]> Pileus Git - ~andy/linux/blob - fs/nfs/nfs4state.c
NFSv4: Clean up nfs4_get_state_owner
[~andy/linux] / 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/fs.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/ratelimit.h>
50 #include <linux/workqueue.h>
51 #include <linux/bitops.h>
52 #include <linux/jiffies.h>
53
54 #include "nfs4_fs.h"
55 #include "callback.h"
56 #include "delegation.h"
57 #include "internal.h"
58 #include "pnfs.h"
59
60 #define OPENOWNER_POOL_SIZE     8
61
62 const nfs4_stateid zero_stateid;
63
64 static LIST_HEAD(nfs4_clientid_list);
65
66 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
67 {
68         struct nfs4_setclientid_res clid = {
69                 .clientid = clp->cl_clientid,
70                 .confirm = clp->cl_confirm,
71         };
72         unsigned short port;
73         int status;
74
75         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
76                 goto do_confirm;
77         port = nfs_callback_tcpport;
78         if (clp->cl_addr.ss_family == AF_INET6)
79                 port = nfs_callback_tcpport6;
80
81         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
82         if (status != 0)
83                 goto out;
84         clp->cl_clientid = clid.clientid;
85         clp->cl_confirm = clid.confirm;
86         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
87 do_confirm:
88         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
89         if (status != 0)
90                 goto out;
91         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
92         nfs4_schedule_state_renewal(clp);
93 out:
94         return status;
95 }
96
97 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
98 {
99         struct rpc_cred *cred = NULL;
100
101         if (clp->cl_machine_cred != NULL)
102                 cred = get_rpccred(clp->cl_machine_cred);
103         return cred;
104 }
105
106 static void nfs4_clear_machine_cred(struct nfs_client *clp)
107 {
108         struct rpc_cred *cred;
109
110         spin_lock(&clp->cl_lock);
111         cred = clp->cl_machine_cred;
112         clp->cl_machine_cred = NULL;
113         spin_unlock(&clp->cl_lock);
114         if (cred != NULL)
115                 put_rpccred(cred);
116 }
117
118 static struct rpc_cred *
119 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
120 {
121         struct rpc_cred *cred = NULL;
122         struct nfs4_state_owner *sp;
123         struct rb_node *pos;
124
125         for (pos = rb_first(&server->state_owners);
126              pos != NULL;
127              pos = rb_next(pos)) {
128                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
129                 if (list_empty(&sp->so_states))
130                         continue;
131                 cred = get_rpccred(sp->so_cred);
132                 break;
133         }
134         return cred;
135 }
136
137 /**
138  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
139  * @clp: client state handle
140  *
141  * Returns an rpc_cred with reference count bumped, or NULL.
142  * Caller must hold clp->cl_lock.
143  */
144 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
145 {
146         struct rpc_cred *cred = NULL;
147         struct nfs_server *server;
148
149         rcu_read_lock();
150         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
151                 cred = nfs4_get_renew_cred_server_locked(server);
152                 if (cred != NULL)
153                         break;
154         }
155         rcu_read_unlock();
156         return cred;
157 }
158
159 #if defined(CONFIG_NFS_V4_1)
160
161 static int nfs41_setup_state_renewal(struct nfs_client *clp)
162 {
163         int status;
164         struct nfs_fsinfo fsinfo;
165
166         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
167                 nfs4_schedule_state_renewal(clp);
168                 return 0;
169         }
170
171         status = nfs4_proc_get_lease_time(clp, &fsinfo);
172         if (status == 0) {
173                 /* Update lease time and schedule renewal */
174                 spin_lock(&clp->cl_lock);
175                 clp->cl_lease_time = fsinfo.lease_time * HZ;
176                 clp->cl_last_renewal = jiffies;
177                 spin_unlock(&clp->cl_lock);
178
179                 nfs4_schedule_state_renewal(clp);
180         }
181
182         return status;
183 }
184
185 /*
186  * Back channel returns NFS4ERR_DELAY for new requests when
187  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
188  * is ended.
189  */
190 static void nfs4_end_drain_session(struct nfs_client *clp)
191 {
192         struct nfs4_session *ses = clp->cl_session;
193         int max_slots;
194
195         if (ses == NULL)
196                 return;
197         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
198                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
199                 max_slots = ses->fc_slot_table.max_slots;
200                 while (max_slots--) {
201                         struct rpc_task *task;
202
203                         task = rpc_wake_up_next(&ses->fc_slot_table.
204                                                 slot_tbl_waitq);
205                         if (!task)
206                                 break;
207                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
208                 }
209                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
210         }
211 }
212
213 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
214 {
215         spin_lock(&tbl->slot_tbl_lock);
216         if (tbl->highest_used_slotid != -1) {
217                 INIT_COMPLETION(tbl->complete);
218                 spin_unlock(&tbl->slot_tbl_lock);
219                 return wait_for_completion_interruptible(&tbl->complete);
220         }
221         spin_unlock(&tbl->slot_tbl_lock);
222         return 0;
223 }
224
225 static int nfs4_begin_drain_session(struct nfs_client *clp)
226 {
227         struct nfs4_session *ses = clp->cl_session;
228         int ret = 0;
229
230         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
231         /* back channel */
232         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
233         if (ret)
234                 return ret;
235         /* fore channel */
236         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
237 }
238
239 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
240 {
241         int status;
242
243         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
244                 goto do_confirm;
245         nfs4_begin_drain_session(clp);
246         status = nfs4_proc_exchange_id(clp, cred);
247         if (status != 0)
248                 goto out;
249         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
250 do_confirm:
251         status = nfs4_proc_create_session(clp);
252         if (status != 0)
253                 goto out;
254         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
255         nfs41_setup_state_renewal(clp);
256         nfs_mark_client_ready(clp, NFS_CS_READY);
257 out:
258         return status;
259 }
260
261 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
262 {
263         struct rpc_cred *cred;
264
265         spin_lock(&clp->cl_lock);
266         cred = nfs4_get_machine_cred_locked(clp);
267         spin_unlock(&clp->cl_lock);
268         return cred;
269 }
270
271 #endif /* CONFIG_NFS_V4_1 */
272
273 static struct rpc_cred *
274 nfs4_get_setclientid_cred_server(struct nfs_server *server)
275 {
276         struct nfs_client *clp = server->nfs_client;
277         struct rpc_cred *cred = NULL;
278         struct nfs4_state_owner *sp;
279         struct rb_node *pos;
280
281         spin_lock(&clp->cl_lock);
282         pos = rb_first(&server->state_owners);
283         if (pos != NULL) {
284                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
285                 cred = get_rpccred(sp->so_cred);
286         }
287         spin_unlock(&clp->cl_lock);
288         return cred;
289 }
290
291 /**
292  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
293  * @clp: client state handle
294  *
295  * Returns an rpc_cred with reference count bumped, or NULL.
296  */
297 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
298 {
299         struct nfs_server *server;
300         struct rpc_cred *cred;
301
302         spin_lock(&clp->cl_lock);
303         cred = nfs4_get_machine_cred_locked(clp);
304         spin_unlock(&clp->cl_lock);
305         if (cred != NULL)
306                 goto out;
307
308         rcu_read_lock();
309         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
310                 cred = nfs4_get_setclientid_cred_server(server);
311                 if (cred != NULL)
312                         break;
313         }
314         rcu_read_unlock();
315
316 out:
317         return cred;
318 }
319
320 static void nfs_alloc_unique_id_locked(struct rb_root *root,
321                                        struct nfs_unique_id *new,
322                                        __u64 minval, int maxbits)
323 {
324         struct rb_node **p, *parent;
325         struct nfs_unique_id *pos;
326         __u64 mask = ~0ULL;
327
328         if (maxbits < 64)
329                 mask = (1ULL << maxbits) - 1ULL;
330
331         /* Ensure distribution is more or less flat */
332         get_random_bytes(&new->id, sizeof(new->id));
333         new->id &= mask;
334         if (new->id < minval)
335                 new->id += minval;
336 retry:
337         p = &root->rb_node;
338         parent = NULL;
339
340         while (*p != NULL) {
341                 parent = *p;
342                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
343
344                 if (new->id < pos->id)
345                         p = &(*p)->rb_left;
346                 else if (new->id > pos->id)
347                         p = &(*p)->rb_right;
348                 else
349                         goto id_exists;
350         }
351         rb_link_node(&new->rb_node, parent, p);
352         rb_insert_color(&new->rb_node, root);
353         return;
354 id_exists:
355         for (;;) {
356                 new->id++;
357                 if (new->id < minval || (new->id & mask) != new->id) {
358                         new->id = minval;
359                         break;
360                 }
361                 parent = rb_next(parent);
362                 if (parent == NULL)
363                         break;
364                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
365                 if (new->id < pos->id)
366                         break;
367         }
368         goto retry;
369 }
370
371 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
372 {
373         rb_erase(&id->rb_node, root);
374 }
375
376 static struct nfs4_state_owner *
377 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
378 {
379         struct rb_node **p = &server->state_owners.rb_node,
380                        *parent = NULL;
381         struct nfs4_state_owner *sp;
382
383         while (*p != NULL) {
384                 parent = *p;
385                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
386
387                 if (cred < sp->so_cred)
388                         p = &parent->rb_left;
389                 else if (cred > sp->so_cred)
390                         p = &parent->rb_right;
391                 else {
392                         if (!list_empty(&sp->so_lru))
393                                 list_del_init(&sp->so_lru);
394                         atomic_inc(&sp->so_count);
395                         return sp;
396                 }
397         }
398         return NULL;
399 }
400
401 static struct nfs4_state_owner *
402 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
403 {
404         struct nfs_server *server = new->so_server;
405         struct rb_node **p = &server->state_owners.rb_node,
406                        *parent = NULL;
407         struct nfs4_state_owner *sp;
408
409         while (*p != NULL) {
410                 parent = *p;
411                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
412
413                 if (new->so_cred < sp->so_cred)
414                         p = &parent->rb_left;
415                 else if (new->so_cred > sp->so_cred)
416                         p = &parent->rb_right;
417                 else {
418                         if (!list_empty(&sp->so_lru))
419                                 list_del_init(&sp->so_lru);
420                         atomic_inc(&sp->so_count);
421                         return sp;
422                 }
423         }
424         nfs_alloc_unique_id_locked(&server->openowner_id,
425                                         &new->so_owner_id, 1, 64);
426         rb_link_node(&new->so_server_node, parent, p);
427         rb_insert_color(&new->so_server_node, &server->state_owners);
428         return new;
429 }
430
431 static void
432 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
433 {
434         struct nfs_server *server = sp->so_server;
435
436         if (!RB_EMPTY_NODE(&sp->so_server_node))
437                 rb_erase(&sp->so_server_node, &server->state_owners);
438         nfs_free_unique_id(&server->openowner_id, &sp->so_owner_id);
439 }
440
441 /*
442  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
443  * create a new state_owner.
444  *
445  */
446 static struct nfs4_state_owner *
447 nfs4_alloc_state_owner(struct nfs_server *server,
448                 struct rpc_cred *cred,
449                 gfp_t gfp_flags)
450 {
451         struct nfs4_state_owner *sp;
452
453         sp = kzalloc(sizeof(*sp), gfp_flags);
454         if (!sp)
455                 return NULL;
456         sp->so_server = server;
457         sp->so_cred = get_rpccred(cred);
458         spin_lock_init(&sp->so_lock);
459         INIT_LIST_HEAD(&sp->so_states);
460         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
461         sp->so_seqid.sequence = &sp->so_sequence;
462         spin_lock_init(&sp->so_sequence.lock);
463         INIT_LIST_HEAD(&sp->so_sequence.list);
464         atomic_set(&sp->so_count, 1);
465         INIT_LIST_HEAD(&sp->so_lru);
466         return sp;
467 }
468
469 static void
470 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
471 {
472         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
473                 struct nfs_server *server = sp->so_server;
474                 struct nfs_client *clp = server->nfs_client;
475
476                 spin_lock(&clp->cl_lock);
477                 rb_erase(&sp->so_server_node, &server->state_owners);
478                 RB_CLEAR_NODE(&sp->so_server_node);
479                 spin_unlock(&clp->cl_lock);
480         }
481 }
482
483 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
484 {
485         rpc_destroy_wait_queue(&sp->so_sequence.wait);
486         put_rpccred(sp->so_cred);
487         kfree(sp);
488 }
489
490 static void nfs4_gc_state_owners(struct nfs_server *server)
491 {
492         struct nfs_client *clp = server->nfs_client;
493         struct nfs4_state_owner *sp, *tmp;
494         unsigned long time_min, time_max;
495         LIST_HEAD(doomed);
496
497         spin_lock(&clp->cl_lock);
498         time_max = jiffies;
499         time_min = (long)time_max - (long)clp->cl_lease_time;
500         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
501                 /* NB: LRU is sorted so that oldest is at the head */
502                 if (time_in_range(sp->so_expires, time_min, time_max))
503                         break;
504                 list_move(&sp->so_lru, &doomed);
505                 nfs4_remove_state_owner_locked(sp);
506         }
507         spin_unlock(&clp->cl_lock);
508
509         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
510                 list_del(&sp->so_lru);
511                 nfs4_free_state_owner(sp);
512         }
513 }
514
515 /**
516  * nfs4_get_state_owner - Look up a state owner given a credential
517  * @server: nfs_server to search
518  * @cred: RPC credential to match
519  *
520  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
521  */
522 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
523                                               struct rpc_cred *cred,
524                                               gfp_t gfp_flags)
525 {
526         struct nfs_client *clp = server->nfs_client;
527         struct nfs4_state_owner *sp, *new;
528
529         spin_lock(&clp->cl_lock);
530         sp = nfs4_find_state_owner_locked(server, cred);
531         spin_unlock(&clp->cl_lock);
532         if (sp != NULL)
533                 goto out;
534         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
535         if (new == NULL)
536                 goto out;
537         sp = nfs4_insert_state_owner_locked(new);
538         spin_unlock(&clp->cl_lock);
539         if (sp != new)
540                 nfs4_free_state_owner(new);
541 out:
542         nfs4_gc_state_owners(server);
543         return sp;
544 }
545
546 /**
547  * nfs4_put_state_owner - Release a nfs4_state_owner
548  * @sp: state owner data to release
549  */
550 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
551 {
552         struct nfs_server *server = sp->so_server;
553         struct nfs_client *clp = server->nfs_client;
554
555         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
556                 return;
557
558         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
559                 sp->so_expires = jiffies;
560                 list_add_tail(&sp->so_lru, &server->state_owners_lru);
561                 spin_unlock(&clp->cl_lock);
562         } else {
563                 nfs4_remove_state_owner_locked(sp);
564                 spin_unlock(&clp->cl_lock);
565                 nfs4_free_state_owner(sp);
566         }
567 }
568
569 /**
570  * nfs4_purge_state_owners - Release all cached state owners
571  * @server: nfs_server with cached state owners to release
572  *
573  * Called at umount time.  Remaining state owners will be on
574  * the LRU with ref count of zero.
575  */
576 void nfs4_purge_state_owners(struct nfs_server *server)
577 {
578         struct nfs_client *clp = server->nfs_client;
579         struct nfs4_state_owner *sp, *tmp;
580         LIST_HEAD(doomed);
581
582         spin_lock(&clp->cl_lock);
583         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
584                 list_move(&sp->so_lru, &doomed);
585                 nfs4_remove_state_owner_locked(sp);
586         }
587         spin_unlock(&clp->cl_lock);
588
589         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
590                 list_del(&sp->so_lru);
591                 nfs4_free_state_owner(sp);
592         }
593 }
594
595 static struct nfs4_state *
596 nfs4_alloc_open_state(void)
597 {
598         struct nfs4_state *state;
599
600         state = kzalloc(sizeof(*state), GFP_NOFS);
601         if (!state)
602                 return NULL;
603         atomic_set(&state->count, 1);
604         INIT_LIST_HEAD(&state->lock_states);
605         spin_lock_init(&state->state_lock);
606         seqlock_init(&state->seqlock);
607         return state;
608 }
609
610 void
611 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
612 {
613         if (state->state == fmode)
614                 return;
615         /* NB! List reordering - see the reclaim code for why.  */
616         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
617                 if (fmode & FMODE_WRITE)
618                         list_move(&state->open_states, &state->owner->so_states);
619                 else
620                         list_move_tail(&state->open_states, &state->owner->so_states);
621         }
622         state->state = fmode;
623 }
624
625 static struct nfs4_state *
626 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
627 {
628         struct nfs_inode *nfsi = NFS_I(inode);
629         struct nfs4_state *state;
630
631         list_for_each_entry(state, &nfsi->open_states, inode_states) {
632                 if (state->owner != owner)
633                         continue;
634                 if (atomic_inc_not_zero(&state->count))
635                         return state;
636         }
637         return NULL;
638 }
639
640 static void
641 nfs4_free_open_state(struct nfs4_state *state)
642 {
643         kfree(state);
644 }
645
646 struct nfs4_state *
647 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
648 {
649         struct nfs4_state *state, *new;
650         struct nfs_inode *nfsi = NFS_I(inode);
651
652         spin_lock(&inode->i_lock);
653         state = __nfs4_find_state_byowner(inode, owner);
654         spin_unlock(&inode->i_lock);
655         if (state)
656                 goto out;
657         new = nfs4_alloc_open_state();
658         spin_lock(&owner->so_lock);
659         spin_lock(&inode->i_lock);
660         state = __nfs4_find_state_byowner(inode, owner);
661         if (state == NULL && new != NULL) {
662                 state = new;
663                 state->owner = owner;
664                 atomic_inc(&owner->so_count);
665                 list_add(&state->inode_states, &nfsi->open_states);
666                 ihold(inode);
667                 state->inode = inode;
668                 spin_unlock(&inode->i_lock);
669                 /* Note: The reclaim code dictates that we add stateless
670                  * and read-only stateids to the end of the list */
671                 list_add_tail(&state->open_states, &owner->so_states);
672                 spin_unlock(&owner->so_lock);
673         } else {
674                 spin_unlock(&inode->i_lock);
675                 spin_unlock(&owner->so_lock);
676                 if (new)
677                         nfs4_free_open_state(new);
678         }
679 out:
680         return state;
681 }
682
683 void nfs4_put_open_state(struct nfs4_state *state)
684 {
685         struct inode *inode = state->inode;
686         struct nfs4_state_owner *owner = state->owner;
687
688         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
689                 return;
690         spin_lock(&inode->i_lock);
691         list_del(&state->inode_states);
692         list_del(&state->open_states);
693         spin_unlock(&inode->i_lock);
694         spin_unlock(&owner->so_lock);
695         iput(inode);
696         nfs4_free_open_state(state);
697         nfs4_put_state_owner(owner);
698 }
699
700 /*
701  * Close the current file.
702  */
703 static void __nfs4_close(struct nfs4_state *state,
704                 fmode_t fmode, gfp_t gfp_mask, int wait)
705 {
706         struct nfs4_state_owner *owner = state->owner;
707         int call_close = 0;
708         fmode_t newstate;
709
710         atomic_inc(&owner->so_count);
711         /* Protect against nfs4_find_state() */
712         spin_lock(&owner->so_lock);
713         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
714                 case FMODE_READ:
715                         state->n_rdonly--;
716                         break;
717                 case FMODE_WRITE:
718                         state->n_wronly--;
719                         break;
720                 case FMODE_READ|FMODE_WRITE:
721                         state->n_rdwr--;
722         }
723         newstate = FMODE_READ|FMODE_WRITE;
724         if (state->n_rdwr == 0) {
725                 if (state->n_rdonly == 0) {
726                         newstate &= ~FMODE_READ;
727                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
728                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
729                 }
730                 if (state->n_wronly == 0) {
731                         newstate &= ~FMODE_WRITE;
732                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
733                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
734                 }
735                 if (newstate == 0)
736                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
737         }
738         nfs4_state_set_mode_locked(state, newstate);
739         spin_unlock(&owner->so_lock);
740
741         if (!call_close) {
742                 nfs4_put_open_state(state);
743                 nfs4_put_state_owner(owner);
744         } else {
745                 bool roc = pnfs_roc(state->inode);
746
747                 nfs4_do_close(state, gfp_mask, wait, roc);
748         }
749 }
750
751 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
752 {
753         __nfs4_close(state, fmode, GFP_NOFS, 0);
754 }
755
756 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
757 {
758         __nfs4_close(state, fmode, GFP_KERNEL, 1);
759 }
760
761 /*
762  * Search the state->lock_states for an existing lock_owner
763  * that is compatible with current->files
764  */
765 static struct nfs4_lock_state *
766 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
767 {
768         struct nfs4_lock_state *pos;
769         list_for_each_entry(pos, &state->lock_states, ls_locks) {
770                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
771                         continue;
772                 switch (pos->ls_owner.lo_type) {
773                 case NFS4_POSIX_LOCK_TYPE:
774                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
775                                 continue;
776                         break;
777                 case NFS4_FLOCK_LOCK_TYPE:
778                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
779                                 continue;
780                 }
781                 atomic_inc(&pos->ls_count);
782                 return pos;
783         }
784         return NULL;
785 }
786
787 /*
788  * Return a compatible lock_state. If no initialized lock_state structure
789  * exists, return an uninitialized one.
790  *
791  */
792 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
793 {
794         struct nfs4_lock_state *lsp;
795         struct nfs_server *server = state->owner->so_server;
796         struct nfs_client *clp = server->nfs_client;
797
798         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
799         if (lsp == NULL)
800                 return NULL;
801         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
802         spin_lock_init(&lsp->ls_sequence.lock);
803         INIT_LIST_HEAD(&lsp->ls_sequence.list);
804         lsp->ls_seqid.sequence = &lsp->ls_sequence;
805         atomic_set(&lsp->ls_count, 1);
806         lsp->ls_state = state;
807         lsp->ls_owner.lo_type = type;
808         switch (lsp->ls_owner.lo_type) {
809         case NFS4_FLOCK_LOCK_TYPE:
810                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
811                 break;
812         case NFS4_POSIX_LOCK_TYPE:
813                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
814                 break;
815         default:
816                 kfree(lsp);
817                 return NULL;
818         }
819         spin_lock(&clp->cl_lock);
820         nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
821         spin_unlock(&clp->cl_lock);
822         INIT_LIST_HEAD(&lsp->ls_locks);
823         return lsp;
824 }
825
826 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
827 {
828         struct nfs_server *server = lsp->ls_state->owner->so_server;
829         struct nfs_client *clp = server->nfs_client;
830
831         spin_lock(&clp->cl_lock);
832         nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
833         spin_unlock(&clp->cl_lock);
834         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
835         kfree(lsp);
836 }
837
838 /*
839  * Return a compatible lock_state. If no initialized lock_state structure
840  * exists, return an uninitialized one.
841  *
842  */
843 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
844 {
845         struct nfs4_lock_state *lsp, *new = NULL;
846         
847         for(;;) {
848                 spin_lock(&state->state_lock);
849                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
850                 if (lsp != NULL)
851                         break;
852                 if (new != NULL) {
853                         list_add(&new->ls_locks, &state->lock_states);
854                         set_bit(LK_STATE_IN_USE, &state->flags);
855                         lsp = new;
856                         new = NULL;
857                         break;
858                 }
859                 spin_unlock(&state->state_lock);
860                 new = nfs4_alloc_lock_state(state, owner, pid, type);
861                 if (new == NULL)
862                         return NULL;
863         }
864         spin_unlock(&state->state_lock);
865         if (new != NULL)
866                 nfs4_free_lock_state(new);
867         return lsp;
868 }
869
870 /*
871  * Release reference to lock_state, and free it if we see that
872  * it is no longer in use
873  */
874 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
875 {
876         struct nfs4_state *state;
877
878         if (lsp == NULL)
879                 return;
880         state = lsp->ls_state;
881         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
882                 return;
883         list_del(&lsp->ls_locks);
884         if (list_empty(&state->lock_states))
885                 clear_bit(LK_STATE_IN_USE, &state->flags);
886         spin_unlock(&state->state_lock);
887         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
888                 nfs4_release_lockowner(lsp);
889         nfs4_free_lock_state(lsp);
890 }
891
892 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
893 {
894         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
895
896         dst->fl_u.nfs4_fl.owner = lsp;
897         atomic_inc(&lsp->ls_count);
898 }
899
900 static void nfs4_fl_release_lock(struct file_lock *fl)
901 {
902         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
903 }
904
905 static const struct file_lock_operations nfs4_fl_lock_ops = {
906         .fl_copy_lock = nfs4_fl_copy_lock,
907         .fl_release_private = nfs4_fl_release_lock,
908 };
909
910 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
911 {
912         struct nfs4_lock_state *lsp;
913
914         if (fl->fl_ops != NULL)
915                 return 0;
916         if (fl->fl_flags & FL_POSIX)
917                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
918         else if (fl->fl_flags & FL_FLOCK)
919                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
920         else
921                 return -EINVAL;
922         if (lsp == NULL)
923                 return -ENOMEM;
924         fl->fl_u.nfs4_fl.owner = lsp;
925         fl->fl_ops = &nfs4_fl_lock_ops;
926         return 0;
927 }
928
929 /*
930  * Byte-range lock aware utility to initialize the stateid of read/write
931  * requests.
932  */
933 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
934 {
935         struct nfs4_lock_state *lsp;
936         int seq;
937
938         do {
939                 seq = read_seqbegin(&state->seqlock);
940                 memcpy(dst, &state->stateid, sizeof(*dst));
941         } while (read_seqretry(&state->seqlock, seq));
942         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
943                 return;
944
945         spin_lock(&state->state_lock);
946         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
947         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
948                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
949         spin_unlock(&state->state_lock);
950         nfs4_put_lock_state(lsp);
951 }
952
953 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
954 {
955         struct nfs_seqid *new;
956
957         new = kmalloc(sizeof(*new), gfp_mask);
958         if (new != NULL) {
959                 new->sequence = counter;
960                 INIT_LIST_HEAD(&new->list);
961         }
962         return new;
963 }
964
965 void nfs_release_seqid(struct nfs_seqid *seqid)
966 {
967         if (!list_empty(&seqid->list)) {
968                 struct rpc_sequence *sequence = seqid->sequence->sequence;
969
970                 spin_lock(&sequence->lock);
971                 list_del_init(&seqid->list);
972                 spin_unlock(&sequence->lock);
973                 rpc_wake_up(&sequence->wait);
974         }
975 }
976
977 void nfs_free_seqid(struct nfs_seqid *seqid)
978 {
979         nfs_release_seqid(seqid);
980         kfree(seqid);
981 }
982
983 /*
984  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
985  * failed with a seqid incrementing error -
986  * see comments nfs_fs.h:seqid_mutating_error()
987  */
988 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
989 {
990         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
991         switch (status) {
992                 case 0:
993                         break;
994                 case -NFS4ERR_BAD_SEQID:
995                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
996                                 return;
997                         printk(KERN_WARNING "NFS: v4 server returned a bad"
998                                         " sequence-id error on an"
999                                         " unconfirmed sequence %p!\n",
1000                                         seqid->sequence);
1001                 case -NFS4ERR_STALE_CLIENTID:
1002                 case -NFS4ERR_STALE_STATEID:
1003                 case -NFS4ERR_BAD_STATEID:
1004                 case -NFS4ERR_BADXDR:
1005                 case -NFS4ERR_RESOURCE:
1006                 case -NFS4ERR_NOFILEHANDLE:
1007                         /* Non-seqid mutating errors */
1008                         return;
1009         };
1010         /*
1011          * Note: no locking needed as we are guaranteed to be first
1012          * on the sequence list
1013          */
1014         seqid->sequence->counter++;
1015 }
1016
1017 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1018 {
1019         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1020                                         struct nfs4_state_owner, so_seqid);
1021         struct nfs_server *server = sp->so_server;
1022
1023         if (status == -NFS4ERR_BAD_SEQID)
1024                 nfs4_drop_state_owner(sp);
1025         if (!nfs4_has_session(server->nfs_client))
1026                 nfs_increment_seqid(status, seqid);
1027 }
1028
1029 /*
1030  * Increment the seqid if the LOCK/LOCKU succeeded, or
1031  * failed with a seqid incrementing error -
1032  * see comments nfs_fs.h:seqid_mutating_error()
1033  */
1034 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1035 {
1036         nfs_increment_seqid(status, seqid);
1037 }
1038
1039 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1040 {
1041         struct rpc_sequence *sequence = seqid->sequence->sequence;
1042         int status = 0;
1043
1044         spin_lock(&sequence->lock);
1045         if (list_empty(&seqid->list))
1046                 list_add_tail(&seqid->list, &sequence->list);
1047         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1048                 goto unlock;
1049         rpc_sleep_on(&sequence->wait, task, NULL);
1050         status = -EAGAIN;
1051 unlock:
1052         spin_unlock(&sequence->lock);
1053         return status;
1054 }
1055
1056 static int nfs4_run_state_manager(void *);
1057
1058 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1059 {
1060         smp_mb__before_clear_bit();
1061         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1062         smp_mb__after_clear_bit();
1063         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1064         rpc_wake_up(&clp->cl_rpcwaitq);
1065 }
1066
1067 /*
1068  * Schedule the nfs_client asynchronous state management routine
1069  */
1070 void nfs4_schedule_state_manager(struct nfs_client *clp)
1071 {
1072         struct task_struct *task;
1073
1074         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1075                 return;
1076         __module_get(THIS_MODULE);
1077         atomic_inc(&clp->cl_count);
1078         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1079                                 rpc_peeraddr2str(clp->cl_rpcclient,
1080                                                         RPC_DISPLAY_ADDR));
1081         if (!IS_ERR(task))
1082                 return;
1083         nfs4_clear_state_manager_bit(clp);
1084         nfs_put_client(clp);
1085         module_put(THIS_MODULE);
1086 }
1087
1088 /*
1089  * Schedule a lease recovery attempt
1090  */
1091 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1092 {
1093         if (!clp)
1094                 return;
1095         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1096                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1097         nfs4_schedule_state_manager(clp);
1098 }
1099
1100 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1101 {
1102         nfs_handle_cb_pathdown(clp);
1103         nfs4_schedule_state_manager(clp);
1104 }
1105
1106 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1107 {
1108
1109         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1110         /* Don't recover state that expired before the reboot */
1111         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1112                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1113                 return 0;
1114         }
1115         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1116         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1117         return 1;
1118 }
1119
1120 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1121 {
1122         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1123         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1124         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1125         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1126         return 1;
1127 }
1128
1129 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1130 {
1131         struct nfs_client *clp = server->nfs_client;
1132
1133         nfs4_state_mark_reclaim_nograce(clp, state);
1134         nfs4_schedule_state_manager(clp);
1135 }
1136
1137 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1138 {
1139         struct inode *inode = state->inode;
1140         struct nfs_inode *nfsi = NFS_I(inode);
1141         struct file_lock *fl;
1142         int status = 0;
1143
1144         if (inode->i_flock == NULL)
1145                 return 0;
1146
1147         /* Guard against delegation returns and new lock/unlock calls */
1148         down_write(&nfsi->rwsem);
1149         /* Protect inode->i_flock using the BKL */
1150         lock_flocks();
1151         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1152                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1153                         continue;
1154                 if (nfs_file_open_context(fl->fl_file)->state != state)
1155                         continue;
1156                 unlock_flocks();
1157                 status = ops->recover_lock(state, fl);
1158                 switch (status) {
1159                         case 0:
1160                                 break;
1161                         case -ESTALE:
1162                         case -NFS4ERR_ADMIN_REVOKED:
1163                         case -NFS4ERR_STALE_STATEID:
1164                         case -NFS4ERR_BAD_STATEID:
1165                         case -NFS4ERR_EXPIRED:
1166                         case -NFS4ERR_NO_GRACE:
1167                         case -NFS4ERR_STALE_CLIENTID:
1168                         case -NFS4ERR_BADSESSION:
1169                         case -NFS4ERR_BADSLOT:
1170                         case -NFS4ERR_BAD_HIGH_SLOT:
1171                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1172                                 goto out;
1173                         default:
1174                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1175                                                 __func__, status);
1176                         case -ENOMEM:
1177                         case -NFS4ERR_DENIED:
1178                         case -NFS4ERR_RECLAIM_BAD:
1179                         case -NFS4ERR_RECLAIM_CONFLICT:
1180                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1181                                 status = 0;
1182                 }
1183                 lock_flocks();
1184         }
1185         unlock_flocks();
1186 out:
1187         up_write(&nfsi->rwsem);
1188         return status;
1189 }
1190
1191 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1192 {
1193         struct nfs4_state *state;
1194         struct nfs4_lock_state *lock;
1195         int status = 0;
1196
1197         /* Note: we rely on the sp->so_states list being ordered 
1198          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1199          * states first.
1200          * This is needed to ensure that the server won't give us any
1201          * read delegations that we have to return if, say, we are
1202          * recovering after a network partition or a reboot from a
1203          * server that doesn't support a grace period.
1204          */
1205 restart:
1206         spin_lock(&sp->so_lock);
1207         list_for_each_entry(state, &sp->so_states, open_states) {
1208                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1209                         continue;
1210                 if (state->state == 0)
1211                         continue;
1212                 atomic_inc(&state->count);
1213                 spin_unlock(&sp->so_lock);
1214                 status = ops->recover_open(sp, state);
1215                 if (status >= 0) {
1216                         status = nfs4_reclaim_locks(state, ops);
1217                         if (status >= 0) {
1218                                 spin_lock(&state->state_lock);
1219                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1220                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1221                                                 printk("%s: Lock reclaim failed!\n",
1222                                                         __func__);
1223                                 }
1224                                 spin_unlock(&state->state_lock);
1225                                 nfs4_put_open_state(state);
1226                                 goto restart;
1227                         }
1228                 }
1229                 switch (status) {
1230                         default:
1231                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1232                                                 __func__, status);
1233                         case -ENOENT:
1234                         case -ENOMEM:
1235                         case -ESTALE:
1236                                 /*
1237                                  * Open state on this file cannot be recovered
1238                                  * All we can do is revert to using the zero stateid.
1239                                  */
1240                                 memset(state->stateid.data, 0,
1241                                         sizeof(state->stateid.data));
1242                                 /* Mark the file as being 'closed' */
1243                                 state->state = 0;
1244                                 break;
1245                         case -EKEYEXPIRED:
1246                                 /*
1247                                  * User RPCSEC_GSS context has expired.
1248                                  * We cannot recover this stateid now, so
1249                                  * skip it and allow recovery thread to
1250                                  * proceed.
1251                                  */
1252                                 break;
1253                         case -NFS4ERR_ADMIN_REVOKED:
1254                         case -NFS4ERR_STALE_STATEID:
1255                         case -NFS4ERR_BAD_STATEID:
1256                         case -NFS4ERR_RECLAIM_BAD:
1257                         case -NFS4ERR_RECLAIM_CONFLICT:
1258                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1259                                 break;
1260                         case -NFS4ERR_EXPIRED:
1261                         case -NFS4ERR_NO_GRACE:
1262                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1263                         case -NFS4ERR_STALE_CLIENTID:
1264                         case -NFS4ERR_BADSESSION:
1265                         case -NFS4ERR_BADSLOT:
1266                         case -NFS4ERR_BAD_HIGH_SLOT:
1267                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1268                                 goto out_err;
1269                 }
1270                 nfs4_put_open_state(state);
1271                 goto restart;
1272         }
1273         spin_unlock(&sp->so_lock);
1274         return 0;
1275 out_err:
1276         nfs4_put_open_state(state);
1277         return status;
1278 }
1279
1280 static void nfs4_clear_open_state(struct nfs4_state *state)
1281 {
1282         struct nfs4_lock_state *lock;
1283
1284         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1285         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1286         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1287         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1288         spin_lock(&state->state_lock);
1289         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1290                 lock->ls_seqid.flags = 0;
1291                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1292         }
1293         spin_unlock(&state->state_lock);
1294 }
1295
1296 static void nfs4_reset_seqids(struct nfs_server *server,
1297         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1298 {
1299         struct nfs_client *clp = server->nfs_client;
1300         struct nfs4_state_owner *sp;
1301         struct rb_node *pos;
1302         struct nfs4_state *state;
1303
1304         spin_lock(&clp->cl_lock);
1305         for (pos = rb_first(&server->state_owners);
1306              pos != NULL;
1307              pos = rb_next(pos)) {
1308                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1309                 sp->so_seqid.flags = 0;
1310                 spin_lock(&sp->so_lock);
1311                 list_for_each_entry(state, &sp->so_states, open_states) {
1312                         if (mark_reclaim(clp, state))
1313                                 nfs4_clear_open_state(state);
1314                 }
1315                 spin_unlock(&sp->so_lock);
1316         }
1317         spin_unlock(&clp->cl_lock);
1318 }
1319
1320 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1321         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1322 {
1323         struct nfs_server *server;
1324
1325         rcu_read_lock();
1326         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1327                 nfs4_reset_seqids(server, mark_reclaim);
1328         rcu_read_unlock();
1329 }
1330
1331 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1332 {
1333         /* Mark all delegations for reclaim */
1334         nfs_delegation_mark_reclaim(clp);
1335         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1336 }
1337
1338 static void nfs4_reclaim_complete(struct nfs_client *clp,
1339                                  const struct nfs4_state_recovery_ops *ops)
1340 {
1341         /* Notify the server we're done reclaiming our state */
1342         if (ops->reclaim_complete)
1343                 (void)ops->reclaim_complete(clp);
1344 }
1345
1346 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1347 {
1348         struct nfs_client *clp = server->nfs_client;
1349         struct nfs4_state_owner *sp;
1350         struct rb_node *pos;
1351         struct nfs4_state *state;
1352
1353         spin_lock(&clp->cl_lock);
1354         for (pos = rb_first(&server->state_owners);
1355              pos != NULL;
1356              pos = rb_next(pos)) {
1357                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1358                 spin_lock(&sp->so_lock);
1359                 list_for_each_entry(state, &sp->so_states, open_states) {
1360                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1361                                                 &state->flags))
1362                                 continue;
1363                         nfs4_state_mark_reclaim_nograce(clp, state);
1364                 }
1365                 spin_unlock(&sp->so_lock);
1366         }
1367         spin_unlock(&clp->cl_lock);
1368 }
1369
1370 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1371 {
1372         struct nfs_server *server;
1373
1374         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1375                 return 0;
1376
1377         rcu_read_lock();
1378         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1379                 nfs4_clear_reclaim_server(server);
1380         rcu_read_unlock();
1381
1382         nfs_delegation_reap_unclaimed(clp);
1383         return 1;
1384 }
1385
1386 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1387 {
1388         if (!nfs4_state_clear_reclaim_reboot(clp))
1389                 return;
1390         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1391 }
1392
1393 static void nfs_delegation_clear_all(struct nfs_client *clp)
1394 {
1395         nfs_delegation_mark_reclaim(clp);
1396         nfs_delegation_reap_unclaimed(clp);
1397 }
1398
1399 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1400 {
1401         nfs_delegation_clear_all(clp);
1402         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1403 }
1404
1405 static void nfs4_warn_keyexpired(const char *s)
1406 {
1407         printk_ratelimited(KERN_WARNING "Error: state manager"
1408                         " encountered RPCSEC_GSS session"
1409                         " expired against NFSv4 server %s.\n",
1410                         s);
1411 }
1412
1413 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1414 {
1415         switch (error) {
1416                 case 0:
1417                         break;
1418                 case -NFS4ERR_CB_PATH_DOWN:
1419                         nfs_handle_cb_pathdown(clp);
1420                         break;
1421                 case -NFS4ERR_NO_GRACE:
1422                         nfs4_state_end_reclaim_reboot(clp);
1423                         break;
1424                 case -NFS4ERR_STALE_CLIENTID:
1425                 case -NFS4ERR_LEASE_MOVED:
1426                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1427                         nfs4_state_clear_reclaim_reboot(clp);
1428                         nfs4_state_start_reclaim_reboot(clp);
1429                         break;
1430                 case -NFS4ERR_EXPIRED:
1431                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1432                         nfs4_state_start_reclaim_nograce(clp);
1433                         break;
1434                 case -NFS4ERR_BADSESSION:
1435                 case -NFS4ERR_BADSLOT:
1436                 case -NFS4ERR_BAD_HIGH_SLOT:
1437                 case -NFS4ERR_DEADSESSION:
1438                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1439                 case -NFS4ERR_SEQ_FALSE_RETRY:
1440                 case -NFS4ERR_SEQ_MISORDERED:
1441                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1442                         /* Zero session reset errors */
1443                         break;
1444                 case -EKEYEXPIRED:
1445                         /* Nothing we can do */
1446                         nfs4_warn_keyexpired(clp->cl_hostname);
1447                         break;
1448                 default:
1449                         return error;
1450         }
1451         return 0;
1452 }
1453
1454 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1455 {
1456         struct nfs4_state_owner *sp;
1457         struct nfs_server *server;
1458         struct rb_node *pos;
1459         int status = 0;
1460
1461 restart:
1462         rcu_read_lock();
1463         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1464                 nfs4_purge_state_owners(server);
1465                 spin_lock(&clp->cl_lock);
1466                 for (pos = rb_first(&server->state_owners);
1467                      pos != NULL;
1468                      pos = rb_next(pos)) {
1469                         sp = rb_entry(pos,
1470                                 struct nfs4_state_owner, so_server_node);
1471                         if (!test_and_clear_bit(ops->owner_flag_bit,
1472                                                         &sp->so_flags))
1473                                 continue;
1474                         atomic_inc(&sp->so_count);
1475                         spin_unlock(&clp->cl_lock);
1476                         rcu_read_unlock();
1477
1478                         status = nfs4_reclaim_open_state(sp, ops);
1479                         if (status < 0) {
1480                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1481                                 nfs4_put_state_owner(sp);
1482                                 return nfs4_recovery_handle_error(clp, status);
1483                         }
1484
1485                         nfs4_put_state_owner(sp);
1486                         goto restart;
1487                 }
1488                 spin_unlock(&clp->cl_lock);
1489         }
1490         rcu_read_unlock();
1491         return status;
1492 }
1493
1494 static int nfs4_check_lease(struct nfs_client *clp)
1495 {
1496         struct rpc_cred *cred;
1497         const struct nfs4_state_maintenance_ops *ops =
1498                 clp->cl_mvops->state_renewal_ops;
1499         int status;
1500
1501         /* Is the client already known to have an expired lease? */
1502         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1503                 return 0;
1504         spin_lock(&clp->cl_lock);
1505         cred = ops->get_state_renewal_cred_locked(clp);
1506         spin_unlock(&clp->cl_lock);
1507         if (cred == NULL) {
1508                 cred = nfs4_get_setclientid_cred(clp);
1509                 status = -ENOKEY;
1510                 if (cred == NULL)
1511                         goto out;
1512         }
1513         status = ops->renew_lease(clp, cred);
1514         put_rpccred(cred);
1515 out:
1516         return nfs4_recovery_handle_error(clp, status);
1517 }
1518
1519 static int nfs4_reclaim_lease(struct nfs_client *clp)
1520 {
1521         struct rpc_cred *cred;
1522         const struct nfs4_state_recovery_ops *ops =
1523                 clp->cl_mvops->reboot_recovery_ops;
1524         int status = -ENOENT;
1525
1526         cred = ops->get_clid_cred(clp);
1527         if (cred != NULL) {
1528                 status = ops->establish_clid(clp, cred);
1529                 put_rpccred(cred);
1530                 /* Handle case where the user hasn't set up machine creds */
1531                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1532                         nfs4_clear_machine_cred(clp);
1533                         status = -EAGAIN;
1534                 }
1535                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1536                         status = -EPROTONOSUPPORT;
1537         }
1538         return status;
1539 }
1540
1541 #ifdef CONFIG_NFS_V4_1
1542 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1543 {
1544         struct nfs_client *clp = session->clp;
1545
1546         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1547         nfs4_schedule_lease_recovery(clp);
1548 }
1549 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1550
1551 void nfs41_handle_recall_slot(struct nfs_client *clp)
1552 {
1553         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1554         nfs4_schedule_state_manager(clp);
1555 }
1556
1557 static void nfs4_reset_all_state(struct nfs_client *clp)
1558 {
1559         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1560                 clp->cl_boot_time = CURRENT_TIME;
1561                 nfs4_state_start_reclaim_nograce(clp);
1562                 nfs4_schedule_state_manager(clp);
1563         }
1564 }
1565
1566 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1567 {
1568         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1569                 nfs4_state_start_reclaim_reboot(clp);
1570                 nfs4_schedule_state_manager(clp);
1571         }
1572 }
1573
1574 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1575 {
1576         /* Temporary */
1577         nfs4_reset_all_state(clp);
1578 }
1579
1580 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1581 {
1582         /* This will need to handle layouts too */
1583         nfs_expire_all_delegations(clp);
1584 }
1585
1586 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1587 {
1588         nfs_expire_all_delegations(clp);
1589         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1590                 nfs4_schedule_state_manager(clp);
1591 }
1592
1593 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1594 {
1595         if (!flags)
1596                 return;
1597         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1598                 nfs41_handle_server_reboot(clp);
1599         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1600                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1601                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1602                             SEQ4_STATUS_LEASE_MOVED))
1603                 nfs41_handle_state_revoked(clp);
1604         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1605                 nfs41_handle_recallable_state_revoked(clp);
1606         if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1607                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1608                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1609                 nfs41_handle_cb_path_down(clp);
1610 }
1611
1612 static int nfs4_reset_session(struct nfs_client *clp)
1613 {
1614         int status;
1615
1616         nfs4_begin_drain_session(clp);
1617         status = nfs4_proc_destroy_session(clp->cl_session);
1618         if (status && status != -NFS4ERR_BADSESSION &&
1619             status != -NFS4ERR_DEADSESSION) {
1620                 status = nfs4_recovery_handle_error(clp, status);
1621                 goto out;
1622         }
1623
1624         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1625         status = nfs4_proc_create_session(clp);
1626         if (status) {
1627                 status = nfs4_recovery_handle_error(clp, status);
1628                 goto out;
1629         }
1630         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1631         /* create_session negotiated new slot table */
1632         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1633
1634          /* Let the state manager reestablish state */
1635         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1636                 nfs41_setup_state_renewal(clp);
1637 out:
1638         return status;
1639 }
1640
1641 static int nfs4_recall_slot(struct nfs_client *clp)
1642 {
1643         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1644         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1645         struct nfs4_slot *new, *old;
1646         int i;
1647
1648         nfs4_begin_drain_session(clp);
1649         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1650                       GFP_NOFS);
1651         if (!new)
1652                 return -ENOMEM;
1653
1654         spin_lock(&fc_tbl->slot_tbl_lock);
1655         for (i = 0; i < fc_tbl->target_max_slots; i++)
1656                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1657         old = fc_tbl->slots;
1658         fc_tbl->slots = new;
1659         fc_tbl->max_slots = fc_tbl->target_max_slots;
1660         fc_tbl->target_max_slots = 0;
1661         fc_attrs->max_reqs = fc_tbl->max_slots;
1662         spin_unlock(&fc_tbl->slot_tbl_lock);
1663
1664         kfree(old);
1665         nfs4_end_drain_session(clp);
1666         return 0;
1667 }
1668
1669 #else /* CONFIG_NFS_V4_1 */
1670 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1671 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1672 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1673 #endif /* CONFIG_NFS_V4_1 */
1674
1675 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1676  * on EXCHANGE_ID for v4.1
1677  */
1678 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1679 {
1680         switch (status) {
1681         case -NFS4ERR_CLID_INUSE:
1682         case -NFS4ERR_STALE_CLIENTID:
1683                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1684                 break;
1685         case -NFS4ERR_DELAY:
1686         case -ETIMEDOUT:
1687         case -EAGAIN:
1688                 ssleep(1);
1689                 break;
1690
1691         case -EKEYEXPIRED:
1692                 nfs4_warn_keyexpired(clp->cl_hostname);
1693         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1694                                  * in nfs4_exchange_id */
1695         default:
1696                 return;
1697         }
1698         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1699 }
1700
1701 static void nfs4_state_manager(struct nfs_client *clp)
1702 {
1703         int status = 0;
1704
1705         /* Ensure exclusive access to NFSv4 state */
1706         do {
1707                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1708                         /* We're going to have to re-establish a clientid */
1709                         status = nfs4_reclaim_lease(clp);
1710                         if (status) {
1711                                 nfs4_set_lease_expired(clp, status);
1712                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1713                                                         &clp->cl_state))
1714                                         continue;
1715                                 if (clp->cl_cons_state ==
1716                                                         NFS_CS_SESSION_INITING)
1717                                         nfs_mark_client_ready(clp, status);
1718                                 goto out_error;
1719                         }
1720                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1721
1722                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1723                                                &clp->cl_state))
1724                                 nfs4_state_start_reclaim_nograce(clp);
1725                         else
1726                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1727                                         &clp->cl_state);
1728
1729                         pnfs_destroy_all_layouts(clp);
1730                 }
1731
1732                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1733                         status = nfs4_check_lease(clp);
1734                         if (status < 0)
1735                                 goto out_error;
1736                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1737                                 continue;
1738                 }
1739
1740                 /* Initialize or reset the session */
1741                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1742                    && nfs4_has_session(clp)) {
1743                         status = nfs4_reset_session(clp);
1744                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1745                                 continue;
1746                         if (status < 0)
1747                                 goto out_error;
1748                 }
1749
1750                 /* First recover reboot state... */
1751                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1752                         status = nfs4_do_reclaim(clp,
1753                                 clp->cl_mvops->reboot_recovery_ops);
1754                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1755                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1756                                 continue;
1757                         nfs4_state_end_reclaim_reboot(clp);
1758                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1759                                 continue;
1760                         if (status < 0)
1761                                 goto out_error;
1762                 }
1763
1764                 /* Now recover expired state... */
1765                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1766                         status = nfs4_do_reclaim(clp,
1767                                 clp->cl_mvops->nograce_recovery_ops);
1768                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1769                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1770                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1771                                 continue;
1772                         if (status < 0)
1773                                 goto out_error;
1774                 }
1775
1776                 nfs4_end_drain_session(clp);
1777                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1778                         nfs_client_return_marked_delegations(clp);
1779                         continue;
1780                 }
1781                 /* Recall session slots */
1782                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1783                    && nfs4_has_session(clp)) {
1784                         status = nfs4_recall_slot(clp);
1785                         if (status < 0)
1786                                 goto out_error;
1787                         continue;
1788                 }
1789
1790
1791                 nfs4_clear_state_manager_bit(clp);
1792                 /* Did we race with an attempt to give us more work? */
1793                 if (clp->cl_state == 0)
1794                         break;
1795                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1796                         break;
1797         } while (atomic_read(&clp->cl_count) > 1);
1798         return;
1799 out_error:
1800         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1801                         " with error %d\n", clp->cl_hostname, -status);
1802         nfs4_end_drain_session(clp);
1803         nfs4_clear_state_manager_bit(clp);
1804 }
1805
1806 static int nfs4_run_state_manager(void *ptr)
1807 {
1808         struct nfs_client *clp = ptr;
1809
1810         allow_signal(SIGKILL);
1811         nfs4_state_manager(clp);
1812         nfs_put_client(clp);
1813         module_put_and_exit(0);
1814         return 0;
1815 }
1816
1817 /*
1818  * Local variables:
1819  *  c-basic-offset: 8
1820  * End:
1821  */