_generation(stateid
, &dp
->dl_stateid
, flags
);
3256 renew_client(dp
->dl_client
);
3258 unhash_delegation(dp
);
3260 nfs4_unlock_state();
3267 * Lock owner state (byte-range locks)
3269 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3270 #define LOCK_HASH_BITS 8
3271 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3272 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3275 end_offset(u64 start
, u64 len
)
3280 return end
>= start
? end
: NFS4_MAX_UINT64
;
3283 /* last octet in a range */
3285 last_byte_offset(u64 start
, u64 len
)
3291 return end
> start
? end
- 1: NFS4_MAX_UINT64
;
3294 #define lockownerid_hashval(id) \
3295 ((id) & LOCK_HASH_MASK)
3297 static inline unsigned int
3298 lock_ownerstr_hashval(struct inode
*inode
, u32 cl_id
,
3299 struct xdr_netobj
*ownername
)
3301 return (file_hashval(inode
) + cl_id
3302 + opaque_hashval(ownername
->data
, ownername
->len
))
3306 static struct list_head lock_ownerid_hashtbl
[LOCK_HASH_SIZE
];
3307 static struct list_head lock_ownerstr_hashtbl
[LOCK_HASH_SIZE
];
3308 static struct list_head lockstateid_hashtbl
[STATEID_HASH_SIZE
];
3310 static struct nfs4_stateid
*
3311 find_stateid(stateid_t
*stid
, int flags
)
3313 struct nfs4_stateid
*local
;
3314 u32 st_id
= stid
->si_stateownerid
;
3315 u32 f_id
= stid
->si_fileid
;
3316 unsigned int hashval
;
3318 dprintk("NFSD: find_stateid flags 0x%x\n",flags
);
3319 if (flags
& (LOCK_STATE
| RD_STATE
| WR_STATE
)) {
3320 hashval
= stateid_hashval(st_id
, f_id
);
3321 list_for_each_entry(local
, &lockstateid_hashtbl
[hashval
], st_hash
) {
3322 if ((local
->st_stateid
.si_stateownerid
== st_id
) &&
3323 (local
->st_stateid
.si_fileid
== f_id
))
3328 if (flags
& (OPEN_STATE
| RD_STATE
| WR_STATE
)) {
3329 hashval
= stateid_hashval(st_id
, f_id
);
3330 list_for_each_entry(local
, &stateid_hashtbl
[hashval
], st_hash
) {
3331 if ((local
->st_stateid
.si_stateownerid
== st_id
) &&
3332 (local
->st_stateid
.si_fileid
== f_id
))
3339 static struct nfs4_delegation
*
3340 find_delegation_stateid(struct inode
*ino
, stateid_t
*stid
)
3342 struct nfs4_file
*fp
;
3343 struct nfs4_delegation
*dl
;
3345 dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3346 stid
->si_boot
, stid
->si_stateownerid
,
3347 stid
->si_fileid
, stid
->si_generation
);
3349 fp
= find_file(ino
);
3352 dl
= find_delegation_file(fp
, stid
);
3358 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3359 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3360 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3361 * locking, this prevents us from being completely protocol-compliant. The
3362 * real solution to this problem is to start using unsigned file offsets in
3363 * the VFS, but this is a very deep change!
3366 nfs4_transform_lock_offset(struct file_lock
*lock
)
3368 if (lock
->fl_start
< 0)
3369 lock
->fl_start
= OFFSET_MAX
;
3370 if (lock
->fl_end
< 0)
3371 lock
->fl_end
= OFFSET_MAX
;
3374 /* Hack!: For now, we're defining this just so we can use a pointer to it
3375 * as a unique cookie to identify our (NFSv4's) posix locks. */
3376 static struct lock_manager_operations nfsd_posix_mng_ops
= {
3380 nfs4_set_lock_denied(struct file_lock
*fl
, struct nfsd4_lock_denied
*deny
)
3382 struct nfs4_stateowner
*sop
;
3385 if (fl
->fl_lmops
== &nfsd_posix_mng_ops
) {
3386 sop
= (struct nfs4_stateowner
*) fl
->fl_owner
;
3387 hval
= lockownerid_hashval(sop
->so_id
);
3388 kref_get(&sop
->so_ref
);
3390 deny
->ld_clientid
= sop
->so_client
->cl_clientid
;
3392 deny
->ld_sop
= NULL
;
3393 deny
->ld_clientid
.cl_boot
= 0;
3394 deny
->ld_clientid
.cl_id
= 0;
3396 deny
->ld_start
= fl
->fl_start
;
3397 deny
->ld_length
= NFS4_MAX_UINT64
;
3398 if (fl
->fl_end
!= NFS4_MAX_UINT64
)
3399 deny
->ld_length
= fl
->fl_end
- fl
->fl_start
+ 1;
3400 deny
->ld_type
= NFS4_READ_LT
;
3401 if (fl
->fl_type
!= F_RDLCK
)
3402 deny
->ld_type
= NFS4_WRITE_LT
;
3405 static struct nfs4_stateowner
*
3406 find_lockstateowner_str(struct inode
*inode
, clientid_t
*clid
,
3407 struct xdr_netobj
*owner
)
3409 unsigned int hashval
= lock_ownerstr_hashval(inode
, clid
->cl_id
, owner
);
3410 struct nfs4_stateowner
*op
;
3412 list_for_each_entry(op
, &lock_ownerstr_hashtbl
[hashval
], so_strhash
) {
3413 if (same_owner_str(op
, owner
, clid
))
3420 * Alloc a lock owner structure.
3421 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3424 * strhashval = lock_ownerstr_hashval
3427 static struct nfs4_stateowner
*
3428 alloc_init_lock_stateowner(unsigned int strhashval
, struct nfs4_client
*clp
, struct nfs4_stateid
*open_stp
, struct nfsd4_lock
*lock
) {
3429 struct nfs4_stateowner
*sop
;
3430 struct nfs4_replay
*rp
;
3431 unsigned int idhashval
;
3433 if (!(sop
= alloc_stateowner(&lock
->lk_new_owner
)))
3435 idhashval
= lockownerid_hashval(current_ownerid
);
3436 INIT_LIST_HEAD(&sop
->so_idhash
);
3437 INIT_LIST_HEAD(&sop
->so_strhash
);
3438 INIT_LIST_HEAD(&sop
->so_perclient
);
3439 INIT_LIST_HEAD(&sop
->so_stateids
);
3440 INIT_LIST_HEAD(&sop
->so_perstateid
);
3441 INIT_LIST_HEAD(&sop
->so_close_lru
); /* not used */
3443 list_add(&sop
->so_idhash
, &lock_ownerid_hashtbl
[idhashval
]);
3444 list_add(&sop
->so_strhash
, &lock_ownerstr_hashtbl
[strhashval
]);
3445 list_add(&sop
->so_perstateid
, &open_stp
->st_lockowners
);
3446 sop
->so_is_open_owner
= 0;
3447 sop
->so_id
= current_ownerid
++;
3448 sop
->so_client
= clp
;
3449 /* It is the openowner seqid that will be incremented in encode in the
3450 * case of new lockowners; so increment the lock seqid manually: */
3451 sop
->so_seqid
= lock
->lk_new_lock_seqid
+ 1;
3452 sop
->so_confirmed
= 1;
3453 rp
= &sop
->so_replay
;
3454 rp
->rp_status
= nfserr_serverfault
;
3456 rp
->rp_buf
= rp
->rp_ibuf
;
3460 static struct nfs4_stateid
*
3461 alloc_init_lock_stateid(struct nfs4_stateowner
*sop
, struct nfs4_file
*fp
, struct nfs4_stateid
*open_stp
)
3463 struct nfs4_stateid
*stp
;
3464 unsigned int hashval
= stateid_hashval(sop
->so_id
, fp
->fi_id
);
3466 stp
= nfs4_alloc_stateid();
3469 INIT_LIST_HEAD(&stp
->st_hash
);
3470 INIT_LIST_HEAD(&stp
->st_perfile
);
3471 INIT_LIST_HEAD(&stp
->st_perstateowner
);
3472 INIT_LIST_HEAD(&stp
->st_lockowners
); /* not used */
3473 list_add(&stp
->st_hash
, &lockstateid_hashtbl
[hashval
]);
3474 list_add(&stp
->st_perfile
, &fp
->fi_stateids
);
3475 list_add(&stp
->st_perstateowner
, &sop
->so_stateids
);
3476 stp
->st_stateowner
= sop
;
3479 stp
->st_stateid
.si_boot
= get_seconds();
3480 stp
->st_stateid
.si_stateownerid
= sop
->so_id
;
3481 stp
->st_stateid
.si_fileid
= fp
->fi_id
;
3482 stp
->st_stateid
.si_generation
= 0;
3483 stp
->st_vfs_file
= open_stp
->st_vfs_file
; /* FIXME refcount?? */
3484 stp
->st_access_bmap
= open_stp
->st_access_bmap
;
3485 stp
->st_deny_bmap
= open_stp
->st_deny_bmap
;
3486 stp
->st_openstp
= open_stp
;
3493 check_lock_length(u64 offset
, u64 length
)
3495 return ((length
== 0) || ((length
!= NFS4_MAX_UINT64
) &&
3496 LOFF_OVERFLOW(offset
, length
)));
3503 nfsd4_lock(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
3504 struct nfsd4_lock
*lock
)
3506 struct nfs4_stateowner
*open_sop
= NULL
;
3507 struct nfs4_stateowner
*lock_sop
= NULL
;
3508 struct nfs4_stateid
*lock_stp
;
3510 struct file_lock file_lock
;
3511 struct file_lock conflock
;
3513 unsigned int strhashval
;
3517 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3518 (long long) lock
->lk_offset
,
3519 (long long) lock
->lk_length
);
3521 if (check_lock_length(lock
->lk_offset
, lock
->lk_length
))
3522 return nfserr_inval
;
3524 if ((status
= fh_verify(rqstp
, &cstate
->current_fh
,
3525 S_IFREG
, NFSD_MAY_LOCK
))) {
3526 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3532 if (lock
->lk_is_new
) {
3534 * Client indicates that this is a new lockowner.
3535 * Use open owner and open stateid to create lock owner and
3538 struct nfs4_stateid
*open_stp
= NULL
;
3539 struct nfs4_file
*fp
;
3541 status
= nfserr_stale_clientid
;
3542 if (!nfsd4_has_session(cstate
) &&
3543 STALE_CLIENTID(&lock
->lk_new_clientid
))
3546 /* validate and update open stateid and open seqid */
3547 status
= nfs4_preprocess_seqid_op(cstate
,
3548 lock
->lk_new_open_seqid
,
3549 &lock
->lk_new_open_stateid
,
3551 &lock
->lk_replay_owner
, &open_stp
,
3555 open_sop
= lock
->lk_replay_owner
;
3556 /* create lockowner and lock stateid */
3557 fp
= open_stp
->st_file
;
3558 strhashval
= lock_ownerstr_hashval(fp
->fi_inode
,
3559 open_sop
->so_client
->cl_clientid
.cl_id
,
3560 &lock
->v
.new.owner
);
3561 /* XXX: Do we need to check for duplicate stateowners on
3562 * the same file, or should they just be allowed (and
3563 * create new stateids)? */
3564 status
= nfserr_resource
;
3565 lock_sop
= alloc_init_lock_stateowner(strhashval
,
3566 open_sop
->so_client
, open_stp
, lock
);
3567 if (lock_sop
== NULL
)
3569 lock_stp
= alloc_init_lock_stateid(lock_sop
, fp
, open_stp
);
3570 if (lock_stp
== NULL
)
3573 /* lock (lock owner + lock stateid) already exists */
3574 status
= nfs4_preprocess_seqid_op(cstate
,
3575 lock
->lk_old_lock_seqid
,
3576 &lock
->lk_old_lock_stateid
,
3578 &lock
->lk_replay_owner
, &lock_stp
, lock
);
3581 lock_sop
= lock
->lk_replay_owner
;
3583 /* lock->lk_replay_owner and lock_stp have been created or found */
3584 filp
= lock_stp
->st_vfs_file
;
3586 status
= nfserr_grace
;
3587 if (locks_in_grace() && !lock
->lk_reclaim
)
3589 status
= nfserr_no_grace
;
3590 if (!locks_in_grace() && lock
->lk_reclaim
)
3593 locks_init_lock(&file_lock
);
3594 switch (lock
->lk_type
) {
3597 file_lock
.fl_type
= F_RDLCK
;
3601 case NFS4_WRITEW_LT
:
3602 file_lock
.fl_type
= F_WRLCK
;
3606 status
= nfserr_inval
;
3609 file_lock
.fl_owner
= (fl_owner_t
)lock_sop
;
3610 file_lock
.fl_pid
= current
->tgid
;
3611 file_lock
.fl_file
= filp
;
3612 file_lock
.fl_flags
= FL_POSIX
;
3613 file_lock
.fl_lmops
= &nfsd_posix_mng_ops
;
3615 file_lock
.fl_start
= lock
->lk_offset
;
3616 file_lock
.fl_end
= last_byte_offset(lock
->lk_offset
, lock
->lk_length
);
3617 nfs4_transform_lock_offset(&file_lock
);
3620 * Try to lock the file in the VFS.
3621 * Note: locks.c uses the BKL to protect the inode's lock list.
3624 err
= vfs_lock_file(filp
, cmd
, &file_lock
, &conflock
);
3626 case 0: /* success! */
3627 update_stateid(&lock_stp
->st_stateid
);
3628 memcpy(&lock
->lk_resp_stateid
, &lock_stp
->st_stateid
,
3632 case (EAGAIN
): /* conflock holds conflicting lock */
3633 status
= nfserr_denied
;
3634 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3635 nfs4_set_lock_denied(&conflock
, &lock
->lk_denied
);
3638 status
= nfserr_deadlock
;
3641 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err
);
3642 status
= nfserr_resource
;
3646 if (status
&& lock
->lk_is_new
&& lock_sop
)
3647 release_lockowner(lock_sop
);
3648 if (lock
->lk_replay_owner
) {
3649 nfs4_get_stateowner(lock
->lk_replay_owner
);
3650 cstate
->replay_owner
= lock
->lk_replay_owner
;
3652 nfs4_unlock_state();
3657 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3658 * so we do a temporary open here just to get an open file to pass to
3659 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
3662 static int nfsd_test_lock(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file_lock
*lock
)
3667 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
3670 err
= vfs_test_lock(file
, lock
);
3679 nfsd4_lockt(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
3680 struct nfsd4_lockt
*lockt
)
3682 struct inode
*inode
;
3683 struct file_lock file_lock
;
3687 if (locks_in_grace())
3688 return nfserr_grace
;
3690 if (check_lock_length(lockt
->lt_offset
, lockt
->lt_length
))
3691 return nfserr_inval
;
3693 lockt
->lt_stateowner
= NULL
;
3696 status
= nfserr_stale_clientid
;
3697 if (!nfsd4_has_session(cstate
) && STALE_CLIENTID(&lockt
->lt_clientid
))
3700 if ((status
= fh_verify(rqstp
, &cstate
->current_fh
, S_IFREG
, 0))) {
3701 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3702 if (status
== nfserr_symlink
)
3703 status
= nfserr_inval
;
3707 inode
= cstate
->current_fh
.fh_dentry
->d_inode
;
3708 locks_init_lock(&file_lock
);
3709 switch (lockt
->lt_type
) {
3712 file_lock
.fl_type
= F_RDLCK
;
3715 case NFS4_WRITEW_LT
:
3716 file_lock
.fl_type
= F_WRLCK
;
3719 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3720 status
= nfserr_inval
;
3724 lockt
->lt_stateowner
= find_lockstateowner_str(inode
,
3725 &lockt
->lt_clientid
, &lockt
->lt_owner
);
3726 if (lockt
->lt_stateowner
)
3727 file_lock
.fl_owner
= (fl_owner_t
)lockt
->lt_stateowner
;
3728 file_lock
.fl_pid
= current
->tgid
;
3729 file_lock
.fl_flags
= FL_POSIX
;
3731 file_lock
.fl_start
= lockt
->lt_offset
;
3732 file_lock
.fl_end
= last_byte_offset(lockt
->lt_offset
, lockt
->lt_length
);
3734 nfs4_transform_lock_offset(&file_lock
);
3737 error
= nfsd_test_lock(rqstp
, &cstate
->current_fh
, &file_lock
);
3739 status
= nfserrno(error
);
3742 if (file_lock
.fl_type
!= F_UNLCK
) {
3743 status
= nfserr_denied
;
3744 nfs4_set_lock_denied(&file_lock
, &lockt
->lt_denied
);
3747 nfs4_unlock_state();
3752 nfsd4_locku(struct svc_rqst
*rqstp
, struct nfsd4_compound_state
*cstate
,
3753 struct nfsd4_locku
*locku
)
3755 struct nfs4_stateid
*stp
;
3756 struct file
*filp
= NULL
;
3757 struct file_lock file_lock
;
3761 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3762 (long long) locku
->lu_offset
,
3763 (long long) locku
->lu_length
);
3765 if (check_lock_length(locku
->lu_offset
, locku
->lu_length
))
3766 return nfserr_inval
;
3770 if ((status
= nfs4_preprocess_seqid_op(cstate
,
3774 &locku
->lu_stateowner
, &stp
, NULL
)))
3777 filp
= stp
->st_vfs_file
;
3779 locks_init_lock(&file_lock
);
3780 file_lock
.fl_type
= F_UNLCK
;
3781 file_lock
.fl_owner
= (fl_owner_t
) locku
->lu_stateowner
;
3782 file_lock
.fl_pid
= current
->tgid
;
3783 file_lock
.fl_file
= filp
;
3784 file_lock
.fl_flags
= FL_POSIX
;
3785 file_lock
.fl_lmops
= &nfsd_posix_mng_ops
;
3786 file_lock
.fl_start
= locku
->lu_offset
;
3788 file_lock
.fl_end
= last_byte_offset(locku
->lu_offset
, locku
->lu_length
);
3789 nfs4_transform_lock_offset(&file_lock
);
3792 * Try to unlock the file in the VFS.
3794 err
= vfs_lock_file(filp
, F_SETLK
, &file_lock
, NULL
);
3796 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3800 * OK, unlock succeeded; the only thing left to do is update the stateid.
3802 update_stateid(&stp
->st_stateid
);
3803 memcpy(&locku
->lu_stateid
, &stp
->st_stateid
, sizeof(stateid_t
));
3806 if (locku
->lu_stateowner
) {
3807 nfs4_get_stateowner(locku
->lu_stateowner
);
3808 cstate
->replay_owner
= locku
->lu_stateowner
;
3810 nfs4_unlock_state();
3814 status
= nfserrno(err
);
3820 * 1: locks held by lockowner
3821 * 0: no locks held by lockowner
3824 check_for_locks(struct file
*filp
, struct nfs4_stateowner
*lowner
)
3826 struct file_lock
**flpp
;
3827 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
3831 for (flpp
= &inode
->i_flock
; *flpp
!= NULL
; flpp
= &(*flpp
)->fl_next
) {
3832 if ((*flpp
)->fl_owner
== (fl_owner_t
)lowner
) {
3843 nfsd4_release_lockowner(struct svc_rqst
*rqstp
,
3844 struct nfsd4_compound_state
*cstate
,
3845 struct nfsd4_release_lockowner
*rlockowner
)
3847 clientid_t
*clid
= &rlockowner
->rl_clientid
;
3848 struct nfs4_stateowner
*sop
;
3849 struct nfs4_stateid
*stp
;
3850 struct xdr_netobj
*owner
= &rlockowner
->rl_owner
;
3851 struct list_head matches
;
3855 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3856 clid
->cl_boot
, clid
->cl_id
);
3858 /* XXX check for lease expiration */
3860 status
= nfserr_stale_clientid
;
3861 if (STALE_CLIENTID(clid
))
3866 status
= nfserr_locks_held
;
3867 /* XXX: we're doing a linear search through all the lockowners.
3868 * Yipes! For now we'll just hope clients aren't really using
3869 * release_lockowner much, but eventually we have to fix these
3870 * data structures. */
3871 INIT_LIST_HEAD(&matches
);
3872 for (i
= 0; i
< LOCK_HASH_SIZE
; i
++) {
3873 list_for_each_entry(sop
, &lock_ownerid_hashtbl
[i
], so_idhash
) {
3874 if (!same_owner_str(sop
, owner
, clid
))
3876 list_for_each_entry(stp
, &sop
->so_stateids
,
3878 if (check_for_locks(stp
->st_vfs_file
, sop
))
3880 /* Note: so_perclient unused for lockowners,
3881 * so it's OK to fool with here. */
3882 list_add(&sop
->so_perclient
, &matches
);
3886 /* Clients probably won't expect us to return with some (but not all)
3887 * of the lockowner state released; so don't release any until all
3888 * have been checked. */
3890 while (!list_empty(&matches
)) {
3891 sop
= list_entry(matches
.next
, struct nfs4_stateowner
,
3893 /* unhash_stateowner deletes so_perclient only
3894 * for openowners. */
3895 list_del(&sop
->so_perclient
);
3896 release_lockowner(sop
);
3899 nfs4_unlock_state();
3903 static inline struct nfs4_client_reclaim
*
3906 return kmalloc(sizeof(struct nfs4_client_reclaim
), GFP_KERNEL
);
3910 nfs4_has_reclaimed_state(const char *name
, bool use_exchange_id
)
3912 unsigned int strhashval
= clientstr_hashval(name
);
3913 struct nfs4_client
*clp
;
3915 clp
= find_confirmed_client_by_str(name
, strhashval
, use_exchange_id
);
3920 * failure => all reset bets are off, nfserr_no_grace...
3923 nfs4_client_to_reclaim(const char *name
)
3925 unsigned int strhashval
;
3926 struct nfs4_client_reclaim
*crp
= NULL
;
3928 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN
, name
);
3929 crp
= alloc_reclaim();
3932 strhashval
= clientstr_hashval(name
);
3933 INIT_LIST_HEAD(&crp
->cr_strhash
);
3934 list_add(&crp
->cr_strhash
, &reclaim_str_hashtbl
[strhashval
]);
3935 memcpy(crp
->cr_recdir
, name
, HEXDIR_LEN
);
3936 reclaim_str_hashtbl_size
++;
3941 nfs4_release_reclaim(void)
3943 struct nfs4_client_reclaim
*crp
= NULL
;
3946 for (i
= 0; i
< CLIENT_HASH_SIZE
; i
++) {
3947 while (!list_empty(&reclaim_str_hashtbl
[i
])) {
3948 crp
= list_entry(reclaim_str_hashtbl
[i
].next
,
3949 struct nfs4_client_reclaim
, cr_strhash
);
3950 list_del(&crp
->cr_strhash
);
3952 reclaim_str_hashtbl_size
--;
3955 BUG_ON(reclaim_str_hashtbl_size
);
3959 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3960 static struct nfs4_client_reclaim
*
3961 nfs4_find_reclaim_client(clientid_t
*clid
)
3963 unsigned int strhashval
;
3964 struct nfs4_client
*clp
;
3965 struct nfs4_client_reclaim
*crp
= NULL
;
3968 /* find clientid in conf_id_hashtbl */
3969 clp
= find_confirmed_client(clid
);
3973 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3974 clp
->cl_name
.len
, clp
->cl_name
.data
,
3977 /* find clp->cl_name in reclaim_str_hashtbl */
3978 strhashval
= clientstr_hashval(clp
->cl_recdir
);
3979 list_for_each_entry(crp
, &reclaim_str_hashtbl
[strhashval
], cr_strhash
) {
3980 if (same_name(crp
->cr_recdir
, clp
->cl_recdir
)) {
3988 * Called from OPEN. Look for clientid in reclaim list.
3991 nfs4_check_open_reclaim(clientid_t
*clid
)
3993 return nfs4_find_reclaim_client(clid
) ? nfs_ok
: nfserr_reclaim_bad
;
3996 /* initialization to perform at module load time: */
3999 nfs4_state_init(void)
4003 status
= nfsd4_init_slabs();
4006 for (i
= 0; i
< CLIENT_HASH_SIZE
; i
++) {
4007 INIT_LIST_HEAD(&conf_id_hashtbl
[i
]);
4008 INIT_LIST_HEAD(&conf_str_hashtbl
[i
]);
4009 INIT_LIST_HEAD(&unconf_str_hashtbl
[i
]);
4010 INIT_LIST_HEAD(&unconf_id_hashtbl
[i
]);
4011 INIT_LIST_HEAD(&reclaim_str_hashtbl
[i
]);
4013 for (i
= 0; i
< SESSION_HASH_SIZE
; i
++)
4014 INIT_LIST_HEAD(&sessionid_hashtbl
[i
]);
4015 for (i
= 0; i
< FILE_HASH_SIZE
; i
++) {
4016 INIT_LIST_HEAD(&file_hashtbl
[i
]);
4018 for (i
= 0; i
< OWNER_HASH_SIZE
; i
++) {
4019 INIT_LIST_HEAD(&ownerstr_hashtbl
[i
]);
4020 INIT_LIST_HEAD(&ownerid_hashtbl
[i
]);
4022 for (i
= 0; i
< STATEID_HASH_SIZE
; i
++) {
4023 INIT_LIST_HEAD(&stateid_hashtbl
[i
]);
4024 INIT_LIST_HEAD(&lockstateid_hashtbl
[i
]);
4026 for (i
= 0; i
< LOCK_HASH_SIZE
; i
++) {
4027 INIT_LIST_HEAD(&lock_ownerid_hashtbl
[i
]);
4028 INIT_LIST_HEAD(&lock_ownerstr_hashtbl
[i
]);
4030 memset(&onestateid
, ~0, sizeof(stateid_t
));
4031 INIT_LIST_HEAD(&close_lru
);
4032 INIT_LIST_HEAD(&client_lru
);
4033 INIT_LIST_HEAD(&del_recall_lru
);
4034 reclaim_str_hashtbl_size
= 0;
4039 nfsd4_load_reboot_recovery_data(void)
4044 nfsd4_init_recdir(user_recovery_dirname
);
4045 status
= nfsd4_recdir_load();
4046 nfs4_unlock_state();
4048 printk("NFSD: Failure reading reboot recovery data\n");
4052 get_nfs4_grace_period(void)
4054 return max(user_lease_time
, lease_time
) * HZ
;
4058 * Since the lifetime of a delegation isn't limited to that of an open, a
4059 * client may quite reasonably hang on to a delegation as long as it has
4060 * the inode cached. This becomes an obvious problem the first time a
4061 * client's inode cache approaches the size of the server's total memory.
4063 * For now we avoid this problem by imposing a hard limit on the number
4064 * of delegations, which varies according to the server's memory size.
4067 set_max_delegations(void)
4070 * Allow at most 4 delegations per megabyte of RAM. Quick
4071 * estimates suggest that in the worst case (where every delegation
4072 * is for a different inode), a delegation could take about 1.5K,
4073 * giving a worst case usage of about 6% of memory.
4075 max_delegations
= nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT
);
4078 /* initialization to perform when the nfsd service is started: */
4081 __nfs4_state_start(void)
4083 unsigned long grace_time
;
4085 boot_time
= get_seconds();
4086 grace_time
= get_nfs4_grace_period();
4087 lease_time
= user_lease_time
;
4088 locks_start_grace(&nfsd4_manager
);
4089 printk(KERN_INFO
"NFSD: starting %ld-second grace period\n",
4091 laundry_wq
= create_singlethread_workqueue("nfsd4");
4092 queue_delayed_work(laundry_wq
, &laundromat_work
, grace_time
);
4093 set_max_delegations();
4097 nfs4_state_start(void)
4101 nfsd4_load_reboot_recovery_data();
4102 __nfs4_state_start();
4108 nfs4_lease_time(void)
4114 __nfs4_state_shutdown(void)
4117 struct nfs4_client
*clp
= NULL
;
4118 struct nfs4_delegation
*dp
= NULL
;
4119 struct list_head
*pos
, *next
, reaplist
;
4121 for (i
= 0; i
< CLIENT_HASH_SIZE
; i
++) {
4122 while (!list_empty(&conf_id_hashtbl
[i
])) {
4123 clp
= list_entry(conf_id_hashtbl
[i
].next
, struct nfs4_client
, cl_idhash
);
4126 while (!list_empty(&unconf_str_hashtbl
[i
])) {
4127 clp
= list_entry(unconf_str_hashtbl
[i
].next
, struct nfs4_client
, cl_strhash
);
4131 INIT_LIST_HEAD(&reaplist
);
4132 spin_lock(&recall_lock
);
4133 list_for_each_safe(pos
, next
, &del_recall_lru
) {
4134 dp
= list_entry (pos
, struct nfs4_delegation
, dl_recall_lru
);
4135 list_move(&dp
->dl_recall_lru
, &reaplist
);
4137 spin_unlock(&recall_lock
);
4138 list_for_each_safe(pos
, next
, &reaplist
) {
4139 dp
= list_entry (pos
, struct nfs4_delegation
, dl_recall_lru
);
4140 list_del_init(&dp
->dl_recall_lru
);
4141 unhash_delegation(dp
);
4144 nfsd4_shutdown_recdir();
4149 nfs4_state_shutdown(void)
4151 cancel_rearming_delayed_workqueue(laundry_wq
, &laundromat_work
);
4152 destroy_workqueue(laundry_wq
);
4153 locks_end_grace(&nfsd4_manager
);
4155 nfs4_release_reclaim();
4156 __nfs4_state_shutdown();
4157 nfs4_unlock_state();
4161 * user_recovery_dirname is protected by the nfsd_mutex since it's only
4162 * accessed when nfsd is starting.
4165 nfs4_set_recdir(char *recdir
)
4167 strcpy(user_recovery_dirname
, recdir
);
4171 * Change the NFSv4 recovery directory to recdir.
4174 nfs4_reset_recoverydir(char *recdir
)
4179 status
= kern_path(recdir
, LOOKUP_FOLLOW
, &path
);
4183 if (S_ISDIR(path
.dentry
->d_inode
->i_mode
)) {
4184 nfs4_set_recdir(recdir
);
4192 nfs4_recoverydir(void)
4194 return user_recovery_dirname
;
4198 * Called when leasetime is changed.
4200 * The only way the protocol gives us to handle on-the-fly lease changes is to
4201 * simulate a reboot. Instead of doing that, we just wait till the next time
4202 * we start to register any changes in lease time. If the administrator
4203 * really wants to change the lease time *now*, they can go ahead and bring
4204 * nfsd down and then back up again after changing the lease time.
4206 * user_lease_time is protected by nfsd_mutex since it's only really accessed
4207 * when nfsd is starting
4210 nfs4_reset_lease(time_t leasetime
)
4212 user_lease_time
= leasetime
;