Merge branch 'bkl/procfs' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
[pandora-kernel.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/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 static void nfs4_end_drain_session(struct nfs_client *clp)
139 {
140         struct nfs4_session *ses = clp->cl_session;
141         int max_slots;
142
143         if (test_and_clear_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state)) {
144                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
145                 max_slots = ses->fc_slot_table.max_slots;
146                 while (max_slots--) {
147                         struct rpc_task *task;
148
149                         task = rpc_wake_up_next(&ses->fc_slot_table.
150                                                 slot_tbl_waitq);
151                         if (!task)
152                                 break;
153                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
154                 }
155                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
156         }
157 }
158
159 static int nfs4_begin_drain_session(struct nfs_client *clp)
160 {
161         struct nfs4_session *ses = clp->cl_session;
162         struct nfs4_slot_table *tbl = &ses->fc_slot_table;
163
164         spin_lock(&tbl->slot_tbl_lock);
165         set_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state);
166         if (tbl->highest_used_slotid != -1) {
167                 INIT_COMPLETION(ses->complete);
168                 spin_unlock(&tbl->slot_tbl_lock);
169                 return wait_for_completion_interruptible(&ses->complete);
170         }
171         spin_unlock(&tbl->slot_tbl_lock);
172         return 0;
173 }
174
175 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
176 {
177         int status;
178
179         nfs4_begin_drain_session(clp);
180         status = nfs4_proc_exchange_id(clp, cred);
181         if (status != 0)
182                 goto out;
183         status = nfs4_proc_create_session(clp);
184         if (status != 0)
185                 goto out;
186         nfs41_setup_state_renewal(clp);
187         nfs_mark_client_ready(clp, NFS_CS_READY);
188 out:
189         return status;
190 }
191
192 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
193 {
194         struct rpc_cred *cred;
195
196         spin_lock(&clp->cl_lock);
197         cred = nfs4_get_machine_cred_locked(clp);
198         spin_unlock(&clp->cl_lock);
199         return cred;
200 }
201
202 #endif /* CONFIG_NFS_V4_1 */
203
204 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
205 {
206         struct nfs4_state_owner *sp;
207         struct rb_node *pos;
208         struct rpc_cred *cred;
209
210         spin_lock(&clp->cl_lock);
211         cred = nfs4_get_machine_cred_locked(clp);
212         if (cred != NULL)
213                 goto out;
214         pos = rb_first(&clp->cl_state_owners);
215         if (pos != NULL) {
216                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
217                 cred = get_rpccred(sp->so_cred);
218         }
219 out:
220         spin_unlock(&clp->cl_lock);
221         return cred;
222 }
223
224 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
225                 __u64 minval, int maxbits)
226 {
227         struct rb_node **p, *parent;
228         struct nfs_unique_id *pos;
229         __u64 mask = ~0ULL;
230
231         if (maxbits < 64)
232                 mask = (1ULL << maxbits) - 1ULL;
233
234         /* Ensure distribution is more or less flat */
235         get_random_bytes(&new->id, sizeof(new->id));
236         new->id &= mask;
237         if (new->id < minval)
238                 new->id += minval;
239 retry:
240         p = &root->rb_node;
241         parent = NULL;
242
243         while (*p != NULL) {
244                 parent = *p;
245                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
246
247                 if (new->id < pos->id)
248                         p = &(*p)->rb_left;
249                 else if (new->id > pos->id)
250                         p = &(*p)->rb_right;
251                 else
252                         goto id_exists;
253         }
254         rb_link_node(&new->rb_node, parent, p);
255         rb_insert_color(&new->rb_node, root);
256         return;
257 id_exists:
258         for (;;) {
259                 new->id++;
260                 if (new->id < minval || (new->id & mask) != new->id) {
261                         new->id = minval;
262                         break;
263                 }
264                 parent = rb_next(parent);
265                 if (parent == NULL)
266                         break;
267                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
268                 if (new->id < pos->id)
269                         break;
270         }
271         goto retry;
272 }
273
274 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
275 {
276         rb_erase(&id->rb_node, root);
277 }
278
279 static struct nfs4_state_owner *
280 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
281 {
282         struct nfs_client *clp = server->nfs_client;
283         struct rb_node **p = &clp->cl_state_owners.rb_node,
284                        *parent = NULL;
285         struct nfs4_state_owner *sp, *res = NULL;
286
287         while (*p != NULL) {
288                 parent = *p;
289                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
290
291                 if (server < sp->so_server) {
292                         p = &parent->rb_left;
293                         continue;
294                 }
295                 if (server > sp->so_server) {
296                         p = &parent->rb_right;
297                         continue;
298                 }
299                 if (cred < sp->so_cred)
300                         p = &parent->rb_left;
301                 else if (cred > sp->so_cred)
302                         p = &parent->rb_right;
303                 else {
304                         atomic_inc(&sp->so_count);
305                         res = sp;
306                         break;
307                 }
308         }
309         return res;
310 }
311
312 static struct nfs4_state_owner *
313 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
314 {
315         struct rb_node **p = &clp->cl_state_owners.rb_node,
316                        *parent = NULL;
317         struct nfs4_state_owner *sp;
318
319         while (*p != NULL) {
320                 parent = *p;
321                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
322
323                 if (new->so_server < sp->so_server) {
324                         p = &parent->rb_left;
325                         continue;
326                 }
327                 if (new->so_server > sp->so_server) {
328                         p = &parent->rb_right;
329                         continue;
330                 }
331                 if (new->so_cred < sp->so_cred)
332                         p = &parent->rb_left;
333                 else if (new->so_cred > sp->so_cred)
334                         p = &parent->rb_right;
335                 else {
336                         atomic_inc(&sp->so_count);
337                         return sp;
338                 }
339         }
340         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
341         rb_link_node(&new->so_client_node, parent, p);
342         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
343         return new;
344 }
345
346 static void
347 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
348 {
349         if (!RB_EMPTY_NODE(&sp->so_client_node))
350                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
351         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
352 }
353
354 /*
355  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
356  * create a new state_owner.
357  *
358  */
359 static struct nfs4_state_owner *
360 nfs4_alloc_state_owner(void)
361 {
362         struct nfs4_state_owner *sp;
363
364         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
365         if (!sp)
366                 return NULL;
367         spin_lock_init(&sp->so_lock);
368         INIT_LIST_HEAD(&sp->so_states);
369         INIT_LIST_HEAD(&sp->so_delegations);
370         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
371         sp->so_seqid.sequence = &sp->so_sequence;
372         spin_lock_init(&sp->so_sequence.lock);
373         INIT_LIST_HEAD(&sp->so_sequence.list);
374         atomic_set(&sp->so_count, 1);
375         return sp;
376 }
377
378 static void
379 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
380 {
381         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
382                 struct nfs_client *clp = sp->so_client;
383
384                 spin_lock(&clp->cl_lock);
385                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
386                 RB_CLEAR_NODE(&sp->so_client_node);
387                 spin_unlock(&clp->cl_lock);
388         }
389 }
390
391 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
392 {
393         struct nfs_client *clp = server->nfs_client;
394         struct nfs4_state_owner *sp, *new;
395
396         spin_lock(&clp->cl_lock);
397         sp = nfs4_find_state_owner(server, cred);
398         spin_unlock(&clp->cl_lock);
399         if (sp != NULL)
400                 return sp;
401         new = nfs4_alloc_state_owner();
402         if (new == NULL)
403                 return NULL;
404         new->so_client = clp;
405         new->so_server = server;
406         new->so_cred = cred;
407         spin_lock(&clp->cl_lock);
408         sp = nfs4_insert_state_owner(clp, new);
409         spin_unlock(&clp->cl_lock);
410         if (sp == new)
411                 get_rpccred(cred);
412         else {
413                 rpc_destroy_wait_queue(&new->so_sequence.wait);
414                 kfree(new);
415         }
416         return sp;
417 }
418
419 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
420 {
421         struct nfs_client *clp = sp->so_client;
422         struct rpc_cred *cred = sp->so_cred;
423
424         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
425                 return;
426         nfs4_remove_state_owner(clp, sp);
427         spin_unlock(&clp->cl_lock);
428         rpc_destroy_wait_queue(&sp->so_sequence.wait);
429         put_rpccred(cred);
430         kfree(sp);
431 }
432
433 static struct nfs4_state *
434 nfs4_alloc_open_state(void)
435 {
436         struct nfs4_state *state;
437
438         state = kzalloc(sizeof(*state), GFP_KERNEL);
439         if (!state)
440                 return NULL;
441         atomic_set(&state->count, 1);
442         INIT_LIST_HEAD(&state->lock_states);
443         spin_lock_init(&state->state_lock);
444         seqlock_init(&state->seqlock);
445         return state;
446 }
447
448 void
449 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
450 {
451         if (state->state == fmode)
452                 return;
453         /* NB! List reordering - see the reclaim code for why.  */
454         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
455                 if (fmode & FMODE_WRITE)
456                         list_move(&state->open_states, &state->owner->so_states);
457                 else
458                         list_move_tail(&state->open_states, &state->owner->so_states);
459         }
460         state->state = fmode;
461 }
462
463 static struct nfs4_state *
464 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
465 {
466         struct nfs_inode *nfsi = NFS_I(inode);
467         struct nfs4_state *state;
468
469         list_for_each_entry(state, &nfsi->open_states, inode_states) {
470                 if (state->owner != owner)
471                         continue;
472                 if (atomic_inc_not_zero(&state->count))
473                         return state;
474         }
475         return NULL;
476 }
477
478 static void
479 nfs4_free_open_state(struct nfs4_state *state)
480 {
481         kfree(state);
482 }
483
484 struct nfs4_state *
485 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
486 {
487         struct nfs4_state *state, *new;
488         struct nfs_inode *nfsi = NFS_I(inode);
489
490         spin_lock(&inode->i_lock);
491         state = __nfs4_find_state_byowner(inode, owner);
492         spin_unlock(&inode->i_lock);
493         if (state)
494                 goto out;
495         new = nfs4_alloc_open_state();
496         spin_lock(&owner->so_lock);
497         spin_lock(&inode->i_lock);
498         state = __nfs4_find_state_byowner(inode, owner);
499         if (state == NULL && new != NULL) {
500                 state = new;
501                 state->owner = owner;
502                 atomic_inc(&owner->so_count);
503                 list_add(&state->inode_states, &nfsi->open_states);
504                 state->inode = igrab(inode);
505                 spin_unlock(&inode->i_lock);
506                 /* Note: The reclaim code dictates that we add stateless
507                  * and read-only stateids to the end of the list */
508                 list_add_tail(&state->open_states, &owner->so_states);
509                 spin_unlock(&owner->so_lock);
510         } else {
511                 spin_unlock(&inode->i_lock);
512                 spin_unlock(&owner->so_lock);
513                 if (new)
514                         nfs4_free_open_state(new);
515         }
516 out:
517         return state;
518 }
519
520 void nfs4_put_open_state(struct nfs4_state *state)
521 {
522         struct inode *inode = state->inode;
523         struct nfs4_state_owner *owner = state->owner;
524
525         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
526                 return;
527         spin_lock(&inode->i_lock);
528         list_del(&state->inode_states);
529         list_del(&state->open_states);
530         spin_unlock(&inode->i_lock);
531         spin_unlock(&owner->so_lock);
532         iput(inode);
533         nfs4_free_open_state(state);
534         nfs4_put_state_owner(owner);
535 }
536
537 /*
538  * Close the current file.
539  */
540 static void __nfs4_close(struct path *path, struct nfs4_state *state, fmode_t fmode, int wait)
541 {
542         struct nfs4_state_owner *owner = state->owner;
543         int call_close = 0;
544         fmode_t newstate;
545
546         atomic_inc(&owner->so_count);
547         /* Protect against nfs4_find_state() */
548         spin_lock(&owner->so_lock);
549         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
550                 case FMODE_READ:
551                         state->n_rdonly--;
552                         break;
553                 case FMODE_WRITE:
554                         state->n_wronly--;
555                         break;
556                 case FMODE_READ|FMODE_WRITE:
557                         state->n_rdwr--;
558         }
559         newstate = FMODE_READ|FMODE_WRITE;
560         if (state->n_rdwr == 0) {
561                 if (state->n_rdonly == 0) {
562                         newstate &= ~FMODE_READ;
563                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
564                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
565                 }
566                 if (state->n_wronly == 0) {
567                         newstate &= ~FMODE_WRITE;
568                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
569                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
570                 }
571                 if (newstate == 0)
572                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
573         }
574         nfs4_state_set_mode_locked(state, newstate);
575         spin_unlock(&owner->so_lock);
576
577         if (!call_close) {
578                 nfs4_put_open_state(state);
579                 nfs4_put_state_owner(owner);
580         } else
581                 nfs4_do_close(path, state, wait);
582 }
583
584 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
585 {
586         __nfs4_close(path, state, fmode, 0);
587 }
588
589 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
590 {
591         __nfs4_close(path, state, fmode, 1);
592 }
593
594 /*
595  * Search the state->lock_states for an existing lock_owner
596  * that is compatible with current->files
597  */
598 static struct nfs4_lock_state *
599 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
600 {
601         struct nfs4_lock_state *pos;
602         list_for_each_entry(pos, &state->lock_states, ls_locks) {
603                 if (pos->ls_owner != fl_owner)
604                         continue;
605                 atomic_inc(&pos->ls_count);
606                 return pos;
607         }
608         return NULL;
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_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
617 {
618         struct nfs4_lock_state *lsp;
619         struct nfs_client *clp = state->owner->so_client;
620
621         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
622         if (lsp == NULL)
623                 return NULL;
624         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
625         spin_lock_init(&lsp->ls_sequence.lock);
626         INIT_LIST_HEAD(&lsp->ls_sequence.list);
627         lsp->ls_seqid.sequence = &lsp->ls_sequence;
628         atomic_set(&lsp->ls_count, 1);
629         lsp->ls_state = state;
630         lsp->ls_owner = fl_owner;
631         spin_lock(&clp->cl_lock);
632         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
633         spin_unlock(&clp->cl_lock);
634         INIT_LIST_HEAD(&lsp->ls_locks);
635         return lsp;
636 }
637
638 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
639 {
640         struct nfs_client *clp = lsp->ls_state->owner->so_client;
641
642         spin_lock(&clp->cl_lock);
643         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
644         spin_unlock(&clp->cl_lock);
645         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
646         kfree(lsp);
647 }
648
649 /*
650  * Return a compatible lock_state. If no initialized lock_state structure
651  * exists, return an uninitialized one.
652  *
653  */
654 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
655 {
656         struct nfs4_lock_state *lsp, *new = NULL;
657         
658         for(;;) {
659                 spin_lock(&state->state_lock);
660                 lsp = __nfs4_find_lock_state(state, owner);
661                 if (lsp != NULL)
662                         break;
663                 if (new != NULL) {
664                         list_add(&new->ls_locks, &state->lock_states);
665                         set_bit(LK_STATE_IN_USE, &state->flags);
666                         lsp = new;
667                         new = NULL;
668                         break;
669                 }
670                 spin_unlock(&state->state_lock);
671                 new = nfs4_alloc_lock_state(state, owner);
672                 if (new == NULL)
673                         return NULL;
674         }
675         spin_unlock(&state->state_lock);
676         if (new != NULL)
677                 nfs4_free_lock_state(new);
678         return lsp;
679 }
680
681 /*
682  * Release reference to lock_state, and free it if we see that
683  * it is no longer in use
684  */
685 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
686 {
687         struct nfs4_state *state;
688
689         if (lsp == NULL)
690                 return;
691         state = lsp->ls_state;
692         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
693                 return;
694         list_del(&lsp->ls_locks);
695         if (list_empty(&state->lock_states))
696                 clear_bit(LK_STATE_IN_USE, &state->flags);
697         spin_unlock(&state->state_lock);
698         nfs4_free_lock_state(lsp);
699 }
700
701 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
702 {
703         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
704
705         dst->fl_u.nfs4_fl.owner = lsp;
706         atomic_inc(&lsp->ls_count);
707 }
708
709 static void nfs4_fl_release_lock(struct file_lock *fl)
710 {
711         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
712 }
713
714 static const struct file_lock_operations nfs4_fl_lock_ops = {
715         .fl_copy_lock = nfs4_fl_copy_lock,
716         .fl_release_private = nfs4_fl_release_lock,
717 };
718
719 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
720 {
721         struct nfs4_lock_state *lsp;
722
723         if (fl->fl_ops != NULL)
724                 return 0;
725         lsp = nfs4_get_lock_state(state, fl->fl_owner);
726         if (lsp == NULL)
727                 return -ENOMEM;
728         fl->fl_u.nfs4_fl.owner = lsp;
729         fl->fl_ops = &nfs4_fl_lock_ops;
730         return 0;
731 }
732
733 /*
734  * Byte-range lock aware utility to initialize the stateid of read/write
735  * requests.
736  */
737 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
738 {
739         struct nfs4_lock_state *lsp;
740         int seq;
741
742         do {
743                 seq = read_seqbegin(&state->seqlock);
744                 memcpy(dst, &state->stateid, sizeof(*dst));
745         } while (read_seqretry(&state->seqlock, seq));
746         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
747                 return;
748
749         spin_lock(&state->state_lock);
750         lsp = __nfs4_find_lock_state(state, fl_owner);
751         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
752                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
753         spin_unlock(&state->state_lock);
754         nfs4_put_lock_state(lsp);
755 }
756
757 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
758 {
759         struct nfs_seqid *new;
760
761         new = kmalloc(sizeof(*new), GFP_KERNEL);
762         if (new != NULL) {
763                 new->sequence = counter;
764                 INIT_LIST_HEAD(&new->list);
765         }
766         return new;
767 }
768
769 void nfs_release_seqid(struct nfs_seqid *seqid)
770 {
771         if (!list_empty(&seqid->list)) {
772                 struct rpc_sequence *sequence = seqid->sequence->sequence;
773
774                 spin_lock(&sequence->lock);
775                 list_del_init(&seqid->list);
776                 spin_unlock(&sequence->lock);
777                 rpc_wake_up(&sequence->wait);
778         }
779 }
780
781 void nfs_free_seqid(struct nfs_seqid *seqid)
782 {
783         nfs_release_seqid(seqid);
784         kfree(seqid);
785 }
786
787 /*
788  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
789  * failed with a seqid incrementing error -
790  * see comments nfs_fs.h:seqid_mutating_error()
791  */
792 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
793 {
794         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
795         switch (status) {
796                 case 0:
797                         break;
798                 case -NFS4ERR_BAD_SEQID:
799                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
800                                 return;
801                         printk(KERN_WARNING "NFS: v4 server returned a bad"
802                                         " sequence-id error on an"
803                                         " unconfirmed sequence %p!\n",
804                                         seqid->sequence);
805                 case -NFS4ERR_STALE_CLIENTID:
806                 case -NFS4ERR_STALE_STATEID:
807                 case -NFS4ERR_BAD_STATEID:
808                 case -NFS4ERR_BADXDR:
809                 case -NFS4ERR_RESOURCE:
810                 case -NFS4ERR_NOFILEHANDLE:
811                         /* Non-seqid mutating errors */
812                         return;
813         };
814         /*
815          * Note: no locking needed as we are guaranteed to be first
816          * on the sequence list
817          */
818         seqid->sequence->counter++;
819 }
820
821 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
822 {
823         struct nfs4_state_owner *sp = container_of(seqid->sequence,
824                                         struct nfs4_state_owner, so_seqid);
825         struct nfs_server *server = sp->so_server;
826
827         if (status == -NFS4ERR_BAD_SEQID)
828                 nfs4_drop_state_owner(sp);
829         if (!nfs4_has_session(server->nfs_client))
830                 nfs_increment_seqid(status, seqid);
831 }
832
833 /*
834  * Increment the seqid if the LOCK/LOCKU succeeded, or
835  * failed with a seqid incrementing error -
836  * see comments nfs_fs.h:seqid_mutating_error()
837  */
838 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
839 {
840         nfs_increment_seqid(status, seqid);
841 }
842
843 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
844 {
845         struct rpc_sequence *sequence = seqid->sequence->sequence;
846         int status = 0;
847
848         spin_lock(&sequence->lock);
849         if (list_empty(&seqid->list))
850                 list_add_tail(&seqid->list, &sequence->list);
851         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
852                 goto unlock;
853         rpc_sleep_on(&sequence->wait, task, NULL);
854         status = -EAGAIN;
855 unlock:
856         spin_unlock(&sequence->lock);
857         return status;
858 }
859
860 static int nfs4_run_state_manager(void *);
861
862 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
863 {
864         smp_mb__before_clear_bit();
865         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
866         smp_mb__after_clear_bit();
867         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
868         rpc_wake_up(&clp->cl_rpcwaitq);
869 }
870
871 /*
872  * Schedule the nfs_client asynchronous state management routine
873  */
874 void nfs4_schedule_state_manager(struct nfs_client *clp)
875 {
876         struct task_struct *task;
877
878         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
879                 return;
880         __module_get(THIS_MODULE);
881         atomic_inc(&clp->cl_count);
882         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
883                                 rpc_peeraddr2str(clp->cl_rpcclient,
884                                                         RPC_DISPLAY_ADDR));
885         if (!IS_ERR(task))
886                 return;
887         nfs4_clear_state_manager_bit(clp);
888         nfs_put_client(clp);
889         module_put(THIS_MODULE);
890 }
891
892 /*
893  * Schedule a state recovery attempt
894  */
895 void nfs4_schedule_state_recovery(struct nfs_client *clp)
896 {
897         if (!clp)
898                 return;
899         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
900                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
901         nfs4_schedule_state_manager(clp);
902 }
903
904 int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
905 {
906
907         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
908         /* Don't recover state that expired before the reboot */
909         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
910                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
911                 return 0;
912         }
913         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
914         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
915         return 1;
916 }
917
918 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
919 {
920         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
921         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
922         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
923         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
924         return 1;
925 }
926
927 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
928 {
929         struct inode *inode = state->inode;
930         struct nfs_inode *nfsi = NFS_I(inode);
931         struct file_lock *fl;
932         int status = 0;
933
934         if (inode->i_flock == NULL)
935                 return 0;
936
937         /* Guard against delegation returns and new lock/unlock calls */
938         down_write(&nfsi->rwsem);
939         /* Protect inode->i_flock using the BKL */
940         lock_kernel();
941         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
942                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
943                         continue;
944                 if (nfs_file_open_context(fl->fl_file)->state != state)
945                         continue;
946                 unlock_kernel();
947                 status = ops->recover_lock(state, fl);
948                 switch (status) {
949                         case 0:
950                                 break;
951                         case -ESTALE:
952                         case -NFS4ERR_ADMIN_REVOKED:
953                         case -NFS4ERR_STALE_STATEID:
954                         case -NFS4ERR_BAD_STATEID:
955                         case -NFS4ERR_EXPIRED:
956                         case -NFS4ERR_NO_GRACE:
957                         case -NFS4ERR_STALE_CLIENTID:
958                         case -NFS4ERR_BADSESSION:
959                         case -NFS4ERR_BADSLOT:
960                         case -NFS4ERR_BAD_HIGH_SLOT:
961                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
962                                 goto out;
963                         default:
964                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
965                                                 __func__, status);
966                         case -ENOMEM:
967                         case -NFS4ERR_DENIED:
968                         case -NFS4ERR_RECLAIM_BAD:
969                         case -NFS4ERR_RECLAIM_CONFLICT:
970                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
971                                 status = 0;
972                 }
973                 lock_kernel();
974         }
975         unlock_kernel();
976 out:
977         up_write(&nfsi->rwsem);
978         return status;
979 }
980
981 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
982 {
983         struct nfs4_state *state;
984         struct nfs4_lock_state *lock;
985         int status = 0;
986
987         /* Note: we rely on the sp->so_states list being ordered 
988          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
989          * states first.
990          * This is needed to ensure that the server won't give us any
991          * read delegations that we have to return if, say, we are
992          * recovering after a network partition or a reboot from a
993          * server that doesn't support a grace period.
994          */
995 restart:
996         spin_lock(&sp->so_lock);
997         list_for_each_entry(state, &sp->so_states, open_states) {
998                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
999                         continue;
1000                 if (state->state == 0)
1001                         continue;
1002                 atomic_inc(&state->count);
1003                 spin_unlock(&sp->so_lock);
1004                 status = ops->recover_open(sp, state);
1005                 if (status >= 0) {
1006                         status = nfs4_reclaim_locks(state, ops);
1007                         if (status >= 0) {
1008                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1009                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1010                                                 printk("%s: Lock reclaim failed!\n",
1011                                                         __func__);
1012                                 }
1013                                 nfs4_put_open_state(state);
1014                                 goto restart;
1015                         }
1016                 }
1017                 switch (status) {
1018                         default:
1019                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1020                                                 __func__, status);
1021                         case -ENOENT:
1022                         case -ENOMEM:
1023                         case -ESTALE:
1024                                 /*
1025                                  * Open state on this file cannot be recovered
1026                                  * All we can do is revert to using the zero stateid.
1027                                  */
1028                                 memset(state->stateid.data, 0,
1029                                         sizeof(state->stateid.data));
1030                                 /* Mark the file as being 'closed' */
1031                                 state->state = 0;
1032                                 break;
1033                         case -NFS4ERR_ADMIN_REVOKED:
1034                         case -NFS4ERR_STALE_STATEID:
1035                         case -NFS4ERR_BAD_STATEID:
1036                         case -NFS4ERR_RECLAIM_BAD:
1037                         case -NFS4ERR_RECLAIM_CONFLICT:
1038                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
1039                                 break;
1040                         case -NFS4ERR_EXPIRED:
1041                         case -NFS4ERR_NO_GRACE:
1042                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
1043                         case -NFS4ERR_STALE_CLIENTID:
1044                         case -NFS4ERR_BADSESSION:
1045                         case -NFS4ERR_BADSLOT:
1046                         case -NFS4ERR_BAD_HIGH_SLOT:
1047                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1048                                 goto out_err;
1049                 }
1050                 nfs4_put_open_state(state);
1051                 goto restart;
1052         }
1053         spin_unlock(&sp->so_lock);
1054         return 0;
1055 out_err:
1056         nfs4_put_open_state(state);
1057         return status;
1058 }
1059
1060 static void nfs4_clear_open_state(struct nfs4_state *state)
1061 {
1062         struct nfs4_lock_state *lock;
1063
1064         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1065         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1066         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1067         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1068         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1069                 lock->ls_seqid.flags = 0;
1070                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1071         }
1072 }
1073
1074 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1075 {
1076         struct nfs4_state_owner *sp;
1077         struct rb_node *pos;
1078         struct nfs4_state *state;
1079
1080         /* Reset all sequence ids to zero */
1081         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1082                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1083                 sp->so_seqid.flags = 0;
1084                 spin_lock(&sp->so_lock);
1085                 list_for_each_entry(state, &sp->so_states, open_states) {
1086                         if (mark_reclaim(clp, state))
1087                                 nfs4_clear_open_state(state);
1088                 }
1089                 spin_unlock(&sp->so_lock);
1090         }
1091 }
1092
1093 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1094 {
1095         /* Mark all delegations for reclaim */
1096         nfs_delegation_mark_reclaim(clp);
1097         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1098 }
1099
1100 static void nfs4_reclaim_complete(struct nfs_client *clp,
1101                                  const struct nfs4_state_recovery_ops *ops)
1102 {
1103         /* Notify the server we're done reclaiming our state */
1104         if (ops->reclaim_complete)
1105                 (void)ops->reclaim_complete(clp);
1106 }
1107
1108 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1109 {
1110         struct nfs4_state_owner *sp;
1111         struct rb_node *pos;
1112         struct nfs4_state *state;
1113
1114         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1115                 return;
1116
1117         nfs4_reclaim_complete(clp,
1118                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1119
1120         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1121                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1122                 spin_lock(&sp->so_lock);
1123                 list_for_each_entry(state, &sp->so_states, open_states) {
1124                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1125                                 continue;
1126                         nfs4_state_mark_reclaim_nograce(clp, state);
1127                 }
1128                 spin_unlock(&sp->so_lock);
1129         }
1130
1131         nfs_delegation_reap_unclaimed(clp);
1132 }
1133
1134 static void nfs_delegation_clear_all(struct nfs_client *clp)
1135 {
1136         nfs_delegation_mark_reclaim(clp);
1137         nfs_delegation_reap_unclaimed(clp);
1138 }
1139
1140 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1141 {
1142         nfs_delegation_clear_all(clp);
1143         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1144 }
1145
1146 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1147 {
1148         switch (error) {
1149                 case -NFS4ERR_CB_PATH_DOWN:
1150                         nfs_handle_cb_pathdown(clp);
1151                         return 0;
1152                 case -NFS4ERR_NO_GRACE:
1153                         nfs4_state_end_reclaim_reboot(clp);
1154                         return 0;
1155                 case -NFS4ERR_STALE_CLIENTID:
1156                 case -NFS4ERR_LEASE_MOVED:
1157                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1158                         nfs4_state_end_reclaim_reboot(clp);
1159                         nfs4_state_start_reclaim_reboot(clp);
1160                         break;
1161                 case -NFS4ERR_EXPIRED:
1162                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1163                         nfs4_state_start_reclaim_nograce(clp);
1164                         break;
1165                 case -NFS4ERR_BADSESSION:
1166                 case -NFS4ERR_BADSLOT:
1167                 case -NFS4ERR_BAD_HIGH_SLOT:
1168                 case -NFS4ERR_DEADSESSION:
1169                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1170                 case -NFS4ERR_SEQ_FALSE_RETRY:
1171                 case -NFS4ERR_SEQ_MISORDERED:
1172                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1173                         /* Zero session reset errors */
1174                         return 0;
1175         }
1176         return error;
1177 }
1178
1179 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1180 {
1181         struct rb_node *pos;
1182         int status = 0;
1183
1184 restart:
1185         spin_lock(&clp->cl_lock);
1186         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1187                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1188                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1189                         continue;
1190                 atomic_inc(&sp->so_count);
1191                 spin_unlock(&clp->cl_lock);
1192                 status = nfs4_reclaim_open_state(sp, ops);
1193                 if (status < 0) {
1194                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1195                         nfs4_put_state_owner(sp);
1196                         return nfs4_recovery_handle_error(clp, status);
1197                 }
1198                 nfs4_put_state_owner(sp);
1199                 goto restart;
1200         }
1201         spin_unlock(&clp->cl_lock);
1202         return status;
1203 }
1204
1205 static int nfs4_check_lease(struct nfs_client *clp)
1206 {
1207         struct rpc_cred *cred;
1208         struct nfs4_state_maintenance_ops *ops =
1209                 nfs4_state_renewal_ops[clp->cl_minorversion];
1210         int status = -NFS4ERR_EXPIRED;
1211
1212         /* Is the client already known to have an expired lease? */
1213         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1214                 return 0;
1215         spin_lock(&clp->cl_lock);
1216         cred = ops->get_state_renewal_cred_locked(clp);
1217         spin_unlock(&clp->cl_lock);
1218         if (cred == NULL) {
1219                 cred = nfs4_get_setclientid_cred(clp);
1220                 if (cred == NULL)
1221                         goto out;
1222         }
1223         status = ops->renew_lease(clp, cred);
1224         put_rpccred(cred);
1225 out:
1226         return nfs4_recovery_handle_error(clp, status);
1227 }
1228
1229 static int nfs4_reclaim_lease(struct nfs_client *clp)
1230 {
1231         struct rpc_cred *cred;
1232         struct nfs4_state_recovery_ops *ops =
1233                 nfs4_reboot_recovery_ops[clp->cl_minorversion];
1234         int status = -ENOENT;
1235
1236         cred = ops->get_clid_cred(clp);
1237         if (cred != NULL) {
1238                 status = ops->establish_clid(clp, cred);
1239                 put_rpccred(cred);
1240                 /* Handle case where the user hasn't set up machine creds */
1241                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1242                         nfs4_clear_machine_cred(clp);
1243                         status = -EAGAIN;
1244                 }
1245                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1246                         status = -EPROTONOSUPPORT;
1247         }
1248         return status;
1249 }
1250
1251 #ifdef CONFIG_NFS_V4_1
1252 void nfs41_handle_recall_slot(struct nfs_client *clp)
1253 {
1254         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1255         nfs4_schedule_state_recovery(clp);
1256 }
1257
1258 static void nfs4_reset_all_state(struct nfs_client *clp)
1259 {
1260         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1261                 clp->cl_boot_time = CURRENT_TIME;
1262                 nfs4_state_start_reclaim_nograce(clp);
1263                 nfs4_schedule_state_recovery(clp);
1264         }
1265 }
1266
1267 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1268 {
1269         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1270                 nfs4_state_start_reclaim_reboot(clp);
1271                 nfs4_schedule_state_recovery(clp);
1272         }
1273 }
1274
1275 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1276 {
1277         /* Temporary */
1278         nfs4_reset_all_state(clp);
1279 }
1280
1281 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1282 {
1283         /* This will need to handle layouts too */
1284         nfs_expire_all_delegations(clp);
1285 }
1286
1287 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1288 {
1289         nfs_expire_all_delegations(clp);
1290         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1291                 nfs4_schedule_state_recovery(clp);
1292 }
1293
1294 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1295 {
1296         if (!flags)
1297                 return;
1298         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1299                 nfs41_handle_server_reboot(clp);
1300         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1301                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1302                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1303                             SEQ4_STATUS_LEASE_MOVED))
1304                 nfs41_handle_state_revoked(clp);
1305         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1306                 nfs41_handle_recallable_state_revoked(clp);
1307         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1308                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1309                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1310                 nfs41_handle_cb_path_down(clp);
1311 }
1312
1313 static int nfs4_reset_session(struct nfs_client *clp)
1314 {
1315         int status;
1316
1317         nfs4_begin_drain_session(clp);
1318         status = nfs4_proc_destroy_session(clp->cl_session);
1319         if (status && status != -NFS4ERR_BADSESSION &&
1320             status != -NFS4ERR_DEADSESSION) {
1321                 status = nfs4_recovery_handle_error(clp, status);
1322                 goto out;
1323         }
1324
1325         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1326         status = nfs4_proc_create_session(clp);
1327         if (status) {
1328                 status = nfs4_recovery_handle_error(clp, status);
1329                 goto out;
1330         }
1331         /* create_session negotiated new slot table */
1332         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1333
1334          /* Let the state manager reestablish state */
1335         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1336                 nfs41_setup_state_renewal(clp);
1337 out:
1338         return status;
1339 }
1340
1341 static int nfs4_recall_slot(struct nfs_client *clp)
1342 {
1343         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1344         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1345         struct nfs4_slot *new, *old;
1346         int i;
1347
1348         nfs4_begin_drain_session(clp);
1349         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1350                       GFP_KERNEL);
1351         if (!new)
1352                 return -ENOMEM;
1353
1354         spin_lock(&fc_tbl->slot_tbl_lock);
1355         for (i = 0; i < fc_tbl->target_max_slots; i++)
1356                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1357         old = fc_tbl->slots;
1358         fc_tbl->slots = new;
1359         fc_tbl->max_slots = fc_tbl->target_max_slots;
1360         fc_tbl->target_max_slots = 0;
1361         fc_attrs->max_reqs = fc_tbl->max_slots;
1362         spin_unlock(&fc_tbl->slot_tbl_lock);
1363
1364         kfree(old);
1365         nfs4_end_drain_session(clp);
1366         return 0;
1367 }
1368
1369 #else /* CONFIG_NFS_V4_1 */
1370 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1371 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1372 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1373 #endif /* CONFIG_NFS_V4_1 */
1374
1375 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1376  * on EXCHANGE_ID for v4.1
1377  */
1378 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1379 {
1380         if (nfs4_has_session(clp)) {
1381                 switch (status) {
1382                 case -NFS4ERR_DELAY:
1383                 case -NFS4ERR_CLID_INUSE:
1384                 case -EAGAIN:
1385                 case -EKEYEXPIRED:
1386                         break;
1387
1388                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1389                                          * in nfs4_exchange_id */
1390                 default:
1391                         return;
1392                 }
1393         }
1394         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1395 }
1396
1397 static void nfs4_state_manager(struct nfs_client *clp)
1398 {
1399         int status = 0;
1400
1401         /* Ensure exclusive access to NFSv4 state */
1402         for(;;) {
1403                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1404                         /* We're going to have to re-establish a clientid */
1405                         status = nfs4_reclaim_lease(clp);
1406                         if (status) {
1407                                 nfs4_set_lease_expired(clp, status);
1408                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1409                                                         &clp->cl_state))
1410                                         continue;
1411                                 if (clp->cl_cons_state ==
1412                                                         NFS_CS_SESSION_INITING)
1413                                         nfs_mark_client_ready(clp, status);
1414                                 goto out_error;
1415                         }
1416                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1417                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1418                 }
1419
1420                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1421                         status = nfs4_check_lease(clp);
1422                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1423                                 continue;
1424                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1425                                 goto out_error;
1426                 }
1427
1428                 /* Initialize or reset the session */
1429                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1430                    && nfs4_has_session(clp)) {
1431                         status = nfs4_reset_session(clp);
1432                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1433                                 continue;
1434                         if (status < 0)
1435                                 goto out_error;
1436                 }
1437
1438                 /* First recover reboot state... */
1439                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1440                         status = nfs4_do_reclaim(clp,
1441                                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1442                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1443                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1444                                 continue;
1445                         nfs4_state_end_reclaim_reboot(clp);
1446                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1447                                 continue;
1448                         if (status < 0)
1449                                 goto out_error;
1450                 }
1451
1452                 /* Now recover expired state... */
1453                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1454                         status = nfs4_do_reclaim(clp,
1455                                 nfs4_nograce_recovery_ops[clp->cl_minorversion]);
1456                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1457                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1458                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1459                                 continue;
1460                         if (status < 0)
1461                                 goto out_error;
1462                 }
1463
1464                 nfs4_end_drain_session(clp);
1465                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1466                         nfs_client_return_marked_delegations(clp);
1467                         continue;
1468                 }
1469                 /* Recall session slots */
1470                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1471                    && nfs4_has_session(clp)) {
1472                         status = nfs4_recall_slot(clp);
1473                         if (status < 0)
1474                                 goto out_error;
1475                         continue;
1476                 }
1477
1478
1479                 nfs4_clear_state_manager_bit(clp);
1480                 /* Did we race with an attempt to give us more work? */
1481                 if (clp->cl_state == 0)
1482                         break;
1483                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1484                         break;
1485         }
1486         return;
1487 out_error:
1488         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1489                         " with error %d\n", clp->cl_hostname, -status);
1490         nfs4_end_drain_session(clp);
1491         nfs4_clear_state_manager_bit(clp);
1492 }
1493
1494 static int nfs4_run_state_manager(void *ptr)
1495 {
1496         struct nfs_client *clp = ptr;
1497
1498         allow_signal(SIGKILL);
1499         nfs4_state_manager(clp);
1500         nfs_put_client(clp);
1501         module_put_and_exit(0);
1502         return 0;
1503 }
1504
1505 /*
1506  * Local variables:
1507  *  c-basic-offset: 8
1508  * End:
1509  */