552706d2d7768649d70b9ad924f07af280098f9d
[firefly-linux-kernel-4.4.55.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                 INIT_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         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1019                 goto out;
1020         if (ret != -ENOENT)
1021                 /* nfs4_copy_delegation_stateid() didn't over-write
1022                  * dst, so it still has the lock stateid which we now
1023                  * choose to use.
1024                  */
1025                 goto out;
1026         ret = nfs4_copy_open_stateid(dst, state);
1027 out:
1028         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1029                 dst->seqid = 0;
1030         return ret;
1031 }
1032
1033 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1034 {
1035         struct nfs_seqid *new;
1036
1037         new = kmalloc(sizeof(*new), gfp_mask);
1038         if (new != NULL) {
1039                 new->sequence = counter;
1040                 INIT_LIST_HEAD(&new->list);
1041                 new->task = NULL;
1042         }
1043         return new;
1044 }
1045
1046 void nfs_release_seqid(struct nfs_seqid *seqid)
1047 {
1048         struct nfs_seqid_counter *sequence;
1049
1050         if (list_empty(&seqid->list))
1051                 return;
1052         sequence = seqid->sequence;
1053         spin_lock(&sequence->lock);
1054         list_del_init(&seqid->list);
1055         if (!list_empty(&sequence->list)) {
1056                 struct nfs_seqid *next;
1057
1058                 next = list_first_entry(&sequence->list,
1059                                 struct nfs_seqid, list);
1060                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1061         }
1062         spin_unlock(&sequence->lock);
1063 }
1064
1065 void nfs_free_seqid(struct nfs_seqid *seqid)
1066 {
1067         nfs_release_seqid(seqid);
1068         kfree(seqid);
1069 }
1070
1071 /*
1072  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1073  * failed with a seqid incrementing error -
1074  * see comments nfs_fs.h:seqid_mutating_error()
1075  */
1076 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1077 {
1078         switch (status) {
1079                 case 0:
1080                         break;
1081                 case -NFS4ERR_BAD_SEQID:
1082                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1083                                 return;
1084                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1085                                         " sequence-id error on an"
1086                                         " unconfirmed sequence %p!\n",
1087                                         seqid->sequence);
1088                 case -NFS4ERR_STALE_CLIENTID:
1089                 case -NFS4ERR_STALE_STATEID:
1090                 case -NFS4ERR_BAD_STATEID:
1091                 case -NFS4ERR_BADXDR:
1092                 case -NFS4ERR_RESOURCE:
1093                 case -NFS4ERR_NOFILEHANDLE:
1094                         /* Non-seqid mutating errors */
1095                         return;
1096         };
1097         /*
1098          * Note: no locking needed as we are guaranteed to be first
1099          * on the sequence list
1100          */
1101         seqid->sequence->counter++;
1102 }
1103
1104 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1105 {
1106         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1107                                         struct nfs4_state_owner, so_seqid);
1108         struct nfs_server *server = sp->so_server;
1109
1110         if (status == -NFS4ERR_BAD_SEQID)
1111                 nfs4_drop_state_owner(sp);
1112         if (!nfs4_has_session(server->nfs_client))
1113                 nfs_increment_seqid(status, seqid);
1114 }
1115
1116 /*
1117  * Increment the seqid if the LOCK/LOCKU succeeded, or
1118  * failed with a seqid incrementing error -
1119  * see comments nfs_fs.h:seqid_mutating_error()
1120  */
1121 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1122 {
1123         nfs_increment_seqid(status, seqid);
1124 }
1125
1126 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1127 {
1128         struct nfs_seqid_counter *sequence = seqid->sequence;
1129         int status = 0;
1130
1131         spin_lock(&sequence->lock);
1132         seqid->task = task;
1133         if (list_empty(&seqid->list))
1134                 list_add_tail(&seqid->list, &sequence->list);
1135         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1136                 goto unlock;
1137         rpc_sleep_on(&sequence->wait, task, NULL);
1138         status = -EAGAIN;
1139 unlock:
1140         spin_unlock(&sequence->lock);
1141         return status;
1142 }
1143
1144 static int nfs4_run_state_manager(void *);
1145
1146 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1147 {
1148         smp_mb__before_clear_bit();
1149         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1150         smp_mb__after_clear_bit();
1151         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1152         rpc_wake_up(&clp->cl_rpcwaitq);
1153 }
1154
1155 /*
1156  * Schedule the nfs_client asynchronous state management routine
1157  */
1158 void nfs4_schedule_state_manager(struct nfs_client *clp)
1159 {
1160         struct task_struct *task;
1161         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1162
1163         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1164                 return;
1165         __module_get(THIS_MODULE);
1166         atomic_inc(&clp->cl_count);
1167
1168         /* The rcu_read_lock() is not strictly necessary, as the state
1169          * manager is the only thread that ever changes the rpc_xprt
1170          * after it's initialized.  At this point, we're single threaded. */
1171         rcu_read_lock();
1172         snprintf(buf, sizeof(buf), "%s-manager",
1173                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1174         rcu_read_unlock();
1175         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1176         if (IS_ERR(task)) {
1177                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1178                         __func__, PTR_ERR(task));
1179                 nfs4_clear_state_manager_bit(clp);
1180                 nfs_put_client(clp);
1181                 module_put(THIS_MODULE);
1182         }
1183 }
1184
1185 /*
1186  * Schedule a lease recovery attempt
1187  */
1188 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1189 {
1190         if (!clp)
1191                 return;
1192         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1193                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1194         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1195                         clp->cl_hostname);
1196         nfs4_schedule_state_manager(clp);
1197 }
1198 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1199
1200 /**
1201  * nfs4_schedule_migration_recovery - trigger migration recovery
1202  *
1203  * @server: FSID that is migrating
1204  *
1205  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1206  * value is returned.
1207  */
1208 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1209 {
1210         struct nfs_client *clp = server->nfs_client;
1211
1212         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1213                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1214                                 clp->cl_hostname);
1215                 return -NFS4ERR_IO;
1216         }
1217
1218         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1219                 return -NFS4ERR_IO;
1220
1221         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1222                         __func__,
1223                         (unsigned long long)server->fsid.major,
1224                         (unsigned long long)server->fsid.minor,
1225                         clp->cl_hostname);
1226
1227         set_bit(NFS_MIG_IN_TRANSITION,
1228                         &((struct nfs_server *)server)->mig_status);
1229         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1230
1231         nfs4_schedule_state_manager(clp);
1232         return 0;
1233 }
1234 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1235
1236 /**
1237  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1238  *
1239  * @clp: server to check for moved leases
1240  *
1241  */
1242 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1243 {
1244         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1245                 __func__, clp->cl_clientid, clp->cl_hostname);
1246
1247         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1248         nfs4_schedule_state_manager(clp);
1249 }
1250 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1251
1252 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1253 {
1254         int res;
1255
1256         might_sleep();
1257
1258         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1259                         nfs_wait_bit_killable, TASK_KILLABLE);
1260         if (res)
1261                 return res;
1262
1263         if (clp->cl_cons_state < 0)
1264                 return clp->cl_cons_state;
1265         return 0;
1266 }
1267
1268 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1269 {
1270         unsigned int loop;
1271         int ret;
1272
1273         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1274                 ret = nfs4_wait_clnt_recover(clp);
1275                 if (ret != 0)
1276                         break;
1277                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1278                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1279                         break;
1280                 nfs4_schedule_state_manager(clp);
1281                 ret = -EIO;
1282         }
1283         return ret;
1284 }
1285
1286 /*
1287  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1288  * @clp: client to process
1289  *
1290  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1291  * resend of the SETCLIENTID and hence re-establish the
1292  * callback channel. Then return all existing delegations.
1293  */
1294 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1295 {
1296         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1297         nfs_expire_all_delegations(clp);
1298         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1299                         clp->cl_hostname);
1300 }
1301
1302 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1303 {
1304         nfs40_handle_cb_pathdown(clp);
1305         nfs4_schedule_state_manager(clp);
1306 }
1307
1308 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1309 {
1310
1311         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1312         /* Don't recover state that expired before the reboot */
1313         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1314                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1315                 return 0;
1316         }
1317         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1318         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1319         return 1;
1320 }
1321
1322 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1323 {
1324         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1325         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1326         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1327         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1328         return 1;
1329 }
1330
1331 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1332 {
1333         struct nfs_client *clp = server->nfs_client;
1334
1335         if (!nfs4_valid_open_stateid(state))
1336                 return -EBADF;
1337         nfs4_state_mark_reclaim_nograce(clp, state);
1338         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1339                         clp->cl_hostname);
1340         nfs4_schedule_state_manager(clp);
1341         return 0;
1342 }
1343 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1344
1345 void nfs_inode_find_state_and_recover(struct inode *inode,
1346                 const nfs4_stateid *stateid)
1347 {
1348         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1349         struct nfs_inode *nfsi = NFS_I(inode);
1350         struct nfs_open_context *ctx;
1351         struct nfs4_state *state;
1352         bool found = false;
1353
1354         spin_lock(&inode->i_lock);
1355         list_for_each_entry(ctx, &nfsi->open_files, list) {
1356                 state = ctx->state;
1357                 if (state == NULL)
1358                         continue;
1359                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1360                         continue;
1361                 if (!nfs4_stateid_match(&state->stateid, stateid))
1362                         continue;
1363                 nfs4_state_mark_reclaim_nograce(clp, state);
1364                 found = true;
1365         }
1366         spin_unlock(&inode->i_lock);
1367         if (found)
1368                 nfs4_schedule_state_manager(clp);
1369 }
1370
1371 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1372 {
1373         struct inode *inode = state->inode;
1374         struct nfs_inode *nfsi = NFS_I(inode);
1375         struct nfs_open_context *ctx;
1376
1377         spin_lock(&inode->i_lock);
1378         list_for_each_entry(ctx, &nfsi->open_files, list) {
1379                 if (ctx->state != state)
1380                         continue;
1381                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1382         }
1383         spin_unlock(&inode->i_lock);
1384 }
1385
1386 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1387 {
1388         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1389         nfs4_state_mark_open_context_bad(state);
1390 }
1391
1392
1393 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1394 {
1395         struct inode *inode = state->inode;
1396         struct nfs_inode *nfsi = NFS_I(inode);
1397         struct file_lock *fl;
1398         int status = 0;
1399
1400         if (inode->i_flock == NULL)
1401                 return 0;
1402
1403         /* Guard against delegation returns and new lock/unlock calls */
1404         down_write(&nfsi->rwsem);
1405         /* Protect inode->i_flock using the BKL */
1406         spin_lock(&inode->i_lock);
1407         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1408                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1409                         continue;
1410                 if (nfs_file_open_context(fl->fl_file)->state != state)
1411                         continue;
1412                 spin_unlock(&inode->i_lock);
1413                 status = ops->recover_lock(state, fl);
1414                 switch (status) {
1415                         case 0:
1416                                 break;
1417                         case -ESTALE:
1418                         case -NFS4ERR_ADMIN_REVOKED:
1419                         case -NFS4ERR_STALE_STATEID:
1420                         case -NFS4ERR_BAD_STATEID:
1421                         case -NFS4ERR_EXPIRED:
1422                         case -NFS4ERR_NO_GRACE:
1423                         case -NFS4ERR_STALE_CLIENTID:
1424                         case -NFS4ERR_BADSESSION:
1425                         case -NFS4ERR_BADSLOT:
1426                         case -NFS4ERR_BAD_HIGH_SLOT:
1427                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1428                                 goto out;
1429                         default:
1430                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1431                                          __func__, status);
1432                         case -ENOMEM:
1433                         case -NFS4ERR_DENIED:
1434                         case -NFS4ERR_RECLAIM_BAD:
1435                         case -NFS4ERR_RECLAIM_CONFLICT:
1436                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1437                                 status = 0;
1438                 }
1439                 spin_lock(&inode->i_lock);
1440         }
1441         spin_unlock(&inode->i_lock);
1442 out:
1443         up_write(&nfsi->rwsem);
1444         return status;
1445 }
1446
1447 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1448 {
1449         struct nfs4_state *state;
1450         struct nfs4_lock_state *lock;
1451         int status = 0;
1452
1453         /* Note: we rely on the sp->so_states list being ordered 
1454          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1455          * states first.
1456          * This is needed to ensure that the server won't give us any
1457          * read delegations that we have to return if, say, we are
1458          * recovering after a network partition or a reboot from a
1459          * server that doesn't support a grace period.
1460          */
1461         spin_lock(&sp->so_lock);
1462         write_seqcount_begin(&sp->so_reclaim_seqcount);
1463 restart:
1464         list_for_each_entry(state, &sp->so_states, open_states) {
1465                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1466                         continue;
1467                 if (!nfs4_valid_open_stateid(state))
1468                         continue;
1469                 if (state->state == 0)
1470                         continue;
1471                 atomic_inc(&state->count);
1472                 spin_unlock(&sp->so_lock);
1473                 status = ops->recover_open(sp, state);
1474                 if (status >= 0) {
1475                         status = nfs4_reclaim_locks(state, ops);
1476                         if (status >= 0) {
1477                                 if (test_bit(NFS_DELEGATED_STATE, &state->flags) != 0) {
1478                                         spin_lock(&state->state_lock);
1479                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1480                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1481                                                         pr_warn_ratelimited("NFS: "
1482                                                                             "%s: Lock reclaim "
1483                                                                             "failed!\n", __func__);
1484                                         }
1485                                         spin_unlock(&state->state_lock);
1486                                 }
1487                                 nfs4_put_open_state(state);
1488                                 spin_lock(&sp->so_lock);
1489                                 goto restart;
1490                         }
1491                 }
1492                 switch (status) {
1493                         default:
1494                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1495                                         __func__, status);
1496                         case -ENOENT:
1497                         case -ENOMEM:
1498                         case -ESTALE:
1499                                 /* Open state on this file cannot be recovered */
1500                                 nfs4_state_mark_recovery_failed(state, status);
1501                                 break;
1502                         case -EAGAIN:
1503                                 ssleep(1);
1504                         case -NFS4ERR_ADMIN_REVOKED:
1505                         case -NFS4ERR_STALE_STATEID:
1506                         case -NFS4ERR_BAD_STATEID:
1507                         case -NFS4ERR_RECLAIM_BAD:
1508                         case -NFS4ERR_RECLAIM_CONFLICT:
1509                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1510                                 break;
1511                         case -NFS4ERR_EXPIRED:
1512                         case -NFS4ERR_NO_GRACE:
1513                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1514                         case -NFS4ERR_STALE_CLIENTID:
1515                         case -NFS4ERR_BADSESSION:
1516                         case -NFS4ERR_BADSLOT:
1517                         case -NFS4ERR_BAD_HIGH_SLOT:
1518                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1519                                 goto out_err;
1520                 }
1521                 nfs4_put_open_state(state);
1522                 spin_lock(&sp->so_lock);
1523                 goto restart;
1524         }
1525         write_seqcount_end(&sp->so_reclaim_seqcount);
1526         spin_unlock(&sp->so_lock);
1527         return 0;
1528 out_err:
1529         nfs4_put_open_state(state);
1530         spin_lock(&sp->so_lock);
1531         write_seqcount_end(&sp->so_reclaim_seqcount);
1532         spin_unlock(&sp->so_lock);
1533         return status;
1534 }
1535
1536 static void nfs4_clear_open_state(struct nfs4_state *state)
1537 {
1538         struct nfs4_lock_state *lock;
1539
1540         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1541         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1542         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1543         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1544         spin_lock(&state->state_lock);
1545         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1546                 lock->ls_seqid.flags = 0;
1547                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1548         }
1549         spin_unlock(&state->state_lock);
1550 }
1551
1552 static void nfs4_reset_seqids(struct nfs_server *server,
1553         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1554 {
1555         struct nfs_client *clp = server->nfs_client;
1556         struct nfs4_state_owner *sp;
1557         struct rb_node *pos;
1558         struct nfs4_state *state;
1559
1560         spin_lock(&clp->cl_lock);
1561         for (pos = rb_first(&server->state_owners);
1562              pos != NULL;
1563              pos = rb_next(pos)) {
1564                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1565                 sp->so_seqid.flags = 0;
1566                 spin_lock(&sp->so_lock);
1567                 list_for_each_entry(state, &sp->so_states, open_states) {
1568                         if (mark_reclaim(clp, state))
1569                                 nfs4_clear_open_state(state);
1570                 }
1571                 spin_unlock(&sp->so_lock);
1572         }
1573         spin_unlock(&clp->cl_lock);
1574 }
1575
1576 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1577         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1578 {
1579         struct nfs_server *server;
1580
1581         rcu_read_lock();
1582         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1583                 nfs4_reset_seqids(server, mark_reclaim);
1584         rcu_read_unlock();
1585 }
1586
1587 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1588 {
1589         /* Mark all delegations for reclaim */
1590         nfs_delegation_mark_reclaim(clp);
1591         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1592 }
1593
1594 static void nfs4_reclaim_complete(struct nfs_client *clp,
1595                                  const struct nfs4_state_recovery_ops *ops,
1596                                  struct rpc_cred *cred)
1597 {
1598         /* Notify the server we're done reclaiming our state */
1599         if (ops->reclaim_complete)
1600                 (void)ops->reclaim_complete(clp, cred);
1601 }
1602
1603 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1604 {
1605         struct nfs_client *clp = server->nfs_client;
1606         struct nfs4_state_owner *sp;
1607         struct rb_node *pos;
1608         struct nfs4_state *state;
1609
1610         spin_lock(&clp->cl_lock);
1611         for (pos = rb_first(&server->state_owners);
1612              pos != NULL;
1613              pos = rb_next(pos)) {
1614                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1615                 spin_lock(&sp->so_lock);
1616                 list_for_each_entry(state, &sp->so_states, open_states) {
1617                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1618                                                 &state->flags))
1619                                 continue;
1620                         nfs4_state_mark_reclaim_nograce(clp, state);
1621                 }
1622                 spin_unlock(&sp->so_lock);
1623         }
1624         spin_unlock(&clp->cl_lock);
1625 }
1626
1627 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1628 {
1629         struct nfs_server *server;
1630
1631         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1632                 return 0;
1633
1634         rcu_read_lock();
1635         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1636                 nfs4_clear_reclaim_server(server);
1637         rcu_read_unlock();
1638
1639         nfs_delegation_reap_unclaimed(clp);
1640         return 1;
1641 }
1642
1643 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1644 {
1645         const struct nfs4_state_recovery_ops *ops;
1646         struct rpc_cred *cred;
1647
1648         if (!nfs4_state_clear_reclaim_reboot(clp))
1649                 return;
1650         ops = clp->cl_mvops->reboot_recovery_ops;
1651         cred = nfs4_get_clid_cred(clp);
1652         nfs4_reclaim_complete(clp, ops, cred);
1653         put_rpccred(cred);
1654 }
1655
1656 static void nfs_delegation_clear_all(struct nfs_client *clp)
1657 {
1658         nfs_delegation_mark_reclaim(clp);
1659         nfs_delegation_reap_unclaimed(clp);
1660 }
1661
1662 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1663 {
1664         nfs_delegation_clear_all(clp);
1665         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1666 }
1667
1668 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1669 {
1670         switch (error) {
1671                 case 0:
1672                         break;
1673                 case -NFS4ERR_CB_PATH_DOWN:
1674                         nfs40_handle_cb_pathdown(clp);
1675                         break;
1676                 case -NFS4ERR_NO_GRACE:
1677                         nfs4_state_end_reclaim_reboot(clp);
1678                         break;
1679                 case -NFS4ERR_STALE_CLIENTID:
1680                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1681                         nfs4_state_clear_reclaim_reboot(clp);
1682                         nfs4_state_start_reclaim_reboot(clp);
1683                         break;
1684                 case -NFS4ERR_EXPIRED:
1685                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1686                         nfs4_state_start_reclaim_nograce(clp);
1687                         break;
1688                 case -NFS4ERR_BADSESSION:
1689                 case -NFS4ERR_BADSLOT:
1690                 case -NFS4ERR_BAD_HIGH_SLOT:
1691                 case -NFS4ERR_DEADSESSION:
1692                 case -NFS4ERR_SEQ_FALSE_RETRY:
1693                 case -NFS4ERR_SEQ_MISORDERED:
1694                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1695                         /* Zero session reset errors */
1696                         break;
1697                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1698                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1699                         break;
1700                 default:
1701                         dprintk("%s: failed to handle error %d for server %s\n",
1702                                         __func__, error, clp->cl_hostname);
1703                         return error;
1704         }
1705         dprintk("%s: handled error %d for server %s\n", __func__, error,
1706                         clp->cl_hostname);
1707         return 0;
1708 }
1709
1710 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1711 {
1712         struct nfs4_state_owner *sp;
1713         struct nfs_server *server;
1714         struct rb_node *pos;
1715         int status = 0;
1716
1717 restart:
1718         rcu_read_lock();
1719         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1720                 nfs4_purge_state_owners(server);
1721                 spin_lock(&clp->cl_lock);
1722                 for (pos = rb_first(&server->state_owners);
1723                      pos != NULL;
1724                      pos = rb_next(pos)) {
1725                         sp = rb_entry(pos,
1726                                 struct nfs4_state_owner, so_server_node);
1727                         if (!test_and_clear_bit(ops->owner_flag_bit,
1728                                                         &sp->so_flags))
1729                                 continue;
1730                         atomic_inc(&sp->so_count);
1731                         spin_unlock(&clp->cl_lock);
1732                         rcu_read_unlock();
1733
1734                         status = nfs4_reclaim_open_state(sp, ops);
1735                         if (status < 0) {
1736                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1737                                 nfs4_put_state_owner(sp);
1738                                 return nfs4_recovery_handle_error(clp, status);
1739                         }
1740
1741                         nfs4_put_state_owner(sp);
1742                         goto restart;
1743                 }
1744                 spin_unlock(&clp->cl_lock);
1745         }
1746         rcu_read_unlock();
1747         return status;
1748 }
1749
1750 static int nfs4_check_lease(struct nfs_client *clp)
1751 {
1752         struct rpc_cred *cred;
1753         const struct nfs4_state_maintenance_ops *ops =
1754                 clp->cl_mvops->state_renewal_ops;
1755         int status;
1756
1757         /* Is the client already known to have an expired lease? */
1758         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1759                 return 0;
1760         spin_lock(&clp->cl_lock);
1761         cred = ops->get_state_renewal_cred_locked(clp);
1762         spin_unlock(&clp->cl_lock);
1763         if (cred == NULL) {
1764                 cred = nfs4_get_clid_cred(clp);
1765                 status = -ENOKEY;
1766                 if (cred == NULL)
1767                         goto out;
1768         }
1769         status = ops->renew_lease(clp, cred);
1770         put_rpccred(cred);
1771         if (status == -ETIMEDOUT) {
1772                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1773                 return 0;
1774         }
1775 out:
1776         return nfs4_recovery_handle_error(clp, status);
1777 }
1778
1779 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1780  * and for recoverable errors on EXCHANGE_ID for v4.1
1781  */
1782 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1783 {
1784         switch (status) {
1785         case -NFS4ERR_SEQ_MISORDERED:
1786                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1787                         return -ESERVERFAULT;
1788                 /* Lease confirmation error: retry after purging the lease */
1789                 ssleep(1);
1790                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1791                 break;
1792         case -NFS4ERR_STALE_CLIENTID:
1793                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1794                 nfs4_state_clear_reclaim_reboot(clp);
1795                 nfs4_state_start_reclaim_reboot(clp);
1796                 break;
1797         case -NFS4ERR_CLID_INUSE:
1798                 pr_err("NFS: Server %s reports our clientid is in use\n",
1799                         clp->cl_hostname);
1800                 nfs_mark_client_ready(clp, -EPERM);
1801                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1802                 return -EPERM;
1803         case -EACCES:
1804         case -NFS4ERR_DELAY:
1805         case -ETIMEDOUT:
1806         case -EAGAIN:
1807                 ssleep(1);
1808                 break;
1809
1810         case -NFS4ERR_MINOR_VERS_MISMATCH:
1811                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1812                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1813                 dprintk("%s: exit with error %d for server %s\n",
1814                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1815                 return -EPROTONOSUPPORT;
1816         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1817                                  * in nfs4_exchange_id */
1818         default:
1819                 dprintk("%s: exit with error %d for server %s\n", __func__,
1820                                 status, clp->cl_hostname);
1821                 return status;
1822         }
1823         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1824         dprintk("%s: handled error %d for server %s\n", __func__, status,
1825                         clp->cl_hostname);
1826         return 0;
1827 }
1828
1829 static int nfs4_establish_lease(struct nfs_client *clp)
1830 {
1831         struct rpc_cred *cred;
1832         const struct nfs4_state_recovery_ops *ops =
1833                 clp->cl_mvops->reboot_recovery_ops;
1834         int status;
1835
1836         cred = nfs4_get_clid_cred(clp);
1837         if (cred == NULL)
1838                 return -ENOENT;
1839         status = ops->establish_clid(clp, cred);
1840         put_rpccred(cred);
1841         if (status != 0)
1842                 return status;
1843         pnfs_destroy_all_layouts(clp);
1844         return 0;
1845 }
1846
1847 /*
1848  * Returns zero or a negative errno.  NFS4ERR values are converted
1849  * to local errno values.
1850  */
1851 static int nfs4_reclaim_lease(struct nfs_client *clp)
1852 {
1853         int status;
1854
1855         status = nfs4_establish_lease(clp);
1856         if (status < 0)
1857                 return nfs4_handle_reclaim_lease_error(clp, status);
1858         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1859                 nfs4_state_start_reclaim_nograce(clp);
1860         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1861                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1862         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1863         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1864         return 0;
1865 }
1866
1867 static int nfs4_purge_lease(struct nfs_client *clp)
1868 {
1869         int status;
1870
1871         status = nfs4_establish_lease(clp);
1872         if (status < 0)
1873                 return nfs4_handle_reclaim_lease_error(clp, status);
1874         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1875         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1876         nfs4_state_start_reclaim_nograce(clp);
1877         return 0;
1878 }
1879
1880 /*
1881  * Try remote migration of one FSID from a source server to a
1882  * destination server.  The source server provides a list of
1883  * potential destinations.
1884  *
1885  * Returns zero or a negative NFS4ERR status code.
1886  */
1887 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1888 {
1889         struct nfs_client *clp = server->nfs_client;
1890         struct nfs4_fs_locations *locations = NULL;
1891         struct inode *inode;
1892         struct page *page;
1893         int status, result;
1894
1895         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1896                         (unsigned long long)server->fsid.major,
1897                         (unsigned long long)server->fsid.minor,
1898                         clp->cl_hostname);
1899
1900         result = 0;
1901         page = alloc_page(GFP_KERNEL);
1902         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1903         if (page == NULL || locations == NULL) {
1904                 dprintk("<-- %s: no memory\n", __func__);
1905                 goto out;
1906         }
1907
1908         inode = server->super->s_root->d_inode;
1909         result = nfs4_proc_get_locations(inode, locations, page, cred);
1910         if (result) {
1911                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1912                         __func__, result);
1913                 goto out;
1914         }
1915
1916         result = -NFS4ERR_NXIO;
1917         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1918                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1919                         __func__);
1920                 goto out;
1921         }
1922
1923         nfs4_begin_drain_session(clp);
1924
1925         status = nfs4_replace_transport(server, locations);
1926         if (status != 0) {
1927                 dprintk("<-- %s: failed to replace transport: %d\n",
1928                         __func__, status);
1929                 goto out;
1930         }
1931
1932         result = 0;
1933         dprintk("<-- %s: migration succeeded\n", __func__);
1934
1935 out:
1936         if (page != NULL)
1937                 __free_page(page);
1938         kfree(locations);
1939         if (result) {
1940                 pr_err("NFS: migration recovery failed (server %s)\n",
1941                                 clp->cl_hostname);
1942                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1943         }
1944         return result;
1945 }
1946
1947 /*
1948  * Returns zero or a negative NFS4ERR status code.
1949  */
1950 static int nfs4_handle_migration(struct nfs_client *clp)
1951 {
1952         const struct nfs4_state_maintenance_ops *ops =
1953                                 clp->cl_mvops->state_renewal_ops;
1954         struct nfs_server *server;
1955         struct rpc_cred *cred;
1956
1957         dprintk("%s: migration reported on \"%s\"\n", __func__,
1958                         clp->cl_hostname);
1959
1960         spin_lock(&clp->cl_lock);
1961         cred = ops->get_state_renewal_cred_locked(clp);
1962         spin_unlock(&clp->cl_lock);
1963         if (cred == NULL)
1964                 return -NFS4ERR_NOENT;
1965
1966         clp->cl_mig_gen++;
1967 restart:
1968         rcu_read_lock();
1969         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1970                 int status;
1971
1972                 if (server->mig_gen == clp->cl_mig_gen)
1973                         continue;
1974                 server->mig_gen = clp->cl_mig_gen;
1975
1976                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1977                                                 &server->mig_status))
1978                         continue;
1979
1980                 rcu_read_unlock();
1981                 status = nfs4_try_migration(server, cred);
1982                 if (status < 0) {
1983                         put_rpccred(cred);
1984                         return status;
1985                 }
1986                 goto restart;
1987         }
1988         rcu_read_unlock();
1989         put_rpccred(cred);
1990         return 0;
1991 }
1992
1993 /*
1994  * Test each nfs_server on the clp's cl_superblocks list to see
1995  * if it's moved to another server.  Stop when the server no longer
1996  * returns NFS4ERR_LEASE_MOVED.
1997  */
1998 static int nfs4_handle_lease_moved(struct nfs_client *clp)
1999 {
2000         const struct nfs4_state_maintenance_ops *ops =
2001                                 clp->cl_mvops->state_renewal_ops;
2002         struct nfs_server *server;
2003         struct rpc_cred *cred;
2004
2005         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2006                         clp->cl_hostname);
2007
2008         spin_lock(&clp->cl_lock);
2009         cred = ops->get_state_renewal_cred_locked(clp);
2010         spin_unlock(&clp->cl_lock);
2011         if (cred == NULL)
2012                 return -NFS4ERR_NOENT;
2013
2014         clp->cl_mig_gen++;
2015 restart:
2016         rcu_read_lock();
2017         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2018                 struct inode *inode;
2019                 int status;
2020
2021                 if (server->mig_gen == clp->cl_mig_gen)
2022                         continue;
2023                 server->mig_gen = clp->cl_mig_gen;
2024
2025                 rcu_read_unlock();
2026
2027                 inode = server->super->s_root->d_inode;
2028                 status = nfs4_proc_fsid_present(inode, cred);
2029                 if (status != -NFS4ERR_MOVED)
2030                         goto restart;   /* wasn't this one */
2031                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2032                         goto restart;   /* there are more */
2033                 goto out;
2034         }
2035         rcu_read_unlock();
2036
2037 out:
2038         put_rpccred(cred);
2039         return 0;
2040 }
2041
2042 /**
2043  * nfs4_discover_server_trunking - Detect server IP address trunking
2044  *
2045  * @clp: nfs_client under test
2046  * @result: OUT: found nfs_client, or clp
2047  *
2048  * Returns zero or a negative errno.  If zero is returned,
2049  * an nfs_client pointer is planted in "result".
2050  *
2051  * Note: since we are invoked in process context, and
2052  * not from inside the state manager, we cannot use
2053  * nfs4_handle_reclaim_lease_error().
2054  */
2055 int nfs4_discover_server_trunking(struct nfs_client *clp,
2056                                   struct nfs_client **result)
2057 {
2058         const struct nfs4_state_recovery_ops *ops =
2059                                 clp->cl_mvops->reboot_recovery_ops;
2060         struct rpc_clnt *clnt;
2061         struct rpc_cred *cred;
2062         int i, status;
2063
2064         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2065
2066         clnt = clp->cl_rpcclient;
2067         i = 0;
2068
2069         mutex_lock(&nfs_clid_init_mutex);
2070 again:
2071         status  = -ENOENT;
2072         cred = nfs4_get_clid_cred(clp);
2073         if (cred == NULL)
2074                 goto out_unlock;
2075
2076         status = ops->detect_trunking(clp, result, cred);
2077         put_rpccred(cred);
2078         switch (status) {
2079         case 0:
2080                 break;
2081         case -NFS4ERR_DELAY:
2082         case -ETIMEDOUT:
2083         case -EAGAIN:
2084                 ssleep(1);
2085         case -NFS4ERR_STALE_CLIENTID:
2086                 dprintk("NFS: %s after status %d, retrying\n",
2087                         __func__, status);
2088                 goto again;
2089         case -EACCES:
2090                 if (i++ == 0) {
2091                         nfs4_root_machine_cred(clp);
2092                         goto again;
2093                 }
2094                 if (i > 2)
2095                         break;
2096         case -NFS4ERR_CLID_INUSE:
2097         case -NFS4ERR_WRONGSEC:
2098                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2099                 if (IS_ERR(clnt)) {
2100                         status = PTR_ERR(clnt);
2101                         break;
2102                 }
2103                 /* Note: this is safe because we haven't yet marked the
2104                  * client as ready, so we are the only user of
2105                  * clp->cl_rpcclient
2106                  */
2107                 clnt = xchg(&clp->cl_rpcclient, clnt);
2108                 rpc_shutdown_client(clnt);
2109                 clnt = clp->cl_rpcclient;
2110                 goto again;
2111
2112         case -NFS4ERR_MINOR_VERS_MISMATCH:
2113                 status = -EPROTONOSUPPORT;
2114                 break;
2115
2116         case -EKEYEXPIRED:
2117         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2118                                  * in nfs4_exchange_id */
2119                 status = -EKEYEXPIRED;
2120                 break;
2121         default:
2122                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2123                                 __func__, status);
2124                 status = -EIO;
2125         }
2126
2127 out_unlock:
2128         mutex_unlock(&nfs_clid_init_mutex);
2129         dprintk("NFS: %s: status = %d\n", __func__, status);
2130         return status;
2131 }
2132
2133 #ifdef CONFIG_NFS_V4_1
2134 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2135 {
2136         struct nfs_client *clp = session->clp;
2137
2138         switch (err) {
2139         default:
2140                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2141                 break;
2142         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2143                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2144         }
2145         nfs4_schedule_lease_recovery(clp);
2146 }
2147 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2148
2149 static void nfs41_ping_server(struct nfs_client *clp)
2150 {
2151         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2152         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2153         nfs4_schedule_state_manager(clp);
2154 }
2155
2156 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2157 {
2158         nfs41_ping_server(clp);
2159 }
2160
2161 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2162 {
2163         nfs41_ping_server(clp);
2164 }
2165
2166 static void nfs4_reset_all_state(struct nfs_client *clp)
2167 {
2168         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2169                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2170                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2171                 nfs4_state_start_reclaim_nograce(clp);
2172                 dprintk("%s: scheduling reset of all state for server %s!\n",
2173                                 __func__, clp->cl_hostname);
2174                 nfs4_schedule_state_manager(clp);
2175         }
2176 }
2177
2178 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2179 {
2180         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2181                 nfs4_state_start_reclaim_reboot(clp);
2182                 dprintk("%s: server %s rebooted!\n", __func__,
2183                                 clp->cl_hostname);
2184                 nfs4_schedule_state_manager(clp);
2185         }
2186 }
2187
2188 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2189 {
2190         nfs4_reset_all_state(clp);
2191         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2192 }
2193
2194 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2195 {
2196         /* This will need to handle layouts too */
2197         nfs_expire_all_delegations(clp);
2198         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2199                         clp->cl_hostname);
2200 }
2201
2202 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2203 {
2204         nfs_expire_all_delegations(clp);
2205         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2206                 nfs4_schedule_state_manager(clp);
2207         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2208                         clp->cl_hostname);
2209 }
2210
2211 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2212 {
2213         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2214                 &clp->cl_state) == 0)
2215                 nfs4_schedule_state_manager(clp);
2216 }
2217
2218 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2219 {
2220         if (!flags)
2221                 return;
2222
2223         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2224                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2225
2226         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2227                 nfs41_handle_server_reboot(clp);
2228         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2229                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2230                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
2231                             SEQ4_STATUS_LEASE_MOVED))
2232                 nfs41_handle_state_revoked(clp);
2233         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2234                 nfs41_handle_recallable_state_revoked(clp);
2235         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2236                 nfs41_handle_backchannel_fault(clp);
2237         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2238                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2239                 nfs41_handle_cb_path_down(clp);
2240 }
2241
2242 static int nfs4_reset_session(struct nfs_client *clp)
2243 {
2244         struct rpc_cred *cred;
2245         int status;
2246
2247         if (!nfs4_has_session(clp))
2248                 return 0;
2249         nfs4_begin_drain_session(clp);
2250         cred = nfs4_get_clid_cred(clp);
2251         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2252         switch (status) {
2253         case 0:
2254         case -NFS4ERR_BADSESSION:
2255         case -NFS4ERR_DEADSESSION:
2256                 break;
2257         case -NFS4ERR_BACK_CHAN_BUSY:
2258         case -NFS4ERR_DELAY:
2259                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2260                 status = 0;
2261                 ssleep(1);
2262                 goto out;
2263         default:
2264                 status = nfs4_recovery_handle_error(clp, status);
2265                 goto out;
2266         }
2267
2268         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2269         status = nfs4_proc_create_session(clp, cred);
2270         if (status) {
2271                 dprintk("%s: session reset failed with status %d for server %s!\n",
2272                         __func__, status, clp->cl_hostname);
2273                 status = nfs4_handle_reclaim_lease_error(clp, status);
2274                 goto out;
2275         }
2276         nfs41_finish_session_reset(clp);
2277         dprintk("%s: session reset was successful for server %s!\n",
2278                         __func__, clp->cl_hostname);
2279 out:
2280         if (cred)
2281                 put_rpccred(cred);
2282         return status;
2283 }
2284
2285 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2286 {
2287         struct rpc_cred *cred;
2288         int ret;
2289
2290         if (!nfs4_has_session(clp))
2291                 return 0;
2292         nfs4_begin_drain_session(clp);
2293         cred = nfs4_get_clid_cred(clp);
2294         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2295         if (cred)
2296                 put_rpccred(cred);
2297         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2298         switch (ret) {
2299         case 0:
2300                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2301                         __func__, clp->cl_hostname);
2302                 break;
2303         case -NFS4ERR_DELAY:
2304                 ssleep(1);
2305                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2306                 break;
2307         default:
2308                 return nfs4_recovery_handle_error(clp, ret);
2309         }
2310         return 0;
2311 }
2312 #else /* CONFIG_NFS_V4_1 */
2313 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2314
2315 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2316 {
2317         return 0;
2318 }
2319 #endif /* CONFIG_NFS_V4_1 */
2320
2321 static void nfs4_state_manager(struct nfs_client *clp)
2322 {
2323         int status = 0;
2324         const char *section = "", *section_sep = "";
2325
2326         /* Ensure exclusive access to NFSv4 state */
2327         do {
2328                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2329                         section = "purge state";
2330                         status = nfs4_purge_lease(clp);
2331                         if (status < 0)
2332                                 goto out_error;
2333                         continue;
2334                 }
2335
2336                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2337                         section = "lease expired";
2338                         /* We're going to have to re-establish a clientid */
2339                         status = nfs4_reclaim_lease(clp);
2340                         if (status < 0)
2341                                 goto out_error;
2342                         continue;
2343                 }
2344
2345                 /* Initialize or reset the session */
2346                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2347                         section = "reset session";
2348                         status = nfs4_reset_session(clp);
2349                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2350                                 continue;
2351                         if (status < 0)
2352                                 goto out_error;
2353                 }
2354
2355                 /* Send BIND_CONN_TO_SESSION */
2356                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2357                                 &clp->cl_state)) {
2358                         section = "bind conn to session";
2359                         status = nfs4_bind_conn_to_session(clp);
2360                         if (status < 0)
2361                                 goto out_error;
2362                         continue;
2363                 }
2364
2365                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2366                         section = "check lease";
2367                         status = nfs4_check_lease(clp);
2368                         if (status < 0)
2369                                 goto out_error;
2370                 }
2371
2372                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2373                         section = "migration";
2374                         status = nfs4_handle_migration(clp);
2375                         if (status < 0)
2376                                 goto out_error;
2377                 }
2378
2379                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2380                         section = "lease moved";
2381                         status = nfs4_handle_lease_moved(clp);
2382                         if (status < 0)
2383                                 goto out_error;
2384                 }
2385
2386                 /* First recover reboot state... */
2387                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2388                         section = "reclaim reboot";
2389                         status = nfs4_do_reclaim(clp,
2390                                 clp->cl_mvops->reboot_recovery_ops);
2391                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2392                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2393                                 continue;
2394                         nfs4_state_end_reclaim_reboot(clp);
2395                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2396                                 continue;
2397                         if (status < 0)
2398                                 goto out_error;
2399                 }
2400
2401                 /* Now recover expired state... */
2402                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2403                         section = "reclaim nograce";
2404                         status = nfs4_do_reclaim(clp,
2405                                 clp->cl_mvops->nograce_recovery_ops);
2406                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2407                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2408                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2409                                 continue;
2410                         if (status < 0)
2411                                 goto out_error;
2412                 }
2413
2414                 nfs4_end_drain_session(clp);
2415                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2416                         nfs_client_return_marked_delegations(clp);
2417                         continue;
2418                 }
2419
2420                 nfs4_clear_state_manager_bit(clp);
2421                 /* Did we race with an attempt to give us more work? */
2422                 if (clp->cl_state == 0)
2423                         break;
2424                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2425                         break;
2426         } while (atomic_read(&clp->cl_count) > 1);
2427         return;
2428 out_error:
2429         if (strlen(section))
2430                 section_sep = ": ";
2431         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2432                         " with error %d\n", section_sep, section,
2433                         clp->cl_hostname, -status);
2434         ssleep(1);
2435         nfs4_end_drain_session(clp);
2436         nfs4_clear_state_manager_bit(clp);
2437 }
2438
2439 static int nfs4_run_state_manager(void *ptr)
2440 {
2441         struct nfs_client *clp = ptr;
2442
2443         allow_signal(SIGKILL);
2444         nfs4_state_manager(clp);
2445         nfs_put_client(clp);
2446         module_put_and_exit(0);
2447         return 0;
2448 }
2449
2450 /*
2451  * Local variables:
2452  *  c-basic-offset: 8
2453  * End:
2454  */