nfs41: add session setup to the state manager
[pandora-kernel.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/smp_lock.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/workqueue.h>
50 #include <linux/bitops.h>
51
52 #include "nfs4_fs.h"
53 #include "callback.h"
54 #include "delegation.h"
55 #include "internal.h"
56
57 #define OPENOWNER_POOL_SIZE     8
58
59 const nfs4_stateid zero_stateid;
60
61 static LIST_HEAD(nfs4_clientid_list);
62
63 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
64 {
65         unsigned short port;
66         int status;
67
68         port = nfs_callback_tcpport;
69         if (clp->cl_addr.ss_family == AF_INET6)
70                 port = nfs_callback_tcpport6;
71
72         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred);
73         if (status == 0)
74                 status = nfs4_proc_setclientid_confirm(clp, cred);
75         if (status == 0)
76                 nfs4_schedule_state_renewal(clp);
77         return status;
78 }
79
80 static struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
81 {
82         struct rpc_cred *cred = NULL;
83
84         if (clp->cl_machine_cred != NULL)
85                 cred = get_rpccred(clp->cl_machine_cred);
86         return cred;
87 }
88
89 static void nfs4_clear_machine_cred(struct nfs_client *clp)
90 {
91         struct rpc_cred *cred;
92
93         spin_lock(&clp->cl_lock);
94         cred = clp->cl_machine_cred;
95         clp->cl_machine_cred = NULL;
96         spin_unlock(&clp->cl_lock);
97         if (cred != NULL)
98                 put_rpccred(cred);
99 }
100
101 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
102 {
103         struct nfs4_state_owner *sp;
104         struct rb_node *pos;
105         struct rpc_cred *cred = NULL;
106
107         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
108                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
109                 if (list_empty(&sp->so_states))
110                         continue;
111                 cred = get_rpccred(sp->so_cred);
112                 break;
113         }
114         return cred;
115 }
116
117 static struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
118 {
119         struct rpc_cred *cred;
120
121         spin_lock(&clp->cl_lock);
122         cred = nfs4_get_renew_cred_locked(clp);
123         spin_unlock(&clp->cl_lock);
124         return cred;
125 }
126
127 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
128 {
129         struct nfs4_state_owner *sp;
130         struct rb_node *pos;
131         struct rpc_cred *cred;
132
133         spin_lock(&clp->cl_lock);
134         cred = nfs4_get_machine_cred_locked(clp);
135         if (cred != NULL)
136                 goto out;
137         pos = rb_first(&clp->cl_state_owners);
138         if (pos != NULL) {
139                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
140                 cred = get_rpccred(sp->so_cred);
141         }
142 out:
143         spin_unlock(&clp->cl_lock);
144         return cred;
145 }
146
147 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
148                 __u64 minval, int maxbits)
149 {
150         struct rb_node **p, *parent;
151         struct nfs_unique_id *pos;
152         __u64 mask = ~0ULL;
153
154         if (maxbits < 64)
155                 mask = (1ULL << maxbits) - 1ULL;
156
157         /* Ensure distribution is more or less flat */
158         get_random_bytes(&new->id, sizeof(new->id));
159         new->id &= mask;
160         if (new->id < minval)
161                 new->id += minval;
162 retry:
163         p = &root->rb_node;
164         parent = NULL;
165
166         while (*p != NULL) {
167                 parent = *p;
168                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
169
170                 if (new->id < pos->id)
171                         p = &(*p)->rb_left;
172                 else if (new->id > pos->id)
173                         p = &(*p)->rb_right;
174                 else
175                         goto id_exists;
176         }
177         rb_link_node(&new->rb_node, parent, p);
178         rb_insert_color(&new->rb_node, root);
179         return;
180 id_exists:
181         for (;;) {
182                 new->id++;
183                 if (new->id < minval || (new->id & mask) != new->id) {
184                         new->id = minval;
185                         break;
186                 }
187                 parent = rb_next(parent);
188                 if (parent == NULL)
189                         break;
190                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
191                 if (new->id < pos->id)
192                         break;
193         }
194         goto retry;
195 }
196
197 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
198 {
199         rb_erase(&id->rb_node, root);
200 }
201
202 static struct nfs4_state_owner *
203 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
204 {
205         struct nfs_client *clp = server->nfs_client;
206         struct rb_node **p = &clp->cl_state_owners.rb_node,
207                        *parent = NULL;
208         struct nfs4_state_owner *sp, *res = NULL;
209
210         while (*p != NULL) {
211                 parent = *p;
212                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
213
214                 if (server < sp->so_server) {
215                         p = &parent->rb_left;
216                         continue;
217                 }
218                 if (server > sp->so_server) {
219                         p = &parent->rb_right;
220                         continue;
221                 }
222                 if (cred < sp->so_cred)
223                         p = &parent->rb_left;
224                 else if (cred > sp->so_cred)
225                         p = &parent->rb_right;
226                 else {
227                         atomic_inc(&sp->so_count);
228                         res = sp;
229                         break;
230                 }
231         }
232         return res;
233 }
234
235 static struct nfs4_state_owner *
236 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
237 {
238         struct rb_node **p = &clp->cl_state_owners.rb_node,
239                        *parent = NULL;
240         struct nfs4_state_owner *sp;
241
242         while (*p != NULL) {
243                 parent = *p;
244                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
245
246                 if (new->so_server < sp->so_server) {
247                         p = &parent->rb_left;
248                         continue;
249                 }
250                 if (new->so_server > sp->so_server) {
251                         p = &parent->rb_right;
252                         continue;
253                 }
254                 if (new->so_cred < sp->so_cred)
255                         p = &parent->rb_left;
256                 else if (new->so_cred > sp->so_cred)
257                         p = &parent->rb_right;
258                 else {
259                         atomic_inc(&sp->so_count);
260                         return sp;
261                 }
262         }
263         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
264         rb_link_node(&new->so_client_node, parent, p);
265         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
266         return new;
267 }
268
269 static void
270 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
271 {
272         if (!RB_EMPTY_NODE(&sp->so_client_node))
273                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
274         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
275 }
276
277 /*
278  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
279  * create a new state_owner.
280  *
281  */
282 static struct nfs4_state_owner *
283 nfs4_alloc_state_owner(void)
284 {
285         struct nfs4_state_owner *sp;
286
287         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
288         if (!sp)
289                 return NULL;
290         spin_lock_init(&sp->so_lock);
291         INIT_LIST_HEAD(&sp->so_states);
292         INIT_LIST_HEAD(&sp->so_delegations);
293         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
294         sp->so_seqid.sequence = &sp->so_sequence;
295         spin_lock_init(&sp->so_sequence.lock);
296         INIT_LIST_HEAD(&sp->so_sequence.list);
297         atomic_set(&sp->so_count, 1);
298         return sp;
299 }
300
301 static void
302 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
303 {
304         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
305                 struct nfs_client *clp = sp->so_client;
306
307                 spin_lock(&clp->cl_lock);
308                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
309                 RB_CLEAR_NODE(&sp->so_client_node);
310                 spin_unlock(&clp->cl_lock);
311         }
312 }
313
314 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
315 {
316         struct nfs_client *clp = server->nfs_client;
317         struct nfs4_state_owner *sp, *new;
318
319         spin_lock(&clp->cl_lock);
320         sp = nfs4_find_state_owner(server, cred);
321         spin_unlock(&clp->cl_lock);
322         if (sp != NULL)
323                 return sp;
324         new = nfs4_alloc_state_owner();
325         if (new == NULL)
326                 return NULL;
327         new->so_client = clp;
328         new->so_server = server;
329         new->so_cred = cred;
330         spin_lock(&clp->cl_lock);
331         sp = nfs4_insert_state_owner(clp, new);
332         spin_unlock(&clp->cl_lock);
333         if (sp == new)
334                 get_rpccred(cred);
335         else {
336                 rpc_destroy_wait_queue(&new->so_sequence.wait);
337                 kfree(new);
338         }
339         return sp;
340 }
341
342 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
343 {
344         struct nfs_client *clp = sp->so_client;
345         struct rpc_cred *cred = sp->so_cred;
346
347         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
348                 return;
349         nfs4_remove_state_owner(clp, sp);
350         spin_unlock(&clp->cl_lock);
351         rpc_destroy_wait_queue(&sp->so_sequence.wait);
352         put_rpccred(cred);
353         kfree(sp);
354 }
355
356 static struct nfs4_state *
357 nfs4_alloc_open_state(void)
358 {
359         struct nfs4_state *state;
360
361         state = kzalloc(sizeof(*state), GFP_KERNEL);
362         if (!state)
363                 return NULL;
364         atomic_set(&state->count, 1);
365         INIT_LIST_HEAD(&state->lock_states);
366         spin_lock_init(&state->state_lock);
367         seqlock_init(&state->seqlock);
368         return state;
369 }
370
371 void
372 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
373 {
374         if (state->state == fmode)
375                 return;
376         /* NB! List reordering - see the reclaim code for why.  */
377         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
378                 if (fmode & FMODE_WRITE)
379                         list_move(&state->open_states, &state->owner->so_states);
380                 else
381                         list_move_tail(&state->open_states, &state->owner->so_states);
382         }
383         state->state = fmode;
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                 if (state->owner != owner)
394                         continue;
395                 if (atomic_inc_not_zero(&state->count))
396                         return state;
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 void nfs4_put_open_state(struct nfs4_state *state)
444 {
445         struct inode *inode = state->inode;
446         struct nfs4_state_owner *owner = state->owner;
447
448         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
449                 return;
450         spin_lock(&inode->i_lock);
451         list_del(&state->inode_states);
452         list_del(&state->open_states);
453         spin_unlock(&inode->i_lock);
454         spin_unlock(&owner->so_lock);
455         iput(inode);
456         nfs4_free_open_state(state);
457         nfs4_put_state_owner(owner);
458 }
459
460 /*
461  * Close the current file.
462  */
463 static void __nfs4_close(struct path *path, struct nfs4_state *state, fmode_t fmode, int wait)
464 {
465         struct nfs4_state_owner *owner = state->owner;
466         int call_close = 0;
467         fmode_t newstate;
468
469         atomic_inc(&owner->so_count);
470         /* Protect against nfs4_find_state() */
471         spin_lock(&owner->so_lock);
472         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
473                 case FMODE_READ:
474                         state->n_rdonly--;
475                         break;
476                 case FMODE_WRITE:
477                         state->n_wronly--;
478                         break;
479                 case FMODE_READ|FMODE_WRITE:
480                         state->n_rdwr--;
481         }
482         newstate = FMODE_READ|FMODE_WRITE;
483         if (state->n_rdwr == 0) {
484                 if (state->n_rdonly == 0) {
485                         newstate &= ~FMODE_READ;
486                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
487                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
488                 }
489                 if (state->n_wronly == 0) {
490                         newstate &= ~FMODE_WRITE;
491                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
492                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
493                 }
494                 if (newstate == 0)
495                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
496         }
497         nfs4_state_set_mode_locked(state, newstate);
498         spin_unlock(&owner->so_lock);
499
500         if (!call_close) {
501                 nfs4_put_open_state(state);
502                 nfs4_put_state_owner(owner);
503         } else
504                 nfs4_do_close(path, state, wait);
505 }
506
507 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
508 {
509         __nfs4_close(path, state, fmode, 0);
510 }
511
512 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
513 {
514         __nfs4_close(path, state, fmode, 1);
515 }
516
517 /*
518  * Search the state->lock_states for an existing lock_owner
519  * that is compatible with current->files
520  */
521 static struct nfs4_lock_state *
522 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
523 {
524         struct nfs4_lock_state *pos;
525         list_for_each_entry(pos, &state->lock_states, ls_locks) {
526                 if (pos->ls_owner != fl_owner)
527                         continue;
528                 atomic_inc(&pos->ls_count);
529                 return pos;
530         }
531         return NULL;
532 }
533
534 /*
535  * Return a compatible lock_state. If no initialized lock_state structure
536  * exists, return an uninitialized one.
537  *
538  */
539 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
540 {
541         struct nfs4_lock_state *lsp;
542         struct nfs_client *clp = state->owner->so_client;
543
544         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
545         if (lsp == NULL)
546                 return NULL;
547         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
548         spin_lock_init(&lsp->ls_sequence.lock);
549         INIT_LIST_HEAD(&lsp->ls_sequence.list);
550         lsp->ls_seqid.sequence = &lsp->ls_sequence;
551         atomic_set(&lsp->ls_count, 1);
552         lsp->ls_owner = fl_owner;
553         spin_lock(&clp->cl_lock);
554         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
555         spin_unlock(&clp->cl_lock);
556         INIT_LIST_HEAD(&lsp->ls_locks);
557         return lsp;
558 }
559
560 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
561 {
562         struct nfs_client *clp = lsp->ls_state->owner->so_client;
563
564         spin_lock(&clp->cl_lock);
565         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
566         spin_unlock(&clp->cl_lock);
567         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
568         kfree(lsp);
569 }
570
571 /*
572  * Return a compatible lock_state. If no initialized lock_state structure
573  * exists, return an uninitialized one.
574  *
575  */
576 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
577 {
578         struct nfs4_lock_state *lsp, *new = NULL;
579         
580         for(;;) {
581                 spin_lock(&state->state_lock);
582                 lsp = __nfs4_find_lock_state(state, owner);
583                 if (lsp != NULL)
584                         break;
585                 if (new != NULL) {
586                         new->ls_state = state;
587                         list_add(&new->ls_locks, &state->lock_states);
588                         set_bit(LK_STATE_IN_USE, &state->flags);
589                         lsp = new;
590                         new = NULL;
591                         break;
592                 }
593                 spin_unlock(&state->state_lock);
594                 new = nfs4_alloc_lock_state(state, owner);
595                 if (new == NULL)
596                         return NULL;
597         }
598         spin_unlock(&state->state_lock);
599         if (new != NULL)
600                 nfs4_free_lock_state(new);
601         return lsp;
602 }
603
604 /*
605  * Release reference to lock_state, and free it if we see that
606  * it is no longer in use
607  */
608 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
609 {
610         struct nfs4_state *state;
611
612         if (lsp == NULL)
613                 return;
614         state = lsp->ls_state;
615         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
616                 return;
617         list_del(&lsp->ls_locks);
618         if (list_empty(&state->lock_states))
619                 clear_bit(LK_STATE_IN_USE, &state->flags);
620         spin_unlock(&state->state_lock);
621         nfs4_free_lock_state(lsp);
622 }
623
624 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
625 {
626         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
627
628         dst->fl_u.nfs4_fl.owner = lsp;
629         atomic_inc(&lsp->ls_count);
630 }
631
632 static void nfs4_fl_release_lock(struct file_lock *fl)
633 {
634         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
635 }
636
637 static struct file_lock_operations nfs4_fl_lock_ops = {
638         .fl_copy_lock = nfs4_fl_copy_lock,
639         .fl_release_private = nfs4_fl_release_lock,
640 };
641
642 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
643 {
644         struct nfs4_lock_state *lsp;
645
646         if (fl->fl_ops != NULL)
647                 return 0;
648         lsp = nfs4_get_lock_state(state, fl->fl_owner);
649         if (lsp == NULL)
650                 return -ENOMEM;
651         fl->fl_u.nfs4_fl.owner = lsp;
652         fl->fl_ops = &nfs4_fl_lock_ops;
653         return 0;
654 }
655
656 /*
657  * Byte-range lock aware utility to initialize the stateid of read/write
658  * requests.
659  */
660 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
661 {
662         struct nfs4_lock_state *lsp;
663         int seq;
664
665         do {
666                 seq = read_seqbegin(&state->seqlock);
667                 memcpy(dst, &state->stateid, sizeof(*dst));
668         } while (read_seqretry(&state->seqlock, seq));
669         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
670                 return;
671
672         spin_lock(&state->state_lock);
673         lsp = __nfs4_find_lock_state(state, fl_owner);
674         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
675                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
676         spin_unlock(&state->state_lock);
677         nfs4_put_lock_state(lsp);
678 }
679
680 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
681 {
682         struct nfs_seqid *new;
683
684         new = kmalloc(sizeof(*new), GFP_KERNEL);
685         if (new != NULL) {
686                 new->sequence = counter;
687                 INIT_LIST_HEAD(&new->list);
688         }
689         return new;
690 }
691
692 void nfs_free_seqid(struct nfs_seqid *seqid)
693 {
694         if (!list_empty(&seqid->list)) {
695                 struct rpc_sequence *sequence = seqid->sequence->sequence;
696
697                 spin_lock(&sequence->lock);
698                 list_del(&seqid->list);
699                 spin_unlock(&sequence->lock);
700                 rpc_wake_up(&sequence->wait);
701         }
702         kfree(seqid);
703 }
704
705 /*
706  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
707  * failed with a seqid incrementing error -
708  * see comments nfs_fs.h:seqid_mutating_error()
709  */
710 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
711 {
712         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
713         switch (status) {
714                 case 0:
715                         break;
716                 case -NFS4ERR_BAD_SEQID:
717                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
718                                 return;
719                         printk(KERN_WARNING "NFS: v4 server returned a bad"
720                                         " sequence-id error on an"
721                                         " unconfirmed sequence %p!\n",
722                                         seqid->sequence);
723                 case -NFS4ERR_STALE_CLIENTID:
724                 case -NFS4ERR_STALE_STATEID:
725                 case -NFS4ERR_BAD_STATEID:
726                 case -NFS4ERR_BADXDR:
727                 case -NFS4ERR_RESOURCE:
728                 case -NFS4ERR_NOFILEHANDLE:
729                         /* Non-seqid mutating errors */
730                         return;
731         };
732         /*
733          * Note: no locking needed as we are guaranteed to be first
734          * on the sequence list
735          */
736         seqid->sequence->counter++;
737 }
738
739 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
740 {
741         if (status == -NFS4ERR_BAD_SEQID) {
742                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
743                                 struct nfs4_state_owner, so_seqid);
744                 nfs4_drop_state_owner(sp);
745         }
746         nfs_increment_seqid(status, seqid);
747 }
748
749 /*
750  * Increment the seqid if the LOCK/LOCKU succeeded, or
751  * failed with a seqid incrementing error -
752  * see comments nfs_fs.h:seqid_mutating_error()
753  */
754 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
755 {
756         nfs_increment_seqid(status, seqid);
757 }
758
759 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
760 {
761         struct rpc_sequence *sequence = seqid->sequence->sequence;
762         int status = 0;
763
764         spin_lock(&sequence->lock);
765         if (list_empty(&seqid->list))
766                 list_add_tail(&seqid->list, &sequence->list);
767         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
768                 goto unlock;
769         rpc_sleep_on(&sequence->wait, task, NULL);
770         status = -EAGAIN;
771 unlock:
772         spin_unlock(&sequence->lock);
773         return status;
774 }
775
776 static int nfs4_run_state_manager(void *);
777
778 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
779 {
780         smp_mb__before_clear_bit();
781         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
782         smp_mb__after_clear_bit();
783         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
784         rpc_wake_up(&clp->cl_rpcwaitq);
785 }
786
787 /*
788  * Schedule the nfs_client asynchronous state management routine
789  */
790 void nfs4_schedule_state_manager(struct nfs_client *clp)
791 {
792         struct task_struct *task;
793
794         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
795                 return;
796         __module_get(THIS_MODULE);
797         atomic_inc(&clp->cl_count);
798         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
799                                 rpc_peeraddr2str(clp->cl_rpcclient,
800                                                         RPC_DISPLAY_ADDR));
801         if (!IS_ERR(task))
802                 return;
803         nfs4_clear_state_manager_bit(clp);
804         nfs_put_client(clp);
805         module_put(THIS_MODULE);
806 }
807
808 /*
809  * Schedule a state recovery attempt
810  */
811 void nfs4_schedule_state_recovery(struct nfs_client *clp)
812 {
813         if (!clp)
814                 return;
815         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
816                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
817         nfs4_schedule_state_manager(clp);
818 }
819
820 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
821 {
822
823         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
824         /* Don't recover state that expired before the reboot */
825         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
826                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
827                 return 0;
828         }
829         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
830         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
831         return 1;
832 }
833
834 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
835 {
836         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
837         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
838         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
839         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
840         return 1;
841 }
842
843 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
844 {
845         struct inode *inode = state->inode;
846         struct nfs_inode *nfsi = NFS_I(inode);
847         struct file_lock *fl;
848         int status = 0;
849
850         down_write(&nfsi->rwsem);
851         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
852                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
853                         continue;
854                 if (nfs_file_open_context(fl->fl_file)->state != state)
855                         continue;
856                 status = ops->recover_lock(state, fl);
857                 if (status >= 0)
858                         continue;
859                 switch (status) {
860                         default:
861                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
862                                                 __func__, status);
863                         case -NFS4ERR_EXPIRED:
864                         case -NFS4ERR_NO_GRACE:
865                         case -NFS4ERR_RECLAIM_BAD:
866                         case -NFS4ERR_RECLAIM_CONFLICT:
867                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
868                                 break;
869                         case -NFS4ERR_STALE_CLIENTID:
870                                 goto out_err;
871                 }
872         }
873         up_write(&nfsi->rwsem);
874         return 0;
875 out_err:
876         up_write(&nfsi->rwsem);
877         return status;
878 }
879
880 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
881 {
882         struct nfs4_state *state;
883         struct nfs4_lock_state *lock;
884         int status = 0;
885
886         /* Note: we rely on the sp->so_states list being ordered 
887          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
888          * states first.
889          * This is needed to ensure that the server won't give us any
890          * read delegations that we have to return if, say, we are
891          * recovering after a network partition or a reboot from a
892          * server that doesn't support a grace period.
893          */
894 restart:
895         spin_lock(&sp->so_lock);
896         list_for_each_entry(state, &sp->so_states, open_states) {
897                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
898                         continue;
899                 if (state->state == 0)
900                         continue;
901                 atomic_inc(&state->count);
902                 spin_unlock(&sp->so_lock);
903                 status = ops->recover_open(sp, state);
904                 if (status >= 0) {
905                         status = nfs4_reclaim_locks(state, ops);
906                         if (status >= 0) {
907                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
908                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
909                                                 printk("%s: Lock reclaim failed!\n",
910                                                         __func__);
911                                 }
912                                 nfs4_put_open_state(state);
913                                 goto restart;
914                         }
915                 }
916                 switch (status) {
917                         default:
918                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
919                                                 __func__, status);
920                         case -ENOENT:
921                         case -ESTALE:
922                                 /*
923                                  * Open state on this file cannot be recovered
924                                  * All we can do is revert to using the zero stateid.
925                                  */
926                                 memset(state->stateid.data, 0,
927                                         sizeof(state->stateid.data));
928                                 /* Mark the file as being 'closed' */
929                                 state->state = 0;
930                                 break;
931                         case -NFS4ERR_RECLAIM_BAD:
932                         case -NFS4ERR_RECLAIM_CONFLICT:
933                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
934                                 break;
935                         case -NFS4ERR_EXPIRED:
936                         case -NFS4ERR_NO_GRACE:
937                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
938                         case -NFS4ERR_STALE_CLIENTID:
939                                 goto out_err;
940                 }
941                 nfs4_put_open_state(state);
942                 goto restart;
943         }
944         spin_unlock(&sp->so_lock);
945         return 0;
946 out_err:
947         nfs4_put_open_state(state);
948         return status;
949 }
950
951 static void nfs4_clear_open_state(struct nfs4_state *state)
952 {
953         struct nfs4_lock_state *lock;
954
955         clear_bit(NFS_DELEGATED_STATE, &state->flags);
956         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
957         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
958         clear_bit(NFS_O_RDWR_STATE, &state->flags);
959         list_for_each_entry(lock, &state->lock_states, ls_locks) {
960                 lock->ls_seqid.flags = 0;
961                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
962         }
963 }
964
965 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
966 {
967         struct nfs4_state_owner *sp;
968         struct rb_node *pos;
969         struct nfs4_state *state;
970
971         /* Reset all sequence ids to zero */
972         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
973                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
974                 sp->so_seqid.flags = 0;
975                 spin_lock(&sp->so_lock);
976                 list_for_each_entry(state, &sp->so_states, open_states) {
977                         if (mark_reclaim(clp, state))
978                                 nfs4_clear_open_state(state);
979                 }
980                 spin_unlock(&sp->so_lock);
981         }
982 }
983
984 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
985 {
986         /* Mark all delegations for reclaim */
987         nfs_delegation_mark_reclaim(clp);
988         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
989 }
990
991 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
992 {
993         struct nfs4_state_owner *sp;
994         struct rb_node *pos;
995         struct nfs4_state *state;
996
997         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
998                 return;
999
1000         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1001                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1002                 spin_lock(&sp->so_lock);
1003                 list_for_each_entry(state, &sp->so_states, open_states) {
1004                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1005                                 continue;
1006                         nfs4_state_mark_reclaim_nograce(clp, state);
1007                 }
1008                 spin_unlock(&sp->so_lock);
1009         }
1010
1011         nfs_delegation_reap_unclaimed(clp);
1012 }
1013
1014 static void nfs_delegation_clear_all(struct nfs_client *clp)
1015 {
1016         nfs_delegation_mark_reclaim(clp);
1017         nfs_delegation_reap_unclaimed(clp);
1018 }
1019
1020 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1021 {
1022         nfs_delegation_clear_all(clp);
1023         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1024 }
1025
1026 static void nfs4_state_end_reclaim_nograce(struct nfs_client *clp)
1027 {
1028         clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1029 }
1030
1031 static void nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1032 {
1033         switch (error) {
1034                 case -NFS4ERR_CB_PATH_DOWN:
1035                         nfs_handle_cb_pathdown(clp);
1036                         break;
1037                 case -NFS4ERR_STALE_CLIENTID:
1038                 case -NFS4ERR_LEASE_MOVED:
1039                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1040                         nfs4_state_start_reclaim_reboot(clp);
1041                         break;
1042                 case -NFS4ERR_EXPIRED:
1043                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1044                         nfs4_state_start_reclaim_nograce(clp);
1045         }
1046 }
1047
1048 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1049 {
1050         struct rb_node *pos;
1051         int status = 0;
1052
1053 restart:
1054         spin_lock(&clp->cl_lock);
1055         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1056                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1057                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1058                         continue;
1059                 atomic_inc(&sp->so_count);
1060                 spin_unlock(&clp->cl_lock);
1061                 status = nfs4_reclaim_open_state(sp, ops);
1062                 if (status < 0) {
1063                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1064                         nfs4_put_state_owner(sp);
1065                         nfs4_recovery_handle_error(clp, status);
1066                         return status;
1067                 }
1068                 nfs4_put_state_owner(sp);
1069                 goto restart;
1070         }
1071         spin_unlock(&clp->cl_lock);
1072         return status;
1073 }
1074
1075 static int nfs4_check_lease(struct nfs_client *clp)
1076 {
1077         struct rpc_cred *cred;
1078         int status = -NFS4ERR_EXPIRED;
1079
1080         /* Is the client already known to have an expired lease? */
1081         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1082                 return 0;
1083         cred = nfs4_get_renew_cred(clp);
1084         if (cred == NULL) {
1085                 cred = nfs4_get_setclientid_cred(clp);
1086                 if (cred == NULL)
1087                         goto out;
1088         }
1089         status = nfs4_proc_renew(clp, cred);
1090         put_rpccred(cred);
1091 out:
1092         nfs4_recovery_handle_error(clp, status);
1093         return status;
1094 }
1095
1096 static int nfs4_reclaim_lease(struct nfs_client *clp)
1097 {
1098         struct rpc_cred *cred;
1099         int status = -ENOENT;
1100
1101         cred = nfs4_get_setclientid_cred(clp);
1102         if (cred != NULL) {
1103                 status = nfs4_init_client(clp, cred);
1104                 put_rpccred(cred);
1105                 /* Handle case where the user hasn't set up machine creds */
1106                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1107                         nfs4_clear_machine_cred(clp);
1108                         status = -EAGAIN;
1109                 }
1110                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1111                         status = -EPROTONOSUPPORT;
1112         }
1113         return status;
1114 }
1115
1116 #ifdef CONFIG_NFS_V4_1
1117
1118 static int nfs4_initialize_session(struct nfs_client *clp)
1119 {
1120         int status;
1121
1122         status = nfs4_proc_create_session(clp, 0);
1123         if (!status) {
1124                 nfs_mark_client_ready(clp, NFS_CS_READY);
1125         } else if (status == -NFS4ERR_STALE_CLIENTID) {
1126                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1127                 set_bit(NFS4CLNT_SESSION_SETUP, &clp->cl_state);
1128         } else {
1129                 nfs_mark_client_ready(clp, status);
1130         }
1131         return status;
1132 }
1133 #else /* CONFIG_NFS_V4_1 */
1134 static int nfs4_initialize_session(struct nfs_client *clp) { return 0; }
1135 #endif /* CONFIG_NFS_V4_1 */
1136
1137 static void nfs4_state_manager(struct nfs_client *clp)
1138 {
1139         int status = 0;
1140
1141         /* Ensure exclusive access to NFSv4 state */
1142         for(;;) {
1143                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1144                         /* We're going to have to re-establish a clientid */
1145                         status = nfs4_reclaim_lease(clp);
1146                         if (status) {
1147                                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1148                                 if (status == -EAGAIN)
1149                                         continue;
1150                                 if (clp->cl_cons_state ==
1151                                                         NFS_CS_SESSION_INITING)
1152                                         nfs_mark_client_ready(clp, status);
1153                                 goto out_error;
1154                         }
1155                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1156                 }
1157
1158                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1159                         status = nfs4_check_lease(clp);
1160                         if (status != 0)
1161                                 continue;
1162                 }
1163                 /* Setup the session */
1164                 if (nfs4_has_session(clp) &&
1165                    test_and_clear_bit(NFS4CLNT_SESSION_SETUP, &clp->cl_state)) {
1166                         status = nfs4_initialize_session(clp);
1167                         if (status) {
1168                                 if (status == -NFS4ERR_STALE_CLIENTID)
1169                                         continue;
1170                                 goto out_error;
1171                         }
1172                 }
1173                 /* First recover reboot state... */
1174                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1175                         status = nfs4_do_reclaim(clp, &nfs4_reboot_recovery_ops);
1176                         if (status == -NFS4ERR_STALE_CLIENTID)
1177                                 continue;
1178                         nfs4_state_end_reclaim_reboot(clp);
1179                         continue;
1180                 }
1181
1182                 /* Now recover expired state... */
1183                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1184                         status = nfs4_do_reclaim(clp, &nfs4_nograce_recovery_ops);
1185                         if (status < 0) {
1186                                 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1187                                 if (status == -NFS4ERR_STALE_CLIENTID)
1188                                         continue;
1189                                 if (status == -NFS4ERR_EXPIRED)
1190                                         continue;
1191                                 goto out_error;
1192                         } else
1193                                 nfs4_state_end_reclaim_nograce(clp);
1194                         continue;
1195                 }
1196
1197                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1198                         nfs_client_return_marked_delegations(clp);
1199                         continue;
1200                 }
1201
1202                 nfs4_clear_state_manager_bit(clp);
1203                 /* Did we race with an attempt to give us more work? */
1204                 if (clp->cl_state == 0)
1205                         break;
1206                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1207                         break;
1208         }
1209         return;
1210 out_error:
1211         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1212                         " with error %d\n", clp->cl_hostname, -status);
1213         if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1214                 nfs4_state_end_reclaim_reboot(clp);
1215         nfs4_clear_state_manager_bit(clp);
1216 }
1217
1218 static int nfs4_run_state_manager(void *ptr)
1219 {
1220         struct nfs_client *clp = ptr;
1221
1222         allow_signal(SIGKILL);
1223         nfs4_state_manager(clp);
1224         nfs_put_client(clp);
1225         module_put_and_exit(0);
1226         return 0;
1227 }
1228
1229 /*
1230  * Local variables:
1231  *  c-basic-offset: 8
1232  * End:
1233  */