de44804d9864e5814005e760f59471087f03ef0e
[linux-2.6.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/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         struct nfs4_slot_table *tbl;
194         int max_slots;
195
196         if (ses == NULL)
197                 return;
198         tbl = &ses->fc_slot_table;
199         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
200                 spin_lock(&tbl->slot_tbl_lock);
201                 max_slots = tbl->max_slots;
202                 while (max_slots--) {
203                         if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
204                                                 nfs4_set_task_privileged,
205                                                 NULL) == NULL)
206                                 break;
207                 }
208                 spin_unlock(&tbl->slot_tbl_lock);
209         }
210 }
211
212 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
213 {
214         spin_lock(&tbl->slot_tbl_lock);
215         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
216                 INIT_COMPLETION(tbl->complete);
217                 spin_unlock(&tbl->slot_tbl_lock);
218                 return wait_for_completion_interruptible(&tbl->complete);
219         }
220         spin_unlock(&tbl->slot_tbl_lock);
221         return 0;
222 }
223
224 static int nfs4_begin_drain_session(struct nfs_client *clp)
225 {
226         struct nfs4_session *ses = clp->cl_session;
227         int ret = 0;
228
229         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
230         /* back channel */
231         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
232         if (ret)
233                 return ret;
234         /* fore channel */
235         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
236 }
237
238 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
239 {
240         int status;
241
242         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
243                 goto do_confirm;
244         nfs4_begin_drain_session(clp);
245         status = nfs4_proc_exchange_id(clp, cred);
246         if (status != 0)
247                 goto out;
248         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
249 do_confirm:
250         status = nfs4_proc_create_session(clp);
251         if (status != 0)
252                 goto out;
253         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
254         nfs41_setup_state_renewal(clp);
255         nfs_mark_client_ready(clp, NFS_CS_READY);
256 out:
257         return status;
258 }
259
260 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
261 {
262         struct rpc_cred *cred;
263
264         spin_lock(&clp->cl_lock);
265         cred = nfs4_get_machine_cred_locked(clp);
266         spin_unlock(&clp->cl_lock);
267         return cred;
268 }
269
270 #endif /* CONFIG_NFS_V4_1 */
271
272 static struct rpc_cred *
273 nfs4_get_setclientid_cred_server(struct nfs_server *server)
274 {
275         struct nfs_client *clp = server->nfs_client;
276         struct rpc_cred *cred = NULL;
277         struct nfs4_state_owner *sp;
278         struct rb_node *pos;
279
280         spin_lock(&clp->cl_lock);
281         pos = rb_first(&server->state_owners);
282         if (pos != NULL) {
283                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
284                 cred = get_rpccred(sp->so_cred);
285         }
286         spin_unlock(&clp->cl_lock);
287         return cred;
288 }
289
290 /**
291  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
292  * @clp: client state handle
293  *
294  * Returns an rpc_cred with reference count bumped, or NULL.
295  */
296 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
297 {
298         struct nfs_server *server;
299         struct rpc_cred *cred;
300
301         spin_lock(&clp->cl_lock);
302         cred = nfs4_get_machine_cred_locked(clp);
303         spin_unlock(&clp->cl_lock);
304         if (cred != NULL)
305                 goto out;
306
307         rcu_read_lock();
308         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
309                 cred = nfs4_get_setclientid_cred_server(server);
310                 if (cred != NULL)
311                         break;
312         }
313         rcu_read_unlock();
314
315 out:
316         return cred;
317 }
318
319 static struct nfs4_state_owner *
320 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
321 {
322         struct rb_node **p = &server->state_owners.rb_node,
323                        *parent = NULL;
324         struct nfs4_state_owner *sp;
325
326         while (*p != NULL) {
327                 parent = *p;
328                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
329
330                 if (cred < sp->so_cred)
331                         p = &parent->rb_left;
332                 else if (cred > sp->so_cred)
333                         p = &parent->rb_right;
334                 else {
335                         if (!list_empty(&sp->so_lru))
336                                 list_del_init(&sp->so_lru);
337                         atomic_inc(&sp->so_count);
338                         return sp;
339                 }
340         }
341         return NULL;
342 }
343
344 static struct nfs4_state_owner *
345 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
346 {
347         struct nfs_server *server = new->so_server;
348         struct rb_node **p = &server->state_owners.rb_node,
349                        *parent = NULL;
350         struct nfs4_state_owner *sp;
351         int err;
352
353         while (*p != NULL) {
354                 parent = *p;
355                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
356
357                 if (new->so_cred < sp->so_cred)
358                         p = &parent->rb_left;
359                 else if (new->so_cred > sp->so_cred)
360                         p = &parent->rb_right;
361                 else {
362                         if (!list_empty(&sp->so_lru))
363                                 list_del_init(&sp->so_lru);
364                         atomic_inc(&sp->so_count);
365                         return sp;
366                 }
367         }
368         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
369         if (err)
370                 return ERR_PTR(err);
371         rb_link_node(&new->so_server_node, parent, p);
372         rb_insert_color(&new->so_server_node, &server->state_owners);
373         return new;
374 }
375
376 static void
377 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
378 {
379         struct nfs_server *server = sp->so_server;
380
381         if (!RB_EMPTY_NODE(&sp->so_server_node))
382                 rb_erase(&sp->so_server_node, &server->state_owners);
383         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
384 }
385
386 static void
387 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
388 {
389         sc->flags = 0;
390         sc->counter = 0;
391         spin_lock_init(&sc->lock);
392         INIT_LIST_HEAD(&sc->list);
393         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
394 }
395
396 static void
397 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
398 {
399         rpc_destroy_wait_queue(&sc->wait);
400 }
401
402 /*
403  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
404  * create a new state_owner.
405  *
406  */
407 static struct nfs4_state_owner *
408 nfs4_alloc_state_owner(struct nfs_server *server,
409                 struct rpc_cred *cred,
410                 gfp_t gfp_flags)
411 {
412         struct nfs4_state_owner *sp;
413
414         sp = kzalloc(sizeof(*sp), gfp_flags);
415         if (!sp)
416                 return NULL;
417         sp->so_server = server;
418         sp->so_cred = get_rpccred(cred);
419         spin_lock_init(&sp->so_lock);
420         INIT_LIST_HEAD(&sp->so_states);
421         nfs4_init_seqid_counter(&sp->so_seqid);
422         atomic_set(&sp->so_count, 1);
423         INIT_LIST_HEAD(&sp->so_lru);
424         return sp;
425 }
426
427 static void
428 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
429 {
430         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
431                 struct nfs_server *server = sp->so_server;
432                 struct nfs_client *clp = server->nfs_client;
433
434                 spin_lock(&clp->cl_lock);
435                 rb_erase(&sp->so_server_node, &server->state_owners);
436                 RB_CLEAR_NODE(&sp->so_server_node);
437                 spin_unlock(&clp->cl_lock);
438         }
439 }
440
441 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
442 {
443         nfs4_destroy_seqid_counter(&sp->so_seqid);
444         put_rpccred(sp->so_cred);
445         kfree(sp);
446 }
447
448 static void nfs4_gc_state_owners(struct nfs_server *server)
449 {
450         struct nfs_client *clp = server->nfs_client;
451         struct nfs4_state_owner *sp, *tmp;
452         unsigned long time_min, time_max;
453         LIST_HEAD(doomed);
454
455         spin_lock(&clp->cl_lock);
456         time_max = jiffies;
457         time_min = (long)time_max - (long)clp->cl_lease_time;
458         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
459                 /* NB: LRU is sorted so that oldest is at the head */
460                 if (time_in_range(sp->so_expires, time_min, time_max))
461                         break;
462                 list_move(&sp->so_lru, &doomed);
463                 nfs4_remove_state_owner_locked(sp);
464         }
465         spin_unlock(&clp->cl_lock);
466
467         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
468                 list_del(&sp->so_lru);
469                 nfs4_free_state_owner(sp);
470         }
471 }
472
473 /**
474  * nfs4_get_state_owner - Look up a state owner given a credential
475  * @server: nfs_server to search
476  * @cred: RPC credential to match
477  *
478  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
479  */
480 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
481                                               struct rpc_cred *cred,
482                                               gfp_t gfp_flags)
483 {
484         struct nfs_client *clp = server->nfs_client;
485         struct nfs4_state_owner *sp, *new;
486
487         spin_lock(&clp->cl_lock);
488         sp = nfs4_find_state_owner_locked(server, cred);
489         spin_unlock(&clp->cl_lock);
490         if (sp != NULL)
491                 goto out;
492         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
493         if (new == NULL)
494                 goto out;
495         do {
496                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
497                         break;
498                 spin_lock(&clp->cl_lock);
499                 sp = nfs4_insert_state_owner_locked(new);
500                 spin_unlock(&clp->cl_lock);
501         } while (sp == ERR_PTR(-EAGAIN));
502         if (sp != new)
503                 nfs4_free_state_owner(new);
504 out:
505         nfs4_gc_state_owners(server);
506         return sp;
507 }
508
509 /**
510  * nfs4_put_state_owner - Release a nfs4_state_owner
511  * @sp: state owner data to release
512  */
513 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
514 {
515         struct nfs_server *server = sp->so_server;
516         struct nfs_client *clp = server->nfs_client;
517
518         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
519                 return;
520
521         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
522                 sp->so_expires = jiffies;
523                 list_add_tail(&sp->so_lru, &server->state_owners_lru);
524                 spin_unlock(&clp->cl_lock);
525         } else {
526                 nfs4_remove_state_owner_locked(sp);
527                 spin_unlock(&clp->cl_lock);
528                 nfs4_free_state_owner(sp);
529         }
530 }
531
532 /**
533  * nfs4_purge_state_owners - Release all cached state owners
534  * @server: nfs_server with cached state owners to release
535  *
536  * Called at umount time.  Remaining state owners will be on
537  * the LRU with ref count of zero.
538  */
539 void nfs4_purge_state_owners(struct nfs_server *server)
540 {
541         struct nfs_client *clp = server->nfs_client;
542         struct nfs4_state_owner *sp, *tmp;
543         LIST_HEAD(doomed);
544
545         spin_lock(&clp->cl_lock);
546         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
547                 list_move(&sp->so_lru, &doomed);
548                 nfs4_remove_state_owner_locked(sp);
549         }
550         spin_unlock(&clp->cl_lock);
551
552         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
553                 list_del(&sp->so_lru);
554                 nfs4_free_state_owner(sp);
555         }
556 }
557
558 static struct nfs4_state *
559 nfs4_alloc_open_state(void)
560 {
561         struct nfs4_state *state;
562
563         state = kzalloc(sizeof(*state), GFP_NOFS);
564         if (!state)
565                 return NULL;
566         atomic_set(&state->count, 1);
567         INIT_LIST_HEAD(&state->lock_states);
568         spin_lock_init(&state->state_lock);
569         seqlock_init(&state->seqlock);
570         return state;
571 }
572
573 void
574 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
575 {
576         if (state->state == fmode)
577                 return;
578         /* NB! List reordering - see the reclaim code for why.  */
579         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
580                 if (fmode & FMODE_WRITE)
581                         list_move(&state->open_states, &state->owner->so_states);
582                 else
583                         list_move_tail(&state->open_states, &state->owner->so_states);
584         }
585         state->state = fmode;
586 }
587
588 static struct nfs4_state *
589 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
590 {
591         struct nfs_inode *nfsi = NFS_I(inode);
592         struct nfs4_state *state;
593
594         list_for_each_entry(state, &nfsi->open_states, inode_states) {
595                 if (state->owner != owner)
596                         continue;
597                 if (atomic_inc_not_zero(&state->count))
598                         return state;
599         }
600         return NULL;
601 }
602
603 static void
604 nfs4_free_open_state(struct nfs4_state *state)
605 {
606         kfree(state);
607 }
608
609 struct nfs4_state *
610 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
611 {
612         struct nfs4_state *state, *new;
613         struct nfs_inode *nfsi = NFS_I(inode);
614
615         spin_lock(&inode->i_lock);
616         state = __nfs4_find_state_byowner(inode, owner);
617         spin_unlock(&inode->i_lock);
618         if (state)
619                 goto out;
620         new = nfs4_alloc_open_state();
621         spin_lock(&owner->so_lock);
622         spin_lock(&inode->i_lock);
623         state = __nfs4_find_state_byowner(inode, owner);
624         if (state == NULL && new != NULL) {
625                 state = new;
626                 state->owner = owner;
627                 atomic_inc(&owner->so_count);
628                 list_add(&state->inode_states, &nfsi->open_states);
629                 ihold(inode);
630                 state->inode = inode;
631                 spin_unlock(&inode->i_lock);
632                 /* Note: The reclaim code dictates that we add stateless
633                  * and read-only stateids to the end of the list */
634                 list_add_tail(&state->open_states, &owner->so_states);
635                 spin_unlock(&owner->so_lock);
636         } else {
637                 spin_unlock(&inode->i_lock);
638                 spin_unlock(&owner->so_lock);
639                 if (new)
640                         nfs4_free_open_state(new);
641         }
642 out:
643         return state;
644 }
645
646 void nfs4_put_open_state(struct nfs4_state *state)
647 {
648         struct inode *inode = state->inode;
649         struct nfs4_state_owner *owner = state->owner;
650
651         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
652                 return;
653         spin_lock(&inode->i_lock);
654         list_del(&state->inode_states);
655         list_del(&state->open_states);
656         spin_unlock(&inode->i_lock);
657         spin_unlock(&owner->so_lock);
658         iput(inode);
659         nfs4_free_open_state(state);
660         nfs4_put_state_owner(owner);
661 }
662
663 /*
664  * Close the current file.
665  */
666 static void __nfs4_close(struct nfs4_state *state,
667                 fmode_t fmode, gfp_t gfp_mask, int wait)
668 {
669         struct nfs4_state_owner *owner = state->owner;
670         int call_close = 0;
671         fmode_t newstate;
672
673         atomic_inc(&owner->so_count);
674         /* Protect against nfs4_find_state() */
675         spin_lock(&owner->so_lock);
676         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
677                 case FMODE_READ:
678                         state->n_rdonly--;
679                         break;
680                 case FMODE_WRITE:
681                         state->n_wronly--;
682                         break;
683                 case FMODE_READ|FMODE_WRITE:
684                         state->n_rdwr--;
685         }
686         newstate = FMODE_READ|FMODE_WRITE;
687         if (state->n_rdwr == 0) {
688                 if (state->n_rdonly == 0) {
689                         newstate &= ~FMODE_READ;
690                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
691                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
692                 }
693                 if (state->n_wronly == 0) {
694                         newstate &= ~FMODE_WRITE;
695                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
696                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
697                 }
698                 if (newstate == 0)
699                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
700         }
701         nfs4_state_set_mode_locked(state, newstate);
702         spin_unlock(&owner->so_lock);
703
704         if (!call_close) {
705                 nfs4_put_open_state(state);
706                 nfs4_put_state_owner(owner);
707         } else {
708                 bool roc = pnfs_roc(state->inode);
709
710                 nfs4_do_close(state, gfp_mask, wait, roc);
711         }
712 }
713
714 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
715 {
716         __nfs4_close(state, fmode, GFP_NOFS, 0);
717 }
718
719 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
720 {
721         __nfs4_close(state, fmode, GFP_KERNEL, 1);
722 }
723
724 /*
725  * Search the state->lock_states for an existing lock_owner
726  * that is compatible with current->files
727  */
728 static struct nfs4_lock_state *
729 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
730 {
731         struct nfs4_lock_state *pos;
732         list_for_each_entry(pos, &state->lock_states, ls_locks) {
733                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
734                         continue;
735                 switch (pos->ls_owner.lo_type) {
736                 case NFS4_POSIX_LOCK_TYPE:
737                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
738                                 continue;
739                         break;
740                 case NFS4_FLOCK_LOCK_TYPE:
741                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
742                                 continue;
743                 }
744                 atomic_inc(&pos->ls_count);
745                 return pos;
746         }
747         return NULL;
748 }
749
750 /*
751  * Return a compatible lock_state. If no initialized lock_state structure
752  * exists, return an uninitialized one.
753  *
754  */
755 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)
756 {
757         struct nfs4_lock_state *lsp;
758         struct nfs_server *server = state->owner->so_server;
759
760         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
761         if (lsp == NULL)
762                 return NULL;
763         nfs4_init_seqid_counter(&lsp->ls_seqid);
764         atomic_set(&lsp->ls_count, 1);
765         lsp->ls_state = state;
766         lsp->ls_owner.lo_type = type;
767         switch (lsp->ls_owner.lo_type) {
768         case NFS4_FLOCK_LOCK_TYPE:
769                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
770                 break;
771         case NFS4_POSIX_LOCK_TYPE:
772                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
773                 break;
774         default:
775                 goto out_free;
776         }
777         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
778         if (lsp->ls_seqid.owner_id < 0)
779                 goto out_free;
780         INIT_LIST_HEAD(&lsp->ls_locks);
781         return lsp;
782 out_free:
783         kfree(lsp);
784         return NULL;
785 }
786
787 void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
788 {
789         struct nfs_server *server = lsp->ls_state->owner->so_server;
790
791         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
792         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
793         kfree(lsp);
794 }
795
796 /*
797  * Return a compatible lock_state. If no initialized lock_state structure
798  * exists, return an uninitialized one.
799  *
800  */
801 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
802 {
803         struct nfs4_lock_state *lsp, *new = NULL;
804         
805         for(;;) {
806                 spin_lock(&state->state_lock);
807                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
808                 if (lsp != NULL)
809                         break;
810                 if (new != NULL) {
811                         list_add(&new->ls_locks, &state->lock_states);
812                         set_bit(LK_STATE_IN_USE, &state->flags);
813                         lsp = new;
814                         new = NULL;
815                         break;
816                 }
817                 spin_unlock(&state->state_lock);
818                 new = nfs4_alloc_lock_state(state, owner, pid, type);
819                 if (new == NULL)
820                         return NULL;
821         }
822         spin_unlock(&state->state_lock);
823         if (new != NULL)
824                 nfs4_free_lock_state(new);
825         return lsp;
826 }
827
828 /*
829  * Release reference to lock_state, and free it if we see that
830  * it is no longer in use
831  */
832 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
833 {
834         struct nfs4_state *state;
835
836         if (lsp == NULL)
837                 return;
838         state = lsp->ls_state;
839         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
840                 return;
841         list_del(&lsp->ls_locks);
842         if (list_empty(&state->lock_states))
843                 clear_bit(LK_STATE_IN_USE, &state->flags);
844         spin_unlock(&state->state_lock);
845         if (lsp->ls_flags & NFS_LOCK_INITIALIZED) {
846                 if (nfs4_release_lockowner(lsp) == 0)
847                         return;
848         }
849         nfs4_free_lock_state(lsp);
850 }
851
852 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
853 {
854         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
855
856         dst->fl_u.nfs4_fl.owner = lsp;
857         atomic_inc(&lsp->ls_count);
858 }
859
860 static void nfs4_fl_release_lock(struct file_lock *fl)
861 {
862         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
863 }
864
865 static const struct file_lock_operations nfs4_fl_lock_ops = {
866         .fl_copy_lock = nfs4_fl_copy_lock,
867         .fl_release_private = nfs4_fl_release_lock,
868 };
869
870 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
871 {
872         struct nfs4_lock_state *lsp;
873
874         if (fl->fl_ops != NULL)
875                 return 0;
876         if (fl->fl_flags & FL_POSIX)
877                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
878         else if (fl->fl_flags & FL_FLOCK)
879                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
880         else
881                 return -EINVAL;
882         if (lsp == NULL)
883                 return -ENOMEM;
884         fl->fl_u.nfs4_fl.owner = lsp;
885         fl->fl_ops = &nfs4_fl_lock_ops;
886         return 0;
887 }
888
889 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
890                 fl_owner_t fl_owner, pid_t fl_pid)
891 {
892         struct nfs4_lock_state *lsp;
893         bool ret = false;
894
895         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
896                 goto out;
897
898         spin_lock(&state->state_lock);
899         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
900         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0) {
901                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
902                 ret = true;
903         }
904         spin_unlock(&state->state_lock);
905         nfs4_put_lock_state(lsp);
906 out:
907         return ret;
908 }
909
910 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
911 {
912         int seq;
913
914         do {
915                 seq = read_seqbegin(&state->seqlock);
916                 nfs4_stateid_copy(dst, &state->stateid);
917         } while (read_seqretry(&state->seqlock, seq));
918 }
919
920 /*
921  * Byte-range lock aware utility to initialize the stateid of read/write
922  * requests.
923  */
924 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
925                 fmode_t fmode, fl_owner_t fl_owner, pid_t fl_pid)
926 {
927         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
928                 return;
929         if (nfs4_copy_lock_stateid(dst, state, fl_owner, fl_pid))
930                 return;
931         nfs4_copy_open_stateid(dst, state);
932 }
933
934 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
935 {
936         struct nfs_seqid *new;
937
938         new = kmalloc(sizeof(*new), gfp_mask);
939         if (new != NULL) {
940                 new->sequence = counter;
941                 INIT_LIST_HEAD(&new->list);
942                 new->task = NULL;
943         }
944         return new;
945 }
946
947 void nfs_release_seqid(struct nfs_seqid *seqid)
948 {
949         struct nfs_seqid_counter *sequence;
950
951         if (list_empty(&seqid->list))
952                 return;
953         sequence = seqid->sequence;
954         spin_lock(&sequence->lock);
955         list_del_init(&seqid->list);
956         if (!list_empty(&sequence->list)) {
957                 struct nfs_seqid *next;
958
959                 next = list_first_entry(&sequence->list,
960                                 struct nfs_seqid, list);
961                 rpc_wake_up_queued_task(&sequence->wait, next->task);
962         }
963         spin_unlock(&sequence->lock);
964 }
965
966 void nfs_free_seqid(struct nfs_seqid *seqid)
967 {
968         nfs_release_seqid(seqid);
969         kfree(seqid);
970 }
971
972 /*
973  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
974  * failed with a seqid incrementing error -
975  * see comments nfs_fs.h:seqid_mutating_error()
976  */
977 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
978 {
979         BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
980         switch (status) {
981                 case 0:
982                         break;
983                 case -NFS4ERR_BAD_SEQID:
984                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
985                                 return;
986                         printk(KERN_WARNING "NFS: v4 server returned a bad"
987                                         " sequence-id error on an"
988                                         " unconfirmed sequence %p!\n",
989                                         seqid->sequence);
990                 case -NFS4ERR_STALE_CLIENTID:
991                 case -NFS4ERR_STALE_STATEID:
992                 case -NFS4ERR_BAD_STATEID:
993                 case -NFS4ERR_BADXDR:
994                 case -NFS4ERR_RESOURCE:
995                 case -NFS4ERR_NOFILEHANDLE:
996                         /* Non-seqid mutating errors */
997                         return;
998         };
999         /*
1000          * Note: no locking needed as we are guaranteed to be first
1001          * on the sequence list
1002          */
1003         seqid->sequence->counter++;
1004 }
1005
1006 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1007 {
1008         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1009                                         struct nfs4_state_owner, so_seqid);
1010         struct nfs_server *server = sp->so_server;
1011
1012         if (status == -NFS4ERR_BAD_SEQID)
1013                 nfs4_drop_state_owner(sp);
1014         if (!nfs4_has_session(server->nfs_client))
1015                 nfs_increment_seqid(status, seqid);
1016 }
1017
1018 /*
1019  * Increment the seqid if the LOCK/LOCKU succeeded, or
1020  * failed with a seqid incrementing error -
1021  * see comments nfs_fs.h:seqid_mutating_error()
1022  */
1023 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1024 {
1025         nfs_increment_seqid(status, seqid);
1026 }
1027
1028 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1029 {
1030         struct nfs_seqid_counter *sequence = seqid->sequence;
1031         int status = 0;
1032
1033         spin_lock(&sequence->lock);
1034         seqid->task = task;
1035         if (list_empty(&seqid->list))
1036                 list_add_tail(&seqid->list, &sequence->list);
1037         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1038                 goto unlock;
1039         rpc_sleep_on(&sequence->wait, task, NULL);
1040         status = -EAGAIN;
1041 unlock:
1042         spin_unlock(&sequence->lock);
1043         return status;
1044 }
1045
1046 static int nfs4_run_state_manager(void *);
1047
1048 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1049 {
1050         smp_mb__before_clear_bit();
1051         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1052         smp_mb__after_clear_bit();
1053         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1054         rpc_wake_up(&clp->cl_rpcwaitq);
1055 }
1056
1057 /*
1058  * Schedule the nfs_client asynchronous state management routine
1059  */
1060 void nfs4_schedule_state_manager(struct nfs_client *clp)
1061 {
1062         struct task_struct *task;
1063         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1064
1065         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1066                 return;
1067         __module_get(THIS_MODULE);
1068         atomic_inc(&clp->cl_count);
1069
1070         /* The rcu_read_lock() is not strictly necessary, as the state
1071          * manager is the only thread that ever changes the rpc_xprt
1072          * after it's initialized.  At this point, we're single threaded. */
1073         rcu_read_lock();
1074         snprintf(buf, sizeof(buf), "%s-manager",
1075                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1076         rcu_read_unlock();
1077         task = kthread_run(nfs4_run_state_manager, clp, buf);
1078         if (IS_ERR(task)) {
1079                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1080                         __func__, PTR_ERR(task));
1081                 nfs4_clear_state_manager_bit(clp);
1082                 nfs_put_client(clp);
1083                 module_put(THIS_MODULE);
1084         }
1085 }
1086
1087 /*
1088  * Schedule a lease recovery attempt
1089  */
1090 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1091 {
1092         if (!clp)
1093                 return;
1094         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1095                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1096         nfs4_schedule_state_manager(clp);
1097 }
1098 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
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 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1137
1138 void nfs_inode_find_state_and_recover(struct inode *inode,
1139                 const nfs4_stateid *stateid)
1140 {
1141         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1142         struct nfs_inode *nfsi = NFS_I(inode);
1143         struct nfs_open_context *ctx;
1144         struct nfs4_state *state;
1145         bool found = false;
1146
1147         spin_lock(&inode->i_lock);
1148         list_for_each_entry(ctx, &nfsi->open_files, list) {
1149                 state = ctx->state;
1150                 if (state == NULL)
1151                         continue;
1152                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1153                         continue;
1154                 if (!nfs4_stateid_match(&state->stateid, stateid))
1155                         continue;
1156                 nfs4_state_mark_reclaim_nograce(clp, state);
1157                 found = true;
1158         }
1159         spin_unlock(&inode->i_lock);
1160         if (found)
1161                 nfs4_schedule_state_manager(clp);
1162 }
1163
1164
1165 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1166 {
1167         struct inode *inode = state->inode;
1168         struct nfs_inode *nfsi = NFS_I(inode);
1169         struct file_lock *fl;
1170         int status = 0;
1171
1172         if (inode->i_flock == NULL)
1173                 return 0;
1174
1175         /* Guard against delegation returns and new lock/unlock calls */
1176         down_write(&nfsi->rwsem);
1177         /* Protect inode->i_flock using the BKL */
1178         lock_flocks();
1179         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1180                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1181                         continue;
1182                 if (nfs_file_open_context(fl->fl_file)->state != state)
1183                         continue;
1184                 unlock_flocks();
1185                 status = ops->recover_lock(state, fl);
1186                 switch (status) {
1187                         case 0:
1188                                 break;
1189                         case -ESTALE:
1190                         case -NFS4ERR_ADMIN_REVOKED:
1191                         case -NFS4ERR_STALE_STATEID:
1192                         case -NFS4ERR_BAD_STATEID:
1193                         case -NFS4ERR_EXPIRED:
1194                         case -NFS4ERR_NO_GRACE:
1195                         case -NFS4ERR_STALE_CLIENTID:
1196                         case -NFS4ERR_BADSESSION:
1197                         case -NFS4ERR_BADSLOT:
1198                         case -NFS4ERR_BAD_HIGH_SLOT:
1199                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1200                                 goto out;
1201                         default:
1202                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1203                                         "Zeroing state\n", __func__, status);
1204                         case -ENOMEM:
1205                         case -NFS4ERR_DENIED:
1206                         case -NFS4ERR_RECLAIM_BAD:
1207                         case -NFS4ERR_RECLAIM_CONFLICT:
1208                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1209                                 status = 0;
1210                 }
1211                 lock_flocks();
1212         }
1213         unlock_flocks();
1214 out:
1215         up_write(&nfsi->rwsem);
1216         return status;
1217 }
1218
1219 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1220 {
1221         struct nfs4_state *state;
1222         struct nfs4_lock_state *lock;
1223         int status = 0;
1224
1225         /* Note: we rely on the sp->so_states list being ordered 
1226          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1227          * states first.
1228          * This is needed to ensure that the server won't give us any
1229          * read delegations that we have to return if, say, we are
1230          * recovering after a network partition or a reboot from a
1231          * server that doesn't support a grace period.
1232          */
1233 restart:
1234         spin_lock(&sp->so_lock);
1235         list_for_each_entry(state, &sp->so_states, open_states) {
1236                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1237                         continue;
1238                 if (state->state == 0)
1239                         continue;
1240                 atomic_inc(&state->count);
1241                 spin_unlock(&sp->so_lock);
1242                 status = ops->recover_open(sp, state);
1243                 if (status >= 0) {
1244                         status = nfs4_reclaim_locks(state, ops);
1245                         if (status >= 0) {
1246                                 spin_lock(&state->state_lock);
1247                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1248                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1249                                                 printk("NFS: %s: Lock reclaim "
1250                                                         "failed!\n", __func__);
1251                                 }
1252                                 spin_unlock(&state->state_lock);
1253                                 nfs4_put_open_state(state);
1254                                 goto restart;
1255                         }
1256                 }
1257                 switch (status) {
1258                         default:
1259                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1260                                         "Zeroing state\n", __func__, status);
1261                         case -ENOENT:
1262                         case -ENOMEM:
1263                         case -ESTALE:
1264                                 /*
1265                                  * Open state on this file cannot be recovered
1266                                  * All we can do is revert to using the zero stateid.
1267                                  */
1268                                 memset(&state->stateid, 0,
1269                                         sizeof(state->stateid));
1270                                 /* Mark the file as being 'closed' */
1271                                 state->state = 0;
1272                                 break;
1273                         case -EKEYEXPIRED:
1274                                 /*
1275                                  * User RPCSEC_GSS context has expired.
1276                                  * We cannot recover this stateid now, so
1277                                  * skip it and allow recovery thread to
1278                                  * proceed.
1279                                  */
1280                                 break;
1281                         case -NFS4ERR_ADMIN_REVOKED:
1282                         case -NFS4ERR_STALE_STATEID:
1283                         case -NFS4ERR_BAD_STATEID:
1284                         case -NFS4ERR_RECLAIM_BAD:
1285                         case -NFS4ERR_RECLAIM_CONFLICT:
1286                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1287                                 break;
1288                         case -NFS4ERR_EXPIRED:
1289                         case -NFS4ERR_NO_GRACE:
1290                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1291                         case -NFS4ERR_STALE_CLIENTID:
1292                         case -NFS4ERR_BADSESSION:
1293                         case -NFS4ERR_BADSLOT:
1294                         case -NFS4ERR_BAD_HIGH_SLOT:
1295                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1296                                 goto out_err;
1297                 }
1298                 nfs4_put_open_state(state);
1299                 goto restart;
1300         }
1301         spin_unlock(&sp->so_lock);
1302         return 0;
1303 out_err:
1304         nfs4_put_open_state(state);
1305         return status;
1306 }
1307
1308 static void nfs4_clear_open_state(struct nfs4_state *state)
1309 {
1310         struct nfs4_lock_state *lock;
1311
1312         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1313         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1314         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1315         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1316         spin_lock(&state->state_lock);
1317         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1318                 lock->ls_seqid.flags = 0;
1319                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1320         }
1321         spin_unlock(&state->state_lock);
1322 }
1323
1324 static void nfs4_reset_seqids(struct nfs_server *server,
1325         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1326 {
1327         struct nfs_client *clp = server->nfs_client;
1328         struct nfs4_state_owner *sp;
1329         struct rb_node *pos;
1330         struct nfs4_state *state;
1331
1332         spin_lock(&clp->cl_lock);
1333         for (pos = rb_first(&server->state_owners);
1334              pos != NULL;
1335              pos = rb_next(pos)) {
1336                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1337                 sp->so_seqid.flags = 0;
1338                 spin_lock(&sp->so_lock);
1339                 list_for_each_entry(state, &sp->so_states, open_states) {
1340                         if (mark_reclaim(clp, state))
1341                                 nfs4_clear_open_state(state);
1342                 }
1343                 spin_unlock(&sp->so_lock);
1344         }
1345         spin_unlock(&clp->cl_lock);
1346 }
1347
1348 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1349         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1350 {
1351         struct nfs_server *server;
1352
1353         rcu_read_lock();
1354         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1355                 nfs4_reset_seqids(server, mark_reclaim);
1356         rcu_read_unlock();
1357 }
1358
1359 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1360 {
1361         /* Mark all delegations for reclaim */
1362         nfs_delegation_mark_reclaim(clp);
1363         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1364 }
1365
1366 static void nfs4_reclaim_complete(struct nfs_client *clp,
1367                                  const struct nfs4_state_recovery_ops *ops)
1368 {
1369         /* Notify the server we're done reclaiming our state */
1370         if (ops->reclaim_complete)
1371                 (void)ops->reclaim_complete(clp);
1372 }
1373
1374 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1375 {
1376         struct nfs_client *clp = server->nfs_client;
1377         struct nfs4_state_owner *sp;
1378         struct rb_node *pos;
1379         struct nfs4_state *state;
1380
1381         spin_lock(&clp->cl_lock);
1382         for (pos = rb_first(&server->state_owners);
1383              pos != NULL;
1384              pos = rb_next(pos)) {
1385                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1386                 spin_lock(&sp->so_lock);
1387                 list_for_each_entry(state, &sp->so_states, open_states) {
1388                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1389                                                 &state->flags))
1390                                 continue;
1391                         nfs4_state_mark_reclaim_nograce(clp, state);
1392                 }
1393                 spin_unlock(&sp->so_lock);
1394         }
1395         spin_unlock(&clp->cl_lock);
1396 }
1397
1398 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1399 {
1400         struct nfs_server *server;
1401
1402         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1403                 return 0;
1404
1405         rcu_read_lock();
1406         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1407                 nfs4_clear_reclaim_server(server);
1408         rcu_read_unlock();
1409
1410         nfs_delegation_reap_unclaimed(clp);
1411         return 1;
1412 }
1413
1414 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1415 {
1416         if (!nfs4_state_clear_reclaim_reboot(clp))
1417                 return;
1418         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1419 }
1420
1421 static void nfs_delegation_clear_all(struct nfs_client *clp)
1422 {
1423         nfs_delegation_mark_reclaim(clp);
1424         nfs_delegation_reap_unclaimed(clp);
1425 }
1426
1427 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1428 {
1429         nfs_delegation_clear_all(clp);
1430         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1431 }
1432
1433 static void nfs4_warn_keyexpired(const char *s)
1434 {
1435         printk_ratelimited(KERN_WARNING "Error: state manager"
1436                         " encountered RPCSEC_GSS session"
1437                         " expired against NFSv4 server %s.\n",
1438                         s);
1439 }
1440
1441 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1442 {
1443         switch (error) {
1444                 case 0:
1445                         break;
1446                 case -NFS4ERR_CB_PATH_DOWN:
1447                         nfs_handle_cb_pathdown(clp);
1448                         break;
1449                 case -NFS4ERR_NO_GRACE:
1450                         nfs4_state_end_reclaim_reboot(clp);
1451                         break;
1452                 case -NFS4ERR_STALE_CLIENTID:
1453                 case -NFS4ERR_LEASE_MOVED:
1454                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1455                         nfs4_state_clear_reclaim_reboot(clp);
1456                         nfs4_state_start_reclaim_reboot(clp);
1457                         break;
1458                 case -NFS4ERR_EXPIRED:
1459                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1460                         nfs4_state_start_reclaim_nograce(clp);
1461                         break;
1462                 case -NFS4ERR_BADSESSION:
1463                 case -NFS4ERR_BADSLOT:
1464                 case -NFS4ERR_BAD_HIGH_SLOT:
1465                 case -NFS4ERR_DEADSESSION:
1466                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1467                 case -NFS4ERR_SEQ_FALSE_RETRY:
1468                 case -NFS4ERR_SEQ_MISORDERED:
1469                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1470                         /* Zero session reset errors */
1471                         break;
1472                 case -EKEYEXPIRED:
1473                         /* Nothing we can do */
1474                         nfs4_warn_keyexpired(clp->cl_hostname);
1475                         break;
1476                 default:
1477                         return error;
1478         }
1479         return 0;
1480 }
1481
1482 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1483 {
1484         struct nfs4_state_owner *sp;
1485         struct nfs_server *server;
1486         struct rb_node *pos;
1487         int status = 0;
1488
1489 restart:
1490         rcu_read_lock();
1491         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1492                 nfs4_purge_state_owners(server);
1493                 spin_lock(&clp->cl_lock);
1494                 for (pos = rb_first(&server->state_owners);
1495                      pos != NULL;
1496                      pos = rb_next(pos)) {
1497                         sp = rb_entry(pos,
1498                                 struct nfs4_state_owner, so_server_node);
1499                         if (!test_and_clear_bit(ops->owner_flag_bit,
1500                                                         &sp->so_flags))
1501                                 continue;
1502                         atomic_inc(&sp->so_count);
1503                         spin_unlock(&clp->cl_lock);
1504                         rcu_read_unlock();
1505
1506                         status = nfs4_reclaim_open_state(sp, ops);
1507                         if (status < 0) {
1508                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1509                                 nfs4_put_state_owner(sp);
1510                                 return nfs4_recovery_handle_error(clp, status);
1511                         }
1512
1513                         nfs4_put_state_owner(sp);
1514                         goto restart;
1515                 }
1516                 spin_unlock(&clp->cl_lock);
1517         }
1518         rcu_read_unlock();
1519         return status;
1520 }
1521
1522 static int nfs4_check_lease(struct nfs_client *clp)
1523 {
1524         struct rpc_cred *cred;
1525         const struct nfs4_state_maintenance_ops *ops =
1526                 clp->cl_mvops->state_renewal_ops;
1527         int status;
1528
1529         /* Is the client already known to have an expired lease? */
1530         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1531                 return 0;
1532         spin_lock(&clp->cl_lock);
1533         cred = ops->get_state_renewal_cred_locked(clp);
1534         spin_unlock(&clp->cl_lock);
1535         if (cred == NULL) {
1536                 cred = nfs4_get_setclientid_cred(clp);
1537                 status = -ENOKEY;
1538                 if (cred == NULL)
1539                         goto out;
1540         }
1541         status = ops->renew_lease(clp, cred);
1542         put_rpccred(cred);
1543 out:
1544         return nfs4_recovery_handle_error(clp, status);
1545 }
1546
1547 static int nfs4_reclaim_lease(struct nfs_client *clp)
1548 {
1549         struct rpc_cred *cred;
1550         const struct nfs4_state_recovery_ops *ops =
1551                 clp->cl_mvops->reboot_recovery_ops;
1552         int status = -ENOENT;
1553
1554         cred = ops->get_clid_cred(clp);
1555         if (cred != NULL) {
1556                 status = ops->establish_clid(clp, cred);
1557                 put_rpccred(cred);
1558                 /* Handle case where the user hasn't set up machine creds */
1559                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1560                         nfs4_clear_machine_cred(clp);
1561                         status = -EAGAIN;
1562                 }
1563                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1564                         status = -EPROTONOSUPPORT;
1565         }
1566         return status;
1567 }
1568
1569 #ifdef CONFIG_NFS_V4_1
1570 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1571 {
1572         struct nfs_client *clp = session->clp;
1573
1574         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1575         nfs4_schedule_lease_recovery(clp);
1576 }
1577 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1578
1579 void nfs41_handle_recall_slot(struct nfs_client *clp)
1580 {
1581         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1582         nfs4_schedule_state_manager(clp);
1583 }
1584
1585 static void nfs4_reset_all_state(struct nfs_client *clp)
1586 {
1587         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1588                 clp->cl_boot_time = CURRENT_TIME;
1589                 nfs4_state_start_reclaim_nograce(clp);
1590                 nfs4_schedule_state_manager(clp);
1591         }
1592 }
1593
1594 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1595 {
1596         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1597                 nfs4_state_start_reclaim_reboot(clp);
1598                 nfs4_schedule_state_manager(clp);
1599         }
1600 }
1601
1602 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1603 {
1604         /* Temporary */
1605         nfs4_reset_all_state(clp);
1606 }
1607
1608 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1609 {
1610         /* This will need to handle layouts too */
1611         nfs_expire_all_delegations(clp);
1612 }
1613
1614 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1615 {
1616         nfs_expire_all_delegations(clp);
1617         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1618                 nfs4_schedule_state_manager(clp);
1619 }
1620
1621 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1622 {
1623         if (!flags)
1624                 return;
1625         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1626                 nfs41_handle_server_reboot(clp);
1627         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1628                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1629                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1630                             SEQ4_STATUS_LEASE_MOVED))
1631                 nfs41_handle_state_revoked(clp);
1632         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1633                 nfs41_handle_recallable_state_revoked(clp);
1634         if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1635                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1636                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1637                 nfs41_handle_cb_path_down(clp);
1638 }
1639
1640 static int nfs4_reset_session(struct nfs_client *clp)
1641 {
1642         int status;
1643
1644         nfs4_begin_drain_session(clp);
1645         status = nfs4_proc_destroy_session(clp->cl_session);
1646         if (status && status != -NFS4ERR_BADSESSION &&
1647             status != -NFS4ERR_DEADSESSION) {
1648                 status = nfs4_recovery_handle_error(clp, status);
1649                 goto out;
1650         }
1651
1652         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1653         status = nfs4_proc_create_session(clp);
1654         if (status) {
1655                 status = nfs4_recovery_handle_error(clp, status);
1656                 goto out;
1657         }
1658         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1659         /* create_session negotiated new slot table */
1660         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1661
1662          /* Let the state manager reestablish state */
1663         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1664                 nfs41_setup_state_renewal(clp);
1665 out:
1666         return status;
1667 }
1668
1669 static int nfs4_recall_slot(struct nfs_client *clp)
1670 {
1671         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1672         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1673         struct nfs4_slot *new, *old;
1674         int i;
1675
1676         nfs4_begin_drain_session(clp);
1677         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1678                       GFP_NOFS);
1679         if (!new)
1680                 return -ENOMEM;
1681
1682         spin_lock(&fc_tbl->slot_tbl_lock);
1683         for (i = 0; i < fc_tbl->target_max_slots; i++)
1684                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1685         old = fc_tbl->slots;
1686         fc_tbl->slots = new;
1687         fc_tbl->max_slots = fc_tbl->target_max_slots;
1688         fc_tbl->target_max_slots = 0;
1689         fc_attrs->max_reqs = fc_tbl->max_slots;
1690         spin_unlock(&fc_tbl->slot_tbl_lock);
1691
1692         kfree(old);
1693         nfs4_end_drain_session(clp);
1694         return 0;
1695 }
1696
1697 #else /* CONFIG_NFS_V4_1 */
1698 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1699 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1700 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1701 #endif /* CONFIG_NFS_V4_1 */
1702
1703 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1704  * on EXCHANGE_ID for v4.1
1705  */
1706 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1707 {
1708         switch (status) {
1709         case -NFS4ERR_CLID_INUSE:
1710         case -NFS4ERR_STALE_CLIENTID:
1711                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1712                 break;
1713         case -NFS4ERR_DELAY:
1714         case -ETIMEDOUT:
1715         case -EAGAIN:
1716                 ssleep(1);
1717                 break;
1718
1719         case -EKEYEXPIRED:
1720                 nfs4_warn_keyexpired(clp->cl_hostname);
1721         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1722                                  * in nfs4_exchange_id */
1723         default:
1724                 return;
1725         }
1726         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1727 }
1728
1729 static void nfs4_state_manager(struct nfs_client *clp)
1730 {
1731         int status = 0;
1732
1733         /* Ensure exclusive access to NFSv4 state */
1734         do {
1735                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1736                         /* We're going to have to re-establish a clientid */
1737                         status = nfs4_reclaim_lease(clp);
1738                         if (status) {
1739                                 nfs4_set_lease_expired(clp, status);
1740                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1741                                                         &clp->cl_state))
1742                                         continue;
1743                                 if (clp->cl_cons_state ==
1744                                                         NFS_CS_SESSION_INITING)
1745                                         nfs_mark_client_ready(clp, status);
1746                                 goto out_error;
1747                         }
1748                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1749
1750                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1751                                                &clp->cl_state))
1752                                 nfs4_state_start_reclaim_nograce(clp);
1753                         else
1754                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1755                                         &clp->cl_state);
1756
1757                         pnfs_destroy_all_layouts(clp);
1758                 }
1759
1760                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1761                         status = nfs4_check_lease(clp);
1762                         if (status < 0)
1763                                 goto out_error;
1764                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1765                                 continue;
1766                 }
1767
1768                 /* Initialize or reset the session */
1769                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1770                    && nfs4_has_session(clp)) {
1771                         status = nfs4_reset_session(clp);
1772                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1773                                 continue;
1774                         if (status < 0)
1775                                 goto out_error;
1776                 }
1777
1778                 /* First recover reboot state... */
1779                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1780                         status = nfs4_do_reclaim(clp,
1781                                 clp->cl_mvops->reboot_recovery_ops);
1782                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1783                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1784                                 continue;
1785                         nfs4_state_end_reclaim_reboot(clp);
1786                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1787                                 continue;
1788                         if (status < 0)
1789                                 goto out_error;
1790                 }
1791
1792                 /* Now recover expired state... */
1793                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1794                         status = nfs4_do_reclaim(clp,
1795                                 clp->cl_mvops->nograce_recovery_ops);
1796                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1797                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1798                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1799                                 continue;
1800                         if (status < 0)
1801                                 goto out_error;
1802                 }
1803
1804                 nfs4_end_drain_session(clp);
1805                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1806                         nfs_client_return_marked_delegations(clp);
1807                         continue;
1808                 }
1809                 /* Recall session slots */
1810                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1811                    && nfs4_has_session(clp)) {
1812                         status = nfs4_recall_slot(clp);
1813                         if (status < 0)
1814                                 goto out_error;
1815                         continue;
1816                 }
1817
1818
1819                 nfs4_clear_state_manager_bit(clp);
1820                 /* Did we race with an attempt to give us more work? */
1821                 if (clp->cl_state == 0)
1822                         break;
1823                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1824                         break;
1825         } while (atomic_read(&clp->cl_count) > 1);
1826         return;
1827 out_error:
1828         printk(KERN_WARNING "NFS: state manager failed on NFSv4 server %s"
1829                         " with error %d\n", clp->cl_hostname, -status);
1830         nfs4_end_drain_session(clp);
1831         nfs4_clear_state_manager_bit(clp);
1832 }
1833
1834 static int nfs4_run_state_manager(void *ptr)
1835 {
1836         struct nfs_client *clp = ptr;
1837
1838         allow_signal(SIGKILL);
1839         nfs4_state_manager(clp);
1840         nfs_put_client(clp);
1841         module_put_and_exit(0);
1842         return 0;
1843 }
1844
1845 /*
1846  * Local variables:
1847  *  c-basic-offset: 8
1848  * End:
1849  */