mm/huge_memory.c: use head to emphasize the purpose of page
[linux-2.6-microblaze.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/kthread.h>
46 #include <linux/module.h>
47 #include <linux/random.h>
48 #include <linux/ratelimit.h>
49 #include <linux/workqueue.h>
50 #include <linux/bitops.h>
51 #include <linux/jiffies.h>
52
53 #include <linux/sunrpc/clnt.h>
54
55 #include "nfs4_fs.h"
56 #include "callback.h"
57 #include "delegation.h"
58 #include "internal.h"
59 #include "nfs4idmap.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
63 #include "nfs4trace.h"
64
65 #define NFSDBG_FACILITY         NFSDBG_STATE
66
67 #define OPENOWNER_POOL_SIZE     8
68
69 const nfs4_stateid zero_stateid = {
70         { .data = { 0 } },
71         .type = NFS4_SPECIAL_STATEID_TYPE,
72 };
73 const nfs4_stateid invalid_stateid = {
74         {
75                 /* Funky initialiser keeps older gcc versions happy */
76                 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
77         },
78         .type = NFS4_INVALID_STATEID_TYPE,
79 };
80
81 const nfs4_stateid current_stateid = {
82         {
83                 /* Funky initialiser keeps older gcc versions happy */
84                 .data = { 0x0, 0x0, 0x0, 0x1, 0 },
85         },
86         .type = NFS4_SPECIAL_STATEID_TYPE,
87 };
88
89 static DEFINE_MUTEX(nfs_clid_init_mutex);
90
91 static int nfs4_setup_state_renewal(struct nfs_client *clp)
92 {
93         int status;
94         struct nfs_fsinfo fsinfo;
95         unsigned long now;
96
97         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
98                 nfs4_schedule_state_renewal(clp);
99                 return 0;
100         }
101
102         now = jiffies;
103         status = nfs4_proc_get_lease_time(clp, &fsinfo);
104         if (status == 0) {
105                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
106                 nfs4_schedule_state_renewal(clp);
107         }
108
109         return status;
110 }
111
112 int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
113 {
114         struct nfs4_setclientid_res clid = {
115                 .clientid = clp->cl_clientid,
116                 .confirm = clp->cl_confirm,
117         };
118         unsigned short port;
119         int status;
120         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
121
122         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
123                 goto do_confirm;
124         port = nn->nfs_callback_tcpport;
125         if (clp->cl_addr.ss_family == AF_INET6)
126                 port = nn->nfs_callback_tcpport6;
127
128         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
129         if (status != 0)
130                 goto out;
131         clp->cl_clientid = clid.clientid;
132         clp->cl_confirm = clid.confirm;
133         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
134 do_confirm:
135         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
136         if (status != 0)
137                 goto out;
138         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
139         nfs4_setup_state_renewal(clp);
140 out:
141         return status;
142 }
143
144 /**
145  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
146  *
147  * @clp: nfs_client under test
148  * @result: OUT: found nfs_client, or clp
149  * @cred: credential to use for trunking test
150  *
151  * Returns zero, a negative errno, or a negative NFS4ERR status.
152  * If zero is returned, an nfs_client pointer is planted in
153  * "result".
154  *
155  * Note: The returned client may not yet be marked ready.
156  */
157 int nfs40_discover_server_trunking(struct nfs_client *clp,
158                                    struct nfs_client **result,
159                                    const struct cred *cred)
160 {
161         struct nfs4_setclientid_res clid = {
162                 .clientid = clp->cl_clientid,
163                 .confirm = clp->cl_confirm,
164         };
165         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
166         unsigned short port;
167         int status;
168
169         port = nn->nfs_callback_tcpport;
170         if (clp->cl_addr.ss_family == AF_INET6)
171                 port = nn->nfs_callback_tcpport6;
172
173         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
174         if (status != 0)
175                 goto out;
176         clp->cl_clientid = clid.clientid;
177         clp->cl_confirm = clid.confirm;
178
179         status = nfs40_walk_client_list(clp, result, cred);
180         if (status == 0) {
181                 /* Sustain the lease, even if it's empty.  If the clientid4
182                  * goes stale it's of no use for trunking discovery. */
183                 nfs4_schedule_state_renewal(*result);
184
185                 /* If the client state need to recover, do it. */
186                 if (clp->cl_state)
187                         nfs4_schedule_state_manager(clp);
188         }
189 out:
190         return status;
191 }
192
193 const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
194 {
195         return get_cred(rpc_machine_cred());
196 }
197
198 static void nfs4_root_machine_cred(struct nfs_client *clp)
199 {
200
201         /* Force root creds instead of machine */
202         clp->cl_principal = NULL;
203         clp->cl_rpcclient->cl_principal = NULL;
204 }
205
206 static const struct cred *
207 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
208 {
209         const struct cred *cred = NULL;
210         struct nfs4_state_owner *sp;
211         struct rb_node *pos;
212
213         for (pos = rb_first(&server->state_owners);
214              pos != NULL;
215              pos = rb_next(pos)) {
216                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
217                 if (list_empty(&sp->so_states))
218                         continue;
219                 cred = get_cred(sp->so_cred);
220                 break;
221         }
222         return cred;
223 }
224
225 /**
226  * nfs4_get_renew_cred - Acquire credential for a renew operation
227  * @clp: client state handle
228  *
229  * Returns an rpc_cred with reference count bumped, or NULL.
230  * Caller must hold clp->cl_lock.
231  */
232 const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
233 {
234         const struct cred *cred = NULL;
235         struct nfs_server *server;
236
237         /* Use machine credentials if available */
238         cred = nfs4_get_machine_cred(clp);
239         if (cred != NULL)
240                 goto out;
241
242         spin_lock(&clp->cl_lock);
243         rcu_read_lock();
244         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
245                 cred = nfs4_get_renew_cred_server_locked(server);
246                 if (cred != NULL)
247                         break;
248         }
249         rcu_read_unlock();
250         spin_unlock(&clp->cl_lock);
251
252 out:
253         return cred;
254 }
255
256 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
257 {
258         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
259                 spin_lock(&tbl->slot_tbl_lock);
260                 nfs41_wake_slot_table(tbl);
261                 spin_unlock(&tbl->slot_tbl_lock);
262         }
263 }
264
265 static void nfs4_end_drain_session(struct nfs_client *clp)
266 {
267         struct nfs4_session *ses = clp->cl_session;
268
269         if (clp->cl_slot_tbl) {
270                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
271                 return;
272         }
273
274         if (ses != NULL) {
275                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
276                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
277         }
278 }
279
280 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
281 {
282         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
283         spin_lock(&tbl->slot_tbl_lock);
284         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
285                 reinit_completion(&tbl->complete);
286                 spin_unlock(&tbl->slot_tbl_lock);
287                 return wait_for_completion_interruptible(&tbl->complete);
288         }
289         spin_unlock(&tbl->slot_tbl_lock);
290         return 0;
291 }
292
293 static int nfs4_begin_drain_session(struct nfs_client *clp)
294 {
295         struct nfs4_session *ses = clp->cl_session;
296         int ret;
297
298         if (clp->cl_slot_tbl)
299                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
300
301         /* back channel */
302         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
303         if (ret)
304                 return ret;
305         /* fore channel */
306         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
307 }
308
309 #if defined(CONFIG_NFS_V4_1)
310
311 static void nfs41_finish_session_reset(struct nfs_client *clp)
312 {
313         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
314         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
315         /* create_session negotiated new slot table */
316         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
317         nfs4_setup_state_renewal(clp);
318 }
319
320 int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
321 {
322         int status;
323
324         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
325                 goto do_confirm;
326         status = nfs4_proc_exchange_id(clp, cred);
327         if (status != 0)
328                 goto out;
329         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
330 do_confirm:
331         status = nfs4_proc_create_session(clp, cred);
332         if (status != 0)
333                 goto out;
334         nfs41_finish_session_reset(clp);
335         nfs_mark_client_ready(clp, NFS_CS_READY);
336 out:
337         return status;
338 }
339
340 /**
341  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
342  *
343  * @clp: nfs_client under test
344  * @result: OUT: found nfs_client, or clp
345  * @cred: credential to use for trunking test
346  *
347  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
348  * If NFS4_OK is returned, an nfs_client pointer is planted in
349  * "result".
350  *
351  * Note: The returned client may not yet be marked ready.
352  */
353 int nfs41_discover_server_trunking(struct nfs_client *clp,
354                                    struct nfs_client **result,
355                                    const struct cred *cred)
356 {
357         int status;
358
359         status = nfs4_proc_exchange_id(clp, cred);
360         if (status != NFS4_OK)
361                 return status;
362
363         status = nfs41_walk_client_list(clp, result, cred);
364         if (status < 0)
365                 return status;
366         if (clp != *result)
367                 return 0;
368
369         /*
370          * Purge state if the client id was established in a prior
371          * instance and the client id could not have arrived on the
372          * server via Transparent State Migration.
373          */
374         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
375                 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
376                         set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
377                 else
378                         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
379         }
380         nfs4_schedule_state_manager(clp);
381         status = nfs_wait_client_init_complete(clp);
382         if (status < 0)
383                 nfs_put_client(clp);
384         return status;
385 }
386
387 #endif /* CONFIG_NFS_V4_1 */
388
389 /**
390  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
391  * @clp: client state handle
392  *
393  * Returns a cred with reference count bumped, or NULL.
394  */
395 const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
396 {
397         const struct cred *cred;
398
399         cred = nfs4_get_machine_cred(clp);
400         return cred;
401 }
402
403 static struct nfs4_state_owner *
404 nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
405 {
406         struct rb_node **p = &server->state_owners.rb_node,
407                        *parent = NULL;
408         struct nfs4_state_owner *sp;
409         int cmp;
410
411         while (*p != NULL) {
412                 parent = *p;
413                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
414                 cmp = cred_fscmp(cred, sp->so_cred);
415
416                 if (cmp < 0)
417                         p = &parent->rb_left;
418                 else if (cmp > 0)
419                         p = &parent->rb_right;
420                 else {
421                         if (!list_empty(&sp->so_lru))
422                                 list_del_init(&sp->so_lru);
423                         atomic_inc(&sp->so_count);
424                         return sp;
425                 }
426         }
427         return NULL;
428 }
429
430 static struct nfs4_state_owner *
431 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
432 {
433         struct nfs_server *server = new->so_server;
434         struct rb_node **p = &server->state_owners.rb_node,
435                        *parent = NULL;
436         struct nfs4_state_owner *sp;
437         int cmp;
438
439         while (*p != NULL) {
440                 parent = *p;
441                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
442                 cmp = cred_fscmp(new->so_cred, sp->so_cred);
443
444                 if (cmp < 0)
445                         p = &parent->rb_left;
446                 else if (cmp > 0)
447                         p = &parent->rb_right;
448                 else {
449                         if (!list_empty(&sp->so_lru))
450                                 list_del_init(&sp->so_lru);
451                         atomic_inc(&sp->so_count);
452                         return sp;
453                 }
454         }
455         rb_link_node(&new->so_server_node, parent, p);
456         rb_insert_color(&new->so_server_node, &server->state_owners);
457         return new;
458 }
459
460 static void
461 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
462 {
463         struct nfs_server *server = sp->so_server;
464
465         if (!RB_EMPTY_NODE(&sp->so_server_node))
466                 rb_erase(&sp->so_server_node, &server->state_owners);
467 }
468
469 static void
470 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
471 {
472         sc->create_time = ktime_get();
473         sc->flags = 0;
474         sc->counter = 0;
475         spin_lock_init(&sc->lock);
476         INIT_LIST_HEAD(&sc->list);
477         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
478 }
479
480 static void
481 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
482 {
483         rpc_destroy_wait_queue(&sc->wait);
484 }
485
486 /*
487  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
488  * create a new state_owner.
489  *
490  */
491 static struct nfs4_state_owner *
492 nfs4_alloc_state_owner(struct nfs_server *server,
493                 const struct cred *cred,
494                 gfp_t gfp_flags)
495 {
496         struct nfs4_state_owner *sp;
497
498         sp = kzalloc(sizeof(*sp), gfp_flags);
499         if (!sp)
500                 return NULL;
501         sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
502                                                 gfp_flags);
503         if (sp->so_seqid.owner_id < 0) {
504                 kfree(sp);
505                 return NULL;
506         }
507         sp->so_server = server;
508         sp->so_cred = get_cred(cred);
509         spin_lock_init(&sp->so_lock);
510         INIT_LIST_HEAD(&sp->so_states);
511         nfs4_init_seqid_counter(&sp->so_seqid);
512         atomic_set(&sp->so_count, 1);
513         INIT_LIST_HEAD(&sp->so_lru);
514         seqcount_init(&sp->so_reclaim_seqcount);
515         mutex_init(&sp->so_delegreturn_mutex);
516         return sp;
517 }
518
519 static void
520 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
521 {
522         /* This state_owner is no longer usable, but must
523          * remain in place so that state recovery can find it
524          * and the opens associated with it.
525          * It may also be used for new 'open' request to
526          * return a delegation to the server.
527          * So update the 'create_time' so that it looks like
528          * a new state_owner.  This will cause the server to
529          * request an OPEN_CONFIRM to start a new sequence.
530          */
531         sp->so_seqid.create_time = ktime_get();
532 }
533
534 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
535 {
536         nfs4_destroy_seqid_counter(&sp->so_seqid);
537         put_cred(sp->so_cred);
538         ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
539         kfree(sp);
540 }
541
542 static void nfs4_gc_state_owners(struct nfs_server *server)
543 {
544         struct nfs_client *clp = server->nfs_client;
545         struct nfs4_state_owner *sp, *tmp;
546         unsigned long time_min, time_max;
547         LIST_HEAD(doomed);
548
549         spin_lock(&clp->cl_lock);
550         time_max = jiffies;
551         time_min = (long)time_max - (long)clp->cl_lease_time;
552         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
553                 /* NB: LRU is sorted so that oldest is at the head */
554                 if (time_in_range(sp->so_expires, time_min, time_max))
555                         break;
556                 list_move(&sp->so_lru, &doomed);
557                 nfs4_remove_state_owner_locked(sp);
558         }
559         spin_unlock(&clp->cl_lock);
560
561         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
562                 list_del(&sp->so_lru);
563                 nfs4_free_state_owner(sp);
564         }
565 }
566
567 /**
568  * nfs4_get_state_owner - Look up a state owner given a credential
569  * @server: nfs_server to search
570  * @cred: RPC credential to match
571  * @gfp_flags: allocation mode
572  *
573  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
574  */
575 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
576                                               const struct cred *cred,
577                                               gfp_t gfp_flags)
578 {
579         struct nfs_client *clp = server->nfs_client;
580         struct nfs4_state_owner *sp, *new;
581
582         spin_lock(&clp->cl_lock);
583         sp = nfs4_find_state_owner_locked(server, cred);
584         spin_unlock(&clp->cl_lock);
585         if (sp != NULL)
586                 goto out;
587         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
588         if (new == NULL)
589                 goto out;
590         spin_lock(&clp->cl_lock);
591         sp = nfs4_insert_state_owner_locked(new);
592         spin_unlock(&clp->cl_lock);
593         if (sp != new)
594                 nfs4_free_state_owner(new);
595 out:
596         nfs4_gc_state_owners(server);
597         return sp;
598 }
599
600 /**
601  * nfs4_put_state_owner - Release a nfs4_state_owner
602  * @sp: state owner data to release
603  *
604  * Note that we keep released state owners on an LRU
605  * list.
606  * This caches valid state owners so that they can be
607  * reused, to avoid the OPEN_CONFIRM on minor version 0.
608  * It also pins the uniquifier of dropped state owners for
609  * a while, to ensure that those state owner names are
610  * never reused.
611  */
612 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
613 {
614         struct nfs_server *server = sp->so_server;
615         struct nfs_client *clp = server->nfs_client;
616
617         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
618                 return;
619
620         sp->so_expires = jiffies;
621         list_add_tail(&sp->so_lru, &server->state_owners_lru);
622         spin_unlock(&clp->cl_lock);
623 }
624
625 /**
626  * nfs4_purge_state_owners - Release all cached state owners
627  * @server: nfs_server with cached state owners to release
628  * @head: resulting list of state owners
629  *
630  * Called at umount time.  Remaining state owners will be on
631  * the LRU with ref count of zero.
632  * Note that the state owners are not freed, but are added
633  * to the list @head, which can later be used as an argument
634  * to nfs4_free_state_owners.
635  */
636 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
637 {
638         struct nfs_client *clp = server->nfs_client;
639         struct nfs4_state_owner *sp, *tmp;
640
641         spin_lock(&clp->cl_lock);
642         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
643                 list_move(&sp->so_lru, head);
644                 nfs4_remove_state_owner_locked(sp);
645         }
646         spin_unlock(&clp->cl_lock);
647 }
648
649 /**
650  * nfs4_purge_state_owners - Release all cached state owners
651  * @head: resulting list of state owners
652  *
653  * Frees a list of state owners that was generated by
654  * nfs4_purge_state_owners
655  */
656 void nfs4_free_state_owners(struct list_head *head)
657 {
658         struct nfs4_state_owner *sp, *tmp;
659
660         list_for_each_entry_safe(sp, tmp, head, so_lru) {
661                 list_del(&sp->so_lru);
662                 nfs4_free_state_owner(sp);
663         }
664 }
665
666 static struct nfs4_state *
667 nfs4_alloc_open_state(void)
668 {
669         struct nfs4_state *state;
670
671         state = kzalloc(sizeof(*state), GFP_NOFS);
672         if (!state)
673                 return NULL;
674         refcount_set(&state->count, 1);
675         INIT_LIST_HEAD(&state->lock_states);
676         spin_lock_init(&state->state_lock);
677         seqlock_init(&state->seqlock);
678         init_waitqueue_head(&state->waitq);
679         return state;
680 }
681
682 void
683 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
684 {
685         if (state->state == fmode)
686                 return;
687         /* NB! List reordering - see the reclaim code for why.  */
688         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
689                 if (fmode & FMODE_WRITE)
690                         list_move(&state->open_states, &state->owner->so_states);
691                 else
692                         list_move_tail(&state->open_states, &state->owner->so_states);
693         }
694         state->state = fmode;
695 }
696
697 static struct nfs4_state *
698 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
699 {
700         struct nfs_inode *nfsi = NFS_I(inode);
701         struct nfs4_state *state;
702
703         list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
704                 if (state->owner != owner)
705                         continue;
706                 if (!nfs4_valid_open_stateid(state))
707                         continue;
708                 if (refcount_inc_not_zero(&state->count))
709                         return state;
710         }
711         return NULL;
712 }
713
714 static void
715 nfs4_free_open_state(struct nfs4_state *state)
716 {
717         kfree_rcu(state, rcu_head);
718 }
719
720 struct nfs4_state *
721 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
722 {
723         struct nfs4_state *state, *new;
724         struct nfs_inode *nfsi = NFS_I(inode);
725
726         rcu_read_lock();
727         state = __nfs4_find_state_byowner(inode, owner);
728         rcu_read_unlock();
729         if (state)
730                 goto out;
731         new = nfs4_alloc_open_state();
732         spin_lock(&owner->so_lock);
733         spin_lock(&inode->i_lock);
734         state = __nfs4_find_state_byowner(inode, owner);
735         if (state == NULL && new != NULL) {
736                 state = new;
737                 state->owner = owner;
738                 atomic_inc(&owner->so_count);
739                 list_add_rcu(&state->inode_states, &nfsi->open_states);
740                 ihold(inode);
741                 state->inode = inode;
742                 spin_unlock(&inode->i_lock);
743                 /* Note: The reclaim code dictates that we add stateless
744                  * and read-only stateids to the end of the list */
745                 list_add_tail(&state->open_states, &owner->so_states);
746                 spin_unlock(&owner->so_lock);
747         } else {
748                 spin_unlock(&inode->i_lock);
749                 spin_unlock(&owner->so_lock);
750                 if (new)
751                         nfs4_free_open_state(new);
752         }
753 out:
754         return state;
755 }
756
757 void nfs4_put_open_state(struct nfs4_state *state)
758 {
759         struct inode *inode = state->inode;
760         struct nfs4_state_owner *owner = state->owner;
761
762         if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
763                 return;
764         spin_lock(&inode->i_lock);
765         list_del_rcu(&state->inode_states);
766         list_del(&state->open_states);
767         spin_unlock(&inode->i_lock);
768         spin_unlock(&owner->so_lock);
769         iput(inode);
770         nfs4_free_open_state(state);
771         nfs4_put_state_owner(owner);
772 }
773
774 /*
775  * Close the current file.
776  */
777 static void __nfs4_close(struct nfs4_state *state,
778                 fmode_t fmode, gfp_t gfp_mask, int wait)
779 {
780         struct nfs4_state_owner *owner = state->owner;
781         int call_close = 0;
782         fmode_t newstate;
783
784         atomic_inc(&owner->so_count);
785         /* Protect against nfs4_find_state() */
786         spin_lock(&owner->so_lock);
787         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
788                 case FMODE_READ:
789                         state->n_rdonly--;
790                         break;
791                 case FMODE_WRITE:
792                         state->n_wronly--;
793                         break;
794                 case FMODE_READ|FMODE_WRITE:
795                         state->n_rdwr--;
796         }
797         newstate = FMODE_READ|FMODE_WRITE;
798         if (state->n_rdwr == 0) {
799                 if (state->n_rdonly == 0) {
800                         newstate &= ~FMODE_READ;
801                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
802                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
803                 }
804                 if (state->n_wronly == 0) {
805                         newstate &= ~FMODE_WRITE;
806                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
807                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
808                 }
809                 if (newstate == 0)
810                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
811         }
812         nfs4_state_set_mode_locked(state, newstate);
813         spin_unlock(&owner->so_lock);
814
815         if (!call_close) {
816                 nfs4_put_open_state(state);
817                 nfs4_put_state_owner(owner);
818         } else
819                 nfs4_do_close(state, gfp_mask, wait);
820 }
821
822 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
823 {
824         __nfs4_close(state, fmode, GFP_NOFS, 0);
825 }
826
827 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
828 {
829         __nfs4_close(state, fmode, GFP_KERNEL, 1);
830 }
831
832 /*
833  * Search the state->lock_states for an existing lock_owner
834  * that is compatible with either of the given owners.
835  * If the second is non-zero, then the first refers to a Posix-lock
836  * owner (current->files) and the second refers to a flock/OFD
837  * owner (struct file*).  In that case, prefer a match for the first
838  * owner.
839  * If both sorts of locks are held on the one file we cannot know
840  * which stateid was intended to be used, so a "correct" choice cannot
841  * be made.  Failing that, a "consistent" choice is preferable.  The
842  * consistent choice we make is to prefer the first owner, that of a
843  * Posix lock.
844  */
845 static struct nfs4_lock_state *
846 __nfs4_find_lock_state(struct nfs4_state *state,
847                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
848 {
849         struct nfs4_lock_state *pos, *ret = NULL;
850         list_for_each_entry(pos, &state->lock_states, ls_locks) {
851                 if (pos->ls_owner == fl_owner) {
852                         ret = pos;
853                         break;
854                 }
855                 if (pos->ls_owner == fl_owner2)
856                         ret = pos;
857         }
858         if (ret)
859                 refcount_inc(&ret->ls_count);
860         return ret;
861 }
862
863 /*
864  * Return a compatible lock_state. If no initialized lock_state structure
865  * exists, return an uninitialized one.
866  *
867  */
868 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
869 {
870         struct nfs4_lock_state *lsp;
871         struct nfs_server *server = state->owner->so_server;
872
873         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
874         if (lsp == NULL)
875                 return NULL;
876         nfs4_init_seqid_counter(&lsp->ls_seqid);
877         refcount_set(&lsp->ls_count, 1);
878         lsp->ls_state = state;
879         lsp->ls_owner = fl_owner;
880         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
881         if (lsp->ls_seqid.owner_id < 0)
882                 goto out_free;
883         INIT_LIST_HEAD(&lsp->ls_locks);
884         return lsp;
885 out_free:
886         kfree(lsp);
887         return NULL;
888 }
889
890 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
891 {
892         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
893         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
894         kfree(lsp);
895 }
896
897 /*
898  * Return a compatible lock_state. If no initialized lock_state structure
899  * exists, return an uninitialized one.
900  *
901  */
902 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
903 {
904         struct nfs4_lock_state *lsp, *new = NULL;
905         
906         for(;;) {
907                 spin_lock(&state->state_lock);
908                 lsp = __nfs4_find_lock_state(state, owner, NULL);
909                 if (lsp != NULL)
910                         break;
911                 if (new != NULL) {
912                         list_add(&new->ls_locks, &state->lock_states);
913                         set_bit(LK_STATE_IN_USE, &state->flags);
914                         lsp = new;
915                         new = NULL;
916                         break;
917                 }
918                 spin_unlock(&state->state_lock);
919                 new = nfs4_alloc_lock_state(state, owner);
920                 if (new == NULL)
921                         return NULL;
922         }
923         spin_unlock(&state->state_lock);
924         if (new != NULL)
925                 nfs4_free_lock_state(state->owner->so_server, new);
926         return lsp;
927 }
928
929 /*
930  * Release reference to lock_state, and free it if we see that
931  * it is no longer in use
932  */
933 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
934 {
935         struct nfs_server *server;
936         struct nfs4_state *state;
937
938         if (lsp == NULL)
939                 return;
940         state = lsp->ls_state;
941         if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
942                 return;
943         list_del(&lsp->ls_locks);
944         if (list_empty(&state->lock_states))
945                 clear_bit(LK_STATE_IN_USE, &state->flags);
946         spin_unlock(&state->state_lock);
947         server = state->owner->so_server;
948         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
949                 struct nfs_client *clp = server->nfs_client;
950
951                 clp->cl_mvops->free_lock_state(server, lsp);
952         } else
953                 nfs4_free_lock_state(server, lsp);
954 }
955
956 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
957 {
958         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
959
960         dst->fl_u.nfs4_fl.owner = lsp;
961         refcount_inc(&lsp->ls_count);
962 }
963
964 static void nfs4_fl_release_lock(struct file_lock *fl)
965 {
966         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
967 }
968
969 static const struct file_lock_operations nfs4_fl_lock_ops = {
970         .fl_copy_lock = nfs4_fl_copy_lock,
971         .fl_release_private = nfs4_fl_release_lock,
972 };
973
974 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
975 {
976         struct nfs4_lock_state *lsp;
977
978         if (fl->fl_ops != NULL)
979                 return 0;
980         lsp = nfs4_get_lock_state(state, fl->fl_owner);
981         if (lsp == NULL)
982                 return -ENOMEM;
983         fl->fl_u.nfs4_fl.owner = lsp;
984         fl->fl_ops = &nfs4_fl_lock_ops;
985         return 0;
986 }
987
988 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
989                 struct nfs4_state *state,
990                 const struct nfs_lock_context *l_ctx)
991 {
992         struct nfs4_lock_state *lsp;
993         fl_owner_t fl_owner, fl_flock_owner;
994         int ret = -ENOENT;
995
996         if (l_ctx == NULL)
997                 goto out;
998
999         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1000                 goto out;
1001
1002         fl_owner = l_ctx->lockowner;
1003         fl_flock_owner = l_ctx->open_context->flock_owner;
1004
1005         spin_lock(&state->state_lock);
1006         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
1007         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1008                 ret = -EIO;
1009         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1010                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1011                 ret = 0;
1012         }
1013         spin_unlock(&state->state_lock);
1014         nfs4_put_lock_state(lsp);
1015 out:
1016         return ret;
1017 }
1018
1019 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1020 {
1021         bool ret;
1022         const nfs4_stateid *src;
1023         int seq;
1024
1025         do {
1026                 ret = false;
1027                 src = &zero_stateid;
1028                 seq = read_seqbegin(&state->seqlock);
1029                 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1030                         src = &state->open_stateid;
1031                         ret = true;
1032                 }
1033                 nfs4_stateid_copy(dst, src);
1034         } while (read_seqretry(&state->seqlock, seq));
1035         return ret;
1036 }
1037
1038 /*
1039  * Byte-range lock aware utility to initialize the stateid of read/write
1040  * requests.
1041  */
1042 int nfs4_select_rw_stateid(struct nfs4_state *state,
1043                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1044                 nfs4_stateid *dst, const struct cred **cred)
1045 {
1046         int ret;
1047
1048         if (!nfs4_valid_open_stateid(state))
1049                 return -EIO;
1050         if (cred != NULL)
1051                 *cred = NULL;
1052         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1053         if (ret == -EIO)
1054                 /* A lost lock - don't even consider delegations */
1055                 goto out;
1056         /* returns true if delegation stateid found and copied */
1057         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1058                 ret = 0;
1059                 goto out;
1060         }
1061         if (ret != -ENOENT)
1062                 /* nfs4_copy_delegation_stateid() didn't over-write
1063                  * dst, so it still has the lock stateid which we now
1064                  * choose to use.
1065                  */
1066                 goto out;
1067         ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1068 out:
1069         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1070                 dst->seqid = 0;
1071         return ret;
1072 }
1073
1074 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1075 {
1076         struct nfs_seqid *new;
1077
1078         new = kmalloc(sizeof(*new), gfp_mask);
1079         if (new == NULL)
1080                 return ERR_PTR(-ENOMEM);
1081         new->sequence = counter;
1082         INIT_LIST_HEAD(&new->list);
1083         new->task = NULL;
1084         return new;
1085 }
1086
1087 void nfs_release_seqid(struct nfs_seqid *seqid)
1088 {
1089         struct nfs_seqid_counter *sequence;
1090
1091         if (seqid == NULL || list_empty(&seqid->list))
1092                 return;
1093         sequence = seqid->sequence;
1094         spin_lock(&sequence->lock);
1095         list_del_init(&seqid->list);
1096         if (!list_empty(&sequence->list)) {
1097                 struct nfs_seqid *next;
1098
1099                 next = list_first_entry(&sequence->list,
1100                                 struct nfs_seqid, list);
1101                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1102         }
1103         spin_unlock(&sequence->lock);
1104 }
1105
1106 void nfs_free_seqid(struct nfs_seqid *seqid)
1107 {
1108         nfs_release_seqid(seqid);
1109         kfree(seqid);
1110 }
1111
1112 /*
1113  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1114  * failed with a seqid incrementing error -
1115  * see comments nfs4.h:seqid_mutating_error()
1116  */
1117 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1118 {
1119         switch (status) {
1120                 case 0:
1121                         break;
1122                 case -NFS4ERR_BAD_SEQID:
1123                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1124                                 return;
1125                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1126                                         " sequence-id error on an"
1127                                         " unconfirmed sequence %p!\n",
1128                                         seqid->sequence);
1129                 case -NFS4ERR_STALE_CLIENTID:
1130                 case -NFS4ERR_STALE_STATEID:
1131                 case -NFS4ERR_BAD_STATEID:
1132                 case -NFS4ERR_BADXDR:
1133                 case -NFS4ERR_RESOURCE:
1134                 case -NFS4ERR_NOFILEHANDLE:
1135                 case -NFS4ERR_MOVED:
1136                         /* Non-seqid mutating errors */
1137                         return;
1138         };
1139         /*
1140          * Note: no locking needed as we are guaranteed to be first
1141          * on the sequence list
1142          */
1143         seqid->sequence->counter++;
1144 }
1145
1146 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1147 {
1148         struct nfs4_state_owner *sp;
1149
1150         if (seqid == NULL)
1151                 return;
1152
1153         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1154         if (status == -NFS4ERR_BAD_SEQID)
1155                 nfs4_reset_state_owner(sp);
1156         if (!nfs4_has_session(sp->so_server->nfs_client))
1157                 nfs_increment_seqid(status, seqid);
1158 }
1159
1160 /*
1161  * Increment the seqid if the LOCK/LOCKU succeeded, or
1162  * failed with a seqid incrementing error -
1163  * see comments nfs4.h:seqid_mutating_error()
1164  */
1165 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1166 {
1167         if (seqid != NULL)
1168                 nfs_increment_seqid(status, seqid);
1169 }
1170
1171 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1172 {
1173         struct nfs_seqid_counter *sequence;
1174         int status = 0;
1175
1176         if (seqid == NULL)
1177                 goto out;
1178         sequence = seqid->sequence;
1179         spin_lock(&sequence->lock);
1180         seqid->task = task;
1181         if (list_empty(&seqid->list))
1182                 list_add_tail(&seqid->list, &sequence->list);
1183         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1184                 goto unlock;
1185         rpc_sleep_on(&sequence->wait, task, NULL);
1186         status = -EAGAIN;
1187 unlock:
1188         spin_unlock(&sequence->lock);
1189 out:
1190         return status;
1191 }
1192
1193 static int nfs4_run_state_manager(void *);
1194
1195 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1196 {
1197         smp_mb__before_atomic();
1198         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1199         smp_mb__after_atomic();
1200         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1201         rpc_wake_up(&clp->cl_rpcwaitq);
1202 }
1203
1204 /*
1205  * Schedule the nfs_client asynchronous state management routine
1206  */
1207 void nfs4_schedule_state_manager(struct nfs_client *clp)
1208 {
1209         struct task_struct *task;
1210         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1211
1212         set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1213         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1214                 return;
1215         __module_get(THIS_MODULE);
1216         refcount_inc(&clp->cl_count);
1217
1218         /* The rcu_read_lock() is not strictly necessary, as the state
1219          * manager is the only thread that ever changes the rpc_xprt
1220          * after it's initialized.  At this point, we're single threaded. */
1221         rcu_read_lock();
1222         snprintf(buf, sizeof(buf), "%s-manager",
1223                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1224         rcu_read_unlock();
1225         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1226         if (IS_ERR(task)) {
1227                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1228                         __func__, PTR_ERR(task));
1229                 nfs4_clear_state_manager_bit(clp);
1230                 nfs_put_client(clp);
1231                 module_put(THIS_MODULE);
1232         }
1233 }
1234
1235 /*
1236  * Schedule a lease recovery attempt
1237  */
1238 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1239 {
1240         if (!clp)
1241                 return;
1242         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1243                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1244         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1245                         clp->cl_hostname);
1246         nfs4_schedule_state_manager(clp);
1247 }
1248 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1249
1250 /**
1251  * nfs4_schedule_migration_recovery - trigger migration recovery
1252  *
1253  * @server: FSID that is migrating
1254  *
1255  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1256  * value is returned.
1257  */
1258 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1259 {
1260         struct nfs_client *clp = server->nfs_client;
1261
1262         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1263                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1264                                 clp->cl_hostname);
1265                 return -NFS4ERR_IO;
1266         }
1267
1268         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1269                 return -NFS4ERR_IO;
1270
1271         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1272                         __func__,
1273                         (unsigned long long)server->fsid.major,
1274                         (unsigned long long)server->fsid.minor,
1275                         clp->cl_hostname);
1276
1277         set_bit(NFS_MIG_IN_TRANSITION,
1278                         &((struct nfs_server *)server)->mig_status);
1279         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1280
1281         nfs4_schedule_state_manager(clp);
1282         return 0;
1283 }
1284 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1285
1286 /**
1287  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1288  *
1289  * @clp: server to check for moved leases
1290  *
1291  */
1292 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1293 {
1294         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1295                 __func__, clp->cl_clientid, clp->cl_hostname);
1296
1297         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1298         nfs4_schedule_state_manager(clp);
1299 }
1300 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1301
1302 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1303 {
1304         int res;
1305
1306         might_sleep();
1307
1308         refcount_inc(&clp->cl_count);
1309         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1310                                  nfs_wait_bit_killable, TASK_KILLABLE);
1311         if (res)
1312                 goto out;
1313         if (clp->cl_cons_state < 0)
1314                 res = clp->cl_cons_state;
1315 out:
1316         nfs_put_client(clp);
1317         return res;
1318 }
1319
1320 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1321 {
1322         unsigned int loop;
1323         int ret;
1324
1325         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1326                 ret = nfs4_wait_clnt_recover(clp);
1327                 if (ret != 0)
1328                         break;
1329                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1330                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1331                         break;
1332                 nfs4_schedule_state_manager(clp);
1333                 ret = -EIO;
1334         }
1335         return ret;
1336 }
1337
1338 /*
1339  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1340  * @clp: client to process
1341  *
1342  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1343  * resend of the SETCLIENTID and hence re-establish the
1344  * callback channel. Then return all existing delegations.
1345  */
1346 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1347 {
1348         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1349         nfs_expire_all_delegations(clp);
1350         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1351                         clp->cl_hostname);
1352 }
1353
1354 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1355 {
1356         nfs40_handle_cb_pathdown(clp);
1357         nfs4_schedule_state_manager(clp);
1358 }
1359
1360 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1361 {
1362
1363         if (!nfs4_valid_open_stateid(state))
1364                 return 0;
1365         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1366         /* Don't recover state that expired before the reboot */
1367         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1368                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1369                 return 0;
1370         }
1371         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1372         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1373         return 1;
1374 }
1375
1376 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1377 {
1378         if (!nfs4_valid_open_stateid(state))
1379                 return 0;
1380         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1381         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1382         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1383         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1384         return 1;
1385 }
1386
1387 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1388 {
1389         struct nfs_client *clp = server->nfs_client;
1390
1391         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1392                 return -EBADF;
1393         nfs_inode_find_delegation_state_and_recover(state->inode,
1394                         &state->stateid);
1395         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1396                         clp->cl_hostname);
1397         nfs4_schedule_state_manager(clp);
1398         return 0;
1399 }
1400 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1401
1402 static struct nfs4_lock_state *
1403 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1404                 const nfs4_stateid *stateid)
1405 {
1406         struct nfs4_lock_state *pos;
1407
1408         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1409                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1410                         continue;
1411                 if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1412                         return pos;
1413         }
1414         return NULL;
1415 }
1416
1417 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1418                 const nfs4_stateid *stateid)
1419 {
1420         bool found = false;
1421
1422         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1423                 spin_lock(&state->state_lock);
1424                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1425                         found = true;
1426                 spin_unlock(&state->state_lock);
1427         }
1428         return found;
1429 }
1430
1431 void nfs_inode_find_state_and_recover(struct inode *inode,
1432                 const nfs4_stateid *stateid)
1433 {
1434         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1435         struct nfs_inode *nfsi = NFS_I(inode);
1436         struct nfs_open_context *ctx;
1437         struct nfs4_state *state;
1438         bool found = false;
1439
1440         rcu_read_lock();
1441         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1442                 state = ctx->state;
1443                 if (state == NULL)
1444                         continue;
1445                 if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1446                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1447                         found = true;
1448                         continue;
1449                 }
1450                 if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1451                     nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1452                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1453                         found = true;
1454                         continue;
1455                 }
1456                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1457                     nfs4_state_mark_reclaim_nograce(clp, state))
1458                         found = true;
1459         }
1460         rcu_read_unlock();
1461
1462         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1463         if (found)
1464                 nfs4_schedule_state_manager(clp);
1465 }
1466
1467 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1468 {
1469         struct inode *inode = state->inode;
1470         struct nfs_inode *nfsi = NFS_I(inode);
1471         struct nfs_open_context *ctx;
1472
1473         rcu_read_lock();
1474         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1475                 if (ctx->state != state)
1476                         continue;
1477                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1478                 pr_warn("NFSv4: state recovery failed for open file %pd2, "
1479                                 "error = %d\n", ctx->dentry, err);
1480         }
1481         rcu_read_unlock();
1482 }
1483
1484 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1485 {
1486         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1487         nfs4_state_mark_open_context_bad(state, error);
1488 }
1489
1490
1491 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1492 {
1493         struct inode *inode = state->inode;
1494         struct nfs_inode *nfsi = NFS_I(inode);
1495         struct file_lock *fl;
1496         struct nfs4_lock_state *lsp;
1497         int status = 0;
1498         struct file_lock_context *flctx = inode->i_flctx;
1499         struct list_head *list;
1500
1501         if (flctx == NULL)
1502                 return 0;
1503
1504         list = &flctx->flc_posix;
1505
1506         /* Guard against delegation returns and new lock/unlock calls */
1507         down_write(&nfsi->rwsem);
1508         spin_lock(&flctx->flc_lock);
1509 restart:
1510         list_for_each_entry(fl, list, fl_list) {
1511                 if (nfs_file_open_context(fl->fl_file)->state != state)
1512                         continue;
1513                 spin_unlock(&flctx->flc_lock);
1514                 status = ops->recover_lock(state, fl);
1515                 switch (status) {
1516                 case 0:
1517                         break;
1518                 case -ETIMEDOUT:
1519                 case -ESTALE:
1520                 case -NFS4ERR_ADMIN_REVOKED:
1521                 case -NFS4ERR_STALE_STATEID:
1522                 case -NFS4ERR_BAD_STATEID:
1523                 case -NFS4ERR_EXPIRED:
1524                 case -NFS4ERR_NO_GRACE:
1525                 case -NFS4ERR_STALE_CLIENTID:
1526                 case -NFS4ERR_BADSESSION:
1527                 case -NFS4ERR_BADSLOT:
1528                 case -NFS4ERR_BAD_HIGH_SLOT:
1529                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1530                         goto out;
1531                 default:
1532                         pr_err("NFS: %s: unhandled error %d\n",
1533                                         __func__, status);
1534                         /* Fall through */
1535                 case -ENOMEM:
1536                 case -NFS4ERR_DENIED:
1537                 case -NFS4ERR_RECLAIM_BAD:
1538                 case -NFS4ERR_RECLAIM_CONFLICT:
1539                         lsp = fl->fl_u.nfs4_fl.owner;
1540                         if (lsp)
1541                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1542                         status = 0;
1543                 }
1544                 spin_lock(&flctx->flc_lock);
1545         }
1546         if (list == &flctx->flc_posix) {
1547                 list = &flctx->flc_flock;
1548                 goto restart;
1549         }
1550         spin_unlock(&flctx->flc_lock);
1551 out:
1552         up_write(&nfsi->rwsem);
1553         return status;
1554 }
1555
1556 #ifdef CONFIG_NFS_V4_2
1557 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1558 {
1559         struct nfs4_copy_state *copy;
1560
1561         if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1562                 !test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1563                 return;
1564
1565         spin_lock(&sp->so_server->nfs_client->cl_lock);
1566         list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1567                 if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1568                                 !nfs4_stateid_match_other(&state->stateid,
1569                                 &copy->parent_dst_state->stateid)))
1570                                 continue;
1571                 copy->flags = 1;
1572                 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1573                                 &state->flags)) {
1574                         clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1575                         complete(&copy->completion);
1576                 }
1577         }
1578         list_for_each_entry(copy, &sp->so_server->ss_copies, src_copies) {
1579                 if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1580                                 !nfs4_stateid_match_other(&state->stateid,
1581                                 &copy->parent_src_state->stateid)))
1582                                 continue;
1583                 copy->flags = 1;
1584                 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1585                                 &state->flags))
1586                         complete(&copy->completion);
1587         }
1588         spin_unlock(&sp->so_server->nfs_client->cl_lock);
1589 }
1590 #else /* !CONFIG_NFS_V4_2 */
1591 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1592                                          struct nfs4_state *state)
1593 {
1594 }
1595 #endif /* CONFIG_NFS_V4_2 */
1596
1597 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1598                                      const struct nfs4_state_recovery_ops *ops)
1599 {
1600         struct nfs4_lock_state *lock;
1601         int status;
1602
1603         status = ops->recover_open(sp, state);
1604         if (status < 0)
1605                 return status;
1606
1607         status = nfs4_reclaim_locks(state, ops);
1608         if (status < 0)
1609                 return status;
1610
1611         if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1612                 spin_lock(&state->state_lock);
1613                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1614                         trace_nfs4_state_lock_reclaim(state, lock);
1615                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1616                                 pr_warn_ratelimited("NFS: %s: Lock reclaim failed!\n", __func__);
1617                 }
1618                 spin_unlock(&state->state_lock);
1619         }
1620
1621         nfs42_complete_copies(sp, state);
1622         clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1623         return status;
1624 }
1625
1626 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1627 {
1628         struct nfs4_state *state;
1629         unsigned int loop = 0;
1630         int status = 0;
1631 #ifdef CONFIG_NFS_V4_2
1632         bool found_ssc_copy_state = false;
1633 #endif /* CONFIG_NFS_V4_2 */
1634
1635         /* Note: we rely on the sp->so_states list being ordered 
1636          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1637          * states first.
1638          * This is needed to ensure that the server won't give us any
1639          * read delegations that we have to return if, say, we are
1640          * recovering after a network partition or a reboot from a
1641          * server that doesn't support a grace period.
1642          */
1643         spin_lock(&sp->so_lock);
1644         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1645 restart:
1646         list_for_each_entry(state, &sp->so_states, open_states) {
1647                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1648                         continue;
1649                 if (!nfs4_valid_open_stateid(state))
1650                         continue;
1651                 if (state->state == 0)
1652                         continue;
1653 #ifdef CONFIG_NFS_V4_2
1654                 if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1655                         nfs4_state_mark_recovery_failed(state, -EIO);
1656                         found_ssc_copy_state = true;
1657                         continue;
1658                 }
1659 #endif /* CONFIG_NFS_V4_2 */
1660                 refcount_inc(&state->count);
1661                 spin_unlock(&sp->so_lock);
1662                 status = __nfs4_reclaim_open_state(sp, state, ops);
1663
1664                 switch (status) {
1665                 default:
1666                         if (status >= 0) {
1667                                 loop = 0;
1668                                 break;
1669                         }
1670                         printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1671                         /* Fall through */
1672                 case -ENOENT:
1673                 case -ENOMEM:
1674                 case -EACCES:
1675                 case -EROFS:
1676                 case -EIO:
1677                 case -ESTALE:
1678                         /* Open state on this file cannot be recovered */
1679                         nfs4_state_mark_recovery_failed(state, status);
1680                         break;
1681                 case -EAGAIN:
1682                         ssleep(1);
1683                         if (loop++ < 10) {
1684                                 set_bit(ops->state_flag_bit, &state->flags);
1685                                 break;
1686                         }
1687                         /* Fall through */
1688                 case -NFS4ERR_ADMIN_REVOKED:
1689                 case -NFS4ERR_STALE_STATEID:
1690                 case -NFS4ERR_OLD_STATEID:
1691                 case -NFS4ERR_BAD_STATEID:
1692                 case -NFS4ERR_RECLAIM_BAD:
1693                 case -NFS4ERR_RECLAIM_CONFLICT:
1694                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1695                         break;
1696                 case -NFS4ERR_EXPIRED:
1697                 case -NFS4ERR_NO_GRACE:
1698                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1699                         /* Fall through */
1700                 case -NFS4ERR_STALE_CLIENTID:
1701                 case -NFS4ERR_BADSESSION:
1702                 case -NFS4ERR_BADSLOT:
1703                 case -NFS4ERR_BAD_HIGH_SLOT:
1704                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1705                 case -ETIMEDOUT:
1706                         goto out_err;
1707                 }
1708                 nfs4_put_open_state(state);
1709                 spin_lock(&sp->so_lock);
1710                 goto restart;
1711         }
1712         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1713         spin_unlock(&sp->so_lock);
1714 #ifdef CONFIG_NFS_V4_2
1715         if (found_ssc_copy_state)
1716                 return -EIO;
1717 #endif /* CONFIG_NFS_V4_2 */
1718         return 0;
1719 out_err:
1720         nfs4_put_open_state(state);
1721         spin_lock(&sp->so_lock);
1722         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1723         spin_unlock(&sp->so_lock);
1724         return status;
1725 }
1726
1727 static void nfs4_clear_open_state(struct nfs4_state *state)
1728 {
1729         struct nfs4_lock_state *lock;
1730
1731         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1732         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1733         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1734         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1735         spin_lock(&state->state_lock);
1736         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1737                 lock->ls_seqid.flags = 0;
1738                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1739         }
1740         spin_unlock(&state->state_lock);
1741 }
1742
1743 static void nfs4_reset_seqids(struct nfs_server *server,
1744         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1745 {
1746         struct nfs_client *clp = server->nfs_client;
1747         struct nfs4_state_owner *sp;
1748         struct rb_node *pos;
1749         struct nfs4_state *state;
1750
1751         spin_lock(&clp->cl_lock);
1752         for (pos = rb_first(&server->state_owners);
1753              pos != NULL;
1754              pos = rb_next(pos)) {
1755                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1756                 sp->so_seqid.flags = 0;
1757                 spin_lock(&sp->so_lock);
1758                 list_for_each_entry(state, &sp->so_states, open_states) {
1759                         if (mark_reclaim(clp, state))
1760                                 nfs4_clear_open_state(state);
1761                 }
1762                 spin_unlock(&sp->so_lock);
1763         }
1764         spin_unlock(&clp->cl_lock);
1765 }
1766
1767 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1768         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1769 {
1770         struct nfs_server *server;
1771
1772         rcu_read_lock();
1773         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1774                 nfs4_reset_seqids(server, mark_reclaim);
1775         rcu_read_unlock();
1776 }
1777
1778 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1779 {
1780         /* Mark all delegations for reclaim */
1781         nfs_delegation_mark_reclaim(clp);
1782         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1783 }
1784
1785 static int nfs4_reclaim_complete(struct nfs_client *clp,
1786                                  const struct nfs4_state_recovery_ops *ops,
1787                                  const struct cred *cred)
1788 {
1789         /* Notify the server we're done reclaiming our state */
1790         if (ops->reclaim_complete)
1791                 return ops->reclaim_complete(clp, cred);
1792         return 0;
1793 }
1794
1795 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1796 {
1797         struct nfs_client *clp = server->nfs_client;
1798         struct nfs4_state_owner *sp;
1799         struct rb_node *pos;
1800         struct nfs4_state *state;
1801
1802         spin_lock(&clp->cl_lock);
1803         for (pos = rb_first(&server->state_owners);
1804              pos != NULL;
1805              pos = rb_next(pos)) {
1806                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1807                 spin_lock(&sp->so_lock);
1808                 list_for_each_entry(state, &sp->so_states, open_states) {
1809                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1810                                                 &state->flags))
1811                                 continue;
1812                         nfs4_state_mark_reclaim_nograce(clp, state);
1813                 }
1814                 spin_unlock(&sp->so_lock);
1815         }
1816         spin_unlock(&clp->cl_lock);
1817 }
1818
1819 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1820 {
1821         struct nfs_server *server;
1822
1823         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1824                 return 0;
1825
1826         rcu_read_lock();
1827         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1828                 nfs4_clear_reclaim_server(server);
1829         rcu_read_unlock();
1830
1831         nfs_delegation_reap_unclaimed(clp);
1832         return 1;
1833 }
1834
1835 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1836 {
1837         const struct nfs4_state_recovery_ops *ops;
1838         const struct cred *cred;
1839         int err;
1840
1841         if (!nfs4_state_clear_reclaim_reboot(clp))
1842                 return;
1843         ops = clp->cl_mvops->reboot_recovery_ops;
1844         cred = nfs4_get_clid_cred(clp);
1845         err = nfs4_reclaim_complete(clp, ops, cred);
1846         put_cred(cred);
1847         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1848                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1849 }
1850
1851 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1852 {
1853         nfs_mark_test_expired_all_delegations(clp);
1854         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1855 }
1856
1857 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1858 {
1859         switch (error) {
1860         case 0:
1861                 break;
1862         case -NFS4ERR_CB_PATH_DOWN:
1863                 nfs40_handle_cb_pathdown(clp);
1864                 break;
1865         case -NFS4ERR_NO_GRACE:
1866                 nfs4_state_end_reclaim_reboot(clp);
1867                 break;
1868         case -NFS4ERR_STALE_CLIENTID:
1869                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1870                 nfs4_state_start_reclaim_reboot(clp);
1871                 break;
1872         case -NFS4ERR_EXPIRED:
1873                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1874                 nfs4_state_start_reclaim_nograce(clp);
1875                 break;
1876         case -NFS4ERR_BADSESSION:
1877         case -NFS4ERR_BADSLOT:
1878         case -NFS4ERR_BAD_HIGH_SLOT:
1879         case -NFS4ERR_DEADSESSION:
1880         case -NFS4ERR_SEQ_FALSE_RETRY:
1881         case -NFS4ERR_SEQ_MISORDERED:
1882                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1883                 /* Zero session reset errors */
1884                 break;
1885         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1886                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1887                 break;
1888         default:
1889                 dprintk("%s: failed to handle error %d for server %s\n",
1890                                 __func__, error, clp->cl_hostname);
1891                 return error;
1892         }
1893         dprintk("%s: handled error %d for server %s\n", __func__, error,
1894                         clp->cl_hostname);
1895         return 0;
1896 }
1897
1898 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1899 {
1900         struct nfs4_state_owner *sp;
1901         struct nfs_server *server;
1902         struct rb_node *pos;
1903         LIST_HEAD(freeme);
1904         int status = 0;
1905
1906 restart:
1907         rcu_read_lock();
1908         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1909                 nfs4_purge_state_owners(server, &freeme);
1910                 spin_lock(&clp->cl_lock);
1911                 for (pos = rb_first(&server->state_owners);
1912                      pos != NULL;
1913                      pos = rb_next(pos)) {
1914                         sp = rb_entry(pos,
1915                                 struct nfs4_state_owner, so_server_node);
1916                         if (!test_and_clear_bit(ops->owner_flag_bit,
1917                                                         &sp->so_flags))
1918                                 continue;
1919                         if (!atomic_inc_not_zero(&sp->so_count))
1920                                 continue;
1921                         spin_unlock(&clp->cl_lock);
1922                         rcu_read_unlock();
1923
1924                         status = nfs4_reclaim_open_state(sp, ops);
1925                         if (status < 0) {
1926                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1927                                 nfs4_put_state_owner(sp);
1928                                 status = nfs4_recovery_handle_error(clp, status);
1929                                 return (status != 0) ? status : -EAGAIN;
1930                         }
1931
1932                         nfs4_put_state_owner(sp);
1933                         goto restart;
1934                 }
1935                 spin_unlock(&clp->cl_lock);
1936         }
1937         rcu_read_unlock();
1938         nfs4_free_state_owners(&freeme);
1939         return 0;
1940 }
1941
1942 static int nfs4_check_lease(struct nfs_client *clp)
1943 {
1944         const struct cred *cred;
1945         const struct nfs4_state_maintenance_ops *ops =
1946                 clp->cl_mvops->state_renewal_ops;
1947         int status;
1948
1949         /* Is the client already known to have an expired lease? */
1950         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1951                 return 0;
1952         cred = ops->get_state_renewal_cred(clp);
1953         if (cred == NULL) {
1954                 cred = nfs4_get_clid_cred(clp);
1955                 status = -ENOKEY;
1956                 if (cred == NULL)
1957                         goto out;
1958         }
1959         status = ops->renew_lease(clp, cred);
1960         put_cred(cred);
1961         if (status == -ETIMEDOUT) {
1962                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1963                 return 0;
1964         }
1965 out:
1966         return nfs4_recovery_handle_error(clp, status);
1967 }
1968
1969 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1970  * and for recoverable errors on EXCHANGE_ID for v4.1
1971  */
1972 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1973 {
1974         switch (status) {
1975         case -NFS4ERR_SEQ_MISORDERED:
1976                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1977                         return -ESERVERFAULT;
1978                 /* Lease confirmation error: retry after purging the lease */
1979                 ssleep(1);
1980                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1981                 break;
1982         case -NFS4ERR_STALE_CLIENTID:
1983                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1984                 nfs4_state_start_reclaim_reboot(clp);
1985                 break;
1986         case -NFS4ERR_CLID_INUSE:
1987                 pr_err("NFS: Server %s reports our clientid is in use\n",
1988                         clp->cl_hostname);
1989                 nfs_mark_client_ready(clp, -EPERM);
1990                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1991                 return -EPERM;
1992         case -EACCES:
1993         case -NFS4ERR_DELAY:
1994         case -EAGAIN:
1995                 ssleep(1);
1996                 break;
1997
1998         case -NFS4ERR_MINOR_VERS_MISMATCH:
1999                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2000                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2001                 dprintk("%s: exit with error %d for server %s\n",
2002                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
2003                 return -EPROTONOSUPPORT;
2004         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2005                                  * in nfs4_exchange_id */
2006         default:
2007                 dprintk("%s: exit with error %d for server %s\n", __func__,
2008                                 status, clp->cl_hostname);
2009                 return status;
2010         }
2011         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2012         dprintk("%s: handled error %d for server %s\n", __func__, status,
2013                         clp->cl_hostname);
2014         return 0;
2015 }
2016
2017 static int nfs4_establish_lease(struct nfs_client *clp)
2018 {
2019         const struct cred *cred;
2020         const struct nfs4_state_recovery_ops *ops =
2021                 clp->cl_mvops->reboot_recovery_ops;
2022         int status;
2023
2024         status = nfs4_begin_drain_session(clp);
2025         if (status != 0)
2026                 return status;
2027         cred = nfs4_get_clid_cred(clp);
2028         if (cred == NULL)
2029                 return -ENOENT;
2030         status = ops->establish_clid(clp, cred);
2031         put_cred(cred);
2032         if (status != 0)
2033                 return status;
2034         pnfs_destroy_all_layouts(clp);
2035         return 0;
2036 }
2037
2038 /*
2039  * Returns zero or a negative errno.  NFS4ERR values are converted
2040  * to local errno values.
2041  */
2042 static int nfs4_reclaim_lease(struct nfs_client *clp)
2043 {
2044         int status;
2045
2046         status = nfs4_establish_lease(clp);
2047         if (status < 0)
2048                 return nfs4_handle_reclaim_lease_error(clp, status);
2049         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2050                 nfs4_state_start_reclaim_nograce(clp);
2051         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2052                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2053         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2054         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2055         return 0;
2056 }
2057
2058 static int nfs4_purge_lease(struct nfs_client *clp)
2059 {
2060         int status;
2061
2062         status = nfs4_establish_lease(clp);
2063         if (status < 0)
2064                 return nfs4_handle_reclaim_lease_error(clp, status);
2065         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2066         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2067         nfs4_state_start_reclaim_nograce(clp);
2068         return 0;
2069 }
2070
2071 /*
2072  * Try remote migration of one FSID from a source server to a
2073  * destination server.  The source server provides a list of
2074  * potential destinations.
2075  *
2076  * Returns zero or a negative NFS4ERR status code.
2077  */
2078 static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2079 {
2080         struct nfs_client *clp = server->nfs_client;
2081         struct nfs4_fs_locations *locations = NULL;
2082         struct inode *inode;
2083         struct page *page;
2084         int status, result;
2085
2086         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2087                         (unsigned long long)server->fsid.major,
2088                         (unsigned long long)server->fsid.minor,
2089                         clp->cl_hostname);
2090
2091         result = 0;
2092         page = alloc_page(GFP_KERNEL);
2093         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2094         if (page == NULL || locations == NULL) {
2095                 dprintk("<-- %s: no memory\n", __func__);
2096                 goto out;
2097         }
2098
2099         inode = d_inode(server->super->s_root);
2100         result = nfs4_proc_get_locations(inode, locations, page, cred);
2101         if (result) {
2102                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2103                         __func__, result);
2104                 goto out;
2105         }
2106
2107         result = -NFS4ERR_NXIO;
2108         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2109                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2110                         __func__);
2111                 goto out;
2112         }
2113
2114         status = nfs4_begin_drain_session(clp);
2115         if (status != 0) {
2116                 result = status;
2117                 goto out;
2118         }
2119
2120         status = nfs4_replace_transport(server, locations);
2121         if (status != 0) {
2122                 dprintk("<-- %s: failed to replace transport: %d\n",
2123                         __func__, status);
2124                 goto out;
2125         }
2126
2127         result = 0;
2128         dprintk("<-- %s: migration succeeded\n", __func__);
2129
2130 out:
2131         if (page != NULL)
2132                 __free_page(page);
2133         kfree(locations);
2134         if (result) {
2135                 pr_err("NFS: migration recovery failed (server %s)\n",
2136                                 clp->cl_hostname);
2137                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2138         }
2139         return result;
2140 }
2141
2142 /*
2143  * Returns zero or a negative NFS4ERR status code.
2144  */
2145 static int nfs4_handle_migration(struct nfs_client *clp)
2146 {
2147         const struct nfs4_state_maintenance_ops *ops =
2148                                 clp->cl_mvops->state_renewal_ops;
2149         struct nfs_server *server;
2150         const struct cred *cred;
2151
2152         dprintk("%s: migration reported on \"%s\"\n", __func__,
2153                         clp->cl_hostname);
2154
2155         cred = ops->get_state_renewal_cred(clp);
2156         if (cred == NULL)
2157                 return -NFS4ERR_NOENT;
2158
2159         clp->cl_mig_gen++;
2160 restart:
2161         rcu_read_lock();
2162         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2163                 int status;
2164
2165                 if (server->mig_gen == clp->cl_mig_gen)
2166                         continue;
2167                 server->mig_gen = clp->cl_mig_gen;
2168
2169                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2170                                                 &server->mig_status))
2171                         continue;
2172
2173                 rcu_read_unlock();
2174                 status = nfs4_try_migration(server, cred);
2175                 if (status < 0) {
2176                         put_cred(cred);
2177                         return status;
2178                 }
2179                 goto restart;
2180         }
2181         rcu_read_unlock();
2182         put_cred(cred);
2183         return 0;
2184 }
2185
2186 /*
2187  * Test each nfs_server on the clp's cl_superblocks list to see
2188  * if it's moved to another server.  Stop when the server no longer
2189  * returns NFS4ERR_LEASE_MOVED.
2190  */
2191 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2192 {
2193         const struct nfs4_state_maintenance_ops *ops =
2194                                 clp->cl_mvops->state_renewal_ops;
2195         struct nfs_server *server;
2196         const struct cred *cred;
2197
2198         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2199                         clp->cl_hostname);
2200
2201         cred = ops->get_state_renewal_cred(clp);
2202         if (cred == NULL)
2203                 return -NFS4ERR_NOENT;
2204
2205         clp->cl_mig_gen++;
2206 restart:
2207         rcu_read_lock();
2208         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2209                 struct inode *inode;
2210                 int status;
2211
2212                 if (server->mig_gen == clp->cl_mig_gen)
2213                         continue;
2214                 server->mig_gen = clp->cl_mig_gen;
2215
2216                 rcu_read_unlock();
2217
2218                 inode = d_inode(server->super->s_root);
2219                 status = nfs4_proc_fsid_present(inode, cred);
2220                 if (status != -NFS4ERR_MOVED)
2221                         goto restart;   /* wasn't this one */
2222                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2223                         goto restart;   /* there are more */
2224                 goto out;
2225         }
2226         rcu_read_unlock();
2227
2228 out:
2229         put_cred(cred);
2230         return 0;
2231 }
2232
2233 /**
2234  * nfs4_discover_server_trunking - Detect server IP address trunking
2235  *
2236  * @clp: nfs_client under test
2237  * @result: OUT: found nfs_client, or clp
2238  *
2239  * Returns zero or a negative errno.  If zero is returned,
2240  * an nfs_client pointer is planted in "result".
2241  *
2242  * Note: since we are invoked in process context, and
2243  * not from inside the state manager, we cannot use
2244  * nfs4_handle_reclaim_lease_error().
2245  */
2246 int nfs4_discover_server_trunking(struct nfs_client *clp,
2247                                   struct nfs_client **result)
2248 {
2249         const struct nfs4_state_recovery_ops *ops =
2250                                 clp->cl_mvops->reboot_recovery_ops;
2251         struct rpc_clnt *clnt;
2252         const struct cred *cred;
2253         int i, status;
2254
2255         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2256
2257         clnt = clp->cl_rpcclient;
2258         i = 0;
2259
2260         mutex_lock(&nfs_clid_init_mutex);
2261 again:
2262         status  = -ENOENT;
2263         cred = nfs4_get_clid_cred(clp);
2264         if (cred == NULL)
2265                 goto out_unlock;
2266
2267         status = ops->detect_trunking(clp, result, cred);
2268         put_cred(cred);
2269         switch (status) {
2270         case 0:
2271         case -EINTR:
2272         case -ERESTARTSYS:
2273                 break;
2274         case -ETIMEDOUT:
2275                 if (clnt->cl_softrtry)
2276                         break;
2277                 /* Fall through */
2278         case -NFS4ERR_DELAY:
2279         case -EAGAIN:
2280                 ssleep(1);
2281                 /* Fall through */
2282         case -NFS4ERR_STALE_CLIENTID:
2283                 dprintk("NFS: %s after status %d, retrying\n",
2284                         __func__, status);
2285                 goto again;
2286         case -EACCES:
2287                 if (i++ == 0) {
2288                         nfs4_root_machine_cred(clp);
2289                         goto again;
2290                 }
2291                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2292                         break;
2293                 /* Fall through */
2294         case -NFS4ERR_CLID_INUSE:
2295         case -NFS4ERR_WRONGSEC:
2296                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2297                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2298                         status = -EPERM;
2299                         break;
2300                 }
2301                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2302                 if (IS_ERR(clnt)) {
2303                         status = PTR_ERR(clnt);
2304                         break;
2305                 }
2306                 /* Note: this is safe because we haven't yet marked the
2307                  * client as ready, so we are the only user of
2308                  * clp->cl_rpcclient
2309                  */
2310                 clnt = xchg(&clp->cl_rpcclient, clnt);
2311                 rpc_shutdown_client(clnt);
2312                 clnt = clp->cl_rpcclient;
2313                 goto again;
2314
2315         case -NFS4ERR_MINOR_VERS_MISMATCH:
2316                 status = -EPROTONOSUPPORT;
2317                 break;
2318
2319         case -EKEYEXPIRED:
2320         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2321                                  * in nfs4_exchange_id */
2322                 status = -EKEYEXPIRED;
2323                 break;
2324         default:
2325                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2326                                 __func__, status);
2327                 status = -EIO;
2328         }
2329
2330 out_unlock:
2331         mutex_unlock(&nfs_clid_init_mutex);
2332         dprintk("NFS: %s: status = %d\n", __func__, status);
2333         return status;
2334 }
2335
2336 #ifdef CONFIG_NFS_V4_1
2337 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2338 {
2339         struct nfs_client *clp = session->clp;
2340
2341         switch (err) {
2342         default:
2343                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2344                 break;
2345         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2346                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2347         }
2348         nfs4_schedule_state_manager(clp);
2349 }
2350 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2351
2352 void nfs41_notify_server(struct nfs_client *clp)
2353 {
2354         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2355         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2356         nfs4_schedule_state_manager(clp);
2357 }
2358
2359 static void nfs4_reset_all_state(struct nfs_client *clp)
2360 {
2361         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2362                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2363                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2364                 nfs4_state_start_reclaim_nograce(clp);
2365                 dprintk("%s: scheduling reset of all state for server %s!\n",
2366                                 __func__, clp->cl_hostname);
2367                 nfs4_schedule_state_manager(clp);
2368         }
2369 }
2370
2371 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2372 {
2373         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2374                 nfs4_state_start_reclaim_reboot(clp);
2375                 dprintk("%s: server %s rebooted!\n", __func__,
2376                                 clp->cl_hostname);
2377                 nfs4_schedule_state_manager(clp);
2378         }
2379 }
2380
2381 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2382 {
2383         nfs4_reset_all_state(clp);
2384         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2385 }
2386
2387 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2388 {
2389         nfs4_state_start_reclaim_nograce(clp);
2390         nfs4_schedule_state_manager(clp);
2391
2392         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2393 }
2394
2395 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2396 {
2397         /* FIXME: For now, we destroy all layouts. */
2398         pnfs_destroy_all_layouts(clp);
2399         nfs_test_expired_all_delegations(clp);
2400         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2401                         clp->cl_hostname);
2402 }
2403
2404 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2405 {
2406         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2407         nfs4_schedule_state_manager(clp);
2408
2409         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2410                         clp->cl_hostname);
2411 }
2412
2413 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2414 {
2415         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2416                 &clp->cl_state) == 0)
2417                 nfs4_schedule_state_manager(clp);
2418 }
2419
2420 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2421                 bool recovery)
2422 {
2423         if (!flags)
2424                 return;
2425
2426         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2427                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2428         /*
2429          * If we're called from the state manager thread, then assume we're
2430          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2431          * Those flags are expected to remain set until we're done
2432          * recovering (see RFC5661, section 18.46.3).
2433          */
2434         if (recovery)
2435                 goto out_recovery;
2436
2437         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2438                 nfs41_handle_server_reboot(clp);
2439         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2440                 nfs41_handle_all_state_revoked(clp);
2441         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2442                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2443                 nfs41_handle_some_state_revoked(clp);
2444         if (flags & SEQ4_STATUS_LEASE_MOVED)
2445                 nfs4_schedule_lease_moved_recovery(clp);
2446         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2447                 nfs41_handle_recallable_state_revoked(clp);
2448 out_recovery:
2449         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2450                 nfs41_handle_backchannel_fault(clp);
2451         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2452                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2453                 nfs41_handle_cb_path_down(clp);
2454 }
2455
2456 static int nfs4_reset_session(struct nfs_client *clp)
2457 {
2458         const struct cred *cred;
2459         int status;
2460
2461         if (!nfs4_has_session(clp))
2462                 return 0;
2463         status = nfs4_begin_drain_session(clp);
2464         if (status != 0)
2465                 return status;
2466         cred = nfs4_get_clid_cred(clp);
2467         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2468         switch (status) {
2469         case 0:
2470         case -NFS4ERR_BADSESSION:
2471         case -NFS4ERR_DEADSESSION:
2472                 break;
2473         case -NFS4ERR_BACK_CHAN_BUSY:
2474         case -NFS4ERR_DELAY:
2475                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2476                 status = 0;
2477                 ssleep(1);
2478                 goto out;
2479         default:
2480                 status = nfs4_recovery_handle_error(clp, status);
2481                 goto out;
2482         }
2483
2484         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2485         status = nfs4_proc_create_session(clp, cred);
2486         if (status) {
2487                 dprintk("%s: session reset failed with status %d for server %s!\n",
2488                         __func__, status, clp->cl_hostname);
2489                 status = nfs4_handle_reclaim_lease_error(clp, status);
2490                 goto out;
2491         }
2492         nfs41_finish_session_reset(clp);
2493         dprintk("%s: session reset was successful for server %s!\n",
2494                         __func__, clp->cl_hostname);
2495 out:
2496         put_cred(cred);
2497         return status;
2498 }
2499
2500 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2501 {
2502         const struct cred *cred;
2503         int ret;
2504
2505         if (!nfs4_has_session(clp))
2506                 return 0;
2507         ret = nfs4_begin_drain_session(clp);
2508         if (ret != 0)
2509                 return ret;
2510         cred = nfs4_get_clid_cred(clp);
2511         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2512         put_cred(cred);
2513         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2514         switch (ret) {
2515         case 0:
2516                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2517                         __func__, clp->cl_hostname);
2518                 break;
2519         case -NFS4ERR_DELAY:
2520                 ssleep(1);
2521                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2522                 break;
2523         default:
2524                 return nfs4_recovery_handle_error(clp, ret);
2525         }
2526         return 0;
2527 }
2528 #else /* CONFIG_NFS_V4_1 */
2529 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2530
2531 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2532 {
2533         return 0;
2534 }
2535 #endif /* CONFIG_NFS_V4_1 */
2536
2537 static void nfs4_state_manager(struct nfs_client *clp)
2538 {
2539         int status = 0;
2540         const char *section = "", *section_sep = "";
2541
2542         /* Ensure exclusive access to NFSv4 state */
2543         do {
2544                 trace_nfs4_state_mgr(clp);
2545                 clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2546                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2547                         section = "purge state";
2548                         status = nfs4_purge_lease(clp);
2549                         if (status < 0)
2550                                 goto out_error;
2551                         continue;
2552                 }
2553
2554                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2555                         section = "lease expired";
2556                         /* We're going to have to re-establish a clientid */
2557                         status = nfs4_reclaim_lease(clp);
2558                         if (status < 0)
2559                                 goto out_error;
2560                         continue;
2561                 }
2562
2563                 /* Initialize or reset the session */
2564                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2565                         section = "reset session";
2566                         status = nfs4_reset_session(clp);
2567                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2568                                 continue;
2569                         if (status < 0)
2570                                 goto out_error;
2571                 }
2572
2573                 /* Send BIND_CONN_TO_SESSION */
2574                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2575                                 &clp->cl_state)) {
2576                         section = "bind conn to session";
2577                         status = nfs4_bind_conn_to_session(clp);
2578                         if (status < 0)
2579                                 goto out_error;
2580                         continue;
2581                 }
2582
2583                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2584                         section = "check lease";
2585                         status = nfs4_check_lease(clp);
2586                         if (status < 0)
2587                                 goto out_error;
2588                         continue;
2589                 }
2590
2591                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2592                         section = "migration";
2593                         status = nfs4_handle_migration(clp);
2594                         if (status < 0)
2595                                 goto out_error;
2596                 }
2597
2598                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2599                         section = "lease moved";
2600                         status = nfs4_handle_lease_moved(clp);
2601                         if (status < 0)
2602                                 goto out_error;
2603                 }
2604
2605                 /* First recover reboot state... */
2606                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2607                         section = "reclaim reboot";
2608                         status = nfs4_do_reclaim(clp,
2609                                 clp->cl_mvops->reboot_recovery_ops);
2610                         if (status == -EAGAIN)
2611                                 continue;
2612                         if (status < 0)
2613                                 goto out_error;
2614                         nfs4_state_end_reclaim_reboot(clp);
2615                 }
2616
2617                 /* Detect expired delegations... */
2618                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2619                         section = "detect expired delegations";
2620                         nfs_reap_expired_delegations(clp);
2621                         continue;
2622                 }
2623
2624                 /* Now recover expired state... */
2625                 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2626                         section = "reclaim nograce";
2627                         status = nfs4_do_reclaim(clp,
2628                                 clp->cl_mvops->nograce_recovery_ops);
2629                         if (status == -EAGAIN)
2630                                 continue;
2631                         if (status < 0)
2632                                 goto out_error;
2633                         clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2634                 }
2635
2636                 nfs4_end_drain_session(clp);
2637                 nfs4_clear_state_manager_bit(clp);
2638
2639                 if (!test_and_set_bit(NFS4CLNT_DELEGRETURN_RUNNING, &clp->cl_state)) {
2640                         if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2641                                 nfs_client_return_marked_delegations(clp);
2642                                 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2643                         }
2644                         clear_bit(NFS4CLNT_DELEGRETURN_RUNNING, &clp->cl_state);
2645                 }
2646
2647                 /* Did we race with an attempt to give us more work? */
2648                 if (!test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state))
2649                         return;
2650                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2651                         return;
2652         } while (refcount_read(&clp->cl_count) > 1 && !signalled());
2653         goto out_drain;
2654
2655 out_error:
2656         if (strlen(section))
2657                 section_sep = ": ";
2658         trace_nfs4_state_mgr_failed(clp, section, status);
2659         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2660                         " with error %d\n", section_sep, section,
2661                         clp->cl_hostname, -status);
2662         ssleep(1);
2663 out_drain:
2664         nfs4_end_drain_session(clp);
2665         nfs4_clear_state_manager_bit(clp);
2666 }
2667
2668 static int nfs4_run_state_manager(void *ptr)
2669 {
2670         struct nfs_client *clp = ptr;
2671
2672         allow_signal(SIGKILL);
2673         nfs4_state_manager(clp);
2674         nfs_put_client(clp);
2675         module_put_and_exit(0);
2676         return 0;
2677 }
2678
2679 /*
2680  * Local variables:
2681  *  c-basic-offset: 8
2682  * End:
2683  */