Merge branch 'srp' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[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         if (server->mnt_path) {
73                 kfree(server->mnt_path);
74                 server->mnt_path = NULL;
75         }
76         if (server->nfs4_state) {
77                 nfs4_put_client(server->nfs4_state);
78                 server->nfs4_state = NULL;
79         }
80 }
81
82 /*
83  * nfs4_get_client(): returns an empty client structure
84  * nfs4_put_client(): drops reference to client structure
85  *
86  * Since these are allocated/deallocated very rarely, we don't
87  * bother putting them in a slab cache...
88  */
89 static struct nfs4_client *
90 nfs4_alloc_client(struct in_addr *addr)
91 {
92         struct nfs4_client *clp;
93
94         if (nfs_callback_up() < 0)
95                 return NULL;
96         if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
97                 nfs_callback_down();
98                 return NULL;
99         }
100         memset(clp, 0, sizeof(*clp));
101         memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
102         init_rwsem(&clp->cl_sem);
103         INIT_LIST_HEAD(&clp->cl_delegations);
104         INIT_LIST_HEAD(&clp->cl_state_owners);
105         INIT_LIST_HEAD(&clp->cl_unused);
106         spin_lock_init(&clp->cl_lock);
107         atomic_set(&clp->cl_count, 1);
108         INIT_WORK(&clp->cl_recoverd, nfs4_recover_state, clp);
109         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
110         INIT_LIST_HEAD(&clp->cl_superblocks);
111         init_waitqueue_head(&clp->cl_waitq);
112         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
113         clp->cl_rpcclient = ERR_PTR(-EINVAL);
114         clp->cl_boot_time = CURRENT_TIME;
115         clp->cl_state = 1 << NFS4CLNT_OK;
116         return clp;
117 }
118
119 static void
120 nfs4_free_client(struct nfs4_client *clp)
121 {
122         struct nfs4_state_owner *sp;
123
124         while (!list_empty(&clp->cl_unused)) {
125                 sp = list_entry(clp->cl_unused.next,
126                                 struct nfs4_state_owner,
127                                 so_list);
128                 list_del(&sp->so_list);
129                 kfree(sp);
130         }
131         BUG_ON(!list_empty(&clp->cl_state_owners));
132         if (clp->cl_cred)
133                 put_rpccred(clp->cl_cred);
134         nfs_idmap_delete(clp);
135         if (!IS_ERR(clp->cl_rpcclient))
136                 rpc_shutdown_client(clp->cl_rpcclient);
137         kfree(clp);
138         nfs_callback_down();
139 }
140
141 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
142 {
143         struct nfs4_client *clp;
144         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
145                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
146                         atomic_inc(&clp->cl_count);
147                         return clp;
148                 }
149         }
150         return NULL;
151 }
152
153 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
154 {
155         struct nfs4_client *clp;
156         spin_lock(&state_spinlock);
157         clp = __nfs4_find_client(addr);
158         spin_unlock(&state_spinlock);
159         return clp;
160 }
161
162 struct nfs4_client *
163 nfs4_get_client(struct in_addr *addr)
164 {
165         struct nfs4_client *clp, *new = NULL;
166
167         spin_lock(&state_spinlock);
168         for (;;) {
169                 clp = __nfs4_find_client(addr);
170                 if (clp != NULL)
171                         break;
172                 clp = new;
173                 if (clp != NULL) {
174                         list_add(&clp->cl_servers, &nfs4_clientid_list);
175                         new = NULL;
176                         break;
177                 }
178                 spin_unlock(&state_spinlock);
179                 new = nfs4_alloc_client(addr);
180                 spin_lock(&state_spinlock);
181                 if (new == NULL)
182                         break;
183         }
184         spin_unlock(&state_spinlock);
185         if (new)
186                 nfs4_free_client(new);
187         return clp;
188 }
189
190 void
191 nfs4_put_client(struct nfs4_client *clp)
192 {
193         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
194                 return;
195         list_del(&clp->cl_servers);
196         spin_unlock(&state_spinlock);
197         BUG_ON(!list_empty(&clp->cl_superblocks));
198         wake_up_all(&clp->cl_waitq);
199         rpc_wake_up(&clp->cl_rpcwaitq);
200         nfs4_kill_renewd(clp);
201         nfs4_free_client(clp);
202 }
203
204 static int __nfs4_init_client(struct nfs4_client *clp)
205 {
206         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
207         if (status == 0)
208                 status = nfs4_proc_setclientid_confirm(clp);
209         if (status == 0)
210                 nfs4_schedule_state_renewal(clp);
211         return status;
212 }
213
214 int nfs4_init_client(struct nfs4_client *clp)
215 {
216         return nfs4_map_errors(__nfs4_init_client(clp));
217 }
218
219 u32
220 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
221 {
222         return clp->cl_lockowner_id ++;
223 }
224
225 static struct nfs4_state_owner *
226 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
227 {
228         struct nfs4_state_owner *sp = NULL;
229
230         if (!list_empty(&clp->cl_unused)) {
231                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
232                 atomic_inc(&sp->so_count);
233                 sp->so_cred = cred;
234                 list_move(&sp->so_list, &clp->cl_state_owners);
235                 clp->cl_nunused--;
236         }
237         return sp;
238 }
239
240 static struct nfs4_state_owner *
241 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
242 {
243         struct nfs4_state_owner *sp, *res = NULL;
244
245         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
246                 if (sp->so_cred != cred)
247                         continue;
248                 atomic_inc(&sp->so_count);
249                 /* Move to the head of the list */
250                 list_move(&sp->so_list, &clp->cl_state_owners);
251                 res = sp;
252                 break;
253         }
254         return res;
255 }
256
257 /*
258  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
259  * create a new state_owner.
260  *
261  */
262 static struct nfs4_state_owner *
263 nfs4_alloc_state_owner(void)
264 {
265         struct nfs4_state_owner *sp;
266
267         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
268         if (!sp)
269                 return NULL;
270         spin_lock_init(&sp->so_lock);
271         INIT_LIST_HEAD(&sp->so_states);
272         INIT_LIST_HEAD(&sp->so_delegations);
273         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
274         sp->so_seqid.sequence = &sp->so_sequence;
275         spin_lock_init(&sp->so_sequence.lock);
276         INIT_LIST_HEAD(&sp->so_sequence.list);
277         atomic_set(&sp->so_count, 1);
278         return sp;
279 }
280
281 void
282 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
283 {
284         struct nfs4_client *clp = sp->so_client;
285         spin_lock(&clp->cl_lock);
286         list_del_init(&sp->so_list);
287         spin_unlock(&clp->cl_lock);
288 }
289
290 /*
291  * Note: must be called with clp->cl_sem held in order to prevent races
292  *       with reboot recovery!
293  */
294 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
295 {
296         struct nfs4_client *clp = server->nfs4_state;
297         struct nfs4_state_owner *sp, *new;
298
299         get_rpccred(cred);
300         new = nfs4_alloc_state_owner();
301         spin_lock(&clp->cl_lock);
302         sp = nfs4_find_state_owner(clp, cred);
303         if (sp == NULL)
304                 sp = nfs4_client_grab_unused(clp, cred);
305         if (sp == NULL && new != NULL) {
306                 list_add(&new->so_list, &clp->cl_state_owners);
307                 new->so_client = clp;
308                 new->so_id = nfs4_alloc_lockowner_id(clp);
309                 new->so_cred = cred;
310                 sp = new;
311                 new = NULL;
312         }
313         spin_unlock(&clp->cl_lock);
314         if (new)
315                 kfree(new);
316         if (sp != NULL)
317                 return sp;
318         put_rpccred(cred);
319         return NULL;
320 }
321
322 /*
323  * Must be called with clp->cl_sem held in order to avoid races
324  * with state recovery...
325  */
326 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
327 {
328         struct nfs4_client *clp = sp->so_client;
329         struct rpc_cred *cred = sp->so_cred;
330
331         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
332                 return;
333         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
334                 goto out_free;
335         if (list_empty(&sp->so_list))
336                 goto out_free;
337         list_move(&sp->so_list, &clp->cl_unused);
338         clp->cl_nunused++;
339         spin_unlock(&clp->cl_lock);
340         put_rpccred(cred);
341         cred = NULL;
342         return;
343 out_free:
344         list_del(&sp->so_list);
345         spin_unlock(&clp->cl_lock);
346         put_rpccred(cred);
347         kfree(sp);
348 }
349
350 static struct nfs4_state *
351 nfs4_alloc_open_state(void)
352 {
353         struct nfs4_state *state;
354
355         state = kmalloc(sizeof(*state), GFP_KERNEL);
356         if (!state)
357                 return NULL;
358         state->state = 0;
359         state->nreaders = 0;
360         state->nwriters = 0;
361         state->flags = 0;
362         memset(state->stateid.data, 0, sizeof(state->stateid.data));
363         atomic_set(&state->count, 1);
364         INIT_LIST_HEAD(&state->lock_states);
365         spin_lock_init(&state->state_lock);
366         return state;
367 }
368
369 void
370 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
371 {
372         if (state->state == mode)
373                 return;
374         /* NB! List reordering - see the reclaim code for why.  */
375         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
376                 if (mode & FMODE_WRITE)
377                         list_move(&state->open_states, &state->owner->so_states);
378                 else
379                         list_move_tail(&state->open_states, &state->owner->so_states);
380         }
381         if (mode == 0)
382                 list_del_init(&state->inode_states);
383         state->state = mode;
384 }
385
386 static struct nfs4_state *
387 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
388 {
389         struct nfs_inode *nfsi = NFS_I(inode);
390         struct nfs4_state *state;
391
392         list_for_each_entry(state, &nfsi->open_states, inode_states) {
393                 /* Is this in the process of being freed? */
394                 if (state->state == 0)
395                         continue;
396                 if (state->owner == owner) {
397                         atomic_inc(&state->count);
398                         return state;
399                 }
400         }
401         return NULL;
402 }
403
404 static void
405 nfs4_free_open_state(struct nfs4_state *state)
406 {
407         kfree(state);
408 }
409
410 struct nfs4_state *
411 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
412 {
413         struct nfs4_state *state, *new;
414         struct nfs_inode *nfsi = NFS_I(inode);
415
416         spin_lock(&inode->i_lock);
417         state = __nfs4_find_state_byowner(inode, owner);
418         spin_unlock(&inode->i_lock);
419         if (state)
420                 goto out;
421         new = nfs4_alloc_open_state();
422         spin_lock(&owner->so_lock);
423         spin_lock(&inode->i_lock);
424         state = __nfs4_find_state_byowner(inode, owner);
425         if (state == NULL && new != NULL) {
426                 state = new;
427                 state->owner = owner;
428                 atomic_inc(&owner->so_count);
429                 list_add(&state->inode_states, &nfsi->open_states);
430                 state->inode = igrab(inode);
431                 spin_unlock(&inode->i_lock);
432                 /* Note: The reclaim code dictates that we add stateless
433                  * and read-only stateids to the end of the list */
434                 list_add_tail(&state->open_states, &owner->so_states);
435                 spin_unlock(&owner->so_lock);
436         } else {
437                 spin_unlock(&inode->i_lock);
438                 spin_unlock(&owner->so_lock);
439                 if (new)
440                         nfs4_free_open_state(new);
441         }
442 out:
443         return state;
444 }
445
446 /*
447  * Beware! Caller must be holding exactly one
448  * reference to clp->cl_sem!
449  */
450 void nfs4_put_open_state(struct nfs4_state *state)
451 {
452         struct inode *inode = state->inode;
453         struct nfs4_state_owner *owner = state->owner;
454
455         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
456                 return;
457         spin_lock(&inode->i_lock);
458         if (!list_empty(&state->inode_states))
459                 list_del(&state->inode_states);
460         list_del(&state->open_states);
461         spin_unlock(&inode->i_lock);
462         spin_unlock(&owner->so_lock);
463         iput(inode);
464         nfs4_free_open_state(state);
465         nfs4_put_state_owner(owner);
466 }
467
468 /*
469  * Close the current file.
470  */
471 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
472 {
473         struct inode *inode = state->inode;
474         struct nfs4_state_owner *owner = state->owner;
475         int oldstate, newstate = 0;
476
477         atomic_inc(&owner->so_count);
478         /* Protect against nfs4_find_state() */
479         spin_lock(&owner->so_lock);
480         spin_lock(&inode->i_lock);
481         if (mode & FMODE_READ)
482                 state->nreaders--;
483         if (mode & FMODE_WRITE)
484                 state->nwriters--;
485         oldstate = newstate = state->state;
486         if (state->nreaders == 0)
487                 newstate &= ~FMODE_READ;
488         if (state->nwriters == 0)
489                 newstate &= ~FMODE_WRITE;
490         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
491                 nfs4_state_set_mode_locked(state, newstate);
492                 oldstate = newstate;
493         }
494         spin_unlock(&inode->i_lock);
495         spin_unlock(&owner->so_lock);
496
497         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
498                 return;
499         nfs4_put_open_state(state);
500         nfs4_put_state_owner(owner);
501 }
502
503 /*
504  * Search the state->lock_states for an existing lock_owner
505  * that is compatible with current->files
506  */
507 static struct nfs4_lock_state *
508 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
509 {
510         struct nfs4_lock_state *pos;
511         list_for_each_entry(pos, &state->lock_states, ls_locks) {
512                 if (pos->ls_owner != fl_owner)
513                         continue;
514                 atomic_inc(&pos->ls_count);
515                 return pos;
516         }
517         return NULL;
518 }
519
520 /*
521  * Return a compatible lock_state. If no initialized lock_state structure
522  * exists, return an uninitialized one.
523  *
524  */
525 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
526 {
527         struct nfs4_lock_state *lsp;
528         struct nfs4_client *clp = state->owner->so_client;
529
530         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
531         if (lsp == NULL)
532                 return NULL;
533         lsp->ls_seqid.sequence = &state->owner->so_sequence;
534         atomic_set(&lsp->ls_count, 1);
535         lsp->ls_owner = fl_owner;
536         spin_lock(&clp->cl_lock);
537         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
538         spin_unlock(&clp->cl_lock);
539         INIT_LIST_HEAD(&lsp->ls_locks);
540         return lsp;
541 }
542
543 /*
544  * Return a compatible lock_state. If no initialized lock_state structure
545  * exists, return an uninitialized one.
546  *
547  * The caller must be holding clp->cl_sem
548  */
549 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
550 {
551         struct nfs4_lock_state *lsp, *new = NULL;
552         
553         for(;;) {
554                 spin_lock(&state->state_lock);
555                 lsp = __nfs4_find_lock_state(state, owner);
556                 if (lsp != NULL)
557                         break;
558                 if (new != NULL) {
559                         new->ls_state = state;
560                         list_add(&new->ls_locks, &state->lock_states);
561                         set_bit(LK_STATE_IN_USE, &state->flags);
562                         lsp = new;
563                         new = NULL;
564                         break;
565                 }
566                 spin_unlock(&state->state_lock);
567                 new = nfs4_alloc_lock_state(state, owner);
568                 if (new == NULL)
569                         return NULL;
570         }
571         spin_unlock(&state->state_lock);
572         kfree(new);
573         return lsp;
574 }
575
576 /*
577  * Release reference to lock_state, and free it if we see that
578  * it is no longer in use
579  */
580 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
581 {
582         struct nfs4_state *state;
583
584         if (lsp == NULL)
585                 return;
586         state = lsp->ls_state;
587         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
588                 return;
589         list_del(&lsp->ls_locks);
590         if (list_empty(&state->lock_states))
591                 clear_bit(LK_STATE_IN_USE, &state->flags);
592         spin_unlock(&state->state_lock);
593         kfree(lsp);
594 }
595
596 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
597 {
598         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
599
600         dst->fl_u.nfs4_fl.owner = lsp;
601         atomic_inc(&lsp->ls_count);
602 }
603
604 static void nfs4_fl_release_lock(struct file_lock *fl)
605 {
606         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
607 }
608
609 static struct file_lock_operations nfs4_fl_lock_ops = {
610         .fl_copy_lock = nfs4_fl_copy_lock,
611         .fl_release_private = nfs4_fl_release_lock,
612 };
613
614 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
615 {
616         struct nfs4_lock_state *lsp;
617
618         if (fl->fl_ops != NULL)
619                 return 0;
620         lsp = nfs4_get_lock_state(state, fl->fl_owner);
621         if (lsp == NULL)
622                 return -ENOMEM;
623         fl->fl_u.nfs4_fl.owner = lsp;
624         fl->fl_ops = &nfs4_fl_lock_ops;
625         return 0;
626 }
627
628 /*
629  * Byte-range lock aware utility to initialize the stateid of read/write
630  * requests.
631  */
632 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
633 {
634         struct nfs4_lock_state *lsp;
635
636         memcpy(dst, &state->stateid, sizeof(*dst));
637         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
638                 return;
639
640         spin_lock(&state->state_lock);
641         lsp = __nfs4_find_lock_state(state, fl_owner);
642         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
643                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
644         spin_unlock(&state->state_lock);
645         nfs4_put_lock_state(lsp);
646 }
647
648 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
649 {
650         struct nfs_seqid *new;
651
652         new = kmalloc(sizeof(*new), GFP_KERNEL);
653         if (new != NULL) {
654                 new->sequence = counter;
655                 INIT_LIST_HEAD(&new->list);
656         }
657         return new;
658 }
659
660 void nfs_free_seqid(struct nfs_seqid *seqid)
661 {
662         struct rpc_sequence *sequence = seqid->sequence->sequence;
663
664         if (!list_empty(&seqid->list)) {
665                 spin_lock(&sequence->lock);
666                 list_del(&seqid->list);
667                 spin_unlock(&sequence->lock);
668         }
669         rpc_wake_up_next(&sequence->wait);
670         kfree(seqid);
671 }
672
673 /*
674  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
675  * failed with a seqid incrementing error -
676  * see comments nfs_fs.h:seqid_mutating_error()
677  */
678 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
679 {
680         switch (status) {
681                 case 0:
682                         break;
683                 case -NFS4ERR_BAD_SEQID:
684                 case -NFS4ERR_STALE_CLIENTID:
685                 case -NFS4ERR_STALE_STATEID:
686                 case -NFS4ERR_BAD_STATEID:
687                 case -NFS4ERR_BADXDR:
688                 case -NFS4ERR_RESOURCE:
689                 case -NFS4ERR_NOFILEHANDLE:
690                         /* Non-seqid mutating errors */
691                         return;
692         };
693         /*
694          * Note: no locking needed as we are guaranteed to be first
695          * on the sequence list
696          */
697         seqid->sequence->counter++;
698 }
699
700 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
701 {
702         if (status == -NFS4ERR_BAD_SEQID) {
703                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
704                                 struct nfs4_state_owner, so_seqid);
705                 nfs4_drop_state_owner(sp);
706         }
707         return nfs_increment_seqid(status, seqid);
708 }
709
710 /*
711  * Increment the seqid if the LOCK/LOCKU succeeded, or
712  * failed with a seqid incrementing error -
713  * see comments nfs_fs.h:seqid_mutating_error()
714  */
715 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
716 {
717         return nfs_increment_seqid(status, seqid);
718 }
719
720 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
721 {
722         struct rpc_sequence *sequence = seqid->sequence->sequence;
723         int status = 0;
724
725         if (sequence->list.next == &seqid->list)
726                 goto out;
727         spin_lock(&sequence->lock);
728         if (!list_empty(&sequence->list)) {
729                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
730                 status = -EAGAIN;
731         } else
732                 list_add(&seqid->list, &sequence->list);
733         spin_unlock(&sequence->lock);
734 out:
735         return status;
736 }
737
738 static int reclaimer(void *);
739 struct reclaimer_args {
740         struct nfs4_client *clp;
741         struct completion complete;
742 };
743
744 /*
745  * State recovery routine
746  */
747 void
748 nfs4_recover_state(void *data)
749 {
750         struct nfs4_client *clp = (struct nfs4_client *)data;
751         struct reclaimer_args args = {
752                 .clp = clp,
753         };
754         might_sleep();
755
756         init_completion(&args.complete);
757
758         if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
759                 goto out_failed_clear;
760         wait_for_completion(&args.complete);
761         return;
762 out_failed_clear:
763         set_bit(NFS4CLNT_OK, &clp->cl_state);
764         wake_up_all(&clp->cl_waitq);
765         rpc_wake_up(&clp->cl_rpcwaitq);
766 }
767
768 /*
769  * Schedule a state recovery attempt
770  */
771 void
772 nfs4_schedule_state_recovery(struct nfs4_client *clp)
773 {
774         if (!clp)
775                 return;
776         if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state))
777                 schedule_work(&clp->cl_recoverd);
778 }
779
780 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
781 {
782         struct inode *inode = state->inode;
783         struct file_lock *fl;
784         int status = 0;
785
786         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
787                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
788                         continue;
789                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
790                         continue;
791                 status = ops->recover_lock(state, fl);
792                 if (status >= 0)
793                         continue;
794                 switch (status) {
795                         default:
796                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
797                                                 __FUNCTION__, status);
798                         case -NFS4ERR_EXPIRED:
799                         case -NFS4ERR_NO_GRACE:
800                         case -NFS4ERR_RECLAIM_BAD:
801                         case -NFS4ERR_RECLAIM_CONFLICT:
802                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
803                                 break;
804                         case -NFS4ERR_STALE_CLIENTID:
805                                 goto out_err;
806                 }
807         }
808         return 0;
809 out_err:
810         return status;
811 }
812
813 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
814 {
815         struct nfs4_state *state;
816         struct nfs4_lock_state *lock;
817         int status = 0;
818
819         /* Note: we rely on the sp->so_states list being ordered 
820          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
821          * states first.
822          * This is needed to ensure that the server won't give us any
823          * read delegations that we have to return if, say, we are
824          * recovering after a network partition or a reboot from a
825          * server that doesn't support a grace period.
826          */
827         list_for_each_entry(state, &sp->so_states, open_states) {
828                 if (state->state == 0)
829                         continue;
830                 status = ops->recover_open(sp, state);
831                 if (status >= 0) {
832                         status = nfs4_reclaim_locks(ops, state);
833                         if (status < 0)
834                                 goto out_err;
835                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
836                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
837                                         printk("%s: Lock reclaim failed!\n",
838                                                         __FUNCTION__);
839                         }
840                         continue;
841                 }
842                 switch (status) {
843                         default:
844                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
845                                                 __FUNCTION__, status);
846                         case -ENOENT:
847                         case -NFS4ERR_RECLAIM_BAD:
848                         case -NFS4ERR_RECLAIM_CONFLICT:
849                                 /*
850                                  * Open state on this file cannot be recovered
851                                  * All we can do is revert to using the zero stateid.
852                                  */
853                                 memset(state->stateid.data, 0,
854                                         sizeof(state->stateid.data));
855                                 /* Mark the file as being 'closed' */
856                                 state->state = 0;
857                                 break;
858                         case -NFS4ERR_EXPIRED:
859                         case -NFS4ERR_NO_GRACE:
860                         case -NFS4ERR_STALE_CLIENTID:
861                                 goto out_err;
862                 }
863         }
864         return 0;
865 out_err:
866         return status;
867 }
868
869 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
870 {
871         struct nfs4_state_owner *sp;
872         struct nfs4_state *state;
873         struct nfs4_lock_state *lock;
874
875         /* Reset all sequence ids to zero */
876         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
877                 sp->so_seqid.counter = 0;
878                 sp->so_seqid.flags = 0;
879                 spin_lock(&sp->so_lock);
880                 list_for_each_entry(state, &sp->so_states, open_states) {
881                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
882                                 lock->ls_seqid.counter = 0;
883                                 lock->ls_seqid.flags = 0;
884                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
885                         }
886                 }
887                 spin_unlock(&sp->so_lock);
888         }
889 }
890
891 static int reclaimer(void *ptr)
892 {
893         struct reclaimer_args *args = (struct reclaimer_args *)ptr;
894         struct nfs4_client *clp = args->clp;
895         struct nfs4_state_owner *sp;
896         struct nfs4_state_recovery_ops *ops;
897         int status = 0;
898
899         daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
900         allow_signal(SIGKILL);
901
902         atomic_inc(&clp->cl_count);
903         complete(&args->complete);
904
905         /* Ensure exclusive access to NFSv4 state */
906         lock_kernel();
907         down_write(&clp->cl_sem);
908         /* Are there any NFS mounts out there? */
909         if (list_empty(&clp->cl_superblocks))
910                 goto out;
911 restart_loop:
912         status = nfs4_proc_renew(clp);
913         switch (status) {
914                 case 0:
915                 case -NFS4ERR_CB_PATH_DOWN:
916                         goto out;
917                 case -NFS4ERR_STALE_CLIENTID:
918                 case -NFS4ERR_LEASE_MOVED:
919                         ops = &nfs4_reboot_recovery_ops;
920                         break;
921                 default:
922                         ops = &nfs4_network_partition_recovery_ops;
923         };
924         nfs4_state_mark_reclaim(clp);
925         status = __nfs4_init_client(clp);
926         if (status)
927                 goto out_error;
928         /* Mark all delegations for reclaim */
929         nfs_delegation_mark_reclaim(clp);
930         /* Note: list is protected by exclusive lock on cl->cl_sem */
931         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
932                 status = nfs4_reclaim_open_state(ops, sp);
933                 if (status < 0) {
934                         if (status == -NFS4ERR_NO_GRACE) {
935                                 ops = &nfs4_network_partition_recovery_ops;
936                                 status = nfs4_reclaim_open_state(ops, sp);
937                         }
938                         if (status == -NFS4ERR_STALE_CLIENTID)
939                                 goto restart_loop;
940                         if (status == -NFS4ERR_EXPIRED)
941                                 goto restart_loop;
942                 }
943         }
944         nfs_delegation_reap_unclaimed(clp);
945 out:
946         set_bit(NFS4CLNT_OK, &clp->cl_state);
947         up_write(&clp->cl_sem);
948         unlock_kernel();
949         wake_up_all(&clp->cl_waitq);
950         rpc_wake_up(&clp->cl_rpcwaitq);
951         if (status == -NFS4ERR_CB_PATH_DOWN)
952                 nfs_handle_cb_pathdown(clp);
953         nfs4_put_client(clp);
954         return 0;
955 out_error:
956         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
957                                 NIPQUAD(clp->cl_addr.s_addr), -status);
958         goto out;
959 }
960
961 /*
962  * Local variables:
963  *  c-basic-offset: 8
964  * End:
965  */