Merge commit 'origin' into devel
[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                 rpc_destroy_wait_queue(&new->so_sequence.wait);
297                 kfree(new);
298         }
299         return sp;
300 }
301
302 /*
303  * Must be called with clp->cl_sem held in order to avoid races
304  * with state recovery...
305  */
306 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
307 {
308         struct nfs_client *clp = sp->so_client;
309         struct rpc_cred *cred = sp->so_cred;
310
311         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
312                 return;
313         nfs4_remove_state_owner(clp, sp);
314         spin_unlock(&clp->cl_lock);
315         rpc_destroy_wait_queue(&sp->so_sequence.wait);
316         put_rpccred(cred);
317         kfree(sp);
318 }
319
320 static struct nfs4_state *
321 nfs4_alloc_open_state(void)
322 {
323         struct nfs4_state *state;
324
325         state = kzalloc(sizeof(*state), GFP_KERNEL);
326         if (!state)
327                 return NULL;
328         atomic_set(&state->count, 1);
329         INIT_LIST_HEAD(&state->lock_states);
330         spin_lock_init(&state->state_lock);
331         seqlock_init(&state->seqlock);
332         return state;
333 }
334
335 void
336 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
337 {
338         if (state->state == mode)
339                 return;
340         /* NB! List reordering - see the reclaim code for why.  */
341         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
342                 if (mode & FMODE_WRITE)
343                         list_move(&state->open_states, &state->owner->so_states);
344                 else
345                         list_move_tail(&state->open_states, &state->owner->so_states);
346         }
347         state->state = mode;
348 }
349
350 static struct nfs4_state *
351 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
352 {
353         struct nfs_inode *nfsi = NFS_I(inode);
354         struct nfs4_state *state;
355
356         list_for_each_entry(state, &nfsi->open_states, inode_states) {
357                 if (state->owner != owner)
358                         continue;
359                 if (atomic_inc_not_zero(&state->count))
360                         return state;
361         }
362         return NULL;
363 }
364
365 static void
366 nfs4_free_open_state(struct nfs4_state *state)
367 {
368         kfree(state);
369 }
370
371 struct nfs4_state *
372 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
373 {
374         struct nfs4_state *state, *new;
375         struct nfs_inode *nfsi = NFS_I(inode);
376
377         spin_lock(&inode->i_lock);
378         state = __nfs4_find_state_byowner(inode, owner);
379         spin_unlock(&inode->i_lock);
380         if (state)
381                 goto out;
382         new = nfs4_alloc_open_state();
383         spin_lock(&owner->so_lock);
384         spin_lock(&inode->i_lock);
385         state = __nfs4_find_state_byowner(inode, owner);
386         if (state == NULL && new != NULL) {
387                 state = new;
388                 state->owner = owner;
389                 atomic_inc(&owner->so_count);
390                 list_add(&state->inode_states, &nfsi->open_states);
391                 state->inode = igrab(inode);
392                 spin_unlock(&inode->i_lock);
393                 /* Note: The reclaim code dictates that we add stateless
394                  * and read-only stateids to the end of the list */
395                 list_add_tail(&state->open_states, &owner->so_states);
396                 spin_unlock(&owner->so_lock);
397         } else {
398                 spin_unlock(&inode->i_lock);
399                 spin_unlock(&owner->so_lock);
400                 if (new)
401                         nfs4_free_open_state(new);
402         }
403 out:
404         return state;
405 }
406
407 /*
408  * Beware! Caller must be holding exactly one
409  * reference to clp->cl_sem!
410  */
411 void nfs4_put_open_state(struct nfs4_state *state)
412 {
413         struct inode *inode = state->inode;
414         struct nfs4_state_owner *owner = state->owner;
415
416         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
417                 return;
418         spin_lock(&inode->i_lock);
419         list_del(&state->inode_states);
420         list_del(&state->open_states);
421         spin_unlock(&inode->i_lock);
422         spin_unlock(&owner->so_lock);
423         iput(inode);
424         nfs4_free_open_state(state);
425         nfs4_put_state_owner(owner);
426 }
427
428 /*
429  * Close the current file.
430  */
431 static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
432 {
433         struct nfs4_state_owner *owner = state->owner;
434         int call_close = 0;
435         int newstate;
436
437         atomic_inc(&owner->so_count);
438         /* Protect against nfs4_find_state() */
439         spin_lock(&owner->so_lock);
440         switch (mode & (FMODE_READ | FMODE_WRITE)) {
441                 case FMODE_READ:
442                         state->n_rdonly--;
443                         break;
444                 case FMODE_WRITE:
445                         state->n_wronly--;
446                         break;
447                 case FMODE_READ|FMODE_WRITE:
448                         state->n_rdwr--;
449         }
450         newstate = FMODE_READ|FMODE_WRITE;
451         if (state->n_rdwr == 0) {
452                 if (state->n_rdonly == 0) {
453                         newstate &= ~FMODE_READ;
454                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
455                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
456                 }
457                 if (state->n_wronly == 0) {
458                         newstate &= ~FMODE_WRITE;
459                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
460                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
461                 }
462                 if (newstate == 0)
463                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
464         }
465         nfs4_state_set_mode_locked(state, newstate);
466         spin_unlock(&owner->so_lock);
467
468         if (!call_close) {
469                 nfs4_put_open_state(state);
470                 nfs4_put_state_owner(owner);
471         } else
472                 nfs4_do_close(path, state, wait);
473 }
474
475 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
476 {
477         __nfs4_close(path, state, mode, 0);
478 }
479
480 void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
481 {
482         __nfs4_close(path, state, mode, 1);
483 }
484
485 /*
486  * Search the state->lock_states for an existing lock_owner
487  * that is compatible with current->files
488  */
489 static struct nfs4_lock_state *
490 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
491 {
492         struct nfs4_lock_state *pos;
493         list_for_each_entry(pos, &state->lock_states, ls_locks) {
494                 if (pos->ls_owner != fl_owner)
495                         continue;
496                 atomic_inc(&pos->ls_count);
497                 return pos;
498         }
499         return NULL;
500 }
501
502 /*
503  * Return a compatible lock_state. If no initialized lock_state structure
504  * exists, return an uninitialized one.
505  *
506  */
507 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
508 {
509         struct nfs4_lock_state *lsp;
510         struct nfs_client *clp = state->owner->so_client;
511
512         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
513         if (lsp == NULL)
514                 return NULL;
515         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
516         spin_lock_init(&lsp->ls_sequence.lock);
517         INIT_LIST_HEAD(&lsp->ls_sequence.list);
518         lsp->ls_seqid.sequence = &lsp->ls_sequence;
519         atomic_set(&lsp->ls_count, 1);
520         lsp->ls_owner = fl_owner;
521         spin_lock(&clp->cl_lock);
522         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
523         spin_unlock(&clp->cl_lock);
524         INIT_LIST_HEAD(&lsp->ls_locks);
525         return lsp;
526 }
527
528 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
529 {
530         struct nfs_client *clp = lsp->ls_state->owner->so_client;
531
532         spin_lock(&clp->cl_lock);
533         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
534         spin_unlock(&clp->cl_lock);
535         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
536         kfree(lsp);
537 }
538
539 /*
540  * Return a compatible lock_state. If no initialized lock_state structure
541  * exists, return an uninitialized one.
542  *
543  * The caller must be holding clp->cl_sem
544  */
545 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
546 {
547         struct nfs4_lock_state *lsp, *new = NULL;
548         
549         for(;;) {
550                 spin_lock(&state->state_lock);
551                 lsp = __nfs4_find_lock_state(state, owner);
552                 if (lsp != NULL)
553                         break;
554                 if (new != NULL) {
555                         new->ls_state = state;
556                         list_add(&new->ls_locks, &state->lock_states);
557                         set_bit(LK_STATE_IN_USE, &state->flags);
558                         lsp = new;
559                         new = NULL;
560                         break;
561                 }
562                 spin_unlock(&state->state_lock);
563                 new = nfs4_alloc_lock_state(state, owner);
564                 if (new == NULL)
565                         return NULL;
566         }
567         spin_unlock(&state->state_lock);
568         if (new != NULL)
569                 nfs4_free_lock_state(new);
570         return lsp;
571 }
572
573 /*
574  * Release reference to lock_state, and free it if we see that
575  * it is no longer in use
576  */
577 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
578 {
579         struct nfs4_state *state;
580
581         if (lsp == NULL)
582                 return;
583         state = lsp->ls_state;
584         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
585                 return;
586         list_del(&lsp->ls_locks);
587         if (list_empty(&state->lock_states))
588                 clear_bit(LK_STATE_IN_USE, &state->flags);
589         spin_unlock(&state->state_lock);
590         nfs4_free_lock_state(lsp);
591 }
592
593 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
594 {
595         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
596
597         dst->fl_u.nfs4_fl.owner = lsp;
598         atomic_inc(&lsp->ls_count);
599 }
600
601 static void nfs4_fl_release_lock(struct file_lock *fl)
602 {
603         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
604 }
605
606 static struct file_lock_operations nfs4_fl_lock_ops = {
607         .fl_copy_lock = nfs4_fl_copy_lock,
608         .fl_release_private = nfs4_fl_release_lock,
609 };
610
611 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
612 {
613         struct nfs4_lock_state *lsp;
614
615         if (fl->fl_ops != NULL)
616                 return 0;
617         lsp = nfs4_get_lock_state(state, fl->fl_owner);
618         if (lsp == NULL)
619                 return -ENOMEM;
620         fl->fl_u.nfs4_fl.owner = lsp;
621         fl->fl_ops = &nfs4_fl_lock_ops;
622         return 0;
623 }
624
625 /*
626  * Byte-range lock aware utility to initialize the stateid of read/write
627  * requests.
628  */
629 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
630 {
631         struct nfs4_lock_state *lsp;
632         int seq;
633
634         do {
635                 seq = read_seqbegin(&state->seqlock);
636                 memcpy(dst, &state->stateid, sizeof(*dst));
637         } while (read_seqretry(&state->seqlock, seq));
638         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
639                 return;
640
641         spin_lock(&state->state_lock);
642         lsp = __nfs4_find_lock_state(state, fl_owner);
643         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
644                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
645         spin_unlock(&state->state_lock);
646         nfs4_put_lock_state(lsp);
647 }
648
649 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
650 {
651         struct nfs_seqid *new;
652
653         new = kmalloc(sizeof(*new), GFP_KERNEL);
654         if (new != NULL) {
655                 new->sequence = counter;
656                 INIT_LIST_HEAD(&new->list);
657         }
658         return new;
659 }
660
661 void nfs_free_seqid(struct nfs_seqid *seqid)
662 {
663         if (!list_empty(&seqid->list)) {
664                 struct rpc_sequence *sequence = seqid->sequence->sequence;
665
666                 spin_lock(&sequence->lock);
667                 list_del(&seqid->list);
668                 spin_unlock(&sequence->lock);
669                 rpc_wake_up(&sequence->wait);
670         }
671         kfree(seqid);
672 }
673
674 /*
675  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
676  * failed with a seqid incrementing error -
677  * see comments nfs_fs.h:seqid_mutating_error()
678  */
679 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
680 {
681         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
682         switch (status) {
683                 case 0:
684                         break;
685                 case -NFS4ERR_BAD_SEQID:
686                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
687                                 return;
688                         printk(KERN_WARNING "NFS: v4 server returned a bad"
689                                         " sequence-id error on an"
690                                         " unconfirmed sequence %p!\n",
691                                         seqid->sequence);
692                 case -NFS4ERR_STALE_CLIENTID:
693                 case -NFS4ERR_STALE_STATEID:
694                 case -NFS4ERR_BAD_STATEID:
695                 case -NFS4ERR_BADXDR:
696                 case -NFS4ERR_RESOURCE:
697                 case -NFS4ERR_NOFILEHANDLE:
698                         /* Non-seqid mutating errors */
699                         return;
700         };
701         /*
702          * Note: no locking needed as we are guaranteed to be first
703          * on the sequence list
704          */
705         seqid->sequence->counter++;
706 }
707
708 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
709 {
710         if (status == -NFS4ERR_BAD_SEQID) {
711                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
712                                 struct nfs4_state_owner, so_seqid);
713                 nfs4_drop_state_owner(sp);
714         }
715         nfs_increment_seqid(status, seqid);
716 }
717
718 /*
719  * Increment the seqid if the LOCK/LOCKU succeeded, or
720  * failed with a seqid incrementing error -
721  * see comments nfs_fs.h:seqid_mutating_error()
722  */
723 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
724 {
725         nfs_increment_seqid(status, seqid);
726 }
727
728 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
729 {
730         struct rpc_sequence *sequence = seqid->sequence->sequence;
731         int status = 0;
732
733         spin_lock(&sequence->lock);
734         if (list_empty(&seqid->list))
735                 list_add_tail(&seqid->list, &sequence->list);
736         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
737                 goto unlock;
738         rpc_sleep_on(&sequence->wait, task, NULL);
739         status = -EAGAIN;
740 unlock:
741         spin_unlock(&sequence->lock);
742         return status;
743 }
744
745 static int reclaimer(void *);
746
747 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
748 {
749         smp_mb__before_clear_bit();
750         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
751         smp_mb__after_clear_bit();
752         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
753         rpc_wake_up(&clp->cl_rpcwaitq);
754 }
755
756 /*
757  * State recovery routine
758  */
759 static void nfs4_recover_state(struct nfs_client *clp)
760 {
761         struct task_struct *task;
762
763         __module_get(THIS_MODULE);
764         atomic_inc(&clp->cl_count);
765         task = kthread_run(reclaimer, clp, "%s-reclaim",
766                                 rpc_peeraddr2str(clp->cl_rpcclient,
767                                                         RPC_DISPLAY_ADDR));
768         if (!IS_ERR(task))
769                 return;
770         nfs4_clear_recover_bit(clp);
771         nfs_put_client(clp);
772         module_put(THIS_MODULE);
773 }
774
775 /*
776  * Schedule a state recovery attempt
777  */
778 void nfs4_schedule_state_recovery(struct nfs_client *clp)
779 {
780         if (!clp)
781                 return;
782         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
783                 nfs4_recover_state(clp);
784 }
785
786 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
787 {
788         struct inode *inode = state->inode;
789         struct file_lock *fl;
790         int status = 0;
791
792         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
793                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
794                         continue;
795                 if (nfs_file_open_context(fl->fl_file)->state != state)
796                         continue;
797                 status = ops->recover_lock(state, fl);
798                 if (status >= 0)
799                         continue;
800                 switch (status) {
801                         default:
802                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
803                                                 __FUNCTION__, status);
804                         case -NFS4ERR_EXPIRED:
805                         case -NFS4ERR_NO_GRACE:
806                         case -NFS4ERR_RECLAIM_BAD:
807                         case -NFS4ERR_RECLAIM_CONFLICT:
808                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
809                                 break;
810                         case -NFS4ERR_STALE_CLIENTID:
811                                 goto out_err;
812                 }
813         }
814         return 0;
815 out_err:
816         return status;
817 }
818
819 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
820 {
821         struct nfs4_state *state;
822         struct nfs4_lock_state *lock;
823         int status = 0;
824
825         /* Note: we rely on the sp->so_states list being ordered 
826          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
827          * states first.
828          * This is needed to ensure that the server won't give us any
829          * read delegations that we have to return if, say, we are
830          * recovering after a network partition or a reboot from a
831          * server that doesn't support a grace period.
832          */
833         list_for_each_entry(state, &sp->so_states, open_states) {
834                 if (state->state == 0)
835                         continue;
836                 status = ops->recover_open(sp, state);
837                 if (status >= 0) {
838                         status = nfs4_reclaim_locks(ops, state);
839                         if (status < 0)
840                                 goto out_err;
841                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
842                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
843                                         printk("%s: Lock reclaim failed!\n",
844                                                         __FUNCTION__);
845                         }
846                         continue;
847                 }
848                 switch (status) {
849                         default:
850                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
851                                                 __FUNCTION__, status);
852                         case -ENOENT:
853                         case -NFS4ERR_RECLAIM_BAD:
854                         case -NFS4ERR_RECLAIM_CONFLICT:
855                                 /*
856                                  * Open state on this file cannot be recovered
857                                  * All we can do is revert to using the zero stateid.
858                                  */
859                                 memset(state->stateid.data, 0,
860                                         sizeof(state->stateid.data));
861                                 /* Mark the file as being 'closed' */
862                                 state->state = 0;
863                                 break;
864                         case -NFS4ERR_EXPIRED:
865                         case -NFS4ERR_NO_GRACE:
866                         case -NFS4ERR_STALE_CLIENTID:
867                                 goto out_err;
868                 }
869         }
870         return 0;
871 out_err:
872         return status;
873 }
874
875 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
876 {
877         struct nfs4_state_owner *sp;
878         struct rb_node *pos;
879         struct nfs4_state *state;
880         struct nfs4_lock_state *lock;
881
882         /* Reset all sequence ids to zero */
883         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
884                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
885                 sp->so_seqid.counter = 0;
886                 sp->so_seqid.flags = 0;
887                 spin_lock(&sp->so_lock);
888                 list_for_each_entry(state, &sp->so_states, open_states) {
889                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
890                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
891                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
892                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
893                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
894                                 lock->ls_seqid.counter = 0;
895                                 lock->ls_seqid.flags = 0;
896                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
897                         }
898                 }
899                 spin_unlock(&sp->so_lock);
900         }
901 }
902
903 static int reclaimer(void *ptr)
904 {
905         struct nfs_client *clp = ptr;
906         struct nfs4_state_owner *sp;
907         struct rb_node *pos;
908         struct nfs4_state_recovery_ops *ops;
909         struct rpc_cred *cred;
910         int status = 0;
911
912         allow_signal(SIGKILL);
913
914         /* Ensure exclusive access to NFSv4 state */
915         lock_kernel();
916         down_write(&clp->cl_sem);
917         /* Are there any NFS mounts out there? */
918         if (list_empty(&clp->cl_superblocks))
919                 goto out;
920 restart_loop:
921         ops = &nfs4_network_partition_recovery_ops;
922         /* Are there any open files on this volume? */
923         cred = nfs4_get_renew_cred(clp);
924         if (cred != NULL) {
925                 /* Yes there are: try to renew the old lease */
926                 status = nfs4_proc_renew(clp, cred);
927                 switch (status) {
928                         case 0:
929                         case -NFS4ERR_CB_PATH_DOWN:
930                                 put_rpccred(cred);
931                                 goto out;
932                         case -NFS4ERR_STALE_CLIENTID:
933                         case -NFS4ERR_LEASE_MOVED:
934                                 ops = &nfs4_reboot_recovery_ops;
935                 }
936         } else {
937                 /* "reboot" to ensure we clear all state on the server */
938                 clp->cl_boot_time = CURRENT_TIME;
939                 cred = nfs4_get_setclientid_cred(clp);
940         }
941         /* We're going to have to re-establish a clientid */
942         nfs4_state_mark_reclaim(clp);
943         status = -ENOENT;
944         if (cred != NULL) {
945                 status = nfs4_init_client(clp, cred);
946                 put_rpccred(cred);
947         }
948         if (status)
949                 goto out_error;
950         /* Mark all delegations for reclaim */
951         nfs_delegation_mark_reclaim(clp);
952         /* Note: list is protected by exclusive lock on cl->cl_sem */
953         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
954                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
955                 status = nfs4_reclaim_open_state(ops, sp);
956                 if (status < 0) {
957                         if (status == -NFS4ERR_NO_GRACE) {
958                                 ops = &nfs4_network_partition_recovery_ops;
959                                 status = nfs4_reclaim_open_state(ops, sp);
960                         }
961                         if (status == -NFS4ERR_STALE_CLIENTID)
962                                 goto restart_loop;
963                         if (status == -NFS4ERR_EXPIRED)
964                                 goto restart_loop;
965                 }
966         }
967         nfs_delegation_reap_unclaimed(clp);
968 out:
969         up_write(&clp->cl_sem);
970         unlock_kernel();
971         if (status == -NFS4ERR_CB_PATH_DOWN)
972                 nfs_handle_cb_pathdown(clp);
973         nfs4_clear_recover_bit(clp);
974         nfs_put_client(clp);
975         module_put_and_exit(0);
976         return 0;
977 out_error:
978         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %s"
979                         " with error %d\n", clp->cl_hostname, -status);
980         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
981         goto out;
982 }
983
984 /*
985  * Local variables:
986  *  c-basic-offset: 8
987  * End:
988  */