NFSv4: Clean up NFS4ERR_CB_PATH_DOWN error management...
[linux-2.6] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/smp_lock.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/workqueue.h>
50 #include <linux/bitops.h>
51
52 #include "nfs4_fs.h"
53 #include "callback.h"
54 #include "delegation.h"
55 #include "internal.h"
56
57 #define OPENOWNER_POOL_SIZE     8
58
59 const nfs4_stateid zero_stateid;
60
61 static LIST_HEAD(nfs4_clientid_list);
62
63 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
64 {
65         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
66                         nfs_callback_tcpport, cred);
67         if (status == 0)
68                 status = nfs4_proc_setclientid_confirm(clp, cred);
69         if (status == 0)
70                 nfs4_schedule_state_renewal(clp);
71         return status;
72 }
73
74 static struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
75 {
76         struct rpc_cred *cred = NULL;
77
78         if (clp->cl_machine_cred != NULL)
79                 cred = get_rpccred(clp->cl_machine_cred);
80         return cred;
81 }
82
83 static void nfs4_clear_machine_cred(struct nfs_client *clp)
84 {
85         struct rpc_cred *cred;
86
87         spin_lock(&clp->cl_lock);
88         cred = clp->cl_machine_cred;
89         clp->cl_machine_cred = NULL;
90         spin_unlock(&clp->cl_lock);
91         if (cred != NULL)
92                 put_rpccred(cred);
93 }
94
95 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
96 {
97         struct nfs4_state_owner *sp;
98         struct rb_node *pos;
99         struct rpc_cred *cred = NULL;
100
101         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
102                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
103                 if (list_empty(&sp->so_states))
104                         continue;
105                 cred = get_rpccred(sp->so_cred);
106                 break;
107         }
108         return cred;
109 }
110
111 static struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
112 {
113         struct rpc_cred *cred;
114
115         spin_lock(&clp->cl_lock);
116         cred = nfs4_get_renew_cred_locked(clp);
117         spin_unlock(&clp->cl_lock);
118         return cred;
119 }
120
121 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
122 {
123         struct nfs4_state_owner *sp;
124         struct rb_node *pos;
125         struct rpc_cred *cred;
126
127         spin_lock(&clp->cl_lock);
128         cred = nfs4_get_machine_cred_locked(clp);
129         if (cred != NULL)
130                 goto out;
131         pos = rb_first(&clp->cl_state_owners);
132         if (pos != NULL) {
133                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
134                 cred = get_rpccred(sp->so_cred);
135         }
136 out:
137         spin_unlock(&clp->cl_lock);
138         return cred;
139 }
140
141 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
142                 __u64 minval, int maxbits)
143 {
144         struct rb_node **p, *parent;
145         struct nfs_unique_id *pos;
146         __u64 mask = ~0ULL;
147
148         if (maxbits < 64)
149                 mask = (1ULL << maxbits) - 1ULL;
150
151         /* Ensure distribution is more or less flat */
152         get_random_bytes(&new->id, sizeof(new->id));
153         new->id &= mask;
154         if (new->id < minval)
155                 new->id += minval;
156 retry:
157         p = &root->rb_node;
158         parent = NULL;
159
160         while (*p != NULL) {
161                 parent = *p;
162                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
163
164                 if (new->id < pos->id)
165                         p = &(*p)->rb_left;
166                 else if (new->id > pos->id)
167                         p = &(*p)->rb_right;
168                 else
169                         goto id_exists;
170         }
171         rb_link_node(&new->rb_node, parent, p);
172         rb_insert_color(&new->rb_node, root);
173         return;
174 id_exists:
175         for (;;) {
176                 new->id++;
177                 if (new->id < minval || (new->id & mask) != new->id) {
178                         new->id = minval;
179                         break;
180                 }
181                 parent = rb_next(parent);
182                 if (parent == NULL)
183                         break;
184                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
185                 if (new->id < pos->id)
186                         break;
187         }
188         goto retry;
189 }
190
191 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
192 {
193         rb_erase(&id->rb_node, root);
194 }
195
196 static struct nfs4_state_owner *
197 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
198 {
199         struct nfs_client *clp = server->nfs_client;
200         struct rb_node **p = &clp->cl_state_owners.rb_node,
201                        *parent = NULL;
202         struct nfs4_state_owner *sp, *res = NULL;
203
204         while (*p != NULL) {
205                 parent = *p;
206                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
207
208                 if (server < sp->so_server) {
209                         p = &parent->rb_left;
210                         continue;
211                 }
212                 if (server > sp->so_server) {
213                         p = &parent->rb_right;
214                         continue;
215                 }
216                 if (cred < sp->so_cred)
217                         p = &parent->rb_left;
218                 else if (cred > sp->so_cred)
219                         p = &parent->rb_right;
220                 else {
221                         atomic_inc(&sp->so_count);
222                         res = sp;
223                         break;
224                 }
225         }
226         return res;
227 }
228
229 static struct nfs4_state_owner *
230 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
231 {
232         struct rb_node **p = &clp->cl_state_owners.rb_node,
233                        *parent = NULL;
234         struct nfs4_state_owner *sp;
235
236         while (*p != NULL) {
237                 parent = *p;
238                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
239
240                 if (new->so_server < sp->so_server) {
241                         p = &parent->rb_left;
242                         continue;
243                 }
244                 if (new->so_server > sp->so_server) {
245                         p = &parent->rb_right;
246                         continue;
247                 }
248                 if (new->so_cred < sp->so_cred)
249                         p = &parent->rb_left;
250                 else if (new->so_cred > sp->so_cred)
251                         p = &parent->rb_right;
252                 else {
253                         atomic_inc(&sp->so_count);
254                         return sp;
255                 }
256         }
257         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
258         rb_link_node(&new->so_client_node, parent, p);
259         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
260         return new;
261 }
262
263 static void
264 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
265 {
266         if (!RB_EMPTY_NODE(&sp->so_client_node))
267                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
268         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
269 }
270
271 /*
272  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
273  * create a new state_owner.
274  *
275  */
276 static struct nfs4_state_owner *
277 nfs4_alloc_state_owner(void)
278 {
279         struct nfs4_state_owner *sp;
280
281         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
282         if (!sp)
283                 return NULL;
284         spin_lock_init(&sp->so_lock);
285         INIT_LIST_HEAD(&sp->so_states);
286         INIT_LIST_HEAD(&sp->so_delegations);
287         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
288         sp->so_seqid.sequence = &sp->so_sequence;
289         spin_lock_init(&sp->so_sequence.lock);
290         INIT_LIST_HEAD(&sp->so_sequence.list);
291         atomic_set(&sp->so_count, 1);
292         return sp;
293 }
294
295 static void
296 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
297 {
298         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
299                 struct nfs_client *clp = sp->so_client;
300
301                 spin_lock(&clp->cl_lock);
302                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
303                 RB_CLEAR_NODE(&sp->so_client_node);
304                 spin_unlock(&clp->cl_lock);
305         }
306 }
307
308 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
309 {
310         struct nfs_client *clp = server->nfs_client;
311         struct nfs4_state_owner *sp, *new;
312
313         spin_lock(&clp->cl_lock);
314         sp = nfs4_find_state_owner(server, cred);
315         spin_unlock(&clp->cl_lock);
316         if (sp != NULL)
317                 return sp;
318         new = nfs4_alloc_state_owner();
319         if (new == NULL)
320                 return NULL;
321         new->so_client = clp;
322         new->so_server = server;
323         new->so_cred = cred;
324         spin_lock(&clp->cl_lock);
325         sp = nfs4_insert_state_owner(clp, new);
326         spin_unlock(&clp->cl_lock);
327         if (sp == new)
328                 get_rpccred(cred);
329         else {
330                 rpc_destroy_wait_queue(&new->so_sequence.wait);
331                 kfree(new);
332         }
333         return sp;
334 }
335
336 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
337 {
338         struct nfs_client *clp = sp->so_client;
339         struct rpc_cred *cred = sp->so_cred;
340
341         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
342                 return;
343         nfs4_remove_state_owner(clp, sp);
344         spin_unlock(&clp->cl_lock);
345         rpc_destroy_wait_queue(&sp->so_sequence.wait);
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 = kzalloc(sizeof(*state), GFP_KERNEL);
356         if (!state)
357                 return NULL;
358         atomic_set(&state->count, 1);
359         INIT_LIST_HEAD(&state->lock_states);
360         spin_lock_init(&state->state_lock);
361         seqlock_init(&state->seqlock);
362         return state;
363 }
364
365 void
366 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
367 {
368         if (state->state == mode)
369                 return;
370         /* NB! List reordering - see the reclaim code for why.  */
371         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
372                 if (mode & FMODE_WRITE)
373                         list_move(&state->open_states, &state->owner->so_states);
374                 else
375                         list_move_tail(&state->open_states, &state->owner->so_states);
376         }
377         state->state = mode;
378 }
379
380 static struct nfs4_state *
381 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
382 {
383         struct nfs_inode *nfsi = NFS_I(inode);
384         struct nfs4_state *state;
385
386         list_for_each_entry(state, &nfsi->open_states, inode_states) {
387                 if (state->owner != owner)
388                         continue;
389                 if (atomic_inc_not_zero(&state->count))
390                         return state;
391         }
392         return NULL;
393 }
394
395 static void
396 nfs4_free_open_state(struct nfs4_state *state)
397 {
398         kfree(state);
399 }
400
401 struct nfs4_state *
402 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
403 {
404         struct nfs4_state *state, *new;
405         struct nfs_inode *nfsi = NFS_I(inode);
406
407         spin_lock(&inode->i_lock);
408         state = __nfs4_find_state_byowner(inode, owner);
409         spin_unlock(&inode->i_lock);
410         if (state)
411                 goto out;
412         new = nfs4_alloc_open_state();
413         spin_lock(&owner->so_lock);
414         spin_lock(&inode->i_lock);
415         state = __nfs4_find_state_byowner(inode, owner);
416         if (state == NULL && new != NULL) {
417                 state = new;
418                 state->owner = owner;
419                 atomic_inc(&owner->so_count);
420                 list_add(&state->inode_states, &nfsi->open_states);
421                 state->inode = igrab(inode);
422                 spin_unlock(&inode->i_lock);
423                 /* Note: The reclaim code dictates that we add stateless
424                  * and read-only stateids to the end of the list */
425                 list_add_tail(&state->open_states, &owner->so_states);
426                 spin_unlock(&owner->so_lock);
427         } else {
428                 spin_unlock(&inode->i_lock);
429                 spin_unlock(&owner->so_lock);
430                 if (new)
431                         nfs4_free_open_state(new);
432         }
433 out:
434         return state;
435 }
436
437 void nfs4_put_open_state(struct nfs4_state *state)
438 {
439         struct inode *inode = state->inode;
440         struct nfs4_state_owner *owner = state->owner;
441
442         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
443                 return;
444         spin_lock(&inode->i_lock);
445         list_del(&state->inode_states);
446         list_del(&state->open_states);
447         spin_unlock(&inode->i_lock);
448         spin_unlock(&owner->so_lock);
449         iput(inode);
450         nfs4_free_open_state(state);
451         nfs4_put_state_owner(owner);
452 }
453
454 /*
455  * Close the current file.
456  */
457 static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
458 {
459         struct nfs4_state_owner *owner = state->owner;
460         int call_close = 0;
461         int newstate;
462
463         atomic_inc(&owner->so_count);
464         /* Protect against nfs4_find_state() */
465         spin_lock(&owner->so_lock);
466         switch (mode & (FMODE_READ | FMODE_WRITE)) {
467                 case FMODE_READ:
468                         state->n_rdonly--;
469                         break;
470                 case FMODE_WRITE:
471                         state->n_wronly--;
472                         break;
473                 case FMODE_READ|FMODE_WRITE:
474                         state->n_rdwr--;
475         }
476         newstate = FMODE_READ|FMODE_WRITE;
477         if (state->n_rdwr == 0) {
478                 if (state->n_rdonly == 0) {
479                         newstate &= ~FMODE_READ;
480                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
481                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
482                 }
483                 if (state->n_wronly == 0) {
484                         newstate &= ~FMODE_WRITE;
485                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
486                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
487                 }
488                 if (newstate == 0)
489                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
490         }
491         nfs4_state_set_mode_locked(state, newstate);
492         spin_unlock(&owner->so_lock);
493
494         if (!call_close) {
495                 nfs4_put_open_state(state);
496                 nfs4_put_state_owner(owner);
497         } else
498                 nfs4_do_close(path, state, wait);
499 }
500
501 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
502 {
503         __nfs4_close(path, state, mode, 0);
504 }
505
506 void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
507 {
508         __nfs4_close(path, state, mode, 1);
509 }
510
511 /*
512  * Search the state->lock_states for an existing lock_owner
513  * that is compatible with current->files
514  */
515 static struct nfs4_lock_state *
516 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
517 {
518         struct nfs4_lock_state *pos;
519         list_for_each_entry(pos, &state->lock_states, ls_locks) {
520                 if (pos->ls_owner != fl_owner)
521                         continue;
522                 atomic_inc(&pos->ls_count);
523                 return pos;
524         }
525         return NULL;
526 }
527
528 /*
529  * Return a compatible lock_state. If no initialized lock_state structure
530  * exists, return an uninitialized one.
531  *
532  */
533 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
534 {
535         struct nfs4_lock_state *lsp;
536         struct nfs_client *clp = state->owner->so_client;
537
538         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
539         if (lsp == NULL)
540                 return NULL;
541         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
542         spin_lock_init(&lsp->ls_sequence.lock);
543         INIT_LIST_HEAD(&lsp->ls_sequence.list);
544         lsp->ls_seqid.sequence = &lsp->ls_sequence;
545         atomic_set(&lsp->ls_count, 1);
546         lsp->ls_owner = fl_owner;
547         spin_lock(&clp->cl_lock);
548         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
549         spin_unlock(&clp->cl_lock);
550         INIT_LIST_HEAD(&lsp->ls_locks);
551         return lsp;
552 }
553
554 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
555 {
556         struct nfs_client *clp = lsp->ls_state->owner->so_client;
557
558         spin_lock(&clp->cl_lock);
559         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
560         spin_unlock(&clp->cl_lock);
561         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
562         kfree(lsp);
563 }
564
565 /*
566  * Return a compatible lock_state. If no initialized lock_state structure
567  * exists, return an uninitialized one.
568  *
569  */
570 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
571 {
572         struct nfs4_lock_state *lsp, *new = NULL;
573         
574         for(;;) {
575                 spin_lock(&state->state_lock);
576                 lsp = __nfs4_find_lock_state(state, owner);
577                 if (lsp != NULL)
578                         break;
579                 if (new != NULL) {
580                         new->ls_state = state;
581                         list_add(&new->ls_locks, &state->lock_states);
582                         set_bit(LK_STATE_IN_USE, &state->flags);
583                         lsp = new;
584                         new = NULL;
585                         break;
586                 }
587                 spin_unlock(&state->state_lock);
588                 new = nfs4_alloc_lock_state(state, owner);
589                 if (new == NULL)
590                         return NULL;
591         }
592         spin_unlock(&state->state_lock);
593         if (new != NULL)
594                 nfs4_free_lock_state(new);
595         return lsp;
596 }
597
598 /*
599  * Release reference to lock_state, and free it if we see that
600  * it is no longer in use
601  */
602 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
603 {
604         struct nfs4_state *state;
605
606         if (lsp == NULL)
607                 return;
608         state = lsp->ls_state;
609         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
610                 return;
611         list_del(&lsp->ls_locks);
612         if (list_empty(&state->lock_states))
613                 clear_bit(LK_STATE_IN_USE, &state->flags);
614         spin_unlock(&state->state_lock);
615         nfs4_free_lock_state(lsp);
616 }
617
618 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
619 {
620         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
621
622         dst->fl_u.nfs4_fl.owner = lsp;
623         atomic_inc(&lsp->ls_count);
624 }
625
626 static void nfs4_fl_release_lock(struct file_lock *fl)
627 {
628         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
629 }
630
631 static struct file_lock_operations nfs4_fl_lock_ops = {
632         .fl_copy_lock = nfs4_fl_copy_lock,
633         .fl_release_private = nfs4_fl_release_lock,
634 };
635
636 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
637 {
638         struct nfs4_lock_state *lsp;
639
640         if (fl->fl_ops != NULL)
641                 return 0;
642         lsp = nfs4_get_lock_state(state, fl->fl_owner);
643         if (lsp == NULL)
644                 return -ENOMEM;
645         fl->fl_u.nfs4_fl.owner = lsp;
646         fl->fl_ops = &nfs4_fl_lock_ops;
647         return 0;
648 }
649
650 /*
651  * Byte-range lock aware utility to initialize the stateid of read/write
652  * requests.
653  */
654 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
655 {
656         struct nfs4_lock_state *lsp;
657         int seq;
658
659         do {
660                 seq = read_seqbegin(&state->seqlock);
661                 memcpy(dst, &state->stateid, sizeof(*dst));
662         } while (read_seqretry(&state->seqlock, seq));
663         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
664                 return;
665
666         spin_lock(&state->state_lock);
667         lsp = __nfs4_find_lock_state(state, fl_owner);
668         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
669                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
670         spin_unlock(&state->state_lock);
671         nfs4_put_lock_state(lsp);
672 }
673
674 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
675 {
676         struct nfs_seqid *new;
677
678         new = kmalloc(sizeof(*new), GFP_KERNEL);
679         if (new != NULL) {
680                 new->sequence = counter;
681                 INIT_LIST_HEAD(&new->list);
682         }
683         return new;
684 }
685
686 void nfs_free_seqid(struct nfs_seqid *seqid)
687 {
688         if (!list_empty(&seqid->list)) {
689                 struct rpc_sequence *sequence = seqid->sequence->sequence;
690
691                 spin_lock(&sequence->lock);
692                 list_del(&seqid->list);
693                 spin_unlock(&sequence->lock);
694                 rpc_wake_up(&sequence->wait);
695         }
696         kfree(seqid);
697 }
698
699 /*
700  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
701  * failed with a seqid incrementing error -
702  * see comments nfs_fs.h:seqid_mutating_error()
703  */
704 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
705 {
706         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
707         switch (status) {
708                 case 0:
709                         break;
710                 case -NFS4ERR_BAD_SEQID:
711                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
712                                 return;
713                         printk(KERN_WARNING "NFS: v4 server returned a bad"
714                                         " sequence-id error on an"
715                                         " unconfirmed sequence %p!\n",
716                                         seqid->sequence);
717                 case -NFS4ERR_STALE_CLIENTID:
718                 case -NFS4ERR_STALE_STATEID:
719                 case -NFS4ERR_BAD_STATEID:
720                 case -NFS4ERR_BADXDR:
721                 case -NFS4ERR_RESOURCE:
722                 case -NFS4ERR_NOFILEHANDLE:
723                         /* Non-seqid mutating errors */
724                         return;
725         };
726         /*
727          * Note: no locking needed as we are guaranteed to be first
728          * on the sequence list
729          */
730         seqid->sequence->counter++;
731 }
732
733 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
734 {
735         if (status == -NFS4ERR_BAD_SEQID) {
736                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
737                                 struct nfs4_state_owner, so_seqid);
738                 nfs4_drop_state_owner(sp);
739         }
740         nfs_increment_seqid(status, seqid);
741 }
742
743 /*
744  * Increment the seqid if the LOCK/LOCKU succeeded, or
745  * failed with a seqid incrementing error -
746  * see comments nfs_fs.h:seqid_mutating_error()
747  */
748 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
749 {
750         nfs_increment_seqid(status, seqid);
751 }
752
753 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
754 {
755         struct rpc_sequence *sequence = seqid->sequence->sequence;
756         int status = 0;
757
758         spin_lock(&sequence->lock);
759         if (list_empty(&seqid->list))
760                 list_add_tail(&seqid->list, &sequence->list);
761         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
762                 goto unlock;
763         rpc_sleep_on(&sequence->wait, task, NULL);
764         status = -EAGAIN;
765 unlock:
766         spin_unlock(&sequence->lock);
767         return status;
768 }
769
770 static int reclaimer(void *);
771
772 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
773 {
774         smp_mb__before_clear_bit();
775         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
776         smp_mb__after_clear_bit();
777         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
778         rpc_wake_up(&clp->cl_rpcwaitq);
779 }
780
781 /*
782  * State recovery routine
783  */
784 static void nfs4_recover_state(struct nfs_client *clp)
785 {
786         struct task_struct *task;
787
788         __module_get(THIS_MODULE);
789         atomic_inc(&clp->cl_count);
790         task = kthread_run(reclaimer, clp, "%s-reclaim",
791                                 rpc_peeraddr2str(clp->cl_rpcclient,
792                                                         RPC_DISPLAY_ADDR));
793         if (!IS_ERR(task))
794                 return;
795         nfs4_clear_recover_bit(clp);
796         nfs_put_client(clp);
797         module_put(THIS_MODULE);
798 }
799
800 /*
801  * Schedule a state recovery attempt
802  */
803 void nfs4_schedule_state_recovery(struct nfs_client *clp)
804 {
805         if (!clp)
806                 return;
807         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
808                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
809         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
810                 nfs4_recover_state(clp);
811 }
812
813 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
814 {
815
816         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
817         /* Don't recover state that expired before the reboot */
818         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
819                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
820                 return 0;
821         }
822         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
823         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
824         return 1;
825 }
826
827 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
828 {
829         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
830         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
831         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
832         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
833         return 1;
834 }
835
836 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
837 {
838         struct inode *inode = state->inode;
839         struct nfs_inode *nfsi = NFS_I(inode);
840         struct file_lock *fl;
841         int status = 0;
842
843         down_write(&nfsi->rwsem);
844         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
845                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
846                         continue;
847                 if (nfs_file_open_context(fl->fl_file)->state != state)
848                         continue;
849                 status = ops->recover_lock(state, fl);
850                 if (status >= 0)
851                         continue;
852                 switch (status) {
853                         default:
854                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
855                                                 __func__, status);
856                         case -NFS4ERR_EXPIRED:
857                         case -NFS4ERR_NO_GRACE:
858                         case -NFS4ERR_RECLAIM_BAD:
859                         case -NFS4ERR_RECLAIM_CONFLICT:
860                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
861                                 break;
862                         case -NFS4ERR_STALE_CLIENTID:
863                                 goto out_err;
864                 }
865         }
866         up_write(&nfsi->rwsem);
867         return 0;
868 out_err:
869         up_write(&nfsi->rwsem);
870         return status;
871 }
872
873 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
874 {
875         struct nfs4_state *state;
876         struct nfs4_lock_state *lock;
877         int status = 0;
878
879         /* Note: we rely on the sp->so_states list being ordered 
880          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
881          * states first.
882          * This is needed to ensure that the server won't give us any
883          * read delegations that we have to return if, say, we are
884          * recovering after a network partition or a reboot from a
885          * server that doesn't support a grace period.
886          */
887 restart:
888         spin_lock(&sp->so_lock);
889         list_for_each_entry(state, &sp->so_states, open_states) {
890                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
891                         continue;
892                 if (state->state == 0)
893                         continue;
894                 atomic_inc(&state->count);
895                 spin_unlock(&sp->so_lock);
896                 status = ops->recover_open(sp, state);
897                 if (status >= 0) {
898                         status = nfs4_reclaim_locks(state, ops);
899                         if (status >= 0) {
900                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
901                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
902                                                 printk("%s: Lock reclaim failed!\n",
903                                                         __func__);
904                                 }
905                                 nfs4_put_open_state(state);
906                                 goto restart;
907                         }
908                 }
909                 switch (status) {
910                         default:
911                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
912                                                 __func__, status);
913                         case -ENOENT:
914                         case -ESTALE:
915                                 /*
916                                  * Open state on this file cannot be recovered
917                                  * All we can do is revert to using the zero stateid.
918                                  */
919                                 memset(state->stateid.data, 0,
920                                         sizeof(state->stateid.data));
921                                 /* Mark the file as being 'closed' */
922                                 state->state = 0;
923                                 break;
924                         case -NFS4ERR_RECLAIM_BAD:
925                         case -NFS4ERR_RECLAIM_CONFLICT:
926                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
927                                 break;
928                         case -NFS4ERR_EXPIRED:
929                         case -NFS4ERR_NO_GRACE:
930                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
931                         case -NFS4ERR_STALE_CLIENTID:
932                                 goto out_err;
933                 }
934                 nfs4_put_open_state(state);
935                 goto restart;
936         }
937         spin_unlock(&sp->so_lock);
938         return 0;
939 out_err:
940         nfs4_put_open_state(state);
941         return status;
942 }
943
944 static void nfs4_clear_open_state(struct nfs4_state *state)
945 {
946         struct nfs4_lock_state *lock;
947
948         clear_bit(NFS_DELEGATED_STATE, &state->flags);
949         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
950         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
951         clear_bit(NFS_O_RDWR_STATE, &state->flags);
952         list_for_each_entry(lock, &state->lock_states, ls_locks) {
953                 lock->ls_seqid.counter = 0;
954                 lock->ls_seqid.flags = 0;
955                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
956         }
957 }
958
959 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
960 {
961         struct nfs4_state_owner *sp;
962         struct rb_node *pos;
963         struct nfs4_state *state;
964
965         /* Reset all sequence ids to zero */
966         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
967                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
968                 sp->so_seqid.counter = 0;
969                 sp->so_seqid.flags = 0;
970                 spin_lock(&sp->so_lock);
971                 list_for_each_entry(state, &sp->so_states, open_states) {
972                         if (mark_reclaim(clp, state))
973                                 nfs4_clear_open_state(state);
974                 }
975                 spin_unlock(&sp->so_lock);
976         }
977 }
978
979 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
980 {
981         /* Mark all delegations for reclaim */
982         nfs_delegation_mark_reclaim(clp);
983         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
984 }
985
986 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
987 {
988         struct nfs4_state_owner *sp;
989         struct rb_node *pos;
990         struct nfs4_state *state;
991
992         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
993                 return;
994
995         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
996                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
997                 spin_lock(&sp->so_lock);
998                 list_for_each_entry(state, &sp->so_states, open_states) {
999                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1000                                 continue;
1001                         nfs4_state_mark_reclaim_nograce(clp, state);
1002                 }
1003                 spin_unlock(&sp->so_lock);
1004         }
1005
1006         nfs_delegation_reap_unclaimed(clp);
1007 }
1008
1009 static void nfs_delegation_clear_all(struct nfs_client *clp)
1010 {
1011         nfs_delegation_mark_reclaim(clp);
1012         nfs_delegation_reap_unclaimed(clp);
1013 }
1014
1015 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1016 {
1017         nfs_delegation_clear_all(clp);
1018         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1019 }
1020
1021 static void nfs4_state_end_reclaim_nograce(struct nfs_client *clp)
1022 {
1023         clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1024 }
1025
1026 static void nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1027 {
1028         switch (error) {
1029                 case -NFS4ERR_CB_PATH_DOWN:
1030                         nfs_handle_cb_pathdown(clp);
1031                         break;
1032                 case -NFS4ERR_STALE_CLIENTID:
1033                 case -NFS4ERR_LEASE_MOVED:
1034                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1035                         nfs4_state_start_reclaim_reboot(clp);
1036                         break;
1037                 case -NFS4ERR_EXPIRED:
1038                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1039                         nfs4_state_start_reclaim_nograce(clp);
1040         }
1041 }
1042
1043 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1044 {
1045         struct rb_node *pos;
1046         int status = 0;
1047
1048 restart:
1049         spin_lock(&clp->cl_lock);
1050         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1051                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1052                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1053                         continue;
1054                 atomic_inc(&sp->so_count);
1055                 spin_unlock(&clp->cl_lock);
1056                 status = nfs4_reclaim_open_state(sp, ops);
1057                 if (status < 0) {
1058                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1059                         nfs4_put_state_owner(sp);
1060                         nfs4_recovery_handle_error(clp, status);
1061                         return status;
1062                 }
1063                 nfs4_put_state_owner(sp);
1064                 goto restart;
1065         }
1066         spin_unlock(&clp->cl_lock);
1067         return status;
1068 }
1069
1070 static int nfs4_check_lease(struct nfs_client *clp)
1071 {
1072         struct rpc_cred *cred;
1073         int status = -NFS4ERR_EXPIRED;
1074
1075         /* Is the client already known to have an expired lease? */
1076         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1077                 return 0;
1078         cred = nfs4_get_renew_cred(clp);
1079         if (cred == NULL) {
1080                 cred = nfs4_get_setclientid_cred(clp);
1081                 if (cred == NULL)
1082                         goto out;
1083         }
1084         status = nfs4_proc_renew(clp, cred);
1085         put_rpccred(cred);
1086 out:
1087         nfs4_recovery_handle_error(clp, status);
1088         return status;
1089 }
1090
1091 static int nfs4_reclaim_lease(struct nfs_client *clp)
1092 {
1093         struct rpc_cred *cred;
1094         int status = -ENOENT;
1095
1096         cred = nfs4_get_setclientid_cred(clp);
1097         if (cred != NULL) {
1098                 status = nfs4_init_client(clp, cred);
1099                 put_rpccred(cred);
1100                 /* Handle case where the user hasn't set up machine creds */
1101                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1102                         nfs4_clear_machine_cred(clp);
1103                         status = -EAGAIN;
1104                 }
1105         }
1106         return status;
1107 }
1108
1109 static int reclaimer(void *ptr)
1110 {
1111         struct nfs_client *clp = ptr;
1112         int status = 0;
1113
1114         allow_signal(SIGKILL);
1115
1116         /* Ensure exclusive access to NFSv4 state */
1117         while (!list_empty(&clp->cl_superblocks)) {
1118                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1119                         /* We're going to have to re-establish a clientid */
1120                         status = nfs4_reclaim_lease(clp);
1121                         if (status) {
1122                                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1123                                 if (status == -EAGAIN)
1124                                         continue;
1125                                 goto out_error;
1126                         }
1127                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1128                 }
1129
1130                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1131                         status = nfs4_check_lease(clp);
1132                         if (status != 0)
1133                                 continue;
1134                 }
1135
1136                 /* First recover reboot state... */
1137                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1138                         status = nfs4_do_reclaim(clp, &nfs4_reboot_recovery_ops);
1139                         if (status == -NFS4ERR_STALE_CLIENTID)
1140                                 continue;
1141                         nfs4_state_end_reclaim_reboot(clp);
1142                         continue;
1143                 }
1144
1145                 /* Now recover expired state... */
1146                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1147                         status = nfs4_do_reclaim(clp, &nfs4_nograce_recovery_ops);
1148                         if (status < 0) {
1149                                 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1150                                 if (status == -NFS4ERR_STALE_CLIENTID)
1151                                         continue;
1152                                 if (status == -NFS4ERR_EXPIRED)
1153                                         continue;
1154                                 goto out_error;
1155                         } else
1156                                 nfs4_state_end_reclaim_nograce(clp);
1157                         continue;
1158                 }
1159
1160                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1161                         nfs_client_return_marked_delegations(clp);
1162                         continue;
1163                 }
1164                 break;
1165         }
1166 out:
1167         nfs4_clear_recover_bit(clp);
1168         nfs_put_client(clp);
1169         module_put_and_exit(0);
1170         return 0;
1171 out_error:
1172         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %s"
1173                         " with error %d\n", clp->cl_hostname, -status);
1174         if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1175                 nfs4_state_end_reclaim_reboot(clp);
1176         goto out;
1177 }
1178
1179 /*
1180  * Local variables:
1181  *  c-basic-offset: 8
1182  * End:
1183  */