GNU Linux-libre 4.9.309-gnu1
[releases.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
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.
23  *
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.
35  *
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.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/kthread.h>
46 #include <linux/module.h>
47 #include <linux/random.h>
48 #include <linux/ratelimit.h>
49 #include <linux/workqueue.h>
50 #include <linux/bitops.h>
51 #include <linux/jiffies.h>
52
53 #include <linux/sunrpc/clnt.h>
54
55 #include "nfs4_fs.h"
56 #include "callback.h"
57 #include "delegation.h"
58 #include "internal.h"
59 #include "nfs4idmap.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
63
64 #define NFSDBG_FACILITY         NFSDBG_STATE
65
66 #define OPENOWNER_POOL_SIZE     8
67
68 const nfs4_stateid zero_stateid = {
69         { .data = { 0 } },
70         .type = NFS4_SPECIAL_STATEID_TYPE,
71 };
72 static DEFINE_MUTEX(nfs_clid_init_mutex);
73
74 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
75 {
76         struct nfs4_setclientid_res clid = {
77                 .clientid = clp->cl_clientid,
78                 .confirm = clp->cl_confirm,
79         };
80         unsigned short port;
81         int status;
82         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
83
84         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
85                 goto do_confirm;
86         port = nn->nfs_callback_tcpport;
87         if (clp->cl_addr.ss_family == AF_INET6)
88                 port = nn->nfs_callback_tcpport6;
89
90         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
91         if (status != 0)
92                 goto out;
93         clp->cl_clientid = clid.clientid;
94         clp->cl_confirm = clid.confirm;
95         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
96 do_confirm:
97         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
98         if (status != 0)
99                 goto out;
100         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
101         nfs4_schedule_state_renewal(clp);
102 out:
103         return status;
104 }
105
106 /**
107  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
108  *
109  * @clp: nfs_client under test
110  * @result: OUT: found nfs_client, or clp
111  * @cred: credential to use for trunking test
112  *
113  * Returns zero, a negative errno, or a negative NFS4ERR status.
114  * If zero is returned, an nfs_client pointer is planted in
115  * "result".
116  *
117  * Note: The returned client may not yet be marked ready.
118  */
119 int nfs40_discover_server_trunking(struct nfs_client *clp,
120                                    struct nfs_client **result,
121                                    struct rpc_cred *cred)
122 {
123         struct nfs4_setclientid_res clid = {
124                 .clientid = clp->cl_clientid,
125                 .confirm = clp->cl_confirm,
126         };
127         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
128         unsigned short port;
129         int status;
130
131         port = nn->nfs_callback_tcpport;
132         if (clp->cl_addr.ss_family == AF_INET6)
133                 port = nn->nfs_callback_tcpport6;
134
135         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
136         if (status != 0)
137                 goto out;
138         clp->cl_clientid = clid.clientid;
139         clp->cl_confirm = clid.confirm;
140
141         status = nfs40_walk_client_list(clp, result, cred);
142         if (status == 0) {
143                 /* Sustain the lease, even if it's empty.  If the clientid4
144                  * goes stale it's of no use for trunking discovery. */
145                 nfs4_schedule_state_renewal(*result);
146
147                 /* If the client state need to recover, do it. */
148                 if (clp->cl_state)
149                         nfs4_schedule_state_manager(clp);
150         }
151 out:
152         return status;
153 }
154
155 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
156 {
157         struct rpc_cred *cred = NULL;
158
159         if (clp->cl_machine_cred != NULL)
160                 cred = get_rpccred(clp->cl_machine_cred);
161         return cred;
162 }
163
164 static void nfs4_root_machine_cred(struct nfs_client *clp)
165 {
166         struct rpc_cred *cred, *new;
167
168         new = rpc_lookup_machine_cred(NULL);
169         spin_lock(&clp->cl_lock);
170         cred = clp->cl_machine_cred;
171         clp->cl_machine_cred = new;
172         spin_unlock(&clp->cl_lock);
173         if (cred != NULL)
174                 put_rpccred(cred);
175 }
176
177 static struct rpc_cred *
178 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
179 {
180         struct rpc_cred *cred = NULL;
181         struct nfs4_state_owner *sp;
182         struct rb_node *pos;
183
184         for (pos = rb_first(&server->state_owners);
185              pos != NULL;
186              pos = rb_next(pos)) {
187                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
188                 if (list_empty(&sp->so_states))
189                         continue;
190                 cred = get_rpccred(sp->so_cred);
191                 break;
192         }
193         return cred;
194 }
195
196 /**
197  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
198  * @clp: client state handle
199  *
200  * Returns an rpc_cred with reference count bumped, or NULL.
201  * Caller must hold clp->cl_lock.
202  */
203 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
204 {
205         struct rpc_cred *cred = NULL;
206         struct nfs_server *server;
207
208         /* Use machine credentials if available */
209         cred = nfs4_get_machine_cred_locked(clp);
210         if (cred != NULL)
211                 goto out;
212
213         rcu_read_lock();
214         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
215                 cred = nfs4_get_renew_cred_server_locked(server);
216                 if (cred != NULL)
217                         break;
218         }
219         rcu_read_unlock();
220
221 out:
222         return cred;
223 }
224
225 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
226 {
227         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
228                 spin_lock(&tbl->slot_tbl_lock);
229                 nfs41_wake_slot_table(tbl);
230                 spin_unlock(&tbl->slot_tbl_lock);
231         }
232 }
233
234 static void nfs4_end_drain_session(struct nfs_client *clp)
235 {
236         struct nfs4_session *ses = clp->cl_session;
237
238         if (clp->cl_slot_tbl) {
239                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
240                 return;
241         }
242
243         if (ses != NULL) {
244                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
245                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
246         }
247 }
248
249 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
250 {
251         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
252         spin_lock(&tbl->slot_tbl_lock);
253         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
254                 reinit_completion(&tbl->complete);
255                 spin_unlock(&tbl->slot_tbl_lock);
256                 return wait_for_completion_interruptible(&tbl->complete);
257         }
258         spin_unlock(&tbl->slot_tbl_lock);
259         return 0;
260 }
261
262 static int nfs4_begin_drain_session(struct nfs_client *clp)
263 {
264         struct nfs4_session *ses = clp->cl_session;
265         int ret = 0;
266
267         if (clp->cl_slot_tbl)
268                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
269
270         /* back channel */
271         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
272         if (ret)
273                 return ret;
274         /* fore channel */
275         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
276 }
277
278 #if defined(CONFIG_NFS_V4_1)
279
280 static int nfs41_setup_state_renewal(struct nfs_client *clp)
281 {
282         int status;
283         struct nfs_fsinfo fsinfo;
284         unsigned long now;
285
286         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
287                 nfs4_schedule_state_renewal(clp);
288                 return 0;
289         }
290
291         now = jiffies;
292         status = nfs4_proc_get_lease_time(clp, &fsinfo);
293         if (status == 0) {
294                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
295                 nfs4_schedule_state_renewal(clp);
296         }
297
298         return status;
299 }
300
301 static void nfs41_finish_session_reset(struct nfs_client *clp)
302 {
303         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
304         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
305         /* create_session negotiated new slot table */
306         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
307         nfs41_setup_state_renewal(clp);
308 }
309
310 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
311 {
312         int status;
313
314         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
315                 goto do_confirm;
316         status = nfs4_proc_exchange_id(clp, cred);
317         if (status != 0)
318                 goto out;
319         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
320 do_confirm:
321         status = nfs4_proc_create_session(clp, cred);
322         if (status != 0)
323                 goto out;
324         nfs41_finish_session_reset(clp);
325         nfs_mark_client_ready(clp, NFS_CS_READY);
326 out:
327         return status;
328 }
329
330 /**
331  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
332  *
333  * @clp: nfs_client under test
334  * @result: OUT: found nfs_client, or clp
335  * @cred: credential to use for trunking test
336  *
337  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
338  * If NFS4_OK is returned, an nfs_client pointer is planted in
339  * "result".
340  *
341  * Note: The returned client may not yet be marked ready.
342  */
343 int nfs41_discover_server_trunking(struct nfs_client *clp,
344                                    struct nfs_client **result,
345                                    struct rpc_cred *cred)
346 {
347         int status;
348
349         status = nfs4_proc_exchange_id(clp, cred);
350         if (status != NFS4_OK)
351                 return status;
352
353         status = nfs41_walk_client_list(clp, result, cred);
354         if (status < 0)
355                 return status;
356         if (clp != *result)
357                 return 0;
358
359         /* Purge state if the client id was established in a prior instance */
360         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
361                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
362         else
363                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
364         nfs4_schedule_state_manager(clp);
365         status = nfs_wait_client_init_complete(clp);
366         if (status < 0)
367                 nfs_put_client(clp);
368         return status;
369 }
370
371 #endif /* CONFIG_NFS_V4_1 */
372
373 /**
374  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
375  * @clp: client state handle
376  *
377  * Returns an rpc_cred with reference count bumped, or NULL.
378  */
379 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
380 {
381         struct rpc_cred *cred;
382
383         spin_lock(&clp->cl_lock);
384         cred = nfs4_get_machine_cred_locked(clp);
385         spin_unlock(&clp->cl_lock);
386         return cred;
387 }
388
389 static struct nfs4_state_owner *
390 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
391 {
392         struct rb_node **p = &server->state_owners.rb_node,
393                        *parent = NULL;
394         struct nfs4_state_owner *sp;
395
396         while (*p != NULL) {
397                 parent = *p;
398                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
399
400                 if (cred < sp->so_cred)
401                         p = &parent->rb_left;
402                 else if (cred > sp->so_cred)
403                         p = &parent->rb_right;
404                 else {
405                         if (!list_empty(&sp->so_lru))
406                                 list_del_init(&sp->so_lru);
407                         atomic_inc(&sp->so_count);
408                         return sp;
409                 }
410         }
411         return NULL;
412 }
413
414 static struct nfs4_state_owner *
415 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
416 {
417         struct nfs_server *server = new->so_server;
418         struct rb_node **p = &server->state_owners.rb_node,
419                        *parent = NULL;
420         struct nfs4_state_owner *sp;
421         int err;
422
423         while (*p != NULL) {
424                 parent = *p;
425                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
426
427                 if (new->so_cred < sp->so_cred)
428                         p = &parent->rb_left;
429                 else if (new->so_cred > sp->so_cred)
430                         p = &parent->rb_right;
431                 else {
432                         if (!list_empty(&sp->so_lru))
433                                 list_del_init(&sp->so_lru);
434                         atomic_inc(&sp->so_count);
435                         return sp;
436                 }
437         }
438         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
439         if (err)
440                 return ERR_PTR(err);
441         rb_link_node(&new->so_server_node, parent, p);
442         rb_insert_color(&new->so_server_node, &server->state_owners);
443         return new;
444 }
445
446 static void
447 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
448 {
449         struct nfs_server *server = sp->so_server;
450
451         if (!RB_EMPTY_NODE(&sp->so_server_node))
452                 rb_erase(&sp->so_server_node, &server->state_owners);
453         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
454 }
455
456 static void
457 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
458 {
459         sc->create_time = ktime_get();
460         sc->flags = 0;
461         sc->counter = 0;
462         spin_lock_init(&sc->lock);
463         INIT_LIST_HEAD(&sc->list);
464         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
465 }
466
467 static void
468 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
469 {
470         rpc_destroy_wait_queue(&sc->wait);
471 }
472
473 /*
474  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
475  * create a new state_owner.
476  *
477  */
478 static struct nfs4_state_owner *
479 nfs4_alloc_state_owner(struct nfs_server *server,
480                 struct rpc_cred *cred,
481                 gfp_t gfp_flags)
482 {
483         struct nfs4_state_owner *sp;
484
485         sp = kzalloc(sizeof(*sp), gfp_flags);
486         if (!sp)
487                 return NULL;
488         sp->so_server = server;
489         sp->so_cred = get_rpccred(cred);
490         spin_lock_init(&sp->so_lock);
491         INIT_LIST_HEAD(&sp->so_states);
492         nfs4_init_seqid_counter(&sp->so_seqid);
493         atomic_set(&sp->so_count, 1);
494         INIT_LIST_HEAD(&sp->so_lru);
495         seqcount_init(&sp->so_reclaim_seqcount);
496         mutex_init(&sp->so_delegreturn_mutex);
497         return sp;
498 }
499
500 static void
501 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
502 {
503         struct rb_node *rb_node = &sp->so_server_node;
504
505         if (!RB_EMPTY_NODE(rb_node)) {
506                 struct nfs_server *server = sp->so_server;
507                 struct nfs_client *clp = server->nfs_client;
508
509                 spin_lock(&clp->cl_lock);
510                 if (!RB_EMPTY_NODE(rb_node)) {
511                         rb_erase(rb_node, &server->state_owners);
512                         RB_CLEAR_NODE(rb_node);
513                 }
514                 spin_unlock(&clp->cl_lock);
515         }
516 }
517
518 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
519 {
520         nfs4_destroy_seqid_counter(&sp->so_seqid);
521         put_rpccred(sp->so_cred);
522         kfree(sp);
523 }
524
525 static void nfs4_gc_state_owners(struct nfs_server *server)
526 {
527         struct nfs_client *clp = server->nfs_client;
528         struct nfs4_state_owner *sp, *tmp;
529         unsigned long time_min, time_max;
530         LIST_HEAD(doomed);
531
532         spin_lock(&clp->cl_lock);
533         time_max = jiffies;
534         time_min = (long)time_max - (long)clp->cl_lease_time;
535         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
536                 /* NB: LRU is sorted so that oldest is at the head */
537                 if (time_in_range(sp->so_expires, time_min, time_max))
538                         break;
539                 list_move(&sp->so_lru, &doomed);
540                 nfs4_remove_state_owner_locked(sp);
541         }
542         spin_unlock(&clp->cl_lock);
543
544         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
545                 list_del(&sp->so_lru);
546                 nfs4_free_state_owner(sp);
547         }
548 }
549
550 /**
551  * nfs4_get_state_owner - Look up a state owner given a credential
552  * @server: nfs_server to search
553  * @cred: RPC credential to match
554  *
555  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
556  */
557 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
558                                               struct rpc_cred *cred,
559                                               gfp_t gfp_flags)
560 {
561         struct nfs_client *clp = server->nfs_client;
562         struct nfs4_state_owner *sp, *new;
563
564         spin_lock(&clp->cl_lock);
565         sp = nfs4_find_state_owner_locked(server, cred);
566         spin_unlock(&clp->cl_lock);
567         if (sp != NULL)
568                 goto out;
569         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
570         if (new == NULL)
571                 goto out;
572         do {
573                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
574                         break;
575                 spin_lock(&clp->cl_lock);
576                 sp = nfs4_insert_state_owner_locked(new);
577                 spin_unlock(&clp->cl_lock);
578         } while (sp == ERR_PTR(-EAGAIN));
579         if (sp != new)
580                 nfs4_free_state_owner(new);
581 out:
582         nfs4_gc_state_owners(server);
583         return sp;
584 }
585
586 /**
587  * nfs4_put_state_owner - Release a nfs4_state_owner
588  * @sp: state owner data to release
589  *
590  * Note that we keep released state owners on an LRU
591  * list.
592  * This caches valid state owners so that they can be
593  * reused, to avoid the OPEN_CONFIRM on minor version 0.
594  * It also pins the uniquifier of dropped state owners for
595  * a while, to ensure that those state owner names are
596  * never reused.
597  */
598 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
599 {
600         struct nfs_server *server = sp->so_server;
601         struct nfs_client *clp = server->nfs_client;
602
603         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
604                 return;
605
606         sp->so_expires = jiffies;
607         list_add_tail(&sp->so_lru, &server->state_owners_lru);
608         spin_unlock(&clp->cl_lock);
609 }
610
611 /**
612  * nfs4_purge_state_owners - Release all cached state owners
613  * @server: nfs_server with cached state owners to release
614  * @head: resulting list of state owners
615  *
616  * Called at umount time.  Remaining state owners will be on
617  * the LRU with ref count of zero.
618  * Note that the state owners are not freed, but are added
619  * to the list @head, which can later be used as an argument
620  * to nfs4_free_state_owners.
621  */
622 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
623 {
624         struct nfs_client *clp = server->nfs_client;
625         struct nfs4_state_owner *sp, *tmp;
626
627         spin_lock(&clp->cl_lock);
628         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
629                 list_move(&sp->so_lru, head);
630                 nfs4_remove_state_owner_locked(sp);
631         }
632         spin_unlock(&clp->cl_lock);
633 }
634
635 /**
636  * nfs4_purge_state_owners - Release all cached state owners
637  * @head: resulting list of state owners
638  *
639  * Frees a list of state owners that was generated by
640  * nfs4_purge_state_owners
641  */
642 void nfs4_free_state_owners(struct list_head *head)
643 {
644         struct nfs4_state_owner *sp, *tmp;
645
646         list_for_each_entry_safe(sp, tmp, head, so_lru) {
647                 list_del(&sp->so_lru);
648                 nfs4_free_state_owner(sp);
649         }
650 }
651
652 static struct nfs4_state *
653 nfs4_alloc_open_state(void)
654 {
655         struct nfs4_state *state;
656
657         state = kzalloc(sizeof(*state), GFP_NOFS);
658         if (!state)
659                 return NULL;
660         atomic_set(&state->count, 1);
661         INIT_LIST_HEAD(&state->lock_states);
662         spin_lock_init(&state->state_lock);
663         seqlock_init(&state->seqlock);
664         return state;
665 }
666
667 void
668 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
669 {
670         if (state->state == fmode)
671                 return;
672         /* NB! List reordering - see the reclaim code for why.  */
673         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
674                 if (fmode & FMODE_WRITE)
675                         list_move(&state->open_states, &state->owner->so_states);
676                 else
677                         list_move_tail(&state->open_states, &state->owner->so_states);
678         }
679         state->state = fmode;
680 }
681
682 static struct nfs4_state *
683 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
684 {
685         struct nfs_inode *nfsi = NFS_I(inode);
686         struct nfs4_state *state;
687
688         list_for_each_entry(state, &nfsi->open_states, inode_states) {
689                 if (state->owner != owner)
690                         continue;
691                 if (!nfs4_valid_open_stateid(state))
692                         continue;
693                 if (atomic_inc_not_zero(&state->count))
694                         return state;
695         }
696         return NULL;
697 }
698
699 static void
700 nfs4_free_open_state(struct nfs4_state *state)
701 {
702         kfree(state);
703 }
704
705 struct nfs4_state *
706 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
707 {
708         struct nfs4_state *state, *new;
709         struct nfs_inode *nfsi = NFS_I(inode);
710
711         spin_lock(&inode->i_lock);
712         state = __nfs4_find_state_byowner(inode, owner);
713         spin_unlock(&inode->i_lock);
714         if (state)
715                 goto out;
716         new = nfs4_alloc_open_state();
717         spin_lock(&owner->so_lock);
718         spin_lock(&inode->i_lock);
719         state = __nfs4_find_state_byowner(inode, owner);
720         if (state == NULL && new != NULL) {
721                 state = new;
722                 state->owner = owner;
723                 atomic_inc(&owner->so_count);
724                 list_add(&state->inode_states, &nfsi->open_states);
725                 ihold(inode);
726                 state->inode = inode;
727                 spin_unlock(&inode->i_lock);
728                 /* Note: The reclaim code dictates that we add stateless
729                  * and read-only stateids to the end of the list */
730                 list_add_tail(&state->open_states, &owner->so_states);
731                 spin_unlock(&owner->so_lock);
732         } else {
733                 spin_unlock(&inode->i_lock);
734                 spin_unlock(&owner->so_lock);
735                 if (new)
736                         nfs4_free_open_state(new);
737         }
738 out:
739         return state;
740 }
741
742 void nfs4_put_open_state(struct nfs4_state *state)
743 {
744         struct inode *inode = state->inode;
745         struct nfs4_state_owner *owner = state->owner;
746
747         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
748                 return;
749         spin_lock(&inode->i_lock);
750         list_del(&state->inode_states);
751         list_del(&state->open_states);
752         spin_unlock(&inode->i_lock);
753         spin_unlock(&owner->so_lock);
754         iput(inode);
755         nfs4_free_open_state(state);
756         nfs4_put_state_owner(owner);
757 }
758
759 /*
760  * Close the current file.
761  */
762 static void __nfs4_close(struct nfs4_state *state,
763                 fmode_t fmode, gfp_t gfp_mask, int wait)
764 {
765         struct nfs4_state_owner *owner = state->owner;
766         int call_close = 0;
767         fmode_t newstate;
768
769         atomic_inc(&owner->so_count);
770         /* Protect against nfs4_find_state() */
771         spin_lock(&owner->so_lock);
772         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
773                 case FMODE_READ:
774                         state->n_rdonly--;
775                         break;
776                 case FMODE_WRITE:
777                         state->n_wronly--;
778                         break;
779                 case FMODE_READ|FMODE_WRITE:
780                         state->n_rdwr--;
781         }
782         newstate = FMODE_READ|FMODE_WRITE;
783         if (state->n_rdwr == 0) {
784                 if (state->n_rdonly == 0) {
785                         newstate &= ~FMODE_READ;
786                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
787                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
788                 }
789                 if (state->n_wronly == 0) {
790                         newstate &= ~FMODE_WRITE;
791                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
792                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
793                 }
794                 if (newstate == 0)
795                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
796         }
797         nfs4_state_set_mode_locked(state, newstate);
798         spin_unlock(&owner->so_lock);
799
800         if (!call_close) {
801                 nfs4_put_open_state(state);
802                 nfs4_put_state_owner(owner);
803         } else
804                 nfs4_do_close(state, gfp_mask, wait);
805 }
806
807 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
808 {
809         __nfs4_close(state, fmode, GFP_NOFS, 0);
810 }
811
812 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
813 {
814         __nfs4_close(state, fmode, GFP_KERNEL, 1);
815 }
816
817 /*
818  * Search the state->lock_states for an existing lock_owner
819  * that is compatible with current->files
820  */
821 static struct nfs4_lock_state *
822 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
823 {
824         struct nfs4_lock_state *pos;
825         list_for_each_entry(pos, &state->lock_states, ls_locks) {
826                 if (pos->ls_owner != fl_owner)
827                         continue;
828                 atomic_inc(&pos->ls_count);
829                 return pos;
830         }
831         return NULL;
832 }
833
834 /*
835  * Return a compatible lock_state. If no initialized lock_state structure
836  * exists, return an uninitialized one.
837  *
838  */
839 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
840 {
841         struct nfs4_lock_state *lsp;
842         struct nfs_server *server = state->owner->so_server;
843
844         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
845         if (lsp == NULL)
846                 return NULL;
847         nfs4_init_seqid_counter(&lsp->ls_seqid);
848         atomic_set(&lsp->ls_count, 1);
849         lsp->ls_state = state;
850         lsp->ls_owner = fl_owner;
851         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
852         if (lsp->ls_seqid.owner_id < 0)
853                 goto out_free;
854         INIT_LIST_HEAD(&lsp->ls_locks);
855         return lsp;
856 out_free:
857         kfree(lsp);
858         return NULL;
859 }
860
861 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
862 {
863         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
864         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
865         kfree(lsp);
866 }
867
868 /*
869  * Return a compatible lock_state. If no initialized lock_state structure
870  * exists, return an uninitialized one.
871  *
872  */
873 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
874 {
875         struct nfs4_lock_state *lsp, *new = NULL;
876         
877         for(;;) {
878                 spin_lock(&state->state_lock);
879                 lsp = __nfs4_find_lock_state(state, owner);
880                 if (lsp != NULL)
881                         break;
882                 if (new != NULL) {
883                         list_add(&new->ls_locks, &state->lock_states);
884                         set_bit(LK_STATE_IN_USE, &state->flags);
885                         lsp = new;
886                         new = NULL;
887                         break;
888                 }
889                 spin_unlock(&state->state_lock);
890                 new = nfs4_alloc_lock_state(state, owner);
891                 if (new == NULL)
892                         return NULL;
893         }
894         spin_unlock(&state->state_lock);
895         if (new != NULL)
896                 nfs4_free_lock_state(state->owner->so_server, new);
897         return lsp;
898 }
899
900 /*
901  * Release reference to lock_state, and free it if we see that
902  * it is no longer in use
903  */
904 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
905 {
906         struct nfs_server *server;
907         struct nfs4_state *state;
908
909         if (lsp == NULL)
910                 return;
911         state = lsp->ls_state;
912         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
913                 return;
914         list_del(&lsp->ls_locks);
915         if (list_empty(&state->lock_states))
916                 clear_bit(LK_STATE_IN_USE, &state->flags);
917         spin_unlock(&state->state_lock);
918         server = state->owner->so_server;
919         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
920                 struct nfs_client *clp = server->nfs_client;
921
922                 clp->cl_mvops->free_lock_state(server, lsp);
923         } else
924                 nfs4_free_lock_state(server, lsp);
925 }
926
927 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
928 {
929         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
930
931         dst->fl_u.nfs4_fl.owner = lsp;
932         atomic_inc(&lsp->ls_count);
933 }
934
935 static void nfs4_fl_release_lock(struct file_lock *fl)
936 {
937         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
938 }
939
940 static const struct file_lock_operations nfs4_fl_lock_ops = {
941         .fl_copy_lock = nfs4_fl_copy_lock,
942         .fl_release_private = nfs4_fl_release_lock,
943 };
944
945 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
946 {
947         struct nfs4_lock_state *lsp;
948
949         if (fl->fl_ops != NULL)
950                 return 0;
951         lsp = nfs4_get_lock_state(state, fl->fl_owner);
952         if (lsp == NULL)
953                 return -ENOMEM;
954         fl->fl_u.nfs4_fl.owner = lsp;
955         fl->fl_ops = &nfs4_fl_lock_ops;
956         return 0;
957 }
958
959 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
960                 struct nfs4_state *state,
961                 const struct nfs_lockowner *lockowner)
962 {
963         struct nfs4_lock_state *lsp;
964         fl_owner_t fl_owner;
965         int ret = -ENOENT;
966
967
968         if (lockowner == NULL)
969                 goto out;
970
971         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
972                 goto out;
973
974         fl_owner = lockowner->l_owner;
975         spin_lock(&state->state_lock);
976         lsp = __nfs4_find_lock_state(state, fl_owner);
977         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
978                 ret = -EIO;
979         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
980                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
981                 ret = 0;
982         }
983         spin_unlock(&state->state_lock);
984         nfs4_put_lock_state(lsp);
985 out:
986         return ret;
987 }
988
989 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
990 {
991         const nfs4_stateid *src;
992         int seq;
993
994         do {
995                 src = &zero_stateid;
996                 seq = read_seqbegin(&state->seqlock);
997                 if (test_bit(NFS_OPEN_STATE, &state->flags))
998                         src = &state->open_stateid;
999                 nfs4_stateid_copy(dst, src);
1000         } while (read_seqretry(&state->seqlock, seq));
1001 }
1002
1003 /*
1004  * Byte-range lock aware utility to initialize the stateid of read/write
1005  * requests.
1006  */
1007 int nfs4_select_rw_stateid(struct nfs4_state *state,
1008                 fmode_t fmode, const struct nfs_lockowner *lockowner,
1009                 nfs4_stateid *dst, struct rpc_cred **cred)
1010 {
1011         int ret;
1012
1013         if (!nfs4_valid_open_stateid(state))
1014                 return -EIO;
1015         if (cred != NULL)
1016                 *cred = NULL;
1017         ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1018         if (ret == -EIO)
1019                 /* A lost lock - don't even consider delegations */
1020                 goto out;
1021         /* returns true if delegation stateid found and copied */
1022         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1023                 ret = 0;
1024                 goto out;
1025         }
1026         if (ret != -ENOENT)
1027                 /* nfs4_copy_delegation_stateid() didn't over-write
1028                  * dst, so it still has the lock stateid which we now
1029                  * choose to use.
1030                  */
1031                 goto out;
1032         nfs4_copy_open_stateid(dst, state);
1033         ret = 0;
1034 out:
1035         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1036                 dst->seqid = 0;
1037         return ret;
1038 }
1039
1040 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1041 {
1042         struct nfs_seqid *new;
1043
1044         new = kmalloc(sizeof(*new), gfp_mask);
1045         if (new == NULL)
1046                 return ERR_PTR(-ENOMEM);
1047         new->sequence = counter;
1048         INIT_LIST_HEAD(&new->list);
1049         new->task = NULL;
1050         return new;
1051 }
1052
1053 void nfs_release_seqid(struct nfs_seqid *seqid)
1054 {
1055         struct nfs_seqid_counter *sequence;
1056
1057         if (seqid == NULL || list_empty(&seqid->list))
1058                 return;
1059         sequence = seqid->sequence;
1060         spin_lock(&sequence->lock);
1061         list_del_init(&seqid->list);
1062         if (!list_empty(&sequence->list)) {
1063                 struct nfs_seqid *next;
1064
1065                 next = list_first_entry(&sequence->list,
1066                                 struct nfs_seqid, list);
1067                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1068         }
1069         spin_unlock(&sequence->lock);
1070 }
1071
1072 void nfs_free_seqid(struct nfs_seqid *seqid)
1073 {
1074         nfs_release_seqid(seqid);
1075         kfree(seqid);
1076 }
1077
1078 /*
1079  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1080  * failed with a seqid incrementing error -
1081  * see comments nfs4.h:seqid_mutating_error()
1082  */
1083 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1084 {
1085         switch (status) {
1086                 case 0:
1087                         break;
1088                 case -NFS4ERR_BAD_SEQID:
1089                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1090                                 return;
1091                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1092                                         " sequence-id error on an"
1093                                         " unconfirmed sequence %p!\n",
1094                                         seqid->sequence);
1095                 case -NFS4ERR_STALE_CLIENTID:
1096                 case -NFS4ERR_STALE_STATEID:
1097                 case -NFS4ERR_BAD_STATEID:
1098                 case -NFS4ERR_BADXDR:
1099                 case -NFS4ERR_RESOURCE:
1100                 case -NFS4ERR_NOFILEHANDLE:
1101                 case -NFS4ERR_MOVED:
1102                         /* Non-seqid mutating errors */
1103                         return;
1104         };
1105         /*
1106          * Note: no locking needed as we are guaranteed to be first
1107          * on the sequence list
1108          */
1109         seqid->sequence->counter++;
1110 }
1111
1112 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1113 {
1114         struct nfs4_state_owner *sp;
1115
1116         if (seqid == NULL)
1117                 return;
1118
1119         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1120         if (status == -NFS4ERR_BAD_SEQID)
1121                 nfs4_drop_state_owner(sp);
1122         if (!nfs4_has_session(sp->so_server->nfs_client))
1123                 nfs_increment_seqid(status, seqid);
1124 }
1125
1126 /*
1127  * Increment the seqid if the LOCK/LOCKU succeeded, or
1128  * failed with a seqid incrementing error -
1129  * see comments nfs4.h:seqid_mutating_error()
1130  */
1131 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1132 {
1133         if (seqid != NULL)
1134                 nfs_increment_seqid(status, seqid);
1135 }
1136
1137 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1138 {
1139         struct nfs_seqid_counter *sequence;
1140         int status = 0;
1141
1142         if (seqid == NULL)
1143                 goto out;
1144         sequence = seqid->sequence;
1145         spin_lock(&sequence->lock);
1146         seqid->task = task;
1147         if (list_empty(&seqid->list))
1148                 list_add_tail(&seqid->list, &sequence->list);
1149         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1150                 goto unlock;
1151         rpc_sleep_on(&sequence->wait, task, NULL);
1152         status = -EAGAIN;
1153 unlock:
1154         spin_unlock(&sequence->lock);
1155 out:
1156         return status;
1157 }
1158
1159 static int nfs4_run_state_manager(void *);
1160
1161 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1162 {
1163         smp_mb__before_atomic();
1164         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1165         smp_mb__after_atomic();
1166         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1167         rpc_wake_up(&clp->cl_rpcwaitq);
1168 }
1169
1170 /*
1171  * Schedule the nfs_client asynchronous state management routine
1172  */
1173 void nfs4_schedule_state_manager(struct nfs_client *clp)
1174 {
1175         struct task_struct *task;
1176         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1177
1178         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1179                 return;
1180         __module_get(THIS_MODULE);
1181         atomic_inc(&clp->cl_count);
1182
1183         /* The rcu_read_lock() is not strictly necessary, as the state
1184          * manager is the only thread that ever changes the rpc_xprt
1185          * after it's initialized.  At this point, we're single threaded. */
1186         rcu_read_lock();
1187         snprintf(buf, sizeof(buf), "%s-manager",
1188                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1189         rcu_read_unlock();
1190         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1191         if (IS_ERR(task)) {
1192                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1193                         __func__, PTR_ERR(task));
1194                 nfs4_clear_state_manager_bit(clp);
1195                 nfs_put_client(clp);
1196                 module_put(THIS_MODULE);
1197         }
1198 }
1199
1200 /*
1201  * Schedule a lease recovery attempt
1202  */
1203 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1204 {
1205         if (!clp)
1206                 return;
1207         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1208                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1209         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1210                         clp->cl_hostname);
1211         nfs4_schedule_state_manager(clp);
1212 }
1213 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1214
1215 /**
1216  * nfs4_schedule_migration_recovery - trigger migration recovery
1217  *
1218  * @server: FSID that is migrating
1219  *
1220  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1221  * value is returned.
1222  */
1223 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1224 {
1225         struct nfs_client *clp = server->nfs_client;
1226
1227         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1228                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1229                                 clp->cl_hostname);
1230                 return -NFS4ERR_IO;
1231         }
1232
1233         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1234                 return -NFS4ERR_IO;
1235
1236         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1237                         __func__,
1238                         (unsigned long long)server->fsid.major,
1239                         (unsigned long long)server->fsid.minor,
1240                         clp->cl_hostname);
1241
1242         set_bit(NFS_MIG_IN_TRANSITION,
1243                         &((struct nfs_server *)server)->mig_status);
1244         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1245
1246         nfs4_schedule_state_manager(clp);
1247         return 0;
1248 }
1249 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1250
1251 /**
1252  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1253  *
1254  * @clp: server to check for moved leases
1255  *
1256  */
1257 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1258 {
1259         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1260                 __func__, clp->cl_clientid, clp->cl_hostname);
1261
1262         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1263         nfs4_schedule_state_manager(clp);
1264 }
1265 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1266
1267 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1268 {
1269         int res;
1270
1271         might_sleep();
1272
1273         atomic_inc(&clp->cl_count);
1274         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1275                                  nfs_wait_bit_killable, TASK_KILLABLE);
1276         if (res)
1277                 goto out;
1278         if (clp->cl_cons_state < 0)
1279                 res = clp->cl_cons_state;
1280 out:
1281         nfs_put_client(clp);
1282         return res;
1283 }
1284
1285 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1286 {
1287         unsigned int loop;
1288         int ret;
1289
1290         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1291                 ret = nfs4_wait_clnt_recover(clp);
1292                 if (ret != 0)
1293                         break;
1294                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1295                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1296                         break;
1297                 nfs4_schedule_state_manager(clp);
1298                 ret = -EIO;
1299         }
1300         return ret;
1301 }
1302
1303 /*
1304  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1305  * @clp: client to process
1306  *
1307  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1308  * resend of the SETCLIENTID and hence re-establish the
1309  * callback channel. Then return all existing delegations.
1310  */
1311 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1312 {
1313         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1314         nfs_expire_all_delegations(clp);
1315         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1316                         clp->cl_hostname);
1317 }
1318
1319 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1320 {
1321         nfs40_handle_cb_pathdown(clp);
1322         nfs4_schedule_state_manager(clp);
1323 }
1324
1325 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1326 {
1327
1328         if (!nfs4_valid_open_stateid(state))
1329                 return 0;
1330         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1331         /* Don't recover state that expired before the reboot */
1332         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1333                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1334                 return 0;
1335         }
1336         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1337         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1338         return 1;
1339 }
1340
1341 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1342 {
1343         if (!nfs4_valid_open_stateid(state))
1344                 return 0;
1345         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1346         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1347         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1348         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1349         return 1;
1350 }
1351
1352 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1353 {
1354         struct nfs_client *clp = server->nfs_client;
1355
1356         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1357                 return -EBADF;
1358         nfs_inode_find_delegation_state_and_recover(state->inode,
1359                         &state->stateid);
1360         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1361                         clp->cl_hostname);
1362         nfs4_schedule_state_manager(clp);
1363         return 0;
1364 }
1365 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1366
1367 static struct nfs4_lock_state *
1368 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1369                 const nfs4_stateid *stateid)
1370 {
1371         struct nfs4_lock_state *pos;
1372
1373         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1374                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1375                         continue;
1376                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1377                         return pos;
1378         }
1379         return NULL;
1380 }
1381
1382 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1383                 const nfs4_stateid *stateid)
1384 {
1385         bool found = false;
1386
1387         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1388                 spin_lock(&state->state_lock);
1389                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1390                         found = true;
1391                 spin_unlock(&state->state_lock);
1392         }
1393         return found;
1394 }
1395
1396 void nfs_inode_find_state_and_recover(struct inode *inode,
1397                 const nfs4_stateid *stateid)
1398 {
1399         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1400         struct nfs_inode *nfsi = NFS_I(inode);
1401         struct nfs_open_context *ctx;
1402         struct nfs4_state *state;
1403         bool found = false;
1404
1405         spin_lock(&inode->i_lock);
1406         list_for_each_entry(ctx, &nfsi->open_files, list) {
1407                 state = ctx->state;
1408                 if (state == NULL)
1409                         continue;
1410                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1411                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1412                         found = true;
1413                         continue;
1414                 }
1415                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1416                     nfs4_state_mark_reclaim_nograce(clp, state))
1417                         found = true;
1418         }
1419         spin_unlock(&inode->i_lock);
1420
1421         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1422         if (found)
1423                 nfs4_schedule_state_manager(clp);
1424 }
1425
1426 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1427 {
1428         struct inode *inode = state->inode;
1429         struct nfs_inode *nfsi = NFS_I(inode);
1430         struct nfs_open_context *ctx;
1431
1432         spin_lock(&inode->i_lock);
1433         list_for_each_entry(ctx, &nfsi->open_files, list) {
1434                 if (ctx->state != state)
1435                         continue;
1436                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1437         }
1438         spin_unlock(&inode->i_lock);
1439 }
1440
1441 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1442 {
1443         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1444         nfs4_state_mark_open_context_bad(state);
1445 }
1446
1447
1448 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1449 {
1450         struct inode *inode = state->inode;
1451         struct nfs_inode *nfsi = NFS_I(inode);
1452         struct file_lock *fl;
1453         struct nfs4_lock_state *lsp;
1454         int status = 0;
1455         struct file_lock_context *flctx = inode->i_flctx;
1456         struct list_head *list;
1457
1458         if (flctx == NULL)
1459                 return 0;
1460
1461         list = &flctx->flc_posix;
1462
1463         /* Guard against delegation returns and new lock/unlock calls */
1464         down_write(&nfsi->rwsem);
1465         spin_lock(&flctx->flc_lock);
1466 restart:
1467         list_for_each_entry(fl, list, fl_list) {
1468                 if (nfs_file_open_context(fl->fl_file)->state != state)
1469                         continue;
1470                 spin_unlock(&flctx->flc_lock);
1471                 status = ops->recover_lock(state, fl);
1472                 switch (status) {
1473                 case 0:
1474                         break;
1475                 case -ESTALE:
1476                 case -NFS4ERR_ADMIN_REVOKED:
1477                 case -NFS4ERR_STALE_STATEID:
1478                 case -NFS4ERR_BAD_STATEID:
1479                 case -NFS4ERR_EXPIRED:
1480                 case -NFS4ERR_NO_GRACE:
1481                 case -NFS4ERR_STALE_CLIENTID:
1482                 case -NFS4ERR_BADSESSION:
1483                 case -NFS4ERR_BADSLOT:
1484                 case -NFS4ERR_BAD_HIGH_SLOT:
1485                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1486                         goto out;
1487                 default:
1488                         pr_err("NFS: %s: unhandled error %d\n",
1489                                         __func__, status);
1490                 case -ENOMEM:
1491                 case -NFS4ERR_DENIED:
1492                 case -NFS4ERR_RECLAIM_BAD:
1493                 case -NFS4ERR_RECLAIM_CONFLICT:
1494                         lsp = fl->fl_u.nfs4_fl.owner;
1495                         if (lsp)
1496                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1497                         status = 0;
1498                 }
1499                 spin_lock(&flctx->flc_lock);
1500         }
1501         if (list == &flctx->flc_posix) {
1502                 list = &flctx->flc_flock;
1503                 goto restart;
1504         }
1505         spin_unlock(&flctx->flc_lock);
1506 out:
1507         up_write(&nfsi->rwsem);
1508         return status;
1509 }
1510
1511 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1512 {
1513         struct nfs4_state *state;
1514         struct nfs4_lock_state *lock;
1515         int status = 0;
1516
1517         /* Note: we rely on the sp->so_states list being ordered 
1518          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1519          * states first.
1520          * This is needed to ensure that the server won't give us any
1521          * read delegations that we have to return if, say, we are
1522          * recovering after a network partition or a reboot from a
1523          * server that doesn't support a grace period.
1524          */
1525         spin_lock(&sp->so_lock);
1526         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1527 restart:
1528         list_for_each_entry(state, &sp->so_states, open_states) {
1529                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1530                         continue;
1531                 if (!nfs4_valid_open_stateid(state))
1532                         continue;
1533                 if (state->state == 0)
1534                         continue;
1535                 atomic_inc(&state->count);
1536                 spin_unlock(&sp->so_lock);
1537                 status = ops->recover_open(sp, state);
1538                 if (status >= 0) {
1539                         status = nfs4_reclaim_locks(state, ops);
1540                         if (status >= 0) {
1541                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1542                                         spin_lock(&state->state_lock);
1543                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1544                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1545                                                         pr_warn_ratelimited("NFS: "
1546                                                                             "%s: Lock reclaim "
1547                                                                             "failed!\n", __func__);
1548                                         }
1549                                         spin_unlock(&state->state_lock);
1550                                 }
1551                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1552                                         &state->flags);
1553                                 nfs4_put_open_state(state);
1554                                 spin_lock(&sp->so_lock);
1555                                 goto restart;
1556                         }
1557                 }
1558                 switch (status) {
1559                         default:
1560                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1561                                         __func__, status);
1562                         case -ENOENT:
1563                         case -ENOMEM:
1564                         case -EACCES:
1565                         case -EROFS:
1566                         case -EIO:
1567                         case -ESTALE:
1568                                 /* Open state on this file cannot be recovered */
1569                                 nfs4_state_mark_recovery_failed(state, status);
1570                                 break;
1571                         case -EAGAIN:
1572                                 ssleep(1);
1573                         case -NFS4ERR_ADMIN_REVOKED:
1574                         case -NFS4ERR_STALE_STATEID:
1575                         case -NFS4ERR_OLD_STATEID:
1576                         case -NFS4ERR_BAD_STATEID:
1577                         case -NFS4ERR_RECLAIM_BAD:
1578                         case -NFS4ERR_RECLAIM_CONFLICT:
1579                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1580                                 break;
1581                         case -NFS4ERR_EXPIRED:
1582                         case -NFS4ERR_NO_GRACE:
1583                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1584                         case -NFS4ERR_STALE_CLIENTID:
1585                         case -NFS4ERR_BADSESSION:
1586                         case -NFS4ERR_BADSLOT:
1587                         case -NFS4ERR_BAD_HIGH_SLOT:
1588                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1589                                 goto out_err;
1590                 }
1591                 nfs4_put_open_state(state);
1592                 spin_lock(&sp->so_lock);
1593                 goto restart;
1594         }
1595         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1596         spin_unlock(&sp->so_lock);
1597         return 0;
1598 out_err:
1599         nfs4_put_open_state(state);
1600         spin_lock(&sp->so_lock);
1601         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1602         spin_unlock(&sp->so_lock);
1603         return status;
1604 }
1605
1606 static void nfs4_clear_open_state(struct nfs4_state *state)
1607 {
1608         struct nfs4_lock_state *lock;
1609
1610         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1611         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1612         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1613         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1614         spin_lock(&state->state_lock);
1615         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1616                 lock->ls_seqid.flags = 0;
1617                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1618         }
1619         spin_unlock(&state->state_lock);
1620 }
1621
1622 static void nfs4_reset_seqids(struct nfs_server *server,
1623         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1624 {
1625         struct nfs_client *clp = server->nfs_client;
1626         struct nfs4_state_owner *sp;
1627         struct rb_node *pos;
1628         struct nfs4_state *state;
1629
1630         spin_lock(&clp->cl_lock);
1631         for (pos = rb_first(&server->state_owners);
1632              pos != NULL;
1633              pos = rb_next(pos)) {
1634                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1635                 sp->so_seqid.flags = 0;
1636                 spin_lock(&sp->so_lock);
1637                 list_for_each_entry(state, &sp->so_states, open_states) {
1638                         if (mark_reclaim(clp, state))
1639                                 nfs4_clear_open_state(state);
1640                 }
1641                 spin_unlock(&sp->so_lock);
1642         }
1643         spin_unlock(&clp->cl_lock);
1644 }
1645
1646 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1647         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1648 {
1649         struct nfs_server *server;
1650
1651         rcu_read_lock();
1652         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1653                 nfs4_reset_seqids(server, mark_reclaim);
1654         rcu_read_unlock();
1655 }
1656
1657 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1658 {
1659         /* Mark all delegations for reclaim */
1660         nfs_delegation_mark_reclaim(clp);
1661         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1662 }
1663
1664 static int nfs4_reclaim_complete(struct nfs_client *clp,
1665                                  const struct nfs4_state_recovery_ops *ops,
1666                                  struct rpc_cred *cred)
1667 {
1668         /* Notify the server we're done reclaiming our state */
1669         if (ops->reclaim_complete)
1670                 return ops->reclaim_complete(clp, cred);
1671         return 0;
1672 }
1673
1674 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1675 {
1676         struct nfs_client *clp = server->nfs_client;
1677         struct nfs4_state_owner *sp;
1678         struct rb_node *pos;
1679         struct nfs4_state *state;
1680
1681         spin_lock(&clp->cl_lock);
1682         for (pos = rb_first(&server->state_owners);
1683              pos != NULL;
1684              pos = rb_next(pos)) {
1685                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1686                 spin_lock(&sp->so_lock);
1687                 list_for_each_entry(state, &sp->so_states, open_states) {
1688                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1689                                                 &state->flags))
1690                                 continue;
1691                         nfs4_state_mark_reclaim_nograce(clp, state);
1692                 }
1693                 spin_unlock(&sp->so_lock);
1694         }
1695         spin_unlock(&clp->cl_lock);
1696 }
1697
1698 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1699 {
1700         struct nfs_server *server;
1701
1702         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1703                 return 0;
1704
1705         rcu_read_lock();
1706         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1707                 nfs4_clear_reclaim_server(server);
1708         rcu_read_unlock();
1709
1710         nfs_delegation_reap_unclaimed(clp);
1711         return 1;
1712 }
1713
1714 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1715 {
1716         const struct nfs4_state_recovery_ops *ops;
1717         struct rpc_cred *cred;
1718         int err;
1719
1720         if (!nfs4_state_clear_reclaim_reboot(clp))
1721                 return;
1722         ops = clp->cl_mvops->reboot_recovery_ops;
1723         cred = nfs4_get_clid_cred(clp);
1724         err = nfs4_reclaim_complete(clp, ops, cred);
1725         put_rpccred(cred);
1726         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1727                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1728 }
1729
1730 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1731 {
1732         nfs_mark_test_expired_all_delegations(clp);
1733         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1734 }
1735
1736 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1737 {
1738         switch (error) {
1739                 case 0:
1740                         break;
1741                 case -NFS4ERR_CB_PATH_DOWN:
1742                         nfs40_handle_cb_pathdown(clp);
1743                         break;
1744                 case -NFS4ERR_NO_GRACE:
1745                         nfs4_state_end_reclaim_reboot(clp);
1746                         break;
1747                 case -NFS4ERR_STALE_CLIENTID:
1748                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1749                         nfs4_state_start_reclaim_reboot(clp);
1750                         break;
1751                 case -NFS4ERR_EXPIRED:
1752                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1753                         nfs4_state_start_reclaim_nograce(clp);
1754                         break;
1755                 case -NFS4ERR_BADSESSION:
1756                 case -NFS4ERR_BADSLOT:
1757                 case -NFS4ERR_BAD_HIGH_SLOT:
1758                 case -NFS4ERR_DEADSESSION:
1759                 case -NFS4ERR_SEQ_FALSE_RETRY:
1760                 case -NFS4ERR_SEQ_MISORDERED:
1761                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1762                         /* Zero session reset errors */
1763                         break;
1764                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1765                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1766                         break;
1767                 default:
1768                         dprintk("%s: failed to handle error %d for server %s\n",
1769                                         __func__, error, clp->cl_hostname);
1770                         return error;
1771         }
1772         dprintk("%s: handled error %d for server %s\n", __func__, error,
1773                         clp->cl_hostname);
1774         return 0;
1775 }
1776
1777 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1778 {
1779         struct nfs4_state_owner *sp;
1780         struct nfs_server *server;
1781         struct rb_node *pos;
1782         LIST_HEAD(freeme);
1783         int status = 0;
1784
1785 restart:
1786         rcu_read_lock();
1787         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1788                 nfs4_purge_state_owners(server, &freeme);
1789                 spin_lock(&clp->cl_lock);
1790                 for (pos = rb_first(&server->state_owners);
1791                      pos != NULL;
1792                      pos = rb_next(pos)) {
1793                         sp = rb_entry(pos,
1794                                 struct nfs4_state_owner, so_server_node);
1795                         if (!test_and_clear_bit(ops->owner_flag_bit,
1796                                                         &sp->so_flags))
1797                                 continue;
1798                         if (!atomic_inc_not_zero(&sp->so_count))
1799                                 continue;
1800                         spin_unlock(&clp->cl_lock);
1801                         rcu_read_unlock();
1802
1803                         status = nfs4_reclaim_open_state(sp, ops);
1804                         if (status < 0) {
1805                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1806                                 nfs4_put_state_owner(sp);
1807                                 status = nfs4_recovery_handle_error(clp, status);
1808                                 return (status != 0) ? status : -EAGAIN;
1809                         }
1810
1811                         nfs4_put_state_owner(sp);
1812                         goto restart;
1813                 }
1814                 spin_unlock(&clp->cl_lock);
1815         }
1816         rcu_read_unlock();
1817         nfs4_free_state_owners(&freeme);
1818         return 0;
1819 }
1820
1821 static int nfs4_check_lease(struct nfs_client *clp)
1822 {
1823         struct rpc_cred *cred;
1824         const struct nfs4_state_maintenance_ops *ops =
1825                 clp->cl_mvops->state_renewal_ops;
1826         int status;
1827
1828         /* Is the client already known to have an expired lease? */
1829         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1830                 return 0;
1831         spin_lock(&clp->cl_lock);
1832         cred = ops->get_state_renewal_cred_locked(clp);
1833         spin_unlock(&clp->cl_lock);
1834         if (cred == NULL) {
1835                 cred = nfs4_get_clid_cred(clp);
1836                 status = -ENOKEY;
1837                 if (cred == NULL)
1838                         goto out;
1839         }
1840         status = ops->renew_lease(clp, cred);
1841         put_rpccred(cred);
1842         if (status == -ETIMEDOUT) {
1843                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1844                 return 0;
1845         }
1846 out:
1847         return nfs4_recovery_handle_error(clp, status);
1848 }
1849
1850 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1851  * and for recoverable errors on EXCHANGE_ID for v4.1
1852  */
1853 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1854 {
1855         switch (status) {
1856         case -NFS4ERR_SEQ_MISORDERED:
1857                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1858                         return -ESERVERFAULT;
1859                 /* Lease confirmation error: retry after purging the lease */
1860                 ssleep(1);
1861                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1862                 break;
1863         case -NFS4ERR_STALE_CLIENTID:
1864                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1865                 nfs4_state_start_reclaim_reboot(clp);
1866                 break;
1867         case -NFS4ERR_CLID_INUSE:
1868                 pr_err("NFS: Server %s reports our clientid is in use\n",
1869                         clp->cl_hostname);
1870                 nfs_mark_client_ready(clp, -EPERM);
1871                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1872                 return -EPERM;
1873         case -EACCES:
1874         case -NFS4ERR_DELAY:
1875         case -ETIMEDOUT:
1876         case -EAGAIN:
1877                 ssleep(1);
1878                 break;
1879
1880         case -NFS4ERR_MINOR_VERS_MISMATCH:
1881                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1882                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1883                 dprintk("%s: exit with error %d for server %s\n",
1884                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1885                 return -EPROTONOSUPPORT;
1886         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1887                                  * in nfs4_exchange_id */
1888         default:
1889                 dprintk("%s: exit with error %d for server %s\n", __func__,
1890                                 status, clp->cl_hostname);
1891                 return status;
1892         }
1893         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1894         dprintk("%s: handled error %d for server %s\n", __func__, status,
1895                         clp->cl_hostname);
1896         return 0;
1897 }
1898
1899 static int nfs4_establish_lease(struct nfs_client *clp)
1900 {
1901         struct rpc_cred *cred;
1902         const struct nfs4_state_recovery_ops *ops =
1903                 clp->cl_mvops->reboot_recovery_ops;
1904         int status;
1905
1906         nfs4_begin_drain_session(clp);
1907         cred = nfs4_get_clid_cred(clp);
1908         if (cred == NULL)
1909                 return -ENOENT;
1910         status = ops->establish_clid(clp, cred);
1911         put_rpccred(cred);
1912         if (status != 0)
1913                 return status;
1914         pnfs_destroy_all_layouts(clp);
1915         return 0;
1916 }
1917
1918 /*
1919  * Returns zero or a negative errno.  NFS4ERR values are converted
1920  * to local errno values.
1921  */
1922 static int nfs4_reclaim_lease(struct nfs_client *clp)
1923 {
1924         int status;
1925
1926         status = nfs4_establish_lease(clp);
1927         if (status < 0)
1928                 return nfs4_handle_reclaim_lease_error(clp, status);
1929         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1930                 nfs4_state_start_reclaim_nograce(clp);
1931         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1932                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1933         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1934         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1935         return 0;
1936 }
1937
1938 static int nfs4_purge_lease(struct nfs_client *clp)
1939 {
1940         int status;
1941
1942         status = nfs4_establish_lease(clp);
1943         if (status < 0)
1944                 return nfs4_handle_reclaim_lease_error(clp, status);
1945         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1946         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1947         nfs4_state_start_reclaim_nograce(clp);
1948         return 0;
1949 }
1950
1951 /*
1952  * Try remote migration of one FSID from a source server to a
1953  * destination server.  The source server provides a list of
1954  * potential destinations.
1955  *
1956  * Returns zero or a negative NFS4ERR status code.
1957  */
1958 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1959 {
1960         struct nfs_client *clp = server->nfs_client;
1961         struct nfs4_fs_locations *locations = NULL;
1962         struct inode *inode;
1963         struct page *page;
1964         int status, result;
1965
1966         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1967                         (unsigned long long)server->fsid.major,
1968                         (unsigned long long)server->fsid.minor,
1969                         clp->cl_hostname);
1970
1971         result = 0;
1972         page = alloc_page(GFP_KERNEL);
1973         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1974         if (page == NULL || locations == NULL) {
1975                 dprintk("<-- %s: no memory\n", __func__);
1976                 goto out;
1977         }
1978
1979         inode = d_inode(server->super->s_root);
1980         result = nfs4_proc_get_locations(inode, locations, page, cred);
1981         if (result) {
1982                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1983                         __func__, result);
1984                 goto out;
1985         }
1986
1987         result = -NFS4ERR_NXIO;
1988         if (!locations->nlocations)
1989                 goto out;
1990
1991         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1992                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1993                         __func__);
1994                 goto out;
1995         }
1996
1997         nfs4_begin_drain_session(clp);
1998
1999         status = nfs4_replace_transport(server, locations);
2000         if (status != 0) {
2001                 dprintk("<-- %s: failed to replace transport: %d\n",
2002                         __func__, status);
2003                 goto out;
2004         }
2005
2006         result = 0;
2007         dprintk("<-- %s: migration succeeded\n", __func__);
2008
2009 out:
2010         if (page != NULL)
2011                 __free_page(page);
2012         kfree(locations);
2013         if (result) {
2014                 pr_err("NFS: migration recovery failed (server %s)\n",
2015                                 clp->cl_hostname);
2016                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2017         }
2018         return result;
2019 }
2020
2021 /*
2022  * Returns zero or a negative NFS4ERR status code.
2023  */
2024 static int nfs4_handle_migration(struct nfs_client *clp)
2025 {
2026         const struct nfs4_state_maintenance_ops *ops =
2027                                 clp->cl_mvops->state_renewal_ops;
2028         struct nfs_server *server;
2029         struct rpc_cred *cred;
2030
2031         dprintk("%s: migration reported on \"%s\"\n", __func__,
2032                         clp->cl_hostname);
2033
2034         spin_lock(&clp->cl_lock);
2035         cred = ops->get_state_renewal_cred_locked(clp);
2036         spin_unlock(&clp->cl_lock);
2037         if (cred == NULL)
2038                 return -NFS4ERR_NOENT;
2039
2040         clp->cl_mig_gen++;
2041 restart:
2042         rcu_read_lock();
2043         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2044                 int status;
2045
2046                 if (server->mig_gen == clp->cl_mig_gen)
2047                         continue;
2048                 server->mig_gen = clp->cl_mig_gen;
2049
2050                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2051                                                 &server->mig_status))
2052                         continue;
2053
2054                 rcu_read_unlock();
2055                 status = nfs4_try_migration(server, cred);
2056                 if (status < 0) {
2057                         put_rpccred(cred);
2058                         return status;
2059                 }
2060                 goto restart;
2061         }
2062         rcu_read_unlock();
2063         put_rpccred(cred);
2064         return 0;
2065 }
2066
2067 /*
2068  * Test each nfs_server on the clp's cl_superblocks list to see
2069  * if it's moved to another server.  Stop when the server no longer
2070  * returns NFS4ERR_LEASE_MOVED.
2071  */
2072 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2073 {
2074         const struct nfs4_state_maintenance_ops *ops =
2075                                 clp->cl_mvops->state_renewal_ops;
2076         struct nfs_server *server;
2077         struct rpc_cred *cred;
2078
2079         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2080                         clp->cl_hostname);
2081
2082         spin_lock(&clp->cl_lock);
2083         cred = ops->get_state_renewal_cred_locked(clp);
2084         spin_unlock(&clp->cl_lock);
2085         if (cred == NULL)
2086                 return -NFS4ERR_NOENT;
2087
2088         clp->cl_mig_gen++;
2089 restart:
2090         rcu_read_lock();
2091         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2092                 struct inode *inode;
2093                 int status;
2094
2095                 if (server->mig_gen == clp->cl_mig_gen)
2096                         continue;
2097                 server->mig_gen = clp->cl_mig_gen;
2098
2099                 rcu_read_unlock();
2100
2101                 inode = d_inode(server->super->s_root);
2102                 status = nfs4_proc_fsid_present(inode, cred);
2103                 if (status != -NFS4ERR_MOVED)
2104                         goto restart;   /* wasn't this one */
2105                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2106                         goto restart;   /* there are more */
2107                 goto out;
2108         }
2109         rcu_read_unlock();
2110
2111 out:
2112         put_rpccred(cred);
2113         return 0;
2114 }
2115
2116 /**
2117  * nfs4_discover_server_trunking - Detect server IP address trunking
2118  *
2119  * @clp: nfs_client under test
2120  * @result: OUT: found nfs_client, or clp
2121  *
2122  * Returns zero or a negative errno.  If zero is returned,
2123  * an nfs_client pointer is planted in "result".
2124  *
2125  * Note: since we are invoked in process context, and
2126  * not from inside the state manager, we cannot use
2127  * nfs4_handle_reclaim_lease_error().
2128  */
2129 int nfs4_discover_server_trunking(struct nfs_client *clp,
2130                                   struct nfs_client **result)
2131 {
2132         const struct nfs4_state_recovery_ops *ops =
2133                                 clp->cl_mvops->reboot_recovery_ops;
2134         struct rpc_clnt *clnt;
2135         struct rpc_cred *cred;
2136         int i, status;
2137
2138         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2139
2140         clnt = clp->cl_rpcclient;
2141         i = 0;
2142
2143         mutex_lock(&nfs_clid_init_mutex);
2144 again:
2145         status  = -ENOENT;
2146         cred = nfs4_get_clid_cred(clp);
2147         if (cred == NULL)
2148                 goto out_unlock;
2149
2150         status = ops->detect_trunking(clp, result, cred);
2151         put_rpccred(cred);
2152         switch (status) {
2153         case 0:
2154                 break;
2155         case -ETIMEDOUT:
2156                 if (clnt->cl_softrtry)
2157                         break;
2158         case -NFS4ERR_DELAY:
2159         case -EAGAIN:
2160                 ssleep(1);
2161         case -NFS4ERR_STALE_CLIENTID:
2162                 dprintk("NFS: %s after status %d, retrying\n",
2163                         __func__, status);
2164                 goto again;
2165         case -EACCES:
2166                 if (i++ == 0) {
2167                         nfs4_root_machine_cred(clp);
2168                         goto again;
2169                 }
2170                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2171                         break;
2172         case -NFS4ERR_CLID_INUSE:
2173         case -NFS4ERR_WRONGSEC:
2174                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2175                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2176                         status = -EPERM;
2177                         break;
2178                 }
2179                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2180                 if (IS_ERR(clnt)) {
2181                         status = PTR_ERR(clnt);
2182                         break;
2183                 }
2184                 /* Note: this is safe because we haven't yet marked the
2185                  * client as ready, so we are the only user of
2186                  * clp->cl_rpcclient
2187                  */
2188                 clnt = xchg(&clp->cl_rpcclient, clnt);
2189                 rpc_shutdown_client(clnt);
2190                 clnt = clp->cl_rpcclient;
2191                 goto again;
2192
2193         case -NFS4ERR_MINOR_VERS_MISMATCH:
2194                 status = -EPROTONOSUPPORT;
2195                 break;
2196
2197         case -EKEYEXPIRED:
2198         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2199                                  * in nfs4_exchange_id */
2200                 status = -EKEYEXPIRED;
2201                 break;
2202         default:
2203                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2204                                 __func__, status);
2205                 status = -EIO;
2206         }
2207
2208 out_unlock:
2209         mutex_unlock(&nfs_clid_init_mutex);
2210         dprintk("NFS: %s: status = %d\n", __func__, status);
2211         return status;
2212 }
2213
2214 #ifdef CONFIG_NFS_V4_1
2215 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2216 {
2217         struct nfs_client *clp = session->clp;
2218
2219         switch (err) {
2220         default:
2221                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2222                 break;
2223         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2224                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2225         }
2226         nfs4_schedule_lease_recovery(clp);
2227 }
2228 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2229
2230 void nfs41_notify_server(struct nfs_client *clp)
2231 {
2232         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2233         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2234         nfs4_schedule_state_manager(clp);
2235 }
2236
2237 static void nfs4_reset_all_state(struct nfs_client *clp)
2238 {
2239         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2240                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2241                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2242                 nfs4_state_start_reclaim_nograce(clp);
2243                 dprintk("%s: scheduling reset of all state for server %s!\n",
2244                                 __func__, clp->cl_hostname);
2245                 nfs4_schedule_state_manager(clp);
2246         }
2247 }
2248
2249 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2250 {
2251         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2252                 nfs4_state_start_reclaim_reboot(clp);
2253                 dprintk("%s: server %s rebooted!\n", __func__,
2254                                 clp->cl_hostname);
2255                 nfs4_schedule_state_manager(clp);
2256         }
2257 }
2258
2259 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2260 {
2261         nfs4_reset_all_state(clp);
2262         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2263 }
2264
2265 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2266 {
2267         nfs4_state_start_reclaim_nograce(clp);
2268         nfs4_schedule_state_manager(clp);
2269
2270         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2271 }
2272
2273 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2274 {
2275         /* FIXME: For now, we destroy all layouts. */
2276         pnfs_destroy_all_layouts(clp);
2277         /* FIXME: For now, we test all delegations+open state+locks. */
2278         nfs41_handle_some_state_revoked(clp);
2279         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2280                         clp->cl_hostname);
2281 }
2282
2283 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2284 {
2285         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2286         nfs4_schedule_state_manager(clp);
2287
2288         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2289                         clp->cl_hostname);
2290 }
2291
2292 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2293 {
2294         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2295                 &clp->cl_state) == 0)
2296                 nfs4_schedule_state_manager(clp);
2297 }
2298
2299 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2300                 bool recovery)
2301 {
2302         if (!flags)
2303                 return;
2304
2305         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2306                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2307         /*
2308          * If we're called from the state manager thread, then assume we're
2309          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2310          * Those flags are expected to remain set until we're done
2311          * recovering (see RFC5661, section 18.46.3).
2312          */
2313         if (recovery)
2314                 goto out_recovery;
2315
2316         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2317                 nfs41_handle_server_reboot(clp);
2318         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2319                 nfs41_handle_all_state_revoked(clp);
2320         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2321                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2322                 nfs41_handle_some_state_revoked(clp);
2323         if (flags & SEQ4_STATUS_LEASE_MOVED)
2324                 nfs4_schedule_lease_moved_recovery(clp);
2325         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2326                 nfs41_handle_recallable_state_revoked(clp);
2327 out_recovery:
2328         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2329                 nfs41_handle_backchannel_fault(clp);
2330         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2331                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2332                 nfs41_handle_cb_path_down(clp);
2333 }
2334
2335 static int nfs4_reset_session(struct nfs_client *clp)
2336 {
2337         struct rpc_cred *cred;
2338         int status;
2339
2340         if (!nfs4_has_session(clp))
2341                 return 0;
2342         nfs4_begin_drain_session(clp);
2343         cred = nfs4_get_clid_cred(clp);
2344         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2345         switch (status) {
2346         case 0:
2347         case -NFS4ERR_BADSESSION:
2348         case -NFS4ERR_DEADSESSION:
2349                 break;
2350         case -NFS4ERR_BACK_CHAN_BUSY:
2351         case -NFS4ERR_DELAY:
2352                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2353                 status = 0;
2354                 ssleep(1);
2355                 goto out;
2356         default:
2357                 status = nfs4_recovery_handle_error(clp, status);
2358                 goto out;
2359         }
2360
2361         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2362         status = nfs4_proc_create_session(clp, cred);
2363         if (status) {
2364                 dprintk("%s: session reset failed with status %d for server %s!\n",
2365                         __func__, status, clp->cl_hostname);
2366                 status = nfs4_handle_reclaim_lease_error(clp, status);
2367                 goto out;
2368         }
2369         nfs41_finish_session_reset(clp);
2370         dprintk("%s: session reset was successful for server %s!\n",
2371                         __func__, clp->cl_hostname);
2372 out:
2373         if (cred)
2374                 put_rpccred(cred);
2375         return status;
2376 }
2377
2378 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2379 {
2380         struct rpc_cred *cred;
2381         int ret;
2382
2383         if (!nfs4_has_session(clp))
2384                 return 0;
2385         nfs4_begin_drain_session(clp);
2386         cred = nfs4_get_clid_cred(clp);
2387         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2388         if (cred)
2389                 put_rpccred(cred);
2390         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2391         switch (ret) {
2392         case 0:
2393                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2394                         __func__, clp->cl_hostname);
2395                 break;
2396         case -NFS4ERR_DELAY:
2397                 ssleep(1);
2398                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2399                 break;
2400         default:
2401                 return nfs4_recovery_handle_error(clp, ret);
2402         }
2403         return 0;
2404 }
2405 #else /* CONFIG_NFS_V4_1 */
2406 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2407
2408 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2409 {
2410         return 0;
2411 }
2412 #endif /* CONFIG_NFS_V4_1 */
2413
2414 static void nfs4_state_manager(struct nfs_client *clp)
2415 {
2416         int status = 0;
2417         const char *section = "", *section_sep = "";
2418
2419         /* Ensure exclusive access to NFSv4 state */
2420         do {
2421                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2422                         section = "purge state";
2423                         status = nfs4_purge_lease(clp);
2424                         if (status < 0)
2425                                 goto out_error;
2426                         continue;
2427                 }
2428
2429                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2430                         section = "lease expired";
2431                         /* We're going to have to re-establish a clientid */
2432                         status = nfs4_reclaim_lease(clp);
2433                         if (status < 0)
2434                                 goto out_error;
2435                         continue;
2436                 }
2437
2438                 /* Initialize or reset the session */
2439                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2440                         section = "reset session";
2441                         status = nfs4_reset_session(clp);
2442                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2443                                 continue;
2444                         if (status < 0)
2445                                 goto out_error;
2446                 }
2447
2448                 /* Send BIND_CONN_TO_SESSION */
2449                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2450                                 &clp->cl_state)) {
2451                         section = "bind conn to session";
2452                         status = nfs4_bind_conn_to_session(clp);
2453                         if (status < 0)
2454                                 goto out_error;
2455                         continue;
2456                 }
2457
2458                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2459                         section = "check lease";
2460                         status = nfs4_check_lease(clp);
2461                         if (status < 0)
2462                                 goto out_error;
2463                         continue;
2464                 }
2465
2466                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2467                         section = "migration";
2468                         status = nfs4_handle_migration(clp);
2469                         if (status < 0)
2470                                 goto out_error;
2471                 }
2472
2473                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2474                         section = "lease moved";
2475                         status = nfs4_handle_lease_moved(clp);
2476                         if (status < 0)
2477                                 goto out_error;
2478                 }
2479
2480                 /* First recover reboot state... */
2481                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2482                         section = "reclaim reboot";
2483                         status = nfs4_do_reclaim(clp,
2484                                 clp->cl_mvops->reboot_recovery_ops);
2485                         if (status == -EAGAIN)
2486                                 continue;
2487                         if (status < 0)
2488                                 goto out_error;
2489                         nfs4_state_end_reclaim_reboot(clp);
2490                 }
2491
2492                 /* Detect expired delegations... */
2493                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2494                         section = "detect expired delegations";
2495                         nfs_reap_expired_delegations(clp);
2496                         continue;
2497                 }
2498
2499                 /* Now recover expired state... */
2500                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2501                         section = "reclaim nograce";
2502                         status = nfs4_do_reclaim(clp,
2503                                 clp->cl_mvops->nograce_recovery_ops);
2504                         if (status == -EAGAIN)
2505                                 continue;
2506                         if (status < 0)
2507                                 goto out_error;
2508                 }
2509
2510                 nfs4_end_drain_session(clp);
2511                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2512                         nfs_client_return_marked_delegations(clp);
2513                         continue;
2514                 }
2515
2516                 nfs4_clear_state_manager_bit(clp);
2517                 /* Did we race with an attempt to give us more work? */
2518                 if (clp->cl_state == 0)
2519                         break;
2520                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2521                         break;
2522         } while (atomic_read(&clp->cl_count) > 1);
2523         return;
2524 out_error:
2525         if (strlen(section))
2526                 section_sep = ": ";
2527         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2528                         " with error %d\n", section_sep, section,
2529                         clp->cl_hostname, -status);
2530         ssleep(1);
2531         nfs4_end_drain_session(clp);
2532         nfs4_clear_state_manager_bit(clp);
2533 }
2534
2535 static int nfs4_run_state_manager(void *ptr)
2536 {
2537         struct nfs_client *clp = ptr;
2538
2539         allow_signal(SIGKILL);
2540         nfs4_state_manager(clp);
2541         nfs_put_client(clp);
2542         module_put_and_exit(0);
2543         return 0;
2544 }
2545
2546 /*
2547  * Local variables:
2548  *  c-basic-offset: 8
2549  * End:
2550  */