NFSv4: Clean up for the state loss reclaimer
[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 static struct rpc_cred *nfs4_get_machine_cred(struct nfs_client *clp)
75 {
76         struct rpc_cred *cred = NULL;
77
78         spin_lock(&clp->cl_lock);
79         if (clp->cl_machine_cred != NULL)
80                 cred = get_rpccred(clp->cl_machine_cred);
81         spin_unlock(&clp->cl_lock);
82         return cred;
83 }
84
85 static void nfs4_clear_machine_cred(struct nfs_client *clp)
86 {
87         struct rpc_cred *cred;
88
89         spin_lock(&clp->cl_lock);
90         cred = clp->cl_machine_cred;
91         clp->cl_machine_cred = NULL;
92         spin_unlock(&clp->cl_lock);
93         if (cred != NULL)
94                 put_rpccred(cred);
95 }
96
97 struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
98 {
99         struct nfs4_state_owner *sp;
100         struct rb_node *pos;
101         struct rpc_cred *cred = NULL;
102
103         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
104                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
105                 if (list_empty(&sp->so_states))
106                         continue;
107                 cred = get_rpccred(sp->so_cred);
108                 break;
109         }
110         return cred;
111 }
112
113 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
114 {
115         struct nfs4_state_owner *sp;
116         struct rb_node *pos;
117         struct rpc_cred *cred;
118
119         cred = nfs4_get_machine_cred(clp);
120         if (cred != NULL)
121                 goto out;
122         pos = rb_first(&clp->cl_state_owners);
123         if (pos != NULL) {
124                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
125                 cred = get_rpccred(sp->so_cred);
126         }
127 out:
128         return cred;
129 }
130
131 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
132                 __u64 minval, int maxbits)
133 {
134         struct rb_node **p, *parent;
135         struct nfs_unique_id *pos;
136         __u64 mask = ~0ULL;
137
138         if (maxbits < 64)
139                 mask = (1ULL << maxbits) - 1ULL;
140
141         /* Ensure distribution is more or less flat */
142         get_random_bytes(&new->id, sizeof(new->id));
143         new->id &= mask;
144         if (new->id < minval)
145                 new->id += minval;
146 retry:
147         p = &root->rb_node;
148         parent = NULL;
149
150         while (*p != NULL) {
151                 parent = *p;
152                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
153
154                 if (new->id < pos->id)
155                         p = &(*p)->rb_left;
156                 else if (new->id > pos->id)
157                         p = &(*p)->rb_right;
158                 else
159                         goto id_exists;
160         }
161         rb_link_node(&new->rb_node, parent, p);
162         rb_insert_color(&new->rb_node, root);
163         return;
164 id_exists:
165         for (;;) {
166                 new->id++;
167                 if (new->id < minval || (new->id & mask) != new->id) {
168                         new->id = minval;
169                         break;
170                 }
171                 parent = rb_next(parent);
172                 if (parent == NULL)
173                         break;
174                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
175                 if (new->id < pos->id)
176                         break;
177         }
178         goto retry;
179 }
180
181 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
182 {
183         rb_erase(&id->rb_node, root);
184 }
185
186 static struct nfs4_state_owner *
187 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
188 {
189         struct nfs_client *clp = server->nfs_client;
190         struct rb_node **p = &clp->cl_state_owners.rb_node,
191                        *parent = NULL;
192         struct nfs4_state_owner *sp, *res = NULL;
193
194         while (*p != NULL) {
195                 parent = *p;
196                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
197
198                 if (server < sp->so_server) {
199                         p = &parent->rb_left;
200                         continue;
201                 }
202                 if (server > sp->so_server) {
203                         p = &parent->rb_right;
204                         continue;
205                 }
206                 if (cred < sp->so_cred)
207                         p = &parent->rb_left;
208                 else if (cred > sp->so_cred)
209                         p = &parent->rb_right;
210                 else {
211                         atomic_inc(&sp->so_count);
212                         res = sp;
213                         break;
214                 }
215         }
216         return res;
217 }
218
219 static struct nfs4_state_owner *
220 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
221 {
222         struct rb_node **p = &clp->cl_state_owners.rb_node,
223                        *parent = NULL;
224         struct nfs4_state_owner *sp;
225
226         while (*p != NULL) {
227                 parent = *p;
228                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
229
230                 if (new->so_server < sp->so_server) {
231                         p = &parent->rb_left;
232                         continue;
233                 }
234                 if (new->so_server > sp->so_server) {
235                         p = &parent->rb_right;
236                         continue;
237                 }
238                 if (new->so_cred < sp->so_cred)
239                         p = &parent->rb_left;
240                 else if (new->so_cred > sp->so_cred)
241                         p = &parent->rb_right;
242                 else {
243                         atomic_inc(&sp->so_count);
244                         return sp;
245                 }
246         }
247         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
248         rb_link_node(&new->so_client_node, parent, p);
249         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
250         return new;
251 }
252
253 static void
254 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
255 {
256         if (!RB_EMPTY_NODE(&sp->so_client_node))
257                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
258         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
259 }
260
261 /*
262  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
263  * create a new state_owner.
264  *
265  */
266 static struct nfs4_state_owner *
267 nfs4_alloc_state_owner(void)
268 {
269         struct nfs4_state_owner *sp;
270
271         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
272         if (!sp)
273                 return NULL;
274         spin_lock_init(&sp->so_lock);
275         INIT_LIST_HEAD(&sp->so_states);
276         INIT_LIST_HEAD(&sp->so_delegations);
277         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
278         sp->so_seqid.sequence = &sp->so_sequence;
279         spin_lock_init(&sp->so_sequence.lock);
280         INIT_LIST_HEAD(&sp->so_sequence.list);
281         atomic_set(&sp->so_count, 1);
282         return sp;
283 }
284
285 static void
286 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
287 {
288         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
289                 struct nfs_client *clp = sp->so_client;
290
291                 spin_lock(&clp->cl_lock);
292                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
293                 RB_CLEAR_NODE(&sp->so_client_node);
294                 spin_unlock(&clp->cl_lock);
295         }
296 }
297
298 /*
299  * Note: must be called with clp->cl_sem held in order to prevent races
300  *       with reboot recovery!
301  */
302 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
303 {
304         struct nfs_client *clp = server->nfs_client;
305         struct nfs4_state_owner *sp, *new;
306
307         spin_lock(&clp->cl_lock);
308         sp = nfs4_find_state_owner(server, cred);
309         spin_unlock(&clp->cl_lock);
310         if (sp != NULL)
311                 return sp;
312         new = nfs4_alloc_state_owner();
313         if (new == NULL)
314                 return NULL;
315         new->so_client = clp;
316         new->so_server = server;
317         new->so_cred = cred;
318         spin_lock(&clp->cl_lock);
319         sp = nfs4_insert_state_owner(clp, new);
320         spin_unlock(&clp->cl_lock);
321         if (sp == new)
322                 get_rpccred(cred);
323         else {
324                 rpc_destroy_wait_queue(&new->so_sequence.wait);
325                 kfree(new);
326         }
327         return sp;
328 }
329
330 /*
331  * Must be called with clp->cl_sem held in order to avoid races
332  * with state recovery...
333  */
334 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
335 {
336         struct nfs_client *clp = sp->so_client;
337         struct rpc_cred *cred = sp->so_cred;
338
339         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
340                 return;
341         nfs4_remove_state_owner(clp, sp);
342         spin_unlock(&clp->cl_lock);
343         rpc_destroy_wait_queue(&sp->so_sequence.wait);
344         put_rpccred(cred);
345         kfree(sp);
346 }
347
348 static struct nfs4_state *
349 nfs4_alloc_open_state(void)
350 {
351         struct nfs4_state *state;
352
353         state = kzalloc(sizeof(*state), GFP_KERNEL);
354         if (!state)
355                 return NULL;
356         atomic_set(&state->count, 1);
357         INIT_LIST_HEAD(&state->lock_states);
358         spin_lock_init(&state->state_lock);
359         seqlock_init(&state->seqlock);
360         return state;
361 }
362
363 void
364 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
365 {
366         if (state->state == mode)
367                 return;
368         /* NB! List reordering - see the reclaim code for why.  */
369         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
370                 if (mode & FMODE_WRITE)
371                         list_move(&state->open_states, &state->owner->so_states);
372                 else
373                         list_move_tail(&state->open_states, &state->owner->so_states);
374         }
375         state->state = mode;
376 }
377
378 static struct nfs4_state *
379 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
380 {
381         struct nfs_inode *nfsi = NFS_I(inode);
382         struct nfs4_state *state;
383
384         list_for_each_entry(state, &nfsi->open_states, inode_states) {
385                 if (state->owner != owner)
386                         continue;
387                 if (atomic_inc_not_zero(&state->count))
388                         return state;
389         }
390         return NULL;
391 }
392
393 static void
394 nfs4_free_open_state(struct nfs4_state *state)
395 {
396         kfree(state);
397 }
398
399 struct nfs4_state *
400 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
401 {
402         struct nfs4_state *state, *new;
403         struct nfs_inode *nfsi = NFS_I(inode);
404
405         spin_lock(&inode->i_lock);
406         state = __nfs4_find_state_byowner(inode, owner);
407         spin_unlock(&inode->i_lock);
408         if (state)
409                 goto out;
410         new = nfs4_alloc_open_state();
411         spin_lock(&owner->so_lock);
412         spin_lock(&inode->i_lock);
413         state = __nfs4_find_state_byowner(inode, owner);
414         if (state == NULL && new != NULL) {
415                 state = new;
416                 state->owner = owner;
417                 atomic_inc(&owner->so_count);
418                 list_add(&state->inode_states, &nfsi->open_states);
419                 state->inode = igrab(inode);
420                 spin_unlock(&inode->i_lock);
421                 /* Note: The reclaim code dictates that we add stateless
422                  * and read-only stateids to the end of the list */
423                 list_add_tail(&state->open_states, &owner->so_states);
424                 spin_unlock(&owner->so_lock);
425         } else {
426                 spin_unlock(&inode->i_lock);
427                 spin_unlock(&owner->so_lock);
428                 if (new)
429                         nfs4_free_open_state(new);
430         }
431 out:
432         return state;
433 }
434
435 /*
436  * Beware! Caller must be holding exactly one
437  * reference to clp->cl_sem!
438  */
439 void nfs4_put_open_state(struct nfs4_state *state)
440 {
441         struct inode *inode = state->inode;
442         struct nfs4_state_owner *owner = state->owner;
443
444         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
445                 return;
446         spin_lock(&inode->i_lock);
447         list_del(&state->inode_states);
448         list_del(&state->open_states);
449         spin_unlock(&inode->i_lock);
450         spin_unlock(&owner->so_lock);
451         iput(inode);
452         nfs4_free_open_state(state);
453         nfs4_put_state_owner(owner);
454 }
455
456 /*
457  * Close the current file.
458  */
459 static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
460 {
461         struct nfs4_state_owner *owner = state->owner;
462         int call_close = 0;
463         int newstate;
464
465         atomic_inc(&owner->so_count);
466         /* Protect against nfs4_find_state() */
467         spin_lock(&owner->so_lock);
468         switch (mode & (FMODE_READ | FMODE_WRITE)) {
469                 case FMODE_READ:
470                         state->n_rdonly--;
471                         break;
472                 case FMODE_WRITE:
473                         state->n_wronly--;
474                         break;
475                 case FMODE_READ|FMODE_WRITE:
476                         state->n_rdwr--;
477         }
478         newstate = FMODE_READ|FMODE_WRITE;
479         if (state->n_rdwr == 0) {
480                 if (state->n_rdonly == 0) {
481                         newstate &= ~FMODE_READ;
482                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
483                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
484                 }
485                 if (state->n_wronly == 0) {
486                         newstate &= ~FMODE_WRITE;
487                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
488                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
489                 }
490                 if (newstate == 0)
491                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
492         }
493         nfs4_state_set_mode_locked(state, newstate);
494         spin_unlock(&owner->so_lock);
495
496         if (!call_close) {
497                 nfs4_put_open_state(state);
498                 nfs4_put_state_owner(owner);
499         } else
500                 nfs4_do_close(path, state, wait);
501 }
502
503 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
504 {
505         __nfs4_close(path, state, mode, 0);
506 }
507
508 void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
509 {
510         __nfs4_close(path, state, mode, 1);
511 }
512
513 /*
514  * Search the state->lock_states for an existing lock_owner
515  * that is compatible with current->files
516  */
517 static struct nfs4_lock_state *
518 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
519 {
520         struct nfs4_lock_state *pos;
521         list_for_each_entry(pos, &state->lock_states, ls_locks) {
522                 if (pos->ls_owner != fl_owner)
523                         continue;
524                 atomic_inc(&pos->ls_count);
525                 return pos;
526         }
527         return NULL;
528 }
529
530 /*
531  * Return a compatible lock_state. If no initialized lock_state structure
532  * exists, return an uninitialized one.
533  *
534  */
535 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
536 {
537         struct nfs4_lock_state *lsp;
538         struct nfs_client *clp = state->owner->so_client;
539
540         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
541         if (lsp == NULL)
542                 return NULL;
543         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
544         spin_lock_init(&lsp->ls_sequence.lock);
545         INIT_LIST_HEAD(&lsp->ls_sequence.list);
546         lsp->ls_seqid.sequence = &lsp->ls_sequence;
547         atomic_set(&lsp->ls_count, 1);
548         lsp->ls_owner = fl_owner;
549         spin_lock(&clp->cl_lock);
550         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
551         spin_unlock(&clp->cl_lock);
552         INIT_LIST_HEAD(&lsp->ls_locks);
553         return lsp;
554 }
555
556 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
557 {
558         struct nfs_client *clp = lsp->ls_state->owner->so_client;
559
560         spin_lock(&clp->cl_lock);
561         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
562         spin_unlock(&clp->cl_lock);
563         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
564         kfree(lsp);
565 }
566
567 /*
568  * Return a compatible lock_state. If no initialized lock_state structure
569  * exists, return an uninitialized one.
570  *
571  * The caller must be holding clp->cl_sem
572  */
573 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
574 {
575         struct nfs4_lock_state *lsp, *new = NULL;
576         
577         for(;;) {
578                 spin_lock(&state->state_lock);
579                 lsp = __nfs4_find_lock_state(state, owner);
580                 if (lsp != NULL)
581                         break;
582                 if (new != NULL) {
583                         new->ls_state = state;
584                         list_add(&new->ls_locks, &state->lock_states);
585                         set_bit(LK_STATE_IN_USE, &state->flags);
586                         lsp = new;
587                         new = NULL;
588                         break;
589                 }
590                 spin_unlock(&state->state_lock);
591                 new = nfs4_alloc_lock_state(state, owner);
592                 if (new == NULL)
593                         return NULL;
594         }
595         spin_unlock(&state->state_lock);
596         if (new != NULL)
597                 nfs4_free_lock_state(new);
598         return lsp;
599 }
600
601 /*
602  * Release reference to lock_state, and free it if we see that
603  * it is no longer in use
604  */
605 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
606 {
607         struct nfs4_state *state;
608
609         if (lsp == NULL)
610                 return;
611         state = lsp->ls_state;
612         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
613                 return;
614         list_del(&lsp->ls_locks);
615         if (list_empty(&state->lock_states))
616                 clear_bit(LK_STATE_IN_USE, &state->flags);
617         spin_unlock(&state->state_lock);
618         nfs4_free_lock_state(lsp);
619 }
620
621 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
622 {
623         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
624
625         dst->fl_u.nfs4_fl.owner = lsp;
626         atomic_inc(&lsp->ls_count);
627 }
628
629 static void nfs4_fl_release_lock(struct file_lock *fl)
630 {
631         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
632 }
633
634 static struct file_lock_operations nfs4_fl_lock_ops = {
635         .fl_copy_lock = nfs4_fl_copy_lock,
636         .fl_release_private = nfs4_fl_release_lock,
637 };
638
639 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
640 {
641         struct nfs4_lock_state *lsp;
642
643         if (fl->fl_ops != NULL)
644                 return 0;
645         lsp = nfs4_get_lock_state(state, fl->fl_owner);
646         if (lsp == NULL)
647                 return -ENOMEM;
648         fl->fl_u.nfs4_fl.owner = lsp;
649         fl->fl_ops = &nfs4_fl_lock_ops;
650         return 0;
651 }
652
653 /*
654  * Byte-range lock aware utility to initialize the stateid of read/write
655  * requests.
656  */
657 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
658 {
659         struct nfs4_lock_state *lsp;
660         int seq;
661
662         do {
663                 seq = read_seqbegin(&state->seqlock);
664                 memcpy(dst, &state->stateid, sizeof(*dst));
665         } while (read_seqretry(&state->seqlock, seq));
666         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
667                 return;
668
669         spin_lock(&state->state_lock);
670         lsp = __nfs4_find_lock_state(state, fl_owner);
671         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
672                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
673         spin_unlock(&state->state_lock);
674         nfs4_put_lock_state(lsp);
675 }
676
677 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
678 {
679         struct nfs_seqid *new;
680
681         new = kmalloc(sizeof(*new), GFP_KERNEL);
682         if (new != NULL) {
683                 new->sequence = counter;
684                 INIT_LIST_HEAD(&new->list);
685         }
686         return new;
687 }
688
689 void nfs_free_seqid(struct nfs_seqid *seqid)
690 {
691         if (!list_empty(&seqid->list)) {
692                 struct rpc_sequence *sequence = seqid->sequence->sequence;
693
694                 spin_lock(&sequence->lock);
695                 list_del(&seqid->list);
696                 spin_unlock(&sequence->lock);
697                 rpc_wake_up(&sequence->wait);
698         }
699         kfree(seqid);
700 }
701
702 /*
703  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
704  * failed with a seqid incrementing error -
705  * see comments nfs_fs.h:seqid_mutating_error()
706  */
707 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
708 {
709         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
710         switch (status) {
711                 case 0:
712                         break;
713                 case -NFS4ERR_BAD_SEQID:
714                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
715                                 return;
716                         printk(KERN_WARNING "NFS: v4 server returned a bad"
717                                         " sequence-id error on an"
718                                         " unconfirmed sequence %p!\n",
719                                         seqid->sequence);
720                 case -NFS4ERR_STALE_CLIENTID:
721                 case -NFS4ERR_STALE_STATEID:
722                 case -NFS4ERR_BAD_STATEID:
723                 case -NFS4ERR_BADXDR:
724                 case -NFS4ERR_RESOURCE:
725                 case -NFS4ERR_NOFILEHANDLE:
726                         /* Non-seqid mutating errors */
727                         return;
728         };
729         /*
730          * Note: no locking needed as we are guaranteed to be first
731          * on the sequence list
732          */
733         seqid->sequence->counter++;
734 }
735
736 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
737 {
738         if (status == -NFS4ERR_BAD_SEQID) {
739                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
740                                 struct nfs4_state_owner, so_seqid);
741                 nfs4_drop_state_owner(sp);
742         }
743         nfs_increment_seqid(status, seqid);
744 }
745
746 /*
747  * Increment the seqid if the LOCK/LOCKU succeeded, or
748  * failed with a seqid incrementing error -
749  * see comments nfs_fs.h:seqid_mutating_error()
750  */
751 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
752 {
753         nfs_increment_seqid(status, seqid);
754 }
755
756 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
757 {
758         struct rpc_sequence *sequence = seqid->sequence->sequence;
759         int status = 0;
760
761         spin_lock(&sequence->lock);
762         if (list_empty(&seqid->list))
763                 list_add_tail(&seqid->list, &sequence->list);
764         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
765                 goto unlock;
766         rpc_sleep_on(&sequence->wait, task, NULL);
767         status = -EAGAIN;
768 unlock:
769         spin_unlock(&sequence->lock);
770         return status;
771 }
772
773 static int reclaimer(void *);
774
775 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
776 {
777         smp_mb__before_clear_bit();
778         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
779         smp_mb__after_clear_bit();
780         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
781         rpc_wake_up(&clp->cl_rpcwaitq);
782 }
783
784 /*
785  * State recovery routine
786  */
787 static void nfs4_recover_state(struct nfs_client *clp)
788 {
789         struct task_struct *task;
790
791         __module_get(THIS_MODULE);
792         atomic_inc(&clp->cl_count);
793         task = kthread_run(reclaimer, clp, "%s-reclaim",
794                                 rpc_peeraddr2str(clp->cl_rpcclient,
795                                                         RPC_DISPLAY_ADDR));
796         if (!IS_ERR(task))
797                 return;
798         nfs4_clear_recover_bit(clp);
799         nfs_put_client(clp);
800         module_put(THIS_MODULE);
801 }
802
803 /*
804  * Schedule a state recovery attempt
805  */
806 void nfs4_schedule_state_recovery(struct nfs_client *clp)
807 {
808         if (!clp)
809                 return;
810         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
811                 nfs4_recover_state(clp);
812 }
813
814 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
815 {
816         struct inode *inode = state->inode;
817         struct file_lock *fl;
818         int status = 0;
819
820         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
821                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
822                         continue;
823                 if (nfs_file_open_context(fl->fl_file)->state != state)
824                         continue;
825                 status = ops->recover_lock(state, fl);
826                 if (status >= 0)
827                         continue;
828                 switch (status) {
829                         default:
830                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
831                                                 __func__, status);
832                         case -NFS4ERR_EXPIRED:
833                         case -NFS4ERR_NO_GRACE:
834                         case -NFS4ERR_RECLAIM_BAD:
835                         case -NFS4ERR_RECLAIM_CONFLICT:
836                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
837                                 break;
838                         case -NFS4ERR_STALE_CLIENTID:
839                                 goto out_err;
840                 }
841         }
842         return 0;
843 out_err:
844         return status;
845 }
846
847 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
848 {
849         struct nfs4_state *state;
850         struct nfs4_lock_state *lock;
851         int status = 0;
852
853         /* Note: we rely on the sp->so_states list being ordered 
854          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
855          * states first.
856          * This is needed to ensure that the server won't give us any
857          * read delegations that we have to return if, say, we are
858          * recovering after a network partition or a reboot from a
859          * server that doesn't support a grace period.
860          */
861         list_for_each_entry(state, &sp->so_states, open_states) {
862                 if (state->state == 0)
863                         continue;
864                 status = ops->recover_open(sp, state);
865                 if (status >= 0) {
866                         status = nfs4_reclaim_locks(state, ops);
867                         if (status >= 0) {
868                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
869                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
870                                                 printk("%s: Lock reclaim failed!\n",
871                                                         __func__);
872                                 }
873                                 continue;
874                         }
875                 }
876                 switch (status) {
877                         default:
878                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
879                                                 __func__, status);
880                         case -ENOENT:
881                         case -NFS4ERR_RECLAIM_BAD:
882                         case -NFS4ERR_RECLAIM_CONFLICT:
883                                 /*
884                                  * Open state on this file cannot be recovered
885                                  * All we can do is revert to using the zero stateid.
886                                  */
887                                 memset(state->stateid.data, 0,
888                                         sizeof(state->stateid.data));
889                                 /* Mark the file as being 'closed' */
890                                 state->state = 0;
891                                 break;
892                         case -NFS4ERR_EXPIRED:
893                         case -NFS4ERR_NO_GRACE:
894                         case -NFS4ERR_STALE_CLIENTID:
895                                 goto out_err;
896                 }
897         }
898         return 0;
899 out_err:
900         return status;
901 }
902
903 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
904 {
905         struct nfs4_state_owner *sp;
906         struct rb_node *pos;
907         struct nfs4_state *state;
908         struct nfs4_lock_state *lock;
909
910         /* Reset all sequence ids to zero */
911         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
912                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
913                 sp->so_seqid.counter = 0;
914                 sp->so_seqid.flags = 0;
915                 spin_lock(&sp->so_lock);
916                 list_for_each_entry(state, &sp->so_states, open_states) {
917                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
918                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
919                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
920                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
921                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
922                                 lock->ls_seqid.counter = 0;
923                                 lock->ls_seqid.flags = 0;
924                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
925                         }
926                 }
927                 spin_unlock(&sp->so_lock);
928         }
929 }
930
931 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
932 {
933         struct rb_node *pos;
934         int status = 0;
935
936         /* Note: list is protected by exclusive lock on cl->cl_sem */
937         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
938                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
939                 status = nfs4_reclaim_open_state(sp, ops);
940                 if (status < 0)
941                         break;
942         }
943         return status;
944 }
945
946 static int nfs4_check_lease(struct nfs_client *clp)
947 {
948         struct rpc_cred *cred;
949         int status = -NFS4ERR_EXPIRED;
950
951         /* Are there any open files on this volume? */
952         cred = nfs4_get_renew_cred(clp);
953         if (cred != NULL) {
954                 /* Yes there are: try to renew the old lease */
955                 status = nfs4_proc_renew(clp, cred);
956                 put_rpccred(cred);
957                 return status;
958         }
959
960         /* "reboot" to ensure we clear all state on the server */
961         clp->cl_boot_time = CURRENT_TIME;
962         return status;
963 }
964
965 static int nfs4_reclaim_lease(struct nfs_client *clp)
966 {
967         struct rpc_cred *cred;
968         int status = -ENOENT;
969
970         cred = nfs4_get_setclientid_cred(clp);
971         if (cred != NULL) {
972                 status = nfs4_init_client(clp, cred);
973                 put_rpccred(cred);
974                 /* Handle case where the user hasn't set up machine creds */
975                 if (status == -EACCES && cred == clp->cl_machine_cred) {
976                         nfs4_clear_machine_cred(clp);
977                         status = -EAGAIN;
978                 }
979         }
980         return status;
981 }
982
983 static int reclaimer(void *ptr)
984 {
985         struct nfs_client *clp = ptr;
986         const struct nfs4_state_recovery_ops *ops;
987         int status = 0;
988
989         allow_signal(SIGKILL);
990
991         /* Ensure exclusive access to NFSv4 state */
992         down_write(&clp->cl_sem);
993         while (!list_empty(&clp->cl_superblocks)) {
994                 ops = &nfs4_network_partition_recovery_ops;
995                 status = nfs4_check_lease(clp);
996                 switch (status) {
997                         case 0:
998                         case -NFS4ERR_CB_PATH_DOWN:
999                                 goto out;
1000                         case -NFS4ERR_STALE_CLIENTID:
1001                         case -NFS4ERR_LEASE_MOVED:
1002                                 ops = &nfs4_reboot_recovery_ops;
1003                 }
1004
1005                 /* We're going to have to re-establish a clientid */
1006                 nfs4_state_mark_reclaim(clp);
1007
1008                 status = nfs4_reclaim_lease(clp);
1009                 if (status) {
1010                         if (status == -EAGAIN)
1011                                 continue;
1012                         goto out_error;
1013                 }
1014
1015                 /* Mark all delegations for reclaim */
1016                 nfs_delegation_mark_reclaim(clp);
1017                 /* Note: list is protected by exclusive lock on cl->cl_sem */
1018                 status = nfs4_do_reclaim(clp, ops);
1019                 if (status < 0) {
1020                         if (status == -NFS4ERR_NO_GRACE) {
1021                                 ops = &nfs4_network_partition_recovery_ops;
1022                                 status = nfs4_do_reclaim(clp, ops);
1023                         }
1024                         if (status == -NFS4ERR_STALE_CLIENTID)
1025                                 continue;
1026                         if (status == -NFS4ERR_EXPIRED)
1027                                 continue;
1028                 }
1029                 nfs_delegation_reap_unclaimed(clp);
1030                 break;
1031         }
1032 out:
1033         up_write(&clp->cl_sem);
1034         if (status == -NFS4ERR_CB_PATH_DOWN)
1035                 nfs_handle_cb_pathdown(clp);
1036         nfs4_clear_recover_bit(clp);
1037         nfs_put_client(clp);
1038         module_put_and_exit(0);
1039         return 0;
1040 out_error:
1041         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %s"
1042                         " with error %d\n", clp->cl_hostname, -status);
1043         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1044         goto out;
1045 }
1046
1047 /*
1048  * Local variables:
1049  *  c-basic-offset: 8
1050  * End:
1051  */