Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / fs / nfs / nfs4state.c
blobe5be72518bd7ebe813522b1c24f7fa16b539421e
1 /*
2 * fs/nfs/nfs4state.c
4 * Client-side XDR for NFSv4.
6 * Copyright (c) 2002 The Regents of the University of Michigan.
7 * All rights reserved.
9 * Kendrick Smith <kmsmith@umich.edu>
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its
21 * contributors may be used to endorse or promote products derived
22 * from this software without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 * Implementation of the NFSv4 state model. For the time being,
37 * this is minimal, but will be made much more complex in a
38 * subsequent patch.
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/ratelimit.h>
50 #include <linux/workqueue.h>
51 #include <linux/bitops.h>
52 #include <linux/jiffies.h>
54 #include <linux/sunrpc/clnt.h>
56 #include "nfs4_fs.h"
57 #include "callback.h"
58 #include "delegation.h"
59 #include "internal.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
64 #define NFSDBG_FACILITY NFSDBG_STATE
66 #define OPENOWNER_POOL_SIZE 8
68 const nfs4_stateid zero_stateid;
69 static DEFINE_MUTEX(nfs_clid_init_mutex);
71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
73 struct nfs4_setclientid_res clid = {
74 .clientid = clp->cl_clientid,
75 .confirm = clp->cl_confirm,
77 unsigned short port;
78 int status;
79 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
81 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
82 goto do_confirm;
83 port = nn->nfs_callback_tcpport;
84 if (clp->cl_addr.ss_family == AF_INET6)
85 port = nn->nfs_callback_tcpport6;
87 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
88 if (status != 0)
89 goto out;
90 clp->cl_clientid = clid.clientid;
91 clp->cl_confirm = clid.confirm;
92 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
93 do_confirm:
94 status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
95 if (status != 0)
96 goto out;
97 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
98 nfs4_schedule_state_renewal(clp);
99 out:
100 return status;
104 * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
106 * @clp: nfs_client under test
107 * @result: OUT: found nfs_client, or clp
108 * @cred: credential to use for trunking test
110 * Returns zero, a negative errno, or a negative NFS4ERR status.
111 * If zero is returned, an nfs_client pointer is planted in
112 * "result".
114 * Note: The returned client may not yet be marked ready.
116 int nfs40_discover_server_trunking(struct nfs_client *clp,
117 struct nfs_client **result,
118 struct rpc_cred *cred)
120 struct nfs4_setclientid_res clid = {
121 .clientid = clp->cl_clientid,
122 .confirm = clp->cl_confirm,
124 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
125 unsigned short port;
126 int status;
128 port = nn->nfs_callback_tcpport;
129 if (clp->cl_addr.ss_family == AF_INET6)
130 port = nn->nfs_callback_tcpport6;
132 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
133 if (status != 0)
134 goto out;
135 clp->cl_clientid = clid.clientid;
136 clp->cl_confirm = clid.confirm;
138 status = nfs40_walk_client_list(clp, result, cred);
139 if (status == 0) {
140 /* Sustain the lease, even if it's empty. If the clientid4
141 * goes stale it's of no use for trunking discovery. */
142 nfs4_schedule_state_renewal(*result);
144 out:
145 return status;
148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
150 struct rpc_cred *cred = NULL;
152 if (clp->cl_machine_cred != NULL)
153 cred = get_rpccred(clp->cl_machine_cred);
154 return cred;
157 static void nfs4_root_machine_cred(struct nfs_client *clp)
159 struct rpc_cred *cred, *new;
161 new = rpc_lookup_machine_cred(NULL);
162 spin_lock(&clp->cl_lock);
163 cred = clp->cl_machine_cred;
164 clp->cl_machine_cred = new;
165 spin_unlock(&clp->cl_lock);
166 if (cred != NULL)
167 put_rpccred(cred);
170 static struct rpc_cred *
171 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
173 struct rpc_cred *cred = NULL;
174 struct nfs4_state_owner *sp;
175 struct rb_node *pos;
177 for (pos = rb_first(&server->state_owners);
178 pos != NULL;
179 pos = rb_next(pos)) {
180 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
181 if (list_empty(&sp->so_states))
182 continue;
183 cred = get_rpccred(sp->so_cred);
184 break;
186 return cred;
190 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
191 * @clp: client state handle
193 * Returns an rpc_cred with reference count bumped, or NULL.
194 * Caller must hold clp->cl_lock.
196 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
198 struct rpc_cred *cred = NULL;
199 struct nfs_server *server;
201 /* Use machine credentials if available */
202 cred = nfs4_get_machine_cred_locked(clp);
203 if (cred != NULL)
204 goto out;
206 rcu_read_lock();
207 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
208 cred = nfs4_get_renew_cred_server_locked(server);
209 if (cred != NULL)
210 break;
212 rcu_read_unlock();
214 out:
215 return cred;
218 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
220 if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
221 spin_lock(&tbl->slot_tbl_lock);
222 nfs41_wake_slot_table(tbl);
223 spin_unlock(&tbl->slot_tbl_lock);
227 static void nfs4_end_drain_session(struct nfs_client *clp)
229 struct nfs4_session *ses = clp->cl_session;
231 if (clp->cl_slot_tbl) {
232 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
233 return;
236 if (ses != NULL) {
237 nfs4_end_drain_slot_table(&ses->bc_slot_table);
238 nfs4_end_drain_slot_table(&ses->fc_slot_table);
242 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
244 set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
245 spin_lock(&tbl->slot_tbl_lock);
246 if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
247 reinit_completion(&tbl->complete);
248 spin_unlock(&tbl->slot_tbl_lock);
249 return wait_for_completion_interruptible(&tbl->complete);
251 spin_unlock(&tbl->slot_tbl_lock);
252 return 0;
255 static int nfs4_begin_drain_session(struct nfs_client *clp)
257 struct nfs4_session *ses = clp->cl_session;
258 int ret = 0;
260 if (clp->cl_slot_tbl)
261 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
263 /* back channel */
264 ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
265 if (ret)
266 return ret;
267 /* fore channel */
268 return nfs4_drain_slot_tbl(&ses->fc_slot_table);
271 #if defined(CONFIG_NFS_V4_1)
273 static int nfs41_setup_state_renewal(struct nfs_client *clp)
275 int status;
276 struct nfs_fsinfo fsinfo;
278 if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
279 nfs4_schedule_state_renewal(clp);
280 return 0;
283 status = nfs4_proc_get_lease_time(clp, &fsinfo);
284 if (status == 0) {
285 /* Update lease time and schedule renewal */
286 spin_lock(&clp->cl_lock);
287 clp->cl_lease_time = fsinfo.lease_time * HZ;
288 clp->cl_last_renewal = jiffies;
289 spin_unlock(&clp->cl_lock);
291 nfs4_schedule_state_renewal(clp);
294 return status;
297 static void nfs41_finish_session_reset(struct nfs_client *clp)
299 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300 clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301 /* create_session negotiated new slot table */
302 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303 nfs41_setup_state_renewal(clp);
306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
308 int status;
310 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311 goto do_confirm;
312 nfs4_begin_drain_session(clp);
313 status = nfs4_proc_exchange_id(clp, cred);
314 if (status != 0)
315 goto out;
316 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
317 do_confirm:
318 status = nfs4_proc_create_session(clp, cred);
319 if (status != 0)
320 goto out;
321 nfs41_finish_session_reset(clp);
322 nfs_mark_client_ready(clp, NFS_CS_READY);
323 out:
324 return status;
328 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
330 * @clp: nfs_client under test
331 * @result: OUT: found nfs_client, or clp
332 * @cred: credential to use for trunking test
334 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
335 * If NFS4_OK is returned, an nfs_client pointer is planted in
336 * "result".
338 * Note: The returned client may not yet be marked ready.
340 int nfs41_discover_server_trunking(struct nfs_client *clp,
341 struct nfs_client **result,
342 struct rpc_cred *cred)
344 int status;
346 status = nfs4_proc_exchange_id(clp, cred);
347 if (status != NFS4_OK)
348 return status;
349 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
351 return nfs41_walk_client_list(clp, result, cred);
354 #endif /* CONFIG_NFS_V4_1 */
357 * nfs4_get_clid_cred - Acquire credential for a setclientid operation
358 * @clp: client state handle
360 * Returns an rpc_cred with reference count bumped, or NULL.
362 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
364 struct rpc_cred *cred;
366 spin_lock(&clp->cl_lock);
367 cred = nfs4_get_machine_cred_locked(clp);
368 spin_unlock(&clp->cl_lock);
369 return cred;
372 static struct nfs4_state_owner *
373 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
375 struct rb_node **p = &server->state_owners.rb_node,
376 *parent = NULL;
377 struct nfs4_state_owner *sp;
379 while (*p != NULL) {
380 parent = *p;
381 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
383 if (cred < sp->so_cred)
384 p = &parent->rb_left;
385 else if (cred > sp->so_cred)
386 p = &parent->rb_right;
387 else {
388 if (!list_empty(&sp->so_lru))
389 list_del_init(&sp->so_lru);
390 atomic_inc(&sp->so_count);
391 return sp;
394 return NULL;
397 static struct nfs4_state_owner *
398 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
400 struct nfs_server *server = new->so_server;
401 struct rb_node **p = &server->state_owners.rb_node,
402 *parent = NULL;
403 struct nfs4_state_owner *sp;
404 int err;
406 while (*p != NULL) {
407 parent = *p;
408 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
410 if (new->so_cred < sp->so_cred)
411 p = &parent->rb_left;
412 else if (new->so_cred > sp->so_cred)
413 p = &parent->rb_right;
414 else {
415 if (!list_empty(&sp->so_lru))
416 list_del_init(&sp->so_lru);
417 atomic_inc(&sp->so_count);
418 return sp;
421 err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
422 if (err)
423 return ERR_PTR(err);
424 rb_link_node(&new->so_server_node, parent, p);
425 rb_insert_color(&new->so_server_node, &server->state_owners);
426 return new;
429 static void
430 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
432 struct nfs_server *server = sp->so_server;
434 if (!RB_EMPTY_NODE(&sp->so_server_node))
435 rb_erase(&sp->so_server_node, &server->state_owners);
436 ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
439 static void
440 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
442 sc->create_time = ktime_get();
443 sc->flags = 0;
444 sc->counter = 0;
445 spin_lock_init(&sc->lock);
446 INIT_LIST_HEAD(&sc->list);
447 rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
450 static void
451 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
453 rpc_destroy_wait_queue(&sc->wait);
457 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
458 * create a new state_owner.
461 static struct nfs4_state_owner *
462 nfs4_alloc_state_owner(struct nfs_server *server,
463 struct rpc_cred *cred,
464 gfp_t gfp_flags)
466 struct nfs4_state_owner *sp;
468 sp = kzalloc(sizeof(*sp), gfp_flags);
469 if (!sp)
470 return NULL;
471 sp->so_server = server;
472 sp->so_cred = get_rpccred(cred);
473 spin_lock_init(&sp->so_lock);
474 INIT_LIST_HEAD(&sp->so_states);
475 nfs4_init_seqid_counter(&sp->so_seqid);
476 atomic_set(&sp->so_count, 1);
477 INIT_LIST_HEAD(&sp->so_lru);
478 seqcount_init(&sp->so_reclaim_seqcount);
479 mutex_init(&sp->so_delegreturn_mutex);
480 return sp;
483 static void
484 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
486 struct rb_node *rb_node = &sp->so_server_node;
488 if (!RB_EMPTY_NODE(rb_node)) {
489 struct nfs_server *server = sp->so_server;
490 struct nfs_client *clp = server->nfs_client;
492 spin_lock(&clp->cl_lock);
493 if (!RB_EMPTY_NODE(rb_node)) {
494 rb_erase(rb_node, &server->state_owners);
495 RB_CLEAR_NODE(rb_node);
497 spin_unlock(&clp->cl_lock);
501 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
503 nfs4_destroy_seqid_counter(&sp->so_seqid);
504 put_rpccred(sp->so_cred);
505 kfree(sp);
508 static void nfs4_gc_state_owners(struct nfs_server *server)
510 struct nfs_client *clp = server->nfs_client;
511 struct nfs4_state_owner *sp, *tmp;
512 unsigned long time_min, time_max;
513 LIST_HEAD(doomed);
515 spin_lock(&clp->cl_lock);
516 time_max = jiffies;
517 time_min = (long)time_max - (long)clp->cl_lease_time;
518 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
519 /* NB: LRU is sorted so that oldest is at the head */
520 if (time_in_range(sp->so_expires, time_min, time_max))
521 break;
522 list_move(&sp->so_lru, &doomed);
523 nfs4_remove_state_owner_locked(sp);
525 spin_unlock(&clp->cl_lock);
527 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
528 list_del(&sp->so_lru);
529 nfs4_free_state_owner(sp);
534 * nfs4_get_state_owner - Look up a state owner given a credential
535 * @server: nfs_server to search
536 * @cred: RPC credential to match
538 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
540 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
541 struct rpc_cred *cred,
542 gfp_t gfp_flags)
544 struct nfs_client *clp = server->nfs_client;
545 struct nfs4_state_owner *sp, *new;
547 spin_lock(&clp->cl_lock);
548 sp = nfs4_find_state_owner_locked(server, cred);
549 spin_unlock(&clp->cl_lock);
550 if (sp != NULL)
551 goto out;
552 new = nfs4_alloc_state_owner(server, cred, gfp_flags);
553 if (new == NULL)
554 goto out;
555 do {
556 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
557 break;
558 spin_lock(&clp->cl_lock);
559 sp = nfs4_insert_state_owner_locked(new);
560 spin_unlock(&clp->cl_lock);
561 } while (sp == ERR_PTR(-EAGAIN));
562 if (sp != new)
563 nfs4_free_state_owner(new);
564 out:
565 nfs4_gc_state_owners(server);
566 return sp;
570 * nfs4_put_state_owner - Release a nfs4_state_owner
571 * @sp: state owner data to release
573 * Note that we keep released state owners on an LRU
574 * list.
575 * This caches valid state owners so that they can be
576 * reused, to avoid the OPEN_CONFIRM on minor version 0.
577 * It also pins the uniquifier of dropped state owners for
578 * a while, to ensure that those state owner names are
579 * never reused.
581 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
583 struct nfs_server *server = sp->so_server;
584 struct nfs_client *clp = server->nfs_client;
586 if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
587 return;
589 sp->so_expires = jiffies;
590 list_add_tail(&sp->so_lru, &server->state_owners_lru);
591 spin_unlock(&clp->cl_lock);
595 * nfs4_purge_state_owners - Release all cached state owners
596 * @server: nfs_server with cached state owners to release
598 * Called at umount time. Remaining state owners will be on
599 * the LRU with ref count of zero.
601 void nfs4_purge_state_owners(struct nfs_server *server)
603 struct nfs_client *clp = server->nfs_client;
604 struct nfs4_state_owner *sp, *tmp;
605 LIST_HEAD(doomed);
607 spin_lock(&clp->cl_lock);
608 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
609 list_move(&sp->so_lru, &doomed);
610 nfs4_remove_state_owner_locked(sp);
612 spin_unlock(&clp->cl_lock);
614 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
615 list_del(&sp->so_lru);
616 nfs4_free_state_owner(sp);
620 static struct nfs4_state *
621 nfs4_alloc_open_state(void)
623 struct nfs4_state *state;
625 state = kzalloc(sizeof(*state), GFP_NOFS);
626 if (!state)
627 return NULL;
628 atomic_set(&state->count, 1);
629 INIT_LIST_HEAD(&state->lock_states);
630 spin_lock_init(&state->state_lock);
631 seqlock_init(&state->seqlock);
632 return state;
635 void
636 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
638 if (state->state == fmode)
639 return;
640 /* NB! List reordering - see the reclaim code for why. */
641 if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
642 if (fmode & FMODE_WRITE)
643 list_move(&state->open_states, &state->owner->so_states);
644 else
645 list_move_tail(&state->open_states, &state->owner->so_states);
647 state->state = fmode;
650 static struct nfs4_state *
651 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
653 struct nfs_inode *nfsi = NFS_I(inode);
654 struct nfs4_state *state;
656 list_for_each_entry(state, &nfsi->open_states, inode_states) {
657 if (state->owner != owner)
658 continue;
659 if (!nfs4_valid_open_stateid(state))
660 continue;
661 if (atomic_inc_not_zero(&state->count))
662 return state;
664 return NULL;
667 static void
668 nfs4_free_open_state(struct nfs4_state *state)
670 kfree(state);
673 struct nfs4_state *
674 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
676 struct nfs4_state *state, *new;
677 struct nfs_inode *nfsi = NFS_I(inode);
679 spin_lock(&inode->i_lock);
680 state = __nfs4_find_state_byowner(inode, owner);
681 spin_unlock(&inode->i_lock);
682 if (state)
683 goto out;
684 new = nfs4_alloc_open_state();
685 spin_lock(&owner->so_lock);
686 spin_lock(&inode->i_lock);
687 state = __nfs4_find_state_byowner(inode, owner);
688 if (state == NULL && new != NULL) {
689 state = new;
690 state->owner = owner;
691 atomic_inc(&owner->so_count);
692 list_add(&state->inode_states, &nfsi->open_states);
693 ihold(inode);
694 state->inode = inode;
695 spin_unlock(&inode->i_lock);
696 /* Note: The reclaim code dictates that we add stateless
697 * and read-only stateids to the end of the list */
698 list_add_tail(&state->open_states, &owner->so_states);
699 spin_unlock(&owner->so_lock);
700 } else {
701 spin_unlock(&inode->i_lock);
702 spin_unlock(&owner->so_lock);
703 if (new)
704 nfs4_free_open_state(new);
706 out:
707 return state;
710 void nfs4_put_open_state(struct nfs4_state *state)
712 struct inode *inode = state->inode;
713 struct nfs4_state_owner *owner = state->owner;
715 if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
716 return;
717 spin_lock(&inode->i_lock);
718 list_del(&state->inode_states);
719 list_del(&state->open_states);
720 spin_unlock(&inode->i_lock);
721 spin_unlock(&owner->so_lock);
722 iput(inode);
723 nfs4_free_open_state(state);
724 nfs4_put_state_owner(owner);
728 * Close the current file.
730 static void __nfs4_close(struct nfs4_state *state,
731 fmode_t fmode, gfp_t gfp_mask, int wait)
733 struct nfs4_state_owner *owner = state->owner;
734 int call_close = 0;
735 fmode_t newstate;
737 atomic_inc(&owner->so_count);
738 /* Protect against nfs4_find_state() */
739 spin_lock(&owner->so_lock);
740 switch (fmode & (FMODE_READ | FMODE_WRITE)) {
741 case FMODE_READ:
742 state->n_rdonly--;
743 break;
744 case FMODE_WRITE:
745 state->n_wronly--;
746 break;
747 case FMODE_READ|FMODE_WRITE:
748 state->n_rdwr--;
750 newstate = FMODE_READ|FMODE_WRITE;
751 if (state->n_rdwr == 0) {
752 if (state->n_rdonly == 0) {
753 newstate &= ~FMODE_READ;
754 call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
755 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
757 if (state->n_wronly == 0) {
758 newstate &= ~FMODE_WRITE;
759 call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
760 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
762 if (newstate == 0)
763 clear_bit(NFS_DELEGATED_STATE, &state->flags);
765 nfs4_state_set_mode_locked(state, newstate);
766 spin_unlock(&owner->so_lock);
768 if (!call_close) {
769 nfs4_put_open_state(state);
770 nfs4_put_state_owner(owner);
771 } else
772 nfs4_do_close(state, gfp_mask, wait);
775 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
777 __nfs4_close(state, fmode, GFP_NOFS, 0);
780 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
782 __nfs4_close(state, fmode, GFP_KERNEL, 1);
786 * Search the state->lock_states for an existing lock_owner
787 * that is compatible with current->files
789 static struct nfs4_lock_state *
790 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
792 struct nfs4_lock_state *pos;
793 list_for_each_entry(pos, &state->lock_states, ls_locks) {
794 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
795 continue;
796 switch (pos->ls_owner.lo_type) {
797 case NFS4_POSIX_LOCK_TYPE:
798 if (pos->ls_owner.lo_u.posix_owner != fl_owner)
799 continue;
800 break;
801 case NFS4_FLOCK_LOCK_TYPE:
802 if (pos->ls_owner.lo_u.flock_owner != fl_pid)
803 continue;
805 atomic_inc(&pos->ls_count);
806 return pos;
808 return NULL;
812 * Return a compatible lock_state. If no initialized lock_state structure
813 * exists, return an uninitialized one.
816 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
818 struct nfs4_lock_state *lsp;
819 struct nfs_server *server = state->owner->so_server;
821 lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
822 if (lsp == NULL)
823 return NULL;
824 nfs4_init_seqid_counter(&lsp->ls_seqid);
825 atomic_set(&lsp->ls_count, 1);
826 lsp->ls_state = state;
827 lsp->ls_owner.lo_type = type;
828 switch (lsp->ls_owner.lo_type) {
829 case NFS4_FLOCK_LOCK_TYPE:
830 lsp->ls_owner.lo_u.flock_owner = fl_pid;
831 break;
832 case NFS4_POSIX_LOCK_TYPE:
833 lsp->ls_owner.lo_u.posix_owner = fl_owner;
834 break;
835 default:
836 goto out_free;
838 lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
839 if (lsp->ls_seqid.owner_id < 0)
840 goto out_free;
841 INIT_LIST_HEAD(&lsp->ls_locks);
842 return lsp;
843 out_free:
844 kfree(lsp);
845 return NULL;
848 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
850 ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
851 nfs4_destroy_seqid_counter(&lsp->ls_seqid);
852 kfree(lsp);
856 * Return a compatible lock_state. If no initialized lock_state structure
857 * exists, return an uninitialized one.
860 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
862 struct nfs4_lock_state *lsp, *new = NULL;
864 for(;;) {
865 spin_lock(&state->state_lock);
866 lsp = __nfs4_find_lock_state(state, owner, pid, type);
867 if (lsp != NULL)
868 break;
869 if (new != NULL) {
870 list_add(&new->ls_locks, &state->lock_states);
871 set_bit(LK_STATE_IN_USE, &state->flags);
872 lsp = new;
873 new = NULL;
874 break;
876 spin_unlock(&state->state_lock);
877 new = nfs4_alloc_lock_state(state, owner, pid, type);
878 if (new == NULL)
879 return NULL;
881 spin_unlock(&state->state_lock);
882 if (new != NULL)
883 nfs4_free_lock_state(state->owner->so_server, new);
884 return lsp;
888 * Release reference to lock_state, and free it if we see that
889 * it is no longer in use
891 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
893 struct nfs_server *server;
894 struct nfs4_state *state;
896 if (lsp == NULL)
897 return;
898 state = lsp->ls_state;
899 if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
900 return;
901 list_del(&lsp->ls_locks);
902 if (list_empty(&state->lock_states))
903 clear_bit(LK_STATE_IN_USE, &state->flags);
904 spin_unlock(&state->state_lock);
905 server = state->owner->so_server;
906 if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
907 struct nfs_client *clp = server->nfs_client;
909 clp->cl_mvops->free_lock_state(server, lsp);
910 } else
911 nfs4_free_lock_state(server, lsp);
914 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
916 struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
918 dst->fl_u.nfs4_fl.owner = lsp;
919 atomic_inc(&lsp->ls_count);
922 static void nfs4_fl_release_lock(struct file_lock *fl)
924 nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
927 static const struct file_lock_operations nfs4_fl_lock_ops = {
928 .fl_copy_lock = nfs4_fl_copy_lock,
929 .fl_release_private = nfs4_fl_release_lock,
932 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
934 struct nfs4_lock_state *lsp;
936 if (fl->fl_ops != NULL)
937 return 0;
938 if (fl->fl_flags & FL_POSIX)
939 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
940 else if (fl->fl_flags & FL_FLOCK)
941 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
942 NFS4_FLOCK_LOCK_TYPE);
943 else
944 return -EINVAL;
945 if (lsp == NULL)
946 return -ENOMEM;
947 fl->fl_u.nfs4_fl.owner = lsp;
948 fl->fl_ops = &nfs4_fl_lock_ops;
949 return 0;
952 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
953 struct nfs4_state *state,
954 const struct nfs_lockowner *lockowner)
956 struct nfs4_lock_state *lsp;
957 fl_owner_t fl_owner;
958 pid_t fl_pid;
959 int ret = -ENOENT;
962 if (lockowner == NULL)
963 goto out;
965 if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
966 goto out;
968 fl_owner = lockowner->l_owner;
969 fl_pid = lockowner->l_pid;
970 spin_lock(&state->state_lock);
971 lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
972 if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
973 ret = -EIO;
974 else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
975 nfs4_stateid_copy(dst, &lsp->ls_stateid);
976 ret = 0;
977 smp_rmb();
978 if (!list_empty(&lsp->ls_seqid.list))
979 ret = -EWOULDBLOCK;
981 spin_unlock(&state->state_lock);
982 nfs4_put_lock_state(lsp);
983 out:
984 return ret;
987 static int nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
989 const nfs4_stateid *src;
990 int ret;
991 int seq;
993 do {
994 src = &zero_stateid;
995 seq = read_seqbegin(&state->seqlock);
996 if (test_bit(NFS_OPEN_STATE, &state->flags))
997 src = &state->open_stateid;
998 nfs4_stateid_copy(dst, src);
999 ret = 0;
1000 smp_rmb();
1001 if (!list_empty(&state->owner->so_seqid.list))
1002 ret = -EWOULDBLOCK;
1003 } while (read_seqretry(&state->seqlock, seq));
1004 return ret;
1008 * Byte-range lock aware utility to initialize the stateid of read/write
1009 * requests.
1011 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1012 fmode_t fmode, const struct nfs_lockowner *lockowner)
1014 int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1015 if (ret == -EIO)
1016 /* A lost lock - don't even consider delegations */
1017 goto out;
1018 if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1019 goto out;
1020 if (ret != -ENOENT)
1021 /* nfs4_copy_delegation_stateid() didn't over-write
1022 * dst, so it still has the lock stateid which we now
1023 * choose to use.
1025 goto out;
1026 ret = nfs4_copy_open_stateid(dst, state);
1027 out:
1028 if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1029 dst->seqid = 0;
1030 return ret;
1033 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1035 struct nfs_seqid *new;
1037 new = kmalloc(sizeof(*new), gfp_mask);
1038 if (new != NULL) {
1039 new->sequence = counter;
1040 INIT_LIST_HEAD(&new->list);
1041 new->task = NULL;
1043 return new;
1046 void nfs_release_seqid(struct nfs_seqid *seqid)
1048 struct nfs_seqid_counter *sequence;
1050 if (list_empty(&seqid->list))
1051 return;
1052 sequence = seqid->sequence;
1053 spin_lock(&sequence->lock);
1054 list_del_init(&seqid->list);
1055 if (!list_empty(&sequence->list)) {
1056 struct nfs_seqid *next;
1058 next = list_first_entry(&sequence->list,
1059 struct nfs_seqid, list);
1060 rpc_wake_up_queued_task(&sequence->wait, next->task);
1062 spin_unlock(&sequence->lock);
1065 void nfs_free_seqid(struct nfs_seqid *seqid)
1067 nfs_release_seqid(seqid);
1068 kfree(seqid);
1072 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1073 * failed with a seqid incrementing error -
1074 * see comments nfs4.h:seqid_mutating_error()
1076 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1078 switch (status) {
1079 case 0:
1080 break;
1081 case -NFS4ERR_BAD_SEQID:
1082 if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1083 return;
1084 pr_warn_ratelimited("NFS: v4 server returned a bad"
1085 " sequence-id error on an"
1086 " unconfirmed sequence %p!\n",
1087 seqid->sequence);
1088 case -NFS4ERR_STALE_CLIENTID:
1089 case -NFS4ERR_STALE_STATEID:
1090 case -NFS4ERR_BAD_STATEID:
1091 case -NFS4ERR_BADXDR:
1092 case -NFS4ERR_RESOURCE:
1093 case -NFS4ERR_NOFILEHANDLE:
1094 /* Non-seqid mutating errors */
1095 return;
1098 * Note: no locking needed as we are guaranteed to be first
1099 * on the sequence list
1101 seqid->sequence->counter++;
1104 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1106 struct nfs4_state_owner *sp = container_of(seqid->sequence,
1107 struct nfs4_state_owner, so_seqid);
1108 struct nfs_server *server = sp->so_server;
1110 if (status == -NFS4ERR_BAD_SEQID)
1111 nfs4_drop_state_owner(sp);
1112 if (!nfs4_has_session(server->nfs_client))
1113 nfs_increment_seqid(status, seqid);
1117 * Increment the seqid if the LOCK/LOCKU succeeded, or
1118 * failed with a seqid incrementing error -
1119 * see comments nfs4.h:seqid_mutating_error()
1121 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1123 nfs_increment_seqid(status, seqid);
1126 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1128 struct nfs_seqid_counter *sequence = seqid->sequence;
1129 int status = 0;
1131 spin_lock(&sequence->lock);
1132 seqid->task = task;
1133 if (list_empty(&seqid->list))
1134 list_add_tail(&seqid->list, &sequence->list);
1135 if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1136 goto unlock;
1137 rpc_sleep_on(&sequence->wait, task, NULL);
1138 status = -EAGAIN;
1139 unlock:
1140 spin_unlock(&sequence->lock);
1141 return status;
1144 static int nfs4_run_state_manager(void *);
1146 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1148 smp_mb__before_clear_bit();
1149 clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1150 smp_mb__after_clear_bit();
1151 wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1152 rpc_wake_up(&clp->cl_rpcwaitq);
1156 * Schedule the nfs_client asynchronous state management routine
1158 void nfs4_schedule_state_manager(struct nfs_client *clp)
1160 struct task_struct *task;
1161 char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1163 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1164 return;
1165 __module_get(THIS_MODULE);
1166 atomic_inc(&clp->cl_count);
1168 /* The rcu_read_lock() is not strictly necessary, as the state
1169 * manager is the only thread that ever changes the rpc_xprt
1170 * after it's initialized. At this point, we're single threaded. */
1171 rcu_read_lock();
1172 snprintf(buf, sizeof(buf), "%s-manager",
1173 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1174 rcu_read_unlock();
1175 task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1176 if (IS_ERR(task)) {
1177 printk(KERN_ERR "%s: kthread_run: %ld\n",
1178 __func__, PTR_ERR(task));
1179 nfs4_clear_state_manager_bit(clp);
1180 nfs_put_client(clp);
1181 module_put(THIS_MODULE);
1186 * Schedule a lease recovery attempt
1188 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1190 if (!clp)
1191 return;
1192 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1193 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1194 dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1195 clp->cl_hostname);
1196 nfs4_schedule_state_manager(clp);
1198 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1201 * nfs4_schedule_migration_recovery - trigger migration recovery
1203 * @server: FSID that is migrating
1205 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1206 * value is returned.
1208 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1210 struct nfs_client *clp = server->nfs_client;
1212 if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1213 pr_err("NFS: volatile file handles not supported (server %s)\n",
1214 clp->cl_hostname);
1215 return -NFS4ERR_IO;
1218 if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1219 return -NFS4ERR_IO;
1221 dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1222 __func__,
1223 (unsigned long long)server->fsid.major,
1224 (unsigned long long)server->fsid.minor,
1225 clp->cl_hostname);
1227 set_bit(NFS_MIG_IN_TRANSITION,
1228 &((struct nfs_server *)server)->mig_status);
1229 set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1231 nfs4_schedule_state_manager(clp);
1232 return 0;
1234 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1237 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1239 * @clp: server to check for moved leases
1242 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1244 dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1245 __func__, clp->cl_clientid, clp->cl_hostname);
1247 set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1248 nfs4_schedule_state_manager(clp);
1250 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1252 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1254 int res;
1256 might_sleep();
1258 atomic_inc(&clp->cl_count);
1259 res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1260 nfs_wait_bit_killable, TASK_KILLABLE);
1261 if (res)
1262 goto out;
1263 if (clp->cl_cons_state < 0)
1264 res = clp->cl_cons_state;
1265 out:
1266 nfs_put_client(clp);
1267 return res;
1270 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1272 unsigned int loop;
1273 int ret;
1275 for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1276 ret = nfs4_wait_clnt_recover(clp);
1277 if (ret != 0)
1278 break;
1279 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1280 !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1281 break;
1282 nfs4_schedule_state_manager(clp);
1283 ret = -EIO;
1285 return ret;
1289 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1290 * @clp: client to process
1292 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1293 * resend of the SETCLIENTID and hence re-establish the
1294 * callback channel. Then return all existing delegations.
1296 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1298 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1299 nfs_expire_all_delegations(clp);
1300 dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1301 clp->cl_hostname);
1304 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1306 nfs40_handle_cb_pathdown(clp);
1307 nfs4_schedule_state_manager(clp);
1310 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1313 set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1314 /* Don't recover state that expired before the reboot */
1315 if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1316 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1317 return 0;
1319 set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1320 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1321 return 1;
1324 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1326 set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1327 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1328 set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1329 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1330 return 1;
1333 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1335 struct nfs_client *clp = server->nfs_client;
1337 if (!nfs4_valid_open_stateid(state))
1338 return -EBADF;
1339 nfs4_state_mark_reclaim_nograce(clp, state);
1340 dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1341 clp->cl_hostname);
1342 nfs4_schedule_state_manager(clp);
1343 return 0;
1345 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1347 void nfs_inode_find_state_and_recover(struct inode *inode,
1348 const nfs4_stateid *stateid)
1350 struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1351 struct nfs_inode *nfsi = NFS_I(inode);
1352 struct nfs_open_context *ctx;
1353 struct nfs4_state *state;
1354 bool found = false;
1356 spin_lock(&inode->i_lock);
1357 list_for_each_entry(ctx, &nfsi->open_files, list) {
1358 state = ctx->state;
1359 if (state == NULL)
1360 continue;
1361 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1362 continue;
1363 if (!nfs4_stateid_match(&state->stateid, stateid))
1364 continue;
1365 nfs4_state_mark_reclaim_nograce(clp, state);
1366 found = true;
1368 spin_unlock(&inode->i_lock);
1369 if (found)
1370 nfs4_schedule_state_manager(clp);
1373 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1375 struct inode *inode = state->inode;
1376 struct nfs_inode *nfsi = NFS_I(inode);
1377 struct nfs_open_context *ctx;
1379 spin_lock(&inode->i_lock);
1380 list_for_each_entry(ctx, &nfsi->open_files, list) {
1381 if (ctx->state != state)
1382 continue;
1383 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1385 spin_unlock(&inode->i_lock);
1388 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1390 set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1391 nfs4_state_mark_open_context_bad(state);
1395 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1397 struct inode *inode = state->inode;
1398 struct nfs_inode *nfsi = NFS_I(inode);
1399 struct file_lock *fl;
1400 int status = 0;
1402 if (inode->i_flock == NULL)
1403 return 0;
1405 /* Guard against delegation returns and new lock/unlock calls */
1406 down_write(&nfsi->rwsem);
1407 /* Protect inode->i_flock using the BKL */
1408 spin_lock(&inode->i_lock);
1409 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1410 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1411 continue;
1412 if (nfs_file_open_context(fl->fl_file)->state != state)
1413 continue;
1414 spin_unlock(&inode->i_lock);
1415 status = ops->recover_lock(state, fl);
1416 switch (status) {
1417 case 0:
1418 break;
1419 case -ESTALE:
1420 case -NFS4ERR_ADMIN_REVOKED:
1421 case -NFS4ERR_STALE_STATEID:
1422 case -NFS4ERR_BAD_STATEID:
1423 case -NFS4ERR_EXPIRED:
1424 case -NFS4ERR_NO_GRACE:
1425 case -NFS4ERR_STALE_CLIENTID:
1426 case -NFS4ERR_BADSESSION:
1427 case -NFS4ERR_BADSLOT:
1428 case -NFS4ERR_BAD_HIGH_SLOT:
1429 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1430 goto out;
1431 default:
1432 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1433 __func__, status);
1434 case -ENOMEM:
1435 case -NFS4ERR_DENIED:
1436 case -NFS4ERR_RECLAIM_BAD:
1437 case -NFS4ERR_RECLAIM_CONFLICT:
1438 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1439 status = 0;
1441 spin_lock(&inode->i_lock);
1443 spin_unlock(&inode->i_lock);
1444 out:
1445 up_write(&nfsi->rwsem);
1446 return status;
1449 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1451 struct nfs4_state *state;
1452 struct nfs4_lock_state *lock;
1453 int status = 0;
1455 /* Note: we rely on the sp->so_states list being ordered
1456 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1457 * states first.
1458 * This is needed to ensure that the server won't give us any
1459 * read delegations that we have to return if, say, we are
1460 * recovering after a network partition or a reboot from a
1461 * server that doesn't support a grace period.
1463 spin_lock(&sp->so_lock);
1464 write_seqcount_begin(&sp->so_reclaim_seqcount);
1465 restart:
1466 list_for_each_entry(state, &sp->so_states, open_states) {
1467 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1468 continue;
1469 if (!nfs4_valid_open_stateid(state))
1470 continue;
1471 if (state->state == 0)
1472 continue;
1473 atomic_inc(&state->count);
1474 spin_unlock(&sp->so_lock);
1475 status = ops->recover_open(sp, state);
1476 if (status >= 0) {
1477 status = nfs4_reclaim_locks(state, ops);
1478 if (status >= 0) {
1479 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1480 spin_lock(&state->state_lock);
1481 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1482 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1483 pr_warn_ratelimited("NFS: "
1484 "%s: Lock reclaim "
1485 "failed!\n", __func__);
1487 spin_unlock(&state->state_lock);
1489 nfs4_put_open_state(state);
1490 spin_lock(&sp->so_lock);
1491 goto restart;
1494 switch (status) {
1495 default:
1496 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1497 __func__, status);
1498 case -ENOENT:
1499 case -ENOMEM:
1500 case -ESTALE:
1501 /* Open state on this file cannot be recovered */
1502 nfs4_state_mark_recovery_failed(state, status);
1503 break;
1504 case -EAGAIN:
1505 ssleep(1);
1506 case -NFS4ERR_ADMIN_REVOKED:
1507 case -NFS4ERR_STALE_STATEID:
1508 case -NFS4ERR_BAD_STATEID:
1509 case -NFS4ERR_RECLAIM_BAD:
1510 case -NFS4ERR_RECLAIM_CONFLICT:
1511 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1512 break;
1513 case -NFS4ERR_EXPIRED:
1514 case -NFS4ERR_NO_GRACE:
1515 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1516 case -NFS4ERR_STALE_CLIENTID:
1517 case -NFS4ERR_BADSESSION:
1518 case -NFS4ERR_BADSLOT:
1519 case -NFS4ERR_BAD_HIGH_SLOT:
1520 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1521 goto out_err;
1523 nfs4_put_open_state(state);
1524 spin_lock(&sp->so_lock);
1525 goto restart;
1527 write_seqcount_end(&sp->so_reclaim_seqcount);
1528 spin_unlock(&sp->so_lock);
1529 return 0;
1530 out_err:
1531 nfs4_put_open_state(state);
1532 spin_lock(&sp->so_lock);
1533 write_seqcount_end(&sp->so_reclaim_seqcount);
1534 spin_unlock(&sp->so_lock);
1535 return status;
1538 static void nfs4_clear_open_state(struct nfs4_state *state)
1540 struct nfs4_lock_state *lock;
1542 clear_bit(NFS_DELEGATED_STATE, &state->flags);
1543 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1544 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1545 clear_bit(NFS_O_RDWR_STATE, &state->flags);
1546 spin_lock(&state->state_lock);
1547 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1548 lock->ls_seqid.flags = 0;
1549 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1551 spin_unlock(&state->state_lock);
1554 static void nfs4_reset_seqids(struct nfs_server *server,
1555 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1557 struct nfs_client *clp = server->nfs_client;
1558 struct nfs4_state_owner *sp;
1559 struct rb_node *pos;
1560 struct nfs4_state *state;
1562 spin_lock(&clp->cl_lock);
1563 for (pos = rb_first(&server->state_owners);
1564 pos != NULL;
1565 pos = rb_next(pos)) {
1566 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1567 sp->so_seqid.flags = 0;
1568 spin_lock(&sp->so_lock);
1569 list_for_each_entry(state, &sp->so_states, open_states) {
1570 if (mark_reclaim(clp, state))
1571 nfs4_clear_open_state(state);
1573 spin_unlock(&sp->so_lock);
1575 spin_unlock(&clp->cl_lock);
1578 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1579 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1581 struct nfs_server *server;
1583 rcu_read_lock();
1584 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1585 nfs4_reset_seqids(server, mark_reclaim);
1586 rcu_read_unlock();
1589 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1591 /* Mark all delegations for reclaim */
1592 nfs_delegation_mark_reclaim(clp);
1593 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1596 static void nfs4_reclaim_complete(struct nfs_client *clp,
1597 const struct nfs4_state_recovery_ops *ops,
1598 struct rpc_cred *cred)
1600 /* Notify the server we're done reclaiming our state */
1601 if (ops->reclaim_complete)
1602 (void)ops->reclaim_complete(clp, cred);
1605 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1607 struct nfs_client *clp = server->nfs_client;
1608 struct nfs4_state_owner *sp;
1609 struct rb_node *pos;
1610 struct nfs4_state *state;
1612 spin_lock(&clp->cl_lock);
1613 for (pos = rb_first(&server->state_owners);
1614 pos != NULL;
1615 pos = rb_next(pos)) {
1616 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1617 spin_lock(&sp->so_lock);
1618 list_for_each_entry(state, &sp->so_states, open_states) {
1619 if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1620 &state->flags))
1621 continue;
1622 nfs4_state_mark_reclaim_nograce(clp, state);
1624 spin_unlock(&sp->so_lock);
1626 spin_unlock(&clp->cl_lock);
1629 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1631 struct nfs_server *server;
1633 if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1634 return 0;
1636 rcu_read_lock();
1637 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1638 nfs4_clear_reclaim_server(server);
1639 rcu_read_unlock();
1641 nfs_delegation_reap_unclaimed(clp);
1642 return 1;
1645 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1647 const struct nfs4_state_recovery_ops *ops;
1648 struct rpc_cred *cred;
1650 if (!nfs4_state_clear_reclaim_reboot(clp))
1651 return;
1652 ops = clp->cl_mvops->reboot_recovery_ops;
1653 cred = nfs4_get_clid_cred(clp);
1654 nfs4_reclaim_complete(clp, ops, cred);
1655 put_rpccred(cred);
1658 static void nfs_delegation_clear_all(struct nfs_client *clp)
1660 nfs_delegation_mark_reclaim(clp);
1661 nfs_delegation_reap_unclaimed(clp);
1664 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1666 nfs_delegation_clear_all(clp);
1667 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1670 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1672 switch (error) {
1673 case 0:
1674 break;
1675 case -NFS4ERR_CB_PATH_DOWN:
1676 nfs40_handle_cb_pathdown(clp);
1677 break;
1678 case -NFS4ERR_NO_GRACE:
1679 nfs4_state_end_reclaim_reboot(clp);
1680 break;
1681 case -NFS4ERR_STALE_CLIENTID:
1682 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1683 nfs4_state_clear_reclaim_reboot(clp);
1684 nfs4_state_start_reclaim_reboot(clp);
1685 break;
1686 case -NFS4ERR_EXPIRED:
1687 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1688 nfs4_state_start_reclaim_nograce(clp);
1689 break;
1690 case -NFS4ERR_BADSESSION:
1691 case -NFS4ERR_BADSLOT:
1692 case -NFS4ERR_BAD_HIGH_SLOT:
1693 case -NFS4ERR_DEADSESSION:
1694 case -NFS4ERR_SEQ_FALSE_RETRY:
1695 case -NFS4ERR_SEQ_MISORDERED:
1696 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1697 /* Zero session reset errors */
1698 break;
1699 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1700 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1701 break;
1702 default:
1703 dprintk("%s: failed to handle error %d for server %s\n",
1704 __func__, error, clp->cl_hostname);
1705 return error;
1707 dprintk("%s: handled error %d for server %s\n", __func__, error,
1708 clp->cl_hostname);
1709 return 0;
1712 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1714 struct nfs4_state_owner *sp;
1715 struct nfs_server *server;
1716 struct rb_node *pos;
1717 int status = 0;
1719 restart:
1720 rcu_read_lock();
1721 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1722 nfs4_purge_state_owners(server);
1723 spin_lock(&clp->cl_lock);
1724 for (pos = rb_first(&server->state_owners);
1725 pos != NULL;
1726 pos = rb_next(pos)) {
1727 sp = rb_entry(pos,
1728 struct nfs4_state_owner, so_server_node);
1729 if (!test_and_clear_bit(ops->owner_flag_bit,
1730 &sp->so_flags))
1731 continue;
1732 atomic_inc(&sp->so_count);
1733 spin_unlock(&clp->cl_lock);
1734 rcu_read_unlock();
1736 status = nfs4_reclaim_open_state(sp, ops);
1737 if (status < 0) {
1738 set_bit(ops->owner_flag_bit, &sp->so_flags);
1739 nfs4_put_state_owner(sp);
1740 return nfs4_recovery_handle_error(clp, status);
1743 nfs4_put_state_owner(sp);
1744 goto restart;
1746 spin_unlock(&clp->cl_lock);
1748 rcu_read_unlock();
1749 return status;
1752 static int nfs4_check_lease(struct nfs_client *clp)
1754 struct rpc_cred *cred;
1755 const struct nfs4_state_maintenance_ops *ops =
1756 clp->cl_mvops->state_renewal_ops;
1757 int status;
1759 /* Is the client already known to have an expired lease? */
1760 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1761 return 0;
1762 spin_lock(&clp->cl_lock);
1763 cred = ops->get_state_renewal_cred_locked(clp);
1764 spin_unlock(&clp->cl_lock);
1765 if (cred == NULL) {
1766 cred = nfs4_get_clid_cred(clp);
1767 status = -ENOKEY;
1768 if (cred == NULL)
1769 goto out;
1771 status = ops->renew_lease(clp, cred);
1772 put_rpccred(cred);
1773 if (status == -ETIMEDOUT) {
1774 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1775 return 0;
1777 out:
1778 return nfs4_recovery_handle_error(clp, status);
1781 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1782 * and for recoverable errors on EXCHANGE_ID for v4.1
1784 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1786 switch (status) {
1787 case -NFS4ERR_SEQ_MISORDERED:
1788 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1789 return -ESERVERFAULT;
1790 /* Lease confirmation error: retry after purging the lease */
1791 ssleep(1);
1792 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1793 break;
1794 case -NFS4ERR_STALE_CLIENTID:
1795 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1796 nfs4_state_clear_reclaim_reboot(clp);
1797 nfs4_state_start_reclaim_reboot(clp);
1798 break;
1799 case -NFS4ERR_CLID_INUSE:
1800 pr_err("NFS: Server %s reports our clientid is in use\n",
1801 clp->cl_hostname);
1802 nfs_mark_client_ready(clp, -EPERM);
1803 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1804 return -EPERM;
1805 case -EACCES:
1806 case -NFS4ERR_DELAY:
1807 case -ETIMEDOUT:
1808 case -EAGAIN:
1809 ssleep(1);
1810 break;
1812 case -NFS4ERR_MINOR_VERS_MISMATCH:
1813 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1814 nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1815 dprintk("%s: exit with error %d for server %s\n",
1816 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1817 return -EPROTONOSUPPORT;
1818 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1819 * in nfs4_exchange_id */
1820 default:
1821 dprintk("%s: exit with error %d for server %s\n", __func__,
1822 status, clp->cl_hostname);
1823 return status;
1825 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1826 dprintk("%s: handled error %d for server %s\n", __func__, status,
1827 clp->cl_hostname);
1828 return 0;
1831 static int nfs4_establish_lease(struct nfs_client *clp)
1833 struct rpc_cred *cred;
1834 const struct nfs4_state_recovery_ops *ops =
1835 clp->cl_mvops->reboot_recovery_ops;
1836 int status;
1838 cred = nfs4_get_clid_cred(clp);
1839 if (cred == NULL)
1840 return -ENOENT;
1841 status = ops->establish_clid(clp, cred);
1842 put_rpccred(cred);
1843 if (status != 0)
1844 return status;
1845 pnfs_destroy_all_layouts(clp);
1846 return 0;
1850 * Returns zero or a negative errno. NFS4ERR values are converted
1851 * to local errno values.
1853 static int nfs4_reclaim_lease(struct nfs_client *clp)
1855 int status;
1857 status = nfs4_establish_lease(clp);
1858 if (status < 0)
1859 return nfs4_handle_reclaim_lease_error(clp, status);
1860 if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1861 nfs4_state_start_reclaim_nograce(clp);
1862 if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1863 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1864 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1865 clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1866 return 0;
1869 static int nfs4_purge_lease(struct nfs_client *clp)
1871 int status;
1873 status = nfs4_establish_lease(clp);
1874 if (status < 0)
1875 return nfs4_handle_reclaim_lease_error(clp, status);
1876 clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1877 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1878 nfs4_state_start_reclaim_nograce(clp);
1879 return 0;
1883 * Try remote migration of one FSID from a source server to a
1884 * destination server. The source server provides a list of
1885 * potential destinations.
1887 * Returns zero or a negative NFS4ERR status code.
1889 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1891 struct nfs_client *clp = server->nfs_client;
1892 struct nfs4_fs_locations *locations = NULL;
1893 struct inode *inode;
1894 struct page *page;
1895 int status, result;
1897 dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1898 (unsigned long long)server->fsid.major,
1899 (unsigned long long)server->fsid.minor,
1900 clp->cl_hostname);
1902 result = 0;
1903 page = alloc_page(GFP_KERNEL);
1904 locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1905 if (page == NULL || locations == NULL) {
1906 dprintk("<-- %s: no memory\n", __func__);
1907 goto out;
1910 inode = server->super->s_root->d_inode;
1911 result = nfs4_proc_get_locations(inode, locations, page, cred);
1912 if (result) {
1913 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1914 __func__, result);
1915 goto out;
1918 result = -NFS4ERR_NXIO;
1919 if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1920 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1921 __func__);
1922 goto out;
1925 nfs4_begin_drain_session(clp);
1927 status = nfs4_replace_transport(server, locations);
1928 if (status != 0) {
1929 dprintk("<-- %s: failed to replace transport: %d\n",
1930 __func__, status);
1931 goto out;
1934 result = 0;
1935 dprintk("<-- %s: migration succeeded\n", __func__);
1937 out:
1938 if (page != NULL)
1939 __free_page(page);
1940 kfree(locations);
1941 if (result) {
1942 pr_err("NFS: migration recovery failed (server %s)\n",
1943 clp->cl_hostname);
1944 set_bit(NFS_MIG_FAILED, &server->mig_status);
1946 return result;
1950 * Returns zero or a negative NFS4ERR status code.
1952 static int nfs4_handle_migration(struct nfs_client *clp)
1954 const struct nfs4_state_maintenance_ops *ops =
1955 clp->cl_mvops->state_renewal_ops;
1956 struct nfs_server *server;
1957 struct rpc_cred *cred;
1959 dprintk("%s: migration reported on \"%s\"\n", __func__,
1960 clp->cl_hostname);
1962 spin_lock(&clp->cl_lock);
1963 cred = ops->get_state_renewal_cred_locked(clp);
1964 spin_unlock(&clp->cl_lock);
1965 if (cred == NULL)
1966 return -NFS4ERR_NOENT;
1968 clp->cl_mig_gen++;
1969 restart:
1970 rcu_read_lock();
1971 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1972 int status;
1974 if (server->mig_gen == clp->cl_mig_gen)
1975 continue;
1976 server->mig_gen = clp->cl_mig_gen;
1978 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1979 &server->mig_status))
1980 continue;
1982 rcu_read_unlock();
1983 status = nfs4_try_migration(server, cred);
1984 if (status < 0) {
1985 put_rpccred(cred);
1986 return status;
1988 goto restart;
1990 rcu_read_unlock();
1991 put_rpccred(cred);
1992 return 0;
1996 * Test each nfs_server on the clp's cl_superblocks list to see
1997 * if it's moved to another server. Stop when the server no longer
1998 * returns NFS4ERR_LEASE_MOVED.
2000 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2002 const struct nfs4_state_maintenance_ops *ops =
2003 clp->cl_mvops->state_renewal_ops;
2004 struct nfs_server *server;
2005 struct rpc_cred *cred;
2007 dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2008 clp->cl_hostname);
2010 spin_lock(&clp->cl_lock);
2011 cred = ops->get_state_renewal_cred_locked(clp);
2012 spin_unlock(&clp->cl_lock);
2013 if (cred == NULL)
2014 return -NFS4ERR_NOENT;
2016 clp->cl_mig_gen++;
2017 restart:
2018 rcu_read_lock();
2019 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2020 struct inode *inode;
2021 int status;
2023 if (server->mig_gen == clp->cl_mig_gen)
2024 continue;
2025 server->mig_gen = clp->cl_mig_gen;
2027 rcu_read_unlock();
2029 inode = server->super->s_root->d_inode;
2030 status = nfs4_proc_fsid_present(inode, cred);
2031 if (status != -NFS4ERR_MOVED)
2032 goto restart; /* wasn't this one */
2033 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2034 goto restart; /* there are more */
2035 goto out;
2037 rcu_read_unlock();
2039 out:
2040 put_rpccred(cred);
2041 return 0;
2045 * nfs4_discover_server_trunking - Detect server IP address trunking
2047 * @clp: nfs_client under test
2048 * @result: OUT: found nfs_client, or clp
2050 * Returns zero or a negative errno. If zero is returned,
2051 * an nfs_client pointer is planted in "result".
2053 * Note: since we are invoked in process context, and
2054 * not from inside the state manager, we cannot use
2055 * nfs4_handle_reclaim_lease_error().
2057 int nfs4_discover_server_trunking(struct nfs_client *clp,
2058 struct nfs_client **result)
2060 const struct nfs4_state_recovery_ops *ops =
2061 clp->cl_mvops->reboot_recovery_ops;
2062 struct rpc_clnt *clnt;
2063 struct rpc_cred *cred;
2064 int i, status;
2066 dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2068 clnt = clp->cl_rpcclient;
2069 i = 0;
2071 mutex_lock(&nfs_clid_init_mutex);
2072 again:
2073 status = -ENOENT;
2074 cred = nfs4_get_clid_cred(clp);
2075 if (cred == NULL)
2076 goto out_unlock;
2078 status = ops->detect_trunking(clp, result, cred);
2079 put_rpccred(cred);
2080 switch (status) {
2081 case 0:
2082 break;
2083 case -NFS4ERR_DELAY:
2084 case -ETIMEDOUT:
2085 case -EAGAIN:
2086 ssleep(1);
2087 case -NFS4ERR_STALE_CLIENTID:
2088 dprintk("NFS: %s after status %d, retrying\n",
2089 __func__, status);
2090 goto again;
2091 case -EACCES:
2092 if (i++ == 0) {
2093 nfs4_root_machine_cred(clp);
2094 goto again;
2096 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2097 break;
2098 case -NFS4ERR_CLID_INUSE:
2099 case -NFS4ERR_WRONGSEC:
2100 /* No point in retrying if we already used RPC_AUTH_UNIX */
2101 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2102 status = -EPERM;
2103 break;
2105 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2106 if (IS_ERR(clnt)) {
2107 status = PTR_ERR(clnt);
2108 break;
2110 /* Note: this is safe because we haven't yet marked the
2111 * client as ready, so we are the only user of
2112 * clp->cl_rpcclient
2114 clnt = xchg(&clp->cl_rpcclient, clnt);
2115 rpc_shutdown_client(clnt);
2116 clnt = clp->cl_rpcclient;
2117 goto again;
2119 case -NFS4ERR_MINOR_VERS_MISMATCH:
2120 status = -EPROTONOSUPPORT;
2121 break;
2123 case -EKEYEXPIRED:
2124 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2125 * in nfs4_exchange_id */
2126 status = -EKEYEXPIRED;
2127 break;
2128 default:
2129 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2130 __func__, status);
2131 status = -EIO;
2134 out_unlock:
2135 mutex_unlock(&nfs_clid_init_mutex);
2136 dprintk("NFS: %s: status = %d\n", __func__, status);
2137 return status;
2140 #ifdef CONFIG_NFS_V4_1
2141 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2143 struct nfs_client *clp = session->clp;
2145 switch (err) {
2146 default:
2147 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2148 break;
2149 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2150 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2152 nfs4_schedule_lease_recovery(clp);
2154 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2156 static void nfs41_ping_server(struct nfs_client *clp)
2158 /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2159 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2160 nfs4_schedule_state_manager(clp);
2163 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2165 nfs41_ping_server(clp);
2168 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2170 nfs41_ping_server(clp);
2173 static void nfs4_reset_all_state(struct nfs_client *clp)
2175 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2176 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2177 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2178 nfs4_state_start_reclaim_nograce(clp);
2179 dprintk("%s: scheduling reset of all state for server %s!\n",
2180 __func__, clp->cl_hostname);
2181 nfs4_schedule_state_manager(clp);
2185 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2187 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2188 nfs4_state_start_reclaim_reboot(clp);
2189 dprintk("%s: server %s rebooted!\n", __func__,
2190 clp->cl_hostname);
2191 nfs4_schedule_state_manager(clp);
2195 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2197 nfs4_reset_all_state(clp);
2198 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2201 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2203 /* This will need to handle layouts too */
2204 nfs_expire_all_delegations(clp);
2205 dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2206 clp->cl_hostname);
2209 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2211 nfs_expire_all_delegations(clp);
2212 if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2213 nfs4_schedule_state_manager(clp);
2214 dprintk("%s: server %s declared a backchannel fault\n", __func__,
2215 clp->cl_hostname);
2218 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2220 if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2221 &clp->cl_state) == 0)
2222 nfs4_schedule_state_manager(clp);
2225 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2227 if (!flags)
2228 return;
2230 dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2231 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2233 if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2234 nfs41_handle_server_reboot(clp);
2235 if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2236 SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2237 SEQ4_STATUS_ADMIN_STATE_REVOKED))
2238 nfs41_handle_state_revoked(clp);
2239 if (flags & SEQ4_STATUS_LEASE_MOVED)
2240 nfs4_schedule_lease_moved_recovery(clp);
2241 if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2242 nfs41_handle_recallable_state_revoked(clp);
2243 if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2244 nfs41_handle_backchannel_fault(clp);
2245 else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2246 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2247 nfs41_handle_cb_path_down(clp);
2250 static int nfs4_reset_session(struct nfs_client *clp)
2252 struct rpc_cred *cred;
2253 int status;
2255 if (!nfs4_has_session(clp))
2256 return 0;
2257 nfs4_begin_drain_session(clp);
2258 cred = nfs4_get_clid_cred(clp);
2259 status = nfs4_proc_destroy_session(clp->cl_session, cred);
2260 switch (status) {
2261 case 0:
2262 case -NFS4ERR_BADSESSION:
2263 case -NFS4ERR_DEADSESSION:
2264 break;
2265 case -NFS4ERR_BACK_CHAN_BUSY:
2266 case -NFS4ERR_DELAY:
2267 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2268 status = 0;
2269 ssleep(1);
2270 goto out;
2271 default:
2272 status = nfs4_recovery_handle_error(clp, status);
2273 goto out;
2276 memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2277 status = nfs4_proc_create_session(clp, cred);
2278 if (status) {
2279 dprintk("%s: session reset failed with status %d for server %s!\n",
2280 __func__, status, clp->cl_hostname);
2281 status = nfs4_handle_reclaim_lease_error(clp, status);
2282 goto out;
2284 nfs41_finish_session_reset(clp);
2285 dprintk("%s: session reset was successful for server %s!\n",
2286 __func__, clp->cl_hostname);
2287 out:
2288 if (cred)
2289 put_rpccred(cred);
2290 return status;
2293 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2295 struct rpc_cred *cred;
2296 int ret;
2298 if (!nfs4_has_session(clp))
2299 return 0;
2300 nfs4_begin_drain_session(clp);
2301 cred = nfs4_get_clid_cred(clp);
2302 ret = nfs4_proc_bind_conn_to_session(clp, cred);
2303 if (cred)
2304 put_rpccred(cred);
2305 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2306 switch (ret) {
2307 case 0:
2308 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2309 __func__, clp->cl_hostname);
2310 break;
2311 case -NFS4ERR_DELAY:
2312 ssleep(1);
2313 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2314 break;
2315 default:
2316 return nfs4_recovery_handle_error(clp, ret);
2318 return 0;
2320 #else /* CONFIG_NFS_V4_1 */
2321 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2323 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2325 return 0;
2327 #endif /* CONFIG_NFS_V4_1 */
2329 static void nfs4_state_manager(struct nfs_client *clp)
2331 int status = 0;
2332 const char *section = "", *section_sep = "";
2334 /* Ensure exclusive access to NFSv4 state */
2335 do {
2336 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2337 section = "purge state";
2338 status = nfs4_purge_lease(clp);
2339 if (status < 0)
2340 goto out_error;
2341 continue;
2344 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2345 section = "lease expired";
2346 /* We're going to have to re-establish a clientid */
2347 status = nfs4_reclaim_lease(clp);
2348 if (status < 0)
2349 goto out_error;
2350 continue;
2353 /* Initialize or reset the session */
2354 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2355 section = "reset session";
2356 status = nfs4_reset_session(clp);
2357 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2358 continue;
2359 if (status < 0)
2360 goto out_error;
2363 /* Send BIND_CONN_TO_SESSION */
2364 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2365 &clp->cl_state)) {
2366 section = "bind conn to session";
2367 status = nfs4_bind_conn_to_session(clp);
2368 if (status < 0)
2369 goto out_error;
2370 continue;
2373 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2374 section = "check lease";
2375 status = nfs4_check_lease(clp);
2376 if (status < 0)
2377 goto out_error;
2380 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2381 section = "migration";
2382 status = nfs4_handle_migration(clp);
2383 if (status < 0)
2384 goto out_error;
2387 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2388 section = "lease moved";
2389 status = nfs4_handle_lease_moved(clp);
2390 if (status < 0)
2391 goto out_error;
2394 /* First recover reboot state... */
2395 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2396 section = "reclaim reboot";
2397 status = nfs4_do_reclaim(clp,
2398 clp->cl_mvops->reboot_recovery_ops);
2399 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2400 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2401 continue;
2402 nfs4_state_end_reclaim_reboot(clp);
2403 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2404 continue;
2405 if (status < 0)
2406 goto out_error;
2409 /* Now recover expired state... */
2410 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2411 section = "reclaim nograce";
2412 status = nfs4_do_reclaim(clp,
2413 clp->cl_mvops->nograce_recovery_ops);
2414 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2415 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2416 test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2417 continue;
2418 if (status < 0)
2419 goto out_error;
2422 nfs4_end_drain_session(clp);
2423 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2424 nfs_client_return_marked_delegations(clp);
2425 continue;
2428 nfs4_clear_state_manager_bit(clp);
2429 /* Did we race with an attempt to give us more work? */
2430 if (clp->cl_state == 0)
2431 break;
2432 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2433 break;
2434 } while (atomic_read(&clp->cl_count) > 1);
2435 return;
2436 out_error:
2437 if (strlen(section))
2438 section_sep = ": ";
2439 pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2440 " with error %d\n", section_sep, section,
2441 clp->cl_hostname, -status);
2442 ssleep(1);
2443 nfs4_end_drain_session(clp);
2444 nfs4_clear_state_manager_bit(clp);
2447 static int nfs4_run_state_manager(void *ptr)
2449 struct nfs_client *clp = ptr;
2451 allow_signal(SIGKILL);
2452 nfs4_state_manager(clp);
2453 nfs_put_client(clp);
2454 module_put_and_exit(0);
2455 return 0;
2459 * Local variables:
2460 * c-basic-offset: 8
2461 * End: