40a08cd483f051072508b53eee362167ce26904f
[linux-2.6-microblaze.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 const nfs4_stateid invalid_stateid = {
73         {
74                 /* Funky initialiser keeps older gcc versions happy */
75                 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
76         },
77         .type = NFS4_INVALID_STATEID_TYPE,
78 };
79
80 const nfs4_stateid current_stateid = {
81         {
82                 /* Funky initialiser keeps older gcc versions happy */
83                 .data = { 0x0, 0x0, 0x0, 0x1, 0 },
84         },
85         .type = NFS4_SPECIAL_STATEID_TYPE,
86 };
87
88 static DEFINE_MUTEX(nfs_clid_init_mutex);
89
90 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
91 {
92         struct nfs4_setclientid_res clid = {
93                 .clientid = clp->cl_clientid,
94                 .confirm = clp->cl_confirm,
95         };
96         unsigned short port;
97         int status;
98         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
99
100         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
101                 goto do_confirm;
102         port = nn->nfs_callback_tcpport;
103         if (clp->cl_addr.ss_family == AF_INET6)
104                 port = nn->nfs_callback_tcpport6;
105
106         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
107         if (status != 0)
108                 goto out;
109         clp->cl_clientid = clid.clientid;
110         clp->cl_confirm = clid.confirm;
111         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
112 do_confirm:
113         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
114         if (status != 0)
115                 goto out;
116         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
117         nfs4_schedule_state_renewal(clp);
118 out:
119         return status;
120 }
121
122 /**
123  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
124  *
125  * @clp: nfs_client under test
126  * @result: OUT: found nfs_client, or clp
127  * @cred: credential to use for trunking test
128  *
129  * Returns zero, a negative errno, or a negative NFS4ERR status.
130  * If zero is returned, an nfs_client pointer is planted in
131  * "result".
132  *
133  * Note: The returned client may not yet be marked ready.
134  */
135 int nfs40_discover_server_trunking(struct nfs_client *clp,
136                                    struct nfs_client **result,
137                                    struct rpc_cred *cred)
138 {
139         struct nfs4_setclientid_res clid = {
140                 .clientid = clp->cl_clientid,
141                 .confirm = clp->cl_confirm,
142         };
143         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
144         unsigned short port;
145         int status;
146
147         port = nn->nfs_callback_tcpport;
148         if (clp->cl_addr.ss_family == AF_INET6)
149                 port = nn->nfs_callback_tcpport6;
150
151         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
152         if (status != 0)
153                 goto out;
154         clp->cl_clientid = clid.clientid;
155         clp->cl_confirm = clid.confirm;
156
157         status = nfs40_walk_client_list(clp, result, cred);
158         if (status == 0) {
159                 /* Sustain the lease, even if it's empty.  If the clientid4
160                  * goes stale it's of no use for trunking discovery. */
161                 nfs4_schedule_state_renewal(*result);
162         }
163 out:
164         return status;
165 }
166
167 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
168 {
169         struct rpc_cred *cred = NULL;
170
171         if (clp->cl_machine_cred != NULL)
172                 cred = get_rpccred(clp->cl_machine_cred);
173         return cred;
174 }
175
176 static void nfs4_root_machine_cred(struct nfs_client *clp)
177 {
178         struct rpc_cred *cred, *new;
179
180         new = rpc_lookup_machine_cred(NULL);
181         spin_lock(&clp->cl_lock);
182         cred = clp->cl_machine_cred;
183         clp->cl_machine_cred = new;
184         spin_unlock(&clp->cl_lock);
185         if (cred != NULL)
186                 put_rpccred(cred);
187 }
188
189 static struct rpc_cred *
190 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
191 {
192         struct rpc_cred *cred = NULL;
193         struct nfs4_state_owner *sp;
194         struct rb_node *pos;
195
196         for (pos = rb_first(&server->state_owners);
197              pos != NULL;
198              pos = rb_next(pos)) {
199                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
200                 if (list_empty(&sp->so_states))
201                         continue;
202                 cred = get_rpccred(sp->so_cred);
203                 break;
204         }
205         return cred;
206 }
207
208 /**
209  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
210  * @clp: client state handle
211  *
212  * Returns an rpc_cred with reference count bumped, or NULL.
213  * Caller must hold clp->cl_lock.
214  */
215 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
216 {
217         struct rpc_cred *cred = NULL;
218         struct nfs_server *server;
219
220         /* Use machine credentials if available */
221         cred = nfs4_get_machine_cred_locked(clp);
222         if (cred != NULL)
223                 goto out;
224
225         rcu_read_lock();
226         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
227                 cred = nfs4_get_renew_cred_server_locked(server);
228                 if (cred != NULL)
229                         break;
230         }
231         rcu_read_unlock();
232
233 out:
234         return cred;
235 }
236
237 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
238 {
239         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
240                 spin_lock(&tbl->slot_tbl_lock);
241                 nfs41_wake_slot_table(tbl);
242                 spin_unlock(&tbl->slot_tbl_lock);
243         }
244 }
245
246 static void nfs4_end_drain_session(struct nfs_client *clp)
247 {
248         struct nfs4_session *ses = clp->cl_session;
249
250         if (clp->cl_slot_tbl) {
251                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
252                 return;
253         }
254
255         if (ses != NULL) {
256                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
257                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
258         }
259 }
260
261 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
262 {
263         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
264         spin_lock(&tbl->slot_tbl_lock);
265         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
266                 reinit_completion(&tbl->complete);
267                 spin_unlock(&tbl->slot_tbl_lock);
268                 return wait_for_completion_interruptible(&tbl->complete);
269         }
270         spin_unlock(&tbl->slot_tbl_lock);
271         return 0;
272 }
273
274 static int nfs4_begin_drain_session(struct nfs_client *clp)
275 {
276         struct nfs4_session *ses = clp->cl_session;
277         int ret;
278
279         if (clp->cl_slot_tbl)
280                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
281
282         /* back channel */
283         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
284         if (ret)
285                 return ret;
286         /* fore channel */
287         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
288 }
289
290 #if defined(CONFIG_NFS_V4_1)
291
292 static int nfs41_setup_state_renewal(struct nfs_client *clp)
293 {
294         int status;
295         struct nfs_fsinfo fsinfo;
296         unsigned long now;
297
298         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
299                 nfs4_schedule_state_renewal(clp);
300                 return 0;
301         }
302
303         now = jiffies;
304         status = nfs4_proc_get_lease_time(clp, &fsinfo);
305         if (status == 0) {
306                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
307                 nfs4_schedule_state_renewal(clp);
308         }
309
310         return status;
311 }
312
313 static void nfs41_finish_session_reset(struct nfs_client *clp)
314 {
315         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
316         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
317         /* create_session negotiated new slot table */
318         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
319         nfs41_setup_state_renewal(clp);
320 }
321
322 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
323 {
324         int status;
325
326         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
327                 goto do_confirm;
328         status = nfs4_proc_exchange_id(clp, cred);
329         if (status != 0)
330                 goto out;
331         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
332 do_confirm:
333         status = nfs4_proc_create_session(clp, cred);
334         if (status != 0)
335                 goto out;
336         nfs41_finish_session_reset(clp);
337         nfs_mark_client_ready(clp, NFS_CS_READY);
338 out:
339         return status;
340 }
341
342 /**
343  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
344  *
345  * @clp: nfs_client under test
346  * @result: OUT: found nfs_client, or clp
347  * @cred: credential to use for trunking test
348  *
349  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
350  * If NFS4_OK is returned, an nfs_client pointer is planted in
351  * "result".
352  *
353  * Note: The returned client may not yet be marked ready.
354  */
355 int nfs41_discover_server_trunking(struct nfs_client *clp,
356                                    struct nfs_client **result,
357                                    struct rpc_cred *cred)
358 {
359         int status;
360
361         status = nfs4_proc_exchange_id(clp, cred);
362         if (status != NFS4_OK)
363                 return status;
364
365         status = nfs41_walk_client_list(clp, result, cred);
366         if (status < 0)
367                 return status;
368         if (clp != *result)
369                 return 0;
370
371         /*
372          * Purge state if the client id was established in a prior
373          * instance and the client id could not have arrived on the
374          * server via Transparent State Migration.
375          */
376         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
377                 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
378                         set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
379                 else
380                         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
381         }
382         nfs4_schedule_state_manager(clp);
383         status = nfs_wait_client_init_complete(clp);
384         if (status < 0)
385                 nfs_put_client(clp);
386         return status;
387 }
388
389 #endif /* CONFIG_NFS_V4_1 */
390
391 /**
392  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
393  * @clp: client state handle
394  *
395  * Returns an rpc_cred with reference count bumped, or NULL.
396  */
397 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
398 {
399         struct rpc_cred *cred;
400
401         spin_lock(&clp->cl_lock);
402         cred = nfs4_get_machine_cred_locked(clp);
403         spin_unlock(&clp->cl_lock);
404         return cred;
405 }
406
407 static struct nfs4_state_owner *
408 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
409 {
410         struct rb_node **p = &server->state_owners.rb_node,
411                        *parent = NULL;
412         struct nfs4_state_owner *sp;
413
414         while (*p != NULL) {
415                 parent = *p;
416                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
417
418                 if (cred < sp->so_cred)
419                         p = &parent->rb_left;
420                 else if (cred > sp->so_cred)
421                         p = &parent->rb_right;
422                 else {
423                         if (!list_empty(&sp->so_lru))
424                                 list_del_init(&sp->so_lru);
425                         atomic_inc(&sp->so_count);
426                         return sp;
427                 }
428         }
429         return NULL;
430 }
431
432 static struct nfs4_state_owner *
433 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
434 {
435         struct nfs_server *server = new->so_server;
436         struct rb_node **p = &server->state_owners.rb_node,
437                        *parent = NULL;
438         struct nfs4_state_owner *sp;
439
440         while (*p != NULL) {
441                 parent = *p;
442                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
443
444                 if (new->so_cred < sp->so_cred)
445                         p = &parent->rb_left;
446                 else if (new->so_cred > sp->so_cred)
447                         p = &parent->rb_right;
448                 else {
449                         if (!list_empty(&sp->so_lru))
450                                 list_del_init(&sp->so_lru);
451                         atomic_inc(&sp->so_count);
452                         return sp;
453                 }
454         }
455         rb_link_node(&new->so_server_node, parent, p);
456         rb_insert_color(&new->so_server_node, &server->state_owners);
457         return new;
458 }
459
460 static void
461 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
462 {
463         struct nfs_server *server = sp->so_server;
464
465         if (!RB_EMPTY_NODE(&sp->so_server_node))
466                 rb_erase(&sp->so_server_node, &server->state_owners);
467 }
468
469 static void
470 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
471 {
472         sc->create_time = ktime_get();
473         sc->flags = 0;
474         sc->counter = 0;
475         spin_lock_init(&sc->lock);
476         INIT_LIST_HEAD(&sc->list);
477         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
478 }
479
480 static void
481 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
482 {
483         rpc_destroy_wait_queue(&sc->wait);
484 }
485
486 /*
487  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
488  * create a new state_owner.
489  *
490  */
491 static struct nfs4_state_owner *
492 nfs4_alloc_state_owner(struct nfs_server *server,
493                 struct rpc_cred *cred,
494                 gfp_t gfp_flags)
495 {
496         struct nfs4_state_owner *sp;
497
498         sp = kzalloc(sizeof(*sp), gfp_flags);
499         if (!sp)
500                 return NULL;
501         sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
502                                                 gfp_flags);
503         if (sp->so_seqid.owner_id < 0) {
504                 kfree(sp);
505                 return NULL;
506         }
507         sp->so_server = server;
508         sp->so_cred = get_rpccred(cred);
509         spin_lock_init(&sp->so_lock);
510         INIT_LIST_HEAD(&sp->so_states);
511         nfs4_init_seqid_counter(&sp->so_seqid);
512         atomic_set(&sp->so_count, 1);
513         INIT_LIST_HEAD(&sp->so_lru);
514         seqcount_init(&sp->so_reclaim_seqcount);
515         mutex_init(&sp->so_delegreturn_mutex);
516         return sp;
517 }
518
519 static void
520 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
521 {
522         /* This state_owner is no longer usable, but must
523          * remain in place so that state recovery can find it
524          * and the opens associated with it.
525          * It may also be used for new 'open' request to
526          * return a delegation to the server.
527          * So update the 'create_time' so that it looks like
528          * a new state_owner.  This will cause the server to
529          * request an OPEN_CONFIRM to start a new sequence.
530          */
531         sp->so_seqid.create_time = ktime_get();
532 }
533
534 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
535 {
536         nfs4_destroy_seqid_counter(&sp->so_seqid);
537         put_rpccred(sp->so_cred);
538         ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
539         kfree(sp);
540 }
541
542 static void nfs4_gc_state_owners(struct nfs_server *server)
543 {
544         struct nfs_client *clp = server->nfs_client;
545         struct nfs4_state_owner *sp, *tmp;
546         unsigned long time_min, time_max;
547         LIST_HEAD(doomed);
548
549         spin_lock(&clp->cl_lock);
550         time_max = jiffies;
551         time_min = (long)time_max - (long)clp->cl_lease_time;
552         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
553                 /* NB: LRU is sorted so that oldest is at the head */
554                 if (time_in_range(sp->so_expires, time_min, time_max))
555                         break;
556                 list_move(&sp->so_lru, &doomed);
557                 nfs4_remove_state_owner_locked(sp);
558         }
559         spin_unlock(&clp->cl_lock);
560
561         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
562                 list_del(&sp->so_lru);
563                 nfs4_free_state_owner(sp);
564         }
565 }
566
567 /**
568  * nfs4_get_state_owner - Look up a state owner given a credential
569  * @server: nfs_server to search
570  * @cred: RPC credential to match
571  *
572  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
573  */
574 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
575                                               struct rpc_cred *cred,
576                                               gfp_t gfp_flags)
577 {
578         struct nfs_client *clp = server->nfs_client;
579         struct nfs4_state_owner *sp, *new;
580
581         spin_lock(&clp->cl_lock);
582         sp = nfs4_find_state_owner_locked(server, cred);
583         spin_unlock(&clp->cl_lock);
584         if (sp != NULL)
585                 goto out;
586         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
587         if (new == NULL)
588                 goto out;
589         spin_lock(&clp->cl_lock);
590         sp = nfs4_insert_state_owner_locked(new);
591         spin_unlock(&clp->cl_lock);
592         if (sp != new)
593                 nfs4_free_state_owner(new);
594 out:
595         nfs4_gc_state_owners(server);
596         return sp;
597 }
598
599 /**
600  * nfs4_put_state_owner - Release a nfs4_state_owner
601  * @sp: state owner data to release
602  *
603  * Note that we keep released state owners on an LRU
604  * list.
605  * This caches valid state owners so that they can be
606  * reused, to avoid the OPEN_CONFIRM on minor version 0.
607  * It also pins the uniquifier of dropped state owners for
608  * a while, to ensure that those state owner names are
609  * never reused.
610  */
611 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
612 {
613         struct nfs_server *server = sp->so_server;
614         struct nfs_client *clp = server->nfs_client;
615
616         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
617                 return;
618
619         sp->so_expires = jiffies;
620         list_add_tail(&sp->so_lru, &server->state_owners_lru);
621         spin_unlock(&clp->cl_lock);
622 }
623
624 /**
625  * nfs4_purge_state_owners - Release all cached state owners
626  * @server: nfs_server with cached state owners to release
627  *
628  * Called at umount time.  Remaining state owners will be on
629  * the LRU with ref count of zero.
630  */
631 void nfs4_purge_state_owners(struct nfs_server *server)
632 {
633         struct nfs_client *clp = server->nfs_client;
634         struct nfs4_state_owner *sp, *tmp;
635         LIST_HEAD(doomed);
636
637         spin_lock(&clp->cl_lock);
638         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
639                 list_move(&sp->so_lru, &doomed);
640                 nfs4_remove_state_owner_locked(sp);
641         }
642         spin_unlock(&clp->cl_lock);
643
644         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
645                 list_del(&sp->so_lru);
646                 nfs4_free_state_owner(sp);
647         }
648 }
649
650 static struct nfs4_state *
651 nfs4_alloc_open_state(void)
652 {
653         struct nfs4_state *state;
654
655         state = kzalloc(sizeof(*state), GFP_NOFS);
656         if (!state)
657                 return NULL;
658         atomic_set(&state->count, 1);
659         INIT_LIST_HEAD(&state->lock_states);
660         spin_lock_init(&state->state_lock);
661         seqlock_init(&state->seqlock);
662         init_waitqueue_head(&state->waitq);
663         return state;
664 }
665
666 void
667 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
668 {
669         if (state->state == fmode)
670                 return;
671         /* NB! List reordering - see the reclaim code for why.  */
672         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
673                 if (fmode & FMODE_WRITE)
674                         list_move(&state->open_states, &state->owner->so_states);
675                 else
676                         list_move_tail(&state->open_states, &state->owner->so_states);
677         }
678         state->state = fmode;
679 }
680
681 static struct nfs4_state *
682 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
683 {
684         struct nfs_inode *nfsi = NFS_I(inode);
685         struct nfs4_state *state;
686
687         list_for_each_entry(state, &nfsi->open_states, inode_states) {
688                 if (state->owner != owner)
689                         continue;
690                 if (!nfs4_valid_open_stateid(state))
691                         continue;
692                 if (atomic_inc_not_zero(&state->count))
693                         return state;
694         }
695         return NULL;
696 }
697
698 static void
699 nfs4_free_open_state(struct nfs4_state *state)
700 {
701         kfree(state);
702 }
703
704 struct nfs4_state *
705 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
706 {
707         struct nfs4_state *state, *new;
708         struct nfs_inode *nfsi = NFS_I(inode);
709
710         spin_lock(&inode->i_lock);
711         state = __nfs4_find_state_byowner(inode, owner);
712         spin_unlock(&inode->i_lock);
713         if (state)
714                 goto out;
715         new = nfs4_alloc_open_state();
716         spin_lock(&owner->so_lock);
717         spin_lock(&inode->i_lock);
718         state = __nfs4_find_state_byowner(inode, owner);
719         if (state == NULL && new != NULL) {
720                 state = new;
721                 state->owner = owner;
722                 atomic_inc(&owner->so_count);
723                 list_add(&state->inode_states, &nfsi->open_states);
724                 ihold(inode);
725                 state->inode = inode;
726                 spin_unlock(&inode->i_lock);
727                 /* Note: The reclaim code dictates that we add stateless
728                  * and read-only stateids to the end of the list */
729                 list_add_tail(&state->open_states, &owner->so_states);
730                 spin_unlock(&owner->so_lock);
731         } else {
732                 spin_unlock(&inode->i_lock);
733                 spin_unlock(&owner->so_lock);
734                 if (new)
735                         nfs4_free_open_state(new);
736         }
737 out:
738         return state;
739 }
740
741 void nfs4_put_open_state(struct nfs4_state *state)
742 {
743         struct inode *inode = state->inode;
744         struct nfs4_state_owner *owner = state->owner;
745
746         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
747                 return;
748         spin_lock(&inode->i_lock);
749         list_del(&state->inode_states);
750         list_del(&state->open_states);
751         spin_unlock(&inode->i_lock);
752         spin_unlock(&owner->so_lock);
753         iput(inode);
754         nfs4_free_open_state(state);
755         nfs4_put_state_owner(owner);
756 }
757
758 /*
759  * Close the current file.
760  */
761 static void __nfs4_close(struct nfs4_state *state,
762                 fmode_t fmode, gfp_t gfp_mask, int wait)
763 {
764         struct nfs4_state_owner *owner = state->owner;
765         int call_close = 0;
766         fmode_t newstate;
767
768         atomic_inc(&owner->so_count);
769         /* Protect against nfs4_find_state() */
770         spin_lock(&owner->so_lock);
771         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
772                 case FMODE_READ:
773                         state->n_rdonly--;
774                         break;
775                 case FMODE_WRITE:
776                         state->n_wronly--;
777                         break;
778                 case FMODE_READ|FMODE_WRITE:
779                         state->n_rdwr--;
780         }
781         newstate = FMODE_READ|FMODE_WRITE;
782         if (state->n_rdwr == 0) {
783                 if (state->n_rdonly == 0) {
784                         newstate &= ~FMODE_READ;
785                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
786                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
787                 }
788                 if (state->n_wronly == 0) {
789                         newstate &= ~FMODE_WRITE;
790                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
791                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
792                 }
793                 if (newstate == 0)
794                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
795         }
796         nfs4_state_set_mode_locked(state, newstate);
797         spin_unlock(&owner->so_lock);
798
799         if (!call_close) {
800                 nfs4_put_open_state(state);
801                 nfs4_put_state_owner(owner);
802         } else
803                 nfs4_do_close(state, gfp_mask, wait);
804 }
805
806 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
807 {
808         __nfs4_close(state, fmode, GFP_NOFS, 0);
809 }
810
811 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
812 {
813         __nfs4_close(state, fmode, GFP_KERNEL, 1);
814 }
815
816 /*
817  * Search the state->lock_states for an existing lock_owner
818  * that is compatible with either of the given owners.
819  * If the second is non-zero, then the first refers to a Posix-lock
820  * owner (current->files) and the second refers to a flock/OFD
821  * owner (struct file*).  In that case, prefer a match for the first
822  * owner.
823  * If both sorts of locks are held on the one file we cannot know
824  * which stateid was intended to be used, so a "correct" choice cannot
825  * be made.  Failing that, a "consistent" choice is preferable.  The
826  * consistent choice we make is to prefer the first owner, that of a
827  * Posix lock.
828  */
829 static struct nfs4_lock_state *
830 __nfs4_find_lock_state(struct nfs4_state *state,
831                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
832 {
833         struct nfs4_lock_state *pos, *ret = NULL;
834         list_for_each_entry(pos, &state->lock_states, ls_locks) {
835                 if (pos->ls_owner == fl_owner) {
836                         ret = pos;
837                         break;
838                 }
839                 if (pos->ls_owner == fl_owner2)
840                         ret = pos;
841         }
842         if (ret)
843                 refcount_inc(&ret->ls_count);
844         return ret;
845 }
846
847 /*
848  * Return a compatible lock_state. If no initialized lock_state structure
849  * exists, return an uninitialized one.
850  *
851  */
852 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
853 {
854         struct nfs4_lock_state *lsp;
855         struct nfs_server *server = state->owner->so_server;
856
857         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
858         if (lsp == NULL)
859                 return NULL;
860         nfs4_init_seqid_counter(&lsp->ls_seqid);
861         refcount_set(&lsp->ls_count, 1);
862         lsp->ls_state = state;
863         lsp->ls_owner = fl_owner;
864         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
865         if (lsp->ls_seqid.owner_id < 0)
866                 goto out_free;
867         INIT_LIST_HEAD(&lsp->ls_locks);
868         return lsp;
869 out_free:
870         kfree(lsp);
871         return NULL;
872 }
873
874 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
875 {
876         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
877         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
878         kfree(lsp);
879 }
880
881 /*
882  * Return a compatible lock_state. If no initialized lock_state structure
883  * exists, return an uninitialized one.
884  *
885  */
886 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
887 {
888         struct nfs4_lock_state *lsp, *new = NULL;
889         
890         for(;;) {
891                 spin_lock(&state->state_lock);
892                 lsp = __nfs4_find_lock_state(state, owner, NULL);
893                 if (lsp != NULL)
894                         break;
895                 if (new != NULL) {
896                         list_add(&new->ls_locks, &state->lock_states);
897                         set_bit(LK_STATE_IN_USE, &state->flags);
898                         lsp = new;
899                         new = NULL;
900                         break;
901                 }
902                 spin_unlock(&state->state_lock);
903                 new = nfs4_alloc_lock_state(state, owner);
904                 if (new == NULL)
905                         return NULL;
906         }
907         spin_unlock(&state->state_lock);
908         if (new != NULL)
909                 nfs4_free_lock_state(state->owner->so_server, new);
910         return lsp;
911 }
912
913 /*
914  * Release reference to lock_state, and free it if we see that
915  * it is no longer in use
916  */
917 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
918 {
919         struct nfs_server *server;
920         struct nfs4_state *state;
921
922         if (lsp == NULL)
923                 return;
924         state = lsp->ls_state;
925         if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
926                 return;
927         list_del(&lsp->ls_locks);
928         if (list_empty(&state->lock_states))
929                 clear_bit(LK_STATE_IN_USE, &state->flags);
930         spin_unlock(&state->state_lock);
931         server = state->owner->so_server;
932         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
933                 struct nfs_client *clp = server->nfs_client;
934
935                 clp->cl_mvops->free_lock_state(server, lsp);
936         } else
937                 nfs4_free_lock_state(server, lsp);
938 }
939
940 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
941 {
942         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
943
944         dst->fl_u.nfs4_fl.owner = lsp;
945         refcount_inc(&lsp->ls_count);
946 }
947
948 static void nfs4_fl_release_lock(struct file_lock *fl)
949 {
950         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
951 }
952
953 static const struct file_lock_operations nfs4_fl_lock_ops = {
954         .fl_copy_lock = nfs4_fl_copy_lock,
955         .fl_release_private = nfs4_fl_release_lock,
956 };
957
958 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
959 {
960         struct nfs4_lock_state *lsp;
961
962         if (fl->fl_ops != NULL)
963                 return 0;
964         lsp = nfs4_get_lock_state(state, fl->fl_owner);
965         if (lsp == NULL)
966                 return -ENOMEM;
967         fl->fl_u.nfs4_fl.owner = lsp;
968         fl->fl_ops = &nfs4_fl_lock_ops;
969         return 0;
970 }
971
972 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
973                 struct nfs4_state *state,
974                 const struct nfs_lock_context *l_ctx)
975 {
976         struct nfs4_lock_state *lsp;
977         fl_owner_t fl_owner, fl_flock_owner;
978         int ret = -ENOENT;
979
980         if (l_ctx == NULL)
981                 goto out;
982
983         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
984                 goto out;
985
986         fl_owner = l_ctx->lockowner;
987         fl_flock_owner = l_ctx->open_context->flock_owner;
988
989         spin_lock(&state->state_lock);
990         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
991         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
992                 ret = -EIO;
993         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
994                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
995                 ret = 0;
996         }
997         spin_unlock(&state->state_lock);
998         nfs4_put_lock_state(lsp);
999 out:
1000         return ret;
1001 }
1002
1003 bool nfs4_refresh_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1004 {
1005         bool ret;
1006         int seq;
1007
1008         do {
1009                 ret = false;
1010                 seq = read_seqbegin(&state->seqlock);
1011                 if (nfs4_state_match_open_stateid_other(state, dst)) {
1012                         dst->seqid = state->open_stateid.seqid;
1013                         ret = true;
1014                 }
1015         } while (read_seqretry(&state->seqlock, seq));
1016         return ret;
1017 }
1018
1019 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1020 {
1021         bool ret;
1022         const nfs4_stateid *src;
1023         int seq;
1024
1025         do {
1026                 ret = false;
1027                 src = &zero_stateid;
1028                 seq = read_seqbegin(&state->seqlock);
1029                 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1030                         src = &state->open_stateid;
1031                         ret = true;
1032                 }
1033                 nfs4_stateid_copy(dst, src);
1034         } while (read_seqretry(&state->seqlock, seq));
1035         return ret;
1036 }
1037
1038 /*
1039  * Byte-range lock aware utility to initialize the stateid of read/write
1040  * requests.
1041  */
1042 int nfs4_select_rw_stateid(struct nfs4_state *state,
1043                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1044                 nfs4_stateid *dst, struct rpc_cred **cred)
1045 {
1046         int ret;
1047
1048         if (!nfs4_valid_open_stateid(state))
1049                 return -EIO;
1050         if (cred != NULL)
1051                 *cred = NULL;
1052         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1053         if (ret == -EIO)
1054                 /* A lost lock - don't even consider delegations */
1055                 goto out;
1056         /* returns true if delegation stateid found and copied */
1057         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1058                 ret = 0;
1059                 goto out;
1060         }
1061         if (ret != -ENOENT)
1062                 /* nfs4_copy_delegation_stateid() didn't over-write
1063                  * dst, so it still has the lock stateid which we now
1064                  * choose to use.
1065                  */
1066                 goto out;
1067         nfs4_copy_open_stateid(dst, state);
1068         ret = 0;
1069 out:
1070         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1071                 dst->seqid = 0;
1072         return ret;
1073 }
1074
1075 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1076 {
1077         struct nfs_seqid *new;
1078
1079         new = kmalloc(sizeof(*new), gfp_mask);
1080         if (new == NULL)
1081                 return ERR_PTR(-ENOMEM);
1082         new->sequence = counter;
1083         INIT_LIST_HEAD(&new->list);
1084         new->task = NULL;
1085         return new;
1086 }
1087
1088 void nfs_release_seqid(struct nfs_seqid *seqid)
1089 {
1090         struct nfs_seqid_counter *sequence;
1091
1092         if (seqid == NULL || list_empty(&seqid->list))
1093                 return;
1094         sequence = seqid->sequence;
1095         spin_lock(&sequence->lock);
1096         list_del_init(&seqid->list);
1097         if (!list_empty(&sequence->list)) {
1098                 struct nfs_seqid *next;
1099
1100                 next = list_first_entry(&sequence->list,
1101                                 struct nfs_seqid, list);
1102                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1103         }
1104         spin_unlock(&sequence->lock);
1105 }
1106
1107 void nfs_free_seqid(struct nfs_seqid *seqid)
1108 {
1109         nfs_release_seqid(seqid);
1110         kfree(seqid);
1111 }
1112
1113 /*
1114  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1115  * failed with a seqid incrementing error -
1116  * see comments nfs4.h:seqid_mutating_error()
1117  */
1118 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1119 {
1120         switch (status) {
1121                 case 0:
1122                         break;
1123                 case -NFS4ERR_BAD_SEQID:
1124                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1125                                 return;
1126                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1127                                         " sequence-id error on an"
1128                                         " unconfirmed sequence %p!\n",
1129                                         seqid->sequence);
1130                 case -NFS4ERR_STALE_CLIENTID:
1131                 case -NFS4ERR_STALE_STATEID:
1132                 case -NFS4ERR_BAD_STATEID:
1133                 case -NFS4ERR_BADXDR:
1134                 case -NFS4ERR_RESOURCE:
1135                 case -NFS4ERR_NOFILEHANDLE:
1136                 case -NFS4ERR_MOVED:
1137                         /* Non-seqid mutating errors */
1138                         return;
1139         };
1140         /*
1141          * Note: no locking needed as we are guaranteed to be first
1142          * on the sequence list
1143          */
1144         seqid->sequence->counter++;
1145 }
1146
1147 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1148 {
1149         struct nfs4_state_owner *sp;
1150
1151         if (seqid == NULL)
1152                 return;
1153
1154         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1155         if (status == -NFS4ERR_BAD_SEQID)
1156                 nfs4_reset_state_owner(sp);
1157         if (!nfs4_has_session(sp->so_server->nfs_client))
1158                 nfs_increment_seqid(status, seqid);
1159 }
1160
1161 /*
1162  * Increment the seqid if the LOCK/LOCKU succeeded, or
1163  * failed with a seqid incrementing error -
1164  * see comments nfs4.h:seqid_mutating_error()
1165  */
1166 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1167 {
1168         if (seqid != NULL)
1169                 nfs_increment_seqid(status, seqid);
1170 }
1171
1172 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1173 {
1174         struct nfs_seqid_counter *sequence;
1175         int status = 0;
1176
1177         if (seqid == NULL)
1178                 goto out;
1179         sequence = seqid->sequence;
1180         spin_lock(&sequence->lock);
1181         seqid->task = task;
1182         if (list_empty(&seqid->list))
1183                 list_add_tail(&seqid->list, &sequence->list);
1184         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1185                 goto unlock;
1186         rpc_sleep_on(&sequence->wait, task, NULL);
1187         status = -EAGAIN;
1188 unlock:
1189         spin_unlock(&sequence->lock);
1190 out:
1191         return status;
1192 }
1193
1194 static int nfs4_run_state_manager(void *);
1195
1196 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1197 {
1198         smp_mb__before_atomic();
1199         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1200         smp_mb__after_atomic();
1201         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1202         rpc_wake_up(&clp->cl_rpcwaitq);
1203 }
1204
1205 /*
1206  * Schedule the nfs_client asynchronous state management routine
1207  */
1208 void nfs4_schedule_state_manager(struct nfs_client *clp)
1209 {
1210         struct task_struct *task;
1211         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1212
1213         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1214                 return;
1215         __module_get(THIS_MODULE);
1216         refcount_inc(&clp->cl_count);
1217
1218         /* The rcu_read_lock() is not strictly necessary, as the state
1219          * manager is the only thread that ever changes the rpc_xprt
1220          * after it's initialized.  At this point, we're single threaded. */
1221         rcu_read_lock();
1222         snprintf(buf, sizeof(buf), "%s-manager",
1223                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1224         rcu_read_unlock();
1225         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1226         if (IS_ERR(task)) {
1227                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1228                         __func__, PTR_ERR(task));
1229                 nfs4_clear_state_manager_bit(clp);
1230                 nfs_put_client(clp);
1231                 module_put(THIS_MODULE);
1232         }
1233 }
1234
1235 /*
1236  * Schedule a lease recovery attempt
1237  */
1238 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1239 {
1240         if (!clp)
1241                 return;
1242         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1243                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1244         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1245                         clp->cl_hostname);
1246         nfs4_schedule_state_manager(clp);
1247 }
1248 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1249
1250 /**
1251  * nfs4_schedule_migration_recovery - trigger migration recovery
1252  *
1253  * @server: FSID that is migrating
1254  *
1255  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1256  * value is returned.
1257  */
1258 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1259 {
1260         struct nfs_client *clp = server->nfs_client;
1261
1262         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1263                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1264                                 clp->cl_hostname);
1265                 return -NFS4ERR_IO;
1266         }
1267
1268         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1269                 return -NFS4ERR_IO;
1270
1271         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1272                         __func__,
1273                         (unsigned long long)server->fsid.major,
1274                         (unsigned long long)server->fsid.minor,
1275                         clp->cl_hostname);
1276
1277         set_bit(NFS_MIG_IN_TRANSITION,
1278                         &((struct nfs_server *)server)->mig_status);
1279         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1280
1281         nfs4_schedule_state_manager(clp);
1282         return 0;
1283 }
1284 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1285
1286 /**
1287  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1288  *
1289  * @clp: server to check for moved leases
1290  *
1291  */
1292 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1293 {
1294         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1295                 __func__, clp->cl_clientid, clp->cl_hostname);
1296
1297         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1298         nfs4_schedule_state_manager(clp);
1299 }
1300 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1301
1302 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1303 {
1304         int res;
1305
1306         might_sleep();
1307
1308         refcount_inc(&clp->cl_count);
1309         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1310                                  nfs_wait_bit_killable, TASK_KILLABLE);
1311         if (res)
1312                 goto out;
1313         if (clp->cl_cons_state < 0)
1314                 res = clp->cl_cons_state;
1315 out:
1316         nfs_put_client(clp);
1317         return res;
1318 }
1319
1320 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1321 {
1322         unsigned int loop;
1323         int ret;
1324
1325         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1326                 ret = nfs4_wait_clnt_recover(clp);
1327                 if (ret != 0)
1328                         break;
1329                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1330                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1331                         break;
1332                 nfs4_schedule_state_manager(clp);
1333                 ret = -EIO;
1334         }
1335         return ret;
1336 }
1337
1338 /*
1339  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1340  * @clp: client to process
1341  *
1342  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1343  * resend of the SETCLIENTID and hence re-establish the
1344  * callback channel. Then return all existing delegations.
1345  */
1346 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1347 {
1348         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1349         nfs_expire_all_delegations(clp);
1350         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1351                         clp->cl_hostname);
1352 }
1353
1354 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1355 {
1356         nfs40_handle_cb_pathdown(clp);
1357         nfs4_schedule_state_manager(clp);
1358 }
1359
1360 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1361 {
1362
1363         if (!nfs4_valid_open_stateid(state))
1364                 return 0;
1365         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1366         /* Don't recover state that expired before the reboot */
1367         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1368                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1369                 return 0;
1370         }
1371         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1372         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1373         return 1;
1374 }
1375
1376 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1377 {
1378         if (!nfs4_valid_open_stateid(state))
1379                 return 0;
1380         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1381         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1382         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1383         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1384         return 1;
1385 }
1386
1387 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1388 {
1389         struct nfs_client *clp = server->nfs_client;
1390
1391         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1392                 return -EBADF;
1393         nfs_inode_find_delegation_state_and_recover(state->inode,
1394                         &state->stateid);
1395         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1396                         clp->cl_hostname);
1397         nfs4_schedule_state_manager(clp);
1398         return 0;
1399 }
1400 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1401
1402 static struct nfs4_lock_state *
1403 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1404                 const nfs4_stateid *stateid)
1405 {
1406         struct nfs4_lock_state *pos;
1407
1408         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1409                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1410                         continue;
1411                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1412                         return pos;
1413         }
1414         return NULL;
1415 }
1416
1417 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1418                 const nfs4_stateid *stateid)
1419 {
1420         bool found = false;
1421
1422         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1423                 spin_lock(&state->state_lock);
1424                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1425                         found = true;
1426                 spin_unlock(&state->state_lock);
1427         }
1428         return found;
1429 }
1430
1431 void nfs_inode_find_state_and_recover(struct inode *inode,
1432                 const nfs4_stateid *stateid)
1433 {
1434         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1435         struct nfs_inode *nfsi = NFS_I(inode);
1436         struct nfs_open_context *ctx;
1437         struct nfs4_state *state;
1438         bool found = false;
1439
1440         spin_lock(&inode->i_lock);
1441         list_for_each_entry(ctx, &nfsi->open_files, list) {
1442                 state = ctx->state;
1443                 if (state == NULL)
1444                         continue;
1445                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1446                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1447                         found = true;
1448                         continue;
1449                 }
1450                 if (nfs4_stateid_match_other(&state->open_stateid, stateid) &&
1451                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1452                         found = true;
1453                         continue;
1454                 }
1455                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1456                     nfs4_state_mark_reclaim_nograce(clp, state))
1457                         found = true;
1458         }
1459         spin_unlock(&inode->i_lock);
1460
1461         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1462         if (found)
1463                 nfs4_schedule_state_manager(clp);
1464 }
1465
1466 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1467 {
1468         struct inode *inode = state->inode;
1469         struct nfs_inode *nfsi = NFS_I(inode);
1470         struct nfs_open_context *ctx;
1471
1472         spin_lock(&inode->i_lock);
1473         list_for_each_entry(ctx, &nfsi->open_files, list) {
1474                 if (ctx->state != state)
1475                         continue;
1476                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1477         }
1478         spin_unlock(&inode->i_lock);
1479 }
1480
1481 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1482 {
1483         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1484         nfs4_state_mark_open_context_bad(state);
1485 }
1486
1487
1488 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1489 {
1490         struct inode *inode = state->inode;
1491         struct nfs_inode *nfsi = NFS_I(inode);
1492         struct file_lock *fl;
1493         struct nfs4_lock_state *lsp;
1494         int status = 0;
1495         struct file_lock_context *flctx = inode->i_flctx;
1496         struct list_head *list;
1497
1498         if (flctx == NULL)
1499                 return 0;
1500
1501         list = &flctx->flc_posix;
1502
1503         /* Guard against delegation returns and new lock/unlock calls */
1504         down_write(&nfsi->rwsem);
1505         spin_lock(&flctx->flc_lock);
1506 restart:
1507         list_for_each_entry(fl, list, fl_list) {
1508                 if (nfs_file_open_context(fl->fl_file)->state != state)
1509                         continue;
1510                 spin_unlock(&flctx->flc_lock);
1511                 status = ops->recover_lock(state, fl);
1512                 switch (status) {
1513                 case 0:
1514                         break;
1515                 case -ESTALE:
1516                 case -NFS4ERR_ADMIN_REVOKED:
1517                 case -NFS4ERR_STALE_STATEID:
1518                 case -NFS4ERR_BAD_STATEID:
1519                 case -NFS4ERR_EXPIRED:
1520                 case -NFS4ERR_NO_GRACE:
1521                 case -NFS4ERR_STALE_CLIENTID:
1522                 case -NFS4ERR_BADSESSION:
1523                 case -NFS4ERR_BADSLOT:
1524                 case -NFS4ERR_BAD_HIGH_SLOT:
1525                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1526                         goto out;
1527                 default:
1528                         pr_err("NFS: %s: unhandled error %d\n",
1529                                         __func__, status);
1530                         /* Fall through */
1531                 case -ENOMEM:
1532                 case -NFS4ERR_DENIED:
1533                 case -NFS4ERR_RECLAIM_BAD:
1534                 case -NFS4ERR_RECLAIM_CONFLICT:
1535                         lsp = fl->fl_u.nfs4_fl.owner;
1536                         if (lsp)
1537                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1538                         status = 0;
1539                 }
1540                 spin_lock(&flctx->flc_lock);
1541         }
1542         if (list == &flctx->flc_posix) {
1543                 list = &flctx->flc_flock;
1544                 goto restart;
1545         }
1546         spin_unlock(&flctx->flc_lock);
1547 out:
1548         up_write(&nfsi->rwsem);
1549         return status;
1550 }
1551
1552 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1553 {
1554         struct nfs4_state *state;
1555         struct nfs4_lock_state *lock;
1556         int status = 0;
1557
1558         /* Note: we rely on the sp->so_states list being ordered 
1559          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1560          * states first.
1561          * This is needed to ensure that the server won't give us any
1562          * read delegations that we have to return if, say, we are
1563          * recovering after a network partition or a reboot from a
1564          * server that doesn't support a grace period.
1565          */
1566         spin_lock(&sp->so_lock);
1567         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1568 restart:
1569         list_for_each_entry(state, &sp->so_states, open_states) {
1570                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1571                         continue;
1572                 if (!nfs4_valid_open_stateid(state))
1573                         continue;
1574                 if (state->state == 0)
1575                         continue;
1576                 atomic_inc(&state->count);
1577                 spin_unlock(&sp->so_lock);
1578                 status = ops->recover_open(sp, state);
1579                 if (status >= 0) {
1580                         status = nfs4_reclaim_locks(state, ops);
1581                         if (status >= 0) {
1582                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1583                                         spin_lock(&state->state_lock);
1584                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1585                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1586                                                         pr_warn_ratelimited("NFS: "
1587                                                                             "%s: Lock reclaim "
1588                                                                             "failed!\n", __func__);
1589                                         }
1590                                         spin_unlock(&state->state_lock);
1591                                 }
1592                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1593                                         &state->flags);
1594 #ifdef CONFIG_NFS_V4_2
1595                                 if (test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags)) {
1596                                         struct nfs4_copy_state *copy;
1597
1598                                         spin_lock(&sp->so_server->nfs_client->cl_lock);
1599                                         list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1600                                                 if (memcmp(&state->stateid.other, &copy->parent_state->stateid.other, NFS4_STATEID_SIZE))
1601                                                         continue;
1602                                                 copy->flags = 1;
1603                                                 complete(&copy->completion);
1604                                                 printk("AGLO: server rebooted waking up the copy\n");
1605                                                 break;
1606                                         }
1607                                         spin_unlock(&sp->so_server->nfs_client->cl_lock);
1608                                 }
1609 #endif /* CONFIG_NFS_V4_2 */
1610                                 nfs4_put_open_state(state);
1611                                 spin_lock(&sp->so_lock);
1612                                 goto restart;
1613                         }
1614                 }
1615                 switch (status) {
1616                         default:
1617                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1618                                         __func__, status);
1619                                 /* Fall through */
1620                         case -ENOENT:
1621                         case -ENOMEM:
1622                         case -EACCES:
1623                         case -EROFS:
1624                         case -EIO:
1625                         case -ESTALE:
1626                                 /* Open state on this file cannot be recovered */
1627                                 nfs4_state_mark_recovery_failed(state, status);
1628                                 break;
1629                         case -EAGAIN:
1630                                 ssleep(1);
1631                                 /* Fall through */
1632                         case -NFS4ERR_ADMIN_REVOKED:
1633                         case -NFS4ERR_STALE_STATEID:
1634                         case -NFS4ERR_OLD_STATEID:
1635                         case -NFS4ERR_BAD_STATEID:
1636                         case -NFS4ERR_RECLAIM_BAD:
1637                         case -NFS4ERR_RECLAIM_CONFLICT:
1638                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1639                                 break;
1640                         case -NFS4ERR_EXPIRED:
1641                         case -NFS4ERR_NO_GRACE:
1642                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1643                         case -NFS4ERR_STALE_CLIENTID:
1644                         case -NFS4ERR_BADSESSION:
1645                         case -NFS4ERR_BADSLOT:
1646                         case -NFS4ERR_BAD_HIGH_SLOT:
1647                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1648                                 goto out_err;
1649                 }
1650                 nfs4_put_open_state(state);
1651                 spin_lock(&sp->so_lock);
1652                 goto restart;
1653         }
1654         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1655         spin_unlock(&sp->so_lock);
1656         return 0;
1657 out_err:
1658         nfs4_put_open_state(state);
1659         spin_lock(&sp->so_lock);
1660         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1661         spin_unlock(&sp->so_lock);
1662         return status;
1663 }
1664
1665 static void nfs4_clear_open_state(struct nfs4_state *state)
1666 {
1667         struct nfs4_lock_state *lock;
1668
1669         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1670         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1671         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1672         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1673         spin_lock(&state->state_lock);
1674         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1675                 lock->ls_seqid.flags = 0;
1676                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1677         }
1678         spin_unlock(&state->state_lock);
1679 }
1680
1681 static void nfs4_reset_seqids(struct nfs_server *server,
1682         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1683 {
1684         struct nfs_client *clp = server->nfs_client;
1685         struct nfs4_state_owner *sp;
1686         struct rb_node *pos;
1687         struct nfs4_state *state;
1688
1689         spin_lock(&clp->cl_lock);
1690         for (pos = rb_first(&server->state_owners);
1691              pos != NULL;
1692              pos = rb_next(pos)) {
1693                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1694                 sp->so_seqid.flags = 0;
1695                 spin_lock(&sp->so_lock);
1696                 list_for_each_entry(state, &sp->so_states, open_states) {
1697                         if (mark_reclaim(clp, state))
1698                                 nfs4_clear_open_state(state);
1699                 }
1700                 spin_unlock(&sp->so_lock);
1701         }
1702         spin_unlock(&clp->cl_lock);
1703 }
1704
1705 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1706         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1707 {
1708         struct nfs_server *server;
1709
1710         rcu_read_lock();
1711         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1712                 nfs4_reset_seqids(server, mark_reclaim);
1713         rcu_read_unlock();
1714 }
1715
1716 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1717 {
1718         /* Mark all delegations for reclaim */
1719         nfs_delegation_mark_reclaim(clp);
1720         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1721 }
1722
1723 static int nfs4_reclaim_complete(struct nfs_client *clp,
1724                                  const struct nfs4_state_recovery_ops *ops,
1725                                  struct rpc_cred *cred)
1726 {
1727         /* Notify the server we're done reclaiming our state */
1728         if (ops->reclaim_complete)
1729                 return ops->reclaim_complete(clp, cred);
1730         return 0;
1731 }
1732
1733 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1734 {
1735         struct nfs_client *clp = server->nfs_client;
1736         struct nfs4_state_owner *sp;
1737         struct rb_node *pos;
1738         struct nfs4_state *state;
1739
1740         spin_lock(&clp->cl_lock);
1741         for (pos = rb_first(&server->state_owners);
1742              pos != NULL;
1743              pos = rb_next(pos)) {
1744                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1745                 spin_lock(&sp->so_lock);
1746                 list_for_each_entry(state, &sp->so_states, open_states) {
1747                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1748                                                 &state->flags))
1749                                 continue;
1750                         nfs4_state_mark_reclaim_nograce(clp, state);
1751                 }
1752                 spin_unlock(&sp->so_lock);
1753         }
1754         spin_unlock(&clp->cl_lock);
1755 }
1756
1757 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1758 {
1759         struct nfs_server *server;
1760
1761         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1762                 return 0;
1763
1764         rcu_read_lock();
1765         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1766                 nfs4_clear_reclaim_server(server);
1767         rcu_read_unlock();
1768
1769         nfs_delegation_reap_unclaimed(clp);
1770         return 1;
1771 }
1772
1773 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1774 {
1775         const struct nfs4_state_recovery_ops *ops;
1776         struct rpc_cred *cred;
1777         int err;
1778
1779         if (!nfs4_state_clear_reclaim_reboot(clp))
1780                 return;
1781         ops = clp->cl_mvops->reboot_recovery_ops;
1782         cred = nfs4_get_clid_cred(clp);
1783         err = nfs4_reclaim_complete(clp, ops, cred);
1784         put_rpccred(cred);
1785         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1786                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1787 }
1788
1789 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1790 {
1791         nfs_mark_test_expired_all_delegations(clp);
1792         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1793 }
1794
1795 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1796 {
1797         switch (error) {
1798                 case 0:
1799                         break;
1800                 case -NFS4ERR_CB_PATH_DOWN:
1801                         nfs40_handle_cb_pathdown(clp);
1802                         break;
1803                 case -NFS4ERR_NO_GRACE:
1804                         nfs4_state_end_reclaim_reboot(clp);
1805                         break;
1806                 case -NFS4ERR_STALE_CLIENTID:
1807                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1808                         nfs4_state_start_reclaim_reboot(clp);
1809                         break;
1810                 case -NFS4ERR_EXPIRED:
1811                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1812                         nfs4_state_start_reclaim_nograce(clp);
1813                         break;
1814                 case -NFS4ERR_BADSESSION:
1815                 case -NFS4ERR_BADSLOT:
1816                 case -NFS4ERR_BAD_HIGH_SLOT:
1817                 case -NFS4ERR_DEADSESSION:
1818                 case -NFS4ERR_SEQ_FALSE_RETRY:
1819                 case -NFS4ERR_SEQ_MISORDERED:
1820                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1821                         /* Zero session reset errors */
1822                         break;
1823                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1824                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1825                         break;
1826                 default:
1827                         dprintk("%s: failed to handle error %d for server %s\n",
1828                                         __func__, error, clp->cl_hostname);
1829                         return error;
1830         }
1831         dprintk("%s: handled error %d for server %s\n", __func__, error,
1832                         clp->cl_hostname);
1833         return 0;
1834 }
1835
1836 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1837 {
1838         struct nfs4_state_owner *sp;
1839         struct nfs_server *server;
1840         struct rb_node *pos;
1841         int status = 0;
1842
1843 restart:
1844         rcu_read_lock();
1845         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1846                 nfs4_purge_state_owners(server);
1847                 spin_lock(&clp->cl_lock);
1848                 for (pos = rb_first(&server->state_owners);
1849                      pos != NULL;
1850                      pos = rb_next(pos)) {
1851                         sp = rb_entry(pos,
1852                                 struct nfs4_state_owner, so_server_node);
1853                         if (!test_and_clear_bit(ops->owner_flag_bit,
1854                                                         &sp->so_flags))
1855                                 continue;
1856                         if (!atomic_inc_not_zero(&sp->so_count))
1857                                 continue;
1858                         spin_unlock(&clp->cl_lock);
1859                         rcu_read_unlock();
1860
1861                         status = nfs4_reclaim_open_state(sp, ops);
1862                         if (status < 0) {
1863                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1864                                 nfs4_put_state_owner(sp);
1865                                 status = nfs4_recovery_handle_error(clp, status);
1866                                 return (status != 0) ? status : -EAGAIN;
1867                         }
1868
1869                         nfs4_put_state_owner(sp);
1870                         goto restart;
1871                 }
1872                 spin_unlock(&clp->cl_lock);
1873         }
1874         rcu_read_unlock();
1875         return 0;
1876 }
1877
1878 static int nfs4_check_lease(struct nfs_client *clp)
1879 {
1880         struct rpc_cred *cred;
1881         const struct nfs4_state_maintenance_ops *ops =
1882                 clp->cl_mvops->state_renewal_ops;
1883         int status;
1884
1885         /* Is the client already known to have an expired lease? */
1886         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1887                 return 0;
1888         spin_lock(&clp->cl_lock);
1889         cred = ops->get_state_renewal_cred_locked(clp);
1890         spin_unlock(&clp->cl_lock);
1891         if (cred == NULL) {
1892                 cred = nfs4_get_clid_cred(clp);
1893                 status = -ENOKEY;
1894                 if (cred == NULL)
1895                         goto out;
1896         }
1897         status = ops->renew_lease(clp, cred);
1898         put_rpccred(cred);
1899         if (status == -ETIMEDOUT) {
1900                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1901                 return 0;
1902         }
1903 out:
1904         return nfs4_recovery_handle_error(clp, status);
1905 }
1906
1907 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1908  * and for recoverable errors on EXCHANGE_ID for v4.1
1909  */
1910 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1911 {
1912         switch (status) {
1913         case -NFS4ERR_SEQ_MISORDERED:
1914                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1915                         return -ESERVERFAULT;
1916                 /* Lease confirmation error: retry after purging the lease */
1917                 ssleep(1);
1918                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1919                 break;
1920         case -NFS4ERR_STALE_CLIENTID:
1921                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1922                 nfs4_state_start_reclaim_reboot(clp);
1923                 break;
1924         case -NFS4ERR_CLID_INUSE:
1925                 pr_err("NFS: Server %s reports our clientid is in use\n",
1926                         clp->cl_hostname);
1927                 nfs_mark_client_ready(clp, -EPERM);
1928                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1929                 return -EPERM;
1930         case -EACCES:
1931         case -NFS4ERR_DELAY:
1932         case -ETIMEDOUT:
1933         case -EAGAIN:
1934                 ssleep(1);
1935                 break;
1936
1937         case -NFS4ERR_MINOR_VERS_MISMATCH:
1938                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1939                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1940                 dprintk("%s: exit with error %d for server %s\n",
1941                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1942                 return -EPROTONOSUPPORT;
1943         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1944                                  * in nfs4_exchange_id */
1945         default:
1946                 dprintk("%s: exit with error %d for server %s\n", __func__,
1947                                 status, clp->cl_hostname);
1948                 return status;
1949         }
1950         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1951         dprintk("%s: handled error %d for server %s\n", __func__, status,
1952                         clp->cl_hostname);
1953         return 0;
1954 }
1955
1956 static int nfs4_establish_lease(struct nfs_client *clp)
1957 {
1958         struct rpc_cred *cred;
1959         const struct nfs4_state_recovery_ops *ops =
1960                 clp->cl_mvops->reboot_recovery_ops;
1961         int status;
1962
1963         status = nfs4_begin_drain_session(clp);
1964         if (status != 0)
1965                 return status;
1966         cred = nfs4_get_clid_cred(clp);
1967         if (cred == NULL)
1968                 return -ENOENT;
1969         status = ops->establish_clid(clp, cred);
1970         put_rpccred(cred);
1971         if (status != 0)
1972                 return status;
1973         pnfs_destroy_all_layouts(clp);
1974         return 0;
1975 }
1976
1977 /*
1978  * Returns zero or a negative errno.  NFS4ERR values are converted
1979  * to local errno values.
1980  */
1981 static int nfs4_reclaim_lease(struct nfs_client *clp)
1982 {
1983         int status;
1984
1985         status = nfs4_establish_lease(clp);
1986         if (status < 0)
1987                 return nfs4_handle_reclaim_lease_error(clp, status);
1988         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1989                 nfs4_state_start_reclaim_nograce(clp);
1990         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1991                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1992         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1993         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1994         return 0;
1995 }
1996
1997 static int nfs4_purge_lease(struct nfs_client *clp)
1998 {
1999         int status;
2000
2001         status = nfs4_establish_lease(clp);
2002         if (status < 0)
2003                 return nfs4_handle_reclaim_lease_error(clp, status);
2004         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2005         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2006         nfs4_state_start_reclaim_nograce(clp);
2007         return 0;
2008 }
2009
2010 /*
2011  * Try remote migration of one FSID from a source server to a
2012  * destination server.  The source server provides a list of
2013  * potential destinations.
2014  *
2015  * Returns zero or a negative NFS4ERR status code.
2016  */
2017 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
2018 {
2019         struct nfs_client *clp = server->nfs_client;
2020         struct nfs4_fs_locations *locations = NULL;
2021         struct inode *inode;
2022         struct page *page;
2023         int status, result;
2024
2025         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2026                         (unsigned long long)server->fsid.major,
2027                         (unsigned long long)server->fsid.minor,
2028                         clp->cl_hostname);
2029
2030         result = 0;
2031         page = alloc_page(GFP_KERNEL);
2032         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2033         if (page == NULL || locations == NULL) {
2034                 dprintk("<-- %s: no memory\n", __func__);
2035                 goto out;
2036         }
2037
2038         inode = d_inode(server->super->s_root);
2039         result = nfs4_proc_get_locations(inode, locations, page, cred);
2040         if (result) {
2041                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2042                         __func__, result);
2043                 goto out;
2044         }
2045
2046         result = -NFS4ERR_NXIO;
2047         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2048                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2049                         __func__);
2050                 goto out;
2051         }
2052
2053         status = nfs4_begin_drain_session(clp);
2054         if (status != 0)
2055                 return status;
2056
2057         status = nfs4_replace_transport(server, locations);
2058         if (status != 0) {
2059                 dprintk("<-- %s: failed to replace transport: %d\n",
2060                         __func__, status);
2061                 goto out;
2062         }
2063
2064         result = 0;
2065         dprintk("<-- %s: migration succeeded\n", __func__);
2066
2067 out:
2068         if (page != NULL)
2069                 __free_page(page);
2070         kfree(locations);
2071         if (result) {
2072                 pr_err("NFS: migration recovery failed (server %s)\n",
2073                                 clp->cl_hostname);
2074                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2075         }
2076         return result;
2077 }
2078
2079 /*
2080  * Returns zero or a negative NFS4ERR status code.
2081  */
2082 static int nfs4_handle_migration(struct nfs_client *clp)
2083 {
2084         const struct nfs4_state_maintenance_ops *ops =
2085                                 clp->cl_mvops->state_renewal_ops;
2086         struct nfs_server *server;
2087         struct rpc_cred *cred;
2088
2089         dprintk("%s: migration reported on \"%s\"\n", __func__,
2090                         clp->cl_hostname);
2091
2092         spin_lock(&clp->cl_lock);
2093         cred = ops->get_state_renewal_cred_locked(clp);
2094         spin_unlock(&clp->cl_lock);
2095         if (cred == NULL)
2096                 return -NFS4ERR_NOENT;
2097
2098         clp->cl_mig_gen++;
2099 restart:
2100         rcu_read_lock();
2101         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2102                 int status;
2103
2104                 if (server->mig_gen == clp->cl_mig_gen)
2105                         continue;
2106                 server->mig_gen = clp->cl_mig_gen;
2107
2108                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2109                                                 &server->mig_status))
2110                         continue;
2111
2112                 rcu_read_unlock();
2113                 status = nfs4_try_migration(server, cred);
2114                 if (status < 0) {
2115                         put_rpccred(cred);
2116                         return status;
2117                 }
2118                 goto restart;
2119         }
2120         rcu_read_unlock();
2121         put_rpccred(cred);
2122         return 0;
2123 }
2124
2125 /*
2126  * Test each nfs_server on the clp's cl_superblocks list to see
2127  * if it's moved to another server.  Stop when the server no longer
2128  * returns NFS4ERR_LEASE_MOVED.
2129  */
2130 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2131 {
2132         const struct nfs4_state_maintenance_ops *ops =
2133                                 clp->cl_mvops->state_renewal_ops;
2134         struct nfs_server *server;
2135         struct rpc_cred *cred;
2136
2137         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2138                         clp->cl_hostname);
2139
2140         spin_lock(&clp->cl_lock);
2141         cred = ops->get_state_renewal_cred_locked(clp);
2142         spin_unlock(&clp->cl_lock);
2143         if (cred == NULL)
2144                 return -NFS4ERR_NOENT;
2145
2146         clp->cl_mig_gen++;
2147 restart:
2148         rcu_read_lock();
2149         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2150                 struct inode *inode;
2151                 int status;
2152
2153                 if (server->mig_gen == clp->cl_mig_gen)
2154                         continue;
2155                 server->mig_gen = clp->cl_mig_gen;
2156
2157                 rcu_read_unlock();
2158
2159                 inode = d_inode(server->super->s_root);
2160                 status = nfs4_proc_fsid_present(inode, cred);
2161                 if (status != -NFS4ERR_MOVED)
2162                         goto restart;   /* wasn't this one */
2163                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2164                         goto restart;   /* there are more */
2165                 goto out;
2166         }
2167         rcu_read_unlock();
2168
2169 out:
2170         put_rpccred(cred);
2171         return 0;
2172 }
2173
2174 /**
2175  * nfs4_discover_server_trunking - Detect server IP address trunking
2176  *
2177  * @clp: nfs_client under test
2178  * @result: OUT: found nfs_client, or clp
2179  *
2180  * Returns zero or a negative errno.  If zero is returned,
2181  * an nfs_client pointer is planted in "result".
2182  *
2183  * Note: since we are invoked in process context, and
2184  * not from inside the state manager, we cannot use
2185  * nfs4_handle_reclaim_lease_error().
2186  */
2187 int nfs4_discover_server_trunking(struct nfs_client *clp,
2188                                   struct nfs_client **result)
2189 {
2190         const struct nfs4_state_recovery_ops *ops =
2191                                 clp->cl_mvops->reboot_recovery_ops;
2192         struct rpc_clnt *clnt;
2193         struct rpc_cred *cred;
2194         int i, status;
2195
2196         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2197
2198         clnt = clp->cl_rpcclient;
2199         i = 0;
2200
2201         mutex_lock(&nfs_clid_init_mutex);
2202 again:
2203         status  = -ENOENT;
2204         cred = nfs4_get_clid_cred(clp);
2205         if (cred == NULL)
2206                 goto out_unlock;
2207
2208         status = ops->detect_trunking(clp, result, cred);
2209         put_rpccred(cred);
2210         switch (status) {
2211         case 0:
2212         case -EINTR:
2213         case -ERESTARTSYS:
2214                 break;
2215         case -ETIMEDOUT:
2216                 if (clnt->cl_softrtry)
2217                         break;
2218                 /* Fall through */
2219         case -NFS4ERR_DELAY:
2220         case -EAGAIN:
2221                 ssleep(1);
2222                 /* Fall through */
2223         case -NFS4ERR_STALE_CLIENTID:
2224                 dprintk("NFS: %s after status %d, retrying\n",
2225                         __func__, status);
2226                 goto again;
2227         case -EACCES:
2228                 if (i++ == 0) {
2229                         nfs4_root_machine_cred(clp);
2230                         goto again;
2231                 }
2232                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2233                         break;
2234                 /* Fall through */
2235         case -NFS4ERR_CLID_INUSE:
2236         case -NFS4ERR_WRONGSEC:
2237                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2238                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2239                         status = -EPERM;
2240                         break;
2241                 }
2242                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2243                 if (IS_ERR(clnt)) {
2244                         status = PTR_ERR(clnt);
2245                         break;
2246                 }
2247                 /* Note: this is safe because we haven't yet marked the
2248                  * client as ready, so we are the only user of
2249                  * clp->cl_rpcclient
2250                  */
2251                 clnt = xchg(&clp->cl_rpcclient, clnt);
2252                 rpc_shutdown_client(clnt);
2253                 clnt = clp->cl_rpcclient;
2254                 goto again;
2255
2256         case -NFS4ERR_MINOR_VERS_MISMATCH:
2257                 status = -EPROTONOSUPPORT;
2258                 break;
2259
2260         case -EKEYEXPIRED:
2261         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2262                                  * in nfs4_exchange_id */
2263                 status = -EKEYEXPIRED;
2264                 break;
2265         default:
2266                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2267                                 __func__, status);
2268                 status = -EIO;
2269         }
2270
2271 out_unlock:
2272         mutex_unlock(&nfs_clid_init_mutex);
2273         dprintk("NFS: %s: status = %d\n", __func__, status);
2274         return status;
2275 }
2276
2277 #ifdef CONFIG_NFS_V4_1
2278 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2279 {
2280         struct nfs_client *clp = session->clp;
2281
2282         switch (err) {
2283         default:
2284                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2285                 break;
2286         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2287                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2288         }
2289         nfs4_schedule_state_manager(clp);
2290 }
2291 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2292
2293 void nfs41_notify_server(struct nfs_client *clp)
2294 {
2295         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2296         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2297         nfs4_schedule_state_manager(clp);
2298 }
2299
2300 static void nfs4_reset_all_state(struct nfs_client *clp)
2301 {
2302         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2303                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2304                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2305                 nfs4_state_start_reclaim_nograce(clp);
2306                 dprintk("%s: scheduling reset of all state for server %s!\n",
2307                                 __func__, clp->cl_hostname);
2308                 nfs4_schedule_state_manager(clp);
2309         }
2310 }
2311
2312 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2313 {
2314         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2315                 nfs4_state_start_reclaim_reboot(clp);
2316                 dprintk("%s: server %s rebooted!\n", __func__,
2317                                 clp->cl_hostname);
2318                 nfs4_schedule_state_manager(clp);
2319         }
2320 }
2321
2322 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2323 {
2324         nfs4_reset_all_state(clp);
2325         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2326 }
2327
2328 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2329 {
2330         nfs4_state_start_reclaim_nograce(clp);
2331         nfs4_schedule_state_manager(clp);
2332
2333         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2334 }
2335
2336 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2337 {
2338         /* FIXME: For now, we destroy all layouts. */
2339         pnfs_destroy_all_layouts(clp);
2340         /* FIXME: For now, we test all delegations+open state+locks. */
2341         nfs41_handle_some_state_revoked(clp);
2342         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2343                         clp->cl_hostname);
2344 }
2345
2346 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2347 {
2348         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2349         nfs4_schedule_state_manager(clp);
2350
2351         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2352                         clp->cl_hostname);
2353 }
2354
2355 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2356 {
2357         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2358                 &clp->cl_state) == 0)
2359                 nfs4_schedule_state_manager(clp);
2360 }
2361
2362 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2363                 bool recovery)
2364 {
2365         if (!flags)
2366                 return;
2367
2368         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2369                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2370         /*
2371          * If we're called from the state manager thread, then assume we're
2372          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2373          * Those flags are expected to remain set until we're done
2374          * recovering (see RFC5661, section 18.46.3).
2375          */
2376         if (recovery)
2377                 goto out_recovery;
2378
2379         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2380                 nfs41_handle_server_reboot(clp);
2381         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2382                 nfs41_handle_all_state_revoked(clp);
2383         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2384                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2385                 nfs41_handle_some_state_revoked(clp);
2386         if (flags & SEQ4_STATUS_LEASE_MOVED)
2387                 nfs4_schedule_lease_moved_recovery(clp);
2388         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2389                 nfs41_handle_recallable_state_revoked(clp);
2390 out_recovery:
2391         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2392                 nfs41_handle_backchannel_fault(clp);
2393         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2394                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2395                 nfs41_handle_cb_path_down(clp);
2396 }
2397
2398 static int nfs4_reset_session(struct nfs_client *clp)
2399 {
2400         struct rpc_cred *cred;
2401         int status;
2402
2403         if (!nfs4_has_session(clp))
2404                 return 0;
2405         status = nfs4_begin_drain_session(clp);
2406         if (status != 0)
2407                 return status;
2408         cred = nfs4_get_clid_cred(clp);
2409         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2410         switch (status) {
2411         case 0:
2412         case -NFS4ERR_BADSESSION:
2413         case -NFS4ERR_DEADSESSION:
2414                 break;
2415         case -NFS4ERR_BACK_CHAN_BUSY:
2416         case -NFS4ERR_DELAY:
2417                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2418                 status = 0;
2419                 ssleep(1);
2420                 goto out;
2421         default:
2422                 status = nfs4_recovery_handle_error(clp, status);
2423                 goto out;
2424         }
2425
2426         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2427         status = nfs4_proc_create_session(clp, cred);
2428         if (status) {
2429                 dprintk("%s: session reset failed with status %d for server %s!\n",
2430                         __func__, status, clp->cl_hostname);
2431                 status = nfs4_handle_reclaim_lease_error(clp, status);
2432                 goto out;
2433         }
2434         nfs41_finish_session_reset(clp);
2435         dprintk("%s: session reset was successful for server %s!\n",
2436                         __func__, clp->cl_hostname);
2437 out:
2438         if (cred)
2439                 put_rpccred(cred);
2440         return status;
2441 }
2442
2443 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2444 {
2445         struct rpc_cred *cred;
2446         int ret;
2447
2448         if (!nfs4_has_session(clp))
2449                 return 0;
2450         ret = nfs4_begin_drain_session(clp);
2451         if (ret != 0)
2452                 return ret;
2453         cred = nfs4_get_clid_cred(clp);
2454         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2455         if (cred)
2456                 put_rpccred(cred);
2457         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2458         switch (ret) {
2459         case 0:
2460                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2461                         __func__, clp->cl_hostname);
2462                 break;
2463         case -NFS4ERR_DELAY:
2464                 ssleep(1);
2465                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2466                 break;
2467         default:
2468                 return nfs4_recovery_handle_error(clp, ret);
2469         }
2470         return 0;
2471 }
2472 #else /* CONFIG_NFS_V4_1 */
2473 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2474
2475 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2476 {
2477         return 0;
2478 }
2479 #endif /* CONFIG_NFS_V4_1 */
2480
2481 static void nfs4_state_manager(struct nfs_client *clp)
2482 {
2483         int status = 0;
2484         const char *section = "", *section_sep = "";
2485
2486         /* Ensure exclusive access to NFSv4 state */
2487         do {
2488                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2489                         section = "purge state";
2490                         status = nfs4_purge_lease(clp);
2491                         if (status < 0)
2492                                 goto out_error;
2493                         continue;
2494                 }
2495
2496                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2497                         section = "lease expired";
2498                         /* We're going to have to re-establish a clientid */
2499                         status = nfs4_reclaim_lease(clp);
2500                         if (status < 0)
2501                                 goto out_error;
2502                         continue;
2503                 }
2504
2505                 /* Initialize or reset the session */
2506                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2507                         section = "reset session";
2508                         status = nfs4_reset_session(clp);
2509                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2510                                 continue;
2511                         if (status < 0)
2512                                 goto out_error;
2513                 }
2514
2515                 /* Send BIND_CONN_TO_SESSION */
2516                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2517                                 &clp->cl_state)) {
2518                         section = "bind conn to session";
2519                         status = nfs4_bind_conn_to_session(clp);
2520                         if (status < 0)
2521                                 goto out_error;
2522                         continue;
2523                 }
2524
2525                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2526                         section = "check lease";
2527                         status = nfs4_check_lease(clp);
2528                         if (status < 0)
2529                                 goto out_error;
2530                         continue;
2531                 }
2532
2533                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2534                         section = "migration";
2535                         status = nfs4_handle_migration(clp);
2536                         if (status < 0)
2537                                 goto out_error;
2538                 }
2539
2540                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2541                         section = "lease moved";
2542                         status = nfs4_handle_lease_moved(clp);
2543                         if (status < 0)
2544                                 goto out_error;
2545                 }
2546
2547                 /* First recover reboot state... */
2548                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2549                         section = "reclaim reboot";
2550                         status = nfs4_do_reclaim(clp,
2551                                 clp->cl_mvops->reboot_recovery_ops);
2552                         if (status == -EAGAIN)
2553                                 continue;
2554                         if (status < 0)
2555                                 goto out_error;
2556                         nfs4_state_end_reclaim_reboot(clp);
2557                 }
2558
2559                 /* Detect expired delegations... */
2560                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2561                         section = "detect expired delegations";
2562                         nfs_reap_expired_delegations(clp);
2563                         continue;
2564                 }
2565
2566                 /* Now recover expired state... */
2567                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2568                         section = "reclaim nograce";
2569                         status = nfs4_do_reclaim(clp,
2570                                 clp->cl_mvops->nograce_recovery_ops);
2571                         if (status == -EAGAIN)
2572                                 continue;
2573                         if (status < 0)
2574                                 goto out_error;
2575                 }
2576
2577                 nfs4_end_drain_session(clp);
2578                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2579                         nfs_client_return_marked_delegations(clp);
2580                         continue;
2581                 }
2582
2583                 nfs4_clear_state_manager_bit(clp);
2584                 /* Did we race with an attempt to give us more work? */
2585                 if (clp->cl_state == 0)
2586                         break;
2587                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2588                         break;
2589         } while (refcount_read(&clp->cl_count) > 1);
2590         return;
2591 out_error:
2592         if (strlen(section))
2593                 section_sep = ": ";
2594         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2595                         " with error %d\n", section_sep, section,
2596                         clp->cl_hostname, -status);
2597         ssleep(1);
2598         nfs4_end_drain_session(clp);
2599         nfs4_clear_state_manager_bit(clp);
2600 }
2601
2602 static int nfs4_run_state_manager(void *ptr)
2603 {
2604         struct nfs_client *clp = ptr;
2605
2606         allow_signal(SIGKILL);
2607         nfs4_state_manager(clp);
2608         nfs_put_client(clp);
2609         module_put_and_exit(0);
2610         return 0;
2611 }
2612
2613 /*
2614  * Local variables:
2615  *  c-basic-offset: 8
2616  * End:
2617  */