Merge tag 'nfsd-4.9' of git://linux-nfs.org/~bfields/linux
[cascardo/linux.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/kthread.h>
46 #include <linux/module.h>
47 #include <linux/random.h>
48 #include <linux/ratelimit.h>
49 #include <linux/workqueue.h>
50 #include <linux/bitops.h>
51 #include <linux/jiffies.h>
52
53 #include <linux/sunrpc/clnt.h>
54
55 #include "nfs4_fs.h"
56 #include "callback.h"
57 #include "delegation.h"
58 #include "internal.h"
59 #include "nfs4idmap.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
63
64 #define NFSDBG_FACILITY         NFSDBG_STATE
65
66 #define OPENOWNER_POOL_SIZE     8
67
68 const nfs4_stateid zero_stateid = {
69         { .data = { 0 } },
70         .type = NFS4_SPECIAL_STATEID_TYPE,
71 };
72 static DEFINE_MUTEX(nfs_clid_init_mutex);
73
74 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
75 {
76         struct nfs4_setclientid_res clid = {
77                 .clientid = clp->cl_clientid,
78                 .confirm = clp->cl_confirm,
79         };
80         unsigned short port;
81         int status;
82         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
83
84         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
85                 goto do_confirm;
86         port = nn->nfs_callback_tcpport;
87         if (clp->cl_addr.ss_family == AF_INET6)
88                 port = nn->nfs_callback_tcpport6;
89
90         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
91         if (status != 0)
92                 goto out;
93         clp->cl_clientid = clid.clientid;
94         clp->cl_confirm = clid.confirm;
95         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
96 do_confirm:
97         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
98         if (status != 0)
99                 goto out;
100         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
101         nfs4_schedule_state_renewal(clp);
102 out:
103         return status;
104 }
105
106 /**
107  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
108  *
109  * @clp: nfs_client under test
110  * @result: OUT: found nfs_client, or clp
111  * @cred: credential to use for trunking test
112  *
113  * Returns zero, a negative errno, or a negative NFS4ERR status.
114  * If zero is returned, an nfs_client pointer is planted in
115  * "result".
116  *
117  * Note: The returned client may not yet be marked ready.
118  */
119 int nfs40_discover_server_trunking(struct nfs_client *clp,
120                                    struct nfs_client **result,
121                                    struct rpc_cred *cred)
122 {
123         struct nfs4_setclientid_res clid = {
124                 .clientid = clp->cl_clientid,
125                 .confirm = clp->cl_confirm,
126         };
127         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
128         unsigned short port;
129         int status;
130
131         port = nn->nfs_callback_tcpport;
132         if (clp->cl_addr.ss_family == AF_INET6)
133                 port = nn->nfs_callback_tcpport6;
134
135         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
136         if (status != 0)
137                 goto out;
138         clp->cl_clientid = clid.clientid;
139         clp->cl_confirm = clid.confirm;
140
141         status = nfs40_walk_client_list(clp, result, cred);
142         if (status == 0) {
143                 /* Sustain the lease, even if it's empty.  If the clientid4
144                  * goes stale it's of no use for trunking discovery. */
145                 nfs4_schedule_state_renewal(*result);
146         }
147 out:
148         return status;
149 }
150
151 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
152 {
153         struct rpc_cred *cred = NULL;
154
155         if (clp->cl_machine_cred != NULL)
156                 cred = get_rpccred(clp->cl_machine_cred);
157         return cred;
158 }
159
160 static void nfs4_root_machine_cred(struct nfs_client *clp)
161 {
162         struct rpc_cred *cred, *new;
163
164         new = rpc_lookup_machine_cred(NULL);
165         spin_lock(&clp->cl_lock);
166         cred = clp->cl_machine_cred;
167         clp->cl_machine_cred = new;
168         spin_unlock(&clp->cl_lock);
169         if (cred != NULL)
170                 put_rpccred(cred);
171 }
172
173 static struct rpc_cred *
174 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
175 {
176         struct rpc_cred *cred = NULL;
177         struct nfs4_state_owner *sp;
178         struct rb_node *pos;
179
180         for (pos = rb_first(&server->state_owners);
181              pos != NULL;
182              pos = rb_next(pos)) {
183                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
184                 if (list_empty(&sp->so_states))
185                         continue;
186                 cred = get_rpccred(sp->so_cred);
187                 break;
188         }
189         return cred;
190 }
191
192 /**
193  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
194  * @clp: client state handle
195  *
196  * Returns an rpc_cred with reference count bumped, or NULL.
197  * Caller must hold clp->cl_lock.
198  */
199 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
200 {
201         struct rpc_cred *cred = NULL;
202         struct nfs_server *server;
203
204         /* Use machine credentials if available */
205         cred = nfs4_get_machine_cred_locked(clp);
206         if (cred != NULL)
207                 goto out;
208
209         rcu_read_lock();
210         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
211                 cred = nfs4_get_renew_cred_server_locked(server);
212                 if (cred != NULL)
213                         break;
214         }
215         rcu_read_unlock();
216
217 out:
218         return cred;
219 }
220
221 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
222 {
223         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
224                 spin_lock(&tbl->slot_tbl_lock);
225                 nfs41_wake_slot_table(tbl);
226                 spin_unlock(&tbl->slot_tbl_lock);
227         }
228 }
229
230 static void nfs4_end_drain_session(struct nfs_client *clp)
231 {
232         struct nfs4_session *ses = clp->cl_session;
233
234         if (clp->cl_slot_tbl) {
235                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
236                 return;
237         }
238
239         if (ses != NULL) {
240                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
241                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
242         }
243 }
244
245 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
246 {
247         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
248         spin_lock(&tbl->slot_tbl_lock);
249         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
250                 reinit_completion(&tbl->complete);
251                 spin_unlock(&tbl->slot_tbl_lock);
252                 return wait_for_completion_interruptible(&tbl->complete);
253         }
254         spin_unlock(&tbl->slot_tbl_lock);
255         return 0;
256 }
257
258 static int nfs4_begin_drain_session(struct nfs_client *clp)
259 {
260         struct nfs4_session *ses = clp->cl_session;
261         int ret = 0;
262
263         if (clp->cl_slot_tbl)
264                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
265
266         /* back channel */
267         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
268         if (ret)
269                 return ret;
270         /* fore channel */
271         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
272 }
273
274 #if defined(CONFIG_NFS_V4_1)
275
276 static int nfs41_setup_state_renewal(struct nfs_client *clp)
277 {
278         int status;
279         struct nfs_fsinfo fsinfo;
280         unsigned long now;
281
282         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
283                 nfs4_schedule_state_renewal(clp);
284                 return 0;
285         }
286
287         now = jiffies;
288         status = nfs4_proc_get_lease_time(clp, &fsinfo);
289         if (status == 0) {
290                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
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         status = nfs4_proc_exchange_id(clp, cred);
313         if (status != 0)
314                 goto out;
315         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
316 do_confirm:
317         status = nfs4_proc_create_session(clp, cred);
318         if (status != 0)
319                 goto out;
320         nfs41_finish_session_reset(clp);
321         nfs_mark_client_ready(clp, NFS_CS_READY);
322 out:
323         return status;
324 }
325
326 /**
327  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
328  *
329  * @clp: nfs_client under test
330  * @result: OUT: found nfs_client, or clp
331  * @cred: credential to use for trunking test
332  *
333  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
334  * If NFS4_OK is returned, an nfs_client pointer is planted in
335  * "result".
336  *
337  * Note: The returned client may not yet be marked ready.
338  */
339 int nfs41_discover_server_trunking(struct nfs_client *clp,
340                                    struct nfs_client **result,
341                                    struct rpc_cred *cred)
342 {
343         int status;
344
345         status = nfs4_proc_exchange_id(clp, cred);
346         if (status != NFS4_OK)
347                 return status;
348
349         status = nfs41_walk_client_list(clp, result, cred);
350         if (status < 0)
351                 return status;
352         if (clp != *result)
353                 return 0;
354
355         /* Purge state if the client id was established in a prior instance */
356         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
357                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
358         else
359                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
360         nfs4_schedule_state_manager(clp);
361         status = nfs_wait_client_init_complete(clp);
362         if (status < 0)
363                 nfs_put_client(clp);
364         return status;
365 }
366
367 #endif /* CONFIG_NFS_V4_1 */
368
369 /**
370  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
371  * @clp: client state handle
372  *
373  * Returns an rpc_cred with reference count bumped, or NULL.
374  */
375 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
376 {
377         struct rpc_cred *cred;
378
379         spin_lock(&clp->cl_lock);
380         cred = nfs4_get_machine_cred_locked(clp);
381         spin_unlock(&clp->cl_lock);
382         return cred;
383 }
384
385 static struct nfs4_state_owner *
386 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
387 {
388         struct rb_node **p = &server->state_owners.rb_node,
389                        *parent = NULL;
390         struct nfs4_state_owner *sp;
391
392         while (*p != NULL) {
393                 parent = *p;
394                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
395
396                 if (cred < sp->so_cred)
397                         p = &parent->rb_left;
398                 else if (cred > sp->so_cred)
399                         p = &parent->rb_right;
400                 else {
401                         if (!list_empty(&sp->so_lru))
402                                 list_del_init(&sp->so_lru);
403                         atomic_inc(&sp->so_count);
404                         return sp;
405                 }
406         }
407         return NULL;
408 }
409
410 static struct nfs4_state_owner *
411 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
412 {
413         struct nfs_server *server = new->so_server;
414         struct rb_node **p = &server->state_owners.rb_node,
415                        *parent = NULL;
416         struct nfs4_state_owner *sp;
417         int err;
418
419         while (*p != NULL) {
420                 parent = *p;
421                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
422
423                 if (new->so_cred < sp->so_cred)
424                         p = &parent->rb_left;
425                 else if (new->so_cred > sp->so_cred)
426                         p = &parent->rb_right;
427                 else {
428                         if (!list_empty(&sp->so_lru))
429                                 list_del_init(&sp->so_lru);
430                         atomic_inc(&sp->so_count);
431                         return sp;
432                 }
433         }
434         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
435         if (err)
436                 return ERR_PTR(err);
437         rb_link_node(&new->so_server_node, parent, p);
438         rb_insert_color(&new->so_server_node, &server->state_owners);
439         return new;
440 }
441
442 static void
443 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
444 {
445         struct nfs_server *server = sp->so_server;
446
447         if (!RB_EMPTY_NODE(&sp->so_server_node))
448                 rb_erase(&sp->so_server_node, &server->state_owners);
449         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
450 }
451
452 static void
453 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
454 {
455         sc->create_time = ktime_get();
456         sc->flags = 0;
457         sc->counter = 0;
458         spin_lock_init(&sc->lock);
459         INIT_LIST_HEAD(&sc->list);
460         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
461 }
462
463 static void
464 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
465 {
466         rpc_destroy_wait_queue(&sc->wait);
467 }
468
469 /*
470  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
471  * create a new state_owner.
472  *
473  */
474 static struct nfs4_state_owner *
475 nfs4_alloc_state_owner(struct nfs_server *server,
476                 struct rpc_cred *cred,
477                 gfp_t gfp_flags)
478 {
479         struct nfs4_state_owner *sp;
480
481         sp = kzalloc(sizeof(*sp), gfp_flags);
482         if (!sp)
483                 return NULL;
484         sp->so_server = server;
485         sp->so_cred = get_rpccred(cred);
486         spin_lock_init(&sp->so_lock);
487         INIT_LIST_HEAD(&sp->so_states);
488         nfs4_init_seqid_counter(&sp->so_seqid);
489         atomic_set(&sp->so_count, 1);
490         INIT_LIST_HEAD(&sp->so_lru);
491         seqcount_init(&sp->so_reclaim_seqcount);
492         mutex_init(&sp->so_delegreturn_mutex);
493         return sp;
494 }
495
496 static void
497 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
498 {
499         struct rb_node *rb_node = &sp->so_server_node;
500
501         if (!RB_EMPTY_NODE(rb_node)) {
502                 struct nfs_server *server = sp->so_server;
503                 struct nfs_client *clp = server->nfs_client;
504
505                 spin_lock(&clp->cl_lock);
506                 if (!RB_EMPTY_NODE(rb_node)) {
507                         rb_erase(rb_node, &server->state_owners);
508                         RB_CLEAR_NODE(rb_node);
509                 }
510                 spin_unlock(&clp->cl_lock);
511         }
512 }
513
514 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
515 {
516         nfs4_destroy_seqid_counter(&sp->so_seqid);
517         put_rpccred(sp->so_cred);
518         kfree(sp);
519 }
520
521 static void nfs4_gc_state_owners(struct nfs_server *server)
522 {
523         struct nfs_client *clp = server->nfs_client;
524         struct nfs4_state_owner *sp, *tmp;
525         unsigned long time_min, time_max;
526         LIST_HEAD(doomed);
527
528         spin_lock(&clp->cl_lock);
529         time_max = jiffies;
530         time_min = (long)time_max - (long)clp->cl_lease_time;
531         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
532                 /* NB: LRU is sorted so that oldest is at the head */
533                 if (time_in_range(sp->so_expires, time_min, time_max))
534                         break;
535                 list_move(&sp->so_lru, &doomed);
536                 nfs4_remove_state_owner_locked(sp);
537         }
538         spin_unlock(&clp->cl_lock);
539
540         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
541                 list_del(&sp->so_lru);
542                 nfs4_free_state_owner(sp);
543         }
544 }
545
546 /**
547  * nfs4_get_state_owner - Look up a state owner given a credential
548  * @server: nfs_server to search
549  * @cred: RPC credential to match
550  *
551  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
552  */
553 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
554                                               struct rpc_cred *cred,
555                                               gfp_t gfp_flags)
556 {
557         struct nfs_client *clp = server->nfs_client;
558         struct nfs4_state_owner *sp, *new;
559
560         spin_lock(&clp->cl_lock);
561         sp = nfs4_find_state_owner_locked(server, cred);
562         spin_unlock(&clp->cl_lock);
563         if (sp != NULL)
564                 goto out;
565         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
566         if (new == NULL)
567                 goto out;
568         do {
569                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
570                         break;
571                 spin_lock(&clp->cl_lock);
572                 sp = nfs4_insert_state_owner_locked(new);
573                 spin_unlock(&clp->cl_lock);
574         } while (sp == ERR_PTR(-EAGAIN));
575         if (sp != new)
576                 nfs4_free_state_owner(new);
577 out:
578         nfs4_gc_state_owners(server);
579         return sp;
580 }
581
582 /**
583  * nfs4_put_state_owner - Release a nfs4_state_owner
584  * @sp: state owner data to release
585  *
586  * Note that we keep released state owners on an LRU
587  * list.
588  * This caches valid state owners so that they can be
589  * reused, to avoid the OPEN_CONFIRM on minor version 0.
590  * It also pins the uniquifier of dropped state owners for
591  * a while, to ensure that those state owner names are
592  * never reused.
593  */
594 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
595 {
596         struct nfs_server *server = sp->so_server;
597         struct nfs_client *clp = server->nfs_client;
598
599         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
600                 return;
601
602         sp->so_expires = jiffies;
603         list_add_tail(&sp->so_lru, &server->state_owners_lru);
604         spin_unlock(&clp->cl_lock);
605 }
606
607 /**
608  * nfs4_purge_state_owners - Release all cached state owners
609  * @server: nfs_server with cached state owners to release
610  *
611  * Called at umount time.  Remaining state owners will be on
612  * the LRU with ref count of zero.
613  */
614 void nfs4_purge_state_owners(struct nfs_server *server)
615 {
616         struct nfs_client *clp = server->nfs_client;
617         struct nfs4_state_owner *sp, *tmp;
618         LIST_HEAD(doomed);
619
620         spin_lock(&clp->cl_lock);
621         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
622                 list_move(&sp->so_lru, &doomed);
623                 nfs4_remove_state_owner_locked(sp);
624         }
625         spin_unlock(&clp->cl_lock);
626
627         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
628                 list_del(&sp->so_lru);
629                 nfs4_free_state_owner(sp);
630         }
631 }
632
633 static struct nfs4_state *
634 nfs4_alloc_open_state(void)
635 {
636         struct nfs4_state *state;
637
638         state = kzalloc(sizeof(*state), GFP_NOFS);
639         if (!state)
640                 return NULL;
641         atomic_set(&state->count, 1);
642         INIT_LIST_HEAD(&state->lock_states);
643         spin_lock_init(&state->state_lock);
644         seqlock_init(&state->seqlock);
645         return state;
646 }
647
648 void
649 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
650 {
651         if (state->state == fmode)
652                 return;
653         /* NB! List reordering - see the reclaim code for why.  */
654         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
655                 if (fmode & FMODE_WRITE)
656                         list_move(&state->open_states, &state->owner->so_states);
657                 else
658                         list_move_tail(&state->open_states, &state->owner->so_states);
659         }
660         state->state = fmode;
661 }
662
663 static struct nfs4_state *
664 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
665 {
666         struct nfs_inode *nfsi = NFS_I(inode);
667         struct nfs4_state *state;
668
669         list_for_each_entry(state, &nfsi->open_states, inode_states) {
670                 if (state->owner != owner)
671                         continue;
672                 if (!nfs4_valid_open_stateid(state))
673                         continue;
674                 if (atomic_inc_not_zero(&state->count))
675                         return state;
676         }
677         return NULL;
678 }
679
680 static void
681 nfs4_free_open_state(struct nfs4_state *state)
682 {
683         kfree(state);
684 }
685
686 struct nfs4_state *
687 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
688 {
689         struct nfs4_state *state, *new;
690         struct nfs_inode *nfsi = NFS_I(inode);
691
692         spin_lock(&inode->i_lock);
693         state = __nfs4_find_state_byowner(inode, owner);
694         spin_unlock(&inode->i_lock);
695         if (state)
696                 goto out;
697         new = nfs4_alloc_open_state();
698         spin_lock(&owner->so_lock);
699         spin_lock(&inode->i_lock);
700         state = __nfs4_find_state_byowner(inode, owner);
701         if (state == NULL && new != NULL) {
702                 state = new;
703                 state->owner = owner;
704                 atomic_inc(&owner->so_count);
705                 list_add(&state->inode_states, &nfsi->open_states);
706                 ihold(inode);
707                 state->inode = inode;
708                 spin_unlock(&inode->i_lock);
709                 /* Note: The reclaim code dictates that we add stateless
710                  * and read-only stateids to the end of the list */
711                 list_add_tail(&state->open_states, &owner->so_states);
712                 spin_unlock(&owner->so_lock);
713         } else {
714                 spin_unlock(&inode->i_lock);
715                 spin_unlock(&owner->so_lock);
716                 if (new)
717                         nfs4_free_open_state(new);
718         }
719 out:
720         return state;
721 }
722
723 void nfs4_put_open_state(struct nfs4_state *state)
724 {
725         struct inode *inode = state->inode;
726         struct nfs4_state_owner *owner = state->owner;
727
728         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
729                 return;
730         spin_lock(&inode->i_lock);
731         list_del(&state->inode_states);
732         list_del(&state->open_states);
733         spin_unlock(&inode->i_lock);
734         spin_unlock(&owner->so_lock);
735         iput(inode);
736         nfs4_free_open_state(state);
737         nfs4_put_state_owner(owner);
738 }
739
740 /*
741  * Close the current file.
742  */
743 static void __nfs4_close(struct nfs4_state *state,
744                 fmode_t fmode, gfp_t gfp_mask, int wait)
745 {
746         struct nfs4_state_owner *owner = state->owner;
747         int call_close = 0;
748         fmode_t newstate;
749
750         atomic_inc(&owner->so_count);
751         /* Protect against nfs4_find_state() */
752         spin_lock(&owner->so_lock);
753         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
754                 case FMODE_READ:
755                         state->n_rdonly--;
756                         break;
757                 case FMODE_WRITE:
758                         state->n_wronly--;
759                         break;
760                 case FMODE_READ|FMODE_WRITE:
761                         state->n_rdwr--;
762         }
763         newstate = FMODE_READ|FMODE_WRITE;
764         if (state->n_rdwr == 0) {
765                 if (state->n_rdonly == 0) {
766                         newstate &= ~FMODE_READ;
767                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
768                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
769                 }
770                 if (state->n_wronly == 0) {
771                         newstate &= ~FMODE_WRITE;
772                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
773                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
774                 }
775                 if (newstate == 0)
776                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
777         }
778         nfs4_state_set_mode_locked(state, newstate);
779         spin_unlock(&owner->so_lock);
780
781         if (!call_close) {
782                 nfs4_put_open_state(state);
783                 nfs4_put_state_owner(owner);
784         } else
785                 nfs4_do_close(state, gfp_mask, wait);
786 }
787
788 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
789 {
790         __nfs4_close(state, fmode, GFP_NOFS, 0);
791 }
792
793 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
794 {
795         __nfs4_close(state, fmode, GFP_KERNEL, 1);
796 }
797
798 /*
799  * Search the state->lock_states for an existing lock_owner
800  * that is compatible with current->files
801  */
802 static struct nfs4_lock_state *
803 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
804 {
805         struct nfs4_lock_state *pos;
806         list_for_each_entry(pos, &state->lock_states, ls_locks) {
807                 if (pos->ls_owner != fl_owner)
808                         continue;
809                 atomic_inc(&pos->ls_count);
810                 return pos;
811         }
812         return NULL;
813 }
814
815 /*
816  * Return a compatible lock_state. If no initialized lock_state structure
817  * exists, return an uninitialized one.
818  *
819  */
820 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
821 {
822         struct nfs4_lock_state *lsp;
823         struct nfs_server *server = state->owner->so_server;
824
825         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
826         if (lsp == NULL)
827                 return NULL;
828         nfs4_init_seqid_counter(&lsp->ls_seqid);
829         atomic_set(&lsp->ls_count, 1);
830         lsp->ls_state = state;
831         lsp->ls_owner = fl_owner;
832         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
833         if (lsp->ls_seqid.owner_id < 0)
834                 goto out_free;
835         INIT_LIST_HEAD(&lsp->ls_locks);
836         return lsp;
837 out_free:
838         kfree(lsp);
839         return NULL;
840 }
841
842 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
843 {
844         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
845         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
846         kfree(lsp);
847 }
848
849 /*
850  * Return a compatible lock_state. If no initialized lock_state structure
851  * exists, return an uninitialized one.
852  *
853  */
854 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
855 {
856         struct nfs4_lock_state *lsp, *new = NULL;
857         
858         for(;;) {
859                 spin_lock(&state->state_lock);
860                 lsp = __nfs4_find_lock_state(state, owner);
861                 if (lsp != NULL)
862                         break;
863                 if (new != NULL) {
864                         list_add(&new->ls_locks, &state->lock_states);
865                         set_bit(LK_STATE_IN_USE, &state->flags);
866                         lsp = new;
867                         new = NULL;
868                         break;
869                 }
870                 spin_unlock(&state->state_lock);
871                 new = nfs4_alloc_lock_state(state, owner);
872                 if (new == NULL)
873                         return NULL;
874         }
875         spin_unlock(&state->state_lock);
876         if (new != NULL)
877                 nfs4_free_lock_state(state->owner->so_server, new);
878         return lsp;
879 }
880
881 /*
882  * Release reference to lock_state, and free it if we see that
883  * it is no longer in use
884  */
885 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
886 {
887         struct nfs_server *server;
888         struct nfs4_state *state;
889
890         if (lsp == NULL)
891                 return;
892         state = lsp->ls_state;
893         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
894                 return;
895         list_del(&lsp->ls_locks);
896         if (list_empty(&state->lock_states))
897                 clear_bit(LK_STATE_IN_USE, &state->flags);
898         spin_unlock(&state->state_lock);
899         server = state->owner->so_server;
900         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
901                 struct nfs_client *clp = server->nfs_client;
902
903                 clp->cl_mvops->free_lock_state(server, lsp);
904         } else
905                 nfs4_free_lock_state(server, lsp);
906 }
907
908 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
909 {
910         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
911
912         dst->fl_u.nfs4_fl.owner = lsp;
913         atomic_inc(&lsp->ls_count);
914 }
915
916 static void nfs4_fl_release_lock(struct file_lock *fl)
917 {
918         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
919 }
920
921 static const struct file_lock_operations nfs4_fl_lock_ops = {
922         .fl_copy_lock = nfs4_fl_copy_lock,
923         .fl_release_private = nfs4_fl_release_lock,
924 };
925
926 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
927 {
928         struct nfs4_lock_state *lsp;
929
930         if (fl->fl_ops != NULL)
931                 return 0;
932         lsp = nfs4_get_lock_state(state, fl->fl_owner);
933         if (lsp == NULL)
934                 return -ENOMEM;
935         fl->fl_u.nfs4_fl.owner = lsp;
936         fl->fl_ops = &nfs4_fl_lock_ops;
937         return 0;
938 }
939
940 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
941                 struct nfs4_state *state,
942                 const struct nfs_lockowner *lockowner)
943 {
944         struct nfs4_lock_state *lsp;
945         fl_owner_t fl_owner;
946         int ret = -ENOENT;
947
948
949         if (lockowner == NULL)
950                 goto out;
951
952         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
953                 goto out;
954
955         fl_owner = lockowner->l_owner;
956         spin_lock(&state->state_lock);
957         lsp = __nfs4_find_lock_state(state, fl_owner);
958         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
959                 ret = -EIO;
960         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
961                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
962                 ret = 0;
963         }
964         spin_unlock(&state->state_lock);
965         nfs4_put_lock_state(lsp);
966 out:
967         return ret;
968 }
969
970 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
971 {
972         const nfs4_stateid *src;
973         int seq;
974
975         do {
976                 src = &zero_stateid;
977                 seq = read_seqbegin(&state->seqlock);
978                 if (test_bit(NFS_OPEN_STATE, &state->flags))
979                         src = &state->open_stateid;
980                 nfs4_stateid_copy(dst, src);
981         } while (read_seqretry(&state->seqlock, seq));
982 }
983
984 /*
985  * Byte-range lock aware utility to initialize the stateid of read/write
986  * requests.
987  */
988 int nfs4_select_rw_stateid(struct nfs4_state *state,
989                 fmode_t fmode, const struct nfs_lockowner *lockowner,
990                 nfs4_stateid *dst, struct rpc_cred **cred)
991 {
992         int ret;
993
994         if (cred != NULL)
995                 *cred = NULL;
996         ret = nfs4_copy_lock_stateid(dst, state, lockowner);
997         if (ret == -EIO)
998                 /* A lost lock - don't even consider delegations */
999                 goto out;
1000         /* returns true if delegation stateid found and copied */
1001         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1002                 ret = 0;
1003                 goto out;
1004         }
1005         if (ret != -ENOENT)
1006                 /* nfs4_copy_delegation_stateid() didn't over-write
1007                  * dst, so it still has the lock stateid which we now
1008                  * choose to use.
1009                  */
1010                 goto out;
1011         nfs4_copy_open_stateid(dst, state);
1012         ret = 0;
1013 out:
1014         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1015                 dst->seqid = 0;
1016         return ret;
1017 }
1018
1019 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1020 {
1021         struct nfs_seqid *new;
1022
1023         new = kmalloc(sizeof(*new), gfp_mask);
1024         if (new == NULL)
1025                 return ERR_PTR(-ENOMEM);
1026         new->sequence = counter;
1027         INIT_LIST_HEAD(&new->list);
1028         new->task = NULL;
1029         return new;
1030 }
1031
1032 void nfs_release_seqid(struct nfs_seqid *seqid)
1033 {
1034         struct nfs_seqid_counter *sequence;
1035
1036         if (seqid == NULL || list_empty(&seqid->list))
1037                 return;
1038         sequence = seqid->sequence;
1039         spin_lock(&sequence->lock);
1040         list_del_init(&seqid->list);
1041         if (!list_empty(&sequence->list)) {
1042                 struct nfs_seqid *next;
1043
1044                 next = list_first_entry(&sequence->list,
1045                                 struct nfs_seqid, list);
1046                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1047         }
1048         spin_unlock(&sequence->lock);
1049 }
1050
1051 void nfs_free_seqid(struct nfs_seqid *seqid)
1052 {
1053         nfs_release_seqid(seqid);
1054         kfree(seqid);
1055 }
1056
1057 /*
1058  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1059  * failed with a seqid incrementing error -
1060  * see comments nfs4.h:seqid_mutating_error()
1061  */
1062 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1063 {
1064         switch (status) {
1065                 case 0:
1066                         break;
1067                 case -NFS4ERR_BAD_SEQID:
1068                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1069                                 return;
1070                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1071                                         " sequence-id error on an"
1072                                         " unconfirmed sequence %p!\n",
1073                                         seqid->sequence);
1074                 case -NFS4ERR_STALE_CLIENTID:
1075                 case -NFS4ERR_STALE_STATEID:
1076                 case -NFS4ERR_BAD_STATEID:
1077                 case -NFS4ERR_BADXDR:
1078                 case -NFS4ERR_RESOURCE:
1079                 case -NFS4ERR_NOFILEHANDLE:
1080                         /* Non-seqid mutating errors */
1081                         return;
1082         };
1083         /*
1084          * Note: no locking needed as we are guaranteed to be first
1085          * on the sequence list
1086          */
1087         seqid->sequence->counter++;
1088 }
1089
1090 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1091 {
1092         struct nfs4_state_owner *sp;
1093
1094         if (seqid == NULL)
1095                 return;
1096
1097         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1098         if (status == -NFS4ERR_BAD_SEQID)
1099                 nfs4_drop_state_owner(sp);
1100         if (!nfs4_has_session(sp->so_server->nfs_client))
1101                 nfs_increment_seqid(status, seqid);
1102 }
1103
1104 /*
1105  * Increment the seqid if the LOCK/LOCKU succeeded, or
1106  * failed with a seqid incrementing error -
1107  * see comments nfs4.h:seqid_mutating_error()
1108  */
1109 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1110 {
1111         if (seqid != NULL)
1112                 nfs_increment_seqid(status, seqid);
1113 }
1114
1115 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1116 {
1117         struct nfs_seqid_counter *sequence;
1118         int status = 0;
1119
1120         if (seqid == NULL)
1121                 goto out;
1122         sequence = seqid->sequence;
1123         spin_lock(&sequence->lock);
1124         seqid->task = task;
1125         if (list_empty(&seqid->list))
1126                 list_add_tail(&seqid->list, &sequence->list);
1127         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1128                 goto unlock;
1129         rpc_sleep_on(&sequence->wait, task, NULL);
1130         status = -EAGAIN;
1131 unlock:
1132         spin_unlock(&sequence->lock);
1133 out:
1134         return status;
1135 }
1136
1137 static int nfs4_run_state_manager(void *);
1138
1139 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1140 {
1141         smp_mb__before_atomic();
1142         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1143         smp_mb__after_atomic();
1144         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1145         rpc_wake_up(&clp->cl_rpcwaitq);
1146 }
1147
1148 /*
1149  * Schedule the nfs_client asynchronous state management routine
1150  */
1151 void nfs4_schedule_state_manager(struct nfs_client *clp)
1152 {
1153         struct task_struct *task;
1154         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1155
1156         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1157                 return;
1158         __module_get(THIS_MODULE);
1159         atomic_inc(&clp->cl_count);
1160
1161         /* The rcu_read_lock() is not strictly necessary, as the state
1162          * manager is the only thread that ever changes the rpc_xprt
1163          * after it's initialized.  At this point, we're single threaded. */
1164         rcu_read_lock();
1165         snprintf(buf, sizeof(buf), "%s-manager",
1166                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1167         rcu_read_unlock();
1168         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1169         if (IS_ERR(task)) {
1170                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1171                         __func__, PTR_ERR(task));
1172                 nfs4_clear_state_manager_bit(clp);
1173                 nfs_put_client(clp);
1174                 module_put(THIS_MODULE);
1175         }
1176 }
1177
1178 /*
1179  * Schedule a lease recovery attempt
1180  */
1181 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1182 {
1183         if (!clp)
1184                 return;
1185         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1186                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1187         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1188                         clp->cl_hostname);
1189         nfs4_schedule_state_manager(clp);
1190 }
1191 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1192
1193 /**
1194  * nfs4_schedule_migration_recovery - trigger migration recovery
1195  *
1196  * @server: FSID that is migrating
1197  *
1198  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1199  * value is returned.
1200  */
1201 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1202 {
1203         struct nfs_client *clp = server->nfs_client;
1204
1205         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1206                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1207                                 clp->cl_hostname);
1208                 return -NFS4ERR_IO;
1209         }
1210
1211         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1212                 return -NFS4ERR_IO;
1213
1214         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1215                         __func__,
1216                         (unsigned long long)server->fsid.major,
1217                         (unsigned long long)server->fsid.minor,
1218                         clp->cl_hostname);
1219
1220         set_bit(NFS_MIG_IN_TRANSITION,
1221                         &((struct nfs_server *)server)->mig_status);
1222         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1223
1224         nfs4_schedule_state_manager(clp);
1225         return 0;
1226 }
1227 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1228
1229 /**
1230  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1231  *
1232  * @clp: server to check for moved leases
1233  *
1234  */
1235 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1236 {
1237         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1238                 __func__, clp->cl_clientid, clp->cl_hostname);
1239
1240         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1241         nfs4_schedule_state_manager(clp);
1242 }
1243 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1244
1245 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1246 {
1247         int res;
1248
1249         might_sleep();
1250
1251         atomic_inc(&clp->cl_count);
1252         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1253                                  nfs_wait_bit_killable, TASK_KILLABLE);
1254         if (res)
1255                 goto out;
1256         if (clp->cl_cons_state < 0)
1257                 res = clp->cl_cons_state;
1258 out:
1259         nfs_put_client(clp);
1260         return res;
1261 }
1262
1263 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1264 {
1265         unsigned int loop;
1266         int ret;
1267
1268         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1269                 ret = nfs4_wait_clnt_recover(clp);
1270                 if (ret != 0)
1271                         break;
1272                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1273                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1274                         break;
1275                 nfs4_schedule_state_manager(clp);
1276                 ret = -EIO;
1277         }
1278         return ret;
1279 }
1280
1281 /*
1282  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1283  * @clp: client to process
1284  *
1285  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1286  * resend of the SETCLIENTID and hence re-establish the
1287  * callback channel. Then return all existing delegations.
1288  */
1289 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1290 {
1291         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1292         nfs_expire_all_delegations(clp);
1293         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1294                         clp->cl_hostname);
1295 }
1296
1297 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1298 {
1299         nfs40_handle_cb_pathdown(clp);
1300         nfs4_schedule_state_manager(clp);
1301 }
1302
1303 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1304 {
1305
1306         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1307         /* Don't recover state that expired before the reboot */
1308         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1309                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1310                 return 0;
1311         }
1312         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1313         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1314         return 1;
1315 }
1316
1317 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1318 {
1319         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1320         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1321         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1322         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1323         return 1;
1324 }
1325
1326 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1327 {
1328         struct nfs_client *clp = server->nfs_client;
1329
1330         if (!nfs4_valid_open_stateid(state))
1331                 return -EBADF;
1332         nfs4_state_mark_reclaim_nograce(clp, state);
1333         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1334                         clp->cl_hostname);
1335         nfs4_schedule_state_manager(clp);
1336         return 0;
1337 }
1338 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1339
1340 void nfs_inode_find_state_and_recover(struct inode *inode,
1341                 const nfs4_stateid *stateid)
1342 {
1343         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1344         struct nfs_inode *nfsi = NFS_I(inode);
1345         struct nfs_open_context *ctx;
1346         struct nfs4_state *state;
1347         bool found = false;
1348
1349         spin_lock(&inode->i_lock);
1350         list_for_each_entry(ctx, &nfsi->open_files, list) {
1351                 state = ctx->state;
1352                 if (state == NULL)
1353                         continue;
1354                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1355                         continue;
1356                 if (!nfs4_stateid_match(&state->stateid, stateid))
1357                         continue;
1358                 nfs4_state_mark_reclaim_nograce(clp, state);
1359                 found = true;
1360         }
1361         spin_unlock(&inode->i_lock);
1362         if (found)
1363                 nfs4_schedule_state_manager(clp);
1364 }
1365
1366 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1367 {
1368         struct inode *inode = state->inode;
1369         struct nfs_inode *nfsi = NFS_I(inode);
1370         struct nfs_open_context *ctx;
1371
1372         spin_lock(&inode->i_lock);
1373         list_for_each_entry(ctx, &nfsi->open_files, list) {
1374                 if (ctx->state != state)
1375                         continue;
1376                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1377         }
1378         spin_unlock(&inode->i_lock);
1379 }
1380
1381 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1382 {
1383         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1384         nfs4_state_mark_open_context_bad(state);
1385 }
1386
1387
1388 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1389 {
1390         struct inode *inode = state->inode;
1391         struct nfs_inode *nfsi = NFS_I(inode);
1392         struct file_lock *fl;
1393         int status = 0;
1394         struct file_lock_context *flctx = inode->i_flctx;
1395         struct list_head *list;
1396
1397         if (flctx == NULL)
1398                 return 0;
1399
1400         list = &flctx->flc_posix;
1401
1402         /* Guard against delegation returns and new lock/unlock calls */
1403         down_write(&nfsi->rwsem);
1404         spin_lock(&flctx->flc_lock);
1405 restart:
1406         list_for_each_entry(fl, list, fl_list) {
1407                 if (nfs_file_open_context(fl->fl_file)->state != state)
1408                         continue;
1409                 spin_unlock(&flctx->flc_lock);
1410                 status = ops->recover_lock(state, fl);
1411                 switch (status) {
1412                 case 0:
1413                         break;
1414                 case -ESTALE:
1415                 case -NFS4ERR_ADMIN_REVOKED:
1416                 case -NFS4ERR_STALE_STATEID:
1417                 case -NFS4ERR_BAD_STATEID:
1418                 case -NFS4ERR_EXPIRED:
1419                 case -NFS4ERR_NO_GRACE:
1420                 case -NFS4ERR_STALE_CLIENTID:
1421                 case -NFS4ERR_BADSESSION:
1422                 case -NFS4ERR_BADSLOT:
1423                 case -NFS4ERR_BAD_HIGH_SLOT:
1424                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1425                         goto out;
1426                 default:
1427                         pr_err("NFS: %s: unhandled error %d\n",
1428                                         __func__, status);
1429                 case -ENOMEM:
1430                 case -NFS4ERR_DENIED:
1431                 case -NFS4ERR_RECLAIM_BAD:
1432                 case -NFS4ERR_RECLAIM_CONFLICT:
1433                         /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1434                         status = 0;
1435                 }
1436                 spin_lock(&flctx->flc_lock);
1437         }
1438         if (list == &flctx->flc_posix) {
1439                 list = &flctx->flc_flock;
1440                 goto restart;
1441         }
1442         spin_unlock(&flctx->flc_lock);
1443 out:
1444         up_write(&nfsi->rwsem);
1445         return status;
1446 }
1447
1448 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1449 {
1450         struct nfs4_state *state;
1451         struct nfs4_lock_state *lock;
1452         int status = 0;
1453
1454         /* Note: we rely on the sp->so_states list being ordered 
1455          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1456          * states first.
1457          * This is needed to ensure that the server won't give us any
1458          * read delegations that we have to return if, say, we are
1459          * recovering after a network partition or a reboot from a
1460          * server that doesn't support a grace period.
1461          */
1462         spin_lock(&sp->so_lock);
1463         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1464 restart:
1465         list_for_each_entry(state, &sp->so_states, open_states) {
1466                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1467                         continue;
1468                 if (!nfs4_valid_open_stateid(state))
1469                         continue;
1470                 if (state->state == 0)
1471                         continue;
1472                 atomic_inc(&state->count);
1473                 spin_unlock(&sp->so_lock);
1474                 status = ops->recover_open(sp, state);
1475                 if (status >= 0) {
1476                         status = nfs4_reclaim_locks(state, ops);
1477                         if (status >= 0) {
1478                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1479                                         spin_lock(&state->state_lock);
1480                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1481                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1482                                                         pr_warn_ratelimited("NFS: "
1483                                                                             "%s: Lock reclaim "
1484                                                                             "failed!\n", __func__);
1485                                         }
1486                                         spin_unlock(&state->state_lock);
1487                                 }
1488                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1489                                         &state->flags);
1490                                 nfs4_put_open_state(state);
1491                                 spin_lock(&sp->so_lock);
1492                                 goto restart;
1493                         }
1494                 }
1495                 switch (status) {
1496                         default:
1497                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1498                                         __func__, status);
1499                         case -ENOENT:
1500                         case -ENOMEM:
1501                         case -ESTALE:
1502                                 /* Open state on this file cannot be recovered */
1503                                 nfs4_state_mark_recovery_failed(state, status);
1504                                 break;
1505                         case -EAGAIN:
1506                                 ssleep(1);
1507                         case -NFS4ERR_ADMIN_REVOKED:
1508                         case -NFS4ERR_STALE_STATEID:
1509                         case -NFS4ERR_BAD_STATEID:
1510                         case -NFS4ERR_RECLAIM_BAD:
1511                         case -NFS4ERR_RECLAIM_CONFLICT:
1512                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1513                                 break;
1514                         case -NFS4ERR_EXPIRED:
1515                         case -NFS4ERR_NO_GRACE:
1516                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1517                         case -NFS4ERR_STALE_CLIENTID:
1518                         case -NFS4ERR_BADSESSION:
1519                         case -NFS4ERR_BADSLOT:
1520                         case -NFS4ERR_BAD_HIGH_SLOT:
1521                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1522                                 goto out_err;
1523                 }
1524                 nfs4_put_open_state(state);
1525                 spin_lock(&sp->so_lock);
1526                 goto restart;
1527         }
1528         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1529         spin_unlock(&sp->so_lock);
1530         return 0;
1531 out_err:
1532         nfs4_put_open_state(state);
1533         spin_lock(&sp->so_lock);
1534         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1535         spin_unlock(&sp->so_lock);
1536         return status;
1537 }
1538
1539 static void nfs4_clear_open_state(struct nfs4_state *state)
1540 {
1541         struct nfs4_lock_state *lock;
1542
1543         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1544         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1545         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1546         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1547         spin_lock(&state->state_lock);
1548         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1549                 lock->ls_seqid.flags = 0;
1550                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1551         }
1552         spin_unlock(&state->state_lock);
1553 }
1554
1555 static void nfs4_reset_seqids(struct nfs_server *server,
1556         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1557 {
1558         struct nfs_client *clp = server->nfs_client;
1559         struct nfs4_state_owner *sp;
1560         struct rb_node *pos;
1561         struct nfs4_state *state;
1562
1563         spin_lock(&clp->cl_lock);
1564         for (pos = rb_first(&server->state_owners);
1565              pos != NULL;
1566              pos = rb_next(pos)) {
1567                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1568                 sp->so_seqid.flags = 0;
1569                 spin_lock(&sp->so_lock);
1570                 list_for_each_entry(state, &sp->so_states, open_states) {
1571                         if (mark_reclaim(clp, state))
1572                                 nfs4_clear_open_state(state);
1573                 }
1574                 spin_unlock(&sp->so_lock);
1575         }
1576         spin_unlock(&clp->cl_lock);
1577 }
1578
1579 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1580         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1581 {
1582         struct nfs_server *server;
1583
1584         rcu_read_lock();
1585         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1586                 nfs4_reset_seqids(server, mark_reclaim);
1587         rcu_read_unlock();
1588 }
1589
1590 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1591 {
1592         /* Mark all delegations for reclaim */
1593         nfs_delegation_mark_reclaim(clp);
1594         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1595 }
1596
1597 static void nfs4_reclaim_complete(struct nfs_client *clp,
1598                                  const struct nfs4_state_recovery_ops *ops,
1599                                  struct rpc_cred *cred)
1600 {
1601         /* Notify the server we're done reclaiming our state */
1602         if (ops->reclaim_complete)
1603                 (void)ops->reclaim_complete(clp, cred);
1604 }
1605
1606 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1607 {
1608         struct nfs_client *clp = server->nfs_client;
1609         struct nfs4_state_owner *sp;
1610         struct rb_node *pos;
1611         struct nfs4_state *state;
1612
1613         spin_lock(&clp->cl_lock);
1614         for (pos = rb_first(&server->state_owners);
1615              pos != NULL;
1616              pos = rb_next(pos)) {
1617                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1618                 spin_lock(&sp->so_lock);
1619                 list_for_each_entry(state, &sp->so_states, open_states) {
1620                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1621                                                 &state->flags))
1622                                 continue;
1623                         nfs4_state_mark_reclaim_nograce(clp, state);
1624                 }
1625                 spin_unlock(&sp->so_lock);
1626         }
1627         spin_unlock(&clp->cl_lock);
1628 }
1629
1630 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1631 {
1632         struct nfs_server *server;
1633
1634         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1635                 return 0;
1636
1637         rcu_read_lock();
1638         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1639                 nfs4_clear_reclaim_server(server);
1640         rcu_read_unlock();
1641
1642         nfs_delegation_reap_unclaimed(clp);
1643         return 1;
1644 }
1645
1646 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1647 {
1648         const struct nfs4_state_recovery_ops *ops;
1649         struct rpc_cred *cred;
1650
1651         if (!nfs4_state_clear_reclaim_reboot(clp))
1652                 return;
1653         ops = clp->cl_mvops->reboot_recovery_ops;
1654         cred = nfs4_get_clid_cred(clp);
1655         nfs4_reclaim_complete(clp, ops, cred);
1656         put_rpccred(cred);
1657 }
1658
1659 static void nfs_delegation_clear_all(struct nfs_client *clp)
1660 {
1661         nfs_delegation_mark_reclaim(clp);
1662         nfs_delegation_reap_unclaimed(clp);
1663 }
1664
1665 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1666 {
1667         nfs_delegation_clear_all(clp);
1668         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1669 }
1670
1671 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1672 {
1673         switch (error) {
1674                 case 0:
1675                         break;
1676                 case -NFS4ERR_CB_PATH_DOWN:
1677                         nfs40_handle_cb_pathdown(clp);
1678                         break;
1679                 case -NFS4ERR_NO_GRACE:
1680                         nfs4_state_end_reclaim_reboot(clp);
1681                         break;
1682                 case -NFS4ERR_STALE_CLIENTID:
1683                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1684                         nfs4_state_clear_reclaim_reboot(clp);
1685                         nfs4_state_start_reclaim_reboot(clp);
1686                         break;
1687                 case -NFS4ERR_EXPIRED:
1688                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1689                         nfs4_state_start_reclaim_nograce(clp);
1690                         break;
1691                 case -NFS4ERR_BADSESSION:
1692                 case -NFS4ERR_BADSLOT:
1693                 case -NFS4ERR_BAD_HIGH_SLOT:
1694                 case -NFS4ERR_DEADSESSION:
1695                 case -NFS4ERR_SEQ_FALSE_RETRY:
1696                 case -NFS4ERR_SEQ_MISORDERED:
1697                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1698                         /* Zero session reset errors */
1699                         break;
1700                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1701                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1702                         break;
1703                 default:
1704                         dprintk("%s: failed to handle error %d for server %s\n",
1705                                         __func__, error, clp->cl_hostname);
1706                         return error;
1707         }
1708         dprintk("%s: handled error %d for server %s\n", __func__, error,
1709                         clp->cl_hostname);
1710         return 0;
1711 }
1712
1713 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1714 {
1715         struct nfs4_state_owner *sp;
1716         struct nfs_server *server;
1717         struct rb_node *pos;
1718         int status = 0;
1719
1720 restart:
1721         rcu_read_lock();
1722         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1723                 nfs4_purge_state_owners(server);
1724                 spin_lock(&clp->cl_lock);
1725                 for (pos = rb_first(&server->state_owners);
1726                      pos != NULL;
1727                      pos = rb_next(pos)) {
1728                         sp = rb_entry(pos,
1729                                 struct nfs4_state_owner, so_server_node);
1730                         if (!test_and_clear_bit(ops->owner_flag_bit,
1731                                                         &sp->so_flags))
1732                                 continue;
1733                         if (!atomic_inc_not_zero(&sp->so_count))
1734                                 continue;
1735                         spin_unlock(&clp->cl_lock);
1736                         rcu_read_unlock();
1737
1738                         status = nfs4_reclaim_open_state(sp, ops);
1739                         if (status < 0) {
1740                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1741                                 nfs4_put_state_owner(sp);
1742                                 status = nfs4_recovery_handle_error(clp, status);
1743                                 return (status != 0) ? status : -EAGAIN;
1744                         }
1745
1746                         nfs4_put_state_owner(sp);
1747                         goto restart;
1748                 }
1749                 spin_unlock(&clp->cl_lock);
1750         }
1751         rcu_read_unlock();
1752         return 0;
1753 }
1754
1755 static int nfs4_check_lease(struct nfs_client *clp)
1756 {
1757         struct rpc_cred *cred;
1758         const struct nfs4_state_maintenance_ops *ops =
1759                 clp->cl_mvops->state_renewal_ops;
1760         int status;
1761
1762         /* Is the client already known to have an expired lease? */
1763         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1764                 return 0;
1765         spin_lock(&clp->cl_lock);
1766         cred = ops->get_state_renewal_cred_locked(clp);
1767         spin_unlock(&clp->cl_lock);
1768         if (cred == NULL) {
1769                 cred = nfs4_get_clid_cred(clp);
1770                 status = -ENOKEY;
1771                 if (cred == NULL)
1772                         goto out;
1773         }
1774         status = ops->renew_lease(clp, cred);
1775         put_rpccred(cred);
1776         if (status == -ETIMEDOUT) {
1777                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1778                 return 0;
1779         }
1780 out:
1781         return nfs4_recovery_handle_error(clp, status);
1782 }
1783
1784 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1785  * and for recoverable errors on EXCHANGE_ID for v4.1
1786  */
1787 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1788 {
1789         switch (status) {
1790         case -NFS4ERR_SEQ_MISORDERED:
1791                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1792                         return -ESERVERFAULT;
1793                 /* Lease confirmation error: retry after purging the lease */
1794                 ssleep(1);
1795                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1796                 break;
1797         case -NFS4ERR_STALE_CLIENTID:
1798                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1799                 nfs4_state_start_reclaim_reboot(clp);
1800                 break;
1801         case -NFS4ERR_CLID_INUSE:
1802                 pr_err("NFS: Server %s reports our clientid is in use\n",
1803                         clp->cl_hostname);
1804                 nfs_mark_client_ready(clp, -EPERM);
1805                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1806                 return -EPERM;
1807         case -EACCES:
1808         case -NFS4ERR_DELAY:
1809         case -ETIMEDOUT:
1810         case -EAGAIN:
1811                 ssleep(1);
1812                 break;
1813
1814         case -NFS4ERR_MINOR_VERS_MISMATCH:
1815                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1816                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1817                 dprintk("%s: exit with error %d for server %s\n",
1818                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1819                 return -EPROTONOSUPPORT;
1820         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1821                                  * in nfs4_exchange_id */
1822         default:
1823                 dprintk("%s: exit with error %d for server %s\n", __func__,
1824                                 status, clp->cl_hostname);
1825                 return status;
1826         }
1827         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1828         dprintk("%s: handled error %d for server %s\n", __func__, status,
1829                         clp->cl_hostname);
1830         return 0;
1831 }
1832
1833 static int nfs4_establish_lease(struct nfs_client *clp)
1834 {
1835         struct rpc_cred *cred;
1836         const struct nfs4_state_recovery_ops *ops =
1837                 clp->cl_mvops->reboot_recovery_ops;
1838         int status;
1839
1840         nfs4_begin_drain_session(clp);
1841         cred = nfs4_get_clid_cred(clp);
1842         if (cred == NULL)
1843                 return -ENOENT;
1844         status = ops->establish_clid(clp, cred);
1845         put_rpccred(cred);
1846         if (status != 0)
1847                 return status;
1848         pnfs_destroy_all_layouts(clp);
1849         return 0;
1850 }
1851
1852 /*
1853  * Returns zero or a negative errno.  NFS4ERR values are converted
1854  * to local errno values.
1855  */
1856 static int nfs4_reclaim_lease(struct nfs_client *clp)
1857 {
1858         int status;
1859
1860         status = nfs4_establish_lease(clp);
1861         if (status < 0)
1862                 return nfs4_handle_reclaim_lease_error(clp, status);
1863         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1864                 nfs4_state_start_reclaim_nograce(clp);
1865         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1866                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1867         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1868         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1869         return 0;
1870 }
1871
1872 static int nfs4_purge_lease(struct nfs_client *clp)
1873 {
1874         int status;
1875
1876         status = nfs4_establish_lease(clp);
1877         if (status < 0)
1878                 return nfs4_handle_reclaim_lease_error(clp, status);
1879         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1880         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1881         nfs4_state_start_reclaim_nograce(clp);
1882         return 0;
1883 }
1884
1885 /*
1886  * Try remote migration of one FSID from a source server to a
1887  * destination server.  The source server provides a list of
1888  * potential destinations.
1889  *
1890  * Returns zero or a negative NFS4ERR status code.
1891  */
1892 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1893 {
1894         struct nfs_client *clp = server->nfs_client;
1895         struct nfs4_fs_locations *locations = NULL;
1896         struct inode *inode;
1897         struct page *page;
1898         int status, result;
1899
1900         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1901                         (unsigned long long)server->fsid.major,
1902                         (unsigned long long)server->fsid.minor,
1903                         clp->cl_hostname);
1904
1905         result = 0;
1906         page = alloc_page(GFP_KERNEL);
1907         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1908         if (page == NULL || locations == NULL) {
1909                 dprintk("<-- %s: no memory\n", __func__);
1910                 goto out;
1911         }
1912
1913         inode = d_inode(server->super->s_root);
1914         result = nfs4_proc_get_locations(inode, locations, page, cred);
1915         if (result) {
1916                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1917                         __func__, result);
1918                 goto out;
1919         }
1920
1921         result = -NFS4ERR_NXIO;
1922         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1923                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1924                         __func__);
1925                 goto out;
1926         }
1927
1928         nfs4_begin_drain_session(clp);
1929
1930         status = nfs4_replace_transport(server, locations);
1931         if (status != 0) {
1932                 dprintk("<-- %s: failed to replace transport: %d\n",
1933                         __func__, status);
1934                 goto out;
1935         }
1936
1937         result = 0;
1938         dprintk("<-- %s: migration succeeded\n", __func__);
1939
1940 out:
1941         if (page != NULL)
1942                 __free_page(page);
1943         kfree(locations);
1944         if (result) {
1945                 pr_err("NFS: migration recovery failed (server %s)\n",
1946                                 clp->cl_hostname);
1947                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1948         }
1949         return result;
1950 }
1951
1952 /*
1953  * Returns zero or a negative NFS4ERR status code.
1954  */
1955 static int nfs4_handle_migration(struct nfs_client *clp)
1956 {
1957         const struct nfs4_state_maintenance_ops *ops =
1958                                 clp->cl_mvops->state_renewal_ops;
1959         struct nfs_server *server;
1960         struct rpc_cred *cred;
1961
1962         dprintk("%s: migration reported on \"%s\"\n", __func__,
1963                         clp->cl_hostname);
1964
1965         spin_lock(&clp->cl_lock);
1966         cred = ops->get_state_renewal_cred_locked(clp);
1967         spin_unlock(&clp->cl_lock);
1968         if (cred == NULL)
1969                 return -NFS4ERR_NOENT;
1970
1971         clp->cl_mig_gen++;
1972 restart:
1973         rcu_read_lock();
1974         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1975                 int status;
1976
1977                 if (server->mig_gen == clp->cl_mig_gen)
1978                         continue;
1979                 server->mig_gen = clp->cl_mig_gen;
1980
1981                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1982                                                 &server->mig_status))
1983                         continue;
1984
1985                 rcu_read_unlock();
1986                 status = nfs4_try_migration(server, cred);
1987                 if (status < 0) {
1988                         put_rpccred(cred);
1989                         return status;
1990                 }
1991                 goto restart;
1992         }
1993         rcu_read_unlock();
1994         put_rpccred(cred);
1995         return 0;
1996 }
1997
1998 /*
1999  * Test each nfs_server on the clp's cl_superblocks list to see
2000  * if it's moved to another server.  Stop when the server no longer
2001  * returns NFS4ERR_LEASE_MOVED.
2002  */
2003 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2004 {
2005         const struct nfs4_state_maintenance_ops *ops =
2006                                 clp->cl_mvops->state_renewal_ops;
2007         struct nfs_server *server;
2008         struct rpc_cred *cred;
2009
2010         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2011                         clp->cl_hostname);
2012
2013         spin_lock(&clp->cl_lock);
2014         cred = ops->get_state_renewal_cred_locked(clp);
2015         spin_unlock(&clp->cl_lock);
2016         if (cred == NULL)
2017                 return -NFS4ERR_NOENT;
2018
2019         clp->cl_mig_gen++;
2020 restart:
2021         rcu_read_lock();
2022         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2023                 struct inode *inode;
2024                 int status;
2025
2026                 if (server->mig_gen == clp->cl_mig_gen)
2027                         continue;
2028                 server->mig_gen = clp->cl_mig_gen;
2029
2030                 rcu_read_unlock();
2031
2032                 inode = d_inode(server->super->s_root);
2033                 status = nfs4_proc_fsid_present(inode, cred);
2034                 if (status != -NFS4ERR_MOVED)
2035                         goto restart;   /* wasn't this one */
2036                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2037                         goto restart;   /* there are more */
2038                 goto out;
2039         }
2040         rcu_read_unlock();
2041
2042 out:
2043         put_rpccred(cred);
2044         return 0;
2045 }
2046
2047 /**
2048  * nfs4_discover_server_trunking - Detect server IP address trunking
2049  *
2050  * @clp: nfs_client under test
2051  * @result: OUT: found nfs_client, or clp
2052  *
2053  * Returns zero or a negative errno.  If zero is returned,
2054  * an nfs_client pointer is planted in "result".
2055  *
2056  * Note: since we are invoked in process context, and
2057  * not from inside the state manager, we cannot use
2058  * nfs4_handle_reclaim_lease_error().
2059  */
2060 int nfs4_discover_server_trunking(struct nfs_client *clp,
2061                                   struct nfs_client **result)
2062 {
2063         const struct nfs4_state_recovery_ops *ops =
2064                                 clp->cl_mvops->reboot_recovery_ops;
2065         struct rpc_clnt *clnt;
2066         struct rpc_cred *cred;
2067         int i, status;
2068
2069         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2070
2071         clnt = clp->cl_rpcclient;
2072         i = 0;
2073
2074         mutex_lock(&nfs_clid_init_mutex);
2075 again:
2076         status  = -ENOENT;
2077         cred = nfs4_get_clid_cred(clp);
2078         if (cred == NULL)
2079                 goto out_unlock;
2080
2081         status = ops->detect_trunking(clp, result, cred);
2082         put_rpccred(cred);
2083         switch (status) {
2084         case 0:
2085                 break;
2086         case -ETIMEDOUT:
2087                 if (clnt->cl_softrtry)
2088                         break;
2089         case -NFS4ERR_DELAY:
2090         case -EAGAIN:
2091                 ssleep(1);
2092         case -NFS4ERR_STALE_CLIENTID:
2093                 dprintk("NFS: %s after status %d, retrying\n",
2094                         __func__, status);
2095                 goto again;
2096         case -EACCES:
2097                 if (i++ == 0) {
2098                         nfs4_root_machine_cred(clp);
2099                         goto again;
2100                 }
2101                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2102                         break;
2103         case -NFS4ERR_CLID_INUSE:
2104         case -NFS4ERR_WRONGSEC:
2105                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2106                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2107                         status = -EPERM;
2108                         break;
2109                 }
2110                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2111                 if (IS_ERR(clnt)) {
2112                         status = PTR_ERR(clnt);
2113                         break;
2114                 }
2115                 /* Note: this is safe because we haven't yet marked the
2116                  * client as ready, so we are the only user of
2117                  * clp->cl_rpcclient
2118                  */
2119                 clnt = xchg(&clp->cl_rpcclient, clnt);
2120                 rpc_shutdown_client(clnt);
2121                 clnt = clp->cl_rpcclient;
2122                 goto again;
2123
2124         case -NFS4ERR_MINOR_VERS_MISMATCH:
2125                 status = -EPROTONOSUPPORT;
2126                 break;
2127
2128         case -EKEYEXPIRED:
2129         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2130                                  * in nfs4_exchange_id */
2131                 status = -EKEYEXPIRED;
2132                 break;
2133         default:
2134                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2135                                 __func__, status);
2136                 status = -EIO;
2137         }
2138
2139 out_unlock:
2140         mutex_unlock(&nfs_clid_init_mutex);
2141         dprintk("NFS: %s: status = %d\n", __func__, status);
2142         return status;
2143 }
2144
2145 #ifdef CONFIG_NFS_V4_1
2146 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2147 {
2148         struct nfs_client *clp = session->clp;
2149
2150         switch (err) {
2151         default:
2152                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2153                 break;
2154         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2155                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2156         }
2157         nfs4_schedule_lease_recovery(clp);
2158 }
2159 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2160
2161 void nfs41_notify_server(struct nfs_client *clp)
2162 {
2163         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2164         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2165         nfs4_schedule_state_manager(clp);
2166 }
2167
2168 static void nfs4_reset_all_state(struct nfs_client *clp)
2169 {
2170         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2171                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2172                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2173                 nfs4_state_start_reclaim_nograce(clp);
2174                 dprintk("%s: scheduling reset of all state for server %s!\n",
2175                                 __func__, clp->cl_hostname);
2176                 nfs4_schedule_state_manager(clp);
2177         }
2178 }
2179
2180 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2181 {
2182         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2183                 nfs4_state_start_reclaim_reboot(clp);
2184                 dprintk("%s: server %s rebooted!\n", __func__,
2185                                 clp->cl_hostname);
2186                 nfs4_schedule_state_manager(clp);
2187         }
2188 }
2189
2190 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2191 {
2192         nfs4_reset_all_state(clp);
2193         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2194 }
2195
2196 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2197 {
2198         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
2199         nfs4_schedule_state_manager(clp);
2200
2201         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2202 }
2203
2204 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2205 {
2206         /* FIXME: For now, we destroy all layouts. */
2207         pnfs_destroy_all_layouts(clp);
2208         /* FIXME: For now, we test all delegations+open state+locks. */
2209         nfs41_handle_some_state_revoked(clp);
2210         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2211                         clp->cl_hostname);
2212 }
2213
2214 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2215 {
2216         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2217         nfs4_schedule_state_manager(clp);
2218
2219         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2220                         clp->cl_hostname);
2221 }
2222
2223 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2224 {
2225         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2226                 &clp->cl_state) == 0)
2227                 nfs4_schedule_state_manager(clp);
2228 }
2229
2230 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2231 {
2232         if (!flags)
2233                 return;
2234
2235         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2236                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2237
2238         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2239                 nfs41_handle_server_reboot(clp);
2240         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2241                 nfs41_handle_all_state_revoked(clp);
2242         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2243                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2244                 nfs41_handle_some_state_revoked(clp);
2245         if (flags & SEQ4_STATUS_LEASE_MOVED)
2246                 nfs4_schedule_lease_moved_recovery(clp);
2247         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2248                 nfs41_handle_recallable_state_revoked(clp);
2249         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2250                 nfs41_handle_backchannel_fault(clp);
2251         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2252                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2253                 nfs41_handle_cb_path_down(clp);
2254 }
2255
2256 static int nfs4_reset_session(struct nfs_client *clp)
2257 {
2258         struct rpc_cred *cred;
2259         int status;
2260
2261         if (!nfs4_has_session(clp))
2262                 return 0;
2263         nfs4_begin_drain_session(clp);
2264         cred = nfs4_get_clid_cred(clp);
2265         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2266         switch (status) {
2267         case 0:
2268         case -NFS4ERR_BADSESSION:
2269         case -NFS4ERR_DEADSESSION:
2270                 break;
2271         case -NFS4ERR_BACK_CHAN_BUSY:
2272         case -NFS4ERR_DELAY:
2273                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2274                 status = 0;
2275                 ssleep(1);
2276                 goto out;
2277         default:
2278                 status = nfs4_recovery_handle_error(clp, status);
2279                 goto out;
2280         }
2281
2282         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2283         status = nfs4_proc_create_session(clp, cred);
2284         if (status) {
2285                 dprintk("%s: session reset failed with status %d for server %s!\n",
2286                         __func__, status, clp->cl_hostname);
2287                 status = nfs4_handle_reclaim_lease_error(clp, status);
2288                 goto out;
2289         }
2290         nfs41_finish_session_reset(clp);
2291         dprintk("%s: session reset was successful for server %s!\n",
2292                         __func__, clp->cl_hostname);
2293 out:
2294         if (cred)
2295                 put_rpccred(cred);
2296         return status;
2297 }
2298
2299 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2300 {
2301         struct rpc_cred *cred;
2302         int ret;
2303
2304         if (!nfs4_has_session(clp))
2305                 return 0;
2306         nfs4_begin_drain_session(clp);
2307         cred = nfs4_get_clid_cred(clp);
2308         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2309         if (cred)
2310                 put_rpccred(cred);
2311         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2312         switch (ret) {
2313         case 0:
2314                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2315                         __func__, clp->cl_hostname);
2316                 break;
2317         case -NFS4ERR_DELAY:
2318                 ssleep(1);
2319                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2320                 break;
2321         default:
2322                 return nfs4_recovery_handle_error(clp, ret);
2323         }
2324         return 0;
2325 }
2326 #else /* CONFIG_NFS_V4_1 */
2327 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2328
2329 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2330 {
2331         return 0;
2332 }
2333 #endif /* CONFIG_NFS_V4_1 */
2334
2335 static void nfs4_state_manager(struct nfs_client *clp)
2336 {
2337         int status = 0;
2338         const char *section = "", *section_sep = "";
2339
2340         /* Ensure exclusive access to NFSv4 state */
2341         do {
2342                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2343                         section = "purge state";
2344                         status = nfs4_purge_lease(clp);
2345                         if (status < 0)
2346                                 goto out_error;
2347                         continue;
2348                 }
2349
2350                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2351                         section = "lease expired";
2352                         /* We're going to have to re-establish a clientid */
2353                         status = nfs4_reclaim_lease(clp);
2354                         if (status < 0)
2355                                 goto out_error;
2356                         continue;
2357                 }
2358
2359                 /* Initialize or reset the session */
2360                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2361                         section = "reset session";
2362                         status = nfs4_reset_session(clp);
2363                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2364                                 continue;
2365                         if (status < 0)
2366                                 goto out_error;
2367                 }
2368
2369                 /* Send BIND_CONN_TO_SESSION */
2370                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2371                                 &clp->cl_state)) {
2372                         section = "bind conn to session";
2373                         status = nfs4_bind_conn_to_session(clp);
2374                         if (status < 0)
2375                                 goto out_error;
2376                         continue;
2377                 }
2378
2379                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2380                         section = "check lease";
2381                         status = nfs4_check_lease(clp);
2382                         if (status < 0)
2383                                 goto out_error;
2384                         continue;
2385                 }
2386
2387                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2388                         section = "migration";
2389                         status = nfs4_handle_migration(clp);
2390                         if (status < 0)
2391                                 goto out_error;
2392                 }
2393
2394                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2395                         section = "lease moved";
2396                         status = nfs4_handle_lease_moved(clp);
2397                         if (status < 0)
2398                                 goto out_error;
2399                 }
2400
2401                 /* First recover reboot state... */
2402                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2403                         section = "reclaim reboot";
2404                         status = nfs4_do_reclaim(clp,
2405                                 clp->cl_mvops->reboot_recovery_ops);
2406                         if (status == -EAGAIN)
2407                                 continue;
2408                         if (status < 0)
2409                                 goto out_error;
2410                         nfs4_state_end_reclaim_reboot(clp);
2411                 }
2412
2413                 /* Now recover expired state... */
2414                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2415                         section = "reclaim nograce";
2416                         status = nfs4_do_reclaim(clp,
2417                                 clp->cl_mvops->nograce_recovery_ops);
2418                         if (status == -EAGAIN)
2419                                 continue;
2420                         if (status < 0)
2421                                 goto out_error;
2422                 }
2423
2424                 nfs4_end_drain_session(clp);
2425                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2426                         nfs_client_return_marked_delegations(clp);
2427                         continue;
2428                 }
2429
2430                 nfs4_clear_state_manager_bit(clp);
2431                 /* Did we race with an attempt to give us more work? */
2432                 if (clp->cl_state == 0)
2433                         break;
2434                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2435                         break;
2436         } while (atomic_read(&clp->cl_count) > 1);
2437         return;
2438 out_error:
2439         if (strlen(section))
2440                 section_sep = ": ";
2441         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2442                         " with error %d\n", section_sep, section,
2443                         clp->cl_hostname, -status);
2444         ssleep(1);
2445         nfs4_end_drain_session(clp);
2446         nfs4_clear_state_manager_bit(clp);
2447 }
2448
2449 static int nfs4_run_state_manager(void *ptr)
2450 {
2451         struct nfs_client *clp = ptr;
2452
2453         allow_signal(SIGKILL);
2454         nfs4_state_manager(clp);
2455         nfs_put_client(clp);
2456         module_put_and_exit(0);
2457         return 0;
2458 }
2459
2460 /*
2461  * Local variables:
2462  *  c-basic-offset: 8
2463  * End:
2464  */