NFSv4: Fix a locking regression in nfs4_set_mode_locked()
[linux-2.6.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 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
64 {
65         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
66                         nfs_callback_tcpport, cred);
67         if (status == 0)
68                 status = nfs4_proc_setclientid_confirm(clp, cred);
69         if (status == 0)
70                 nfs4_schedule_state_renewal(clp);
71         return status;
72 }
73
74 struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
75 {
76         struct nfs4_state_owner *sp;
77         struct rb_node *pos;
78         struct rpc_cred *cred = NULL;
79
80         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
81                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
82                 if (list_empty(&sp->so_states))
83                         continue;
84                 cred = get_rpccred(sp->so_cred);
85                 break;
86         }
87         return cred;
88 }
89
90 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
91 {
92         struct nfs4_state_owner *sp;
93         struct rb_node *pos;
94
95         pos = rb_first(&clp->cl_state_owners);
96         if (pos != NULL) {
97                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
98                 return get_rpccred(sp->so_cred);
99         }
100         return NULL;
101 }
102
103 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
104                 __u64 minval, int maxbits)
105 {
106         struct rb_node **p, *parent;
107         struct nfs_unique_id *pos;
108         __u64 mask = ~0ULL;
109
110         if (maxbits < 64)
111                 mask = (1ULL << maxbits) - 1ULL;
112
113         /* Ensure distribution is more or less flat */
114         get_random_bytes(&new->id, sizeof(new->id));
115         new->id &= mask;
116         if (new->id < minval)
117                 new->id += minval;
118 retry:
119         p = &root->rb_node;
120         parent = NULL;
121
122         while (*p != NULL) {
123                 parent = *p;
124                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
125
126                 if (new->id < pos->id)
127                         p = &(*p)->rb_left;
128                 else if (new->id > pos->id)
129                         p = &(*p)->rb_right;
130                 else
131                         goto id_exists;
132         }
133         rb_link_node(&new->rb_node, parent, p);
134         rb_insert_color(&new->rb_node, root);
135         return;
136 id_exists:
137         for (;;) {
138                 new->id++;
139                 if (new->id < minval || (new->id & mask) != new->id) {
140                         new->id = minval;
141                         break;
142                 }
143                 parent = rb_next(parent);
144                 if (parent == NULL)
145                         break;
146                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
147                 if (new->id < pos->id)
148                         break;
149         }
150         goto retry;
151 }
152
153 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
154 {
155         rb_erase(&id->rb_node, root);
156 }
157
158 static struct nfs4_state_owner *
159 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
160 {
161         struct nfs_client *clp = server->nfs_client;
162         struct rb_node **p = &clp->cl_state_owners.rb_node,
163                        *parent = NULL;
164         struct nfs4_state_owner *sp, *res = NULL;
165
166         while (*p != NULL) {
167                 parent = *p;
168                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
169
170                 if (server < sp->so_server) {
171                         p = &parent->rb_left;
172                         continue;
173                 }
174                 if (server > sp->so_server) {
175                         p = &parent->rb_right;
176                         continue;
177                 }
178                 if (cred < sp->so_cred)
179                         p = &parent->rb_left;
180                 else if (cred > sp->so_cred)
181                         p = &parent->rb_right;
182                 else {
183                         atomic_inc(&sp->so_count);
184                         res = sp;
185                         break;
186                 }
187         }
188         return res;
189 }
190
191 static struct nfs4_state_owner *
192 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
193 {
194         struct rb_node **p = &clp->cl_state_owners.rb_node,
195                        *parent = NULL;
196         struct nfs4_state_owner *sp;
197
198         while (*p != NULL) {
199                 parent = *p;
200                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
201
202                 if (new->so_server < sp->so_server) {
203                         p = &parent->rb_left;
204                         continue;
205                 }
206                 if (new->so_server > sp->so_server) {
207                         p = &parent->rb_right;
208                         continue;
209                 }
210                 if (new->so_cred < sp->so_cred)
211                         p = &parent->rb_left;
212                 else if (new->so_cred > sp->so_cred)
213                         p = &parent->rb_right;
214                 else {
215                         atomic_inc(&sp->so_count);
216                         return sp;
217                 }
218         }
219         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
220         rb_link_node(&new->so_client_node, parent, p);
221         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
222         return new;
223 }
224
225 static void
226 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
227 {
228         if (!RB_EMPTY_NODE(&sp->so_client_node))
229                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
230         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
231 }
232
233 /*
234  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
235  * create a new state_owner.
236  *
237  */
238 static struct nfs4_state_owner *
239 nfs4_alloc_state_owner(void)
240 {
241         struct nfs4_state_owner *sp;
242
243         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
244         if (!sp)
245                 return NULL;
246         spin_lock_init(&sp->so_lock);
247         INIT_LIST_HEAD(&sp->so_states);
248         INIT_LIST_HEAD(&sp->so_delegations);
249         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
250         sp->so_seqid.sequence = &sp->so_sequence;
251         spin_lock_init(&sp->so_sequence.lock);
252         INIT_LIST_HEAD(&sp->so_sequence.list);
253         atomic_set(&sp->so_count, 1);
254         return sp;
255 }
256
257 void
258 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
259 {
260         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
261                 struct nfs_client *clp = sp->so_client;
262
263                 spin_lock(&clp->cl_lock);
264                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
265                 RB_CLEAR_NODE(&sp->so_client_node);
266                 spin_unlock(&clp->cl_lock);
267         }
268 }
269
270 /*
271  * Note: must be called with clp->cl_sem held in order to prevent races
272  *       with reboot recovery!
273  */
274 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
275 {
276         struct nfs_client *clp = server->nfs_client;
277         struct nfs4_state_owner *sp, *new;
278
279         spin_lock(&clp->cl_lock);
280         sp = nfs4_find_state_owner(server, cred);
281         spin_unlock(&clp->cl_lock);
282         if (sp != NULL)
283                 return sp;
284         new = nfs4_alloc_state_owner();
285         if (new == NULL)
286                 return NULL;
287         new->so_client = clp;
288         new->so_server = server;
289         new->so_cred = cred;
290         spin_lock(&clp->cl_lock);
291         sp = nfs4_insert_state_owner(clp, new);
292         spin_unlock(&clp->cl_lock);
293         if (sp == new)
294                 get_rpccred(cred);
295         else
296                 kfree(new);
297         return sp;
298 }
299
300 /*
301  * Must be called with clp->cl_sem held in order to avoid races
302  * with state recovery...
303  */
304 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
305 {
306         struct nfs_client *clp = sp->so_client;
307         struct rpc_cred *cred = sp->so_cred;
308
309         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
310                 return;
311         nfs4_remove_state_owner(clp, sp);
312         spin_unlock(&clp->cl_lock);
313         put_rpccred(cred);
314         kfree(sp);
315 }
316
317 static struct nfs4_state *
318 nfs4_alloc_open_state(void)
319 {
320         struct nfs4_state *state;
321
322         state = kzalloc(sizeof(*state), GFP_KERNEL);
323         if (!state)
324                 return NULL;
325         atomic_set(&state->count, 1);
326         INIT_LIST_HEAD(&state->lock_states);
327         spin_lock_init(&state->state_lock);
328         seqlock_init(&state->seqlock);
329         return state;
330 }
331
332 void
333 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
334 {
335         if (state->state == mode)
336                 return;
337         /* NB! List reordering - see the reclaim code for why.  */
338         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
339                 if (mode & FMODE_WRITE)
340                         list_move(&state->open_states, &state->owner->so_states);
341                 else
342                         list_move_tail(&state->open_states, &state->owner->so_states);
343         }
344         state->state = mode;
345 }
346
347 static struct nfs4_state *
348 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
349 {
350         struct nfs_inode *nfsi = NFS_I(inode);
351         struct nfs4_state *state;
352
353         list_for_each_entry(state, &nfsi->open_states, inode_states) {
354                 if (state->owner != owner)
355                         continue;
356                 if (atomic_inc_not_zero(&state->count))
357                         return state;
358         }
359         return NULL;
360 }
361
362 static void
363 nfs4_free_open_state(struct nfs4_state *state)
364 {
365         kfree(state);
366 }
367
368 struct nfs4_state *
369 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
370 {
371         struct nfs4_state *state, *new;
372         struct nfs_inode *nfsi = NFS_I(inode);
373
374         spin_lock(&inode->i_lock);
375         state = __nfs4_find_state_byowner(inode, owner);
376         spin_unlock(&inode->i_lock);
377         if (state)
378                 goto out;
379         new = nfs4_alloc_open_state();
380         spin_lock(&owner->so_lock);
381         spin_lock(&inode->i_lock);
382         state = __nfs4_find_state_byowner(inode, owner);
383         if (state == NULL && new != NULL) {
384                 state = new;
385                 state->owner = owner;
386                 atomic_inc(&owner->so_count);
387                 list_add(&state->inode_states, &nfsi->open_states);
388                 state->inode = igrab(inode);
389                 spin_unlock(&inode->i_lock);
390                 /* Note: The reclaim code dictates that we add stateless
391                  * and read-only stateids to the end of the list */
392                 list_add_tail(&state->open_states, &owner->so_states);
393                 spin_unlock(&owner->so_lock);
394         } else {
395                 spin_unlock(&inode->i_lock);
396                 spin_unlock(&owner->so_lock);
397                 if (new)
398                         nfs4_free_open_state(new);
399         }
400 out:
401         return state;
402 }
403
404 /*
405  * Beware! Caller must be holding exactly one
406  * reference to clp->cl_sem!
407  */
408 void nfs4_put_open_state(struct nfs4_state *state)
409 {
410         struct inode *inode = state->inode;
411         struct nfs4_state_owner *owner = state->owner;
412
413         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
414                 return;
415         spin_lock(&inode->i_lock);
416         list_del(&state->inode_states);
417         list_del(&state->open_states);
418         spin_unlock(&inode->i_lock);
419         spin_unlock(&owner->so_lock);
420         iput(inode);
421         nfs4_free_open_state(state);
422         nfs4_put_state_owner(owner);
423 }
424
425 /*
426  * Close the current file.
427  */
428 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
429 {
430         struct nfs4_state_owner *owner = state->owner;
431         int call_close = 0;
432         int newstate;
433
434         atomic_inc(&owner->so_count);
435         /* Protect against nfs4_find_state() */
436         spin_lock(&owner->so_lock);
437         switch (mode & (FMODE_READ | FMODE_WRITE)) {
438                 case FMODE_READ:
439                         state->n_rdonly--;
440                         break;
441                 case FMODE_WRITE:
442                         state->n_wronly--;
443                         break;
444                 case FMODE_READ|FMODE_WRITE:
445                         state->n_rdwr--;
446         }
447         newstate = FMODE_READ|FMODE_WRITE;
448         if (state->n_rdwr == 0) {
449                 if (state->n_rdonly == 0) {
450                         newstate &= ~FMODE_READ;
451                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
452                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
453                 }
454                 if (state->n_wronly == 0) {
455                         newstate &= ~FMODE_WRITE;
456                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
457                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
458                 }
459                 if (newstate == 0)
460                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
461         }
462         nfs4_state_set_mode_locked(state, newstate);
463         spin_unlock(&owner->so_lock);
464
465         if (!call_close) {
466                 nfs4_put_open_state(state);
467                 nfs4_put_state_owner(owner);
468         } else
469                 nfs4_do_close(path, state);
470 }
471
472 /*
473  * Search the state->lock_states for an existing lock_owner
474  * that is compatible with current->files
475  */
476 static struct nfs4_lock_state *
477 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
478 {
479         struct nfs4_lock_state *pos;
480         list_for_each_entry(pos, &state->lock_states, ls_locks) {
481                 if (pos->ls_owner != fl_owner)
482                         continue;
483                 atomic_inc(&pos->ls_count);
484                 return pos;
485         }
486         return NULL;
487 }
488
489 /*
490  * Return a compatible lock_state. If no initialized lock_state structure
491  * exists, return an uninitialized one.
492  *
493  */
494 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
495 {
496         struct nfs4_lock_state *lsp;
497         struct nfs_client *clp = state->owner->so_client;
498
499         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
500         if (lsp == NULL)
501                 return NULL;
502         lsp->ls_seqid.sequence = &state->owner->so_sequence;
503         atomic_set(&lsp->ls_count, 1);
504         lsp->ls_owner = fl_owner;
505         spin_lock(&clp->cl_lock);
506         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
507         spin_unlock(&clp->cl_lock);
508         INIT_LIST_HEAD(&lsp->ls_locks);
509         return lsp;
510 }
511
512 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
513 {
514         struct nfs_client *clp = lsp->ls_state->owner->so_client;
515
516         spin_lock(&clp->cl_lock);
517         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
518         spin_unlock(&clp->cl_lock);
519         kfree(lsp);
520 }
521
522 /*
523  * Return a compatible lock_state. If no initialized lock_state structure
524  * exists, return an uninitialized one.
525  *
526  * The caller must be holding clp->cl_sem
527  */
528 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
529 {
530         struct nfs4_lock_state *lsp, *new = NULL;
531         
532         for(;;) {
533                 spin_lock(&state->state_lock);
534                 lsp = __nfs4_find_lock_state(state, owner);
535                 if (lsp != NULL)
536                         break;
537                 if (new != NULL) {
538                         new->ls_state = state;
539                         list_add(&new->ls_locks, &state->lock_states);
540                         set_bit(LK_STATE_IN_USE, &state->flags);
541                         lsp = new;
542                         new = NULL;
543                         break;
544                 }
545                 spin_unlock(&state->state_lock);
546                 new = nfs4_alloc_lock_state(state, owner);
547                 if (new == NULL)
548                         return NULL;
549         }
550         spin_unlock(&state->state_lock);
551         if (new != NULL)
552                 nfs4_free_lock_state(new);
553         return lsp;
554 }
555
556 /*
557  * Release reference to lock_state, and free it if we see that
558  * it is no longer in use
559  */
560 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
561 {
562         struct nfs4_state *state;
563
564         if (lsp == NULL)
565                 return;
566         state = lsp->ls_state;
567         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
568                 return;
569         list_del(&lsp->ls_locks);
570         if (list_empty(&state->lock_states))
571                 clear_bit(LK_STATE_IN_USE, &state->flags);
572         spin_unlock(&state->state_lock);
573         nfs4_free_lock_state(lsp);
574 }
575
576 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
577 {
578         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
579
580         dst->fl_u.nfs4_fl.owner = lsp;
581         atomic_inc(&lsp->ls_count);
582 }
583
584 static void nfs4_fl_release_lock(struct file_lock *fl)
585 {
586         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
587 }
588
589 static struct file_lock_operations nfs4_fl_lock_ops = {
590         .fl_copy_lock = nfs4_fl_copy_lock,
591         .fl_release_private = nfs4_fl_release_lock,
592 };
593
594 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
595 {
596         struct nfs4_lock_state *lsp;
597
598         if (fl->fl_ops != NULL)
599                 return 0;
600         lsp = nfs4_get_lock_state(state, fl->fl_owner);
601         if (lsp == NULL)
602                 return -ENOMEM;
603         fl->fl_u.nfs4_fl.owner = lsp;
604         fl->fl_ops = &nfs4_fl_lock_ops;
605         return 0;
606 }
607
608 /*
609  * Byte-range lock aware utility to initialize the stateid of read/write
610  * requests.
611  */
612 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
613 {
614         struct nfs4_lock_state *lsp;
615         int seq;
616
617         do {
618                 seq = read_seqbegin(&state->seqlock);
619                 memcpy(dst, &state->stateid, sizeof(*dst));
620         } while (read_seqretry(&state->seqlock, seq));
621         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
622                 return;
623
624         spin_lock(&state->state_lock);
625         lsp = __nfs4_find_lock_state(state, fl_owner);
626         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
627                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
628         spin_unlock(&state->state_lock);
629         nfs4_put_lock_state(lsp);
630 }
631
632 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
633 {
634         struct rpc_sequence *sequence = counter->sequence;
635         struct nfs_seqid *new;
636
637         new = kmalloc(sizeof(*new), GFP_KERNEL);
638         if (new != NULL) {
639                 new->sequence = counter;
640                 spin_lock(&sequence->lock);
641                 list_add_tail(&new->list, &sequence->list);
642                 spin_unlock(&sequence->lock);
643         }
644         return new;
645 }
646
647 void nfs_free_seqid(struct nfs_seqid *seqid)
648 {
649         struct rpc_sequence *sequence = seqid->sequence->sequence;
650
651         spin_lock(&sequence->lock);
652         list_del(&seqid->list);
653         spin_unlock(&sequence->lock);
654         rpc_wake_up(&sequence->wait);
655         kfree(seqid);
656 }
657
658 /*
659  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
660  * failed with a seqid incrementing error -
661  * see comments nfs_fs.h:seqid_mutating_error()
662  */
663 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
664 {
665         switch (status) {
666                 case 0:
667                         break;
668                 case -NFS4ERR_BAD_SEQID:
669                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
670                                 return;
671                         printk(KERN_WARNING "NFS: v4 server returned a bad"
672                                         "sequence-id error on an"
673                                         "unconfirmed sequence %p!\n",
674                                         seqid->sequence);
675                 case -NFS4ERR_STALE_CLIENTID:
676                 case -NFS4ERR_STALE_STATEID:
677                 case -NFS4ERR_BAD_STATEID:
678                 case -NFS4ERR_BADXDR:
679                 case -NFS4ERR_RESOURCE:
680                 case -NFS4ERR_NOFILEHANDLE:
681                         /* Non-seqid mutating errors */
682                         return;
683         };
684         /*
685          * Note: no locking needed as we are guaranteed to be first
686          * on the sequence list
687          */
688         seqid->sequence->counter++;
689 }
690
691 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
692 {
693         if (status == -NFS4ERR_BAD_SEQID) {
694                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
695                                 struct nfs4_state_owner, so_seqid);
696                 nfs4_drop_state_owner(sp);
697         }
698         nfs_increment_seqid(status, seqid);
699 }
700
701 /*
702  * Increment the seqid if the LOCK/LOCKU succeeded, or
703  * failed with a seqid incrementing error -
704  * see comments nfs_fs.h:seqid_mutating_error()
705  */
706 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
707 {
708         nfs_increment_seqid(status, seqid);
709 }
710
711 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
712 {
713         struct rpc_sequence *sequence = seqid->sequence->sequence;
714         int status = 0;
715
716         if (sequence->list.next == &seqid->list)
717                 goto out;
718         spin_lock(&sequence->lock);
719         if (sequence->list.next != &seqid->list) {
720                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
721                 status = -EAGAIN;
722         }
723         spin_unlock(&sequence->lock);
724 out:
725         return status;
726 }
727
728 static int reclaimer(void *);
729
730 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
731 {
732         smp_mb__before_clear_bit();
733         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
734         smp_mb__after_clear_bit();
735         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
736         rpc_wake_up(&clp->cl_rpcwaitq);
737 }
738
739 /*
740  * State recovery routine
741  */
742 static void nfs4_recover_state(struct nfs_client *clp)
743 {
744         struct task_struct *task;
745
746         __module_get(THIS_MODULE);
747         atomic_inc(&clp->cl_count);
748         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
749                         NIPQUAD(clp->cl_addr.sin_addr));
750         if (!IS_ERR(task))
751                 return;
752         nfs4_clear_recover_bit(clp);
753         nfs_put_client(clp);
754         module_put(THIS_MODULE);
755 }
756
757 /*
758  * Schedule a state recovery attempt
759  */
760 void nfs4_schedule_state_recovery(struct nfs_client *clp)
761 {
762         if (!clp)
763                 return;
764         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
765                 nfs4_recover_state(clp);
766 }
767
768 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
769 {
770         struct inode *inode = state->inode;
771         struct file_lock *fl;
772         int status = 0;
773
774         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
775                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
776                         continue;
777                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
778                         continue;
779                 status = ops->recover_lock(state, fl);
780                 if (status >= 0)
781                         continue;
782                 switch (status) {
783                         default:
784                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
785                                                 __FUNCTION__, status);
786                         case -NFS4ERR_EXPIRED:
787                         case -NFS4ERR_NO_GRACE:
788                         case -NFS4ERR_RECLAIM_BAD:
789                         case -NFS4ERR_RECLAIM_CONFLICT:
790                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
791                                 break;
792                         case -NFS4ERR_STALE_CLIENTID:
793                                 goto out_err;
794                 }
795         }
796         return 0;
797 out_err:
798         return status;
799 }
800
801 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
802 {
803         struct nfs4_state *state;
804         struct nfs4_lock_state *lock;
805         int status = 0;
806
807         /* Note: we rely on the sp->so_states list being ordered 
808          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
809          * states first.
810          * This is needed to ensure that the server won't give us any
811          * read delegations that we have to return if, say, we are
812          * recovering after a network partition or a reboot from a
813          * server that doesn't support a grace period.
814          */
815         list_for_each_entry(state, &sp->so_states, open_states) {
816                 if (state->state == 0)
817                         continue;
818                 status = ops->recover_open(sp, state);
819                 if (status >= 0) {
820                         status = nfs4_reclaim_locks(ops, state);
821                         if (status < 0)
822                                 goto out_err;
823                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
824                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
825                                         printk("%s: Lock reclaim failed!\n",
826                                                         __FUNCTION__);
827                         }
828                         continue;
829                 }
830                 switch (status) {
831                         default:
832                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
833                                                 __FUNCTION__, status);
834                         case -ENOENT:
835                         case -NFS4ERR_RECLAIM_BAD:
836                         case -NFS4ERR_RECLAIM_CONFLICT:
837                                 /*
838                                  * Open state on this file cannot be recovered
839                                  * All we can do is revert to using the zero stateid.
840                                  */
841                                 memset(state->stateid.data, 0,
842                                         sizeof(state->stateid.data));
843                                 /* Mark the file as being 'closed' */
844                                 state->state = 0;
845                                 break;
846                         case -NFS4ERR_EXPIRED:
847                         case -NFS4ERR_NO_GRACE:
848                         case -NFS4ERR_STALE_CLIENTID:
849                                 goto out_err;
850                 }
851         }
852         return 0;
853 out_err:
854         return status;
855 }
856
857 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
858 {
859         struct nfs4_state_owner *sp;
860         struct rb_node *pos;
861         struct nfs4_state *state;
862         struct nfs4_lock_state *lock;
863
864         /* Reset all sequence ids to zero */
865         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
866                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
867                 sp->so_seqid.counter = 0;
868                 sp->so_seqid.flags = 0;
869                 spin_lock(&sp->so_lock);
870                 list_for_each_entry(state, &sp->so_states, open_states) {
871                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
872                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
873                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
874                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
875                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
876                                 lock->ls_seqid.counter = 0;
877                                 lock->ls_seqid.flags = 0;
878                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
879                         }
880                 }
881                 spin_unlock(&sp->so_lock);
882         }
883 }
884
885 static int reclaimer(void *ptr)
886 {
887         struct nfs_client *clp = ptr;
888         struct nfs4_state_owner *sp;
889         struct rb_node *pos;
890         struct nfs4_state_recovery_ops *ops;
891         struct rpc_cred *cred;
892         int status = 0;
893
894         allow_signal(SIGKILL);
895
896         /* Ensure exclusive access to NFSv4 state */
897         lock_kernel();
898         down_write(&clp->cl_sem);
899         /* Are there any NFS mounts out there? */
900         if (list_empty(&clp->cl_superblocks))
901                 goto out;
902 restart_loop:
903         ops = &nfs4_network_partition_recovery_ops;
904         /* Are there any open files on this volume? */
905         cred = nfs4_get_renew_cred(clp);
906         if (cred != NULL) {
907                 /* Yes there are: try to renew the old lease */
908                 status = nfs4_proc_renew(clp, cred);
909                 switch (status) {
910                         case 0:
911                         case -NFS4ERR_CB_PATH_DOWN:
912                                 put_rpccred(cred);
913                                 goto out;
914                         case -NFS4ERR_STALE_CLIENTID:
915                         case -NFS4ERR_LEASE_MOVED:
916                                 ops = &nfs4_reboot_recovery_ops;
917                 }
918         } else {
919                 /* "reboot" to ensure we clear all state on the server */
920                 clp->cl_boot_time = CURRENT_TIME;
921                 cred = nfs4_get_setclientid_cred(clp);
922         }
923         /* We're going to have to re-establish a clientid */
924         nfs4_state_mark_reclaim(clp);
925         status = -ENOENT;
926         if (cred != NULL) {
927                 status = nfs4_init_client(clp, cred);
928                 put_rpccred(cred);
929         }
930         if (status)
931                 goto out_error;
932         /* Mark all delegations for reclaim */
933         nfs_delegation_mark_reclaim(clp);
934         /* Note: list is protected by exclusive lock on cl->cl_sem */
935         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
936                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
937                 status = nfs4_reclaim_open_state(ops, sp);
938                 if (status < 0) {
939                         if (status == -NFS4ERR_NO_GRACE) {
940                                 ops = &nfs4_network_partition_recovery_ops;
941                                 status = nfs4_reclaim_open_state(ops, sp);
942                         }
943                         if (status == -NFS4ERR_STALE_CLIENTID)
944                                 goto restart_loop;
945                         if (status == -NFS4ERR_EXPIRED)
946                                 goto restart_loop;
947                 }
948         }
949         nfs_delegation_reap_unclaimed(clp);
950 out:
951         up_write(&clp->cl_sem);
952         unlock_kernel();
953         if (status == -NFS4ERR_CB_PATH_DOWN)
954                 nfs_handle_cb_pathdown(clp);
955         nfs4_clear_recover_bit(clp);
956         nfs_put_client(clp);
957         module_put_and_exit(0);
958         return 0;
959 out_error:
960         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
961                                 NIPQUAD(clp->cl_addr.sin_addr), -status);
962         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
963         goto out;
964 }
965
966 /*
967  * Local variables:
968  *  c-basic-offset: 8
969  * End:
970  */