Merge branch 'master'
[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/config.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/workqueue.h>
47 #include <linux/bitops.h>
48
49 #include "nfs4_fs.h"
50 #include "callback.h"
51 #include "delegation.h"
52
53 #define OPENOWNER_POOL_SIZE     8
54
55 const nfs4_stateid zero_stateid;
56
57 static DEFINE_SPINLOCK(state_spinlock);
58 static LIST_HEAD(nfs4_clientid_list);
59
60 static void nfs4_recover_state(void *);
61
62 void
63 init_nfsv4_state(struct nfs_server *server)
64 {
65         server->nfs4_state = NULL;
66         INIT_LIST_HEAD(&server->nfs4_siblings);
67 }
68
69 void
70 destroy_nfsv4_state(struct nfs_server *server)
71 {
72         kfree(server->mnt_path);
73         server->mnt_path = NULL;
74         if (server->nfs4_state) {
75                 nfs4_put_client(server->nfs4_state);
76                 server->nfs4_state = NULL;
77         }
78 }
79
80 /*
81  * nfs4_get_client(): returns an empty client structure
82  * nfs4_put_client(): drops reference to client structure
83  *
84  * Since these are allocated/deallocated very rarely, we don't
85  * bother putting them in a slab cache...
86  */
87 static struct nfs4_client *
88 nfs4_alloc_client(struct in_addr *addr)
89 {
90         struct nfs4_client *clp;
91
92         if (nfs_callback_up() < 0)
93                 return NULL;
94         if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
95                 nfs_callback_down();
96                 return NULL;
97         }
98         memset(clp, 0, sizeof(*clp));
99         memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
100         init_rwsem(&clp->cl_sem);
101         INIT_LIST_HEAD(&clp->cl_delegations);
102         INIT_LIST_HEAD(&clp->cl_state_owners);
103         INIT_LIST_HEAD(&clp->cl_unused);
104         spin_lock_init(&clp->cl_lock);
105         atomic_set(&clp->cl_count, 1);
106         INIT_WORK(&clp->cl_recoverd, nfs4_recover_state, clp);
107         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
108         INIT_LIST_HEAD(&clp->cl_superblocks);
109         init_waitqueue_head(&clp->cl_waitq);
110         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
111         clp->cl_rpcclient = ERR_PTR(-EINVAL);
112         clp->cl_boot_time = CURRENT_TIME;
113         clp->cl_state = 1 << NFS4CLNT_OK;
114         return clp;
115 }
116
117 static void
118 nfs4_free_client(struct nfs4_client *clp)
119 {
120         struct nfs4_state_owner *sp;
121
122         while (!list_empty(&clp->cl_unused)) {
123                 sp = list_entry(clp->cl_unused.next,
124                                 struct nfs4_state_owner,
125                                 so_list);
126                 list_del(&sp->so_list);
127                 kfree(sp);
128         }
129         BUG_ON(!list_empty(&clp->cl_state_owners));
130         if (clp->cl_cred)
131                 put_rpccred(clp->cl_cred);
132         nfs_idmap_delete(clp);
133         if (!IS_ERR(clp->cl_rpcclient))
134                 rpc_shutdown_client(clp->cl_rpcclient);
135         kfree(clp);
136         nfs_callback_down();
137 }
138
139 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
140 {
141         struct nfs4_client *clp;
142         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
143                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
144                         atomic_inc(&clp->cl_count);
145                         return clp;
146                 }
147         }
148         return NULL;
149 }
150
151 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
152 {
153         struct nfs4_client *clp;
154         spin_lock(&state_spinlock);
155         clp = __nfs4_find_client(addr);
156         spin_unlock(&state_spinlock);
157         return clp;
158 }
159
160 struct nfs4_client *
161 nfs4_get_client(struct in_addr *addr)
162 {
163         struct nfs4_client *clp, *new = NULL;
164
165         spin_lock(&state_spinlock);
166         for (;;) {
167                 clp = __nfs4_find_client(addr);
168                 if (clp != NULL)
169                         break;
170                 clp = new;
171                 if (clp != NULL) {
172                         list_add(&clp->cl_servers, &nfs4_clientid_list);
173                         new = NULL;
174                         break;
175                 }
176                 spin_unlock(&state_spinlock);
177                 new = nfs4_alloc_client(addr);
178                 spin_lock(&state_spinlock);
179                 if (new == NULL)
180                         break;
181         }
182         spin_unlock(&state_spinlock);
183         if (new)
184                 nfs4_free_client(new);
185         return clp;
186 }
187
188 void
189 nfs4_put_client(struct nfs4_client *clp)
190 {
191         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
192                 return;
193         list_del(&clp->cl_servers);
194         spin_unlock(&state_spinlock);
195         BUG_ON(!list_empty(&clp->cl_superblocks));
196         wake_up_all(&clp->cl_waitq);
197         rpc_wake_up(&clp->cl_rpcwaitq);
198         nfs4_kill_renewd(clp);
199         nfs4_free_client(clp);
200 }
201
202 static int __nfs4_init_client(struct nfs4_client *clp)
203 {
204         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
205         if (status == 0)
206                 status = nfs4_proc_setclientid_confirm(clp);
207         if (status == 0)
208                 nfs4_schedule_state_renewal(clp);
209         return status;
210 }
211
212 int nfs4_init_client(struct nfs4_client *clp)
213 {
214         return nfs4_map_errors(__nfs4_init_client(clp));
215 }
216
217 u32
218 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
219 {
220         return clp->cl_lockowner_id ++;
221 }
222
223 static struct nfs4_state_owner *
224 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
225 {
226         struct nfs4_state_owner *sp = NULL;
227
228         if (!list_empty(&clp->cl_unused)) {
229                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
230                 atomic_inc(&sp->so_count);
231                 sp->so_cred = cred;
232                 list_move(&sp->so_list, &clp->cl_state_owners);
233                 clp->cl_nunused--;
234         }
235         return sp;
236 }
237
238 static struct nfs4_state_owner *
239 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
240 {
241         struct nfs4_state_owner *sp, *res = NULL;
242
243         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
244                 if (sp->so_cred != cred)
245                         continue;
246                 atomic_inc(&sp->so_count);
247                 /* Move to the head of the list */
248                 list_move(&sp->so_list, &clp->cl_state_owners);
249                 res = sp;
250                 break;
251         }
252         return res;
253 }
254
255 /*
256  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
257  * create a new state_owner.
258  *
259  */
260 static struct nfs4_state_owner *
261 nfs4_alloc_state_owner(void)
262 {
263         struct nfs4_state_owner *sp;
264
265         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
266         if (!sp)
267                 return NULL;
268         spin_lock_init(&sp->so_lock);
269         INIT_LIST_HEAD(&sp->so_states);
270         INIT_LIST_HEAD(&sp->so_delegations);
271         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
272         sp->so_seqid.sequence = &sp->so_sequence;
273         spin_lock_init(&sp->so_sequence.lock);
274         INIT_LIST_HEAD(&sp->so_sequence.list);
275         atomic_set(&sp->so_count, 1);
276         return sp;
277 }
278
279 void
280 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
281 {
282         struct nfs4_client *clp = sp->so_client;
283         spin_lock(&clp->cl_lock);
284         list_del_init(&sp->so_list);
285         spin_unlock(&clp->cl_lock);
286 }
287
288 /*
289  * Note: must be called with clp->cl_sem held in order to prevent races
290  *       with reboot recovery!
291  */
292 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
293 {
294         struct nfs4_client *clp = server->nfs4_state;
295         struct nfs4_state_owner *sp, *new;
296
297         get_rpccred(cred);
298         new = nfs4_alloc_state_owner();
299         spin_lock(&clp->cl_lock);
300         sp = nfs4_find_state_owner(clp, cred);
301         if (sp == NULL)
302                 sp = nfs4_client_grab_unused(clp, cred);
303         if (sp == NULL && new != NULL) {
304                 list_add(&new->so_list, &clp->cl_state_owners);
305                 new->so_client = clp;
306                 new->so_id = nfs4_alloc_lockowner_id(clp);
307                 new->so_cred = cred;
308                 sp = new;
309                 new = NULL;
310         }
311         spin_unlock(&clp->cl_lock);
312         kfree(new);
313         if (sp != NULL)
314                 return sp;
315         put_rpccred(cred);
316         return NULL;
317 }
318
319 /*
320  * Must be called with clp->cl_sem held in order to avoid races
321  * with state recovery...
322  */
323 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
324 {
325         struct nfs4_client *clp = sp->so_client;
326         struct rpc_cred *cred = sp->so_cred;
327
328         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
329                 return;
330         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
331                 goto out_free;
332         if (list_empty(&sp->so_list))
333                 goto out_free;
334         list_move(&sp->so_list, &clp->cl_unused);
335         clp->cl_nunused++;
336         spin_unlock(&clp->cl_lock);
337         put_rpccred(cred);
338         cred = NULL;
339         return;
340 out_free:
341         list_del(&sp->so_list);
342         spin_unlock(&clp->cl_lock);
343         put_rpccred(cred);
344         kfree(sp);
345 }
346
347 static struct nfs4_state *
348 nfs4_alloc_open_state(void)
349 {
350         struct nfs4_state *state;
351
352         state = kmalloc(sizeof(*state), GFP_KERNEL);
353         if (!state)
354                 return NULL;
355         state->state = 0;
356         state->nreaders = 0;
357         state->nwriters = 0;
358         state->flags = 0;
359         memset(state->stateid.data, 0, sizeof(state->stateid.data));
360         atomic_set(&state->count, 1);
361         INIT_LIST_HEAD(&state->lock_states);
362         spin_lock_init(&state->state_lock);
363         return state;
364 }
365
366 void
367 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
368 {
369         if (state->state == mode)
370                 return;
371         /* NB! List reordering - see the reclaim code for why.  */
372         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
373                 if (mode & FMODE_WRITE)
374                         list_move(&state->open_states, &state->owner->so_states);
375                 else
376                         list_move_tail(&state->open_states, &state->owner->so_states);
377         }
378         if (mode == 0)
379                 list_del_init(&state->inode_states);
380         state->state = mode;
381 }
382
383 static struct nfs4_state *
384 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
385 {
386         struct nfs_inode *nfsi = NFS_I(inode);
387         struct nfs4_state *state;
388
389         list_for_each_entry(state, &nfsi->open_states, inode_states) {
390                 /* Is this in the process of being freed? */
391                 if (state->state == 0)
392                         continue;
393                 if (state->owner == owner) {
394                         atomic_inc(&state->count);
395                         return state;
396                 }
397         }
398         return NULL;
399 }
400
401 static void
402 nfs4_free_open_state(struct nfs4_state *state)
403 {
404         kfree(state);
405 }
406
407 struct nfs4_state *
408 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
409 {
410         struct nfs4_state *state, *new;
411         struct nfs_inode *nfsi = NFS_I(inode);
412
413         spin_lock(&inode->i_lock);
414         state = __nfs4_find_state_byowner(inode, owner);
415         spin_unlock(&inode->i_lock);
416         if (state)
417                 goto out;
418         new = nfs4_alloc_open_state();
419         spin_lock(&owner->so_lock);
420         spin_lock(&inode->i_lock);
421         state = __nfs4_find_state_byowner(inode, owner);
422         if (state == NULL && new != NULL) {
423                 state = new;
424                 state->owner = owner;
425                 atomic_inc(&owner->so_count);
426                 list_add(&state->inode_states, &nfsi->open_states);
427                 state->inode = igrab(inode);
428                 spin_unlock(&inode->i_lock);
429                 /* Note: The reclaim code dictates that we add stateless
430                  * and read-only stateids to the end of the list */
431                 list_add_tail(&state->open_states, &owner->so_states);
432                 spin_unlock(&owner->so_lock);
433         } else {
434                 spin_unlock(&inode->i_lock);
435                 spin_unlock(&owner->so_lock);
436                 if (new)
437                         nfs4_free_open_state(new);
438         }
439 out:
440         return state;
441 }
442
443 /*
444  * Beware! Caller must be holding exactly one
445  * reference to clp->cl_sem!
446  */
447 void nfs4_put_open_state(struct nfs4_state *state)
448 {
449         struct inode *inode = state->inode;
450         struct nfs4_state_owner *owner = state->owner;
451
452         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
453                 return;
454         spin_lock(&inode->i_lock);
455         if (!list_empty(&state->inode_states))
456                 list_del(&state->inode_states);
457         list_del(&state->open_states);
458         spin_unlock(&inode->i_lock);
459         spin_unlock(&owner->so_lock);
460         iput(inode);
461         nfs4_free_open_state(state);
462         nfs4_put_state_owner(owner);
463 }
464
465 /*
466  * Close the current file.
467  */
468 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
469 {
470         struct inode *inode = state->inode;
471         struct nfs4_state_owner *owner = state->owner;
472         int oldstate, newstate = 0;
473
474         atomic_inc(&owner->so_count);
475         /* Protect against nfs4_find_state() */
476         spin_lock(&owner->so_lock);
477         spin_lock(&inode->i_lock);
478         if (mode & FMODE_READ)
479                 state->nreaders--;
480         if (mode & FMODE_WRITE)
481                 state->nwriters--;
482         oldstate = newstate = state->state;
483         if (state->nreaders == 0)
484                 newstate &= ~FMODE_READ;
485         if (state->nwriters == 0)
486                 newstate &= ~FMODE_WRITE;
487         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
488                 nfs4_state_set_mode_locked(state, newstate);
489                 oldstate = newstate;
490         }
491         spin_unlock(&inode->i_lock);
492         spin_unlock(&owner->so_lock);
493
494         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
495                 return;
496         nfs4_put_open_state(state);
497         nfs4_put_state_owner(owner);
498 }
499
500 /*
501  * Search the state->lock_states for an existing lock_owner
502  * that is compatible with current->files
503  */
504 static struct nfs4_lock_state *
505 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
506 {
507         struct nfs4_lock_state *pos;
508         list_for_each_entry(pos, &state->lock_states, ls_locks) {
509                 if (pos->ls_owner != fl_owner)
510                         continue;
511                 atomic_inc(&pos->ls_count);
512                 return pos;
513         }
514         return NULL;
515 }
516
517 /*
518  * Return a compatible lock_state. If no initialized lock_state structure
519  * exists, return an uninitialized one.
520  *
521  */
522 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
523 {
524         struct nfs4_lock_state *lsp;
525         struct nfs4_client *clp = state->owner->so_client;
526
527         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
528         if (lsp == NULL)
529                 return NULL;
530         lsp->ls_seqid.sequence = &state->owner->so_sequence;
531         atomic_set(&lsp->ls_count, 1);
532         lsp->ls_owner = fl_owner;
533         spin_lock(&clp->cl_lock);
534         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
535         spin_unlock(&clp->cl_lock);
536         INIT_LIST_HEAD(&lsp->ls_locks);
537         return lsp;
538 }
539
540 /*
541  * Return a compatible lock_state. If no initialized lock_state structure
542  * exists, return an uninitialized one.
543  *
544  * The caller must be holding clp->cl_sem
545  */
546 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
547 {
548         struct nfs4_lock_state *lsp, *new = NULL;
549         
550         for(;;) {
551                 spin_lock(&state->state_lock);
552                 lsp = __nfs4_find_lock_state(state, owner);
553                 if (lsp != NULL)
554                         break;
555                 if (new != NULL) {
556                         new->ls_state = state;
557                         list_add(&new->ls_locks, &state->lock_states);
558                         set_bit(LK_STATE_IN_USE, &state->flags);
559                         lsp = new;
560                         new = NULL;
561                         break;
562                 }
563                 spin_unlock(&state->state_lock);
564                 new = nfs4_alloc_lock_state(state, owner);
565                 if (new == NULL)
566                         return NULL;
567         }
568         spin_unlock(&state->state_lock);
569         kfree(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         kfree(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
633         memcpy(dst, &state->stateid, sizeof(*dst));
634         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
635                 return;
636
637         spin_lock(&state->state_lock);
638         lsp = __nfs4_find_lock_state(state, fl_owner);
639         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
640                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
641         spin_unlock(&state->state_lock);
642         nfs4_put_lock_state(lsp);
643 }
644
645 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
646 {
647         struct nfs_seqid *new;
648
649         new = kmalloc(sizeof(*new), GFP_KERNEL);
650         if (new != NULL) {
651                 new->sequence = counter;
652                 INIT_LIST_HEAD(&new->list);
653         }
654         return new;
655 }
656
657 void nfs_free_seqid(struct nfs_seqid *seqid)
658 {
659         struct rpc_sequence *sequence = seqid->sequence->sequence;
660
661         if (!list_empty(&seqid->list)) {
662                 spin_lock(&sequence->lock);
663                 list_del(&seqid->list);
664                 spin_unlock(&sequence->lock);
665         }
666         rpc_wake_up_next(&sequence->wait);
667         kfree(seqid);
668 }
669
670 /*
671  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
672  * failed with a seqid incrementing error -
673  * see comments nfs_fs.h:seqid_mutating_error()
674  */
675 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
676 {
677         switch (status) {
678                 case 0:
679                         break;
680                 case -NFS4ERR_BAD_SEQID:
681                 case -NFS4ERR_STALE_CLIENTID:
682                 case -NFS4ERR_STALE_STATEID:
683                 case -NFS4ERR_BAD_STATEID:
684                 case -NFS4ERR_BADXDR:
685                 case -NFS4ERR_RESOURCE:
686                 case -NFS4ERR_NOFILEHANDLE:
687                         /* Non-seqid mutating errors */
688                         return;
689         };
690         /*
691          * Note: no locking needed as we are guaranteed to be first
692          * on the sequence list
693          */
694         seqid->sequence->counter++;
695 }
696
697 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
698 {
699         if (status == -NFS4ERR_BAD_SEQID) {
700                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
701                                 struct nfs4_state_owner, so_seqid);
702                 nfs4_drop_state_owner(sp);
703         }
704         return nfs_increment_seqid(status, seqid);
705 }
706
707 /*
708  * Increment the seqid if the LOCK/LOCKU succeeded, or
709  * failed with a seqid incrementing error -
710  * see comments nfs_fs.h:seqid_mutating_error()
711  */
712 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
713 {
714         return nfs_increment_seqid(status, seqid);
715 }
716
717 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
718 {
719         struct rpc_sequence *sequence = seqid->sequence->sequence;
720         int status = 0;
721
722         if (sequence->list.next == &seqid->list)
723                 goto out;
724         spin_lock(&sequence->lock);
725         if (!list_empty(&sequence->list)) {
726                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
727                 status = -EAGAIN;
728         } else
729                 list_add(&seqid->list, &sequence->list);
730         spin_unlock(&sequence->lock);
731 out:
732         return status;
733 }
734
735 static int reclaimer(void *);
736 struct reclaimer_args {
737         struct nfs4_client *clp;
738         struct completion complete;
739 };
740
741 /*
742  * State recovery routine
743  */
744 void
745 nfs4_recover_state(void *data)
746 {
747         struct nfs4_client *clp = (struct nfs4_client *)data;
748         struct reclaimer_args args = {
749                 .clp = clp,
750         };
751         might_sleep();
752
753         init_completion(&args.complete);
754
755         if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
756                 goto out_failed_clear;
757         wait_for_completion(&args.complete);
758         return;
759 out_failed_clear:
760         set_bit(NFS4CLNT_OK, &clp->cl_state);
761         wake_up_all(&clp->cl_waitq);
762         rpc_wake_up(&clp->cl_rpcwaitq);
763 }
764
765 /*
766  * Schedule a state recovery attempt
767  */
768 void
769 nfs4_schedule_state_recovery(struct nfs4_client *clp)
770 {
771         if (!clp)
772                 return;
773         if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state))
774                 schedule_work(&clp->cl_recoverd);
775 }
776
777 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
778 {
779         struct inode *inode = state->inode;
780         struct file_lock *fl;
781         int status = 0;
782
783         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
784                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
785                         continue;
786                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
787                         continue;
788                 status = ops->recover_lock(state, fl);
789                 if (status >= 0)
790                         continue;
791                 switch (status) {
792                         default:
793                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
794                                                 __FUNCTION__, status);
795                         case -NFS4ERR_EXPIRED:
796                         case -NFS4ERR_NO_GRACE:
797                         case -NFS4ERR_RECLAIM_BAD:
798                         case -NFS4ERR_RECLAIM_CONFLICT:
799                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
800                                 break;
801                         case -NFS4ERR_STALE_CLIENTID:
802                                 goto out_err;
803                 }
804         }
805         return 0;
806 out_err:
807         return status;
808 }
809
810 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
811 {
812         struct nfs4_state *state;
813         struct nfs4_lock_state *lock;
814         int status = 0;
815
816         /* Note: we rely on the sp->so_states list being ordered 
817          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
818          * states first.
819          * This is needed to ensure that the server won't give us any
820          * read delegations that we have to return if, say, we are
821          * recovering after a network partition or a reboot from a
822          * server that doesn't support a grace period.
823          */
824         list_for_each_entry(state, &sp->so_states, open_states) {
825                 if (state->state == 0)
826                         continue;
827                 status = ops->recover_open(sp, state);
828                 if (status >= 0) {
829                         status = nfs4_reclaim_locks(ops, state);
830                         if (status < 0)
831                                 goto out_err;
832                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
833                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
834                                         printk("%s: Lock reclaim failed!\n",
835                                                         __FUNCTION__);
836                         }
837                         continue;
838                 }
839                 switch (status) {
840                         default:
841                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
842                                                 __FUNCTION__, status);
843                         case -ENOENT:
844                         case -NFS4ERR_RECLAIM_BAD:
845                         case -NFS4ERR_RECLAIM_CONFLICT:
846                                 /*
847                                  * Open state on this file cannot be recovered
848                                  * All we can do is revert to using the zero stateid.
849                                  */
850                                 memset(state->stateid.data, 0,
851                                         sizeof(state->stateid.data));
852                                 /* Mark the file as being 'closed' */
853                                 state->state = 0;
854                                 break;
855                         case -NFS4ERR_EXPIRED:
856                         case -NFS4ERR_NO_GRACE:
857                         case -NFS4ERR_STALE_CLIENTID:
858                                 goto out_err;
859                 }
860         }
861         return 0;
862 out_err:
863         return status;
864 }
865
866 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
867 {
868         struct nfs4_state_owner *sp;
869         struct nfs4_state *state;
870         struct nfs4_lock_state *lock;
871
872         /* Reset all sequence ids to zero */
873         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
874                 sp->so_seqid.counter = 0;
875                 sp->so_seqid.flags = 0;
876                 spin_lock(&sp->so_lock);
877                 list_for_each_entry(state, &sp->so_states, open_states) {
878                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
879                                 lock->ls_seqid.counter = 0;
880                                 lock->ls_seqid.flags = 0;
881                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
882                         }
883                 }
884                 spin_unlock(&sp->so_lock);
885         }
886 }
887
888 static int reclaimer(void *ptr)
889 {
890         struct reclaimer_args *args = (struct reclaimer_args *)ptr;
891         struct nfs4_client *clp = args->clp;
892         struct nfs4_state_owner *sp;
893         struct nfs4_state_recovery_ops *ops;
894         int status = 0;
895
896         daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
897         allow_signal(SIGKILL);
898
899         atomic_inc(&clp->cl_count);
900         complete(&args->complete);
901
902         /* Ensure exclusive access to NFSv4 state */
903         lock_kernel();
904         down_write(&clp->cl_sem);
905         /* Are there any NFS mounts out there? */
906         if (list_empty(&clp->cl_superblocks))
907                 goto out;
908 restart_loop:
909         status = nfs4_proc_renew(clp);
910         switch (status) {
911                 case 0:
912                 case -NFS4ERR_CB_PATH_DOWN:
913                         goto out;
914                 case -NFS4ERR_STALE_CLIENTID:
915                 case -NFS4ERR_LEASE_MOVED:
916                         ops = &nfs4_reboot_recovery_ops;
917                         break;
918                 default:
919                         ops = &nfs4_network_partition_recovery_ops;
920         };
921         nfs4_state_mark_reclaim(clp);
922         status = __nfs4_init_client(clp);
923         if (status)
924                 goto out_error;
925         /* Mark all delegations for reclaim */
926         nfs_delegation_mark_reclaim(clp);
927         /* Note: list is protected by exclusive lock on cl->cl_sem */
928         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
929                 status = nfs4_reclaim_open_state(ops, sp);
930                 if (status < 0) {
931                         if (status == -NFS4ERR_NO_GRACE) {
932                                 ops = &nfs4_network_partition_recovery_ops;
933                                 status = nfs4_reclaim_open_state(ops, sp);
934                         }
935                         if (status == -NFS4ERR_STALE_CLIENTID)
936                                 goto restart_loop;
937                         if (status == -NFS4ERR_EXPIRED)
938                                 goto restart_loop;
939                 }
940         }
941         nfs_delegation_reap_unclaimed(clp);
942 out:
943         set_bit(NFS4CLNT_OK, &clp->cl_state);
944         up_write(&clp->cl_sem);
945         unlock_kernel();
946         wake_up_all(&clp->cl_waitq);
947         rpc_wake_up(&clp->cl_rpcwaitq);
948         if (status == -NFS4ERR_CB_PATH_DOWN)
949                 nfs_handle_cb_pathdown(clp);
950         nfs4_put_client(clp);
951         return 0;
952 out_error:
953         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
954                                 NIPQUAD(clp->cl_addr.s_addr), -status);
955         goto out;
956 }
957
958 /*
959  * Local variables:
960  *  c-basic-offset: 8
961  * End:
962  */