]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - fs/nfs/nfs4state.c
Merge tag 'pci-v3.17-changes-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[karo-tx-linux.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/ratelimit.h>
50 #include <linux/workqueue.h>
51 #include <linux/bitops.h>
52 #include <linux/jiffies.h>
53
54 #include <linux/sunrpc/clnt.h>
55
56 #include "nfs4_fs.h"
57 #include "callback.h"
58 #include "delegation.h"
59 #include "internal.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
63
64 #define NFSDBG_FACILITY         NFSDBG_STATE
65
66 #define OPENOWNER_POOL_SIZE     8
67
68 const nfs4_stateid zero_stateid;
69 static DEFINE_MUTEX(nfs_clid_init_mutex);
70
71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
72 {
73         struct nfs4_setclientid_res clid = {
74                 .clientid = clp->cl_clientid,
75                 .confirm = clp->cl_confirm,
76         };
77         unsigned short port;
78         int status;
79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
80
81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
82                 goto do_confirm;
83         port = nn->nfs_callback_tcpport;
84         if (clp->cl_addr.ss_family == AF_INET6)
85                 port = nn->nfs_callback_tcpport6;
86
87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
88         if (status != 0)
89                 goto out;
90         clp->cl_clientid = clid.clientid;
91         clp->cl_confirm = clid.confirm;
92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
93 do_confirm:
94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
95         if (status != 0)
96                 goto out;
97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
98         nfs4_schedule_state_renewal(clp);
99 out:
100         return status;
101 }
102
103 /**
104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
105  *
106  * @clp: nfs_client under test
107  * @result: OUT: found nfs_client, or clp
108  * @cred: credential to use for trunking test
109  *
110  * Returns zero, a negative errno, or a negative NFS4ERR status.
111  * If zero is returned, an nfs_client pointer is planted in
112  * "result".
113  *
114  * Note: The returned client may not yet be marked ready.
115  */
116 int nfs40_discover_server_trunking(struct nfs_client *clp,
117                                    struct nfs_client **result,
118                                    struct rpc_cred *cred)
119 {
120         struct nfs4_setclientid_res clid = {
121                 .clientid = clp->cl_clientid,
122                 .confirm = clp->cl_confirm,
123         };
124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
125         unsigned short port;
126         int status;
127
128         port = nn->nfs_callback_tcpport;
129         if (clp->cl_addr.ss_family == AF_INET6)
130                 port = nn->nfs_callback_tcpport6;
131
132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
133         if (status != 0)
134                 goto out;
135         clp->cl_clientid = clid.clientid;
136         clp->cl_confirm = clid.confirm;
137
138         status = nfs40_walk_client_list(clp, result, cred);
139         if (status == 0) {
140                 /* Sustain the lease, even if it's empty.  If the clientid4
141                  * goes stale it's of no use for trunking discovery. */
142                 nfs4_schedule_state_renewal(*result);
143         }
144 out:
145         return status;
146 }
147
148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
149 {
150         struct rpc_cred *cred = NULL;
151
152         if (clp->cl_machine_cred != NULL)
153                 cred = get_rpccred(clp->cl_machine_cred);
154         return cred;
155 }
156
157 static void nfs4_root_machine_cred(struct nfs_client *clp)
158 {
159         struct rpc_cred *cred, *new;
160
161         new = rpc_lookup_machine_cred(NULL);
162         spin_lock(&clp->cl_lock);
163         cred = clp->cl_machine_cred;
164         clp->cl_machine_cred = new;
165         spin_unlock(&clp->cl_lock);
166         if (cred != NULL)
167                 put_rpccred(cred);
168 }
169
170 static struct rpc_cred *
171 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
172 {
173         struct rpc_cred *cred = NULL;
174         struct nfs4_state_owner *sp;
175         struct rb_node *pos;
176
177         for (pos = rb_first(&server->state_owners);
178              pos != NULL;
179              pos = rb_next(pos)) {
180                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
181                 if (list_empty(&sp->so_states))
182                         continue;
183                 cred = get_rpccred(sp->so_cred);
184                 break;
185         }
186         return cred;
187 }
188
189 /**
190  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
191  * @clp: client state handle
192  *
193  * Returns an rpc_cred with reference count bumped, or NULL.
194  * Caller must hold clp->cl_lock.
195  */
196 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
197 {
198         struct rpc_cred *cred = NULL;
199         struct nfs_server *server;
200
201         /* Use machine credentials if available */
202         cred = nfs4_get_machine_cred_locked(clp);
203         if (cred != NULL)
204                 goto out;
205
206         rcu_read_lock();
207         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
208                 cred = nfs4_get_renew_cred_server_locked(server);
209                 if (cred != NULL)
210                         break;
211         }
212         rcu_read_unlock();
213
214 out:
215         return cred;
216 }
217
218 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
219 {
220         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
221                 spin_lock(&tbl->slot_tbl_lock);
222                 nfs41_wake_slot_table(tbl);
223                 spin_unlock(&tbl->slot_tbl_lock);
224         }
225 }
226
227 static void nfs4_end_drain_session(struct nfs_client *clp)
228 {
229         struct nfs4_session *ses = clp->cl_session;
230
231         if (clp->cl_slot_tbl) {
232                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
233                 return;
234         }
235
236         if (ses != NULL) {
237                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
238                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
239         }
240 }
241
242 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
243 {
244         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
245         spin_lock(&tbl->slot_tbl_lock);
246         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
247                 reinit_completion(&tbl->complete);
248                 spin_unlock(&tbl->slot_tbl_lock);
249                 return wait_for_completion_interruptible(&tbl->complete);
250         }
251         spin_unlock(&tbl->slot_tbl_lock);
252         return 0;
253 }
254
255 static int nfs4_begin_drain_session(struct nfs_client *clp)
256 {
257         struct nfs4_session *ses = clp->cl_session;
258         int ret = 0;
259
260         if (clp->cl_slot_tbl)
261                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
262
263         /* back channel */
264         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
265         if (ret)
266                 return ret;
267         /* fore channel */
268         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
269 }
270
271 #if defined(CONFIG_NFS_V4_1)
272
273 static int nfs41_setup_state_renewal(struct nfs_client *clp)
274 {
275         int status;
276         struct nfs_fsinfo fsinfo;
277
278         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
279                 nfs4_schedule_state_renewal(clp);
280                 return 0;
281         }
282
283         status = nfs4_proc_get_lease_time(clp, &fsinfo);
284         if (status == 0) {
285                 /* Update lease time and schedule renewal */
286                 spin_lock(&clp->cl_lock);
287                 clp->cl_lease_time = fsinfo.lease_time * HZ;
288                 clp->cl_last_renewal = jiffies;
289                 spin_unlock(&clp->cl_lock);
290
291                 nfs4_schedule_state_renewal(clp);
292         }
293
294         return status;
295 }
296
297 static void nfs41_finish_session_reset(struct nfs_client *clp)
298 {
299         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301         /* create_session negotiated new slot table */
302         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303         nfs41_setup_state_renewal(clp);
304 }
305
306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
307 {
308         int status;
309
310         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311                 goto do_confirm;
312         nfs4_begin_drain_session(clp);
313         status = nfs4_proc_exchange_id(clp, cred);
314         if (status != 0)
315                 goto out;
316         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
317 do_confirm:
318         status = nfs4_proc_create_session(clp, cred);
319         if (status != 0)
320                 goto out;
321         nfs41_finish_session_reset(clp);
322         nfs_mark_client_ready(clp, NFS_CS_READY);
323 out:
324         return status;
325 }
326
327 /**
328  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
329  *
330  * @clp: nfs_client under test
331  * @result: OUT: found nfs_client, or clp
332  * @cred: credential to use for trunking test
333  *
334  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
335  * If NFS4_OK is returned, an nfs_client pointer is planted in
336  * "result".
337  *
338  * Note: The returned client may not yet be marked ready.
339  */
340 int nfs41_discover_server_trunking(struct nfs_client *clp,
341                                    struct nfs_client **result,
342                                    struct rpc_cred *cred)
343 {
344         int status;
345
346         status = nfs4_proc_exchange_id(clp, cred);
347         if (status != NFS4_OK)
348                 return status;
349         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
350
351         return nfs41_walk_client_list(clp, result, cred);
352 }
353
354 #endif /* CONFIG_NFS_V4_1 */
355
356 /**
357  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
358  * @clp: client state handle
359  *
360  * Returns an rpc_cred with reference count bumped, or NULL.
361  */
362 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
363 {
364         struct rpc_cred *cred;
365
366         spin_lock(&clp->cl_lock);
367         cred = nfs4_get_machine_cred_locked(clp);
368         spin_unlock(&clp->cl_lock);
369         return cred;
370 }
371
372 static struct nfs4_state_owner *
373 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
374 {
375         struct rb_node **p = &server->state_owners.rb_node,
376                        *parent = NULL;
377         struct nfs4_state_owner *sp;
378
379         while (*p != NULL) {
380                 parent = *p;
381                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
382
383                 if (cred < sp->so_cred)
384                         p = &parent->rb_left;
385                 else if (cred > sp->so_cred)
386                         p = &parent->rb_right;
387                 else {
388                         if (!list_empty(&sp->so_lru))
389                                 list_del_init(&sp->so_lru);
390                         atomic_inc(&sp->so_count);
391                         return sp;
392                 }
393         }
394         return NULL;
395 }
396
397 static struct nfs4_state_owner *
398 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
399 {
400         struct nfs_server *server = new->so_server;
401         struct rb_node **p = &server->state_owners.rb_node,
402                        *parent = NULL;
403         struct nfs4_state_owner *sp;
404         int err;
405
406         while (*p != NULL) {
407                 parent = *p;
408                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
409
410                 if (new->so_cred < sp->so_cred)
411                         p = &parent->rb_left;
412                 else if (new->so_cred > sp->so_cred)
413                         p = &parent->rb_right;
414                 else {
415                         if (!list_empty(&sp->so_lru))
416                                 list_del_init(&sp->so_lru);
417                         atomic_inc(&sp->so_count);
418                         return sp;
419                 }
420         }
421         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
422         if (err)
423                 return ERR_PTR(err);
424         rb_link_node(&new->so_server_node, parent, p);
425         rb_insert_color(&new->so_server_node, &server->state_owners);
426         return new;
427 }
428
429 static void
430 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
431 {
432         struct nfs_server *server = sp->so_server;
433
434         if (!RB_EMPTY_NODE(&sp->so_server_node))
435                 rb_erase(&sp->so_server_node, &server->state_owners);
436         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
437 }
438
439 static void
440 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
441 {
442         sc->create_time = ktime_get();
443         sc->flags = 0;
444         sc->counter = 0;
445         spin_lock_init(&sc->lock);
446         INIT_LIST_HEAD(&sc->list);
447         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
448 }
449
450 static void
451 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
452 {
453         rpc_destroy_wait_queue(&sc->wait);
454 }
455
456 /*
457  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
458  * create a new state_owner.
459  *
460  */
461 static struct nfs4_state_owner *
462 nfs4_alloc_state_owner(struct nfs_server *server,
463                 struct rpc_cred *cred,
464                 gfp_t gfp_flags)
465 {
466         struct nfs4_state_owner *sp;
467
468         sp = kzalloc(sizeof(*sp), gfp_flags);
469         if (!sp)
470                 return NULL;
471         sp->so_server = server;
472         sp->so_cred = get_rpccred(cred);
473         spin_lock_init(&sp->so_lock);
474         INIT_LIST_HEAD(&sp->so_states);
475         nfs4_init_seqid_counter(&sp->so_seqid);
476         atomic_set(&sp->so_count, 1);
477         INIT_LIST_HEAD(&sp->so_lru);
478         seqcount_init(&sp->so_reclaim_seqcount);
479         mutex_init(&sp->so_delegreturn_mutex);
480         return sp;
481 }
482
483 static void
484 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
485 {
486         struct rb_node *rb_node = &sp->so_server_node;
487
488         if (!RB_EMPTY_NODE(rb_node)) {
489                 struct nfs_server *server = sp->so_server;
490                 struct nfs_client *clp = server->nfs_client;
491
492                 spin_lock(&clp->cl_lock);
493                 if (!RB_EMPTY_NODE(rb_node)) {
494                         rb_erase(rb_node, &server->state_owners);
495                         RB_CLEAR_NODE(rb_node);
496                 }
497                 spin_unlock(&clp->cl_lock);
498         }
499 }
500
501 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
502 {
503         nfs4_destroy_seqid_counter(&sp->so_seqid);
504         put_rpccred(sp->so_cred);
505         kfree(sp);
506 }
507
508 static void nfs4_gc_state_owners(struct nfs_server *server)
509 {
510         struct nfs_client *clp = server->nfs_client;
511         struct nfs4_state_owner *sp, *tmp;
512         unsigned long time_min, time_max;
513         LIST_HEAD(doomed);
514
515         spin_lock(&clp->cl_lock);
516         time_max = jiffies;
517         time_min = (long)time_max - (long)clp->cl_lease_time;
518         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
519                 /* NB: LRU is sorted so that oldest is at the head */
520                 if (time_in_range(sp->so_expires, time_min, time_max))
521                         break;
522                 list_move(&sp->so_lru, &doomed);
523                 nfs4_remove_state_owner_locked(sp);
524         }
525         spin_unlock(&clp->cl_lock);
526
527         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
528                 list_del(&sp->so_lru);
529                 nfs4_free_state_owner(sp);
530         }
531 }
532
533 /**
534  * nfs4_get_state_owner - Look up a state owner given a credential
535  * @server: nfs_server to search
536  * @cred: RPC credential to match
537  *
538  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
539  */
540 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
541                                               struct rpc_cred *cred,
542                                               gfp_t gfp_flags)
543 {
544         struct nfs_client *clp = server->nfs_client;
545         struct nfs4_state_owner *sp, *new;
546
547         spin_lock(&clp->cl_lock);
548         sp = nfs4_find_state_owner_locked(server, cred);
549         spin_unlock(&clp->cl_lock);
550         if (sp != NULL)
551                 goto out;
552         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
553         if (new == NULL)
554                 goto out;
555         do {
556                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
557                         break;
558                 spin_lock(&clp->cl_lock);
559                 sp = nfs4_insert_state_owner_locked(new);
560                 spin_unlock(&clp->cl_lock);
561         } while (sp == ERR_PTR(-EAGAIN));
562         if (sp != new)
563                 nfs4_free_state_owner(new);
564 out:
565         nfs4_gc_state_owners(server);
566         return sp;
567 }
568
569 /**
570  * nfs4_put_state_owner - Release a nfs4_state_owner
571  * @sp: state owner data to release
572  *
573  * Note that we keep released state owners on an LRU
574  * list.
575  * This caches valid state owners so that they can be
576  * reused, to avoid the OPEN_CONFIRM on minor version 0.
577  * It also pins the uniquifier of dropped state owners for
578  * a while, to ensure that those state owner names are
579  * never reused.
580  */
581 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
582 {
583         struct nfs_server *server = sp->so_server;
584         struct nfs_client *clp = server->nfs_client;
585
586         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
587                 return;
588
589         sp->so_expires = jiffies;
590         list_add_tail(&sp->so_lru, &server->state_owners_lru);
591         spin_unlock(&clp->cl_lock);
592 }
593
594 /**
595  * nfs4_purge_state_owners - Release all cached state owners
596  * @server: nfs_server with cached state owners to release
597  *
598  * Called at umount time.  Remaining state owners will be on
599  * the LRU with ref count of zero.
600  */
601 void nfs4_purge_state_owners(struct nfs_server *server)
602 {
603         struct nfs_client *clp = server->nfs_client;
604         struct nfs4_state_owner *sp, *tmp;
605         LIST_HEAD(doomed);
606
607         spin_lock(&clp->cl_lock);
608         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
609                 list_move(&sp->so_lru, &doomed);
610                 nfs4_remove_state_owner_locked(sp);
611         }
612         spin_unlock(&clp->cl_lock);
613
614         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
615                 list_del(&sp->so_lru);
616                 nfs4_free_state_owner(sp);
617         }
618 }
619
620 static struct nfs4_state *
621 nfs4_alloc_open_state(void)
622 {
623         struct nfs4_state *state;
624
625         state = kzalloc(sizeof(*state), GFP_NOFS);
626         if (!state)
627                 return NULL;
628         atomic_set(&state->count, 1);
629         INIT_LIST_HEAD(&state->lock_states);
630         spin_lock_init(&state->state_lock);
631         seqlock_init(&state->seqlock);
632         return state;
633 }
634
635 void
636 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
637 {
638         if (state->state == fmode)
639                 return;
640         /* NB! List reordering - see the reclaim code for why.  */
641         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
642                 if (fmode & FMODE_WRITE)
643                         list_move(&state->open_states, &state->owner->so_states);
644                 else
645                         list_move_tail(&state->open_states, &state->owner->so_states);
646         }
647         state->state = fmode;
648 }
649
650 static struct nfs4_state *
651 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
652 {
653         struct nfs_inode *nfsi = NFS_I(inode);
654         struct nfs4_state *state;
655
656         list_for_each_entry(state, &nfsi->open_states, inode_states) {
657                 if (state->owner != owner)
658                         continue;
659                 if (!nfs4_valid_open_stateid(state))
660                         continue;
661                 if (atomic_inc_not_zero(&state->count))
662                         return state;
663         }
664         return NULL;
665 }
666
667 static void
668 nfs4_free_open_state(struct nfs4_state *state)
669 {
670         kfree(state);
671 }
672
673 struct nfs4_state *
674 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
675 {
676         struct nfs4_state *state, *new;
677         struct nfs_inode *nfsi = NFS_I(inode);
678
679         spin_lock(&inode->i_lock);
680         state = __nfs4_find_state_byowner(inode, owner);
681         spin_unlock(&inode->i_lock);
682         if (state)
683                 goto out;
684         new = nfs4_alloc_open_state();
685         spin_lock(&owner->so_lock);
686         spin_lock(&inode->i_lock);
687         state = __nfs4_find_state_byowner(inode, owner);
688         if (state == NULL && new != NULL) {
689                 state = new;
690                 state->owner = owner;
691                 atomic_inc(&owner->so_count);
692                 list_add(&state->inode_states, &nfsi->open_states);
693                 ihold(inode);
694                 state->inode = inode;
695                 spin_unlock(&inode->i_lock);
696                 /* Note: The reclaim code dictates that we add stateless
697                  * and read-only stateids to the end of the list */
698                 list_add_tail(&state->open_states, &owner->so_states);
699                 spin_unlock(&owner->so_lock);
700         } else {
701                 spin_unlock(&inode->i_lock);
702                 spin_unlock(&owner->so_lock);
703                 if (new)
704                         nfs4_free_open_state(new);
705         }
706 out:
707         return state;
708 }
709
710 void nfs4_put_open_state(struct nfs4_state *state)
711 {
712         struct inode *inode = state->inode;
713         struct nfs4_state_owner *owner = state->owner;
714
715         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
716                 return;
717         spin_lock(&inode->i_lock);
718         list_del(&state->inode_states);
719         list_del(&state->open_states);
720         spin_unlock(&inode->i_lock);
721         spin_unlock(&owner->so_lock);
722         iput(inode);
723         nfs4_free_open_state(state);
724         nfs4_put_state_owner(owner);
725 }
726
727 /*
728  * Close the current file.
729  */
730 static void __nfs4_close(struct nfs4_state *state,
731                 fmode_t fmode, gfp_t gfp_mask, int wait)
732 {
733         struct nfs4_state_owner *owner = state->owner;
734         int call_close = 0;
735         fmode_t newstate;
736
737         atomic_inc(&owner->so_count);
738         /* Protect against nfs4_find_state() */
739         spin_lock(&owner->so_lock);
740         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
741                 case FMODE_READ:
742                         state->n_rdonly--;
743                         break;
744                 case FMODE_WRITE:
745                         state->n_wronly--;
746                         break;
747                 case FMODE_READ|FMODE_WRITE:
748                         state->n_rdwr--;
749         }
750         newstate = FMODE_READ|FMODE_WRITE;
751         if (state->n_rdwr == 0) {
752                 if (state->n_rdonly == 0) {
753                         newstate &= ~FMODE_READ;
754                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
755                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
756                 }
757                 if (state->n_wronly == 0) {
758                         newstate &= ~FMODE_WRITE;
759                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
760                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
761                 }
762                 if (newstate == 0)
763                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
764         }
765         nfs4_state_set_mode_locked(state, newstate);
766         spin_unlock(&owner->so_lock);
767
768         if (!call_close) {
769                 nfs4_put_open_state(state);
770                 nfs4_put_state_owner(owner);
771         } else
772                 nfs4_do_close(state, gfp_mask, wait);
773 }
774
775 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
776 {
777         __nfs4_close(state, fmode, GFP_NOFS, 0);
778 }
779
780 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
781 {
782         __nfs4_close(state, fmode, GFP_KERNEL, 1);
783 }
784
785 /*
786  * Search the state->lock_states for an existing lock_owner
787  * that is compatible with current->files
788  */
789 static struct nfs4_lock_state *
790 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
791 {
792         struct nfs4_lock_state *pos;
793         list_for_each_entry(pos, &state->lock_states, ls_locks) {
794                 if (pos->ls_owner != fl_owner)
795                         continue;
796                 atomic_inc(&pos->ls_count);
797                 return pos;
798         }
799         return NULL;
800 }
801
802 static void
803 free_lock_state_work(struct work_struct *work)
804 {
805         struct nfs4_lock_state *lsp = container_of(work,
806                                         struct nfs4_lock_state, ls_release);
807         struct nfs4_state *state = lsp->ls_state;
808         struct nfs_server *server = state->owner->so_server;
809         struct nfs_client *clp = server->nfs_client;
810
811         clp->cl_mvops->free_lock_state(server, lsp);
812 }
813
814 /*
815  * Return a compatible lock_state. If no initialized lock_state structure
816  * exists, return an uninitialized one.
817  *
818  */
819 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
820 {
821         struct nfs4_lock_state *lsp;
822         struct nfs_server *server = state->owner->so_server;
823
824         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
825         if (lsp == NULL)
826                 return NULL;
827         nfs4_init_seqid_counter(&lsp->ls_seqid);
828         atomic_set(&lsp->ls_count, 1);
829         lsp->ls_state = state;
830         lsp->ls_owner = fl_owner;
831         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
832         if (lsp->ls_seqid.owner_id < 0)
833                 goto out_free;
834         INIT_LIST_HEAD(&lsp->ls_locks);
835         INIT_WORK(&lsp->ls_release, free_lock_state_work);
836         return lsp;
837 out_free:
838         kfree(lsp);
839         return NULL;
840 }
841
842 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
843 {
844         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
845         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
846         kfree(lsp);
847 }
848
849 /*
850  * Return a compatible lock_state. If no initialized lock_state structure
851  * exists, return an uninitialized one.
852  *
853  */
854 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
855 {
856         struct nfs4_lock_state *lsp, *new = NULL;
857         
858         for(;;) {
859                 spin_lock(&state->state_lock);
860                 lsp = __nfs4_find_lock_state(state, owner);
861                 if (lsp != NULL)
862                         break;
863                 if (new != NULL) {
864                         list_add(&new->ls_locks, &state->lock_states);
865                         set_bit(LK_STATE_IN_USE, &state->flags);
866                         lsp = new;
867                         new = NULL;
868                         break;
869                 }
870                 spin_unlock(&state->state_lock);
871                 new = nfs4_alloc_lock_state(state, owner);
872                 if (new == NULL)
873                         return NULL;
874         }
875         spin_unlock(&state->state_lock);
876         if (new != NULL)
877                 nfs4_free_lock_state(state->owner->so_server, new);
878         return lsp;
879 }
880
881 /*
882  * Release reference to lock_state, and free it if we see that
883  * it is no longer in use
884  */
885 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
886 {
887         struct nfs_server *server;
888         struct nfs4_state *state;
889
890         if (lsp == NULL)
891                 return;
892         state = lsp->ls_state;
893         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
894                 return;
895         list_del(&lsp->ls_locks);
896         if (list_empty(&state->lock_states))
897                 clear_bit(LK_STATE_IN_USE, &state->flags);
898         spin_unlock(&state->state_lock);
899         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags))
900                 queue_work(nfsiod_workqueue, &lsp->ls_release);
901         else {
902                 server = state->owner->so_server;
903                 nfs4_free_lock_state(server, lsp);
904         }
905 }
906
907 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
908 {
909         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
910
911         dst->fl_u.nfs4_fl.owner = lsp;
912         atomic_inc(&lsp->ls_count);
913 }
914
915 static void nfs4_fl_release_lock(struct file_lock *fl)
916 {
917         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
918 }
919
920 static const struct file_lock_operations nfs4_fl_lock_ops = {
921         .fl_copy_lock = nfs4_fl_copy_lock,
922         .fl_release_private = nfs4_fl_release_lock,
923 };
924
925 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
926 {
927         struct nfs4_lock_state *lsp;
928
929         if (fl->fl_ops != NULL)
930                 return 0;
931         lsp = nfs4_get_lock_state(state, fl->fl_owner);
932         if (lsp == NULL)
933                 return -ENOMEM;
934         fl->fl_u.nfs4_fl.owner = lsp;
935         fl->fl_ops = &nfs4_fl_lock_ops;
936         return 0;
937 }
938
939 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
940                 struct nfs4_state *state,
941                 const struct nfs_lockowner *lockowner)
942 {
943         struct nfs4_lock_state *lsp;
944         fl_owner_t fl_owner;
945         int ret = -ENOENT;
946
947
948         if (lockowner == NULL)
949                 goto out;
950
951         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
952                 goto out;
953
954         fl_owner = lockowner->l_owner;
955         spin_lock(&state->state_lock);
956         lsp = __nfs4_find_lock_state(state, fl_owner);
957         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
958                 ret = -EIO;
959         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
960                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
961                 ret = 0;
962         }
963         spin_unlock(&state->state_lock);
964         nfs4_put_lock_state(lsp);
965 out:
966         return ret;
967 }
968
969 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
970 {
971         const nfs4_stateid *src;
972         int seq;
973
974         do {
975                 src = &zero_stateid;
976                 seq = read_seqbegin(&state->seqlock);
977                 if (test_bit(NFS_OPEN_STATE, &state->flags))
978                         src = &state->open_stateid;
979                 nfs4_stateid_copy(dst, src);
980         } while (read_seqretry(&state->seqlock, seq));
981 }
982
983 /*
984  * Byte-range lock aware utility to initialize the stateid of read/write
985  * requests.
986  */
987 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
988                 fmode_t fmode, const struct nfs_lockowner *lockowner)
989 {
990         int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
991         if (ret == -EIO)
992                 /* A lost lock - don't even consider delegations */
993                 goto out;
994         /* returns true if delegation stateid found and copied */
995         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
996                 ret = 0;
997                 goto out;
998         }
999         if (ret != -ENOENT)
1000                 /* nfs4_copy_delegation_stateid() didn't over-write
1001                  * dst, so it still has the lock stateid which we now
1002                  * choose to use.
1003                  */
1004                 goto out;
1005         nfs4_copy_open_stateid(dst, state);
1006         ret = 0;
1007 out:
1008         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1009                 dst->seqid = 0;
1010         return ret;
1011 }
1012
1013 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1014 {
1015         struct nfs_seqid *new;
1016
1017         new = kmalloc(sizeof(*new), gfp_mask);
1018         if (new != NULL) {
1019                 new->sequence = counter;
1020                 INIT_LIST_HEAD(&new->list);
1021                 new->task = NULL;
1022         }
1023         return new;
1024 }
1025
1026 void nfs_release_seqid(struct nfs_seqid *seqid)
1027 {
1028         struct nfs_seqid_counter *sequence;
1029
1030         if (list_empty(&seqid->list))
1031                 return;
1032         sequence = seqid->sequence;
1033         spin_lock(&sequence->lock);
1034         list_del_init(&seqid->list);
1035         if (!list_empty(&sequence->list)) {
1036                 struct nfs_seqid *next;
1037
1038                 next = list_first_entry(&sequence->list,
1039                                 struct nfs_seqid, list);
1040                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1041         }
1042         spin_unlock(&sequence->lock);
1043 }
1044
1045 void nfs_free_seqid(struct nfs_seqid *seqid)
1046 {
1047         nfs_release_seqid(seqid);
1048         kfree(seqid);
1049 }
1050
1051 /*
1052  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1053  * failed with a seqid incrementing error -
1054  * see comments nfs4.h:seqid_mutating_error()
1055  */
1056 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1057 {
1058         switch (status) {
1059                 case 0:
1060                         break;
1061                 case -NFS4ERR_BAD_SEQID:
1062                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1063                                 return;
1064                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1065                                         " sequence-id error on an"
1066                                         " unconfirmed sequence %p!\n",
1067                                         seqid->sequence);
1068                 case -NFS4ERR_STALE_CLIENTID:
1069                 case -NFS4ERR_STALE_STATEID:
1070                 case -NFS4ERR_BAD_STATEID:
1071                 case -NFS4ERR_BADXDR:
1072                 case -NFS4ERR_RESOURCE:
1073                 case -NFS4ERR_NOFILEHANDLE:
1074                         /* Non-seqid mutating errors */
1075                         return;
1076         };
1077         /*
1078          * Note: no locking needed as we are guaranteed to be first
1079          * on the sequence list
1080          */
1081         seqid->sequence->counter++;
1082 }
1083
1084 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1085 {
1086         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1087                                         struct nfs4_state_owner, so_seqid);
1088         struct nfs_server *server = sp->so_server;
1089
1090         if (status == -NFS4ERR_BAD_SEQID)
1091                 nfs4_drop_state_owner(sp);
1092         if (!nfs4_has_session(server->nfs_client))
1093                 nfs_increment_seqid(status, seqid);
1094 }
1095
1096 /*
1097  * Increment the seqid if the LOCK/LOCKU succeeded, or
1098  * failed with a seqid incrementing error -
1099  * see comments nfs4.h:seqid_mutating_error()
1100  */
1101 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1102 {
1103         nfs_increment_seqid(status, seqid);
1104 }
1105
1106 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1107 {
1108         struct nfs_seqid_counter *sequence = seqid->sequence;
1109         int status = 0;
1110
1111         spin_lock(&sequence->lock);
1112         seqid->task = task;
1113         if (list_empty(&seqid->list))
1114                 list_add_tail(&seqid->list, &sequence->list);
1115         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1116                 goto unlock;
1117         rpc_sleep_on(&sequence->wait, task, NULL);
1118         status = -EAGAIN;
1119 unlock:
1120         spin_unlock(&sequence->lock);
1121         return status;
1122 }
1123
1124 static int nfs4_run_state_manager(void *);
1125
1126 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1127 {
1128         smp_mb__before_atomic();
1129         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1130         smp_mb__after_atomic();
1131         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1132         rpc_wake_up(&clp->cl_rpcwaitq);
1133 }
1134
1135 /*
1136  * Schedule the nfs_client asynchronous state management routine
1137  */
1138 void nfs4_schedule_state_manager(struct nfs_client *clp)
1139 {
1140         struct task_struct *task;
1141         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1142
1143         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1144                 return;
1145         __module_get(THIS_MODULE);
1146         atomic_inc(&clp->cl_count);
1147
1148         /* The rcu_read_lock() is not strictly necessary, as the state
1149          * manager is the only thread that ever changes the rpc_xprt
1150          * after it's initialized.  At this point, we're single threaded. */
1151         rcu_read_lock();
1152         snprintf(buf, sizeof(buf), "%s-manager",
1153                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1154         rcu_read_unlock();
1155         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1156         if (IS_ERR(task)) {
1157                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1158                         __func__, PTR_ERR(task));
1159                 nfs4_clear_state_manager_bit(clp);
1160                 nfs_put_client(clp);
1161                 module_put(THIS_MODULE);
1162         }
1163 }
1164
1165 /*
1166  * Schedule a lease recovery attempt
1167  */
1168 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1169 {
1170         if (!clp)
1171                 return;
1172         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1173                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1174         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1175                         clp->cl_hostname);
1176         nfs4_schedule_state_manager(clp);
1177 }
1178 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1179
1180 /**
1181  * nfs4_schedule_migration_recovery - trigger migration recovery
1182  *
1183  * @server: FSID that is migrating
1184  *
1185  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1186  * value is returned.
1187  */
1188 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1189 {
1190         struct nfs_client *clp = server->nfs_client;
1191
1192         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1193                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1194                                 clp->cl_hostname);
1195                 return -NFS4ERR_IO;
1196         }
1197
1198         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1199                 return -NFS4ERR_IO;
1200
1201         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1202                         __func__,
1203                         (unsigned long long)server->fsid.major,
1204                         (unsigned long long)server->fsid.minor,
1205                         clp->cl_hostname);
1206
1207         set_bit(NFS_MIG_IN_TRANSITION,
1208                         &((struct nfs_server *)server)->mig_status);
1209         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1210
1211         nfs4_schedule_state_manager(clp);
1212         return 0;
1213 }
1214 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1215
1216 /**
1217  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1218  *
1219  * @clp: server to check for moved leases
1220  *
1221  */
1222 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1223 {
1224         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1225                 __func__, clp->cl_clientid, clp->cl_hostname);
1226
1227         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1228         nfs4_schedule_state_manager(clp);
1229 }
1230 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1231
1232 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1233 {
1234         int res;
1235
1236         might_sleep();
1237
1238         atomic_inc(&clp->cl_count);
1239         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1240                                  nfs_wait_bit_killable, TASK_KILLABLE);
1241         if (res)
1242                 goto out;
1243         if (clp->cl_cons_state < 0)
1244                 res = clp->cl_cons_state;
1245 out:
1246         nfs_put_client(clp);
1247         return res;
1248 }
1249
1250 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1251 {
1252         unsigned int loop;
1253         int ret;
1254
1255         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1256                 ret = nfs4_wait_clnt_recover(clp);
1257                 if (ret != 0)
1258                         break;
1259                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1260                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1261                         break;
1262                 nfs4_schedule_state_manager(clp);
1263                 ret = -EIO;
1264         }
1265         return ret;
1266 }
1267
1268 /*
1269  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1270  * @clp: client to process
1271  *
1272  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1273  * resend of the SETCLIENTID and hence re-establish the
1274  * callback channel. Then return all existing delegations.
1275  */
1276 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1277 {
1278         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1279         nfs_expire_all_delegations(clp);
1280         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1281                         clp->cl_hostname);
1282 }
1283
1284 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1285 {
1286         nfs40_handle_cb_pathdown(clp);
1287         nfs4_schedule_state_manager(clp);
1288 }
1289
1290 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1291 {
1292
1293         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1294         /* Don't recover state that expired before the reboot */
1295         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1296                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1297                 return 0;
1298         }
1299         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1300         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1301         return 1;
1302 }
1303
1304 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1305 {
1306         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1307         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1308         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1309         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1310         return 1;
1311 }
1312
1313 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1314 {
1315         struct nfs_client *clp = server->nfs_client;
1316
1317         if (!nfs4_valid_open_stateid(state))
1318                 return -EBADF;
1319         nfs4_state_mark_reclaim_nograce(clp, state);
1320         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1321                         clp->cl_hostname);
1322         nfs4_schedule_state_manager(clp);
1323         return 0;
1324 }
1325 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1326
1327 void nfs_inode_find_state_and_recover(struct inode *inode,
1328                 const nfs4_stateid *stateid)
1329 {
1330         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1331         struct nfs_inode *nfsi = NFS_I(inode);
1332         struct nfs_open_context *ctx;
1333         struct nfs4_state *state;
1334         bool found = false;
1335
1336         spin_lock(&inode->i_lock);
1337         list_for_each_entry(ctx, &nfsi->open_files, list) {
1338                 state = ctx->state;
1339                 if (state == NULL)
1340                         continue;
1341                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1342                         continue;
1343                 if (!nfs4_stateid_match(&state->stateid, stateid))
1344                         continue;
1345                 nfs4_state_mark_reclaim_nograce(clp, state);
1346                 found = true;
1347         }
1348         spin_unlock(&inode->i_lock);
1349         if (found)
1350                 nfs4_schedule_state_manager(clp);
1351 }
1352
1353 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1354 {
1355         struct inode *inode = state->inode;
1356         struct nfs_inode *nfsi = NFS_I(inode);
1357         struct nfs_open_context *ctx;
1358
1359         spin_lock(&inode->i_lock);
1360         list_for_each_entry(ctx, &nfsi->open_files, list) {
1361                 if (ctx->state != state)
1362                         continue;
1363                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1364         }
1365         spin_unlock(&inode->i_lock);
1366 }
1367
1368 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1369 {
1370         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1371         nfs4_state_mark_open_context_bad(state);
1372 }
1373
1374
1375 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1376 {
1377         struct inode *inode = state->inode;
1378         struct nfs_inode *nfsi = NFS_I(inode);
1379         struct file_lock *fl;
1380         int status = 0;
1381
1382         if (inode->i_flock == NULL)
1383                 return 0;
1384
1385         /* Guard against delegation returns and new lock/unlock calls */
1386         down_write(&nfsi->rwsem);
1387         /* Protect inode->i_flock using the BKL */
1388         spin_lock(&inode->i_lock);
1389         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1390                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1391                         continue;
1392                 if (nfs_file_open_context(fl->fl_file)->state != state)
1393                         continue;
1394                 spin_unlock(&inode->i_lock);
1395                 status = ops->recover_lock(state, fl);
1396                 switch (status) {
1397                         case 0:
1398                                 break;
1399                         case -ESTALE:
1400                         case -NFS4ERR_ADMIN_REVOKED:
1401                         case -NFS4ERR_STALE_STATEID:
1402                         case -NFS4ERR_BAD_STATEID:
1403                         case -NFS4ERR_EXPIRED:
1404                         case -NFS4ERR_NO_GRACE:
1405                         case -NFS4ERR_STALE_CLIENTID:
1406                         case -NFS4ERR_BADSESSION:
1407                         case -NFS4ERR_BADSLOT:
1408                         case -NFS4ERR_BAD_HIGH_SLOT:
1409                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1410                                 goto out;
1411                         default:
1412                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1413                                          __func__, status);
1414                         case -ENOMEM:
1415                         case -NFS4ERR_DENIED:
1416                         case -NFS4ERR_RECLAIM_BAD:
1417                         case -NFS4ERR_RECLAIM_CONFLICT:
1418                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1419                                 status = 0;
1420                 }
1421                 spin_lock(&inode->i_lock);
1422         }
1423         spin_unlock(&inode->i_lock);
1424 out:
1425         up_write(&nfsi->rwsem);
1426         return status;
1427 }
1428
1429 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1430 {
1431         struct nfs4_state *state;
1432         struct nfs4_lock_state *lock;
1433         int status = 0;
1434
1435         /* Note: we rely on the sp->so_states list being ordered 
1436          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1437          * states first.
1438          * This is needed to ensure that the server won't give us any
1439          * read delegations that we have to return if, say, we are
1440          * recovering after a network partition or a reboot from a
1441          * server that doesn't support a grace period.
1442          */
1443         spin_lock(&sp->so_lock);
1444         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1445 restart:
1446         list_for_each_entry(state, &sp->so_states, open_states) {
1447                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1448                         continue;
1449                 if (!nfs4_valid_open_stateid(state))
1450                         continue;
1451                 if (state->state == 0)
1452                         continue;
1453                 atomic_inc(&state->count);
1454                 spin_unlock(&sp->so_lock);
1455                 status = ops->recover_open(sp, state);
1456                 if (status >= 0) {
1457                         status = nfs4_reclaim_locks(state, ops);
1458                         if (status >= 0) {
1459                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1460                                         spin_lock(&state->state_lock);
1461                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1462                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1463                                                         pr_warn_ratelimited("NFS: "
1464                                                                             "%s: Lock reclaim "
1465                                                                             "failed!\n", __func__);
1466                                         }
1467                                         spin_unlock(&state->state_lock);
1468                                 }
1469                                 nfs4_put_open_state(state);
1470                                 spin_lock(&sp->so_lock);
1471                                 goto restart;
1472                         }
1473                 }
1474                 switch (status) {
1475                         default:
1476                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1477                                         __func__, status);
1478                         case -ENOENT:
1479                         case -ENOMEM:
1480                         case -ESTALE:
1481                                 /* Open state on this file cannot be recovered */
1482                                 nfs4_state_mark_recovery_failed(state, status);
1483                                 break;
1484                         case -EAGAIN:
1485                                 ssleep(1);
1486                         case -NFS4ERR_ADMIN_REVOKED:
1487                         case -NFS4ERR_STALE_STATEID:
1488                         case -NFS4ERR_BAD_STATEID:
1489                         case -NFS4ERR_RECLAIM_BAD:
1490                         case -NFS4ERR_RECLAIM_CONFLICT:
1491                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1492                                 break;
1493                         case -NFS4ERR_EXPIRED:
1494                         case -NFS4ERR_NO_GRACE:
1495                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1496                         case -NFS4ERR_STALE_CLIENTID:
1497                         case -NFS4ERR_BADSESSION:
1498                         case -NFS4ERR_BADSLOT:
1499                         case -NFS4ERR_BAD_HIGH_SLOT:
1500                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1501                                 goto out_err;
1502                 }
1503                 nfs4_put_open_state(state);
1504                 spin_lock(&sp->so_lock);
1505                 goto restart;
1506         }
1507         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1508         spin_unlock(&sp->so_lock);
1509         return 0;
1510 out_err:
1511         nfs4_put_open_state(state);
1512         spin_lock(&sp->so_lock);
1513         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1514         spin_unlock(&sp->so_lock);
1515         return status;
1516 }
1517
1518 static void nfs4_clear_open_state(struct nfs4_state *state)
1519 {
1520         struct nfs4_lock_state *lock;
1521
1522         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1523         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1524         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1525         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1526         spin_lock(&state->state_lock);
1527         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1528                 lock->ls_seqid.flags = 0;
1529                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1530         }
1531         spin_unlock(&state->state_lock);
1532 }
1533
1534 static void nfs4_reset_seqids(struct nfs_server *server,
1535         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1536 {
1537         struct nfs_client *clp = server->nfs_client;
1538         struct nfs4_state_owner *sp;
1539         struct rb_node *pos;
1540         struct nfs4_state *state;
1541
1542         spin_lock(&clp->cl_lock);
1543         for (pos = rb_first(&server->state_owners);
1544              pos != NULL;
1545              pos = rb_next(pos)) {
1546                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1547                 sp->so_seqid.flags = 0;
1548                 spin_lock(&sp->so_lock);
1549                 list_for_each_entry(state, &sp->so_states, open_states) {
1550                         if (mark_reclaim(clp, state))
1551                                 nfs4_clear_open_state(state);
1552                 }
1553                 spin_unlock(&sp->so_lock);
1554         }
1555         spin_unlock(&clp->cl_lock);
1556 }
1557
1558 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1559         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1560 {
1561         struct nfs_server *server;
1562
1563         rcu_read_lock();
1564         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1565                 nfs4_reset_seqids(server, mark_reclaim);
1566         rcu_read_unlock();
1567 }
1568
1569 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1570 {
1571         /* Mark all delegations for reclaim */
1572         nfs_delegation_mark_reclaim(clp);
1573         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1574 }
1575
1576 static void nfs4_reclaim_complete(struct nfs_client *clp,
1577                                  const struct nfs4_state_recovery_ops *ops,
1578                                  struct rpc_cred *cred)
1579 {
1580         /* Notify the server we're done reclaiming our state */
1581         if (ops->reclaim_complete)
1582                 (void)ops->reclaim_complete(clp, cred);
1583 }
1584
1585 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1586 {
1587         struct nfs_client *clp = server->nfs_client;
1588         struct nfs4_state_owner *sp;
1589         struct rb_node *pos;
1590         struct nfs4_state *state;
1591
1592         spin_lock(&clp->cl_lock);
1593         for (pos = rb_first(&server->state_owners);
1594              pos != NULL;
1595              pos = rb_next(pos)) {
1596                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1597                 spin_lock(&sp->so_lock);
1598                 list_for_each_entry(state, &sp->so_states, open_states) {
1599                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1600                                                 &state->flags))
1601                                 continue;
1602                         nfs4_state_mark_reclaim_nograce(clp, state);
1603                 }
1604                 spin_unlock(&sp->so_lock);
1605         }
1606         spin_unlock(&clp->cl_lock);
1607 }
1608
1609 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1610 {
1611         struct nfs_server *server;
1612
1613         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1614                 return 0;
1615
1616         rcu_read_lock();
1617         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1618                 nfs4_clear_reclaim_server(server);
1619         rcu_read_unlock();
1620
1621         nfs_delegation_reap_unclaimed(clp);
1622         return 1;
1623 }
1624
1625 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1626 {
1627         const struct nfs4_state_recovery_ops *ops;
1628         struct rpc_cred *cred;
1629
1630         if (!nfs4_state_clear_reclaim_reboot(clp))
1631                 return;
1632         ops = clp->cl_mvops->reboot_recovery_ops;
1633         cred = nfs4_get_clid_cred(clp);
1634         nfs4_reclaim_complete(clp, ops, cred);
1635         put_rpccred(cred);
1636 }
1637
1638 static void nfs_delegation_clear_all(struct nfs_client *clp)
1639 {
1640         nfs_delegation_mark_reclaim(clp);
1641         nfs_delegation_reap_unclaimed(clp);
1642 }
1643
1644 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1645 {
1646         nfs_delegation_clear_all(clp);
1647         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1648 }
1649
1650 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1651 {
1652         switch (error) {
1653                 case 0:
1654                         break;
1655                 case -NFS4ERR_CB_PATH_DOWN:
1656                         nfs40_handle_cb_pathdown(clp);
1657                         break;
1658                 case -NFS4ERR_NO_GRACE:
1659                         nfs4_state_end_reclaim_reboot(clp);
1660                         break;
1661                 case -NFS4ERR_STALE_CLIENTID:
1662                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1663                         nfs4_state_clear_reclaim_reboot(clp);
1664                         nfs4_state_start_reclaim_reboot(clp);
1665                         break;
1666                 case -NFS4ERR_EXPIRED:
1667                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1668                         nfs4_state_start_reclaim_nograce(clp);
1669                         break;
1670                 case -NFS4ERR_BADSESSION:
1671                 case -NFS4ERR_BADSLOT:
1672                 case -NFS4ERR_BAD_HIGH_SLOT:
1673                 case -NFS4ERR_DEADSESSION:
1674                 case -NFS4ERR_SEQ_FALSE_RETRY:
1675                 case -NFS4ERR_SEQ_MISORDERED:
1676                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1677                         /* Zero session reset errors */
1678                         break;
1679                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1680                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1681                         break;
1682                 default:
1683                         dprintk("%s: failed to handle error %d for server %s\n",
1684                                         __func__, error, clp->cl_hostname);
1685                         return error;
1686         }
1687         dprintk("%s: handled error %d for server %s\n", __func__, error,
1688                         clp->cl_hostname);
1689         return 0;
1690 }
1691
1692 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1693 {
1694         struct nfs4_state_owner *sp;
1695         struct nfs_server *server;
1696         struct rb_node *pos;
1697         int status = 0;
1698
1699 restart:
1700         rcu_read_lock();
1701         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1702                 nfs4_purge_state_owners(server);
1703                 spin_lock(&clp->cl_lock);
1704                 for (pos = rb_first(&server->state_owners);
1705                      pos != NULL;
1706                      pos = rb_next(pos)) {
1707                         sp = rb_entry(pos,
1708                                 struct nfs4_state_owner, so_server_node);
1709                         if (!test_and_clear_bit(ops->owner_flag_bit,
1710                                                         &sp->so_flags))
1711                                 continue;
1712                         atomic_inc(&sp->so_count);
1713                         spin_unlock(&clp->cl_lock);
1714                         rcu_read_unlock();
1715
1716                         status = nfs4_reclaim_open_state(sp, ops);
1717                         if (status < 0) {
1718                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1719                                 nfs4_put_state_owner(sp);
1720                                 return nfs4_recovery_handle_error(clp, status);
1721                         }
1722
1723                         nfs4_put_state_owner(sp);
1724                         goto restart;
1725                 }
1726                 spin_unlock(&clp->cl_lock);
1727         }
1728         rcu_read_unlock();
1729         return status;
1730 }
1731
1732 static int nfs4_check_lease(struct nfs_client *clp)
1733 {
1734         struct rpc_cred *cred;
1735         const struct nfs4_state_maintenance_ops *ops =
1736                 clp->cl_mvops->state_renewal_ops;
1737         int status;
1738
1739         /* Is the client already known to have an expired lease? */
1740         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1741                 return 0;
1742         spin_lock(&clp->cl_lock);
1743         cred = ops->get_state_renewal_cred_locked(clp);
1744         spin_unlock(&clp->cl_lock);
1745         if (cred == NULL) {
1746                 cred = nfs4_get_clid_cred(clp);
1747                 status = -ENOKEY;
1748                 if (cred == NULL)
1749                         goto out;
1750         }
1751         status = ops->renew_lease(clp, cred);
1752         put_rpccred(cred);
1753         if (status == -ETIMEDOUT) {
1754                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1755                 return 0;
1756         }
1757 out:
1758         return nfs4_recovery_handle_error(clp, status);
1759 }
1760
1761 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1762  * and for recoverable errors on EXCHANGE_ID for v4.1
1763  */
1764 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1765 {
1766         switch (status) {
1767         case -NFS4ERR_SEQ_MISORDERED:
1768                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1769                         return -ESERVERFAULT;
1770                 /* Lease confirmation error: retry after purging the lease */
1771                 ssleep(1);
1772                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1773                 break;
1774         case -NFS4ERR_STALE_CLIENTID:
1775                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1776                 nfs4_state_clear_reclaim_reboot(clp);
1777                 nfs4_state_start_reclaim_reboot(clp);
1778                 break;
1779         case -NFS4ERR_CLID_INUSE:
1780                 pr_err("NFS: Server %s reports our clientid is in use\n",
1781                         clp->cl_hostname);
1782                 nfs_mark_client_ready(clp, -EPERM);
1783                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1784                 return -EPERM;
1785         case -EACCES:
1786         case -NFS4ERR_DELAY:
1787         case -ETIMEDOUT:
1788         case -EAGAIN:
1789                 ssleep(1);
1790                 break;
1791
1792         case -NFS4ERR_MINOR_VERS_MISMATCH:
1793                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1794                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1795                 dprintk("%s: exit with error %d for server %s\n",
1796                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1797                 return -EPROTONOSUPPORT;
1798         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1799                                  * in nfs4_exchange_id */
1800         default:
1801                 dprintk("%s: exit with error %d for server %s\n", __func__,
1802                                 status, clp->cl_hostname);
1803                 return status;
1804         }
1805         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1806         dprintk("%s: handled error %d for server %s\n", __func__, status,
1807                         clp->cl_hostname);
1808         return 0;
1809 }
1810
1811 static int nfs4_establish_lease(struct nfs_client *clp)
1812 {
1813         struct rpc_cred *cred;
1814         const struct nfs4_state_recovery_ops *ops =
1815                 clp->cl_mvops->reboot_recovery_ops;
1816         int status;
1817
1818         cred = nfs4_get_clid_cred(clp);
1819         if (cred == NULL)
1820                 return -ENOENT;
1821         status = ops->establish_clid(clp, cred);
1822         put_rpccred(cred);
1823         if (status != 0)
1824                 return status;
1825         pnfs_destroy_all_layouts(clp);
1826         return 0;
1827 }
1828
1829 /*
1830  * Returns zero or a negative errno.  NFS4ERR values are converted
1831  * to local errno values.
1832  */
1833 static int nfs4_reclaim_lease(struct nfs_client *clp)
1834 {
1835         int status;
1836
1837         status = nfs4_establish_lease(clp);
1838         if (status < 0)
1839                 return nfs4_handle_reclaim_lease_error(clp, status);
1840         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1841                 nfs4_state_start_reclaim_nograce(clp);
1842         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1843                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1844         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1845         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1846         return 0;
1847 }
1848
1849 static int nfs4_purge_lease(struct nfs_client *clp)
1850 {
1851         int status;
1852
1853         status = nfs4_establish_lease(clp);
1854         if (status < 0)
1855                 return nfs4_handle_reclaim_lease_error(clp, status);
1856         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1857         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1858         nfs4_state_start_reclaim_nograce(clp);
1859         return 0;
1860 }
1861
1862 /*
1863  * Try remote migration of one FSID from a source server to a
1864  * destination server.  The source server provides a list of
1865  * potential destinations.
1866  *
1867  * Returns zero or a negative NFS4ERR status code.
1868  */
1869 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1870 {
1871         struct nfs_client *clp = server->nfs_client;
1872         struct nfs4_fs_locations *locations = NULL;
1873         struct inode *inode;
1874         struct page *page;
1875         int status, result;
1876
1877         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1878                         (unsigned long long)server->fsid.major,
1879                         (unsigned long long)server->fsid.minor,
1880                         clp->cl_hostname);
1881
1882         result = 0;
1883         page = alloc_page(GFP_KERNEL);
1884         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1885         if (page == NULL || locations == NULL) {
1886                 dprintk("<-- %s: no memory\n", __func__);
1887                 goto out;
1888         }
1889
1890         inode = server->super->s_root->d_inode;
1891         result = nfs4_proc_get_locations(inode, locations, page, cred);
1892         if (result) {
1893                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1894                         __func__, result);
1895                 goto out;
1896         }
1897
1898         result = -NFS4ERR_NXIO;
1899         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1900                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1901                         __func__);
1902                 goto out;
1903         }
1904
1905         nfs4_begin_drain_session(clp);
1906
1907         status = nfs4_replace_transport(server, locations);
1908         if (status != 0) {
1909                 dprintk("<-- %s: failed to replace transport: %d\n",
1910                         __func__, status);
1911                 goto out;
1912         }
1913
1914         result = 0;
1915         dprintk("<-- %s: migration succeeded\n", __func__);
1916
1917 out:
1918         if (page != NULL)
1919                 __free_page(page);
1920         kfree(locations);
1921         if (result) {
1922                 pr_err("NFS: migration recovery failed (server %s)\n",
1923                                 clp->cl_hostname);
1924                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1925         }
1926         return result;
1927 }
1928
1929 /*
1930  * Returns zero or a negative NFS4ERR status code.
1931  */
1932 static int nfs4_handle_migration(struct nfs_client *clp)
1933 {
1934         const struct nfs4_state_maintenance_ops *ops =
1935                                 clp->cl_mvops->state_renewal_ops;
1936         struct nfs_server *server;
1937         struct rpc_cred *cred;
1938
1939         dprintk("%s: migration reported on \"%s\"\n", __func__,
1940                         clp->cl_hostname);
1941
1942         spin_lock(&clp->cl_lock);
1943         cred = ops->get_state_renewal_cred_locked(clp);
1944         spin_unlock(&clp->cl_lock);
1945         if (cred == NULL)
1946                 return -NFS4ERR_NOENT;
1947
1948         clp->cl_mig_gen++;
1949 restart:
1950         rcu_read_lock();
1951         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1952                 int status;
1953
1954                 if (server->mig_gen == clp->cl_mig_gen)
1955                         continue;
1956                 server->mig_gen = clp->cl_mig_gen;
1957
1958                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1959                                                 &server->mig_status))
1960                         continue;
1961
1962                 rcu_read_unlock();
1963                 status = nfs4_try_migration(server, cred);
1964                 if (status < 0) {
1965                         put_rpccred(cred);
1966                         return status;
1967                 }
1968                 goto restart;
1969         }
1970         rcu_read_unlock();
1971         put_rpccred(cred);
1972         return 0;
1973 }
1974
1975 /*
1976  * Test each nfs_server on the clp's cl_superblocks list to see
1977  * if it's moved to another server.  Stop when the server no longer
1978  * returns NFS4ERR_LEASE_MOVED.
1979  */
1980 static int nfs4_handle_lease_moved(struct nfs_client *clp)
1981 {
1982         const struct nfs4_state_maintenance_ops *ops =
1983                                 clp->cl_mvops->state_renewal_ops;
1984         struct nfs_server *server;
1985         struct rpc_cred *cred;
1986
1987         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
1988                         clp->cl_hostname);
1989
1990         spin_lock(&clp->cl_lock);
1991         cred = ops->get_state_renewal_cred_locked(clp);
1992         spin_unlock(&clp->cl_lock);
1993         if (cred == NULL)
1994                 return -NFS4ERR_NOENT;
1995
1996         clp->cl_mig_gen++;
1997 restart:
1998         rcu_read_lock();
1999         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2000                 struct inode *inode;
2001                 int status;
2002
2003                 if (server->mig_gen == clp->cl_mig_gen)
2004                         continue;
2005                 server->mig_gen = clp->cl_mig_gen;
2006
2007                 rcu_read_unlock();
2008
2009                 inode = server->super->s_root->d_inode;
2010                 status = nfs4_proc_fsid_present(inode, cred);
2011                 if (status != -NFS4ERR_MOVED)
2012                         goto restart;   /* wasn't this one */
2013                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2014                         goto restart;   /* there are more */
2015                 goto out;
2016         }
2017         rcu_read_unlock();
2018
2019 out:
2020         put_rpccred(cred);
2021         return 0;
2022 }
2023
2024 /**
2025  * nfs4_discover_server_trunking - Detect server IP address trunking
2026  *
2027  * @clp: nfs_client under test
2028  * @result: OUT: found nfs_client, or clp
2029  *
2030  * Returns zero or a negative errno.  If zero is returned,
2031  * an nfs_client pointer is planted in "result".
2032  *
2033  * Note: since we are invoked in process context, and
2034  * not from inside the state manager, we cannot use
2035  * nfs4_handle_reclaim_lease_error().
2036  */
2037 int nfs4_discover_server_trunking(struct nfs_client *clp,
2038                                   struct nfs_client **result)
2039 {
2040         const struct nfs4_state_recovery_ops *ops =
2041                                 clp->cl_mvops->reboot_recovery_ops;
2042         struct rpc_clnt *clnt;
2043         struct rpc_cred *cred;
2044         int i, status;
2045
2046         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2047
2048         clnt = clp->cl_rpcclient;
2049         i = 0;
2050
2051         mutex_lock(&nfs_clid_init_mutex);
2052 again:
2053         status  = -ENOENT;
2054         cred = nfs4_get_clid_cred(clp);
2055         if (cred == NULL)
2056                 goto out_unlock;
2057
2058         status = ops->detect_trunking(clp, result, cred);
2059         put_rpccred(cred);
2060         switch (status) {
2061         case 0:
2062                 break;
2063         case -ETIMEDOUT:
2064                 if (clnt->cl_softrtry)
2065                         break;
2066         case -NFS4ERR_DELAY:
2067         case -EAGAIN:
2068                 ssleep(1);
2069         case -NFS4ERR_STALE_CLIENTID:
2070                 dprintk("NFS: %s after status %d, retrying\n",
2071                         __func__, status);
2072                 goto again;
2073         case -EACCES:
2074                 if (i++ == 0) {
2075                         nfs4_root_machine_cred(clp);
2076                         goto again;
2077                 }
2078                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2079                         break;
2080         case -NFS4ERR_CLID_INUSE:
2081         case -NFS4ERR_WRONGSEC:
2082                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2083                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2084                         status = -EPERM;
2085                         break;
2086                 }
2087                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2088                 if (IS_ERR(clnt)) {
2089                         status = PTR_ERR(clnt);
2090                         break;
2091                 }
2092                 /* Note: this is safe because we haven't yet marked the
2093                  * client as ready, so we are the only user of
2094                  * clp->cl_rpcclient
2095                  */
2096                 clnt = xchg(&clp->cl_rpcclient, clnt);
2097                 rpc_shutdown_client(clnt);
2098                 clnt = clp->cl_rpcclient;
2099                 goto again;
2100
2101         case -NFS4ERR_MINOR_VERS_MISMATCH:
2102                 status = -EPROTONOSUPPORT;
2103                 break;
2104
2105         case -EKEYEXPIRED:
2106         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2107                                  * in nfs4_exchange_id */
2108                 status = -EKEYEXPIRED;
2109                 break;
2110         default:
2111                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2112                                 __func__, status);
2113                 status = -EIO;
2114         }
2115
2116 out_unlock:
2117         mutex_unlock(&nfs_clid_init_mutex);
2118         dprintk("NFS: %s: status = %d\n", __func__, status);
2119         return status;
2120 }
2121
2122 #ifdef CONFIG_NFS_V4_1
2123 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2124 {
2125         struct nfs_client *clp = session->clp;
2126
2127         switch (err) {
2128         default:
2129                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2130                 break;
2131         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2132                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2133         }
2134         nfs4_schedule_lease_recovery(clp);
2135 }
2136 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2137
2138 static void nfs41_ping_server(struct nfs_client *clp)
2139 {
2140         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2141         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2142         nfs4_schedule_state_manager(clp);
2143 }
2144
2145 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2146 {
2147         nfs41_ping_server(clp);
2148 }
2149
2150 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2151 {
2152         nfs41_ping_server(clp);
2153 }
2154
2155 static void nfs4_reset_all_state(struct nfs_client *clp)
2156 {
2157         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2158                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2159                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2160                 nfs4_state_start_reclaim_nograce(clp);
2161                 dprintk("%s: scheduling reset of all state for server %s!\n",
2162                                 __func__, clp->cl_hostname);
2163                 nfs4_schedule_state_manager(clp);
2164         }
2165 }
2166
2167 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2168 {
2169         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2170                 nfs4_state_start_reclaim_reboot(clp);
2171                 dprintk("%s: server %s rebooted!\n", __func__,
2172                                 clp->cl_hostname);
2173                 nfs4_schedule_state_manager(clp);
2174         }
2175 }
2176
2177 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2178 {
2179         nfs4_reset_all_state(clp);
2180         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2181 }
2182
2183 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2184 {
2185         /* This will need to handle layouts too */
2186         nfs_expire_all_delegations(clp);
2187         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2188                         clp->cl_hostname);
2189 }
2190
2191 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2192 {
2193         nfs_expire_all_delegations(clp);
2194         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2195                 nfs4_schedule_state_manager(clp);
2196         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2197                         clp->cl_hostname);
2198 }
2199
2200 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2201 {
2202         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2203                 &clp->cl_state) == 0)
2204                 nfs4_schedule_state_manager(clp);
2205 }
2206
2207 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2208 {
2209         if (!flags)
2210                 return;
2211
2212         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2213                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2214
2215         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2216                 nfs41_handle_server_reboot(clp);
2217         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2218                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2219                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2220                 nfs41_handle_state_revoked(clp);
2221         if (flags & SEQ4_STATUS_LEASE_MOVED)
2222                 nfs4_schedule_lease_moved_recovery(clp);
2223         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2224                 nfs41_handle_recallable_state_revoked(clp);
2225         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2226                 nfs41_handle_backchannel_fault(clp);
2227         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2228                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2229                 nfs41_handle_cb_path_down(clp);
2230 }
2231
2232 static int nfs4_reset_session(struct nfs_client *clp)
2233 {
2234         struct rpc_cred *cred;
2235         int status;
2236
2237         if (!nfs4_has_session(clp))
2238                 return 0;
2239         nfs4_begin_drain_session(clp);
2240         cred = nfs4_get_clid_cred(clp);
2241         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2242         switch (status) {
2243         case 0:
2244         case -NFS4ERR_BADSESSION:
2245         case -NFS4ERR_DEADSESSION:
2246                 break;
2247         case -NFS4ERR_BACK_CHAN_BUSY:
2248         case -NFS4ERR_DELAY:
2249                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2250                 status = 0;
2251                 ssleep(1);
2252                 goto out;
2253         default:
2254                 status = nfs4_recovery_handle_error(clp, status);
2255                 goto out;
2256         }
2257
2258         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2259         status = nfs4_proc_create_session(clp, cred);
2260         if (status) {
2261                 dprintk("%s: session reset failed with status %d for server %s!\n",
2262                         __func__, status, clp->cl_hostname);
2263                 status = nfs4_handle_reclaim_lease_error(clp, status);
2264                 goto out;
2265         }
2266         nfs41_finish_session_reset(clp);
2267         dprintk("%s: session reset was successful for server %s!\n",
2268                         __func__, clp->cl_hostname);
2269 out:
2270         if (cred)
2271                 put_rpccred(cred);
2272         return status;
2273 }
2274
2275 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2276 {
2277         struct rpc_cred *cred;
2278         int ret;
2279
2280         if (!nfs4_has_session(clp))
2281                 return 0;
2282         nfs4_begin_drain_session(clp);
2283         cred = nfs4_get_clid_cred(clp);
2284         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2285         if (cred)
2286                 put_rpccred(cred);
2287         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2288         switch (ret) {
2289         case 0:
2290                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2291                         __func__, clp->cl_hostname);
2292                 break;
2293         case -NFS4ERR_DELAY:
2294                 ssleep(1);
2295                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2296                 break;
2297         default:
2298                 return nfs4_recovery_handle_error(clp, ret);
2299         }
2300         return 0;
2301 }
2302 #else /* CONFIG_NFS_V4_1 */
2303 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2304
2305 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2306 {
2307         return 0;
2308 }
2309 #endif /* CONFIG_NFS_V4_1 */
2310
2311 static void nfs4_state_manager(struct nfs_client *clp)
2312 {
2313         int status = 0;
2314         const char *section = "", *section_sep = "";
2315
2316         /* Ensure exclusive access to NFSv4 state */
2317         do {
2318                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2319                         section = "purge state";
2320                         status = nfs4_purge_lease(clp);
2321                         if (status < 0)
2322                                 goto out_error;
2323                         continue;
2324                 }
2325
2326                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2327                         section = "lease expired";
2328                         /* We're going to have to re-establish a clientid */
2329                         status = nfs4_reclaim_lease(clp);
2330                         if (status < 0)
2331                                 goto out_error;
2332                         continue;
2333                 }
2334
2335                 /* Initialize or reset the session */
2336                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2337                         section = "reset session";
2338                         status = nfs4_reset_session(clp);
2339                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2340                                 continue;
2341                         if (status < 0)
2342                                 goto out_error;
2343                 }
2344
2345                 /* Send BIND_CONN_TO_SESSION */
2346                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2347                                 &clp->cl_state)) {
2348                         section = "bind conn to session";
2349                         status = nfs4_bind_conn_to_session(clp);
2350                         if (status < 0)
2351                                 goto out_error;
2352                         continue;
2353                 }
2354
2355                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2356                         section = "check lease";
2357                         status = nfs4_check_lease(clp);
2358                         if (status < 0)
2359                                 goto out_error;
2360                 }
2361
2362                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2363                         section = "migration";
2364                         status = nfs4_handle_migration(clp);
2365                         if (status < 0)
2366                                 goto out_error;
2367                 }
2368
2369                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2370                         section = "lease moved";
2371                         status = nfs4_handle_lease_moved(clp);
2372                         if (status < 0)
2373                                 goto out_error;
2374                 }
2375
2376                 /* First recover reboot state... */
2377                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2378                         section = "reclaim reboot";
2379                         status = nfs4_do_reclaim(clp,
2380                                 clp->cl_mvops->reboot_recovery_ops);
2381                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2382                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2383                                 continue;
2384                         nfs4_state_end_reclaim_reboot(clp);
2385                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2386                                 continue;
2387                         if (status < 0)
2388                                 goto out_error;
2389                 }
2390
2391                 /* Now recover expired state... */
2392                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2393                         section = "reclaim nograce";
2394                         status = nfs4_do_reclaim(clp,
2395                                 clp->cl_mvops->nograce_recovery_ops);
2396                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2397                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2398                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2399                                 continue;
2400                         if (status < 0)
2401                                 goto out_error;
2402                 }
2403
2404                 nfs4_end_drain_session(clp);
2405                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2406                         nfs_client_return_marked_delegations(clp);
2407                         continue;
2408                 }
2409
2410                 nfs4_clear_state_manager_bit(clp);
2411                 /* Did we race with an attempt to give us more work? */
2412                 if (clp->cl_state == 0)
2413                         break;
2414                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2415                         break;
2416         } while (atomic_read(&clp->cl_count) > 1);
2417         return;
2418 out_error:
2419         if (strlen(section))
2420                 section_sep = ": ";
2421         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2422                         " with error %d\n", section_sep, section,
2423                         clp->cl_hostname, -status);
2424         ssleep(1);
2425         nfs4_end_drain_session(clp);
2426         nfs4_clear_state_manager_bit(clp);
2427 }
2428
2429 static int nfs4_run_state_manager(void *ptr)
2430 {
2431         struct nfs_client *clp = ptr;
2432
2433         allow_signal(SIGKILL);
2434         nfs4_state_manager(clp);
2435         nfs_put_client(clp);
2436         module_put_and_exit(0);
2437         return 0;
2438 }
2439
2440 /*
2441  * Local variables:
2442  *  c-basic-offset: 8
2443  * End:
2444  */