Merge tag 'exynos-cleanup' of git://git.kernel.org/pub/scm/linux/kernel/git/kgene...
[cascardo/linux.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/ratelimit.h>
50 #include <linux/workqueue.h>
51 #include <linux/bitops.h>
52 #include <linux/jiffies.h>
53
54 #include <linux/sunrpc/clnt.h>
55
56 #include "nfs4_fs.h"
57 #include "callback.h"
58 #include "delegation.h"
59 #include "internal.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
63
64 #define NFSDBG_FACILITY         NFSDBG_STATE
65
66 #define OPENOWNER_POOL_SIZE     8
67
68 const nfs4_stateid zero_stateid;
69 static DEFINE_MUTEX(nfs_clid_init_mutex);
70
71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
72 {
73         struct nfs4_setclientid_res clid = {
74                 .clientid = clp->cl_clientid,
75                 .confirm = clp->cl_confirm,
76         };
77         unsigned short port;
78         int status;
79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
80
81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
82                 goto do_confirm;
83         port = nn->nfs_callback_tcpport;
84         if (clp->cl_addr.ss_family == AF_INET6)
85                 port = nn->nfs_callback_tcpport6;
86
87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
88         if (status != 0)
89                 goto out;
90         clp->cl_clientid = clid.clientid;
91         clp->cl_confirm = clid.confirm;
92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
93 do_confirm:
94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
95         if (status != 0)
96                 goto out;
97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
98         nfs4_schedule_state_renewal(clp);
99 out:
100         return status;
101 }
102
103 /**
104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
105  *
106  * @clp: nfs_client under test
107  * @result: OUT: found nfs_client, or clp
108  * @cred: credential to use for trunking test
109  *
110  * Returns zero, a negative errno, or a negative NFS4ERR status.
111  * If zero is returned, an nfs_client pointer is planted in
112  * "result".
113  *
114  * Note: The returned client may not yet be marked ready.
115  */
116 int nfs40_discover_server_trunking(struct nfs_client *clp,
117                                    struct nfs_client **result,
118                                    struct rpc_cred *cred)
119 {
120         struct nfs4_setclientid_res clid = {
121                 .clientid = clp->cl_clientid,
122                 .confirm = clp->cl_confirm,
123         };
124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
125         unsigned short port;
126         int status;
127
128         port = nn->nfs_callback_tcpport;
129         if (clp->cl_addr.ss_family == AF_INET6)
130                 port = nn->nfs_callback_tcpport6;
131
132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
133         if (status != 0)
134                 goto out;
135         clp->cl_clientid = clid.clientid;
136         clp->cl_confirm = clid.confirm;
137
138         status = nfs40_walk_client_list(clp, result, cred);
139         if (status == 0) {
140                 /* Sustain the lease, even if it's empty.  If the clientid4
141                  * goes stale it's of no use for trunking discovery. */
142                 nfs4_schedule_state_renewal(*result);
143         }
144 out:
145         return status;
146 }
147
148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
149 {
150         struct rpc_cred *cred = NULL;
151
152         if (clp->cl_machine_cred != NULL)
153                 cred = get_rpccred(clp->cl_machine_cred);
154         return cred;
155 }
156
157 static void nfs4_root_machine_cred(struct nfs_client *clp)
158 {
159         struct rpc_cred *cred, *new;
160
161         new = rpc_lookup_machine_cred(NULL);
162         spin_lock(&clp->cl_lock);
163         cred = clp->cl_machine_cred;
164         clp->cl_machine_cred = new;
165         spin_unlock(&clp->cl_lock);
166         if (cred != NULL)
167                 put_rpccred(cred);
168 }
169
170 static struct rpc_cred *
171 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
172 {
173         struct rpc_cred *cred = NULL;
174         struct nfs4_state_owner *sp;
175         struct rb_node *pos;
176
177         for (pos = rb_first(&server->state_owners);
178              pos != NULL;
179              pos = rb_next(pos)) {
180                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
181                 if (list_empty(&sp->so_states))
182                         continue;
183                 cred = get_rpccred(sp->so_cred);
184                 break;
185         }
186         return cred;
187 }
188
189 /**
190  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
191  * @clp: client state handle
192  *
193  * Returns an rpc_cred with reference count bumped, or NULL.
194  * Caller must hold clp->cl_lock.
195  */
196 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
197 {
198         struct rpc_cred *cred = NULL;
199         struct nfs_server *server;
200
201         /* Use machine credentials if available */
202         cred = nfs4_get_machine_cred_locked(clp);
203         if (cred != NULL)
204                 goto out;
205
206         rcu_read_lock();
207         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
208                 cred = nfs4_get_renew_cred_server_locked(server);
209                 if (cred != NULL)
210                         break;
211         }
212         rcu_read_unlock();
213
214 out:
215         return cred;
216 }
217
218 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
219 {
220         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
221                 spin_lock(&tbl->slot_tbl_lock);
222                 nfs41_wake_slot_table(tbl);
223                 spin_unlock(&tbl->slot_tbl_lock);
224         }
225 }
226
227 static void nfs4_end_drain_session(struct nfs_client *clp)
228 {
229         struct nfs4_session *ses = clp->cl_session;
230
231         if (clp->cl_slot_tbl) {
232                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
233                 return;
234         }
235
236         if (ses != NULL) {
237                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
238                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
239         }
240 }
241
242 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
243 {
244         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
245         spin_lock(&tbl->slot_tbl_lock);
246         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
247                 reinit_completion(&tbl->complete);
248                 spin_unlock(&tbl->slot_tbl_lock);
249                 return wait_for_completion_interruptible(&tbl->complete);
250         }
251         spin_unlock(&tbl->slot_tbl_lock);
252         return 0;
253 }
254
255 static int nfs4_begin_drain_session(struct nfs_client *clp)
256 {
257         struct nfs4_session *ses = clp->cl_session;
258         int ret = 0;
259
260         if (clp->cl_slot_tbl)
261                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
262
263         /* back channel */
264         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
265         if (ret)
266                 return ret;
267         /* fore channel */
268         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
269 }
270
271 #if defined(CONFIG_NFS_V4_1)
272
273 static int nfs41_setup_state_renewal(struct nfs_client *clp)
274 {
275         int status;
276         struct nfs_fsinfo fsinfo;
277
278         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
279                 nfs4_schedule_state_renewal(clp);
280                 return 0;
281         }
282
283         status = nfs4_proc_get_lease_time(clp, &fsinfo);
284         if (status == 0) {
285                 /* Update lease time and schedule renewal */
286                 spin_lock(&clp->cl_lock);
287                 clp->cl_lease_time = fsinfo.lease_time * HZ;
288                 clp->cl_last_renewal = jiffies;
289                 spin_unlock(&clp->cl_lock);
290
291                 nfs4_schedule_state_renewal(clp);
292         }
293
294         return status;
295 }
296
297 static void nfs41_finish_session_reset(struct nfs_client *clp)
298 {
299         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301         /* create_session negotiated new slot table */
302         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303         nfs41_setup_state_renewal(clp);
304 }
305
306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
307 {
308         int status;
309
310         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311                 goto do_confirm;
312         nfs4_begin_drain_session(clp);
313         status = nfs4_proc_exchange_id(clp, cred);
314         if (status != 0)
315                 goto out;
316         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
317 do_confirm:
318         status = nfs4_proc_create_session(clp, cred);
319         if (status != 0)
320                 goto out;
321         nfs41_finish_session_reset(clp);
322         nfs_mark_client_ready(clp, NFS_CS_READY);
323 out:
324         return status;
325 }
326
327 /**
328  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
329  *
330  * @clp: nfs_client under test
331  * @result: OUT: found nfs_client, or clp
332  * @cred: credential to use for trunking test
333  *
334  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
335  * If NFS4_OK is returned, an nfs_client pointer is planted in
336  * "result".
337  *
338  * Note: The returned client may not yet be marked ready.
339  */
340 int nfs41_discover_server_trunking(struct nfs_client *clp,
341                                    struct nfs_client **result,
342                                    struct rpc_cred *cred)
343 {
344         int status;
345
346         status = nfs4_proc_exchange_id(clp, cred);
347         if (status != NFS4_OK)
348                 return status;
349         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
350
351         return nfs41_walk_client_list(clp, result, cred);
352 }
353
354 #endif /* CONFIG_NFS_V4_1 */
355
356 /**
357  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
358  * @clp: client state handle
359  *
360  * Returns an rpc_cred with reference count bumped, or NULL.
361  */
362 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
363 {
364         struct rpc_cred *cred;
365
366         spin_lock(&clp->cl_lock);
367         cred = nfs4_get_machine_cred_locked(clp);
368         spin_unlock(&clp->cl_lock);
369         return cred;
370 }
371
372 static struct nfs4_state_owner *
373 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
374 {
375         struct rb_node **p = &server->state_owners.rb_node,
376                        *parent = NULL;
377         struct nfs4_state_owner *sp;
378
379         while (*p != NULL) {
380                 parent = *p;
381                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
382
383                 if (cred < sp->so_cred)
384                         p = &parent->rb_left;
385                 else if (cred > sp->so_cred)
386                         p = &parent->rb_right;
387                 else {
388                         if (!list_empty(&sp->so_lru))
389                                 list_del_init(&sp->so_lru);
390                         atomic_inc(&sp->so_count);
391                         return sp;
392                 }
393         }
394         return NULL;
395 }
396
397 static struct nfs4_state_owner *
398 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
399 {
400         struct nfs_server *server = new->so_server;
401         struct rb_node **p = &server->state_owners.rb_node,
402                        *parent = NULL;
403         struct nfs4_state_owner *sp;
404         int err;
405
406         while (*p != NULL) {
407                 parent = *p;
408                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
409
410                 if (new->so_cred < sp->so_cred)
411                         p = &parent->rb_left;
412                 else if (new->so_cred > sp->so_cred)
413                         p = &parent->rb_right;
414                 else {
415                         if (!list_empty(&sp->so_lru))
416                                 list_del_init(&sp->so_lru);
417                         atomic_inc(&sp->so_count);
418                         return sp;
419                 }
420         }
421         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
422         if (err)
423                 return ERR_PTR(err);
424         rb_link_node(&new->so_server_node, parent, p);
425         rb_insert_color(&new->so_server_node, &server->state_owners);
426         return new;
427 }
428
429 static void
430 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
431 {
432         struct nfs_server *server = sp->so_server;
433
434         if (!RB_EMPTY_NODE(&sp->so_server_node))
435                 rb_erase(&sp->so_server_node, &server->state_owners);
436         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
437 }
438
439 static void
440 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
441 {
442         sc->create_time = ktime_get();
443         sc->flags = 0;
444         sc->counter = 0;
445         spin_lock_init(&sc->lock);
446         INIT_LIST_HEAD(&sc->list);
447         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
448 }
449
450 static void
451 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
452 {
453         rpc_destroy_wait_queue(&sc->wait);
454 }
455
456 /*
457  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
458  * create a new state_owner.
459  *
460  */
461 static struct nfs4_state_owner *
462 nfs4_alloc_state_owner(struct nfs_server *server,
463                 struct rpc_cred *cred,
464                 gfp_t gfp_flags)
465 {
466         struct nfs4_state_owner *sp;
467
468         sp = kzalloc(sizeof(*sp), gfp_flags);
469         if (!sp)
470                 return NULL;
471         sp->so_server = server;
472         sp->so_cred = get_rpccred(cred);
473         spin_lock_init(&sp->so_lock);
474         INIT_LIST_HEAD(&sp->so_states);
475         nfs4_init_seqid_counter(&sp->so_seqid);
476         atomic_set(&sp->so_count, 1);
477         INIT_LIST_HEAD(&sp->so_lru);
478         seqcount_init(&sp->so_reclaim_seqcount);
479         mutex_init(&sp->so_delegreturn_mutex);
480         return sp;
481 }
482
483 static void
484 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
485 {
486         struct rb_node *rb_node = &sp->so_server_node;
487
488         if (!RB_EMPTY_NODE(rb_node)) {
489                 struct nfs_server *server = sp->so_server;
490                 struct nfs_client *clp = server->nfs_client;
491
492                 spin_lock(&clp->cl_lock);
493                 if (!RB_EMPTY_NODE(rb_node)) {
494                         rb_erase(rb_node, &server->state_owners);
495                         RB_CLEAR_NODE(rb_node);
496                 }
497                 spin_unlock(&clp->cl_lock);
498         }
499 }
500
501 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
502 {
503         nfs4_destroy_seqid_counter(&sp->so_seqid);
504         put_rpccred(sp->so_cred);
505         kfree(sp);
506 }
507
508 static void nfs4_gc_state_owners(struct nfs_server *server)
509 {
510         struct nfs_client *clp = server->nfs_client;
511         struct nfs4_state_owner *sp, *tmp;
512         unsigned long time_min, time_max;
513         LIST_HEAD(doomed);
514
515         spin_lock(&clp->cl_lock);
516         time_max = jiffies;
517         time_min = (long)time_max - (long)clp->cl_lease_time;
518         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
519                 /* NB: LRU is sorted so that oldest is at the head */
520                 if (time_in_range(sp->so_expires, time_min, time_max))
521                         break;
522                 list_move(&sp->so_lru, &doomed);
523                 nfs4_remove_state_owner_locked(sp);
524         }
525         spin_unlock(&clp->cl_lock);
526
527         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
528                 list_del(&sp->so_lru);
529                 nfs4_free_state_owner(sp);
530         }
531 }
532
533 /**
534  * nfs4_get_state_owner - Look up a state owner given a credential
535  * @server: nfs_server to search
536  * @cred: RPC credential to match
537  *
538  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
539  */
540 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
541                                               struct rpc_cred *cred,
542                                               gfp_t gfp_flags)
543 {
544         struct nfs_client *clp = server->nfs_client;
545         struct nfs4_state_owner *sp, *new;
546
547         spin_lock(&clp->cl_lock);
548         sp = nfs4_find_state_owner_locked(server, cred);
549         spin_unlock(&clp->cl_lock);
550         if (sp != NULL)
551                 goto out;
552         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
553         if (new == NULL)
554                 goto out;
555         do {
556                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
557                         break;
558                 spin_lock(&clp->cl_lock);
559                 sp = nfs4_insert_state_owner_locked(new);
560                 spin_unlock(&clp->cl_lock);
561         } while (sp == ERR_PTR(-EAGAIN));
562         if (sp != new)
563                 nfs4_free_state_owner(new);
564 out:
565         nfs4_gc_state_owners(server);
566         return sp;
567 }
568
569 /**
570  * nfs4_put_state_owner - Release a nfs4_state_owner
571  * @sp: state owner data to release
572  *
573  * Note that we keep released state owners on an LRU
574  * list.
575  * This caches valid state owners so that they can be
576  * reused, to avoid the OPEN_CONFIRM on minor version 0.
577  * It also pins the uniquifier of dropped state owners for
578  * a while, to ensure that those state owner names are
579  * never reused.
580  */
581 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
582 {
583         struct nfs_server *server = sp->so_server;
584         struct nfs_client *clp = server->nfs_client;
585
586         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
587                 return;
588
589         sp->so_expires = jiffies;
590         list_add_tail(&sp->so_lru, &server->state_owners_lru);
591         spin_unlock(&clp->cl_lock);
592 }
593
594 /**
595  * nfs4_purge_state_owners - Release all cached state owners
596  * @server: nfs_server with cached state owners to release
597  *
598  * Called at umount time.  Remaining state owners will be on
599  * the LRU with ref count of zero.
600  */
601 void nfs4_purge_state_owners(struct nfs_server *server)
602 {
603         struct nfs_client *clp = server->nfs_client;
604         struct nfs4_state_owner *sp, *tmp;
605         LIST_HEAD(doomed);
606
607         spin_lock(&clp->cl_lock);
608         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
609                 list_move(&sp->so_lru, &doomed);
610                 nfs4_remove_state_owner_locked(sp);
611         }
612         spin_unlock(&clp->cl_lock);
613
614         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
615                 list_del(&sp->so_lru);
616                 nfs4_free_state_owner(sp);
617         }
618 }
619
620 static struct nfs4_state *
621 nfs4_alloc_open_state(void)
622 {
623         struct nfs4_state *state;
624
625         state = kzalloc(sizeof(*state), GFP_NOFS);
626         if (!state)
627                 return NULL;
628         atomic_set(&state->count, 1);
629         INIT_LIST_HEAD(&state->lock_states);
630         spin_lock_init(&state->state_lock);
631         seqlock_init(&state->seqlock);
632         return state;
633 }
634
635 void
636 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
637 {
638         if (state->state == fmode)
639                 return;
640         /* NB! List reordering - see the reclaim code for why.  */
641         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
642                 if (fmode & FMODE_WRITE)
643                         list_move(&state->open_states, &state->owner->so_states);
644                 else
645                         list_move_tail(&state->open_states, &state->owner->so_states);
646         }
647         state->state = fmode;
648 }
649
650 static struct nfs4_state *
651 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
652 {
653         struct nfs_inode *nfsi = NFS_I(inode);
654         struct nfs4_state *state;
655
656         list_for_each_entry(state, &nfsi->open_states, inode_states) {
657                 if (state->owner != owner)
658                         continue;
659                 if (!nfs4_valid_open_stateid(state))
660                         continue;
661                 if (atomic_inc_not_zero(&state->count))
662                         return state;
663         }
664         return NULL;
665 }
666
667 static void
668 nfs4_free_open_state(struct nfs4_state *state)
669 {
670         kfree(state);
671 }
672
673 struct nfs4_state *
674 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
675 {
676         struct nfs4_state *state, *new;
677         struct nfs_inode *nfsi = NFS_I(inode);
678
679         spin_lock(&inode->i_lock);
680         state = __nfs4_find_state_byowner(inode, owner);
681         spin_unlock(&inode->i_lock);
682         if (state)
683                 goto out;
684         new = nfs4_alloc_open_state();
685         spin_lock(&owner->so_lock);
686         spin_lock(&inode->i_lock);
687         state = __nfs4_find_state_byowner(inode, owner);
688         if (state == NULL && new != NULL) {
689                 state = new;
690                 state->owner = owner;
691                 atomic_inc(&owner->so_count);
692                 list_add(&state->inode_states, &nfsi->open_states);
693                 ihold(inode);
694                 state->inode = inode;
695                 spin_unlock(&inode->i_lock);
696                 /* Note: The reclaim code dictates that we add stateless
697                  * and read-only stateids to the end of the list */
698                 list_add_tail(&state->open_states, &owner->so_states);
699                 spin_unlock(&owner->so_lock);
700         } else {
701                 spin_unlock(&inode->i_lock);
702                 spin_unlock(&owner->so_lock);
703                 if (new)
704                         nfs4_free_open_state(new);
705         }
706 out:
707         return state;
708 }
709
710 void nfs4_put_open_state(struct nfs4_state *state)
711 {
712         struct inode *inode = state->inode;
713         struct nfs4_state_owner *owner = state->owner;
714
715         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
716                 return;
717         spin_lock(&inode->i_lock);
718         list_del(&state->inode_states);
719         list_del(&state->open_states);
720         spin_unlock(&inode->i_lock);
721         spin_unlock(&owner->so_lock);
722         iput(inode);
723         nfs4_free_open_state(state);
724         nfs4_put_state_owner(owner);
725 }
726
727 /*
728  * Close the current file.
729  */
730 static void __nfs4_close(struct nfs4_state *state,
731                 fmode_t fmode, gfp_t gfp_mask, int wait)
732 {
733         struct nfs4_state_owner *owner = state->owner;
734         int call_close = 0;
735         fmode_t newstate;
736
737         atomic_inc(&owner->so_count);
738         /* Protect against nfs4_find_state() */
739         spin_lock(&owner->so_lock);
740         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
741                 case FMODE_READ:
742                         state->n_rdonly--;
743                         break;
744                 case FMODE_WRITE:
745                         state->n_wronly--;
746                         break;
747                 case FMODE_READ|FMODE_WRITE:
748                         state->n_rdwr--;
749         }
750         newstate = FMODE_READ|FMODE_WRITE;
751         if (state->n_rdwr == 0) {
752                 if (state->n_rdonly == 0) {
753                         newstate &= ~FMODE_READ;
754                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
755                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
756                 }
757                 if (state->n_wronly == 0) {
758                         newstate &= ~FMODE_WRITE;
759                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
760                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
761                 }
762                 if (newstate == 0)
763                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
764         }
765         nfs4_state_set_mode_locked(state, newstate);
766         spin_unlock(&owner->so_lock);
767
768         if (!call_close) {
769                 nfs4_put_open_state(state);
770                 nfs4_put_state_owner(owner);
771         } else
772                 nfs4_do_close(state, gfp_mask, wait);
773 }
774
775 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
776 {
777         __nfs4_close(state, fmode, GFP_NOFS, 0);
778 }
779
780 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
781 {
782         __nfs4_close(state, fmode, GFP_KERNEL, 1);
783 }
784
785 /*
786  * Search the state->lock_states for an existing lock_owner
787  * that is compatible with current->files
788  */
789 static struct nfs4_lock_state *
790 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
791 {
792         struct nfs4_lock_state *pos;
793         list_for_each_entry(pos, &state->lock_states, ls_locks) {
794                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
795                         continue;
796                 switch (pos->ls_owner.lo_type) {
797                 case NFS4_POSIX_LOCK_TYPE:
798                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
799                                 continue;
800                         break;
801                 case NFS4_FLOCK_LOCK_TYPE:
802                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
803                                 continue;
804                 }
805                 atomic_inc(&pos->ls_count);
806                 return pos;
807         }
808         return NULL;
809 }
810
811 /*
812  * Return a compatible lock_state. If no initialized lock_state structure
813  * exists, return an uninitialized one.
814  *
815  */
816 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
817 {
818         struct nfs4_lock_state *lsp;
819         struct nfs_server *server = state->owner->so_server;
820
821         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
822         if (lsp == NULL)
823                 return NULL;
824         nfs4_init_seqid_counter(&lsp->ls_seqid);
825         atomic_set(&lsp->ls_count, 1);
826         lsp->ls_state = state;
827         lsp->ls_owner.lo_type = type;
828         switch (lsp->ls_owner.lo_type) {
829         case NFS4_FLOCK_LOCK_TYPE:
830                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
831                 break;
832         case NFS4_POSIX_LOCK_TYPE:
833                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
834                 break;
835         default:
836                 goto out_free;
837         }
838         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
839         if (lsp->ls_seqid.owner_id < 0)
840                 goto out_free;
841         INIT_LIST_HEAD(&lsp->ls_locks);
842         return lsp;
843 out_free:
844         kfree(lsp);
845         return NULL;
846 }
847
848 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
849 {
850         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
851         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
852         kfree(lsp);
853 }
854
855 /*
856  * Return a compatible lock_state. If no initialized lock_state structure
857  * exists, return an uninitialized one.
858  *
859  */
860 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
861 {
862         struct nfs4_lock_state *lsp, *new = NULL;
863         
864         for(;;) {
865                 spin_lock(&state->state_lock);
866                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
867                 if (lsp != NULL)
868                         break;
869                 if (new != NULL) {
870                         list_add(&new->ls_locks, &state->lock_states);
871                         set_bit(LK_STATE_IN_USE, &state->flags);
872                         lsp = new;
873                         new = NULL;
874                         break;
875                 }
876                 spin_unlock(&state->state_lock);
877                 new = nfs4_alloc_lock_state(state, owner, pid, type);
878                 if (new == NULL)
879                         return NULL;
880         }
881         spin_unlock(&state->state_lock);
882         if (new != NULL)
883                 nfs4_free_lock_state(state->owner->so_server, new);
884         return lsp;
885 }
886
887 /*
888  * Release reference to lock_state, and free it if we see that
889  * it is no longer in use
890  */
891 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
892 {
893         struct nfs_server *server;
894         struct nfs4_state *state;
895
896         if (lsp == NULL)
897                 return;
898         state = lsp->ls_state;
899         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
900                 return;
901         list_del(&lsp->ls_locks);
902         if (list_empty(&state->lock_states))
903                 clear_bit(LK_STATE_IN_USE, &state->flags);
904         spin_unlock(&state->state_lock);
905         server = state->owner->so_server;
906         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
907                 struct nfs_client *clp = server->nfs_client;
908
909                 clp->cl_mvops->free_lock_state(server, lsp);
910         } else
911                 nfs4_free_lock_state(server, lsp);
912 }
913
914 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
915 {
916         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
917
918         dst->fl_u.nfs4_fl.owner = lsp;
919         atomic_inc(&lsp->ls_count);
920 }
921
922 static void nfs4_fl_release_lock(struct file_lock *fl)
923 {
924         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
925 }
926
927 static const struct file_lock_operations nfs4_fl_lock_ops = {
928         .fl_copy_lock = nfs4_fl_copy_lock,
929         .fl_release_private = nfs4_fl_release_lock,
930 };
931
932 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
933 {
934         struct nfs4_lock_state *lsp;
935
936         if (fl->fl_ops != NULL)
937                 return 0;
938         if (fl->fl_flags & FL_POSIX)
939                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
940         else if (fl->fl_flags & FL_FLOCK)
941                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
942                                 NFS4_FLOCK_LOCK_TYPE);
943         else
944                 return -EINVAL;
945         if (lsp == NULL)
946                 return -ENOMEM;
947         fl->fl_u.nfs4_fl.owner = lsp;
948         fl->fl_ops = &nfs4_fl_lock_ops;
949         return 0;
950 }
951
952 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
953                 struct nfs4_state *state,
954                 const struct nfs_lockowner *lockowner)
955 {
956         struct nfs4_lock_state *lsp;
957         fl_owner_t fl_owner;
958         pid_t fl_pid;
959         int ret = -ENOENT;
960
961
962         if (lockowner == NULL)
963                 goto out;
964
965         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
966                 goto out;
967
968         fl_owner = lockowner->l_owner;
969         fl_pid = lockowner->l_pid;
970         spin_lock(&state->state_lock);
971         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
972         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
973                 ret = -EIO;
974         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
975                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
976                 ret = 0;
977                 smp_rmb();
978                 if (!list_empty(&lsp->ls_seqid.list))
979                         ret = -EWOULDBLOCK;
980         }
981         spin_unlock(&state->state_lock);
982         nfs4_put_lock_state(lsp);
983 out:
984         return ret;
985 }
986
987 static int nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
988 {
989         const nfs4_stateid *src;
990         int ret;
991         int seq;
992
993         do {
994                 src = &zero_stateid;
995                 seq = read_seqbegin(&state->seqlock);
996                 if (test_bit(NFS_OPEN_STATE, &state->flags))
997                         src = &state->open_stateid;
998                 nfs4_stateid_copy(dst, src);
999                 ret = 0;
1000                 smp_rmb();
1001                 if (!list_empty(&state->owner->so_seqid.list))
1002                         ret = -EWOULDBLOCK;
1003         } while (read_seqretry(&state->seqlock, seq));
1004         return ret;
1005 }
1006
1007 /*
1008  * Byte-range lock aware utility to initialize the stateid of read/write
1009  * requests.
1010  */
1011 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1012                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1013 {
1014         int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1015         if (ret == -EIO)
1016                 /* A lost lock - don't even consider delegations */
1017                 goto out;
1018         /* returns true if delegation stateid found and copied */
1019         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
1020                 ret = 0;
1021                 goto out;
1022         }
1023         if (ret != -ENOENT)
1024                 /* nfs4_copy_delegation_stateid() didn't over-write
1025                  * dst, so it still has the lock stateid which we now
1026                  * choose to use.
1027                  */
1028                 goto out;
1029         ret = nfs4_copy_open_stateid(dst, state);
1030 out:
1031         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1032                 dst->seqid = 0;
1033         return ret;
1034 }
1035
1036 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1037 {
1038         struct nfs_seqid *new;
1039
1040         new = kmalloc(sizeof(*new), gfp_mask);
1041         if (new != NULL) {
1042                 new->sequence = counter;
1043                 INIT_LIST_HEAD(&new->list);
1044                 new->task = NULL;
1045         }
1046         return new;
1047 }
1048
1049 void nfs_release_seqid(struct nfs_seqid *seqid)
1050 {
1051         struct nfs_seqid_counter *sequence;
1052
1053         if (list_empty(&seqid->list))
1054                 return;
1055         sequence = seqid->sequence;
1056         spin_lock(&sequence->lock);
1057         list_del_init(&seqid->list);
1058         if (!list_empty(&sequence->list)) {
1059                 struct nfs_seqid *next;
1060
1061                 next = list_first_entry(&sequence->list,
1062                                 struct nfs_seqid, list);
1063                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1064         }
1065         spin_unlock(&sequence->lock);
1066 }
1067
1068 void nfs_free_seqid(struct nfs_seqid *seqid)
1069 {
1070         nfs_release_seqid(seqid);
1071         kfree(seqid);
1072 }
1073
1074 /*
1075  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1076  * failed with a seqid incrementing error -
1077  * see comments nfs4.h:seqid_mutating_error()
1078  */
1079 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1080 {
1081         switch (status) {
1082                 case 0:
1083                         break;
1084                 case -NFS4ERR_BAD_SEQID:
1085                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1086                                 return;
1087                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1088                                         " sequence-id error on an"
1089                                         " unconfirmed sequence %p!\n",
1090                                         seqid->sequence);
1091                 case -NFS4ERR_STALE_CLIENTID:
1092                 case -NFS4ERR_STALE_STATEID:
1093                 case -NFS4ERR_BAD_STATEID:
1094                 case -NFS4ERR_BADXDR:
1095                 case -NFS4ERR_RESOURCE:
1096                 case -NFS4ERR_NOFILEHANDLE:
1097                         /* Non-seqid mutating errors */
1098                         return;
1099         };
1100         /*
1101          * Note: no locking needed as we are guaranteed to be first
1102          * on the sequence list
1103          */
1104         seqid->sequence->counter++;
1105 }
1106
1107 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1108 {
1109         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1110                                         struct nfs4_state_owner, so_seqid);
1111         struct nfs_server *server = sp->so_server;
1112
1113         if (status == -NFS4ERR_BAD_SEQID)
1114                 nfs4_drop_state_owner(sp);
1115         if (!nfs4_has_session(server->nfs_client))
1116                 nfs_increment_seqid(status, seqid);
1117 }
1118
1119 /*
1120  * Increment the seqid if the LOCK/LOCKU succeeded, or
1121  * failed with a seqid incrementing error -
1122  * see comments nfs4.h:seqid_mutating_error()
1123  */
1124 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1125 {
1126         nfs_increment_seqid(status, seqid);
1127 }
1128
1129 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1130 {
1131         struct nfs_seqid_counter *sequence = seqid->sequence;
1132         int status = 0;
1133
1134         spin_lock(&sequence->lock);
1135         seqid->task = task;
1136         if (list_empty(&seqid->list))
1137                 list_add_tail(&seqid->list, &sequence->list);
1138         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1139                 goto unlock;
1140         rpc_sleep_on(&sequence->wait, task, NULL);
1141         status = -EAGAIN;
1142 unlock:
1143         spin_unlock(&sequence->lock);
1144         return status;
1145 }
1146
1147 static int nfs4_run_state_manager(void *);
1148
1149 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1150 {
1151         smp_mb__before_clear_bit();
1152         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1153         smp_mb__after_clear_bit();
1154         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1155         rpc_wake_up(&clp->cl_rpcwaitq);
1156 }
1157
1158 /*
1159  * Schedule the nfs_client asynchronous state management routine
1160  */
1161 void nfs4_schedule_state_manager(struct nfs_client *clp)
1162 {
1163         struct task_struct *task;
1164         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1165
1166         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1167                 return;
1168         __module_get(THIS_MODULE);
1169         atomic_inc(&clp->cl_count);
1170
1171         /* The rcu_read_lock() is not strictly necessary, as the state
1172          * manager is the only thread that ever changes the rpc_xprt
1173          * after it's initialized.  At this point, we're single threaded. */
1174         rcu_read_lock();
1175         snprintf(buf, sizeof(buf), "%s-manager",
1176                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1177         rcu_read_unlock();
1178         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1179         if (IS_ERR(task)) {
1180                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1181                         __func__, PTR_ERR(task));
1182                 nfs4_clear_state_manager_bit(clp);
1183                 nfs_put_client(clp);
1184                 module_put(THIS_MODULE);
1185         }
1186 }
1187
1188 /*
1189  * Schedule a lease recovery attempt
1190  */
1191 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1192 {
1193         if (!clp)
1194                 return;
1195         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1196                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1197         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1198                         clp->cl_hostname);
1199         nfs4_schedule_state_manager(clp);
1200 }
1201 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1202
1203 /**
1204  * nfs4_schedule_migration_recovery - trigger migration recovery
1205  *
1206  * @server: FSID that is migrating
1207  *
1208  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1209  * value is returned.
1210  */
1211 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1212 {
1213         struct nfs_client *clp = server->nfs_client;
1214
1215         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1216                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1217                                 clp->cl_hostname);
1218                 return -NFS4ERR_IO;
1219         }
1220
1221         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1222                 return -NFS4ERR_IO;
1223
1224         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1225                         __func__,
1226                         (unsigned long long)server->fsid.major,
1227                         (unsigned long long)server->fsid.minor,
1228                         clp->cl_hostname);
1229
1230         set_bit(NFS_MIG_IN_TRANSITION,
1231                         &((struct nfs_server *)server)->mig_status);
1232         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1233
1234         nfs4_schedule_state_manager(clp);
1235         return 0;
1236 }
1237 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1238
1239 /**
1240  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1241  *
1242  * @clp: server to check for moved leases
1243  *
1244  */
1245 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1246 {
1247         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1248                 __func__, clp->cl_clientid, clp->cl_hostname);
1249
1250         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1251         nfs4_schedule_state_manager(clp);
1252 }
1253 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1254
1255 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1256 {
1257         int res;
1258
1259         might_sleep();
1260
1261         atomic_inc(&clp->cl_count);
1262         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1263                         nfs_wait_bit_killable, TASK_KILLABLE);
1264         if (res)
1265                 goto out;
1266         if (clp->cl_cons_state < 0)
1267                 res = clp->cl_cons_state;
1268 out:
1269         nfs_put_client(clp);
1270         return res;
1271 }
1272
1273 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1274 {
1275         unsigned int loop;
1276         int ret;
1277
1278         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1279                 ret = nfs4_wait_clnt_recover(clp);
1280                 if (ret != 0)
1281                         break;
1282                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1283                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1284                         break;
1285                 nfs4_schedule_state_manager(clp);
1286                 ret = -EIO;
1287         }
1288         return ret;
1289 }
1290
1291 /*
1292  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1293  * @clp: client to process
1294  *
1295  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1296  * resend of the SETCLIENTID and hence re-establish the
1297  * callback channel. Then return all existing delegations.
1298  */
1299 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1300 {
1301         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1302         nfs_expire_all_delegations(clp);
1303         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1304                         clp->cl_hostname);
1305 }
1306
1307 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1308 {
1309         nfs40_handle_cb_pathdown(clp);
1310         nfs4_schedule_state_manager(clp);
1311 }
1312
1313 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1314 {
1315
1316         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1317         /* Don't recover state that expired before the reboot */
1318         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1319                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1320                 return 0;
1321         }
1322         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1323         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1324         return 1;
1325 }
1326
1327 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1328 {
1329         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1330         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1331         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1332         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1333         return 1;
1334 }
1335
1336 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1337 {
1338         struct nfs_client *clp = server->nfs_client;
1339
1340         if (!nfs4_valid_open_stateid(state))
1341                 return -EBADF;
1342         nfs4_state_mark_reclaim_nograce(clp, state);
1343         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1344                         clp->cl_hostname);
1345         nfs4_schedule_state_manager(clp);
1346         return 0;
1347 }
1348 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1349
1350 void nfs_inode_find_state_and_recover(struct inode *inode,
1351                 const nfs4_stateid *stateid)
1352 {
1353         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1354         struct nfs_inode *nfsi = NFS_I(inode);
1355         struct nfs_open_context *ctx;
1356         struct nfs4_state *state;
1357         bool found = false;
1358
1359         spin_lock(&inode->i_lock);
1360         list_for_each_entry(ctx, &nfsi->open_files, list) {
1361                 state = ctx->state;
1362                 if (state == NULL)
1363                         continue;
1364                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1365                         continue;
1366                 if (!nfs4_stateid_match(&state->stateid, stateid))
1367                         continue;
1368                 nfs4_state_mark_reclaim_nograce(clp, state);
1369                 found = true;
1370         }
1371         spin_unlock(&inode->i_lock);
1372         if (found)
1373                 nfs4_schedule_state_manager(clp);
1374 }
1375
1376 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1377 {
1378         struct inode *inode = state->inode;
1379         struct nfs_inode *nfsi = NFS_I(inode);
1380         struct nfs_open_context *ctx;
1381
1382         spin_lock(&inode->i_lock);
1383         list_for_each_entry(ctx, &nfsi->open_files, list) {
1384                 if (ctx->state != state)
1385                         continue;
1386                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1387         }
1388         spin_unlock(&inode->i_lock);
1389 }
1390
1391 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1392 {
1393         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1394         nfs4_state_mark_open_context_bad(state);
1395 }
1396
1397
1398 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1399 {
1400         struct inode *inode = state->inode;
1401         struct nfs_inode *nfsi = NFS_I(inode);
1402         struct file_lock *fl;
1403         int status = 0;
1404
1405         if (inode->i_flock == NULL)
1406                 return 0;
1407
1408         /* Guard against delegation returns and new lock/unlock calls */
1409         down_write(&nfsi->rwsem);
1410         /* Protect inode->i_flock using the BKL */
1411         spin_lock(&inode->i_lock);
1412         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1413                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1414                         continue;
1415                 if (nfs_file_open_context(fl->fl_file)->state != state)
1416                         continue;
1417                 spin_unlock(&inode->i_lock);
1418                 status = ops->recover_lock(state, fl);
1419                 switch (status) {
1420                         case 0:
1421                                 break;
1422                         case -ESTALE:
1423                         case -NFS4ERR_ADMIN_REVOKED:
1424                         case -NFS4ERR_STALE_STATEID:
1425                         case -NFS4ERR_BAD_STATEID:
1426                         case -NFS4ERR_EXPIRED:
1427                         case -NFS4ERR_NO_GRACE:
1428                         case -NFS4ERR_STALE_CLIENTID:
1429                         case -NFS4ERR_BADSESSION:
1430                         case -NFS4ERR_BADSLOT:
1431                         case -NFS4ERR_BAD_HIGH_SLOT:
1432                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1433                                 goto out;
1434                         default:
1435                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1436                                          __func__, status);
1437                         case -ENOMEM:
1438                         case -NFS4ERR_DENIED:
1439                         case -NFS4ERR_RECLAIM_BAD:
1440                         case -NFS4ERR_RECLAIM_CONFLICT:
1441                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1442                                 status = 0;
1443                 }
1444                 spin_lock(&inode->i_lock);
1445         }
1446         spin_unlock(&inode->i_lock);
1447 out:
1448         up_write(&nfsi->rwsem);
1449         return status;
1450 }
1451
1452 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1453 {
1454         struct nfs4_state *state;
1455         struct nfs4_lock_state *lock;
1456         int status = 0;
1457
1458         /* Note: we rely on the sp->so_states list being ordered 
1459          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1460          * states first.
1461          * This is needed to ensure that the server won't give us any
1462          * read delegations that we have to return if, say, we are
1463          * recovering after a network partition or a reboot from a
1464          * server that doesn't support a grace period.
1465          */
1466         spin_lock(&sp->so_lock);
1467         write_seqcount_begin(&sp->so_reclaim_seqcount);
1468 restart:
1469         list_for_each_entry(state, &sp->so_states, open_states) {
1470                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1471                         continue;
1472                 if (!nfs4_valid_open_stateid(state))
1473                         continue;
1474                 if (state->state == 0)
1475                         continue;
1476                 atomic_inc(&state->count);
1477                 spin_unlock(&sp->so_lock);
1478                 status = ops->recover_open(sp, state);
1479                 if (status >= 0) {
1480                         status = nfs4_reclaim_locks(state, ops);
1481                         if (status >= 0) {
1482                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1483                                         spin_lock(&state->state_lock);
1484                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1485                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1486                                                         pr_warn_ratelimited("NFS: "
1487                                                                             "%s: Lock reclaim "
1488                                                                             "failed!\n", __func__);
1489                                         }
1490                                         spin_unlock(&state->state_lock);
1491                                 }
1492                                 nfs4_put_open_state(state);
1493                                 spin_lock(&sp->so_lock);
1494                                 goto restart;
1495                         }
1496                 }
1497                 switch (status) {
1498                         default:
1499                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1500                                         __func__, status);
1501                         case -ENOENT:
1502                         case -ENOMEM:
1503                         case -ESTALE:
1504                                 /* Open state on this file cannot be recovered */
1505                                 nfs4_state_mark_recovery_failed(state, status);
1506                                 break;
1507                         case -EAGAIN:
1508                                 ssleep(1);
1509                         case -NFS4ERR_ADMIN_REVOKED:
1510                         case -NFS4ERR_STALE_STATEID:
1511                         case -NFS4ERR_BAD_STATEID:
1512                         case -NFS4ERR_RECLAIM_BAD:
1513                         case -NFS4ERR_RECLAIM_CONFLICT:
1514                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1515                                 break;
1516                         case -NFS4ERR_EXPIRED:
1517                         case -NFS4ERR_NO_GRACE:
1518                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1519                         case -NFS4ERR_STALE_CLIENTID:
1520                         case -NFS4ERR_BADSESSION:
1521                         case -NFS4ERR_BADSLOT:
1522                         case -NFS4ERR_BAD_HIGH_SLOT:
1523                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1524                                 goto out_err;
1525                 }
1526                 nfs4_put_open_state(state);
1527                 spin_lock(&sp->so_lock);
1528                 goto restart;
1529         }
1530         write_seqcount_end(&sp->so_reclaim_seqcount);
1531         spin_unlock(&sp->so_lock);
1532         return 0;
1533 out_err:
1534         nfs4_put_open_state(state);
1535         spin_lock(&sp->so_lock);
1536         write_seqcount_end(&sp->so_reclaim_seqcount);
1537         spin_unlock(&sp->so_lock);
1538         return status;
1539 }
1540
1541 static void nfs4_clear_open_state(struct nfs4_state *state)
1542 {
1543         struct nfs4_lock_state *lock;
1544
1545         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1546         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1547         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1548         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1549         spin_lock(&state->state_lock);
1550         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1551                 lock->ls_seqid.flags = 0;
1552                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1553         }
1554         spin_unlock(&state->state_lock);
1555 }
1556
1557 static void nfs4_reset_seqids(struct nfs_server *server,
1558         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1559 {
1560         struct nfs_client *clp = server->nfs_client;
1561         struct nfs4_state_owner *sp;
1562         struct rb_node *pos;
1563         struct nfs4_state *state;
1564
1565         spin_lock(&clp->cl_lock);
1566         for (pos = rb_first(&server->state_owners);
1567              pos != NULL;
1568              pos = rb_next(pos)) {
1569                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1570                 sp->so_seqid.flags = 0;
1571                 spin_lock(&sp->so_lock);
1572                 list_for_each_entry(state, &sp->so_states, open_states) {
1573                         if (mark_reclaim(clp, state))
1574                                 nfs4_clear_open_state(state);
1575                 }
1576                 spin_unlock(&sp->so_lock);
1577         }
1578         spin_unlock(&clp->cl_lock);
1579 }
1580
1581 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1582         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1583 {
1584         struct nfs_server *server;
1585
1586         rcu_read_lock();
1587         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1588                 nfs4_reset_seqids(server, mark_reclaim);
1589         rcu_read_unlock();
1590 }
1591
1592 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1593 {
1594         /* Mark all delegations for reclaim */
1595         nfs_delegation_mark_reclaim(clp);
1596         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1597 }
1598
1599 static void nfs4_reclaim_complete(struct nfs_client *clp,
1600                                  const struct nfs4_state_recovery_ops *ops,
1601                                  struct rpc_cred *cred)
1602 {
1603         /* Notify the server we're done reclaiming our state */
1604         if (ops->reclaim_complete)
1605                 (void)ops->reclaim_complete(clp, cred);
1606 }
1607
1608 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1609 {
1610         struct nfs_client *clp = server->nfs_client;
1611         struct nfs4_state_owner *sp;
1612         struct rb_node *pos;
1613         struct nfs4_state *state;
1614
1615         spin_lock(&clp->cl_lock);
1616         for (pos = rb_first(&server->state_owners);
1617              pos != NULL;
1618              pos = rb_next(pos)) {
1619                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1620                 spin_lock(&sp->so_lock);
1621                 list_for_each_entry(state, &sp->so_states, open_states) {
1622                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1623                                                 &state->flags))
1624                                 continue;
1625                         nfs4_state_mark_reclaim_nograce(clp, state);
1626                 }
1627                 spin_unlock(&sp->so_lock);
1628         }
1629         spin_unlock(&clp->cl_lock);
1630 }
1631
1632 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1633 {
1634         struct nfs_server *server;
1635
1636         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1637                 return 0;
1638
1639         rcu_read_lock();
1640         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1641                 nfs4_clear_reclaim_server(server);
1642         rcu_read_unlock();
1643
1644         nfs_delegation_reap_unclaimed(clp);
1645         return 1;
1646 }
1647
1648 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1649 {
1650         const struct nfs4_state_recovery_ops *ops;
1651         struct rpc_cred *cred;
1652
1653         if (!nfs4_state_clear_reclaim_reboot(clp))
1654                 return;
1655         ops = clp->cl_mvops->reboot_recovery_ops;
1656         cred = nfs4_get_clid_cred(clp);
1657         nfs4_reclaim_complete(clp, ops, cred);
1658         put_rpccred(cred);
1659 }
1660
1661 static void nfs_delegation_clear_all(struct nfs_client *clp)
1662 {
1663         nfs_delegation_mark_reclaim(clp);
1664         nfs_delegation_reap_unclaimed(clp);
1665 }
1666
1667 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1668 {
1669         nfs_delegation_clear_all(clp);
1670         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1671 }
1672
1673 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1674 {
1675         switch (error) {
1676                 case 0:
1677                         break;
1678                 case -NFS4ERR_CB_PATH_DOWN:
1679                         nfs40_handle_cb_pathdown(clp);
1680                         break;
1681                 case -NFS4ERR_NO_GRACE:
1682                         nfs4_state_end_reclaim_reboot(clp);
1683                         break;
1684                 case -NFS4ERR_STALE_CLIENTID:
1685                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1686                         nfs4_state_clear_reclaim_reboot(clp);
1687                         nfs4_state_start_reclaim_reboot(clp);
1688                         break;
1689                 case -NFS4ERR_EXPIRED:
1690                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1691                         nfs4_state_start_reclaim_nograce(clp);
1692                         break;
1693                 case -NFS4ERR_BADSESSION:
1694                 case -NFS4ERR_BADSLOT:
1695                 case -NFS4ERR_BAD_HIGH_SLOT:
1696                 case -NFS4ERR_DEADSESSION:
1697                 case -NFS4ERR_SEQ_FALSE_RETRY:
1698                 case -NFS4ERR_SEQ_MISORDERED:
1699                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1700                         /* Zero session reset errors */
1701                         break;
1702                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1703                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1704                         break;
1705                 default:
1706                         dprintk("%s: failed to handle error %d for server %s\n",
1707                                         __func__, error, clp->cl_hostname);
1708                         return error;
1709         }
1710         dprintk("%s: handled error %d for server %s\n", __func__, error,
1711                         clp->cl_hostname);
1712         return 0;
1713 }
1714
1715 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1716 {
1717         struct nfs4_state_owner *sp;
1718         struct nfs_server *server;
1719         struct rb_node *pos;
1720         int status = 0;
1721
1722 restart:
1723         rcu_read_lock();
1724         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1725                 nfs4_purge_state_owners(server);
1726                 spin_lock(&clp->cl_lock);
1727                 for (pos = rb_first(&server->state_owners);
1728                      pos != NULL;
1729                      pos = rb_next(pos)) {
1730                         sp = rb_entry(pos,
1731                                 struct nfs4_state_owner, so_server_node);
1732                         if (!test_and_clear_bit(ops->owner_flag_bit,
1733                                                         &sp->so_flags))
1734                                 continue;
1735                         atomic_inc(&sp->so_count);
1736                         spin_unlock(&clp->cl_lock);
1737                         rcu_read_unlock();
1738
1739                         status = nfs4_reclaim_open_state(sp, ops);
1740                         if (status < 0) {
1741                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1742                                 nfs4_put_state_owner(sp);
1743                                 return nfs4_recovery_handle_error(clp, status);
1744                         }
1745
1746                         nfs4_put_state_owner(sp);
1747                         goto restart;
1748                 }
1749                 spin_unlock(&clp->cl_lock);
1750         }
1751         rcu_read_unlock();
1752         return status;
1753 }
1754
1755 static int nfs4_check_lease(struct nfs_client *clp)
1756 {
1757         struct rpc_cred *cred;
1758         const struct nfs4_state_maintenance_ops *ops =
1759                 clp->cl_mvops->state_renewal_ops;
1760         int status;
1761
1762         /* Is the client already known to have an expired lease? */
1763         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1764                 return 0;
1765         spin_lock(&clp->cl_lock);
1766         cred = ops->get_state_renewal_cred_locked(clp);
1767         spin_unlock(&clp->cl_lock);
1768         if (cred == NULL) {
1769                 cred = nfs4_get_clid_cred(clp);
1770                 status = -ENOKEY;
1771                 if (cred == NULL)
1772                         goto out;
1773         }
1774         status = ops->renew_lease(clp, cred);
1775         put_rpccred(cred);
1776         if (status == -ETIMEDOUT) {
1777                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1778                 return 0;
1779         }
1780 out:
1781         return nfs4_recovery_handle_error(clp, status);
1782 }
1783
1784 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1785  * and for recoverable errors on EXCHANGE_ID for v4.1
1786  */
1787 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1788 {
1789         switch (status) {
1790         case -NFS4ERR_SEQ_MISORDERED:
1791                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1792                         return -ESERVERFAULT;
1793                 /* Lease confirmation error: retry after purging the lease */
1794                 ssleep(1);
1795                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1796                 break;
1797         case -NFS4ERR_STALE_CLIENTID:
1798                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1799                 nfs4_state_clear_reclaim_reboot(clp);
1800                 nfs4_state_start_reclaim_reboot(clp);
1801                 break;
1802         case -NFS4ERR_CLID_INUSE:
1803                 pr_err("NFS: Server %s reports our clientid is in use\n",
1804                         clp->cl_hostname);
1805                 nfs_mark_client_ready(clp, -EPERM);
1806                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1807                 return -EPERM;
1808         case -EACCES:
1809         case -NFS4ERR_DELAY:
1810         case -ETIMEDOUT:
1811         case -EAGAIN:
1812                 ssleep(1);
1813                 break;
1814
1815         case -NFS4ERR_MINOR_VERS_MISMATCH:
1816                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1817                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1818                 dprintk("%s: exit with error %d for server %s\n",
1819                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1820                 return -EPROTONOSUPPORT;
1821         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1822                                  * in nfs4_exchange_id */
1823         default:
1824                 dprintk("%s: exit with error %d for server %s\n", __func__,
1825                                 status, clp->cl_hostname);
1826                 return status;
1827         }
1828         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1829         dprintk("%s: handled error %d for server %s\n", __func__, status,
1830                         clp->cl_hostname);
1831         return 0;
1832 }
1833
1834 static int nfs4_establish_lease(struct nfs_client *clp)
1835 {
1836         struct rpc_cred *cred;
1837         const struct nfs4_state_recovery_ops *ops =
1838                 clp->cl_mvops->reboot_recovery_ops;
1839         int status;
1840
1841         cred = nfs4_get_clid_cred(clp);
1842         if (cred == NULL)
1843                 return -ENOENT;
1844         status = ops->establish_clid(clp, cred);
1845         put_rpccred(cred);
1846         if (status != 0)
1847                 return status;
1848         pnfs_destroy_all_layouts(clp);
1849         return 0;
1850 }
1851
1852 /*
1853  * Returns zero or a negative errno.  NFS4ERR values are converted
1854  * to local errno values.
1855  */
1856 static int nfs4_reclaim_lease(struct nfs_client *clp)
1857 {
1858         int status;
1859
1860         status = nfs4_establish_lease(clp);
1861         if (status < 0)
1862                 return nfs4_handle_reclaim_lease_error(clp, status);
1863         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1864                 nfs4_state_start_reclaim_nograce(clp);
1865         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1866                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1867         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1868         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1869         return 0;
1870 }
1871
1872 static int nfs4_purge_lease(struct nfs_client *clp)
1873 {
1874         int status;
1875
1876         status = nfs4_establish_lease(clp);
1877         if (status < 0)
1878                 return nfs4_handle_reclaim_lease_error(clp, status);
1879         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1880         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1881         nfs4_state_start_reclaim_nograce(clp);
1882         return 0;
1883 }
1884
1885 /*
1886  * Try remote migration of one FSID from a source server to a
1887  * destination server.  The source server provides a list of
1888  * potential destinations.
1889  *
1890  * Returns zero or a negative NFS4ERR status code.
1891  */
1892 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1893 {
1894         struct nfs_client *clp = server->nfs_client;
1895         struct nfs4_fs_locations *locations = NULL;
1896         struct inode *inode;
1897         struct page *page;
1898         int status, result;
1899
1900         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1901                         (unsigned long long)server->fsid.major,
1902                         (unsigned long long)server->fsid.minor,
1903                         clp->cl_hostname);
1904
1905         result = 0;
1906         page = alloc_page(GFP_KERNEL);
1907         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1908         if (page == NULL || locations == NULL) {
1909                 dprintk("<-- %s: no memory\n", __func__);
1910                 goto out;
1911         }
1912
1913         inode = server->super->s_root->d_inode;
1914         result = nfs4_proc_get_locations(inode, locations, page, cred);
1915         if (result) {
1916                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1917                         __func__, result);
1918                 goto out;
1919         }
1920
1921         result = -NFS4ERR_NXIO;
1922         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1923                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1924                         __func__);
1925                 goto out;
1926         }
1927
1928         nfs4_begin_drain_session(clp);
1929
1930         status = nfs4_replace_transport(server, locations);
1931         if (status != 0) {
1932                 dprintk("<-- %s: failed to replace transport: %d\n",
1933                         __func__, status);
1934                 goto out;
1935         }
1936
1937         result = 0;
1938         dprintk("<-- %s: migration succeeded\n", __func__);
1939
1940 out:
1941         if (page != NULL)
1942                 __free_page(page);
1943         kfree(locations);
1944         if (result) {
1945                 pr_err("NFS: migration recovery failed (server %s)\n",
1946                                 clp->cl_hostname);
1947                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1948         }
1949         return result;
1950 }
1951
1952 /*
1953  * Returns zero or a negative NFS4ERR status code.
1954  */
1955 static int nfs4_handle_migration(struct nfs_client *clp)
1956 {
1957         const struct nfs4_state_maintenance_ops *ops =
1958                                 clp->cl_mvops->state_renewal_ops;
1959         struct nfs_server *server;
1960         struct rpc_cred *cred;
1961
1962         dprintk("%s: migration reported on \"%s\"\n", __func__,
1963                         clp->cl_hostname);
1964
1965         spin_lock(&clp->cl_lock);
1966         cred = ops->get_state_renewal_cred_locked(clp);
1967         spin_unlock(&clp->cl_lock);
1968         if (cred == NULL)
1969                 return -NFS4ERR_NOENT;
1970
1971         clp->cl_mig_gen++;
1972 restart:
1973         rcu_read_lock();
1974         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1975                 int status;
1976
1977                 if (server->mig_gen == clp->cl_mig_gen)
1978                         continue;
1979                 server->mig_gen = clp->cl_mig_gen;
1980
1981                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1982                                                 &server->mig_status))
1983                         continue;
1984
1985                 rcu_read_unlock();
1986                 status = nfs4_try_migration(server, cred);
1987                 if (status < 0) {
1988                         put_rpccred(cred);
1989                         return status;
1990                 }
1991                 goto restart;
1992         }
1993         rcu_read_unlock();
1994         put_rpccred(cred);
1995         return 0;
1996 }
1997
1998 /*
1999  * Test each nfs_server on the clp's cl_superblocks list to see
2000  * if it's moved to another server.  Stop when the server no longer
2001  * returns NFS4ERR_LEASE_MOVED.
2002  */
2003 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2004 {
2005         const struct nfs4_state_maintenance_ops *ops =
2006                                 clp->cl_mvops->state_renewal_ops;
2007         struct nfs_server *server;
2008         struct rpc_cred *cred;
2009
2010         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2011                         clp->cl_hostname);
2012
2013         spin_lock(&clp->cl_lock);
2014         cred = ops->get_state_renewal_cred_locked(clp);
2015         spin_unlock(&clp->cl_lock);
2016         if (cred == NULL)
2017                 return -NFS4ERR_NOENT;
2018
2019         clp->cl_mig_gen++;
2020 restart:
2021         rcu_read_lock();
2022         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2023                 struct inode *inode;
2024                 int status;
2025
2026                 if (server->mig_gen == clp->cl_mig_gen)
2027                         continue;
2028                 server->mig_gen = clp->cl_mig_gen;
2029
2030                 rcu_read_unlock();
2031
2032                 inode = server->super->s_root->d_inode;
2033                 status = nfs4_proc_fsid_present(inode, cred);
2034                 if (status != -NFS4ERR_MOVED)
2035                         goto restart;   /* wasn't this one */
2036                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2037                         goto restart;   /* there are more */
2038                 goto out;
2039         }
2040         rcu_read_unlock();
2041
2042 out:
2043         put_rpccred(cred);
2044         return 0;
2045 }
2046
2047 /**
2048  * nfs4_discover_server_trunking - Detect server IP address trunking
2049  *
2050  * @clp: nfs_client under test
2051  * @result: OUT: found nfs_client, or clp
2052  *
2053  * Returns zero or a negative errno.  If zero is returned,
2054  * an nfs_client pointer is planted in "result".
2055  *
2056  * Note: since we are invoked in process context, and
2057  * not from inside the state manager, we cannot use
2058  * nfs4_handle_reclaim_lease_error().
2059  */
2060 int nfs4_discover_server_trunking(struct nfs_client *clp,
2061                                   struct nfs_client **result)
2062 {
2063         const struct nfs4_state_recovery_ops *ops =
2064                                 clp->cl_mvops->reboot_recovery_ops;
2065         struct rpc_clnt *clnt;
2066         struct rpc_cred *cred;
2067         int i, status;
2068
2069         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2070
2071         clnt = clp->cl_rpcclient;
2072         i = 0;
2073
2074         mutex_lock(&nfs_clid_init_mutex);
2075 again:
2076         status  = -ENOENT;
2077         cred = nfs4_get_clid_cred(clp);
2078         if (cred == NULL)
2079                 goto out_unlock;
2080
2081         status = ops->detect_trunking(clp, result, cred);
2082         put_rpccred(cred);
2083         switch (status) {
2084         case 0:
2085                 break;
2086         case -NFS4ERR_DELAY:
2087         case -ETIMEDOUT:
2088         case -EAGAIN:
2089                 ssleep(1);
2090         case -NFS4ERR_STALE_CLIENTID:
2091                 dprintk("NFS: %s after status %d, retrying\n",
2092                         __func__, status);
2093                 goto again;
2094         case -EACCES:
2095                 if (i++ == 0) {
2096                         nfs4_root_machine_cred(clp);
2097                         goto again;
2098                 }
2099                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2100                         break;
2101         case -NFS4ERR_CLID_INUSE:
2102         case -NFS4ERR_WRONGSEC:
2103                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2104                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2105                         status = -EPERM;
2106                         break;
2107                 }
2108                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2109                 if (IS_ERR(clnt)) {
2110                         status = PTR_ERR(clnt);
2111                         break;
2112                 }
2113                 /* Note: this is safe because we haven't yet marked the
2114                  * client as ready, so we are the only user of
2115                  * clp->cl_rpcclient
2116                  */
2117                 clnt = xchg(&clp->cl_rpcclient, clnt);
2118                 rpc_shutdown_client(clnt);
2119                 clnt = clp->cl_rpcclient;
2120                 goto again;
2121
2122         case -NFS4ERR_MINOR_VERS_MISMATCH:
2123                 status = -EPROTONOSUPPORT;
2124                 break;
2125
2126         case -EKEYEXPIRED:
2127         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2128                                  * in nfs4_exchange_id */
2129                 status = -EKEYEXPIRED;
2130                 break;
2131         default:
2132                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2133                                 __func__, status);
2134                 status = -EIO;
2135         }
2136
2137 out_unlock:
2138         mutex_unlock(&nfs_clid_init_mutex);
2139         dprintk("NFS: %s: status = %d\n", __func__, status);
2140         return status;
2141 }
2142
2143 #ifdef CONFIG_NFS_V4_1
2144 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2145 {
2146         struct nfs_client *clp = session->clp;
2147
2148         switch (err) {
2149         default:
2150                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2151                 break;
2152         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2153                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2154         }
2155         nfs4_schedule_lease_recovery(clp);
2156 }
2157 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2158
2159 static void nfs41_ping_server(struct nfs_client *clp)
2160 {
2161         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2162         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2163         nfs4_schedule_state_manager(clp);
2164 }
2165
2166 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2167 {
2168         nfs41_ping_server(clp);
2169 }
2170
2171 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2172 {
2173         nfs41_ping_server(clp);
2174 }
2175
2176 static void nfs4_reset_all_state(struct nfs_client *clp)
2177 {
2178         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2179                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2180                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2181                 nfs4_state_start_reclaim_nograce(clp);
2182                 dprintk("%s: scheduling reset of all state for server %s!\n",
2183                                 __func__, clp->cl_hostname);
2184                 nfs4_schedule_state_manager(clp);
2185         }
2186 }
2187
2188 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2189 {
2190         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2191                 nfs4_state_start_reclaim_reboot(clp);
2192                 dprintk("%s: server %s rebooted!\n", __func__,
2193                                 clp->cl_hostname);
2194                 nfs4_schedule_state_manager(clp);
2195         }
2196 }
2197
2198 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2199 {
2200         nfs4_reset_all_state(clp);
2201         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2202 }
2203
2204 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2205 {
2206         /* This will need to handle layouts too */
2207         nfs_expire_all_delegations(clp);
2208         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2209                         clp->cl_hostname);
2210 }
2211
2212 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2213 {
2214         nfs_expire_all_delegations(clp);
2215         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2216                 nfs4_schedule_state_manager(clp);
2217         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2218                         clp->cl_hostname);
2219 }
2220
2221 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2222 {
2223         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2224                 &clp->cl_state) == 0)
2225                 nfs4_schedule_state_manager(clp);
2226 }
2227
2228 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2229 {
2230         if (!flags)
2231                 return;
2232
2233         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2234                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2235
2236         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2237                 nfs41_handle_server_reboot(clp);
2238         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2239                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2240                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2241                 nfs41_handle_state_revoked(clp);
2242         if (flags & SEQ4_STATUS_LEASE_MOVED)
2243                 nfs4_schedule_lease_moved_recovery(clp);
2244         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2245                 nfs41_handle_recallable_state_revoked(clp);
2246         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2247                 nfs41_handle_backchannel_fault(clp);
2248         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2249                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2250                 nfs41_handle_cb_path_down(clp);
2251 }
2252
2253 static int nfs4_reset_session(struct nfs_client *clp)
2254 {
2255         struct rpc_cred *cred;
2256         int status;
2257
2258         if (!nfs4_has_session(clp))
2259                 return 0;
2260         nfs4_begin_drain_session(clp);
2261         cred = nfs4_get_clid_cred(clp);
2262         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2263         switch (status) {
2264         case 0:
2265         case -NFS4ERR_BADSESSION:
2266         case -NFS4ERR_DEADSESSION:
2267                 break;
2268         case -NFS4ERR_BACK_CHAN_BUSY:
2269         case -NFS4ERR_DELAY:
2270                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2271                 status = 0;
2272                 ssleep(1);
2273                 goto out;
2274         default:
2275                 status = nfs4_recovery_handle_error(clp, status);
2276                 goto out;
2277         }
2278
2279         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2280         status = nfs4_proc_create_session(clp, cred);
2281         if (status) {
2282                 dprintk("%s: session reset failed with status %d for server %s!\n",
2283                         __func__, status, clp->cl_hostname);
2284                 status = nfs4_handle_reclaim_lease_error(clp, status);
2285                 goto out;
2286         }
2287         nfs41_finish_session_reset(clp);
2288         dprintk("%s: session reset was successful for server %s!\n",
2289                         __func__, clp->cl_hostname);
2290 out:
2291         if (cred)
2292                 put_rpccred(cred);
2293         return status;
2294 }
2295
2296 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2297 {
2298         struct rpc_cred *cred;
2299         int ret;
2300
2301         if (!nfs4_has_session(clp))
2302                 return 0;
2303         nfs4_begin_drain_session(clp);
2304         cred = nfs4_get_clid_cred(clp);
2305         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2306         if (cred)
2307                 put_rpccred(cred);
2308         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2309         switch (ret) {
2310         case 0:
2311                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2312                         __func__, clp->cl_hostname);
2313                 break;
2314         case -NFS4ERR_DELAY:
2315                 ssleep(1);
2316                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2317                 break;
2318         default:
2319                 return nfs4_recovery_handle_error(clp, ret);
2320         }
2321         return 0;
2322 }
2323 #else /* CONFIG_NFS_V4_1 */
2324 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2325
2326 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2327 {
2328         return 0;
2329 }
2330 #endif /* CONFIG_NFS_V4_1 */
2331
2332 static void nfs4_state_manager(struct nfs_client *clp)
2333 {
2334         int status = 0;
2335         const char *section = "", *section_sep = "";
2336
2337         /* Ensure exclusive access to NFSv4 state */
2338         do {
2339                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2340                         section = "purge state";
2341                         status = nfs4_purge_lease(clp);
2342                         if (status < 0)
2343                                 goto out_error;
2344                         continue;
2345                 }
2346
2347                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2348                         section = "lease expired";
2349                         /* We're going to have to re-establish a clientid */
2350                         status = nfs4_reclaim_lease(clp);
2351                         if (status < 0)
2352                                 goto out_error;
2353                         continue;
2354                 }
2355
2356                 /* Initialize or reset the session */
2357                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2358                         section = "reset session";
2359                         status = nfs4_reset_session(clp);
2360                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2361                                 continue;
2362                         if (status < 0)
2363                                 goto out_error;
2364                 }
2365
2366                 /* Send BIND_CONN_TO_SESSION */
2367                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2368                                 &clp->cl_state)) {
2369                         section = "bind conn to session";
2370                         status = nfs4_bind_conn_to_session(clp);
2371                         if (status < 0)
2372                                 goto out_error;
2373                         continue;
2374                 }
2375
2376                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2377                         section = "check lease";
2378                         status = nfs4_check_lease(clp);
2379                         if (status < 0)
2380                                 goto out_error;
2381                 }
2382
2383                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2384                         section = "migration";
2385                         status = nfs4_handle_migration(clp);
2386                         if (status < 0)
2387                                 goto out_error;
2388                 }
2389
2390                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2391                         section = "lease moved";
2392                         status = nfs4_handle_lease_moved(clp);
2393                         if (status < 0)
2394                                 goto out_error;
2395                 }
2396
2397                 /* First recover reboot state... */
2398                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2399                         section = "reclaim reboot";
2400                         status = nfs4_do_reclaim(clp,
2401                                 clp->cl_mvops->reboot_recovery_ops);
2402                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2403                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2404                                 continue;
2405                         nfs4_state_end_reclaim_reboot(clp);
2406                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2407                                 continue;
2408                         if (status < 0)
2409                                 goto out_error;
2410                 }
2411
2412                 /* Now recover expired state... */
2413                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2414                         section = "reclaim nograce";
2415                         status = nfs4_do_reclaim(clp,
2416                                 clp->cl_mvops->nograce_recovery_ops);
2417                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2418                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2419                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2420                                 continue;
2421                         if (status < 0)
2422                                 goto out_error;
2423                 }
2424
2425                 nfs4_end_drain_session(clp);
2426                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2427                         nfs_client_return_marked_delegations(clp);
2428                         continue;
2429                 }
2430
2431                 nfs4_clear_state_manager_bit(clp);
2432                 /* Did we race with an attempt to give us more work? */
2433                 if (clp->cl_state == 0)
2434                         break;
2435                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2436                         break;
2437         } while (atomic_read(&clp->cl_count) > 1);
2438         return;
2439 out_error:
2440         if (strlen(section))
2441                 section_sep = ": ";
2442         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2443                         " with error %d\n", section_sep, section,
2444                         clp->cl_hostname, -status);
2445         ssleep(1);
2446         nfs4_end_drain_session(clp);
2447         nfs4_clear_state_manager_bit(clp);
2448 }
2449
2450 static int nfs4_run_state_manager(void *ptr)
2451 {
2452         struct nfs_client *clp = ptr;
2453
2454         allow_signal(SIGKILL);
2455         nfs4_state_manager(clp);
2456         nfs_put_client(clp);
2457         module_put_and_exit(0);
2458         return 0;
2459 }
2460
2461 /*
2462  * Local variables:
2463  *  c-basic-offset: 8
2464  * End:
2465  */