Kernel bump from 4.1.3-rt to 4.1.7-rt.
[kvmfornfv.git] / kernel / 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 static DEFINE_MUTEX(nfs_clid_init_mutex);
70
71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
72 {
73         struct nfs4_setclientid_res clid = {
74                 .clientid = clp->cl_clientid,
75                 .confirm = clp->cl_confirm,
76         };
77         unsigned short port;
78         int status;
79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
80
81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
82                 goto do_confirm;
83         port = nn->nfs_callback_tcpport;
84         if (clp->cl_addr.ss_family == AF_INET6)
85                 port = nn->nfs_callback_tcpport6;
86
87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
88         if (status != 0)
89                 goto out;
90         clp->cl_clientid = clid.clientid;
91         clp->cl_confirm = clid.confirm;
92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
93 do_confirm:
94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
95         if (status != 0)
96                 goto out;
97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
98         nfs4_schedule_state_renewal(clp);
99 out:
100         return status;
101 }
102
103 /**
104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
105  *
106  * @clp: nfs_client under test
107  * @result: OUT: found nfs_client, or clp
108  * @cred: credential to use for trunking test
109  *
110  * Returns zero, a negative errno, or a negative NFS4ERR status.
111  * If zero is returned, an nfs_client pointer is planted in
112  * "result".
113  *
114  * Note: The returned client may not yet be marked ready.
115  */
116 int nfs40_discover_server_trunking(struct nfs_client *clp,
117                                    struct nfs_client **result,
118                                    struct rpc_cred *cred)
119 {
120         struct nfs4_setclientid_res clid = {
121                 .clientid = clp->cl_clientid,
122                 .confirm = clp->cl_confirm,
123         };
124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
125         unsigned short port;
126         int status;
127
128         port = nn->nfs_callback_tcpport;
129         if (clp->cl_addr.ss_family == AF_INET6)
130                 port = nn->nfs_callback_tcpport6;
131
132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
133         if (status != 0)
134                 goto out;
135         clp->cl_clientid = clid.clientid;
136         clp->cl_confirm = clid.confirm;
137
138         status = nfs40_walk_client_list(clp, result, cred);
139         if (status == 0) {
140                 /* Sustain the lease, even if it's empty.  If the clientid4
141                  * goes stale it's of no use for trunking discovery. */
142                 nfs4_schedule_state_renewal(*result);
143         }
144 out:
145         return status;
146 }
147
148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
149 {
150         struct rpc_cred *cred = NULL;
151
152         if (clp->cl_machine_cred != NULL)
153                 cred = get_rpccred(clp->cl_machine_cred);
154         return cred;
155 }
156
157 static void nfs4_root_machine_cred(struct nfs_client *clp)
158 {
159         struct rpc_cred *cred, *new;
160
161         new = rpc_lookup_machine_cred(NULL);
162         spin_lock(&clp->cl_lock);
163         cred = clp->cl_machine_cred;
164         clp->cl_machine_cred = new;
165         spin_unlock(&clp->cl_lock);
166         if (cred != NULL)
167                 put_rpccred(cred);
168 }
169
170 static struct rpc_cred *
171 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
172 {
173         struct rpc_cred *cred = NULL;
174         struct nfs4_state_owner *sp;
175         struct rb_node *pos;
176
177         for (pos = rb_first(&server->state_owners);
178              pos != NULL;
179              pos = rb_next(pos)) {
180                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
181                 if (list_empty(&sp->so_states))
182                         continue;
183                 cred = get_rpccred(sp->so_cred);
184                 break;
185         }
186         return cred;
187 }
188
189 /**
190  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
191  * @clp: client state handle
192  *
193  * Returns an rpc_cred with reference count bumped, or NULL.
194  * Caller must hold clp->cl_lock.
195  */
196 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
197 {
198         struct rpc_cred *cred = NULL;
199         struct nfs_server *server;
200
201         /* Use machine credentials if available */
202         cred = nfs4_get_machine_cred_locked(clp);
203         if (cred != NULL)
204                 goto out;
205
206         rcu_read_lock();
207         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
208                 cred = nfs4_get_renew_cred_server_locked(server);
209                 if (cred != NULL)
210                         break;
211         }
212         rcu_read_unlock();
213
214 out:
215         return cred;
216 }
217
218 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
219 {
220         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
221                 spin_lock(&tbl->slot_tbl_lock);
222                 nfs41_wake_slot_table(tbl);
223                 spin_unlock(&tbl->slot_tbl_lock);
224         }
225 }
226
227 static void nfs4_end_drain_session(struct nfs_client *clp)
228 {
229         struct nfs4_session *ses = clp->cl_session;
230
231         if (clp->cl_slot_tbl) {
232                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
233                 return;
234         }
235
236         if (ses != NULL) {
237                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
238                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
239         }
240 }
241
242 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
243 {
244         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
245         spin_lock(&tbl->slot_tbl_lock);
246         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
247                 reinit_completion(&tbl->complete);
248                 spin_unlock(&tbl->slot_tbl_lock);
249                 return wait_for_completion_interruptible(&tbl->complete);
250         }
251         spin_unlock(&tbl->slot_tbl_lock);
252         return 0;
253 }
254
255 static int nfs4_begin_drain_session(struct nfs_client *clp)
256 {
257         struct nfs4_session *ses = clp->cl_session;
258         int ret = 0;
259
260         if (clp->cl_slot_tbl)
261                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
262
263         /* back channel */
264         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
265         if (ret)
266                 return ret;
267         /* fore channel */
268         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
269 }
270
271 #if defined(CONFIG_NFS_V4_1)
272
273 static int nfs41_setup_state_renewal(struct nfs_client *clp)
274 {
275         int status;
276         struct nfs_fsinfo fsinfo;
277
278         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
279                 nfs4_schedule_state_renewal(clp);
280                 return 0;
281         }
282
283         status = nfs4_proc_get_lease_time(clp, &fsinfo);
284         if (status == 0) {
285                 /* Update lease time and schedule renewal */
286                 spin_lock(&clp->cl_lock);
287                 clp->cl_lease_time = fsinfo.lease_time * HZ;
288                 clp->cl_last_renewal = jiffies;
289                 spin_unlock(&clp->cl_lock);
290
291                 nfs4_schedule_state_renewal(clp);
292         }
293
294         return status;
295 }
296
297 static void nfs41_finish_session_reset(struct nfs_client *clp)
298 {
299         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301         /* create_session negotiated new slot table */
302         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303         nfs41_setup_state_renewal(clp);
304 }
305
306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
307 {
308         int status;
309
310         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311                 goto do_confirm;
312         nfs4_begin_drain_session(clp);
313         status = nfs4_proc_exchange_id(clp, cred);
314         if (status != 0)
315                 goto out;
316         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
317 do_confirm:
318         status = nfs4_proc_create_session(clp, cred);
319         if (status != 0)
320                 goto out;
321         nfs41_finish_session_reset(clp);
322         nfs_mark_client_ready(clp, NFS_CS_READY);
323 out:
324         return status;
325 }
326
327 /**
328  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
329  *
330  * @clp: nfs_client under test
331  * @result: OUT: found nfs_client, or clp
332  * @cred: credential to use for trunking test
333  *
334  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
335  * If NFS4_OK is returned, an nfs_client pointer is planted in
336  * "result".
337  *
338  * Note: The returned client may not yet be marked ready.
339  */
340 int nfs41_discover_server_trunking(struct nfs_client *clp,
341                                    struct nfs_client **result,
342                                    struct rpc_cred *cred)
343 {
344         int status;
345
346         status = nfs4_proc_exchange_id(clp, cred);
347         if (status != NFS4_OK)
348                 return status;
349
350         status = nfs41_walk_client_list(clp, result, cred);
351         if (status < 0)
352                 return status;
353         if (clp != *result)
354                 return 0;
355
356         /* Purge state if the client id was established in a prior instance */
357         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
358                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
359         else
360                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
361         nfs4_schedule_state_manager(clp);
362         status = nfs_wait_client_init_complete(clp);
363         if (status < 0)
364                 nfs_put_client(clp);
365         return status;
366 }
367
368 #endif /* CONFIG_NFS_V4_1 */
369
370 /**
371  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
372  * @clp: client state handle
373  *
374  * Returns an rpc_cred with reference count bumped, or NULL.
375  */
376 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
377 {
378         struct rpc_cred *cred;
379
380         spin_lock(&clp->cl_lock);
381         cred = nfs4_get_machine_cred_locked(clp);
382         spin_unlock(&clp->cl_lock);
383         return cred;
384 }
385
386 static struct nfs4_state_owner *
387 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
388 {
389         struct rb_node **p = &server->state_owners.rb_node,
390                        *parent = NULL;
391         struct nfs4_state_owner *sp;
392
393         while (*p != NULL) {
394                 parent = *p;
395                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
396
397                 if (cred < sp->so_cred)
398                         p = &parent->rb_left;
399                 else if (cred > sp->so_cred)
400                         p = &parent->rb_right;
401                 else {
402                         if (!list_empty(&sp->so_lru))
403                                 list_del_init(&sp->so_lru);
404                         atomic_inc(&sp->so_count);
405                         return sp;
406                 }
407         }
408         return NULL;
409 }
410
411 static struct nfs4_state_owner *
412 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
413 {
414         struct nfs_server *server = new->so_server;
415         struct rb_node **p = &server->state_owners.rb_node,
416                        *parent = NULL;
417         struct nfs4_state_owner *sp;
418         int err;
419
420         while (*p != NULL) {
421                 parent = *p;
422                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
423
424                 if (new->so_cred < sp->so_cred)
425                         p = &parent->rb_left;
426                 else if (new->so_cred > sp->so_cred)
427                         p = &parent->rb_right;
428                 else {
429                         if (!list_empty(&sp->so_lru))
430                                 list_del_init(&sp->so_lru);
431                         atomic_inc(&sp->so_count);
432                         return sp;
433                 }
434         }
435         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
436         if (err)
437                 return ERR_PTR(err);
438         rb_link_node(&new->so_server_node, parent, p);
439         rb_insert_color(&new->so_server_node, &server->state_owners);
440         return new;
441 }
442
443 static void
444 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
445 {
446         struct nfs_server *server = sp->so_server;
447
448         if (!RB_EMPTY_NODE(&sp->so_server_node))
449                 rb_erase(&sp->so_server_node, &server->state_owners);
450         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
451 }
452
453 static void
454 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
455 {
456         sc->create_time = ktime_get();
457         sc->flags = 0;
458         sc->counter = 0;
459         spin_lock_init(&sc->lock);
460         INIT_LIST_HEAD(&sc->list);
461         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
462 }
463
464 static void
465 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
466 {
467         rpc_destroy_wait_queue(&sc->wait);
468 }
469
470 /*
471  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
472  * create a new state_owner.
473  *
474  */
475 static struct nfs4_state_owner *
476 nfs4_alloc_state_owner(struct nfs_server *server,
477                 struct rpc_cred *cred,
478                 gfp_t gfp_flags)
479 {
480         struct nfs4_state_owner *sp;
481
482         sp = kzalloc(sizeof(*sp), gfp_flags);
483         if (!sp)
484                 return NULL;
485         sp->so_server = server;
486         sp->so_cred = get_rpccred(cred);
487         spin_lock_init(&sp->so_lock);
488         INIT_LIST_HEAD(&sp->so_states);
489         nfs4_init_seqid_counter(&sp->so_seqid);
490         atomic_set(&sp->so_count, 1);
491         INIT_LIST_HEAD(&sp->so_lru);
492         seqcount_init(&sp->so_reclaim_seqcount);
493         mutex_init(&sp->so_delegreturn_mutex);
494         return sp;
495 }
496
497 static void
498 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
499 {
500         struct rb_node *rb_node = &sp->so_server_node;
501
502         if (!RB_EMPTY_NODE(rb_node)) {
503                 struct nfs_server *server = sp->so_server;
504                 struct nfs_client *clp = server->nfs_client;
505
506                 spin_lock(&clp->cl_lock);
507                 if (!RB_EMPTY_NODE(rb_node)) {
508                         rb_erase(rb_node, &server->state_owners);
509                         RB_CLEAR_NODE(rb_node);
510                 }
511                 spin_unlock(&clp->cl_lock);
512         }
513 }
514
515 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
516 {
517         nfs4_destroy_seqid_counter(&sp->so_seqid);
518         put_rpccred(sp->so_cred);
519         kfree(sp);
520 }
521
522 static void nfs4_gc_state_owners(struct nfs_server *server)
523 {
524         struct nfs_client *clp = server->nfs_client;
525         struct nfs4_state_owner *sp, *tmp;
526         unsigned long time_min, time_max;
527         LIST_HEAD(doomed);
528
529         spin_lock(&clp->cl_lock);
530         time_max = jiffies;
531         time_min = (long)time_max - (long)clp->cl_lease_time;
532         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
533                 /* NB: LRU is sorted so that oldest is at the head */
534                 if (time_in_range(sp->so_expires, time_min, time_max))
535                         break;
536                 list_move(&sp->so_lru, &doomed);
537                 nfs4_remove_state_owner_locked(sp);
538         }
539         spin_unlock(&clp->cl_lock);
540
541         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
542                 list_del(&sp->so_lru);
543                 nfs4_free_state_owner(sp);
544         }
545 }
546
547 /**
548  * nfs4_get_state_owner - Look up a state owner given a credential
549  * @server: nfs_server to search
550  * @cred: RPC credential to match
551  *
552  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
553  */
554 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
555                                               struct rpc_cred *cred,
556                                               gfp_t gfp_flags)
557 {
558         struct nfs_client *clp = server->nfs_client;
559         struct nfs4_state_owner *sp, *new;
560
561         spin_lock(&clp->cl_lock);
562         sp = nfs4_find_state_owner_locked(server, cred);
563         spin_unlock(&clp->cl_lock);
564         if (sp != NULL)
565                 goto out;
566         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
567         if (new == NULL)
568                 goto out;
569         do {
570                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
571                         break;
572                 spin_lock(&clp->cl_lock);
573                 sp = nfs4_insert_state_owner_locked(new);
574                 spin_unlock(&clp->cl_lock);
575         } while (sp == ERR_PTR(-EAGAIN));
576         if (sp != new)
577                 nfs4_free_state_owner(new);
578 out:
579         nfs4_gc_state_owners(server);
580         return sp;
581 }
582
583 /**
584  * nfs4_put_state_owner - Release a nfs4_state_owner
585  * @sp: state owner data to release
586  *
587  * Note that we keep released state owners on an LRU
588  * list.
589  * This caches valid state owners so that they can be
590  * reused, to avoid the OPEN_CONFIRM on minor version 0.
591  * It also pins the uniquifier of dropped state owners for
592  * a while, to ensure that those state owner names are
593  * never reused.
594  */
595 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
596 {
597         struct nfs_server *server = sp->so_server;
598         struct nfs_client *clp = server->nfs_client;
599
600         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
601                 return;
602
603         sp->so_expires = jiffies;
604         list_add_tail(&sp->so_lru, &server->state_owners_lru);
605         spin_unlock(&clp->cl_lock);
606 }
607
608 /**
609  * nfs4_purge_state_owners - Release all cached state owners
610  * @server: nfs_server with cached state owners to release
611  *
612  * Called at umount time.  Remaining state owners will be on
613  * the LRU with ref count of zero.
614  */
615 void nfs4_purge_state_owners(struct nfs_server *server)
616 {
617         struct nfs_client *clp = server->nfs_client;
618         struct nfs4_state_owner *sp, *tmp;
619         LIST_HEAD(doomed);
620
621         spin_lock(&clp->cl_lock);
622         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
623                 list_move(&sp->so_lru, &doomed);
624                 nfs4_remove_state_owner_locked(sp);
625         }
626         spin_unlock(&clp->cl_lock);
627
628         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
629                 list_del(&sp->so_lru);
630                 nfs4_free_state_owner(sp);
631         }
632 }
633
634 static struct nfs4_state *
635 nfs4_alloc_open_state(void)
636 {
637         struct nfs4_state *state;
638
639         state = kzalloc(sizeof(*state), GFP_NOFS);
640         if (!state)
641                 return NULL;
642         atomic_set(&state->count, 1);
643         INIT_LIST_HEAD(&state->lock_states);
644         spin_lock_init(&state->state_lock);
645         seqlock_init(&state->seqlock);
646         return state;
647 }
648
649 void
650 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
651 {
652         if (state->state == fmode)
653                 return;
654         /* NB! List reordering - see the reclaim code for why.  */
655         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
656                 if (fmode & FMODE_WRITE)
657                         list_move(&state->open_states, &state->owner->so_states);
658                 else
659                         list_move_tail(&state->open_states, &state->owner->so_states);
660         }
661         state->state = fmode;
662 }
663
664 static struct nfs4_state *
665 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
666 {
667         struct nfs_inode *nfsi = NFS_I(inode);
668         struct nfs4_state *state;
669
670         list_for_each_entry(state, &nfsi->open_states, inode_states) {
671                 if (state->owner != owner)
672                         continue;
673                 if (!nfs4_valid_open_stateid(state))
674                         continue;
675                 if (atomic_inc_not_zero(&state->count))
676                         return state;
677         }
678         return NULL;
679 }
680
681 static void
682 nfs4_free_open_state(struct nfs4_state *state)
683 {
684         kfree(state);
685 }
686
687 struct nfs4_state *
688 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
689 {
690         struct nfs4_state *state, *new;
691         struct nfs_inode *nfsi = NFS_I(inode);
692
693         spin_lock(&inode->i_lock);
694         state = __nfs4_find_state_byowner(inode, owner);
695         spin_unlock(&inode->i_lock);
696         if (state)
697                 goto out;
698         new = nfs4_alloc_open_state();
699         spin_lock(&owner->so_lock);
700         spin_lock(&inode->i_lock);
701         state = __nfs4_find_state_byowner(inode, owner);
702         if (state == NULL && new != NULL) {
703                 state = new;
704                 state->owner = owner;
705                 atomic_inc(&owner->so_count);
706                 list_add(&state->inode_states, &nfsi->open_states);
707                 ihold(inode);
708                 state->inode = inode;
709                 spin_unlock(&inode->i_lock);
710                 /* Note: The reclaim code dictates that we add stateless
711                  * and read-only stateids to the end of the list */
712                 list_add_tail(&state->open_states, &owner->so_states);
713                 spin_unlock(&owner->so_lock);
714         } else {
715                 spin_unlock(&inode->i_lock);
716                 spin_unlock(&owner->so_lock);
717                 if (new)
718                         nfs4_free_open_state(new);
719         }
720 out:
721         return state;
722 }
723
724 void nfs4_put_open_state(struct nfs4_state *state)
725 {
726         struct inode *inode = state->inode;
727         struct nfs4_state_owner *owner = state->owner;
728
729         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
730                 return;
731         spin_lock(&inode->i_lock);
732         list_del(&state->inode_states);
733         list_del(&state->open_states);
734         spin_unlock(&inode->i_lock);
735         spin_unlock(&owner->so_lock);
736         iput(inode);
737         nfs4_free_open_state(state);
738         nfs4_put_state_owner(owner);
739 }
740
741 /*
742  * Close the current file.
743  */
744 static void __nfs4_close(struct nfs4_state *state,
745                 fmode_t fmode, gfp_t gfp_mask, int wait)
746 {
747         struct nfs4_state_owner *owner = state->owner;
748         int call_close = 0;
749         fmode_t newstate;
750
751         atomic_inc(&owner->so_count);
752         /* Protect against nfs4_find_state() */
753         spin_lock(&owner->so_lock);
754         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
755                 case FMODE_READ:
756                         state->n_rdonly--;
757                         break;
758                 case FMODE_WRITE:
759                         state->n_wronly--;
760                         break;
761                 case FMODE_READ|FMODE_WRITE:
762                         state->n_rdwr--;
763         }
764         newstate = FMODE_READ|FMODE_WRITE;
765         if (state->n_rdwr == 0) {
766                 if (state->n_rdonly == 0) {
767                         newstate &= ~FMODE_READ;
768                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
769                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
770                 }
771                 if (state->n_wronly == 0) {
772                         newstate &= ~FMODE_WRITE;
773                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
774                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
775                 }
776                 if (newstate == 0)
777                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
778         }
779         nfs4_state_set_mode_locked(state, newstate);
780         spin_unlock(&owner->so_lock);
781
782         if (!call_close) {
783                 nfs4_put_open_state(state);
784                 nfs4_put_state_owner(owner);
785         } else
786                 nfs4_do_close(state, gfp_mask, wait);
787 }
788
789 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
790 {
791         __nfs4_close(state, fmode, GFP_NOFS, 0);
792 }
793
794 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
795 {
796         __nfs4_close(state, fmode, GFP_KERNEL, 1);
797 }
798
799 /*
800  * Search the state->lock_states for an existing lock_owner
801  * that is compatible with current->files
802  */
803 static struct nfs4_lock_state *
804 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
805 {
806         struct nfs4_lock_state *pos;
807         list_for_each_entry(pos, &state->lock_states, ls_locks) {
808                 if (pos->ls_owner != fl_owner)
809                         continue;
810                 atomic_inc(&pos->ls_count);
811                 return pos;
812         }
813         return NULL;
814 }
815
816 /*
817  * Return a compatible lock_state. If no initialized lock_state structure
818  * exists, return an uninitialized one.
819  *
820  */
821 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
822 {
823         struct nfs4_lock_state *lsp;
824         struct nfs_server *server = state->owner->so_server;
825
826         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
827         if (lsp == NULL)
828                 return NULL;
829         nfs4_init_seqid_counter(&lsp->ls_seqid);
830         atomic_set(&lsp->ls_count, 1);
831         lsp->ls_state = state;
832         lsp->ls_owner = fl_owner;
833         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
834         if (lsp->ls_seqid.owner_id < 0)
835                 goto out_free;
836         INIT_LIST_HEAD(&lsp->ls_locks);
837         return lsp;
838 out_free:
839         kfree(lsp);
840         return NULL;
841 }
842
843 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
844 {
845         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
846         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
847         kfree(lsp);
848 }
849
850 /*
851  * Return a compatible lock_state. If no initialized lock_state structure
852  * exists, return an uninitialized one.
853  *
854  */
855 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
856 {
857         struct nfs4_lock_state *lsp, *new = NULL;
858         
859         for(;;) {
860                 spin_lock(&state->state_lock);
861                 lsp = __nfs4_find_lock_state(state, owner);
862                 if (lsp != NULL)
863                         break;
864                 if (new != NULL) {
865                         list_add(&new->ls_locks, &state->lock_states);
866                         set_bit(LK_STATE_IN_USE, &state->flags);
867                         lsp = new;
868                         new = NULL;
869                         break;
870                 }
871                 spin_unlock(&state->state_lock);
872                 new = nfs4_alloc_lock_state(state, owner);
873                 if (new == NULL)
874                         return NULL;
875         }
876         spin_unlock(&state->state_lock);
877         if (new != NULL)
878                 nfs4_free_lock_state(state->owner->so_server, new);
879         return lsp;
880 }
881
882 /*
883  * Release reference to lock_state, and free it if we see that
884  * it is no longer in use
885  */
886 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
887 {
888         struct nfs_server *server;
889         struct nfs4_state *state;
890
891         if (lsp == NULL)
892                 return;
893         state = lsp->ls_state;
894         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
895                 return;
896         list_del(&lsp->ls_locks);
897         if (list_empty(&state->lock_states))
898                 clear_bit(LK_STATE_IN_USE, &state->flags);
899         spin_unlock(&state->state_lock);
900         server = state->owner->so_server;
901         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
902                 struct nfs_client *clp = server->nfs_client;
903
904                 clp->cl_mvops->free_lock_state(server, lsp);
905         } else
906                 nfs4_free_lock_state(server, lsp);
907 }
908
909 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
910 {
911         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
912
913         dst->fl_u.nfs4_fl.owner = lsp;
914         atomic_inc(&lsp->ls_count);
915 }
916
917 static void nfs4_fl_release_lock(struct file_lock *fl)
918 {
919         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
920 }
921
922 static const struct file_lock_operations nfs4_fl_lock_ops = {
923         .fl_copy_lock = nfs4_fl_copy_lock,
924         .fl_release_private = nfs4_fl_release_lock,
925 };
926
927 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
928 {
929         struct nfs4_lock_state *lsp;
930
931         if (fl->fl_ops != NULL)
932                 return 0;
933         lsp = nfs4_get_lock_state(state, fl->fl_owner);
934         if (lsp == NULL)
935                 return -ENOMEM;
936         fl->fl_u.nfs4_fl.owner = lsp;
937         fl->fl_ops = &nfs4_fl_lock_ops;
938         return 0;
939 }
940
941 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
942                 struct nfs4_state *state,
943                 const struct nfs_lockowner *lockowner)
944 {
945         struct nfs4_lock_state *lsp;
946         fl_owner_t fl_owner;
947         int ret = -ENOENT;
948
949
950         if (lockowner == NULL)
951                 goto out;
952
953         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
954                 goto out;
955
956         fl_owner = lockowner->l_owner;
957         spin_lock(&state->state_lock);
958         lsp = __nfs4_find_lock_state(state, fl_owner);
959         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
960                 ret = -EIO;
961         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
962                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
963                 ret = 0;
964         }
965         spin_unlock(&state->state_lock);
966         nfs4_put_lock_state(lsp);
967 out:
968         return ret;
969 }
970
971 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
972 {
973         const nfs4_stateid *src;
974         int seq;
975
976         do {
977                 src = &zero_stateid;
978                 seq = read_seqbegin(&state->seqlock);
979                 if (test_bit(NFS_OPEN_STATE, &state->flags))
980                         src = &state->open_stateid;
981                 nfs4_stateid_copy(dst, src);
982         } while (read_seqretry(&state->seqlock, seq));
983 }
984
985 /*
986  * Byte-range lock aware utility to initialize the stateid of read/write
987  * requests.
988  */
989 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
990                 fmode_t fmode, const struct nfs_lockowner *lockowner)
991 {
992         int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
993         if (ret == -EIO)
994                 /* A lost lock - don't even consider delegations */
995                 goto out;
996         /* returns true if delegation stateid found and copied */
997         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
998                 ret = 0;
999                 goto out;
1000         }
1001         if (ret != -ENOENT)
1002                 /* nfs4_copy_delegation_stateid() didn't over-write
1003                  * dst, so it still has the lock stateid which we now
1004                  * choose to use.
1005                  */
1006                 goto out;
1007         nfs4_copy_open_stateid(dst, state);
1008         ret = 0;
1009 out:
1010         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1011                 dst->seqid = 0;
1012         return ret;
1013 }
1014
1015 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1016 {
1017         struct nfs_seqid *new;
1018
1019         new = kmalloc(sizeof(*new), gfp_mask);
1020         if (new == NULL)
1021                 return ERR_PTR(-ENOMEM);
1022         new->sequence = counter;
1023         INIT_LIST_HEAD(&new->list);
1024         new->task = NULL;
1025         return new;
1026 }
1027
1028 void nfs_release_seqid(struct nfs_seqid *seqid)
1029 {
1030         struct nfs_seqid_counter *sequence;
1031
1032         if (seqid == NULL || list_empty(&seqid->list))
1033                 return;
1034         sequence = seqid->sequence;
1035         spin_lock(&sequence->lock);
1036         list_del_init(&seqid->list);
1037         if (!list_empty(&sequence->list)) {
1038                 struct nfs_seqid *next;
1039
1040                 next = list_first_entry(&sequence->list,
1041                                 struct nfs_seqid, list);
1042                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1043         }
1044         spin_unlock(&sequence->lock);
1045 }
1046
1047 void nfs_free_seqid(struct nfs_seqid *seqid)
1048 {
1049         nfs_release_seqid(seqid);
1050         kfree(seqid);
1051 }
1052
1053 /*
1054  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1055  * failed with a seqid incrementing error -
1056  * see comments nfs4.h:seqid_mutating_error()
1057  */
1058 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1059 {
1060         switch (status) {
1061                 case 0:
1062                         break;
1063                 case -NFS4ERR_BAD_SEQID:
1064                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1065                                 return;
1066                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1067                                         " sequence-id error on an"
1068                                         " unconfirmed sequence %p!\n",
1069                                         seqid->sequence);
1070                 case -NFS4ERR_STALE_CLIENTID:
1071                 case -NFS4ERR_STALE_STATEID:
1072                 case -NFS4ERR_BAD_STATEID:
1073                 case -NFS4ERR_BADXDR:
1074                 case -NFS4ERR_RESOURCE:
1075                 case -NFS4ERR_NOFILEHANDLE:
1076                         /* Non-seqid mutating errors */
1077                         return;
1078         };
1079         /*
1080          * Note: no locking needed as we are guaranteed to be first
1081          * on the sequence list
1082          */
1083         seqid->sequence->counter++;
1084 }
1085
1086 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1087 {
1088         struct nfs4_state_owner *sp;
1089
1090         if (seqid == NULL)
1091                 return;
1092
1093         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1094         if (status == -NFS4ERR_BAD_SEQID)
1095                 nfs4_drop_state_owner(sp);
1096         if (!nfs4_has_session(sp->so_server->nfs_client))
1097                 nfs_increment_seqid(status, seqid);
1098 }
1099
1100 /*
1101  * Increment the seqid if the LOCK/LOCKU succeeded, or
1102  * failed with a seqid incrementing error -
1103  * see comments nfs4.h:seqid_mutating_error()
1104  */
1105 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1106 {
1107         if (seqid != NULL)
1108                 nfs_increment_seqid(status, seqid);
1109 }
1110
1111 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1112 {
1113         struct nfs_seqid_counter *sequence;
1114         int status = 0;
1115
1116         if (seqid == NULL)
1117                 goto out;
1118         sequence = seqid->sequence;
1119         spin_lock(&sequence->lock);
1120         seqid->task = task;
1121         if (list_empty(&seqid->list))
1122                 list_add_tail(&seqid->list, &sequence->list);
1123         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1124                 goto unlock;
1125         rpc_sleep_on(&sequence->wait, task, NULL);
1126         status = -EAGAIN;
1127 unlock:
1128         spin_unlock(&sequence->lock);
1129 out:
1130         return status;
1131 }
1132
1133 static int nfs4_run_state_manager(void *);
1134
1135 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1136 {
1137         smp_mb__before_atomic();
1138         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1139         smp_mb__after_atomic();
1140         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1141         rpc_wake_up(&clp->cl_rpcwaitq);
1142 }
1143
1144 /*
1145  * Schedule the nfs_client asynchronous state management routine
1146  */
1147 void nfs4_schedule_state_manager(struct nfs_client *clp)
1148 {
1149         struct task_struct *task;
1150         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1151
1152         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1153                 return;
1154         __module_get(THIS_MODULE);
1155         atomic_inc(&clp->cl_count);
1156
1157         /* The rcu_read_lock() is not strictly necessary, as the state
1158          * manager is the only thread that ever changes the rpc_xprt
1159          * after it's initialized.  At this point, we're single threaded. */
1160         rcu_read_lock();
1161         snprintf(buf, sizeof(buf), "%s-manager",
1162                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1163         rcu_read_unlock();
1164         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1165         if (IS_ERR(task)) {
1166                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1167                         __func__, PTR_ERR(task));
1168                 nfs4_clear_state_manager_bit(clp);
1169                 nfs_put_client(clp);
1170                 module_put(THIS_MODULE);
1171         }
1172 }
1173
1174 /*
1175  * Schedule a lease recovery attempt
1176  */
1177 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1178 {
1179         if (!clp)
1180                 return;
1181         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1182                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1183         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1184                         clp->cl_hostname);
1185         nfs4_schedule_state_manager(clp);
1186 }
1187 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1188
1189 /**
1190  * nfs4_schedule_migration_recovery - trigger migration recovery
1191  *
1192  * @server: FSID that is migrating
1193  *
1194  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1195  * value is returned.
1196  */
1197 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1198 {
1199         struct nfs_client *clp = server->nfs_client;
1200
1201         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1202                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1203                                 clp->cl_hostname);
1204                 return -NFS4ERR_IO;
1205         }
1206
1207         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1208                 return -NFS4ERR_IO;
1209
1210         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1211                         __func__,
1212                         (unsigned long long)server->fsid.major,
1213                         (unsigned long long)server->fsid.minor,
1214                         clp->cl_hostname);
1215
1216         set_bit(NFS_MIG_IN_TRANSITION,
1217                         &((struct nfs_server *)server)->mig_status);
1218         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1219
1220         nfs4_schedule_state_manager(clp);
1221         return 0;
1222 }
1223 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1224
1225 /**
1226  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1227  *
1228  * @clp: server to check for moved leases
1229  *
1230  */
1231 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1232 {
1233         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1234                 __func__, clp->cl_clientid, clp->cl_hostname);
1235
1236         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1237         nfs4_schedule_state_manager(clp);
1238 }
1239 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1240
1241 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1242 {
1243         int res;
1244
1245         might_sleep();
1246
1247         atomic_inc(&clp->cl_count);
1248         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1249                                  nfs_wait_bit_killable, TASK_KILLABLE);
1250         if (res)
1251                 goto out;
1252         if (clp->cl_cons_state < 0)
1253                 res = clp->cl_cons_state;
1254 out:
1255         nfs_put_client(clp);
1256         return res;
1257 }
1258
1259 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1260 {
1261         unsigned int loop;
1262         int ret;
1263
1264         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1265                 ret = nfs4_wait_clnt_recover(clp);
1266                 if (ret != 0)
1267                         break;
1268                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1269                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1270                         break;
1271                 nfs4_schedule_state_manager(clp);
1272                 ret = -EIO;
1273         }
1274         return ret;
1275 }
1276
1277 /*
1278  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1279  * @clp: client to process
1280  *
1281  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1282  * resend of the SETCLIENTID and hence re-establish the
1283  * callback channel. Then return all existing delegations.
1284  */
1285 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1286 {
1287         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1288         nfs_expire_all_delegations(clp);
1289         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1290                         clp->cl_hostname);
1291 }
1292
1293 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1294 {
1295         nfs40_handle_cb_pathdown(clp);
1296         nfs4_schedule_state_manager(clp);
1297 }
1298
1299 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1300 {
1301
1302         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1303         /* Don't recover state that expired before the reboot */
1304         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1305                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1306                 return 0;
1307         }
1308         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1309         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1310         return 1;
1311 }
1312
1313 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1314 {
1315         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1316         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1317         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1318         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1319         return 1;
1320 }
1321
1322 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1323 {
1324         struct nfs_client *clp = server->nfs_client;
1325
1326         if (!nfs4_valid_open_stateid(state))
1327                 return -EBADF;
1328         nfs4_state_mark_reclaim_nograce(clp, state);
1329         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1330                         clp->cl_hostname);
1331         nfs4_schedule_state_manager(clp);
1332         return 0;
1333 }
1334 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1335
1336 void nfs_inode_find_state_and_recover(struct inode *inode,
1337                 const nfs4_stateid *stateid)
1338 {
1339         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1340         struct nfs_inode *nfsi = NFS_I(inode);
1341         struct nfs_open_context *ctx;
1342         struct nfs4_state *state;
1343         bool found = false;
1344
1345         spin_lock(&inode->i_lock);
1346         list_for_each_entry(ctx, &nfsi->open_files, list) {
1347                 state = ctx->state;
1348                 if (state == NULL)
1349                         continue;
1350                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1351                         continue;
1352                 if (!nfs4_stateid_match(&state->stateid, stateid))
1353                         continue;
1354                 nfs4_state_mark_reclaim_nograce(clp, state);
1355                 found = true;
1356         }
1357         spin_unlock(&inode->i_lock);
1358         if (found)
1359                 nfs4_schedule_state_manager(clp);
1360 }
1361
1362 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1363 {
1364         struct inode *inode = state->inode;
1365         struct nfs_inode *nfsi = NFS_I(inode);
1366         struct nfs_open_context *ctx;
1367
1368         spin_lock(&inode->i_lock);
1369         list_for_each_entry(ctx, &nfsi->open_files, list) {
1370                 if (ctx->state != state)
1371                         continue;
1372                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1373         }
1374         spin_unlock(&inode->i_lock);
1375 }
1376
1377 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1378 {
1379         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1380         nfs4_state_mark_open_context_bad(state);
1381 }
1382
1383
1384 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1385 {
1386         struct inode *inode = state->inode;
1387         struct nfs_inode *nfsi = NFS_I(inode);
1388         struct file_lock *fl;
1389         int status = 0;
1390         struct file_lock_context *flctx = inode->i_flctx;
1391         struct list_head *list;
1392
1393         if (flctx == NULL)
1394                 return 0;
1395
1396         list = &flctx->flc_posix;
1397
1398         /* Guard against delegation returns and new lock/unlock calls */
1399         down_write(&nfsi->rwsem);
1400         spin_lock(&flctx->flc_lock);
1401 restart:
1402         list_for_each_entry(fl, list, fl_list) {
1403                 if (nfs_file_open_context(fl->fl_file)->state != state)
1404                         continue;
1405                 spin_unlock(&flctx->flc_lock);
1406                 status = ops->recover_lock(state, fl);
1407                 switch (status) {
1408                 case 0:
1409                         break;
1410                 case -ESTALE:
1411                 case -NFS4ERR_ADMIN_REVOKED:
1412                 case -NFS4ERR_STALE_STATEID:
1413                 case -NFS4ERR_BAD_STATEID:
1414                 case -NFS4ERR_EXPIRED:
1415                 case -NFS4ERR_NO_GRACE:
1416                 case -NFS4ERR_STALE_CLIENTID:
1417                 case -NFS4ERR_BADSESSION:
1418                 case -NFS4ERR_BADSLOT:
1419                 case -NFS4ERR_BAD_HIGH_SLOT:
1420                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1421                         goto out;
1422                 default:
1423                         pr_err("NFS: %s: unhandled error %d\n",
1424                                         __func__, status);
1425                 case -ENOMEM:
1426                 case -NFS4ERR_DENIED:
1427                 case -NFS4ERR_RECLAIM_BAD:
1428                 case -NFS4ERR_RECLAIM_CONFLICT:
1429                         /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1430                         status = 0;
1431                 }
1432                 spin_lock(&flctx->flc_lock);
1433         }
1434         if (list == &flctx->flc_posix) {
1435                 list = &flctx->flc_flock;
1436                 goto restart;
1437         }
1438         spin_unlock(&flctx->flc_lock);
1439 out:
1440         up_write(&nfsi->rwsem);
1441         return status;
1442 }
1443
1444 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1445 {
1446         struct nfs4_state *state;
1447         struct nfs4_lock_state *lock;
1448         int status = 0;
1449
1450         /* Note: we rely on the sp->so_states list being ordered 
1451          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1452          * states first.
1453          * This is needed to ensure that the server won't give us any
1454          * read delegations that we have to return if, say, we are
1455          * recovering after a network partition or a reboot from a
1456          * server that doesn't support a grace period.
1457          */
1458         spin_lock(&sp->so_lock);
1459         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1460 restart:
1461         list_for_each_entry(state, &sp->so_states, open_states) {
1462                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1463                         continue;
1464                 if (!nfs4_valid_open_stateid(state))
1465                         continue;
1466                 if (state->state == 0)
1467                         continue;
1468                 atomic_inc(&state->count);
1469                 spin_unlock(&sp->so_lock);
1470                 status = ops->recover_open(sp, state);
1471                 if (status >= 0) {
1472                         status = nfs4_reclaim_locks(state, ops);
1473                         if (status >= 0) {
1474                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1475                                         spin_lock(&state->state_lock);
1476                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1477                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1478                                                         pr_warn_ratelimited("NFS: "
1479                                                                             "%s: Lock reclaim "
1480                                                                             "failed!\n", __func__);
1481                                         }
1482                                         spin_unlock(&state->state_lock);
1483                                 }
1484                                 nfs4_put_open_state(state);
1485                                 clear_bit(NFS4CLNT_RECLAIM_NOGRACE,
1486                                         &state->flags);
1487                                 spin_lock(&sp->so_lock);
1488                                 goto restart;
1489                         }
1490                 }
1491                 switch (status) {
1492                         default:
1493                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1494                                         __func__, status);
1495                         case -ENOENT:
1496                         case -ENOMEM:
1497                         case -ESTALE:
1498                                 /* Open state on this file cannot be recovered */
1499                                 nfs4_state_mark_recovery_failed(state, status);
1500                                 break;
1501                         case -EAGAIN:
1502                                 ssleep(1);
1503                         case -NFS4ERR_ADMIN_REVOKED:
1504                         case -NFS4ERR_STALE_STATEID:
1505                         case -NFS4ERR_BAD_STATEID:
1506                         case -NFS4ERR_RECLAIM_BAD:
1507                         case -NFS4ERR_RECLAIM_CONFLICT:
1508                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1509                                 break;
1510                         case -NFS4ERR_EXPIRED:
1511                         case -NFS4ERR_NO_GRACE:
1512                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1513                         case -NFS4ERR_STALE_CLIENTID:
1514                         case -NFS4ERR_BADSESSION:
1515                         case -NFS4ERR_BADSLOT:
1516                         case -NFS4ERR_BAD_HIGH_SLOT:
1517                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1518                                 goto out_err;
1519                 }
1520                 nfs4_put_open_state(state);
1521                 spin_lock(&sp->so_lock);
1522                 goto restart;
1523         }
1524         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1525         spin_unlock(&sp->so_lock);
1526         return 0;
1527 out_err:
1528         nfs4_put_open_state(state);
1529         spin_lock(&sp->so_lock);
1530         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1531         spin_unlock(&sp->so_lock);
1532         return status;
1533 }
1534
1535 static void nfs4_clear_open_state(struct nfs4_state *state)
1536 {
1537         struct nfs4_lock_state *lock;
1538
1539         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1540         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1541         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1542         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1543         spin_lock(&state->state_lock);
1544         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1545                 lock->ls_seqid.flags = 0;
1546                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1547         }
1548         spin_unlock(&state->state_lock);
1549 }
1550
1551 static void nfs4_reset_seqids(struct nfs_server *server,
1552         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1553 {
1554         struct nfs_client *clp = server->nfs_client;
1555         struct nfs4_state_owner *sp;
1556         struct rb_node *pos;
1557         struct nfs4_state *state;
1558
1559         spin_lock(&clp->cl_lock);
1560         for (pos = rb_first(&server->state_owners);
1561              pos != NULL;
1562              pos = rb_next(pos)) {
1563                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1564                 sp->so_seqid.flags = 0;
1565                 spin_lock(&sp->so_lock);
1566                 list_for_each_entry(state, &sp->so_states, open_states) {
1567                         if (mark_reclaim(clp, state))
1568                                 nfs4_clear_open_state(state);
1569                 }
1570                 spin_unlock(&sp->so_lock);
1571         }
1572         spin_unlock(&clp->cl_lock);
1573 }
1574
1575 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1576         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1577 {
1578         struct nfs_server *server;
1579
1580         rcu_read_lock();
1581         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1582                 nfs4_reset_seqids(server, mark_reclaim);
1583         rcu_read_unlock();
1584 }
1585
1586 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1587 {
1588         /* Mark all delegations for reclaim */
1589         nfs_delegation_mark_reclaim(clp);
1590         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1591 }
1592
1593 static void nfs4_reclaim_complete(struct nfs_client *clp,
1594                                  const struct nfs4_state_recovery_ops *ops,
1595                                  struct rpc_cred *cred)
1596 {
1597         /* Notify the server we're done reclaiming our state */
1598         if (ops->reclaim_complete)
1599                 (void)ops->reclaim_complete(clp, cred);
1600 }
1601
1602 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1603 {
1604         struct nfs_client *clp = server->nfs_client;
1605         struct nfs4_state_owner *sp;
1606         struct rb_node *pos;
1607         struct nfs4_state *state;
1608
1609         spin_lock(&clp->cl_lock);
1610         for (pos = rb_first(&server->state_owners);
1611              pos != NULL;
1612              pos = rb_next(pos)) {
1613                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1614                 spin_lock(&sp->so_lock);
1615                 list_for_each_entry(state, &sp->so_states, open_states) {
1616                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1617                                                 &state->flags))
1618                                 continue;
1619                         nfs4_state_mark_reclaim_nograce(clp, state);
1620                 }
1621                 spin_unlock(&sp->so_lock);
1622         }
1623         spin_unlock(&clp->cl_lock);
1624 }
1625
1626 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1627 {
1628         struct nfs_server *server;
1629
1630         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1631                 return 0;
1632
1633         rcu_read_lock();
1634         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1635                 nfs4_clear_reclaim_server(server);
1636         rcu_read_unlock();
1637
1638         nfs_delegation_reap_unclaimed(clp);
1639         return 1;
1640 }
1641
1642 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1643 {
1644         const struct nfs4_state_recovery_ops *ops;
1645         struct rpc_cred *cred;
1646
1647         if (!nfs4_state_clear_reclaim_reboot(clp))
1648                 return;
1649         ops = clp->cl_mvops->reboot_recovery_ops;
1650         cred = nfs4_get_clid_cred(clp);
1651         nfs4_reclaim_complete(clp, ops, cred);
1652         put_rpccred(cred);
1653 }
1654
1655 static void nfs_delegation_clear_all(struct nfs_client *clp)
1656 {
1657         nfs_delegation_mark_reclaim(clp);
1658         nfs_delegation_reap_unclaimed(clp);
1659 }
1660
1661 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1662 {
1663         nfs_delegation_clear_all(clp);
1664         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1665 }
1666
1667 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1668 {
1669         switch (error) {
1670                 case 0:
1671                         break;
1672                 case -NFS4ERR_CB_PATH_DOWN:
1673                         nfs40_handle_cb_pathdown(clp);
1674                         break;
1675                 case -NFS4ERR_NO_GRACE:
1676                         nfs4_state_end_reclaim_reboot(clp);
1677                         break;
1678                 case -NFS4ERR_STALE_CLIENTID:
1679                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1680                         nfs4_state_clear_reclaim_reboot(clp);
1681                         nfs4_state_start_reclaim_reboot(clp);
1682                         break;
1683                 case -NFS4ERR_EXPIRED:
1684                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1685                         nfs4_state_start_reclaim_nograce(clp);
1686                         break;
1687                 case -NFS4ERR_BADSESSION:
1688                 case -NFS4ERR_BADSLOT:
1689                 case -NFS4ERR_BAD_HIGH_SLOT:
1690                 case -NFS4ERR_DEADSESSION:
1691                 case -NFS4ERR_SEQ_FALSE_RETRY:
1692                 case -NFS4ERR_SEQ_MISORDERED:
1693                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1694                         /* Zero session reset errors */
1695                         break;
1696                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1697                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1698                         break;
1699                 default:
1700                         dprintk("%s: failed to handle error %d for server %s\n",
1701                                         __func__, error, clp->cl_hostname);
1702                         return error;
1703         }
1704         dprintk("%s: handled error %d for server %s\n", __func__, error,
1705                         clp->cl_hostname);
1706         return 0;
1707 }
1708
1709 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1710 {
1711         struct nfs4_state_owner *sp;
1712         struct nfs_server *server;
1713         struct rb_node *pos;
1714         int status = 0;
1715
1716 restart:
1717         rcu_read_lock();
1718         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1719                 nfs4_purge_state_owners(server);
1720                 spin_lock(&clp->cl_lock);
1721                 for (pos = rb_first(&server->state_owners);
1722                      pos != NULL;
1723                      pos = rb_next(pos)) {
1724                         sp = rb_entry(pos,
1725                                 struct nfs4_state_owner, so_server_node);
1726                         if (!test_and_clear_bit(ops->owner_flag_bit,
1727                                                         &sp->so_flags))
1728                                 continue;
1729                         atomic_inc(&sp->so_count);
1730                         spin_unlock(&clp->cl_lock);
1731                         rcu_read_unlock();
1732
1733                         status = nfs4_reclaim_open_state(sp, ops);
1734                         if (status < 0) {
1735                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1736                                 nfs4_put_state_owner(sp);
1737                                 status = nfs4_recovery_handle_error(clp, status);
1738                                 return (status != 0) ? status : -EAGAIN;
1739                         }
1740
1741                         nfs4_put_state_owner(sp);
1742                         goto restart;
1743                 }
1744                 spin_unlock(&clp->cl_lock);
1745         }
1746         rcu_read_unlock();
1747         return 0;
1748 }
1749
1750 static int nfs4_check_lease(struct nfs_client *clp)
1751 {
1752         struct rpc_cred *cred;
1753         const struct nfs4_state_maintenance_ops *ops =
1754                 clp->cl_mvops->state_renewal_ops;
1755         int status;
1756
1757         /* Is the client already known to have an expired lease? */
1758         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1759                 return 0;
1760         spin_lock(&clp->cl_lock);
1761         cred = ops->get_state_renewal_cred_locked(clp);
1762         spin_unlock(&clp->cl_lock);
1763         if (cred == NULL) {
1764                 cred = nfs4_get_clid_cred(clp);
1765                 status = -ENOKEY;
1766                 if (cred == NULL)
1767                         goto out;
1768         }
1769         status = ops->renew_lease(clp, cred);
1770         put_rpccred(cred);
1771         if (status == -ETIMEDOUT) {
1772                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1773                 return 0;
1774         }
1775 out:
1776         return nfs4_recovery_handle_error(clp, status);
1777 }
1778
1779 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1780  * and for recoverable errors on EXCHANGE_ID for v4.1
1781  */
1782 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1783 {
1784         switch (status) {
1785         case -NFS4ERR_SEQ_MISORDERED:
1786                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1787                         return -ESERVERFAULT;
1788                 /* Lease confirmation error: retry after purging the lease */
1789                 ssleep(1);
1790                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1791                 break;
1792         case -NFS4ERR_STALE_CLIENTID:
1793                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1794                 nfs4_state_start_reclaim_reboot(clp);
1795                 break;
1796         case -NFS4ERR_CLID_INUSE:
1797                 pr_err("NFS: Server %s reports our clientid is in use\n",
1798                         clp->cl_hostname);
1799                 nfs_mark_client_ready(clp, -EPERM);
1800                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1801                 return -EPERM;
1802         case -EACCES:
1803         case -NFS4ERR_DELAY:
1804         case -ETIMEDOUT:
1805         case -EAGAIN:
1806                 ssleep(1);
1807                 break;
1808
1809         case -NFS4ERR_MINOR_VERS_MISMATCH:
1810                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1811                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1812                 dprintk("%s: exit with error %d for server %s\n",
1813                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1814                 return -EPROTONOSUPPORT;
1815         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1816                                  * in nfs4_exchange_id */
1817         default:
1818                 dprintk("%s: exit with error %d for server %s\n", __func__,
1819                                 status, clp->cl_hostname);
1820                 return status;
1821         }
1822         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1823         dprintk("%s: handled error %d for server %s\n", __func__, status,
1824                         clp->cl_hostname);
1825         return 0;
1826 }
1827
1828 static int nfs4_establish_lease(struct nfs_client *clp)
1829 {
1830         struct rpc_cred *cred;
1831         const struct nfs4_state_recovery_ops *ops =
1832                 clp->cl_mvops->reboot_recovery_ops;
1833         int status;
1834
1835         cred = nfs4_get_clid_cred(clp);
1836         if (cred == NULL)
1837                 return -ENOENT;
1838         status = ops->establish_clid(clp, cred);
1839         put_rpccred(cred);
1840         if (status != 0)
1841                 return status;
1842         pnfs_destroy_all_layouts(clp);
1843         return 0;
1844 }
1845
1846 /*
1847  * Returns zero or a negative errno.  NFS4ERR values are converted
1848  * to local errno values.
1849  */
1850 static int nfs4_reclaim_lease(struct nfs_client *clp)
1851 {
1852         int status;
1853
1854         status = nfs4_establish_lease(clp);
1855         if (status < 0)
1856                 return nfs4_handle_reclaim_lease_error(clp, status);
1857         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1858                 nfs4_state_start_reclaim_nograce(clp);
1859         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1860                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1861         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1862         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1863         return 0;
1864 }
1865
1866 static int nfs4_purge_lease(struct nfs_client *clp)
1867 {
1868         int status;
1869
1870         status = nfs4_establish_lease(clp);
1871         if (status < 0)
1872                 return nfs4_handle_reclaim_lease_error(clp, status);
1873         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1874         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1875         nfs4_state_start_reclaim_nograce(clp);
1876         return 0;
1877 }
1878
1879 /*
1880  * Try remote migration of one FSID from a source server to a
1881  * destination server.  The source server provides a list of
1882  * potential destinations.
1883  *
1884  * Returns zero or a negative NFS4ERR status code.
1885  */
1886 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1887 {
1888         struct nfs_client *clp = server->nfs_client;
1889         struct nfs4_fs_locations *locations = NULL;
1890         struct inode *inode;
1891         struct page *page;
1892         int status, result;
1893
1894         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1895                         (unsigned long long)server->fsid.major,
1896                         (unsigned long long)server->fsid.minor,
1897                         clp->cl_hostname);
1898
1899         result = 0;
1900         page = alloc_page(GFP_KERNEL);
1901         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1902         if (page == NULL || locations == NULL) {
1903                 dprintk("<-- %s: no memory\n", __func__);
1904                 goto out;
1905         }
1906
1907         inode = d_inode(server->super->s_root);
1908         result = nfs4_proc_get_locations(inode, locations, page, cred);
1909         if (result) {
1910                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1911                         __func__, result);
1912                 goto out;
1913         }
1914
1915         result = -NFS4ERR_NXIO;
1916         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1917                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1918                         __func__);
1919                 goto out;
1920         }
1921
1922         nfs4_begin_drain_session(clp);
1923
1924         status = nfs4_replace_transport(server, locations);
1925         if (status != 0) {
1926                 dprintk("<-- %s: failed to replace transport: %d\n",
1927                         __func__, status);
1928                 goto out;
1929         }
1930
1931         result = 0;
1932         dprintk("<-- %s: migration succeeded\n", __func__);
1933
1934 out:
1935         if (page != NULL)
1936                 __free_page(page);
1937         kfree(locations);
1938         if (result) {
1939                 pr_err("NFS: migration recovery failed (server %s)\n",
1940                                 clp->cl_hostname);
1941                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1942         }
1943         return result;
1944 }
1945
1946 /*
1947  * Returns zero or a negative NFS4ERR status code.
1948  */
1949 static int nfs4_handle_migration(struct nfs_client *clp)
1950 {
1951         const struct nfs4_state_maintenance_ops *ops =
1952                                 clp->cl_mvops->state_renewal_ops;
1953         struct nfs_server *server;
1954         struct rpc_cred *cred;
1955
1956         dprintk("%s: migration reported on \"%s\"\n", __func__,
1957                         clp->cl_hostname);
1958
1959         spin_lock(&clp->cl_lock);
1960         cred = ops->get_state_renewal_cred_locked(clp);
1961         spin_unlock(&clp->cl_lock);
1962         if (cred == NULL)
1963                 return -NFS4ERR_NOENT;
1964
1965         clp->cl_mig_gen++;
1966 restart:
1967         rcu_read_lock();
1968         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1969                 int status;
1970
1971                 if (server->mig_gen == clp->cl_mig_gen)
1972                         continue;
1973                 server->mig_gen = clp->cl_mig_gen;
1974
1975                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1976                                                 &server->mig_status))
1977                         continue;
1978
1979                 rcu_read_unlock();
1980                 status = nfs4_try_migration(server, cred);
1981                 if (status < 0) {
1982                         put_rpccred(cred);
1983                         return status;
1984                 }
1985                 goto restart;
1986         }
1987         rcu_read_unlock();
1988         put_rpccred(cred);
1989         return 0;
1990 }
1991
1992 /*
1993  * Test each nfs_server on the clp's cl_superblocks list to see
1994  * if it's moved to another server.  Stop when the server no longer
1995  * returns NFS4ERR_LEASE_MOVED.
1996  */
1997 static int nfs4_handle_lease_moved(struct nfs_client *clp)
1998 {
1999         const struct nfs4_state_maintenance_ops *ops =
2000                                 clp->cl_mvops->state_renewal_ops;
2001         struct nfs_server *server;
2002         struct rpc_cred *cred;
2003
2004         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2005                         clp->cl_hostname);
2006
2007         spin_lock(&clp->cl_lock);
2008         cred = ops->get_state_renewal_cred_locked(clp);
2009         spin_unlock(&clp->cl_lock);
2010         if (cred == NULL)
2011                 return -NFS4ERR_NOENT;
2012
2013         clp->cl_mig_gen++;
2014 restart:
2015         rcu_read_lock();
2016         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2017                 struct inode *inode;
2018                 int status;
2019
2020                 if (server->mig_gen == clp->cl_mig_gen)
2021                         continue;
2022                 server->mig_gen = clp->cl_mig_gen;
2023
2024                 rcu_read_unlock();
2025
2026                 inode = d_inode(server->super->s_root);
2027                 status = nfs4_proc_fsid_present(inode, cred);
2028                 if (status != -NFS4ERR_MOVED)
2029                         goto restart;   /* wasn't this one */
2030                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2031                         goto restart;   /* there are more */
2032                 goto out;
2033         }
2034         rcu_read_unlock();
2035
2036 out:
2037         put_rpccred(cred);
2038         return 0;
2039 }
2040
2041 /**
2042  * nfs4_discover_server_trunking - Detect server IP address trunking
2043  *
2044  * @clp: nfs_client under test
2045  * @result: OUT: found nfs_client, or clp
2046  *
2047  * Returns zero or a negative errno.  If zero is returned,
2048  * an nfs_client pointer is planted in "result".
2049  *
2050  * Note: since we are invoked in process context, and
2051  * not from inside the state manager, we cannot use
2052  * nfs4_handle_reclaim_lease_error().
2053  */
2054 int nfs4_discover_server_trunking(struct nfs_client *clp,
2055                                   struct nfs_client **result)
2056 {
2057         const struct nfs4_state_recovery_ops *ops =
2058                                 clp->cl_mvops->reboot_recovery_ops;
2059         struct rpc_clnt *clnt;
2060         struct rpc_cred *cred;
2061         int i, status;
2062
2063         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2064
2065         clnt = clp->cl_rpcclient;
2066         i = 0;
2067
2068         mutex_lock(&nfs_clid_init_mutex);
2069 again:
2070         status  = -ENOENT;
2071         cred = nfs4_get_clid_cred(clp);
2072         if (cred == NULL)
2073                 goto out_unlock;
2074
2075         status = ops->detect_trunking(clp, result, cred);
2076         put_rpccred(cred);
2077         switch (status) {
2078         case 0:
2079                 break;
2080         case -ETIMEDOUT:
2081                 if (clnt->cl_softrtry)
2082                         break;
2083         case -NFS4ERR_DELAY:
2084         case -EAGAIN:
2085                 ssleep(1);
2086         case -NFS4ERR_STALE_CLIENTID:
2087                 dprintk("NFS: %s after status %d, retrying\n",
2088                         __func__, status);
2089                 goto again;
2090         case -EACCES:
2091                 if (i++ == 0) {
2092                         nfs4_root_machine_cred(clp);
2093                         goto again;
2094                 }
2095                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2096                         break;
2097         case -NFS4ERR_CLID_INUSE:
2098         case -NFS4ERR_WRONGSEC:
2099                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2100                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2101                         status = -EPERM;
2102                         break;
2103                 }
2104                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2105                 if (IS_ERR(clnt)) {
2106                         status = PTR_ERR(clnt);
2107                         break;
2108                 }
2109                 /* Note: this is safe because we haven't yet marked the
2110                  * client as ready, so we are the only user of
2111                  * clp->cl_rpcclient
2112                  */
2113                 clnt = xchg(&clp->cl_rpcclient, clnt);
2114                 rpc_shutdown_client(clnt);
2115                 clnt = clp->cl_rpcclient;
2116                 goto again;
2117
2118         case -NFS4ERR_MINOR_VERS_MISMATCH:
2119                 status = -EPROTONOSUPPORT;
2120                 break;
2121
2122         case -EKEYEXPIRED:
2123         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2124                                  * in nfs4_exchange_id */
2125                 status = -EKEYEXPIRED;
2126                 break;
2127         default:
2128                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2129                                 __func__, status);
2130                 status = -EIO;
2131         }
2132
2133 out_unlock:
2134         mutex_unlock(&nfs_clid_init_mutex);
2135         dprintk("NFS: %s: status = %d\n", __func__, status);
2136         return status;
2137 }
2138
2139 #ifdef CONFIG_NFS_V4_1
2140 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2141 {
2142         struct nfs_client *clp = session->clp;
2143
2144         switch (err) {
2145         default:
2146                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2147                 break;
2148         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2149                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2150         }
2151         nfs4_schedule_lease_recovery(clp);
2152 }
2153 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2154
2155 static void nfs41_ping_server(struct nfs_client *clp)
2156 {
2157         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2158         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2159         nfs4_schedule_state_manager(clp);
2160 }
2161
2162 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2163 {
2164         nfs41_ping_server(clp);
2165 }
2166
2167 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2168 {
2169         nfs41_ping_server(clp);
2170 }
2171
2172 static void nfs4_reset_all_state(struct nfs_client *clp)
2173 {
2174         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2175                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2176                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2177                 nfs4_state_start_reclaim_nograce(clp);
2178                 dprintk("%s: scheduling reset of all state for server %s!\n",
2179                                 __func__, clp->cl_hostname);
2180                 nfs4_schedule_state_manager(clp);
2181         }
2182 }
2183
2184 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2185 {
2186         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2187                 nfs4_state_start_reclaim_reboot(clp);
2188                 dprintk("%s: server %s rebooted!\n", __func__,
2189                                 clp->cl_hostname);
2190                 nfs4_schedule_state_manager(clp);
2191         }
2192 }
2193
2194 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2195 {
2196         nfs4_reset_all_state(clp);
2197         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2198 }
2199
2200 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2201 {
2202         /* This will need to handle layouts too */
2203         nfs_expire_all_delegations(clp);
2204         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2205                         clp->cl_hostname);
2206 }
2207
2208 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2209 {
2210         nfs_expire_all_delegations(clp);
2211         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2212                 nfs4_schedule_state_manager(clp);
2213         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2214                         clp->cl_hostname);
2215 }
2216
2217 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2218 {
2219         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2220                 &clp->cl_state) == 0)
2221                 nfs4_schedule_state_manager(clp);
2222 }
2223
2224 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2225 {
2226         if (!flags)
2227                 return;
2228
2229         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2230                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2231
2232         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2233                 nfs41_handle_server_reboot(clp);
2234         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2235                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2236                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2237                 nfs41_handle_state_revoked(clp);
2238         if (flags & SEQ4_STATUS_LEASE_MOVED)
2239                 nfs4_schedule_lease_moved_recovery(clp);
2240         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2241                 nfs41_handle_recallable_state_revoked(clp);
2242         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2243                 nfs41_handle_backchannel_fault(clp);
2244         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2245                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2246                 nfs41_handle_cb_path_down(clp);
2247 }
2248
2249 static int nfs4_reset_session(struct nfs_client *clp)
2250 {
2251         struct rpc_cred *cred;
2252         int status;
2253
2254         if (!nfs4_has_session(clp))
2255                 return 0;
2256         nfs4_begin_drain_session(clp);
2257         cred = nfs4_get_clid_cred(clp);
2258         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2259         switch (status) {
2260         case 0:
2261         case -NFS4ERR_BADSESSION:
2262         case -NFS4ERR_DEADSESSION:
2263                 break;
2264         case -NFS4ERR_BACK_CHAN_BUSY:
2265         case -NFS4ERR_DELAY:
2266                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2267                 status = 0;
2268                 ssleep(1);
2269                 goto out;
2270         default:
2271                 status = nfs4_recovery_handle_error(clp, status);
2272                 goto out;
2273         }
2274
2275         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2276         status = nfs4_proc_create_session(clp, cred);
2277         if (status) {
2278                 dprintk("%s: session reset failed with status %d for server %s!\n",
2279                         __func__, status, clp->cl_hostname);
2280                 status = nfs4_handle_reclaim_lease_error(clp, status);
2281                 goto out;
2282         }
2283         nfs41_finish_session_reset(clp);
2284         dprintk("%s: session reset was successful for server %s!\n",
2285                         __func__, clp->cl_hostname);
2286 out:
2287         if (cred)
2288                 put_rpccred(cred);
2289         return status;
2290 }
2291
2292 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2293 {
2294         struct rpc_cred *cred;
2295         int ret;
2296
2297         if (!nfs4_has_session(clp))
2298                 return 0;
2299         nfs4_begin_drain_session(clp);
2300         cred = nfs4_get_clid_cred(clp);
2301         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2302         if (cred)
2303                 put_rpccred(cred);
2304         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2305         switch (ret) {
2306         case 0:
2307                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2308                         __func__, clp->cl_hostname);
2309                 break;
2310         case -NFS4ERR_DELAY:
2311                 ssleep(1);
2312                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2313                 break;
2314         default:
2315                 return nfs4_recovery_handle_error(clp, ret);
2316         }
2317         return 0;
2318 }
2319 #else /* CONFIG_NFS_V4_1 */
2320 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2321
2322 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2323 {
2324         return 0;
2325 }
2326 #endif /* CONFIG_NFS_V4_1 */
2327
2328 static void nfs4_state_manager(struct nfs_client *clp)
2329 {
2330         int status = 0;
2331         const char *section = "", *section_sep = "";
2332
2333         /* Ensure exclusive access to NFSv4 state */
2334         do {
2335                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2336                         section = "purge state";
2337                         status = nfs4_purge_lease(clp);
2338                         if (status < 0)
2339                                 goto out_error;
2340                         continue;
2341                 }
2342
2343                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2344                         section = "lease expired";
2345                         /* We're going to have to re-establish a clientid */
2346                         status = nfs4_reclaim_lease(clp);
2347                         if (status < 0)
2348                                 goto out_error;
2349                         continue;
2350                 }
2351
2352                 /* Initialize or reset the session */
2353                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2354                         section = "reset session";
2355                         status = nfs4_reset_session(clp);
2356                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2357                                 continue;
2358                         if (status < 0)
2359                                 goto out_error;
2360                 }
2361
2362                 /* Send BIND_CONN_TO_SESSION */
2363                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2364                                 &clp->cl_state)) {
2365                         section = "bind conn to session";
2366                         status = nfs4_bind_conn_to_session(clp);
2367                         if (status < 0)
2368                                 goto out_error;
2369                         continue;
2370                 }
2371
2372                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2373                         section = "check lease";
2374                         status = nfs4_check_lease(clp);
2375                         if (status < 0)
2376                                 goto out_error;
2377                         continue;
2378                 }
2379
2380                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2381                         section = "migration";
2382                         status = nfs4_handle_migration(clp);
2383                         if (status < 0)
2384                                 goto out_error;
2385                 }
2386
2387                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2388                         section = "lease moved";
2389                         status = nfs4_handle_lease_moved(clp);
2390                         if (status < 0)
2391                                 goto out_error;
2392                 }
2393
2394                 /* First recover reboot state... */
2395                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2396                         section = "reclaim reboot";
2397                         status = nfs4_do_reclaim(clp,
2398                                 clp->cl_mvops->reboot_recovery_ops);
2399                         if (status == -EAGAIN)
2400                                 continue;
2401                         if (status < 0)
2402                                 goto out_error;
2403                         nfs4_state_end_reclaim_reboot(clp);
2404                 }
2405
2406                 /* Now recover expired state... */
2407                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2408                         section = "reclaim nograce";
2409                         status = nfs4_do_reclaim(clp,
2410                                 clp->cl_mvops->nograce_recovery_ops);
2411                         if (status == -EAGAIN)
2412                                 continue;
2413                         if (status < 0)
2414                                 goto out_error;
2415                 }
2416
2417                 nfs4_end_drain_session(clp);
2418                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2419                         nfs_client_return_marked_delegations(clp);
2420                         continue;
2421                 }
2422
2423                 nfs4_clear_state_manager_bit(clp);
2424                 /* Did we race with an attempt to give us more work? */
2425                 if (clp->cl_state == 0)
2426                         break;
2427                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2428                         break;
2429         } while (atomic_read(&clp->cl_count) > 1);
2430         return;
2431 out_error:
2432         if (strlen(section))
2433                 section_sep = ": ";
2434         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2435                         " with error %d\n", section_sep, section,
2436                         clp->cl_hostname, -status);
2437         ssleep(1);
2438         nfs4_end_drain_session(clp);
2439         nfs4_clear_state_manager_bit(clp);
2440 }
2441
2442 static int nfs4_run_state_manager(void *ptr)
2443 {
2444         struct nfs_client *clp = ptr;
2445
2446         allow_signal(SIGKILL);
2447         nfs4_state_manager(clp);
2448         nfs_put_client(clp);
2449         module_put_and_exit(0);
2450         return 0;
2451 }
2452
2453 /*
2454  * Local variables:
2455  *  c-basic-offset: 8
2456  * End:
2457  */