Merge branch 'merge'
[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/slab.h>
42 #include <linux/smp_lock.h>
43 #include <linux/nfs_fs.h>
44 #include <linux/nfs_idmap.h>
45 #include <linux/kthread.h>
46 #include <linux/module.h>
47 #include <linux/workqueue.h>
48 #include <linux/bitops.h>
49
50 #include "nfs4_fs.h"
51 #include "callback.h"
52 #include "delegation.h"
53
54 #define OPENOWNER_POOL_SIZE     8
55
56 const nfs4_stateid zero_stateid;
57
58 static DEFINE_SPINLOCK(state_spinlock);
59 static LIST_HEAD(nfs4_clientid_list);
60
61 void
62 init_nfsv4_state(struct nfs_server *server)
63 {
64         server->nfs4_state = NULL;
65         INIT_LIST_HEAD(&server->nfs4_siblings);
66 }
67
68 void
69 destroy_nfsv4_state(struct nfs_server *server)
70 {
71         kfree(server->mnt_path);
72         server->mnt_path = NULL;
73         if (server->nfs4_state) {
74                 nfs4_put_client(server->nfs4_state);
75                 server->nfs4_state = NULL;
76         }
77 }
78
79 /*
80  * nfs4_get_client(): returns an empty client structure
81  * nfs4_put_client(): drops reference to client structure
82  *
83  * Since these are allocated/deallocated very rarely, we don't
84  * bother putting them in a slab cache...
85  */
86 static struct nfs4_client *
87 nfs4_alloc_client(struct in_addr *addr)
88 {
89         struct nfs4_client *clp;
90
91         if (nfs_callback_up() < 0)
92                 return NULL;
93         if ((clp = kzalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
94                 nfs_callback_down();
95                 return NULL;
96         }
97         memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
98         init_rwsem(&clp->cl_sem);
99         INIT_LIST_HEAD(&clp->cl_delegations);
100         INIT_LIST_HEAD(&clp->cl_state_owners);
101         INIT_LIST_HEAD(&clp->cl_unused);
102         spin_lock_init(&clp->cl_lock);
103         atomic_set(&clp->cl_count, 1);
104         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
105         INIT_LIST_HEAD(&clp->cl_superblocks);
106         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
107         clp->cl_rpcclient = ERR_PTR(-EINVAL);
108         clp->cl_boot_time = CURRENT_TIME;
109         clp->cl_state = 1 << NFS4CLNT_LEASE_EXPIRED;
110         return clp;
111 }
112
113 static void
114 nfs4_free_client(struct nfs4_client *clp)
115 {
116         struct nfs4_state_owner *sp;
117
118         while (!list_empty(&clp->cl_unused)) {
119                 sp = list_entry(clp->cl_unused.next,
120                                 struct nfs4_state_owner,
121                                 so_list);
122                 list_del(&sp->so_list);
123                 kfree(sp);
124         }
125         BUG_ON(!list_empty(&clp->cl_state_owners));
126         nfs_idmap_delete(clp);
127         if (!IS_ERR(clp->cl_rpcclient))
128                 rpc_shutdown_client(clp->cl_rpcclient);
129         kfree(clp);
130         nfs_callback_down();
131 }
132
133 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
134 {
135         struct nfs4_client *clp;
136         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
137                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
138                         atomic_inc(&clp->cl_count);
139                         return clp;
140                 }
141         }
142         return NULL;
143 }
144
145 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
146 {
147         struct nfs4_client *clp;
148         spin_lock(&state_spinlock);
149         clp = __nfs4_find_client(addr);
150         spin_unlock(&state_spinlock);
151         return clp;
152 }
153
154 struct nfs4_client *
155 nfs4_get_client(struct in_addr *addr)
156 {
157         struct nfs4_client *clp, *new = NULL;
158
159         spin_lock(&state_spinlock);
160         for (;;) {
161                 clp = __nfs4_find_client(addr);
162                 if (clp != NULL)
163                         break;
164                 clp = new;
165                 if (clp != NULL) {
166                         list_add(&clp->cl_servers, &nfs4_clientid_list);
167                         new = NULL;
168                         break;
169                 }
170                 spin_unlock(&state_spinlock);
171                 new = nfs4_alloc_client(addr);
172                 spin_lock(&state_spinlock);
173                 if (new == NULL)
174                         break;
175         }
176         spin_unlock(&state_spinlock);
177         if (new)
178                 nfs4_free_client(new);
179         return clp;
180 }
181
182 void
183 nfs4_put_client(struct nfs4_client *clp)
184 {
185         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
186                 return;
187         list_del(&clp->cl_servers);
188         spin_unlock(&state_spinlock);
189         BUG_ON(!list_empty(&clp->cl_superblocks));
190         rpc_wake_up(&clp->cl_rpcwaitq);
191         nfs4_kill_renewd(clp);
192         nfs4_free_client(clp);
193 }
194
195 static int nfs4_init_client(struct nfs4_client *clp, struct rpc_cred *cred)
196 {
197         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
198                         nfs_callback_tcpport, cred);
199         if (status == 0)
200                 status = nfs4_proc_setclientid_confirm(clp, cred);
201         if (status == 0)
202                 nfs4_schedule_state_renewal(clp);
203         return status;
204 }
205
206 u32
207 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
208 {
209         return clp->cl_lockowner_id ++;
210 }
211
212 static struct nfs4_state_owner *
213 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
214 {
215         struct nfs4_state_owner *sp = NULL;
216
217         if (!list_empty(&clp->cl_unused)) {
218                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
219                 atomic_inc(&sp->so_count);
220                 sp->so_cred = cred;
221                 list_move(&sp->so_list, &clp->cl_state_owners);
222                 clp->cl_nunused--;
223         }
224         return sp;
225 }
226
227 struct rpc_cred *nfs4_get_renew_cred(struct nfs4_client *clp)
228 {
229         struct nfs4_state_owner *sp;
230         struct rpc_cred *cred = NULL;
231
232         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
233                 if (list_empty(&sp->so_states))
234                         continue;
235                 cred = get_rpccred(sp->so_cred);
236                 break;
237         }
238         return cred;
239 }
240
241 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs4_client *clp)
242 {
243         struct nfs4_state_owner *sp;
244
245         if (!list_empty(&clp->cl_state_owners)) {
246                 sp = list_entry(clp->cl_state_owners.next,
247                                 struct nfs4_state_owner, so_list);
248                 return get_rpccred(sp->so_cred);
249         }
250         return NULL;
251 }
252
253 static struct nfs4_state_owner *
254 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
255 {
256         struct nfs4_state_owner *sp, *res = NULL;
257
258         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
259                 if (sp->so_cred != cred)
260                         continue;
261                 atomic_inc(&sp->so_count);
262                 /* Move to the head of the list */
263                 list_move(&sp->so_list, &clp->cl_state_owners);
264                 res = sp;
265                 break;
266         }
267         return res;
268 }
269
270 /*
271  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
272  * create a new state_owner.
273  *
274  */
275 static struct nfs4_state_owner *
276 nfs4_alloc_state_owner(void)
277 {
278         struct nfs4_state_owner *sp;
279
280         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
281         if (!sp)
282                 return NULL;
283         spin_lock_init(&sp->so_lock);
284         INIT_LIST_HEAD(&sp->so_states);
285         INIT_LIST_HEAD(&sp->so_delegations);
286         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
287         sp->so_seqid.sequence = &sp->so_sequence;
288         spin_lock_init(&sp->so_sequence.lock);
289         INIT_LIST_HEAD(&sp->so_sequence.list);
290         atomic_set(&sp->so_count, 1);
291         return sp;
292 }
293
294 void
295 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
296 {
297         struct nfs4_client *clp = sp->so_client;
298         spin_lock(&clp->cl_lock);
299         list_del_init(&sp->so_list);
300         spin_unlock(&clp->cl_lock);
301 }
302
303 /*
304  * Note: must be called with clp->cl_sem held in order to prevent races
305  *       with reboot recovery!
306  */
307 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
308 {
309         struct nfs4_client *clp = server->nfs4_state;
310         struct nfs4_state_owner *sp, *new;
311
312         get_rpccred(cred);
313         new = nfs4_alloc_state_owner();
314         spin_lock(&clp->cl_lock);
315         sp = nfs4_find_state_owner(clp, cred);
316         if (sp == NULL)
317                 sp = nfs4_client_grab_unused(clp, cred);
318         if (sp == NULL && new != NULL) {
319                 list_add(&new->so_list, &clp->cl_state_owners);
320                 new->so_client = clp;
321                 new->so_id = nfs4_alloc_lockowner_id(clp);
322                 new->so_cred = cred;
323                 sp = new;
324                 new = NULL;
325         }
326         spin_unlock(&clp->cl_lock);
327         kfree(new);
328         if (sp != NULL)
329                 return sp;
330         put_rpccred(cred);
331         return NULL;
332 }
333
334 /*
335  * Must be called with clp->cl_sem held in order to avoid races
336  * with state recovery...
337  */
338 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
339 {
340         struct nfs4_client *clp = sp->so_client;
341         struct rpc_cred *cred = sp->so_cred;
342
343         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
344                 return;
345         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
346                 goto out_free;
347         if (list_empty(&sp->so_list))
348                 goto out_free;
349         list_move(&sp->so_list, &clp->cl_unused);
350         clp->cl_nunused++;
351         spin_unlock(&clp->cl_lock);
352         put_rpccred(cred);
353         cred = NULL;
354         return;
355 out_free:
356         list_del(&sp->so_list);
357         spin_unlock(&clp->cl_lock);
358         put_rpccred(cred);
359         kfree(sp);
360 }
361
362 static struct nfs4_state *
363 nfs4_alloc_open_state(void)
364 {
365         struct nfs4_state *state;
366
367         state = kzalloc(sizeof(*state), GFP_KERNEL);
368         if (!state)
369                 return NULL;
370         atomic_set(&state->count, 1);
371         INIT_LIST_HEAD(&state->lock_states);
372         spin_lock_init(&state->state_lock);
373         return state;
374 }
375
376 void
377 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
378 {
379         if (state->state == mode)
380                 return;
381         /* NB! List reordering - see the reclaim code for why.  */
382         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
383                 if (mode & FMODE_WRITE)
384                         list_move(&state->open_states, &state->owner->so_states);
385                 else
386                         list_move_tail(&state->open_states, &state->owner->so_states);
387         }
388         if (mode == 0)
389                 list_del_init(&state->inode_states);
390         state->state = mode;
391 }
392
393 static struct nfs4_state *
394 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
395 {
396         struct nfs_inode *nfsi = NFS_I(inode);
397         struct nfs4_state *state;
398
399         list_for_each_entry(state, &nfsi->open_states, inode_states) {
400                 /* Is this in the process of being freed? */
401                 if (state->state == 0)
402                         continue;
403                 if (state->owner == owner) {
404                         atomic_inc(&state->count);
405                         return state;
406                 }
407         }
408         return NULL;
409 }
410
411 static void
412 nfs4_free_open_state(struct nfs4_state *state)
413 {
414         kfree(state);
415 }
416
417 struct nfs4_state *
418 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
419 {
420         struct nfs4_state *state, *new;
421         struct nfs_inode *nfsi = NFS_I(inode);
422
423         spin_lock(&inode->i_lock);
424         state = __nfs4_find_state_byowner(inode, owner);
425         spin_unlock(&inode->i_lock);
426         if (state)
427                 goto out;
428         new = nfs4_alloc_open_state();
429         spin_lock(&owner->so_lock);
430         spin_lock(&inode->i_lock);
431         state = __nfs4_find_state_byowner(inode, owner);
432         if (state == NULL && new != NULL) {
433                 state = new;
434                 state->owner = owner;
435                 atomic_inc(&owner->so_count);
436                 list_add(&state->inode_states, &nfsi->open_states);
437                 state->inode = igrab(inode);
438                 spin_unlock(&inode->i_lock);
439                 /* Note: The reclaim code dictates that we add stateless
440                  * and read-only stateids to the end of the list */
441                 list_add_tail(&state->open_states, &owner->so_states);
442                 spin_unlock(&owner->so_lock);
443         } else {
444                 spin_unlock(&inode->i_lock);
445                 spin_unlock(&owner->so_lock);
446                 if (new)
447                         nfs4_free_open_state(new);
448         }
449 out:
450         return state;
451 }
452
453 /*
454  * Beware! Caller must be holding exactly one
455  * reference to clp->cl_sem!
456  */
457 void nfs4_put_open_state(struct nfs4_state *state)
458 {
459         struct inode *inode = state->inode;
460         struct nfs4_state_owner *owner = state->owner;
461
462         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
463                 return;
464         spin_lock(&inode->i_lock);
465         if (!list_empty(&state->inode_states))
466                 list_del(&state->inode_states);
467         list_del(&state->open_states);
468         spin_unlock(&inode->i_lock);
469         spin_unlock(&owner->so_lock);
470         iput(inode);
471         nfs4_free_open_state(state);
472         nfs4_put_state_owner(owner);
473 }
474
475 /*
476  * Close the current file.
477  */
478 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
479 {
480         struct inode *inode = state->inode;
481         struct nfs4_state_owner *owner = state->owner;
482         int oldstate, newstate = 0;
483
484         atomic_inc(&owner->so_count);
485         /* Protect against nfs4_find_state() */
486         spin_lock(&owner->so_lock);
487         spin_lock(&inode->i_lock);
488         switch (mode & (FMODE_READ | FMODE_WRITE)) {
489                 case FMODE_READ:
490                         state->n_rdonly--;
491                         break;
492                 case FMODE_WRITE:
493                         state->n_wronly--;
494                         break;
495                 case FMODE_READ|FMODE_WRITE:
496                         state->n_rdwr--;
497         }
498         oldstate = newstate = state->state;
499         if (state->n_rdwr == 0) {
500                 if (state->n_rdonly == 0)
501                         newstate &= ~FMODE_READ;
502                 if (state->n_wronly == 0)
503                         newstate &= ~FMODE_WRITE;
504         }
505         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
506                 nfs4_state_set_mode_locked(state, newstate);
507                 oldstate = newstate;
508         }
509         spin_unlock(&inode->i_lock);
510         spin_unlock(&owner->so_lock);
511
512         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
513                 return;
514         nfs4_put_open_state(state);
515         nfs4_put_state_owner(owner);
516 }
517
518 /*
519  * Search the state->lock_states for an existing lock_owner
520  * that is compatible with current->files
521  */
522 static struct nfs4_lock_state *
523 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
524 {
525         struct nfs4_lock_state *pos;
526         list_for_each_entry(pos, &state->lock_states, ls_locks) {
527                 if (pos->ls_owner != fl_owner)
528                         continue;
529                 atomic_inc(&pos->ls_count);
530                 return pos;
531         }
532         return NULL;
533 }
534
535 /*
536  * Return a compatible lock_state. If no initialized lock_state structure
537  * exists, return an uninitialized one.
538  *
539  */
540 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
541 {
542         struct nfs4_lock_state *lsp;
543         struct nfs4_client *clp = state->owner->so_client;
544
545         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
546         if (lsp == NULL)
547                 return NULL;
548         lsp->ls_seqid.sequence = &state->owner->so_sequence;
549         atomic_set(&lsp->ls_count, 1);
550         lsp->ls_owner = fl_owner;
551         spin_lock(&clp->cl_lock);
552         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
553         spin_unlock(&clp->cl_lock);
554         INIT_LIST_HEAD(&lsp->ls_locks);
555         return lsp;
556 }
557
558 /*
559  * Return a compatible lock_state. If no initialized lock_state structure
560  * exists, return an uninitialized one.
561  *
562  * The caller must be holding clp->cl_sem
563  */
564 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
565 {
566         struct nfs4_lock_state *lsp, *new = NULL;
567         
568         for(;;) {
569                 spin_lock(&state->state_lock);
570                 lsp = __nfs4_find_lock_state(state, owner);
571                 if (lsp != NULL)
572                         break;
573                 if (new != NULL) {
574                         new->ls_state = state;
575                         list_add(&new->ls_locks, &state->lock_states);
576                         set_bit(LK_STATE_IN_USE, &state->flags);
577                         lsp = new;
578                         new = NULL;
579                         break;
580                 }
581                 spin_unlock(&state->state_lock);
582                 new = nfs4_alloc_lock_state(state, owner);
583                 if (new == NULL)
584                         return NULL;
585         }
586         spin_unlock(&state->state_lock);
587         kfree(new);
588         return lsp;
589 }
590
591 /*
592  * Release reference to lock_state, and free it if we see that
593  * it is no longer in use
594  */
595 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
596 {
597         struct nfs4_state *state;
598
599         if (lsp == NULL)
600                 return;
601         state = lsp->ls_state;
602         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
603                 return;
604         list_del(&lsp->ls_locks);
605         if (list_empty(&state->lock_states))
606                 clear_bit(LK_STATE_IN_USE, &state->flags);
607         spin_unlock(&state->state_lock);
608         kfree(lsp);
609 }
610
611 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
612 {
613         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
614
615         dst->fl_u.nfs4_fl.owner = lsp;
616         atomic_inc(&lsp->ls_count);
617 }
618
619 static void nfs4_fl_release_lock(struct file_lock *fl)
620 {
621         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
622 }
623
624 static struct file_lock_operations nfs4_fl_lock_ops = {
625         .fl_copy_lock = nfs4_fl_copy_lock,
626         .fl_release_private = nfs4_fl_release_lock,
627 };
628
629 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
630 {
631         struct nfs4_lock_state *lsp;
632
633         if (fl->fl_ops != NULL)
634                 return 0;
635         lsp = nfs4_get_lock_state(state, fl->fl_owner);
636         if (lsp == NULL)
637                 return -ENOMEM;
638         fl->fl_u.nfs4_fl.owner = lsp;
639         fl->fl_ops = &nfs4_fl_lock_ops;
640         return 0;
641 }
642
643 /*
644  * Byte-range lock aware utility to initialize the stateid of read/write
645  * requests.
646  */
647 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
648 {
649         struct nfs4_lock_state *lsp;
650
651         memcpy(dst, &state->stateid, sizeof(*dst));
652         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
653                 return;
654
655         spin_lock(&state->state_lock);
656         lsp = __nfs4_find_lock_state(state, fl_owner);
657         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
658                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
659         spin_unlock(&state->state_lock);
660         nfs4_put_lock_state(lsp);
661 }
662
663 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
664 {
665         struct rpc_sequence *sequence = counter->sequence;
666         struct nfs_seqid *new;
667
668         new = kmalloc(sizeof(*new), GFP_KERNEL);
669         if (new != NULL) {
670                 new->sequence = counter;
671                 spin_lock(&sequence->lock);
672                 list_add_tail(&new->list, &sequence->list);
673                 spin_unlock(&sequence->lock);
674         }
675         return new;
676 }
677
678 void nfs_free_seqid(struct nfs_seqid *seqid)
679 {
680         struct rpc_sequence *sequence = seqid->sequence->sequence;
681
682         spin_lock(&sequence->lock);
683         list_del(&seqid->list);
684         spin_unlock(&sequence->lock);
685         rpc_wake_up(&sequence->wait);
686         kfree(seqid);
687 }
688
689 /*
690  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
691  * failed with a seqid incrementing error -
692  * see comments nfs_fs.h:seqid_mutating_error()
693  */
694 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
695 {
696         switch (status) {
697                 case 0:
698                         break;
699                 case -NFS4ERR_BAD_SEQID:
700                 case -NFS4ERR_STALE_CLIENTID:
701                 case -NFS4ERR_STALE_STATEID:
702                 case -NFS4ERR_BAD_STATEID:
703                 case -NFS4ERR_BADXDR:
704                 case -NFS4ERR_RESOURCE:
705                 case -NFS4ERR_NOFILEHANDLE:
706                         /* Non-seqid mutating errors */
707                         return;
708         };
709         /*
710          * Note: no locking needed as we are guaranteed to be first
711          * on the sequence list
712          */
713         seqid->sequence->counter++;
714 }
715
716 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
717 {
718         if (status == -NFS4ERR_BAD_SEQID) {
719                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
720                                 struct nfs4_state_owner, so_seqid);
721                 nfs4_drop_state_owner(sp);
722         }
723         return nfs_increment_seqid(status, seqid);
724 }
725
726 /*
727  * Increment the seqid if the LOCK/LOCKU succeeded, or
728  * failed with a seqid incrementing error -
729  * see comments nfs_fs.h:seqid_mutating_error()
730  */
731 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
732 {
733         return nfs_increment_seqid(status, seqid);
734 }
735
736 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
737 {
738         struct rpc_sequence *sequence = seqid->sequence->sequence;
739         int status = 0;
740
741         if (sequence->list.next == &seqid->list)
742                 goto out;
743         spin_lock(&sequence->lock);
744         if (sequence->list.next != &seqid->list) {
745                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
746                 status = -EAGAIN;
747         }
748         spin_unlock(&sequence->lock);
749 out:
750         return status;
751 }
752
753 static int reclaimer(void *);
754
755 static inline void nfs4_clear_recover_bit(struct nfs4_client *clp)
756 {
757         smp_mb__before_clear_bit();
758         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
759         smp_mb__after_clear_bit();
760         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
761         rpc_wake_up(&clp->cl_rpcwaitq);
762 }
763
764 /*
765  * State recovery routine
766  */
767 static void nfs4_recover_state(struct nfs4_client *clp)
768 {
769         struct task_struct *task;
770
771         __module_get(THIS_MODULE);
772         atomic_inc(&clp->cl_count);
773         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
774                         NIPQUAD(clp->cl_addr));
775         if (!IS_ERR(task))
776                 return;
777         nfs4_clear_recover_bit(clp);
778         nfs4_put_client(clp);
779         module_put(THIS_MODULE);
780 }
781
782 /*
783  * Schedule a state recovery attempt
784  */
785 void nfs4_schedule_state_recovery(struct nfs4_client *clp)
786 {
787         if (!clp)
788                 return;
789         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
790                 nfs4_recover_state(clp);
791 }
792
793 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
794 {
795         struct inode *inode = state->inode;
796         struct file_lock *fl;
797         int status = 0;
798
799         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
800                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
801                         continue;
802                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
803                         continue;
804                 status = ops->recover_lock(state, fl);
805                 if (status >= 0)
806                         continue;
807                 switch (status) {
808                         default:
809                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
810                                                 __FUNCTION__, status);
811                         case -NFS4ERR_EXPIRED:
812                         case -NFS4ERR_NO_GRACE:
813                         case -NFS4ERR_RECLAIM_BAD:
814                         case -NFS4ERR_RECLAIM_CONFLICT:
815                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
816                                 break;
817                         case -NFS4ERR_STALE_CLIENTID:
818                                 goto out_err;
819                 }
820         }
821         return 0;
822 out_err:
823         return status;
824 }
825
826 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
827 {
828         struct nfs4_state *state;
829         struct nfs4_lock_state *lock;
830         int status = 0;
831
832         /* Note: we rely on the sp->so_states list being ordered 
833          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
834          * states first.
835          * This is needed to ensure that the server won't give us any
836          * read delegations that we have to return if, say, we are
837          * recovering after a network partition or a reboot from a
838          * server that doesn't support a grace period.
839          */
840         list_for_each_entry(state, &sp->so_states, open_states) {
841                 if (state->state == 0)
842                         continue;
843                 status = ops->recover_open(sp, state);
844                 if (status >= 0) {
845                         status = nfs4_reclaim_locks(ops, state);
846                         if (status < 0)
847                                 goto out_err;
848                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
849                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
850                                         printk("%s: Lock reclaim failed!\n",
851                                                         __FUNCTION__);
852                         }
853                         continue;
854                 }
855                 switch (status) {
856                         default:
857                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
858                                                 __FUNCTION__, status);
859                         case -ENOENT:
860                         case -NFS4ERR_RECLAIM_BAD:
861                         case -NFS4ERR_RECLAIM_CONFLICT:
862                                 /*
863                                  * Open state on this file cannot be recovered
864                                  * All we can do is revert to using the zero stateid.
865                                  */
866                                 memset(state->stateid.data, 0,
867                                         sizeof(state->stateid.data));
868                                 /* Mark the file as being 'closed' */
869                                 state->state = 0;
870                                 break;
871                         case -NFS4ERR_EXPIRED:
872                         case -NFS4ERR_NO_GRACE:
873                         case -NFS4ERR_STALE_CLIENTID:
874                                 goto out_err;
875                 }
876         }
877         return 0;
878 out_err:
879         return status;
880 }
881
882 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
883 {
884         struct nfs4_state_owner *sp;
885         struct nfs4_state *state;
886         struct nfs4_lock_state *lock;
887
888         /* Reset all sequence ids to zero */
889         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
890                 sp->so_seqid.counter = 0;
891                 sp->so_seqid.flags = 0;
892                 spin_lock(&sp->so_lock);
893                 list_for_each_entry(state, &sp->so_states, open_states) {
894                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
895                                 lock->ls_seqid.counter = 0;
896                                 lock->ls_seqid.flags = 0;
897                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
898                         }
899                 }
900                 spin_unlock(&sp->so_lock);
901         }
902 }
903
904 static int reclaimer(void *ptr)
905 {
906         struct nfs4_client *clp = ptr;
907         struct nfs4_state_owner *sp;
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         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
954                 status = nfs4_reclaim_open_state(ops, sp);
955                 if (status < 0) {
956                         if (status == -NFS4ERR_NO_GRACE) {
957                                 ops = &nfs4_network_partition_recovery_ops;
958                                 status = nfs4_reclaim_open_state(ops, sp);
959                         }
960                         if (status == -NFS4ERR_STALE_CLIENTID)
961                                 goto restart_loop;
962                         if (status == -NFS4ERR_EXPIRED)
963                                 goto restart_loop;
964                 }
965         }
966         nfs_delegation_reap_unclaimed(clp);
967 out:
968         up_write(&clp->cl_sem);
969         unlock_kernel();
970         if (status == -NFS4ERR_CB_PATH_DOWN)
971                 nfs_handle_cb_pathdown(clp);
972         nfs4_clear_recover_bit(clp);
973         nfs4_put_client(clp);
974         module_put_and_exit(0);
975         return 0;
976 out_error:
977         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
978                                 NIPQUAD(clp->cl_addr.s_addr), -status);
979         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
980         goto out;
981 }
982
983 /*
984  * Local variables:
985  *  c-basic-offset: 8
986  * End:
987  */