nfs: enforce FIFO ordering of operations trying to acquire slot
[linux-2.6-block.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 nfs41_end_drain_session(struct nfs_client *clp,
139                 struct nfs4_session *ses)
140 {
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 nfs41_begin_drain_session(struct nfs_client *clp,
160                 struct nfs4_session *ses)
161 {
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         status = nfs41_begin_drain_session(clp, clp->cl_session);
180         if (status != 0)
181                 goto out;
182         status = nfs4_proc_exchange_id(clp, cred);
183         if (status != 0)
184                 goto out;
185         status = nfs4_proc_create_session(clp);
186         if (status != 0)
187                 goto out;
188         nfs41_end_drain_session(clp, clp->cl_session);
189         nfs41_setup_state_renewal(clp);
190         nfs_mark_client_ready(clp, NFS_CS_READY);
191 out:
192         return status;
193 }
194
195 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
196 {
197         struct rpc_cred *cred;
198
199         spin_lock(&clp->cl_lock);
200         cred = nfs4_get_machine_cred_locked(clp);
201         spin_unlock(&clp->cl_lock);
202         return cred;
203 }
204
205 #endif /* CONFIG_NFS_V4_1 */
206
207 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
208 {
209         struct nfs4_state_owner *sp;
210         struct rb_node *pos;
211         struct rpc_cred *cred;
212
213         spin_lock(&clp->cl_lock);
214         cred = nfs4_get_machine_cred_locked(clp);
215         if (cred != NULL)
216                 goto out;
217         pos = rb_first(&clp->cl_state_owners);
218         if (pos != NULL) {
219                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
220                 cred = get_rpccred(sp->so_cred);
221         }
222 out:
223         spin_unlock(&clp->cl_lock);
224         return cred;
225 }
226
227 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
228                 __u64 minval, int maxbits)
229 {
230         struct rb_node **p, *parent;
231         struct nfs_unique_id *pos;
232         __u64 mask = ~0ULL;
233
234         if (maxbits < 64)
235                 mask = (1ULL << maxbits) - 1ULL;
236
237         /* Ensure distribution is more or less flat */
238         get_random_bytes(&new->id, sizeof(new->id));
239         new->id &= mask;
240         if (new->id < minval)
241                 new->id += minval;
242 retry:
243         p = &root->rb_node;
244         parent = NULL;
245
246         while (*p != NULL) {
247                 parent = *p;
248                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
249
250                 if (new->id < pos->id)
251                         p = &(*p)->rb_left;
252                 else if (new->id > pos->id)
253                         p = &(*p)->rb_right;
254                 else
255                         goto id_exists;
256         }
257         rb_link_node(&new->rb_node, parent, p);
258         rb_insert_color(&new->rb_node, root);
259         return;
260 id_exists:
261         for (;;) {
262                 new->id++;
263                 if (new->id < minval || (new->id & mask) != new->id) {
264                         new->id = minval;
265                         break;
266                 }
267                 parent = rb_next(parent);
268                 if (parent == NULL)
269                         break;
270                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
271                 if (new->id < pos->id)
272                         break;
273         }
274         goto retry;
275 }
276
277 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
278 {
279         rb_erase(&id->rb_node, root);
280 }
281
282 static struct nfs4_state_owner *
283 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
284 {
285         struct nfs_client *clp = server->nfs_client;
286         struct rb_node **p = &clp->cl_state_owners.rb_node,
287                        *parent = NULL;
288         struct nfs4_state_owner *sp, *res = NULL;
289
290         while (*p != NULL) {
291                 parent = *p;
292                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
293
294                 if (server < sp->so_server) {
295                         p = &parent->rb_left;
296                         continue;
297                 }
298                 if (server > sp->so_server) {
299                         p = &parent->rb_right;
300                         continue;
301                 }
302                 if (cred < sp->so_cred)
303                         p = &parent->rb_left;
304                 else if (cred > sp->so_cred)
305                         p = &parent->rb_right;
306                 else {
307                         atomic_inc(&sp->so_count);
308                         res = sp;
309                         break;
310                 }
311         }
312         return res;
313 }
314
315 static struct nfs4_state_owner *
316 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
317 {
318         struct rb_node **p = &clp->cl_state_owners.rb_node,
319                        *parent = NULL;
320         struct nfs4_state_owner *sp;
321
322         while (*p != NULL) {
323                 parent = *p;
324                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
325
326                 if (new->so_server < sp->so_server) {
327                         p = &parent->rb_left;
328                         continue;
329                 }
330                 if (new->so_server > sp->so_server) {
331                         p = &parent->rb_right;
332                         continue;
333                 }
334                 if (new->so_cred < sp->so_cred)
335                         p = &parent->rb_left;
336                 else if (new->so_cred > sp->so_cred)
337                         p = &parent->rb_right;
338                 else {
339                         atomic_inc(&sp->so_count);
340                         return sp;
341                 }
342         }
343         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
344         rb_link_node(&new->so_client_node, parent, p);
345         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
346         return new;
347 }
348
349 static void
350 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
351 {
352         if (!RB_EMPTY_NODE(&sp->so_client_node))
353                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
354         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
355 }
356
357 /*
358  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
359  * create a new state_owner.
360  *
361  */
362 static struct nfs4_state_owner *
363 nfs4_alloc_state_owner(void)
364 {
365         struct nfs4_state_owner *sp;
366
367         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
368         if (!sp)
369                 return NULL;
370         spin_lock_init(&sp->so_lock);
371         INIT_LIST_HEAD(&sp->so_states);
372         INIT_LIST_HEAD(&sp->so_delegations);
373         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
374         sp->so_seqid.sequence = &sp->so_sequence;
375         spin_lock_init(&sp->so_sequence.lock);
376         INIT_LIST_HEAD(&sp->so_sequence.list);
377         atomic_set(&sp->so_count, 1);
378         return sp;
379 }
380
381 static void
382 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
383 {
384         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
385                 struct nfs_client *clp = sp->so_client;
386
387                 spin_lock(&clp->cl_lock);
388                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
389                 RB_CLEAR_NODE(&sp->so_client_node);
390                 spin_unlock(&clp->cl_lock);
391         }
392 }
393
394 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
395 {
396         struct nfs_client *clp = server->nfs_client;
397         struct nfs4_state_owner *sp, *new;
398
399         spin_lock(&clp->cl_lock);
400         sp = nfs4_find_state_owner(server, cred);
401         spin_unlock(&clp->cl_lock);
402         if (sp != NULL)
403                 return sp;
404         new = nfs4_alloc_state_owner();
405         if (new == NULL)
406                 return NULL;
407         new->so_client = clp;
408         new->so_server = server;
409         new->so_cred = cred;
410         spin_lock(&clp->cl_lock);
411         sp = nfs4_insert_state_owner(clp, new);
412         spin_unlock(&clp->cl_lock);
413         if (sp == new)
414                 get_rpccred(cred);
415         else {
416                 rpc_destroy_wait_queue(&new->so_sequence.wait);
417                 kfree(new);
418         }
419         return sp;
420 }
421
422 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
423 {
424         struct nfs_client *clp = sp->so_client;
425         struct rpc_cred *cred = sp->so_cred;
426
427         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
428                 return;
429         nfs4_remove_state_owner(clp, sp);
430         spin_unlock(&clp->cl_lock);
431         rpc_destroy_wait_queue(&sp->so_sequence.wait);
432         put_rpccred(cred);
433         kfree(sp);
434 }
435
436 static struct nfs4_state *
437 nfs4_alloc_open_state(void)
438 {
439         struct nfs4_state *state;
440
441         state = kzalloc(sizeof(*state), GFP_KERNEL);
442         if (!state)
443                 return NULL;
444         atomic_set(&state->count, 1);
445         INIT_LIST_HEAD(&state->lock_states);
446         spin_lock_init(&state->state_lock);
447         seqlock_init(&state->seqlock);
448         return state;
449 }
450
451 void
452 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
453 {
454         if (state->state == fmode)
455                 return;
456         /* NB! List reordering - see the reclaim code for why.  */
457         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
458                 if (fmode & FMODE_WRITE)
459                         list_move(&state->open_states, &state->owner->so_states);
460                 else
461                         list_move_tail(&state->open_states, &state->owner->so_states);
462         }
463         state->state = fmode;
464 }
465
466 static struct nfs4_state *
467 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
468 {
469         struct nfs_inode *nfsi = NFS_I(inode);
470         struct nfs4_state *state;
471
472         list_for_each_entry(state, &nfsi->open_states, inode_states) {
473                 if (state->owner != owner)
474                         continue;
475                 if (atomic_inc_not_zero(&state->count))
476                         return state;
477         }
478         return NULL;
479 }
480
481 static void
482 nfs4_free_open_state(struct nfs4_state *state)
483 {
484         kfree(state);
485 }
486
487 struct nfs4_state *
488 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
489 {
490         struct nfs4_state *state, *new;
491         struct nfs_inode *nfsi = NFS_I(inode);
492
493         spin_lock(&inode->i_lock);
494         state = __nfs4_find_state_byowner(inode, owner);
495         spin_unlock(&inode->i_lock);
496         if (state)
497                 goto out;
498         new = nfs4_alloc_open_state();
499         spin_lock(&owner->so_lock);
500         spin_lock(&inode->i_lock);
501         state = __nfs4_find_state_byowner(inode, owner);
502         if (state == NULL && new != NULL) {
503                 state = new;
504                 state->owner = owner;
505                 atomic_inc(&owner->so_count);
506                 list_add(&state->inode_states, &nfsi->open_states);
507                 state->inode = igrab(inode);
508                 spin_unlock(&inode->i_lock);
509                 /* Note: The reclaim code dictates that we add stateless
510                  * and read-only stateids to the end of the list */
511                 list_add_tail(&state->open_states, &owner->so_states);
512                 spin_unlock(&owner->so_lock);
513         } else {
514                 spin_unlock(&inode->i_lock);
515                 spin_unlock(&owner->so_lock);
516                 if (new)
517                         nfs4_free_open_state(new);
518         }
519 out:
520         return state;
521 }
522
523 void nfs4_put_open_state(struct nfs4_state *state)
524 {
525         struct inode *inode = state->inode;
526         struct nfs4_state_owner *owner = state->owner;
527
528         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
529                 return;
530         spin_lock(&inode->i_lock);
531         list_del(&state->inode_states);
532         list_del(&state->open_states);
533         spin_unlock(&inode->i_lock);
534         spin_unlock(&owner->so_lock);
535         iput(inode);
536         nfs4_free_open_state(state);
537         nfs4_put_state_owner(owner);
538 }
539
540 /*
541  * Close the current file.
542  */
543 static void __nfs4_close(struct path *path, struct nfs4_state *state, fmode_t fmode, int wait)
544 {
545         struct nfs4_state_owner *owner = state->owner;
546         int call_close = 0;
547         fmode_t newstate;
548
549         atomic_inc(&owner->so_count);
550         /* Protect against nfs4_find_state() */
551         spin_lock(&owner->so_lock);
552         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
553                 case FMODE_READ:
554                         state->n_rdonly--;
555                         break;
556                 case FMODE_WRITE:
557                         state->n_wronly--;
558                         break;
559                 case FMODE_READ|FMODE_WRITE:
560                         state->n_rdwr--;
561         }
562         newstate = FMODE_READ|FMODE_WRITE;
563         if (state->n_rdwr == 0) {
564                 if (state->n_rdonly == 0) {
565                         newstate &= ~FMODE_READ;
566                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
567                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
568                 }
569                 if (state->n_wronly == 0) {
570                         newstate &= ~FMODE_WRITE;
571                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
572                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
573                 }
574                 if (newstate == 0)
575                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
576         }
577         nfs4_state_set_mode_locked(state, newstate);
578         spin_unlock(&owner->so_lock);
579
580         if (!call_close) {
581                 nfs4_put_open_state(state);
582                 nfs4_put_state_owner(owner);
583         } else
584                 nfs4_do_close(path, state, wait);
585 }
586
587 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
588 {
589         __nfs4_close(path, state, fmode, 0);
590 }
591
592 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
593 {
594         __nfs4_close(path, state, fmode, 1);
595 }
596
597 /*
598  * Search the state->lock_states for an existing lock_owner
599  * that is compatible with current->files
600  */
601 static struct nfs4_lock_state *
602 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
603 {
604         struct nfs4_lock_state *pos;
605         list_for_each_entry(pos, &state->lock_states, ls_locks) {
606                 if (pos->ls_owner != fl_owner)
607                         continue;
608                 atomic_inc(&pos->ls_count);
609                 return pos;
610         }
611         return NULL;
612 }
613
614 /*
615  * Return a compatible lock_state. If no initialized lock_state structure
616  * exists, return an uninitialized one.
617  *
618  */
619 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
620 {
621         struct nfs4_lock_state *lsp;
622         struct nfs_client *clp = state->owner->so_client;
623
624         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
625         if (lsp == NULL)
626                 return NULL;
627         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
628         spin_lock_init(&lsp->ls_sequence.lock);
629         INIT_LIST_HEAD(&lsp->ls_sequence.list);
630         lsp->ls_seqid.sequence = &lsp->ls_sequence;
631         atomic_set(&lsp->ls_count, 1);
632         lsp->ls_state = state;
633         lsp->ls_owner = fl_owner;
634         spin_lock(&clp->cl_lock);
635         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
636         spin_unlock(&clp->cl_lock);
637         INIT_LIST_HEAD(&lsp->ls_locks);
638         return lsp;
639 }
640
641 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
642 {
643         struct nfs_client *clp = lsp->ls_state->owner->so_client;
644
645         spin_lock(&clp->cl_lock);
646         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
647         spin_unlock(&clp->cl_lock);
648         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
649         kfree(lsp);
650 }
651
652 /*
653  * Return a compatible lock_state. If no initialized lock_state structure
654  * exists, return an uninitialized one.
655  *
656  */
657 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
658 {
659         struct nfs4_lock_state *lsp, *new = NULL;
660         
661         for(;;) {
662                 spin_lock(&state->state_lock);
663                 lsp = __nfs4_find_lock_state(state, owner);
664                 if (lsp != NULL)
665                         break;
666                 if (new != NULL) {
667                         list_add(&new->ls_locks, &state->lock_states);
668                         set_bit(LK_STATE_IN_USE, &state->flags);
669                         lsp = new;
670                         new = NULL;
671                         break;
672                 }
673                 spin_unlock(&state->state_lock);
674                 new = nfs4_alloc_lock_state(state, owner);
675                 if (new == NULL)
676                         return NULL;
677         }
678         spin_unlock(&state->state_lock);
679         if (new != NULL)
680                 nfs4_free_lock_state(new);
681         return lsp;
682 }
683
684 /*
685  * Release reference to lock_state, and free it if we see that
686  * it is no longer in use
687  */
688 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
689 {
690         struct nfs4_state *state;
691
692         if (lsp == NULL)
693                 return;
694         state = lsp->ls_state;
695         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
696                 return;
697         list_del(&lsp->ls_locks);
698         if (list_empty(&state->lock_states))
699                 clear_bit(LK_STATE_IN_USE, &state->flags);
700         spin_unlock(&state->state_lock);
701         nfs4_free_lock_state(lsp);
702 }
703
704 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
705 {
706         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
707
708         dst->fl_u.nfs4_fl.owner = lsp;
709         atomic_inc(&lsp->ls_count);
710 }
711
712 static void nfs4_fl_release_lock(struct file_lock *fl)
713 {
714         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
715 }
716
717 static const struct file_lock_operations nfs4_fl_lock_ops = {
718         .fl_copy_lock = nfs4_fl_copy_lock,
719         .fl_release_private = nfs4_fl_release_lock,
720 };
721
722 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
723 {
724         struct nfs4_lock_state *lsp;
725
726         if (fl->fl_ops != NULL)
727                 return 0;
728         lsp = nfs4_get_lock_state(state, fl->fl_owner);
729         if (lsp == NULL)
730                 return -ENOMEM;
731         fl->fl_u.nfs4_fl.owner = lsp;
732         fl->fl_ops = &nfs4_fl_lock_ops;
733         return 0;
734 }
735
736 /*
737  * Byte-range lock aware utility to initialize the stateid of read/write
738  * requests.
739  */
740 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
741 {
742         struct nfs4_lock_state *lsp;
743         int seq;
744
745         do {
746                 seq = read_seqbegin(&state->seqlock);
747                 memcpy(dst, &state->stateid, sizeof(*dst));
748         } while (read_seqretry(&state->seqlock, seq));
749         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
750                 return;
751
752         spin_lock(&state->state_lock);
753         lsp = __nfs4_find_lock_state(state, fl_owner);
754         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
755                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
756         spin_unlock(&state->state_lock);
757         nfs4_put_lock_state(lsp);
758 }
759
760 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
761 {
762         struct nfs_seqid *new;
763
764         new = kmalloc(sizeof(*new), GFP_KERNEL);
765         if (new != NULL) {
766                 new->sequence = counter;
767                 INIT_LIST_HEAD(&new->list);
768         }
769         return new;
770 }
771
772 void nfs_free_seqid(struct nfs_seqid *seqid)
773 {
774         if (!list_empty(&seqid->list)) {
775                 struct rpc_sequence *sequence = seqid->sequence->sequence;
776
777                 spin_lock(&sequence->lock);
778                 list_del(&seqid->list);
779                 spin_unlock(&sequence->lock);
780                 rpc_wake_up(&sequence->wait);
781         }
782         kfree(seqid);
783 }
784
785 /*
786  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
787  * failed with a seqid incrementing error -
788  * see comments nfs_fs.h:seqid_mutating_error()
789  */
790 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
791 {
792         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
793         switch (status) {
794                 case 0:
795                         break;
796                 case -NFS4ERR_BAD_SEQID:
797                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
798                                 return;
799                         printk(KERN_WARNING "NFS: v4 server returned a bad"
800                                         " sequence-id error on an"
801                                         " unconfirmed sequence %p!\n",
802                                         seqid->sequence);
803                 case -NFS4ERR_STALE_CLIENTID:
804                 case -NFS4ERR_STALE_STATEID:
805                 case -NFS4ERR_BAD_STATEID:
806                 case -NFS4ERR_BADXDR:
807                 case -NFS4ERR_RESOURCE:
808                 case -NFS4ERR_NOFILEHANDLE:
809                         /* Non-seqid mutating errors */
810                         return;
811         };
812         /*
813          * Note: no locking needed as we are guaranteed to be first
814          * on the sequence list
815          */
816         seqid->sequence->counter++;
817 }
818
819 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
820 {
821         struct nfs4_state_owner *sp = container_of(seqid->sequence,
822                                         struct nfs4_state_owner, so_seqid);
823         struct nfs_server *server = sp->so_server;
824
825         if (status == -NFS4ERR_BAD_SEQID)
826                 nfs4_drop_state_owner(sp);
827         if (!nfs4_has_session(server->nfs_client))
828                 nfs_increment_seqid(status, seqid);
829 }
830
831 /*
832  * Increment the seqid if the LOCK/LOCKU succeeded, or
833  * failed with a seqid incrementing error -
834  * see comments nfs_fs.h:seqid_mutating_error()
835  */
836 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
837 {
838         nfs_increment_seqid(status, seqid);
839 }
840
841 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
842 {
843         struct rpc_sequence *sequence = seqid->sequence->sequence;
844         int status = 0;
845
846         spin_lock(&sequence->lock);
847         if (list_empty(&seqid->list))
848                 list_add_tail(&seqid->list, &sequence->list);
849         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
850                 goto unlock;
851         rpc_sleep_on(&sequence->wait, task, NULL);
852         status = -EAGAIN;
853 unlock:
854         spin_unlock(&sequence->lock);
855         return status;
856 }
857
858 static int nfs4_run_state_manager(void *);
859
860 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
861 {
862         smp_mb__before_clear_bit();
863         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
864         smp_mb__after_clear_bit();
865         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
866         rpc_wake_up(&clp->cl_rpcwaitq);
867 }
868
869 /*
870  * Schedule the nfs_client asynchronous state management routine
871  */
872 void nfs4_schedule_state_manager(struct nfs_client *clp)
873 {
874         struct task_struct *task;
875
876         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
877                 return;
878         __module_get(THIS_MODULE);
879         atomic_inc(&clp->cl_count);
880         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
881                                 rpc_peeraddr2str(clp->cl_rpcclient,
882                                                         RPC_DISPLAY_ADDR));
883         if (!IS_ERR(task))
884                 return;
885         nfs4_clear_state_manager_bit(clp);
886         nfs_put_client(clp);
887         module_put(THIS_MODULE);
888 }
889
890 /*
891  * Schedule a state recovery attempt
892  */
893 void nfs4_schedule_state_recovery(struct nfs_client *clp)
894 {
895         if (!clp)
896                 return;
897         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
898                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
899         nfs4_schedule_state_manager(clp);
900 }
901
902 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
903 {
904
905         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
906         /* Don't recover state that expired before the reboot */
907         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
908                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
909                 return 0;
910         }
911         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
912         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
913         return 1;
914 }
915
916 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
917 {
918         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
919         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
920         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
921         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
922         return 1;
923 }
924
925 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
926 {
927         struct inode *inode = state->inode;
928         struct nfs_inode *nfsi = NFS_I(inode);
929         struct file_lock *fl;
930         int status = 0;
931
932         if (inode->i_flock == NULL)
933                 return 0;
934
935         /* Guard against delegation returns and new lock/unlock calls */
936         down_write(&nfsi->rwsem);
937         /* Protect inode->i_flock using the BKL */
938         lock_kernel();
939         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
940                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
941                         continue;
942                 if (nfs_file_open_context(fl->fl_file)->state != state)
943                         continue;
944                 unlock_kernel();
945                 status = ops->recover_lock(state, fl);
946                 switch (status) {
947                         case 0:
948                                 break;
949                         case -ESTALE:
950                         case -NFS4ERR_ADMIN_REVOKED:
951                         case -NFS4ERR_STALE_STATEID:
952                         case -NFS4ERR_BAD_STATEID:
953                         case -NFS4ERR_EXPIRED:
954                         case -NFS4ERR_NO_GRACE:
955                         case -NFS4ERR_STALE_CLIENTID:
956                         case -NFS4ERR_BADSESSION:
957                         case -NFS4ERR_BADSLOT:
958                         case -NFS4ERR_BAD_HIGH_SLOT:
959                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
960                                 goto out;
961                         default:
962                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
963                                                 __func__, status);
964                         case -ENOMEM:
965                         case -NFS4ERR_DENIED:
966                         case -NFS4ERR_RECLAIM_BAD:
967                         case -NFS4ERR_RECLAIM_CONFLICT:
968                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
969                                 status = 0;
970                 }
971                 lock_kernel();
972         }
973         unlock_kernel();
974 out:
975         up_write(&nfsi->rwsem);
976         return status;
977 }
978
979 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
980 {
981         struct nfs4_state *state;
982         struct nfs4_lock_state *lock;
983         int status = 0;
984
985         /* Note: we rely on the sp->so_states list being ordered 
986          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
987          * states first.
988          * This is needed to ensure that the server won't give us any
989          * read delegations that we have to return if, say, we are
990          * recovering after a network partition or a reboot from a
991          * server that doesn't support a grace period.
992          */
993 restart:
994         spin_lock(&sp->so_lock);
995         list_for_each_entry(state, &sp->so_states, open_states) {
996                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
997                         continue;
998                 if (state->state == 0)
999                         continue;
1000                 atomic_inc(&state->count);
1001                 spin_unlock(&sp->so_lock);
1002                 status = ops->recover_open(sp, state);
1003                 if (status >= 0) {
1004                         status = nfs4_reclaim_locks(state, ops);
1005                         if (status >= 0) {
1006                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1007                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1008                                                 printk("%s: Lock reclaim failed!\n",
1009                                                         __func__);
1010                                 }
1011                                 nfs4_put_open_state(state);
1012                                 goto restart;
1013                         }
1014                 }
1015                 switch (status) {
1016                         default:
1017                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1018                                                 __func__, status);
1019                         case -ENOENT:
1020                         case -ENOMEM:
1021                         case -ESTALE:
1022                                 /*
1023                                  * Open state on this file cannot be recovered
1024                                  * All we can do is revert to using the zero stateid.
1025                                  */
1026                                 memset(state->stateid.data, 0,
1027                                         sizeof(state->stateid.data));
1028                                 /* Mark the file as being 'closed' */
1029                                 state->state = 0;
1030                                 break;
1031                         case -NFS4ERR_ADMIN_REVOKED:
1032                         case -NFS4ERR_STALE_STATEID:
1033                         case -NFS4ERR_BAD_STATEID:
1034                         case -NFS4ERR_RECLAIM_BAD:
1035                         case -NFS4ERR_RECLAIM_CONFLICT:
1036                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
1037                                 break;
1038                         case -NFS4ERR_EXPIRED:
1039                         case -NFS4ERR_NO_GRACE:
1040                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
1041                         case -NFS4ERR_STALE_CLIENTID:
1042                         case -NFS4ERR_BADSESSION:
1043                         case -NFS4ERR_BADSLOT:
1044                         case -NFS4ERR_BAD_HIGH_SLOT:
1045                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1046                                 goto out_err;
1047                 }
1048                 nfs4_put_open_state(state);
1049                 goto restart;
1050         }
1051         spin_unlock(&sp->so_lock);
1052         return 0;
1053 out_err:
1054         nfs4_put_open_state(state);
1055         return status;
1056 }
1057
1058 static void nfs4_clear_open_state(struct nfs4_state *state)
1059 {
1060         struct nfs4_lock_state *lock;
1061
1062         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1063         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1064         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1065         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1066         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1067                 lock->ls_seqid.flags = 0;
1068                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1069         }
1070 }
1071
1072 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1073 {
1074         struct nfs4_state_owner *sp;
1075         struct rb_node *pos;
1076         struct nfs4_state *state;
1077
1078         /* Reset all sequence ids to zero */
1079         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1080                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1081                 sp->so_seqid.flags = 0;
1082                 spin_lock(&sp->so_lock);
1083                 list_for_each_entry(state, &sp->so_states, open_states) {
1084                         if (mark_reclaim(clp, state))
1085                                 nfs4_clear_open_state(state);
1086                 }
1087                 spin_unlock(&sp->so_lock);
1088         }
1089 }
1090
1091 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1092 {
1093         /* Mark all delegations for reclaim */
1094         nfs_delegation_mark_reclaim(clp);
1095         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1096 }
1097
1098 static void nfs4_reclaim_complete(struct nfs_client *clp,
1099                                  const struct nfs4_state_recovery_ops *ops)
1100 {
1101         /* Notify the server we're done reclaiming our state */
1102         if (ops->reclaim_complete)
1103                 (void)ops->reclaim_complete(clp);
1104 }
1105
1106 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1107 {
1108         struct nfs4_state_owner *sp;
1109         struct rb_node *pos;
1110         struct nfs4_state *state;
1111
1112         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1113                 return;
1114
1115         nfs4_reclaim_complete(clp,
1116                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1117
1118         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1119                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1120                 spin_lock(&sp->so_lock);
1121                 list_for_each_entry(state, &sp->so_states, open_states) {
1122                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1123                                 continue;
1124                         nfs4_state_mark_reclaim_nograce(clp, state);
1125                 }
1126                 spin_unlock(&sp->so_lock);
1127         }
1128
1129         nfs_delegation_reap_unclaimed(clp);
1130 }
1131
1132 static void nfs_delegation_clear_all(struct nfs_client *clp)
1133 {
1134         nfs_delegation_mark_reclaim(clp);
1135         nfs_delegation_reap_unclaimed(clp);
1136 }
1137
1138 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1139 {
1140         nfs_delegation_clear_all(clp);
1141         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1142 }
1143
1144 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1145 {
1146         switch (error) {
1147                 case -NFS4ERR_CB_PATH_DOWN:
1148                         nfs_handle_cb_pathdown(clp);
1149                         return 0;
1150                 case -NFS4ERR_NO_GRACE:
1151                         nfs4_state_end_reclaim_reboot(clp);
1152                         return 0;
1153                 case -NFS4ERR_STALE_CLIENTID:
1154                 case -NFS4ERR_LEASE_MOVED:
1155                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1156                         nfs4_state_end_reclaim_reboot(clp);
1157                         nfs4_state_start_reclaim_reboot(clp);
1158                         break;
1159                 case -NFS4ERR_EXPIRED:
1160                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1161                         nfs4_state_start_reclaim_nograce(clp);
1162                         break;
1163                 case -NFS4ERR_BADSESSION:
1164                 case -NFS4ERR_BADSLOT:
1165                 case -NFS4ERR_BAD_HIGH_SLOT:
1166                 case -NFS4ERR_DEADSESSION:
1167                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1168                 case -NFS4ERR_SEQ_FALSE_RETRY:
1169                 case -NFS4ERR_SEQ_MISORDERED:
1170                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1171                         /* Zero session reset errors */
1172                         return 0;
1173         }
1174         return error;
1175 }
1176
1177 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1178 {
1179         struct rb_node *pos;
1180         int status = 0;
1181
1182 restart:
1183         spin_lock(&clp->cl_lock);
1184         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1185                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1186                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1187                         continue;
1188                 atomic_inc(&sp->so_count);
1189                 spin_unlock(&clp->cl_lock);
1190                 status = nfs4_reclaim_open_state(sp, ops);
1191                 if (status < 0) {
1192                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1193                         nfs4_put_state_owner(sp);
1194                         return nfs4_recovery_handle_error(clp, status);
1195                 }
1196                 nfs4_put_state_owner(sp);
1197                 goto restart;
1198         }
1199         spin_unlock(&clp->cl_lock);
1200         return status;
1201 }
1202
1203 static int nfs4_check_lease(struct nfs_client *clp)
1204 {
1205         struct rpc_cred *cred;
1206         struct nfs4_state_maintenance_ops *ops =
1207                 nfs4_state_renewal_ops[clp->cl_minorversion];
1208         int status = -NFS4ERR_EXPIRED;
1209
1210         /* Is the client already known to have an expired lease? */
1211         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1212                 return 0;
1213         spin_lock(&clp->cl_lock);
1214         cred = ops->get_state_renewal_cred_locked(clp);
1215         spin_unlock(&clp->cl_lock);
1216         if (cred == NULL) {
1217                 cred = nfs4_get_setclientid_cred(clp);
1218                 if (cred == NULL)
1219                         goto out;
1220         }
1221         status = ops->renew_lease(clp, cred);
1222         put_rpccred(cred);
1223 out:
1224         return nfs4_recovery_handle_error(clp, status);
1225 }
1226
1227 static int nfs4_reclaim_lease(struct nfs_client *clp)
1228 {
1229         struct rpc_cred *cred;
1230         struct nfs4_state_recovery_ops *ops =
1231                 nfs4_reboot_recovery_ops[clp->cl_minorversion];
1232         int status = -ENOENT;
1233
1234         cred = ops->get_clid_cred(clp);
1235         if (cred != NULL) {
1236                 status = ops->establish_clid(clp, cred);
1237                 put_rpccred(cred);
1238                 /* Handle case where the user hasn't set up machine creds */
1239                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1240                         nfs4_clear_machine_cred(clp);
1241                         status = -EAGAIN;
1242                 }
1243                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1244                         status = -EPROTONOSUPPORT;
1245         }
1246         return status;
1247 }
1248
1249 #ifdef CONFIG_NFS_V4_1
1250 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1251 {
1252         if (!flags)
1253                 return;
1254         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED) {
1255                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1256                 nfs4_state_start_reclaim_reboot(clp);
1257                 nfs4_schedule_state_recovery(clp);
1258         } else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1259                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1260                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1261                             SEQ4_STATUS_RECALLABLE_STATE_REVOKED |
1262                             SEQ4_STATUS_LEASE_MOVED)) {
1263                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1264                 nfs4_state_start_reclaim_nograce(clp);
1265                 nfs4_schedule_state_recovery(clp);
1266         } else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1267                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1268                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1269                 nfs_expire_all_delegations(clp);
1270 }
1271
1272 static int nfs4_reset_session(struct nfs_client *clp)
1273 {
1274         struct nfs4_session *ses = clp->cl_session;
1275         int status;
1276
1277         status = nfs41_begin_drain_session(clp, ses);
1278         if (status != 0)
1279                 return status;
1280
1281         status = nfs4_proc_destroy_session(clp->cl_session);
1282         if (status && status != -NFS4ERR_BADSESSION &&
1283             status != -NFS4ERR_DEADSESSION) {
1284                 status = nfs4_recovery_handle_error(clp, status);
1285                 goto out;
1286         }
1287
1288         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1289         status = nfs4_proc_create_session(clp);
1290         if (status)
1291                 status = nfs4_recovery_handle_error(clp, status);
1292
1293 out:
1294         /*
1295          * Let the state manager reestablish state
1296          * without waking other tasks yet.
1297          */
1298         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1299                 /* Wake up the next rpc task */
1300                 nfs41_end_drain_session(clp, ses);
1301                 if (status == 0)
1302                         nfs41_setup_state_renewal(clp);
1303         }
1304         return status;
1305 }
1306
1307 #else /* CONFIG_NFS_V4_1 */
1308 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1309 #endif /* CONFIG_NFS_V4_1 */
1310
1311 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1312  * on EXCHANGE_ID for v4.1
1313  */
1314 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1315 {
1316         if (nfs4_has_session(clp)) {
1317                 switch (status) {
1318                 case -NFS4ERR_DELAY:
1319                 case -NFS4ERR_CLID_INUSE:
1320                 case -EAGAIN:
1321                         break;
1322
1323                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1324                                          * in nfs4_exchange_id */
1325                 default:
1326                         return;
1327                 }
1328         }
1329         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1330 }
1331
1332 static void nfs4_state_manager(struct nfs_client *clp)
1333 {
1334         int status = 0;
1335
1336         /* Ensure exclusive access to NFSv4 state */
1337         for(;;) {
1338                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1339                         /* We're going to have to re-establish a clientid */
1340                         status = nfs4_reclaim_lease(clp);
1341                         if (status) {
1342                                 nfs4_set_lease_expired(clp, status);
1343                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1344                                                         &clp->cl_state))
1345                                         continue;
1346                                 if (clp->cl_cons_state ==
1347                                                         NFS_CS_SESSION_INITING)
1348                                         nfs_mark_client_ready(clp, status);
1349                                 goto out_error;
1350                         }
1351                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1352                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1353                 }
1354
1355                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1356                         status = nfs4_check_lease(clp);
1357                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1358                                 continue;
1359                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1360                                 goto out_error;
1361                 }
1362
1363                 /* Initialize or reset the session */
1364                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1365                    && nfs4_has_session(clp)) {
1366                         status = nfs4_reset_session(clp);
1367                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1368                                 continue;
1369                         if (status < 0)
1370                                 goto out_error;
1371                 }
1372
1373                 /* First recover reboot state... */
1374                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1375                         status = nfs4_do_reclaim(clp,
1376                                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1377                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1378                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1379                                 continue;
1380                         nfs4_state_end_reclaim_reboot(clp);
1381                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1382                                 continue;
1383                         if (status < 0)
1384                                 goto out_error;
1385                 }
1386
1387                 /* Now recover expired state... */
1388                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1389                         status = nfs4_do_reclaim(clp,
1390                                 nfs4_nograce_recovery_ops[clp->cl_minorversion]);
1391                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1392                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1393                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1394                                 continue;
1395                         if (status < 0)
1396                                 goto out_error;
1397                 }
1398
1399                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1400                         nfs_client_return_marked_delegations(clp);
1401                         continue;
1402                 }
1403
1404                 nfs4_clear_state_manager_bit(clp);
1405                 /* Did we race with an attempt to give us more work? */
1406                 if (clp->cl_state == 0)
1407                         break;
1408                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1409                         break;
1410         }
1411         return;
1412 out_error:
1413         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1414                         " with error %d\n", clp->cl_hostname, -status);
1415         nfs4_clear_state_manager_bit(clp);
1416 }
1417
1418 static int nfs4_run_state_manager(void *ptr)
1419 {
1420         struct nfs_client *clp = ptr;
1421
1422         allow_signal(SIGKILL);
1423         nfs4_state_manager(clp);
1424         nfs_put_client(clp);
1425         module_put_and_exit(0);
1426         return 0;
1427 }
1428
1429 /*
1430  * Local variables:
1431  *  c-basic-offset: 8
1432  * End:
1433  */