nfs41: nfs41_setup_state_renewal
[safe/jmp/linux-2.6] / 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/smp_lock.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/workqueue.h>
50 #include <linux/bitops.h>
51
52 #include "nfs4_fs.h"
53 #include "callback.h"
54 #include "delegation.h"
55 #include "internal.h"
56
57 #define OPENOWNER_POOL_SIZE     8
58
59 const nfs4_stateid zero_stateid;
60
61 static LIST_HEAD(nfs4_clientid_list);
62
63 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
64 {
65         unsigned short port;
66         int status;
67
68         port = nfs_callback_tcpport;
69         if (clp->cl_addr.ss_family == AF_INET6)
70                 port = nfs_callback_tcpport6;
71
72         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred);
73         if (status == 0)
74                 status = nfs4_proc_setclientid_confirm(clp, cred);
75         if (status == 0)
76                 nfs4_schedule_state_renewal(clp);
77         return status;
78 }
79
80 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
81 {
82         struct rpc_cred *cred = NULL;
83
84         if (clp->cl_machine_cred != NULL)
85                 cred = get_rpccred(clp->cl_machine_cred);
86         return cred;
87 }
88
89 static void nfs4_clear_machine_cred(struct nfs_client *clp)
90 {
91         struct rpc_cred *cred;
92
93         spin_lock(&clp->cl_lock);
94         cred = clp->cl_machine_cred;
95         clp->cl_machine_cred = NULL;
96         spin_unlock(&clp->cl_lock);
97         if (cred != NULL)
98                 put_rpccred(cred);
99 }
100
101 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
102 {
103         struct nfs4_state_owner *sp;
104         struct rb_node *pos;
105         struct rpc_cred *cred = NULL;
106
107         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
108                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
109                 if (list_empty(&sp->so_states))
110                         continue;
111                 cred = get_rpccred(sp->so_cred);
112                 break;
113         }
114         return cred;
115 }
116
117 #if defined(CONFIG_NFS_V4_1)
118
119 static int nfs41_setup_state_renewal(struct nfs_client *clp)
120 {
121         int status;
122         struct nfs_fsinfo fsinfo;
123
124         status = nfs4_proc_get_lease_time(clp, &fsinfo);
125         if (status == 0) {
126                 /* Update lease time and schedule renewal */
127                 spin_lock(&clp->cl_lock);
128                 clp->cl_lease_time = fsinfo.lease_time * HZ;
129                 clp->cl_last_renewal = jiffies;
130                 spin_unlock(&clp->cl_lock);
131
132                 nfs4_schedule_state_renewal(clp);
133         }
134
135         return status;
136 }
137
138 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
139 {
140         int status;
141
142         status = nfs4_proc_exchange_id(clp, cred);
143         if (status != 0)
144                 goto out;
145         status = nfs4_proc_create_session(clp);
146         if (status != 0)
147                 goto out;
148         nfs41_setup_state_renewal(clp);
149         nfs_mark_client_ready(clp, NFS_CS_READY);
150 out:
151         return status;
152 }
153
154 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
155 {
156         struct rpc_cred *cred;
157
158         spin_lock(&clp->cl_lock);
159         cred = nfs4_get_machine_cred_locked(clp);
160         spin_unlock(&clp->cl_lock);
161         return cred;
162 }
163
164 #endif /* CONFIG_NFS_V4_1 */
165
166 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
167 {
168         struct nfs4_state_owner *sp;
169         struct rb_node *pos;
170         struct rpc_cred *cred;
171
172         spin_lock(&clp->cl_lock);
173         cred = nfs4_get_machine_cred_locked(clp);
174         if (cred != NULL)
175                 goto out;
176         pos = rb_first(&clp->cl_state_owners);
177         if (pos != NULL) {
178                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
179                 cred = get_rpccred(sp->so_cred);
180         }
181 out:
182         spin_unlock(&clp->cl_lock);
183         return cred;
184 }
185
186 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
187                 __u64 minval, int maxbits)
188 {
189         struct rb_node **p, *parent;
190         struct nfs_unique_id *pos;
191         __u64 mask = ~0ULL;
192
193         if (maxbits < 64)
194                 mask = (1ULL << maxbits) - 1ULL;
195
196         /* Ensure distribution is more or less flat */
197         get_random_bytes(&new->id, sizeof(new->id));
198         new->id &= mask;
199         if (new->id < minval)
200                 new->id += minval;
201 retry:
202         p = &root->rb_node;
203         parent = NULL;
204
205         while (*p != NULL) {
206                 parent = *p;
207                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
208
209                 if (new->id < pos->id)
210                         p = &(*p)->rb_left;
211                 else if (new->id > pos->id)
212                         p = &(*p)->rb_right;
213                 else
214                         goto id_exists;
215         }
216         rb_link_node(&new->rb_node, parent, p);
217         rb_insert_color(&new->rb_node, root);
218         return;
219 id_exists:
220         for (;;) {
221                 new->id++;
222                 if (new->id < minval || (new->id & mask) != new->id) {
223                         new->id = minval;
224                         break;
225                 }
226                 parent = rb_next(parent);
227                 if (parent == NULL)
228                         break;
229                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
230                 if (new->id < pos->id)
231                         break;
232         }
233         goto retry;
234 }
235
236 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
237 {
238         rb_erase(&id->rb_node, root);
239 }
240
241 static struct nfs4_state_owner *
242 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
243 {
244         struct nfs_client *clp = server->nfs_client;
245         struct rb_node **p = &clp->cl_state_owners.rb_node,
246                        *parent = NULL;
247         struct nfs4_state_owner *sp, *res = NULL;
248
249         while (*p != NULL) {
250                 parent = *p;
251                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
252
253                 if (server < sp->so_server) {
254                         p = &parent->rb_left;
255                         continue;
256                 }
257                 if (server > sp->so_server) {
258                         p = &parent->rb_right;
259                         continue;
260                 }
261                 if (cred < sp->so_cred)
262                         p = &parent->rb_left;
263                 else if (cred > sp->so_cred)
264                         p = &parent->rb_right;
265                 else {
266                         atomic_inc(&sp->so_count);
267                         res = sp;
268                         break;
269                 }
270         }
271         return res;
272 }
273
274 static struct nfs4_state_owner *
275 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
276 {
277         struct rb_node **p = &clp->cl_state_owners.rb_node,
278                        *parent = NULL;
279         struct nfs4_state_owner *sp;
280
281         while (*p != NULL) {
282                 parent = *p;
283                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
284
285                 if (new->so_server < sp->so_server) {
286                         p = &parent->rb_left;
287                         continue;
288                 }
289                 if (new->so_server > sp->so_server) {
290                         p = &parent->rb_right;
291                         continue;
292                 }
293                 if (new->so_cred < sp->so_cred)
294                         p = &parent->rb_left;
295                 else if (new->so_cred > sp->so_cred)
296                         p = &parent->rb_right;
297                 else {
298                         atomic_inc(&sp->so_count);
299                         return sp;
300                 }
301         }
302         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
303         rb_link_node(&new->so_client_node, parent, p);
304         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
305         return new;
306 }
307
308 static void
309 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
310 {
311         if (!RB_EMPTY_NODE(&sp->so_client_node))
312                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
313         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
314 }
315
316 /*
317  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
318  * create a new state_owner.
319  *
320  */
321 static struct nfs4_state_owner *
322 nfs4_alloc_state_owner(void)
323 {
324         struct nfs4_state_owner *sp;
325
326         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
327         if (!sp)
328                 return NULL;
329         spin_lock_init(&sp->so_lock);
330         INIT_LIST_HEAD(&sp->so_states);
331         INIT_LIST_HEAD(&sp->so_delegations);
332         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
333         sp->so_seqid.sequence = &sp->so_sequence;
334         spin_lock_init(&sp->so_sequence.lock);
335         INIT_LIST_HEAD(&sp->so_sequence.list);
336         atomic_set(&sp->so_count, 1);
337         return sp;
338 }
339
340 static void
341 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
342 {
343         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
344                 struct nfs_client *clp = sp->so_client;
345
346                 spin_lock(&clp->cl_lock);
347                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
348                 RB_CLEAR_NODE(&sp->so_client_node);
349                 spin_unlock(&clp->cl_lock);
350         }
351 }
352
353 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
354 {
355         struct nfs_client *clp = server->nfs_client;
356         struct nfs4_state_owner *sp, *new;
357
358         spin_lock(&clp->cl_lock);
359         sp = nfs4_find_state_owner(server, cred);
360         spin_unlock(&clp->cl_lock);
361         if (sp != NULL)
362                 return sp;
363         new = nfs4_alloc_state_owner();
364         if (new == NULL)
365                 return NULL;
366         new->so_client = clp;
367         new->so_server = server;
368         new->so_cred = cred;
369         spin_lock(&clp->cl_lock);
370         sp = nfs4_insert_state_owner(clp, new);
371         spin_unlock(&clp->cl_lock);
372         if (sp == new)
373                 get_rpccred(cred);
374         else {
375                 rpc_destroy_wait_queue(&new->so_sequence.wait);
376                 kfree(new);
377         }
378         return sp;
379 }
380
381 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
382 {
383         struct nfs_client *clp = sp->so_client;
384         struct rpc_cred *cred = sp->so_cred;
385
386         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
387                 return;
388         nfs4_remove_state_owner(clp, sp);
389         spin_unlock(&clp->cl_lock);
390         rpc_destroy_wait_queue(&sp->so_sequence.wait);
391         put_rpccred(cred);
392         kfree(sp);
393 }
394
395 static struct nfs4_state *
396 nfs4_alloc_open_state(void)
397 {
398         struct nfs4_state *state;
399
400         state = kzalloc(sizeof(*state), GFP_KERNEL);
401         if (!state)
402                 return NULL;
403         atomic_set(&state->count, 1);
404         INIT_LIST_HEAD(&state->lock_states);
405         spin_lock_init(&state->state_lock);
406         seqlock_init(&state->seqlock);
407         return state;
408 }
409
410 void
411 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
412 {
413         if (state->state == fmode)
414                 return;
415         /* NB! List reordering - see the reclaim code for why.  */
416         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
417                 if (fmode & FMODE_WRITE)
418                         list_move(&state->open_states, &state->owner->so_states);
419                 else
420                         list_move_tail(&state->open_states, &state->owner->so_states);
421         }
422         state->state = fmode;
423 }
424
425 static struct nfs4_state *
426 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
427 {
428         struct nfs_inode *nfsi = NFS_I(inode);
429         struct nfs4_state *state;
430
431         list_for_each_entry(state, &nfsi->open_states, inode_states) {
432                 if (state->owner != owner)
433                         continue;
434                 if (atomic_inc_not_zero(&state->count))
435                         return state;
436         }
437         return NULL;
438 }
439
440 static void
441 nfs4_free_open_state(struct nfs4_state *state)
442 {
443         kfree(state);
444 }
445
446 struct nfs4_state *
447 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
448 {
449         struct nfs4_state *state, *new;
450         struct nfs_inode *nfsi = NFS_I(inode);
451
452         spin_lock(&inode->i_lock);
453         state = __nfs4_find_state_byowner(inode, owner);
454         spin_unlock(&inode->i_lock);
455         if (state)
456                 goto out;
457         new = nfs4_alloc_open_state();
458         spin_lock(&owner->so_lock);
459         spin_lock(&inode->i_lock);
460         state = __nfs4_find_state_byowner(inode, owner);
461         if (state == NULL && new != NULL) {
462                 state = new;
463                 state->owner = owner;
464                 atomic_inc(&owner->so_count);
465                 list_add(&state->inode_states, &nfsi->open_states);
466                 state->inode = igrab(inode);
467                 spin_unlock(&inode->i_lock);
468                 /* Note: The reclaim code dictates that we add stateless
469                  * and read-only stateids to the end of the list */
470                 list_add_tail(&state->open_states, &owner->so_states);
471                 spin_unlock(&owner->so_lock);
472         } else {
473                 spin_unlock(&inode->i_lock);
474                 spin_unlock(&owner->so_lock);
475                 if (new)
476                         nfs4_free_open_state(new);
477         }
478 out:
479         return state;
480 }
481
482 void nfs4_put_open_state(struct nfs4_state *state)
483 {
484         struct inode *inode = state->inode;
485         struct nfs4_state_owner *owner = state->owner;
486
487         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
488                 return;
489         spin_lock(&inode->i_lock);
490         list_del(&state->inode_states);
491         list_del(&state->open_states);
492         spin_unlock(&inode->i_lock);
493         spin_unlock(&owner->so_lock);
494         iput(inode);
495         nfs4_free_open_state(state);
496         nfs4_put_state_owner(owner);
497 }
498
499 /*
500  * Close the current file.
501  */
502 static void __nfs4_close(struct path *path, struct nfs4_state *state, fmode_t fmode, int wait)
503 {
504         struct nfs4_state_owner *owner = state->owner;
505         int call_close = 0;
506         fmode_t newstate;
507
508         atomic_inc(&owner->so_count);
509         /* Protect against nfs4_find_state() */
510         spin_lock(&owner->so_lock);
511         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
512                 case FMODE_READ:
513                         state->n_rdonly--;
514                         break;
515                 case FMODE_WRITE:
516                         state->n_wronly--;
517                         break;
518                 case FMODE_READ|FMODE_WRITE:
519                         state->n_rdwr--;
520         }
521         newstate = FMODE_READ|FMODE_WRITE;
522         if (state->n_rdwr == 0) {
523                 if (state->n_rdonly == 0) {
524                         newstate &= ~FMODE_READ;
525                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
526                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
527                 }
528                 if (state->n_wronly == 0) {
529                         newstate &= ~FMODE_WRITE;
530                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
531                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
532                 }
533                 if (newstate == 0)
534                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
535         }
536         nfs4_state_set_mode_locked(state, newstate);
537         spin_unlock(&owner->so_lock);
538
539         if (!call_close) {
540                 nfs4_put_open_state(state);
541                 nfs4_put_state_owner(owner);
542         } else
543                 nfs4_do_close(path, state, wait);
544 }
545
546 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
547 {
548         __nfs4_close(path, state, fmode, 0);
549 }
550
551 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
552 {
553         __nfs4_close(path, state, fmode, 1);
554 }
555
556 /*
557  * Search the state->lock_states for an existing lock_owner
558  * that is compatible with current->files
559  */
560 static struct nfs4_lock_state *
561 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
562 {
563         struct nfs4_lock_state *pos;
564         list_for_each_entry(pos, &state->lock_states, ls_locks) {
565                 if (pos->ls_owner != fl_owner)
566                         continue;
567                 atomic_inc(&pos->ls_count);
568                 return pos;
569         }
570         return NULL;
571 }
572
573 /*
574  * Return a compatible lock_state. If no initialized lock_state structure
575  * exists, return an uninitialized one.
576  *
577  */
578 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
579 {
580         struct nfs4_lock_state *lsp;
581         struct nfs_client *clp = state->owner->so_client;
582
583         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
584         if (lsp == NULL)
585                 return NULL;
586         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
587         spin_lock_init(&lsp->ls_sequence.lock);
588         INIT_LIST_HEAD(&lsp->ls_sequence.list);
589         lsp->ls_seqid.sequence = &lsp->ls_sequence;
590         atomic_set(&lsp->ls_count, 1);
591         lsp->ls_state = state;
592         lsp->ls_owner = fl_owner;
593         spin_lock(&clp->cl_lock);
594         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
595         spin_unlock(&clp->cl_lock);
596         INIT_LIST_HEAD(&lsp->ls_locks);
597         return lsp;
598 }
599
600 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
601 {
602         struct nfs_client *clp = lsp->ls_state->owner->so_client;
603
604         spin_lock(&clp->cl_lock);
605         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
606         spin_unlock(&clp->cl_lock);
607         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
608         kfree(lsp);
609 }
610
611 /*
612  * Return a compatible lock_state. If no initialized lock_state structure
613  * exists, return an uninitialized one.
614  *
615  */
616 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
617 {
618         struct nfs4_lock_state *lsp, *new = NULL;
619         
620         for(;;) {
621                 spin_lock(&state->state_lock);
622                 lsp = __nfs4_find_lock_state(state, owner);
623                 if (lsp != NULL)
624                         break;
625                 if (new != NULL) {
626                         list_add(&new->ls_locks, &state->lock_states);
627                         set_bit(LK_STATE_IN_USE, &state->flags);
628                         lsp = new;
629                         new = NULL;
630                         break;
631                 }
632                 spin_unlock(&state->state_lock);
633                 new = nfs4_alloc_lock_state(state, owner);
634                 if (new == NULL)
635                         return NULL;
636         }
637         spin_unlock(&state->state_lock);
638         if (new != NULL)
639                 nfs4_free_lock_state(new);
640         return lsp;
641 }
642
643 /*
644  * Release reference to lock_state, and free it if we see that
645  * it is no longer in use
646  */
647 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
648 {
649         struct nfs4_state *state;
650
651         if (lsp == NULL)
652                 return;
653         state = lsp->ls_state;
654         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
655                 return;
656         list_del(&lsp->ls_locks);
657         if (list_empty(&state->lock_states))
658                 clear_bit(LK_STATE_IN_USE, &state->flags);
659         spin_unlock(&state->state_lock);
660         nfs4_free_lock_state(lsp);
661 }
662
663 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
664 {
665         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
666
667         dst->fl_u.nfs4_fl.owner = lsp;
668         atomic_inc(&lsp->ls_count);
669 }
670
671 static void nfs4_fl_release_lock(struct file_lock *fl)
672 {
673         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
674 }
675
676 static const struct file_lock_operations nfs4_fl_lock_ops = {
677         .fl_copy_lock = nfs4_fl_copy_lock,
678         .fl_release_private = nfs4_fl_release_lock,
679 };
680
681 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
682 {
683         struct nfs4_lock_state *lsp;
684
685         if (fl->fl_ops != NULL)
686                 return 0;
687         lsp = nfs4_get_lock_state(state, fl->fl_owner);
688         if (lsp == NULL)
689                 return -ENOMEM;
690         fl->fl_u.nfs4_fl.owner = lsp;
691         fl->fl_ops = &nfs4_fl_lock_ops;
692         return 0;
693 }
694
695 /*
696  * Byte-range lock aware utility to initialize the stateid of read/write
697  * requests.
698  */
699 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
700 {
701         struct nfs4_lock_state *lsp;
702         int seq;
703
704         do {
705                 seq = read_seqbegin(&state->seqlock);
706                 memcpy(dst, &state->stateid, sizeof(*dst));
707         } while (read_seqretry(&state->seqlock, seq));
708         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
709                 return;
710
711         spin_lock(&state->state_lock);
712         lsp = __nfs4_find_lock_state(state, fl_owner);
713         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
714                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
715         spin_unlock(&state->state_lock);
716         nfs4_put_lock_state(lsp);
717 }
718
719 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
720 {
721         struct nfs_seqid *new;
722
723         new = kmalloc(sizeof(*new), GFP_KERNEL);
724         if (new != NULL) {
725                 new->sequence = counter;
726                 INIT_LIST_HEAD(&new->list);
727         }
728         return new;
729 }
730
731 void nfs_free_seqid(struct nfs_seqid *seqid)
732 {
733         if (!list_empty(&seqid->list)) {
734                 struct rpc_sequence *sequence = seqid->sequence->sequence;
735
736                 spin_lock(&sequence->lock);
737                 list_del(&seqid->list);
738                 spin_unlock(&sequence->lock);
739                 rpc_wake_up(&sequence->wait);
740         }
741         kfree(seqid);
742 }
743
744 /*
745  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
746  * failed with a seqid incrementing error -
747  * see comments nfs_fs.h:seqid_mutating_error()
748  */
749 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
750 {
751         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
752         switch (status) {
753                 case 0:
754                         break;
755                 case -NFS4ERR_BAD_SEQID:
756                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
757                                 return;
758                         printk(KERN_WARNING "NFS: v4 server returned a bad"
759                                         " sequence-id error on an"
760                                         " unconfirmed sequence %p!\n",
761                                         seqid->sequence);
762                 case -NFS4ERR_STALE_CLIENTID:
763                 case -NFS4ERR_STALE_STATEID:
764                 case -NFS4ERR_BAD_STATEID:
765                 case -NFS4ERR_BADXDR:
766                 case -NFS4ERR_RESOURCE:
767                 case -NFS4ERR_NOFILEHANDLE:
768                         /* Non-seqid mutating errors */
769                         return;
770         };
771         /*
772          * Note: no locking needed as we are guaranteed to be first
773          * on the sequence list
774          */
775         seqid->sequence->counter++;
776 }
777
778 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
779 {
780         struct nfs4_state_owner *sp = container_of(seqid->sequence,
781                                         struct nfs4_state_owner, so_seqid);
782         struct nfs_server *server = sp->so_server;
783
784         if (status == -NFS4ERR_BAD_SEQID)
785                 nfs4_drop_state_owner(sp);
786         if (!nfs4_has_session(server->nfs_client))
787                 nfs_increment_seqid(status, seqid);
788 }
789
790 /*
791  * Increment the seqid if the LOCK/LOCKU succeeded, or
792  * failed with a seqid incrementing error -
793  * see comments nfs_fs.h:seqid_mutating_error()
794  */
795 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
796 {
797         nfs_increment_seqid(status, seqid);
798 }
799
800 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
801 {
802         struct rpc_sequence *sequence = seqid->sequence->sequence;
803         int status = 0;
804
805         spin_lock(&sequence->lock);
806         if (list_empty(&seqid->list))
807                 list_add_tail(&seqid->list, &sequence->list);
808         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
809                 goto unlock;
810         rpc_sleep_on(&sequence->wait, task, NULL);
811         status = -EAGAIN;
812 unlock:
813         spin_unlock(&sequence->lock);
814         return status;
815 }
816
817 static int nfs4_run_state_manager(void *);
818
819 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
820 {
821         smp_mb__before_clear_bit();
822         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
823         smp_mb__after_clear_bit();
824         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
825         rpc_wake_up(&clp->cl_rpcwaitq);
826 }
827
828 /*
829  * Schedule the nfs_client asynchronous state management routine
830  */
831 void nfs4_schedule_state_manager(struct nfs_client *clp)
832 {
833         struct task_struct *task;
834
835         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
836                 return;
837         __module_get(THIS_MODULE);
838         atomic_inc(&clp->cl_count);
839         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
840                                 rpc_peeraddr2str(clp->cl_rpcclient,
841                                                         RPC_DISPLAY_ADDR));
842         if (!IS_ERR(task))
843                 return;
844         nfs4_clear_state_manager_bit(clp);
845         nfs_put_client(clp);
846         module_put(THIS_MODULE);
847 }
848
849 /*
850  * Schedule a state recovery attempt
851  */
852 void nfs4_schedule_state_recovery(struct nfs_client *clp)
853 {
854         if (!clp)
855                 return;
856         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
857                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
858         nfs4_schedule_state_manager(clp);
859 }
860
861 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
862 {
863
864         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
865         /* Don't recover state that expired before the reboot */
866         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
867                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
868                 return 0;
869         }
870         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
871         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
872         return 1;
873 }
874
875 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
876 {
877         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
878         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
879         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
880         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
881         return 1;
882 }
883
884 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
885 {
886         struct inode *inode = state->inode;
887         struct nfs_inode *nfsi = NFS_I(inode);
888         struct file_lock *fl;
889         int status = 0;
890
891         if (inode->i_flock == NULL)
892                 return 0;
893
894         /* Guard against delegation returns and new lock/unlock calls */
895         down_write(&nfsi->rwsem);
896         /* Protect inode->i_flock using the BKL */
897         lock_kernel();
898         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
899                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
900                         continue;
901                 if (nfs_file_open_context(fl->fl_file)->state != state)
902                         continue;
903                 unlock_kernel();
904                 status = ops->recover_lock(state, fl);
905                 switch (status) {
906                         case 0:
907                                 break;
908                         case -ESTALE:
909                         case -NFS4ERR_ADMIN_REVOKED:
910                         case -NFS4ERR_STALE_STATEID:
911                         case -NFS4ERR_BAD_STATEID:
912                         case -NFS4ERR_EXPIRED:
913                         case -NFS4ERR_NO_GRACE:
914                         case -NFS4ERR_STALE_CLIENTID:
915                         case -NFS4ERR_BADSESSION:
916                         case -NFS4ERR_BADSLOT:
917                         case -NFS4ERR_BAD_HIGH_SLOT:
918                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
919                                 goto out;
920                         default:
921                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
922                                                 __func__, status);
923                         case -ENOMEM:
924                         case -NFS4ERR_DENIED:
925                         case -NFS4ERR_RECLAIM_BAD:
926                         case -NFS4ERR_RECLAIM_CONFLICT:
927                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
928                                 status = 0;
929                 }
930                 lock_kernel();
931         }
932         unlock_kernel();
933 out:
934         up_write(&nfsi->rwsem);
935         return status;
936 }
937
938 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
939 {
940         struct nfs4_state *state;
941         struct nfs4_lock_state *lock;
942         int status = 0;
943
944         /* Note: we rely on the sp->so_states list being ordered 
945          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
946          * states first.
947          * This is needed to ensure that the server won't give us any
948          * read delegations that we have to return if, say, we are
949          * recovering after a network partition or a reboot from a
950          * server that doesn't support a grace period.
951          */
952 restart:
953         spin_lock(&sp->so_lock);
954         list_for_each_entry(state, &sp->so_states, open_states) {
955                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
956                         continue;
957                 if (state->state == 0)
958                         continue;
959                 atomic_inc(&state->count);
960                 spin_unlock(&sp->so_lock);
961                 status = ops->recover_open(sp, state);
962                 if (status >= 0) {
963                         status = nfs4_reclaim_locks(state, ops);
964                         if (status >= 0) {
965                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
966                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
967                                                 printk("%s: Lock reclaim failed!\n",
968                                                         __func__);
969                                 }
970                                 nfs4_put_open_state(state);
971                                 goto restart;
972                         }
973                 }
974                 switch (status) {
975                         default:
976                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
977                                                 __func__, status);
978                         case -ENOENT:
979                         case -ENOMEM:
980                         case -ESTALE:
981                                 /*
982                                  * Open state on this file cannot be recovered
983                                  * All we can do is revert to using the zero stateid.
984                                  */
985                                 memset(state->stateid.data, 0,
986                                         sizeof(state->stateid.data));
987                                 /* Mark the file as being 'closed' */
988                                 state->state = 0;
989                                 break;
990                         case -NFS4ERR_ADMIN_REVOKED:
991                         case -NFS4ERR_STALE_STATEID:
992                         case -NFS4ERR_BAD_STATEID:
993                         case -NFS4ERR_RECLAIM_BAD:
994                         case -NFS4ERR_RECLAIM_CONFLICT:
995                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
996                                 break;
997                         case -NFS4ERR_EXPIRED:
998                         case -NFS4ERR_NO_GRACE:
999                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
1000                         case -NFS4ERR_STALE_CLIENTID:
1001                         case -NFS4ERR_BADSESSION:
1002                         case -NFS4ERR_BADSLOT:
1003                         case -NFS4ERR_BAD_HIGH_SLOT:
1004                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1005                                 goto out_err;
1006                 }
1007                 nfs4_put_open_state(state);
1008                 goto restart;
1009         }
1010         spin_unlock(&sp->so_lock);
1011         return 0;
1012 out_err:
1013         nfs4_put_open_state(state);
1014         return status;
1015 }
1016
1017 static void nfs4_clear_open_state(struct nfs4_state *state)
1018 {
1019         struct nfs4_lock_state *lock;
1020
1021         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1022         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1023         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1024         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1025         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1026                 lock->ls_seqid.flags = 0;
1027                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1028         }
1029 }
1030
1031 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1032 {
1033         struct nfs4_state_owner *sp;
1034         struct rb_node *pos;
1035         struct nfs4_state *state;
1036
1037         /* Reset all sequence ids to zero */
1038         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1039                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1040                 sp->so_seqid.flags = 0;
1041                 spin_lock(&sp->so_lock);
1042                 list_for_each_entry(state, &sp->so_states, open_states) {
1043                         if (mark_reclaim(clp, state))
1044                                 nfs4_clear_open_state(state);
1045                 }
1046                 spin_unlock(&sp->so_lock);
1047         }
1048 }
1049
1050 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1051 {
1052         /* Mark all delegations for reclaim */
1053         nfs_delegation_mark_reclaim(clp);
1054         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1055 }
1056
1057 static void nfs4_reclaim_complete(struct nfs_client *clp,
1058                                  const struct nfs4_state_recovery_ops *ops)
1059 {
1060         /* Notify the server we're done reclaiming our state */
1061         if (ops->reclaim_complete)
1062                 (void)ops->reclaim_complete(clp);
1063 }
1064
1065 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1066 {
1067         struct nfs4_state_owner *sp;
1068         struct rb_node *pos;
1069         struct nfs4_state *state;
1070
1071         nfs4_reclaim_complete(clp,
1072                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1073
1074         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1075                 return;
1076
1077         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1078                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1079                 spin_lock(&sp->so_lock);
1080                 list_for_each_entry(state, &sp->so_states, open_states) {
1081                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1082                                 continue;
1083                         nfs4_state_mark_reclaim_nograce(clp, state);
1084                 }
1085                 spin_unlock(&sp->so_lock);
1086         }
1087
1088         nfs_delegation_reap_unclaimed(clp);
1089 }
1090
1091 static void nfs_delegation_clear_all(struct nfs_client *clp)
1092 {
1093         nfs_delegation_mark_reclaim(clp);
1094         nfs_delegation_reap_unclaimed(clp);
1095 }
1096
1097 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1098 {
1099         nfs_delegation_clear_all(clp);
1100         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1101 }
1102
1103 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1104 {
1105         switch (error) {
1106                 case -NFS4ERR_CB_PATH_DOWN:
1107                         nfs_handle_cb_pathdown(clp);
1108                         return 0;
1109                 case -NFS4ERR_NO_GRACE:
1110                         nfs4_state_end_reclaim_reboot(clp);
1111                         return 0;
1112                 case -NFS4ERR_STALE_CLIENTID:
1113                 case -NFS4ERR_LEASE_MOVED:
1114                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1115                         nfs4_state_end_reclaim_reboot(clp);
1116                         nfs4_state_start_reclaim_reboot(clp);
1117                         break;
1118                 case -NFS4ERR_EXPIRED:
1119                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1120                         nfs4_state_start_reclaim_nograce(clp);
1121                         break;
1122                 case -NFS4ERR_BADSESSION:
1123                 case -NFS4ERR_BADSLOT:
1124                 case -NFS4ERR_BAD_HIGH_SLOT:
1125                 case -NFS4ERR_DEADSESSION:
1126                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1127                 case -NFS4ERR_SEQ_FALSE_RETRY:
1128                 case -NFS4ERR_SEQ_MISORDERED:
1129                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1130                         /* Zero session reset errors */
1131                         return 0;
1132         }
1133         return error;
1134 }
1135
1136 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1137 {
1138         struct rb_node *pos;
1139         int status = 0;
1140
1141 restart:
1142         spin_lock(&clp->cl_lock);
1143         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1144                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1145                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1146                         continue;
1147                 atomic_inc(&sp->so_count);
1148                 spin_unlock(&clp->cl_lock);
1149                 status = nfs4_reclaim_open_state(sp, ops);
1150                 if (status < 0) {
1151                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1152                         nfs4_put_state_owner(sp);
1153                         return nfs4_recovery_handle_error(clp, status);
1154                 }
1155                 nfs4_put_state_owner(sp);
1156                 goto restart;
1157         }
1158         spin_unlock(&clp->cl_lock);
1159         return status;
1160 }
1161
1162 static int nfs4_check_lease(struct nfs_client *clp)
1163 {
1164         struct rpc_cred *cred;
1165         struct nfs4_state_maintenance_ops *ops =
1166                 nfs4_state_renewal_ops[clp->cl_minorversion];
1167         int status = -NFS4ERR_EXPIRED;
1168
1169         /* Is the client already known to have an expired lease? */
1170         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1171                 return 0;
1172         spin_lock(&clp->cl_lock);
1173         cred = ops->get_state_renewal_cred_locked(clp);
1174         spin_unlock(&clp->cl_lock);
1175         if (cred == NULL) {
1176                 cred = nfs4_get_setclientid_cred(clp);
1177                 if (cred == NULL)
1178                         goto out;
1179         }
1180         status = ops->renew_lease(clp, cred);
1181         put_rpccred(cred);
1182 out:
1183         return nfs4_recovery_handle_error(clp, status);
1184 }
1185
1186 static int nfs4_reclaim_lease(struct nfs_client *clp)
1187 {
1188         struct rpc_cred *cred;
1189         struct nfs4_state_recovery_ops *ops =
1190                 nfs4_reboot_recovery_ops[clp->cl_minorversion];
1191         int status = -ENOENT;
1192
1193         cred = ops->get_clid_cred(clp);
1194         if (cred != NULL) {
1195                 status = ops->establish_clid(clp, cred);
1196                 put_rpccred(cred);
1197                 /* Handle case where the user hasn't set up machine creds */
1198                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1199                         nfs4_clear_machine_cred(clp);
1200                         status = -EAGAIN;
1201                 }
1202                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1203                         status = -EPROTONOSUPPORT;
1204         }
1205         return status;
1206 }
1207
1208 #ifdef CONFIG_NFS_V4_1
1209 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1210 {
1211         if (!flags)
1212                 return;
1213         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED) {
1214                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1215                 nfs4_state_start_reclaim_reboot(clp);
1216                 nfs4_schedule_state_recovery(clp);
1217         } else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1218                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1219                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1220                             SEQ4_STATUS_RECALLABLE_STATE_REVOKED |
1221                             SEQ4_STATUS_LEASE_MOVED)) {
1222                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1223                 nfs4_state_start_reclaim_nograce(clp);
1224                 nfs4_schedule_state_recovery(clp);
1225         } else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1226                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1227                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1228                 nfs_expire_all_delegations(clp);
1229 }
1230
1231 static void nfs4_session_recovery_handle_error(struct nfs_client *clp, int err)
1232 {
1233         switch (err) {
1234         case -NFS4ERR_STALE_CLIENTID:
1235                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1236         }
1237 }
1238
1239 static int nfs4_reset_session(struct nfs_client *clp)
1240 {
1241         struct nfs4_session *ses = clp->cl_session;
1242         struct nfs4_slot_table *tbl = &ses->fc_slot_table;
1243         int status;
1244
1245         spin_lock(&tbl->slot_tbl_lock);
1246         set_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state);
1247         if (tbl->highest_used_slotid != -1) {
1248                 INIT_COMPLETION(ses->complete);
1249                 spin_unlock(&tbl->slot_tbl_lock);
1250                 status = wait_for_completion_interruptible(&ses->complete);
1251                 if (status) /* -ERESTARTSYS */
1252                         goto out;
1253         } else {
1254                 spin_unlock(&tbl->slot_tbl_lock);
1255         }
1256
1257         status = nfs4_proc_destroy_session(clp->cl_session);
1258         if (status && status != -NFS4ERR_BADSESSION &&
1259             status != -NFS4ERR_DEADSESSION) {
1260                 nfs4_session_recovery_handle_error(clp, status);
1261                 goto out;
1262         }
1263
1264         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1265         status = nfs4_proc_create_session(clp);
1266         if (status)
1267                 nfs4_session_recovery_handle_error(clp, status);
1268                 /* fall through*/
1269 out:
1270         /* Wake up the next rpc task even on error */
1271         clear_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state);
1272         rpc_wake_up(&clp->cl_session->fc_slot_table.slot_tbl_waitq);
1273         if (status == 0)
1274                 nfs41_setup_state_renewal(clp);
1275         return status;
1276 }
1277
1278 #else /* CONFIG_NFS_V4_1 */
1279 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1280 #endif /* CONFIG_NFS_V4_1 */
1281
1282 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1283  * on EXCHANGE_ID for v4.1
1284  */
1285 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1286 {
1287         if (nfs4_has_session(clp)) {
1288                 switch (status) {
1289                 case -NFS4ERR_DELAY:
1290                 case -NFS4ERR_CLID_INUSE:
1291                 case -EAGAIN:
1292                         break;
1293
1294                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1295                                          * in nfs4_exchange_id */
1296                 default:
1297                         return;
1298                 }
1299         }
1300         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1301 }
1302
1303 static void nfs4_state_manager(struct nfs_client *clp)
1304 {
1305         int status = 0;
1306
1307         /* Ensure exclusive access to NFSv4 state */
1308         for(;;) {
1309                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1310                         /* We're going to have to re-establish a clientid */
1311                         status = nfs4_reclaim_lease(clp);
1312                         if (status) {
1313                                 nfs4_set_lease_expired(clp, status);
1314                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1315                                                         &clp->cl_state))
1316                                         continue;
1317                                 if (clp->cl_cons_state ==
1318                                                         NFS_CS_SESSION_INITING)
1319                                         nfs_mark_client_ready(clp, status);
1320                                 goto out_error;
1321                         }
1322                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1323                 }
1324
1325                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1326                         status = nfs4_check_lease(clp);
1327                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1328                                 continue;
1329                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1330                                 goto out_error;
1331                 }
1332
1333                 /* Initialize or reset the session */
1334                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1335                    && nfs4_has_session(clp)) {
1336                         status = nfs4_reset_session(clp);
1337                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1338                                 continue;
1339                         if (status < 0)
1340                                 goto out_error;
1341                 }
1342
1343                 /* First recover reboot state... */
1344                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1345                         status = nfs4_do_reclaim(clp,
1346                                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1347                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1348                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1349                                 continue;
1350                         nfs4_state_end_reclaim_reboot(clp);
1351                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1352                                 continue;
1353                         if (status < 0)
1354                                 goto out_error;
1355                 }
1356
1357                 /* Now recover expired state... */
1358                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1359                         status = nfs4_do_reclaim(clp,
1360                                 nfs4_nograce_recovery_ops[clp->cl_minorversion]);
1361                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1362                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1363                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1364                                 continue;
1365                         if (status < 0)
1366                                 goto out_error;
1367                 }
1368
1369                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1370                         nfs_client_return_marked_delegations(clp);
1371                         continue;
1372                 }
1373
1374                 nfs4_clear_state_manager_bit(clp);
1375                 /* Did we race with an attempt to give us more work? */
1376                 if (clp->cl_state == 0)
1377                         break;
1378                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1379                         break;
1380         }
1381         return;
1382 out_error:
1383         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1384                         " with error %d\n", clp->cl_hostname, -status);
1385         nfs4_clear_state_manager_bit(clp);
1386 }
1387
1388 static int nfs4_run_state_manager(void *ptr)
1389 {
1390         struct nfs_client *clp = ptr;
1391
1392         allow_signal(SIGKILL);
1393         nfs4_state_manager(clp);
1394         nfs_put_client(clp);
1395         module_put_and_exit(0);
1396         return 0;
1397 }
1398
1399 /*
1400  * Local variables:
1401  *  c-basic-offset: 8
1402  * End:
1403  */