Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm
[linux-2.6] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/smp_lock.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/workqueue.h>
50 #include <linux/bitops.h>
51
52 #include "nfs4_fs.h"
53 #include "callback.h"
54 #include "delegation.h"
55 #include "internal.h"
56
57 #define OPENOWNER_POOL_SIZE     8
58
59 const nfs4_stateid zero_stateid;
60
61 static LIST_HEAD(nfs4_clientid_list);
62
63 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 static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
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, wait);
470 }
471
472 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
473 {
474         __nfs4_close(path, state, mode, 0);
475 }
476
477 void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
478 {
479         __nfs4_close(path, state, mode, 1);
480 }
481
482 /*
483  * Search the state->lock_states for an existing lock_owner
484  * that is compatible with current->files
485  */
486 static struct nfs4_lock_state *
487 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
488 {
489         struct nfs4_lock_state *pos;
490         list_for_each_entry(pos, &state->lock_states, ls_locks) {
491                 if (pos->ls_owner != fl_owner)
492                         continue;
493                 atomic_inc(&pos->ls_count);
494                 return pos;
495         }
496         return NULL;
497 }
498
499 /*
500  * Return a compatible lock_state. If no initialized lock_state structure
501  * exists, return an uninitialized one.
502  *
503  */
504 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
505 {
506         struct nfs4_lock_state *lsp;
507         struct nfs_client *clp = state->owner->so_client;
508
509         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
510         if (lsp == NULL)
511                 return NULL;
512         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
513         spin_lock_init(&lsp->ls_sequence.lock);
514         INIT_LIST_HEAD(&lsp->ls_sequence.list);
515         lsp->ls_seqid.sequence = &lsp->ls_sequence;
516         atomic_set(&lsp->ls_count, 1);
517         lsp->ls_owner = fl_owner;
518         spin_lock(&clp->cl_lock);
519         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
520         spin_unlock(&clp->cl_lock);
521         INIT_LIST_HEAD(&lsp->ls_locks);
522         return lsp;
523 }
524
525 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
526 {
527         struct nfs_client *clp = lsp->ls_state->owner->so_client;
528
529         spin_lock(&clp->cl_lock);
530         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
531         spin_unlock(&clp->cl_lock);
532         kfree(lsp);
533 }
534
535 /*
536  * Return a compatible lock_state. If no initialized lock_state structure
537  * exists, return an uninitialized one.
538  *
539  * The caller must be holding clp->cl_sem
540  */
541 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
542 {
543         struct nfs4_lock_state *lsp, *new = NULL;
544         
545         for(;;) {
546                 spin_lock(&state->state_lock);
547                 lsp = __nfs4_find_lock_state(state, owner);
548                 if (lsp != NULL)
549                         break;
550                 if (new != NULL) {
551                         new->ls_state = state;
552                         list_add(&new->ls_locks, &state->lock_states);
553                         set_bit(LK_STATE_IN_USE, &state->flags);
554                         lsp = new;
555                         new = NULL;
556                         break;
557                 }
558                 spin_unlock(&state->state_lock);
559                 new = nfs4_alloc_lock_state(state, owner);
560                 if (new == NULL)
561                         return NULL;
562         }
563         spin_unlock(&state->state_lock);
564         if (new != NULL)
565                 nfs4_free_lock_state(new);
566         return lsp;
567 }
568
569 /*
570  * Release reference to lock_state, and free it if we see that
571  * it is no longer in use
572  */
573 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
574 {
575         struct nfs4_state *state;
576
577         if (lsp == NULL)
578                 return;
579         state = lsp->ls_state;
580         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
581                 return;
582         list_del(&lsp->ls_locks);
583         if (list_empty(&state->lock_states))
584                 clear_bit(LK_STATE_IN_USE, &state->flags);
585         spin_unlock(&state->state_lock);
586         nfs4_free_lock_state(lsp);
587 }
588
589 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
590 {
591         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
592
593         dst->fl_u.nfs4_fl.owner = lsp;
594         atomic_inc(&lsp->ls_count);
595 }
596
597 static void nfs4_fl_release_lock(struct file_lock *fl)
598 {
599         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
600 }
601
602 static struct file_lock_operations nfs4_fl_lock_ops = {
603         .fl_copy_lock = nfs4_fl_copy_lock,
604         .fl_release_private = nfs4_fl_release_lock,
605 };
606
607 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
608 {
609         struct nfs4_lock_state *lsp;
610
611         if (fl->fl_ops != NULL)
612                 return 0;
613         lsp = nfs4_get_lock_state(state, fl->fl_owner);
614         if (lsp == NULL)
615                 return -ENOMEM;
616         fl->fl_u.nfs4_fl.owner = lsp;
617         fl->fl_ops = &nfs4_fl_lock_ops;
618         return 0;
619 }
620
621 /*
622  * Byte-range lock aware utility to initialize the stateid of read/write
623  * requests.
624  */
625 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
626 {
627         struct nfs4_lock_state *lsp;
628         int seq;
629
630         do {
631                 seq = read_seqbegin(&state->seqlock);
632                 memcpy(dst, &state->stateid, sizeof(*dst));
633         } while (read_seqretry(&state->seqlock, seq));
634         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
635                 return;
636
637         spin_lock(&state->state_lock);
638         lsp = __nfs4_find_lock_state(state, fl_owner);
639         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
640                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
641         spin_unlock(&state->state_lock);
642         nfs4_put_lock_state(lsp);
643 }
644
645 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
646 {
647         struct nfs_seqid *new;
648
649         new = kmalloc(sizeof(*new), GFP_KERNEL);
650         if (new != NULL) {
651                 new->sequence = counter;
652                 INIT_LIST_HEAD(&new->list);
653         }
654         return new;
655 }
656
657 void nfs_free_seqid(struct nfs_seqid *seqid)
658 {
659         if (!list_empty(&seqid->list)) {
660                 struct rpc_sequence *sequence = seqid->sequence->sequence;
661
662                 spin_lock(&sequence->lock);
663                 list_del(&seqid->list);
664                 spin_unlock(&sequence->lock);
665                 rpc_wake_up(&sequence->wait);
666         }
667         kfree(seqid);
668 }
669
670 /*
671  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
672  * failed with a seqid incrementing error -
673  * see comments nfs_fs.h:seqid_mutating_error()
674  */
675 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
676 {
677         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
678         switch (status) {
679                 case 0:
680                         break;
681                 case -NFS4ERR_BAD_SEQID:
682                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
683                                 return;
684                         printk(KERN_WARNING "NFS: v4 server returned a bad"
685                                         "sequence-id error on an"
686                                         "unconfirmed sequence %p!\n",
687                                         seqid->sequence);
688                 case -NFS4ERR_STALE_CLIENTID:
689                 case -NFS4ERR_STALE_STATEID:
690                 case -NFS4ERR_BAD_STATEID:
691                 case -NFS4ERR_BADXDR:
692                 case -NFS4ERR_RESOURCE:
693                 case -NFS4ERR_NOFILEHANDLE:
694                         /* Non-seqid mutating errors */
695                         return;
696         };
697         /*
698          * Note: no locking needed as we are guaranteed to be first
699          * on the sequence list
700          */
701         seqid->sequence->counter++;
702 }
703
704 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
705 {
706         if (status == -NFS4ERR_BAD_SEQID) {
707                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
708                                 struct nfs4_state_owner, so_seqid);
709                 nfs4_drop_state_owner(sp);
710         }
711         nfs_increment_seqid(status, seqid);
712 }
713
714 /*
715  * Increment the seqid if the LOCK/LOCKU succeeded, or
716  * failed with a seqid incrementing error -
717  * see comments nfs_fs.h:seqid_mutating_error()
718  */
719 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
720 {
721         nfs_increment_seqid(status, seqid);
722 }
723
724 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
725 {
726         struct rpc_sequence *sequence = seqid->sequence->sequence;
727         int status = 0;
728
729         spin_lock(&sequence->lock);
730         if (list_empty(&seqid->list))
731                 list_add_tail(&seqid->list, &sequence->list);
732         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
733                 goto unlock;
734         rpc_sleep_on(&sequence->wait, task, NULL, NULL);
735         status = -EAGAIN;
736 unlock:
737         spin_unlock(&sequence->lock);
738         return status;
739 }
740
741 static int reclaimer(void *);
742
743 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
744 {
745         smp_mb__before_clear_bit();
746         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
747         smp_mb__after_clear_bit();
748         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
749         rpc_wake_up(&clp->cl_rpcwaitq);
750 }
751
752 /*
753  * State recovery routine
754  */
755 static void nfs4_recover_state(struct nfs_client *clp)
756 {
757         struct task_struct *task;
758
759         __module_get(THIS_MODULE);
760         atomic_inc(&clp->cl_count);
761         task = kthread_run(reclaimer, clp, "%s-reclaim",
762                                 rpc_peeraddr2str(clp->cl_rpcclient,
763                                                         RPC_DISPLAY_ADDR));
764         if (!IS_ERR(task))
765                 return;
766         nfs4_clear_recover_bit(clp);
767         nfs_put_client(clp);
768         module_put(THIS_MODULE);
769 }
770
771 /*
772  * Schedule a state recovery attempt
773  */
774 void nfs4_schedule_state_recovery(struct nfs_client *clp)
775 {
776         if (!clp)
777                 return;
778         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
779                 nfs4_recover_state(clp);
780 }
781
782 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
783 {
784         struct inode *inode = state->inode;
785         struct file_lock *fl;
786         int status = 0;
787
788         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
789                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
790                         continue;
791                 if (nfs_file_open_context(fl->fl_file)->state != state)
792                         continue;
793                 status = ops->recover_lock(state, fl);
794                 if (status >= 0)
795                         continue;
796                 switch (status) {
797                         default:
798                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
799                                                 __FUNCTION__, status);
800                         case -NFS4ERR_EXPIRED:
801                         case -NFS4ERR_NO_GRACE:
802                         case -NFS4ERR_RECLAIM_BAD:
803                         case -NFS4ERR_RECLAIM_CONFLICT:
804                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
805                                 break;
806                         case -NFS4ERR_STALE_CLIENTID:
807                                 goto out_err;
808                 }
809         }
810         return 0;
811 out_err:
812         return status;
813 }
814
815 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
816 {
817         struct nfs4_state *state;
818         struct nfs4_lock_state *lock;
819         int status = 0;
820
821         /* Note: we rely on the sp->so_states list being ordered 
822          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
823          * states first.
824          * This is needed to ensure that the server won't give us any
825          * read delegations that we have to return if, say, we are
826          * recovering after a network partition or a reboot from a
827          * server that doesn't support a grace period.
828          */
829         list_for_each_entry(state, &sp->so_states, open_states) {
830                 if (state->state == 0)
831                         continue;
832                 status = ops->recover_open(sp, state);
833                 if (status >= 0) {
834                         status = nfs4_reclaim_locks(ops, state);
835                         if (status < 0)
836                                 goto out_err;
837                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
838                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
839                                         printk("%s: Lock reclaim failed!\n",
840                                                         __FUNCTION__);
841                         }
842                         continue;
843                 }
844                 switch (status) {
845                         default:
846                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
847                                                 __FUNCTION__, status);
848                         case -ENOENT:
849                         case -NFS4ERR_RECLAIM_BAD:
850                         case -NFS4ERR_RECLAIM_CONFLICT:
851                                 /*
852                                  * Open state on this file cannot be recovered
853                                  * All we can do is revert to using the zero stateid.
854                                  */
855                                 memset(state->stateid.data, 0,
856                                         sizeof(state->stateid.data));
857                                 /* Mark the file as being 'closed' */
858                                 state->state = 0;
859                                 break;
860                         case -NFS4ERR_EXPIRED:
861                         case -NFS4ERR_NO_GRACE:
862                         case -NFS4ERR_STALE_CLIENTID:
863                                 goto out_err;
864                 }
865         }
866         return 0;
867 out_err:
868         return status;
869 }
870
871 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
872 {
873         struct nfs4_state_owner *sp;
874         struct rb_node *pos;
875         struct nfs4_state *state;
876         struct nfs4_lock_state *lock;
877
878         /* Reset all sequence ids to zero */
879         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
880                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
881                 sp->so_seqid.counter = 0;
882                 sp->so_seqid.flags = 0;
883                 spin_lock(&sp->so_lock);
884                 list_for_each_entry(state, &sp->so_states, open_states) {
885                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
886                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
887                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
888                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
889                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
890                                 lock->ls_seqid.counter = 0;
891                                 lock->ls_seqid.flags = 0;
892                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
893                         }
894                 }
895                 spin_unlock(&sp->so_lock);
896         }
897 }
898
899 static int reclaimer(void *ptr)
900 {
901         struct nfs_client *clp = ptr;
902         struct nfs4_state_owner *sp;
903         struct rb_node *pos;
904         struct nfs4_state_recovery_ops *ops;
905         struct rpc_cred *cred;
906         int status = 0;
907
908         allow_signal(SIGKILL);
909
910         /* Ensure exclusive access to NFSv4 state */
911         lock_kernel();
912         down_write(&clp->cl_sem);
913         /* Are there any NFS mounts out there? */
914         if (list_empty(&clp->cl_superblocks))
915                 goto out;
916 restart_loop:
917         ops = &nfs4_network_partition_recovery_ops;
918         /* Are there any open files on this volume? */
919         cred = nfs4_get_renew_cred(clp);
920         if (cred != NULL) {
921                 /* Yes there are: try to renew the old lease */
922                 status = nfs4_proc_renew(clp, cred);
923                 switch (status) {
924                         case 0:
925                         case -NFS4ERR_CB_PATH_DOWN:
926                                 put_rpccred(cred);
927                                 goto out;
928                         case -NFS4ERR_STALE_CLIENTID:
929                         case -NFS4ERR_LEASE_MOVED:
930                                 ops = &nfs4_reboot_recovery_ops;
931                 }
932         } else {
933                 /* "reboot" to ensure we clear all state on the server */
934                 clp->cl_boot_time = CURRENT_TIME;
935                 cred = nfs4_get_setclientid_cred(clp);
936         }
937         /* We're going to have to re-establish a clientid */
938         nfs4_state_mark_reclaim(clp);
939         status = -ENOENT;
940         if (cred != NULL) {
941                 status = nfs4_init_client(clp, cred);
942                 put_rpccred(cred);
943         }
944         if (status)
945                 goto out_error;
946         /* Mark all delegations for reclaim */
947         nfs_delegation_mark_reclaim(clp);
948         /* Note: list is protected by exclusive lock on cl->cl_sem */
949         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
950                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
951                 status = nfs4_reclaim_open_state(ops, sp);
952                 if (status < 0) {
953                         if (status == -NFS4ERR_NO_GRACE) {
954                                 ops = &nfs4_network_partition_recovery_ops;
955                                 status = nfs4_reclaim_open_state(ops, sp);
956                         }
957                         if (status == -NFS4ERR_STALE_CLIENTID)
958                                 goto restart_loop;
959                         if (status == -NFS4ERR_EXPIRED)
960                                 goto restart_loop;
961                 }
962         }
963         nfs_delegation_reap_unclaimed(clp);
964 out:
965         up_write(&clp->cl_sem);
966         unlock_kernel();
967         if (status == -NFS4ERR_CB_PATH_DOWN)
968                 nfs_handle_cb_pathdown(clp);
969         nfs4_clear_recover_bit(clp);
970         nfs_put_client(clp);
971         module_put_and_exit(0);
972         return 0;
973 out_error:
974         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %s"
975                         " with error %d\n", clp->cl_hostname, -status);
976         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
977         goto out;
978 }
979
980 /*
981  * Local variables:
982  *  c-basic-offset: 8
983  * End:
984  */