Merge branch 'master' of git://1984.lsi.us.es/net
[platform/adaptation/renesas_rcar/renesas_kernel.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/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(void)
448 {
449         struct nfs4_state_owner *sp;
450
451         sp = kzalloc(sizeof(*sp),GFP_NOFS);
452         if (!sp)
453                 return NULL;
454         spin_lock_init(&sp->so_lock);
455         INIT_LIST_HEAD(&sp->so_states);
456         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
457         sp->so_seqid.sequence = &sp->so_sequence;
458         spin_lock_init(&sp->so_sequence.lock);
459         INIT_LIST_HEAD(&sp->so_sequence.list);
460         atomic_set(&sp->so_count, 1);
461         INIT_LIST_HEAD(&sp->so_lru);
462         return sp;
463 }
464
465 static void
466 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
467 {
468         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
469                 struct nfs_server *server = sp->so_server;
470                 struct nfs_client *clp = server->nfs_client;
471
472                 spin_lock(&clp->cl_lock);
473                 rb_erase(&sp->so_server_node, &server->state_owners);
474                 RB_CLEAR_NODE(&sp->so_server_node);
475                 spin_unlock(&clp->cl_lock);
476         }
477 }
478
479 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
480 {
481         rpc_destroy_wait_queue(&sp->so_sequence.wait);
482         put_rpccred(sp->so_cred);
483         kfree(sp);
484 }
485
486 static void nfs4_gc_state_owners(struct nfs_server *server)
487 {
488         struct nfs_client *clp = server->nfs_client;
489         struct nfs4_state_owner *sp, *tmp;
490         unsigned long time_min, time_max;
491         LIST_HEAD(doomed);
492
493         spin_lock(&clp->cl_lock);
494         time_max = jiffies;
495         time_min = (long)time_max - (long)clp->cl_lease_time;
496         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
497                 /* NB: LRU is sorted so that oldest is at the head */
498                 if (time_in_range(sp->so_expires, time_min, time_max))
499                         break;
500                 list_move(&sp->so_lru, &doomed);
501                 nfs4_remove_state_owner_locked(sp);
502         }
503         spin_unlock(&clp->cl_lock);
504
505         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
506                 list_del(&sp->so_lru);
507                 nfs4_free_state_owner(sp);
508         }
509 }
510
511 /**
512  * nfs4_get_state_owner - Look up a state owner given a credential
513  * @server: nfs_server to search
514  * @cred: RPC credential to match
515  *
516  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
517  */
518 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
519                                               struct rpc_cred *cred)
520 {
521         struct nfs_client *clp = server->nfs_client;
522         struct nfs4_state_owner *sp, *new;
523
524         spin_lock(&clp->cl_lock);
525         sp = nfs4_find_state_owner_locked(server, cred);
526         spin_unlock(&clp->cl_lock);
527         if (sp != NULL)
528                 goto out;
529         new = nfs4_alloc_state_owner();
530         if (new == NULL)
531                 goto out;
532         new->so_server = server;
533         new->so_cred = cred;
534         spin_lock(&clp->cl_lock);
535         sp = nfs4_insert_state_owner_locked(new);
536         spin_unlock(&clp->cl_lock);
537         if (sp == new)
538                 get_rpccred(cred);
539         else {
540                 rpc_destroy_wait_queue(&new->so_sequence.wait);
541                 kfree(new);
542         }
543 out:
544         nfs4_gc_state_owners(server);
545         return sp;
546 }
547
548 /**
549  * nfs4_put_state_owner - Release a nfs4_state_owner
550  * @sp: state owner data to release
551  */
552 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
553 {
554         struct nfs_server *server = sp->so_server;
555         struct nfs_client *clp = server->nfs_client;
556
557         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
558                 return;
559
560         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
561                 sp->so_expires = jiffies;
562                 list_add_tail(&sp->so_lru, &server->state_owners_lru);
563                 spin_unlock(&clp->cl_lock);
564         } else {
565                 nfs4_remove_state_owner_locked(sp);
566                 spin_unlock(&clp->cl_lock);
567                 nfs4_free_state_owner(sp);
568         }
569 }
570
571 /**
572  * nfs4_purge_state_owners - Release all cached state owners
573  * @server: nfs_server with cached state owners to release
574  *
575  * Called at umount time.  Remaining state owners will be on
576  * the LRU with ref count of zero.
577  */
578 void nfs4_purge_state_owners(struct nfs_server *server)
579 {
580         struct nfs_client *clp = server->nfs_client;
581         struct nfs4_state_owner *sp, *tmp;
582         LIST_HEAD(doomed);
583
584         spin_lock(&clp->cl_lock);
585         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
586                 list_move(&sp->so_lru, &doomed);
587                 nfs4_remove_state_owner_locked(sp);
588         }
589         spin_unlock(&clp->cl_lock);
590
591         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
592                 list_del(&sp->so_lru);
593                 nfs4_free_state_owner(sp);
594         }
595 }
596
597 static struct nfs4_state *
598 nfs4_alloc_open_state(void)
599 {
600         struct nfs4_state *state;
601
602         state = kzalloc(sizeof(*state), GFP_NOFS);
603         if (!state)
604                 return NULL;
605         atomic_set(&state->count, 1);
606         INIT_LIST_HEAD(&state->lock_states);
607         spin_lock_init(&state->state_lock);
608         seqlock_init(&state->seqlock);
609         return state;
610 }
611
612 void
613 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
614 {
615         if (state->state == fmode)
616                 return;
617         /* NB! List reordering - see the reclaim code for why.  */
618         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
619                 if (fmode & FMODE_WRITE)
620                         list_move(&state->open_states, &state->owner->so_states);
621                 else
622                         list_move_tail(&state->open_states, &state->owner->so_states);
623         }
624         state->state = fmode;
625 }
626
627 static struct nfs4_state *
628 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
629 {
630         struct nfs_inode *nfsi = NFS_I(inode);
631         struct nfs4_state *state;
632
633         list_for_each_entry(state, &nfsi->open_states, inode_states) {
634                 if (state->owner != owner)
635                         continue;
636                 if (atomic_inc_not_zero(&state->count))
637                         return state;
638         }
639         return NULL;
640 }
641
642 static void
643 nfs4_free_open_state(struct nfs4_state *state)
644 {
645         kfree(state);
646 }
647
648 struct nfs4_state *
649 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
650 {
651         struct nfs4_state *state, *new;
652         struct nfs_inode *nfsi = NFS_I(inode);
653
654         spin_lock(&inode->i_lock);
655         state = __nfs4_find_state_byowner(inode, owner);
656         spin_unlock(&inode->i_lock);
657         if (state)
658                 goto out;
659         new = nfs4_alloc_open_state();
660         spin_lock(&owner->so_lock);
661         spin_lock(&inode->i_lock);
662         state = __nfs4_find_state_byowner(inode, owner);
663         if (state == NULL && new != NULL) {
664                 state = new;
665                 state->owner = owner;
666                 atomic_inc(&owner->so_count);
667                 list_add(&state->inode_states, &nfsi->open_states);
668                 ihold(inode);
669                 state->inode = inode;
670                 spin_unlock(&inode->i_lock);
671                 /* Note: The reclaim code dictates that we add stateless
672                  * and read-only stateids to the end of the list */
673                 list_add_tail(&state->open_states, &owner->so_states);
674                 spin_unlock(&owner->so_lock);
675         } else {
676                 spin_unlock(&inode->i_lock);
677                 spin_unlock(&owner->so_lock);
678                 if (new)
679                         nfs4_free_open_state(new);
680         }
681 out:
682         return state;
683 }
684
685 void nfs4_put_open_state(struct nfs4_state *state)
686 {
687         struct inode *inode = state->inode;
688         struct nfs4_state_owner *owner = state->owner;
689
690         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
691                 return;
692         spin_lock(&inode->i_lock);
693         list_del(&state->inode_states);
694         list_del(&state->open_states);
695         spin_unlock(&inode->i_lock);
696         spin_unlock(&owner->so_lock);
697         iput(inode);
698         nfs4_free_open_state(state);
699         nfs4_put_state_owner(owner);
700 }
701
702 /*
703  * Close the current file.
704  */
705 static void __nfs4_close(struct nfs4_state *state,
706                 fmode_t fmode, gfp_t gfp_mask, int wait)
707 {
708         struct nfs4_state_owner *owner = state->owner;
709         int call_close = 0;
710         fmode_t newstate;
711
712         atomic_inc(&owner->so_count);
713         /* Protect against nfs4_find_state() */
714         spin_lock(&owner->so_lock);
715         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
716                 case FMODE_READ:
717                         state->n_rdonly--;
718                         break;
719                 case FMODE_WRITE:
720                         state->n_wronly--;
721                         break;
722                 case FMODE_READ|FMODE_WRITE:
723                         state->n_rdwr--;
724         }
725         newstate = FMODE_READ|FMODE_WRITE;
726         if (state->n_rdwr == 0) {
727                 if (state->n_rdonly == 0) {
728                         newstate &= ~FMODE_READ;
729                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
730                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
731                 }
732                 if (state->n_wronly == 0) {
733                         newstate &= ~FMODE_WRITE;
734                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
735                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
736                 }
737                 if (newstate == 0)
738                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
739         }
740         nfs4_state_set_mode_locked(state, newstate);
741         spin_unlock(&owner->so_lock);
742
743         if (!call_close) {
744                 nfs4_put_open_state(state);
745                 nfs4_put_state_owner(owner);
746         } else {
747                 bool roc = pnfs_roc(state->inode);
748
749                 nfs4_do_close(state, gfp_mask, wait, roc);
750         }
751 }
752
753 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
754 {
755         __nfs4_close(state, fmode, GFP_NOFS, 0);
756 }
757
758 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
759 {
760         __nfs4_close(state, fmode, GFP_KERNEL, 1);
761 }
762
763 /*
764  * Search the state->lock_states for an existing lock_owner
765  * that is compatible with current->files
766  */
767 static struct nfs4_lock_state *
768 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
769 {
770         struct nfs4_lock_state *pos;
771         list_for_each_entry(pos, &state->lock_states, ls_locks) {
772                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
773                         continue;
774                 switch (pos->ls_owner.lo_type) {
775                 case NFS4_POSIX_LOCK_TYPE:
776                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
777                                 continue;
778                         break;
779                 case NFS4_FLOCK_LOCK_TYPE:
780                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
781                                 continue;
782                 }
783                 atomic_inc(&pos->ls_count);
784                 return pos;
785         }
786         return NULL;
787 }
788
789 /*
790  * Return a compatible lock_state. If no initialized lock_state structure
791  * exists, return an uninitialized one.
792  *
793  */
794 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)
795 {
796         struct nfs4_lock_state *lsp;
797         struct nfs_server *server = state->owner->so_server;
798         struct nfs_client *clp = server->nfs_client;
799
800         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
801         if (lsp == NULL)
802                 return NULL;
803         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
804         spin_lock_init(&lsp->ls_sequence.lock);
805         INIT_LIST_HEAD(&lsp->ls_sequence.list);
806         lsp->ls_seqid.sequence = &lsp->ls_sequence;
807         atomic_set(&lsp->ls_count, 1);
808         lsp->ls_state = state;
809         lsp->ls_owner.lo_type = type;
810         switch (lsp->ls_owner.lo_type) {
811         case NFS4_FLOCK_LOCK_TYPE:
812                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
813                 break;
814         case NFS4_POSIX_LOCK_TYPE:
815                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
816                 break;
817         default:
818                 kfree(lsp);
819                 return NULL;
820         }
821         spin_lock(&clp->cl_lock);
822         nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
823         spin_unlock(&clp->cl_lock);
824         INIT_LIST_HEAD(&lsp->ls_locks);
825         return lsp;
826 }
827
828 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
829 {
830         struct nfs_server *server = lsp->ls_state->owner->so_server;
831         struct nfs_client *clp = server->nfs_client;
832
833         spin_lock(&clp->cl_lock);
834         nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
835         spin_unlock(&clp->cl_lock);
836         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
837         kfree(lsp);
838 }
839
840 /*
841  * Return a compatible lock_state. If no initialized lock_state structure
842  * exists, return an uninitialized one.
843  *
844  */
845 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
846 {
847         struct nfs4_lock_state *lsp, *new = NULL;
848         
849         for(;;) {
850                 spin_lock(&state->state_lock);
851                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
852                 if (lsp != NULL)
853                         break;
854                 if (new != NULL) {
855                         list_add(&new->ls_locks, &state->lock_states);
856                         set_bit(LK_STATE_IN_USE, &state->flags);
857                         lsp = new;
858                         new = NULL;
859                         break;
860                 }
861                 spin_unlock(&state->state_lock);
862                 new = nfs4_alloc_lock_state(state, owner, pid, type);
863                 if (new == NULL)
864                         return NULL;
865         }
866         spin_unlock(&state->state_lock);
867         if (new != NULL)
868                 nfs4_free_lock_state(new);
869         return lsp;
870 }
871
872 /*
873  * Release reference to lock_state, and free it if we see that
874  * it is no longer in use
875  */
876 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
877 {
878         struct nfs4_state *state;
879
880         if (lsp == NULL)
881                 return;
882         state = lsp->ls_state;
883         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
884                 return;
885         list_del(&lsp->ls_locks);
886         if (list_empty(&state->lock_states))
887                 clear_bit(LK_STATE_IN_USE, &state->flags);
888         spin_unlock(&state->state_lock);
889         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
890                 nfs4_release_lockowner(lsp);
891         nfs4_free_lock_state(lsp);
892 }
893
894 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
895 {
896         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
897
898         dst->fl_u.nfs4_fl.owner = lsp;
899         atomic_inc(&lsp->ls_count);
900 }
901
902 static void nfs4_fl_release_lock(struct file_lock *fl)
903 {
904         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
905 }
906
907 static const struct file_lock_operations nfs4_fl_lock_ops = {
908         .fl_copy_lock = nfs4_fl_copy_lock,
909         .fl_release_private = nfs4_fl_release_lock,
910 };
911
912 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
913 {
914         struct nfs4_lock_state *lsp;
915
916         if (fl->fl_ops != NULL)
917                 return 0;
918         if (fl->fl_flags & FL_POSIX)
919                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
920         else if (fl->fl_flags & FL_FLOCK)
921                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
922         else
923                 return -EINVAL;
924         if (lsp == NULL)
925                 return -ENOMEM;
926         fl->fl_u.nfs4_fl.owner = lsp;
927         fl->fl_ops = &nfs4_fl_lock_ops;
928         return 0;
929 }
930
931 /*
932  * Byte-range lock aware utility to initialize the stateid of read/write
933  * requests.
934  */
935 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
936 {
937         struct nfs4_lock_state *lsp;
938         int seq;
939
940         do {
941                 seq = read_seqbegin(&state->seqlock);
942                 memcpy(dst, &state->stateid, sizeof(*dst));
943         } while (read_seqretry(&state->seqlock, seq));
944         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
945                 return;
946
947         spin_lock(&state->state_lock);
948         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
949         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
950                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
951         spin_unlock(&state->state_lock);
952         nfs4_put_lock_state(lsp);
953 }
954
955 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
956 {
957         struct nfs_seqid *new;
958
959         new = kmalloc(sizeof(*new), gfp_mask);
960         if (new != NULL) {
961                 new->sequence = counter;
962                 INIT_LIST_HEAD(&new->list);
963         }
964         return new;
965 }
966
967 void nfs_release_seqid(struct nfs_seqid *seqid)
968 {
969         if (!list_empty(&seqid->list)) {
970                 struct rpc_sequence *sequence = seqid->sequence->sequence;
971
972                 spin_lock(&sequence->lock);
973                 list_del_init(&seqid->list);
974                 spin_unlock(&sequence->lock);
975                 rpc_wake_up(&sequence->wait);
976         }
977 }
978
979 void nfs_free_seqid(struct nfs_seqid *seqid)
980 {
981         nfs_release_seqid(seqid);
982         kfree(seqid);
983 }
984
985 /*
986  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
987  * failed with a seqid incrementing error -
988  * see comments nfs_fs.h:seqid_mutating_error()
989  */
990 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
991 {
992         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
993         switch (status) {
994                 case 0:
995                         break;
996                 case -NFS4ERR_BAD_SEQID:
997                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
998                                 return;
999                         printk(KERN_WARNING "NFS: v4 server returned a bad"
1000                                         " sequence-id error on an"
1001                                         " unconfirmed sequence %p!\n",
1002                                         seqid->sequence);
1003                 case -NFS4ERR_STALE_CLIENTID:
1004                 case -NFS4ERR_STALE_STATEID:
1005                 case -NFS4ERR_BAD_STATEID:
1006                 case -NFS4ERR_BADXDR:
1007                 case -NFS4ERR_RESOURCE:
1008                 case -NFS4ERR_NOFILEHANDLE:
1009                         /* Non-seqid mutating errors */
1010                         return;
1011         };
1012         /*
1013          * Note: no locking needed as we are guaranteed to be first
1014          * on the sequence list
1015          */
1016         seqid->sequence->counter++;
1017 }
1018
1019 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1020 {
1021         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1022                                         struct nfs4_state_owner, so_seqid);
1023         struct nfs_server *server = sp->so_server;
1024
1025         if (status == -NFS4ERR_BAD_SEQID)
1026                 nfs4_drop_state_owner(sp);
1027         if (!nfs4_has_session(server->nfs_client))
1028                 nfs_increment_seqid(status, seqid);
1029 }
1030
1031 /*
1032  * Increment the seqid if the LOCK/LOCKU succeeded, or
1033  * failed with a seqid incrementing error -
1034  * see comments nfs_fs.h:seqid_mutating_error()
1035  */
1036 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1037 {
1038         nfs_increment_seqid(status, seqid);
1039 }
1040
1041 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1042 {
1043         struct rpc_sequence *sequence = seqid->sequence->sequence;
1044         int status = 0;
1045
1046         spin_lock(&sequence->lock);
1047         if (list_empty(&seqid->list))
1048                 list_add_tail(&seqid->list, &sequence->list);
1049         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1050                 goto unlock;
1051         rpc_sleep_on(&sequence->wait, task, NULL);
1052         status = -EAGAIN;
1053 unlock:
1054         spin_unlock(&sequence->lock);
1055         return status;
1056 }
1057
1058 static int nfs4_run_state_manager(void *);
1059
1060 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1061 {
1062         smp_mb__before_clear_bit();
1063         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1064         smp_mb__after_clear_bit();
1065         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1066         rpc_wake_up(&clp->cl_rpcwaitq);
1067 }
1068
1069 /*
1070  * Schedule the nfs_client asynchronous state management routine
1071  */
1072 void nfs4_schedule_state_manager(struct nfs_client *clp)
1073 {
1074         struct task_struct *task;
1075
1076         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1077                 return;
1078         __module_get(THIS_MODULE);
1079         atomic_inc(&clp->cl_count);
1080         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1081                                 rpc_peeraddr2str(clp->cl_rpcclient,
1082                                                         RPC_DISPLAY_ADDR));
1083         if (!IS_ERR(task))
1084                 return;
1085         nfs4_clear_state_manager_bit(clp);
1086         nfs_put_client(clp);
1087         module_put(THIS_MODULE);
1088 }
1089
1090 /*
1091  * Schedule a lease recovery attempt
1092  */
1093 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1094 {
1095         if (!clp)
1096                 return;
1097         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1098                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1099         nfs4_schedule_state_manager(clp);
1100 }
1101
1102 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1103 {
1104         nfs_handle_cb_pathdown(clp);
1105         nfs4_schedule_state_manager(clp);
1106 }
1107
1108 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1109 {
1110
1111         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1112         /* Don't recover state that expired before the reboot */
1113         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1114                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1115                 return 0;
1116         }
1117         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1118         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1119         return 1;
1120 }
1121
1122 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1123 {
1124         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1125         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1126         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1127         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1128         return 1;
1129 }
1130
1131 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1132 {
1133         struct nfs_client *clp = server->nfs_client;
1134
1135         if (test_and_clear_bit(NFS_DELEGATED_STATE, &state->flags))
1136                 nfs_async_inode_return_delegation(state->inode, &state->stateid);
1137         nfs4_state_mark_reclaim_nograce(clp, state);
1138         nfs4_schedule_state_manager(clp);
1139 }
1140
1141 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1142 {
1143         struct inode *inode = state->inode;
1144         struct nfs_inode *nfsi = NFS_I(inode);
1145         struct file_lock *fl;
1146         int status = 0;
1147
1148         if (inode->i_flock == NULL)
1149                 return 0;
1150
1151         /* Guard against delegation returns and new lock/unlock calls */
1152         down_write(&nfsi->rwsem);
1153         /* Protect inode->i_flock using the BKL */
1154         lock_flocks();
1155         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1156                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1157                         continue;
1158                 if (nfs_file_open_context(fl->fl_file)->state != state)
1159                         continue;
1160                 unlock_flocks();
1161                 status = ops->recover_lock(state, fl);
1162                 switch (status) {
1163                         case 0:
1164                                 break;
1165                         case -ESTALE:
1166                         case -NFS4ERR_ADMIN_REVOKED:
1167                         case -NFS4ERR_STALE_STATEID:
1168                         case -NFS4ERR_BAD_STATEID:
1169                         case -NFS4ERR_EXPIRED:
1170                         case -NFS4ERR_NO_GRACE:
1171                         case -NFS4ERR_STALE_CLIENTID:
1172                         case -NFS4ERR_BADSESSION:
1173                         case -NFS4ERR_BADSLOT:
1174                         case -NFS4ERR_BAD_HIGH_SLOT:
1175                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1176                                 goto out;
1177                         default:
1178                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1179                                                 __func__, status);
1180                         case -ENOMEM:
1181                         case -NFS4ERR_DENIED:
1182                         case -NFS4ERR_RECLAIM_BAD:
1183                         case -NFS4ERR_RECLAIM_CONFLICT:
1184                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1185                                 status = 0;
1186                 }
1187                 lock_flocks();
1188         }
1189         unlock_flocks();
1190 out:
1191         up_write(&nfsi->rwsem);
1192         return status;
1193 }
1194
1195 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1196 {
1197         struct nfs4_state *state;
1198         struct nfs4_lock_state *lock;
1199         int status = 0;
1200
1201         /* Note: we rely on the sp->so_states list being ordered 
1202          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1203          * states first.
1204          * This is needed to ensure that the server won't give us any
1205          * read delegations that we have to return if, say, we are
1206          * recovering after a network partition or a reboot from a
1207          * server that doesn't support a grace period.
1208          */
1209 restart:
1210         spin_lock(&sp->so_lock);
1211         list_for_each_entry(state, &sp->so_states, open_states) {
1212                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1213                         continue;
1214                 if (state->state == 0)
1215                         continue;
1216                 atomic_inc(&state->count);
1217                 spin_unlock(&sp->so_lock);
1218                 status = ops->recover_open(sp, state);
1219                 if (status >= 0) {
1220                         status = nfs4_reclaim_locks(state, ops);
1221                         if (status >= 0) {
1222                                 spin_lock(&state->state_lock);
1223                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1224                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1225                                                 printk("%s: Lock reclaim failed!\n",
1226                                                         __func__);
1227                                 }
1228                                 spin_unlock(&state->state_lock);
1229                                 nfs4_put_open_state(state);
1230                                 goto restart;
1231                         }
1232                 }
1233                 switch (status) {
1234                         default:
1235                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1236                                                 __func__, status);
1237                         case -ENOENT:
1238                         case -ENOMEM:
1239                         case -ESTALE:
1240                                 /*
1241                                  * Open state on this file cannot be recovered
1242                                  * All we can do is revert to using the zero stateid.
1243                                  */
1244                                 memset(state->stateid.data, 0,
1245                                         sizeof(state->stateid.data));
1246                                 /* Mark the file as being 'closed' */
1247                                 state->state = 0;
1248                                 break;
1249                         case -EKEYEXPIRED:
1250                                 /*
1251                                  * User RPCSEC_GSS context has expired.
1252                                  * We cannot recover this stateid now, so
1253                                  * skip it and allow recovery thread to
1254                                  * proceed.
1255                                  */
1256                                 break;
1257                         case -NFS4ERR_ADMIN_REVOKED:
1258                         case -NFS4ERR_STALE_STATEID:
1259                         case -NFS4ERR_BAD_STATEID:
1260                         case -NFS4ERR_RECLAIM_BAD:
1261                         case -NFS4ERR_RECLAIM_CONFLICT:
1262                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1263                                 break;
1264                         case -NFS4ERR_EXPIRED:
1265                         case -NFS4ERR_NO_GRACE:
1266                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1267                         case -NFS4ERR_STALE_CLIENTID:
1268                         case -NFS4ERR_BADSESSION:
1269                         case -NFS4ERR_BADSLOT:
1270                         case -NFS4ERR_BAD_HIGH_SLOT:
1271                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1272                                 goto out_err;
1273                 }
1274                 nfs4_put_open_state(state);
1275                 goto restart;
1276         }
1277         spin_unlock(&sp->so_lock);
1278         return 0;
1279 out_err:
1280         nfs4_put_open_state(state);
1281         return status;
1282 }
1283
1284 static void nfs4_clear_open_state(struct nfs4_state *state)
1285 {
1286         struct nfs4_lock_state *lock;
1287
1288         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1289         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1290         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1291         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1292         spin_lock(&state->state_lock);
1293         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1294                 lock->ls_seqid.flags = 0;
1295                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1296         }
1297         spin_unlock(&state->state_lock);
1298 }
1299
1300 static void nfs4_reset_seqids(struct nfs_server *server,
1301         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1302 {
1303         struct nfs_client *clp = server->nfs_client;
1304         struct nfs4_state_owner *sp;
1305         struct rb_node *pos;
1306         struct nfs4_state *state;
1307
1308         spin_lock(&clp->cl_lock);
1309         for (pos = rb_first(&server->state_owners);
1310              pos != NULL;
1311              pos = rb_next(pos)) {
1312                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1313                 sp->so_seqid.flags = 0;
1314                 spin_lock(&sp->so_lock);
1315                 list_for_each_entry(state, &sp->so_states, open_states) {
1316                         if (mark_reclaim(clp, state))
1317                                 nfs4_clear_open_state(state);
1318                 }
1319                 spin_unlock(&sp->so_lock);
1320         }
1321         spin_unlock(&clp->cl_lock);
1322 }
1323
1324 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1325         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1326 {
1327         struct nfs_server *server;
1328
1329         rcu_read_lock();
1330         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1331                 nfs4_reset_seqids(server, mark_reclaim);
1332         rcu_read_unlock();
1333 }
1334
1335 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1336 {
1337         /* Mark all delegations for reclaim */
1338         nfs_delegation_mark_reclaim(clp);
1339         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1340 }
1341
1342 static void nfs4_reclaim_complete(struct nfs_client *clp,
1343                                  const struct nfs4_state_recovery_ops *ops)
1344 {
1345         /* Notify the server we're done reclaiming our state */
1346         if (ops->reclaim_complete)
1347                 (void)ops->reclaim_complete(clp);
1348 }
1349
1350 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1351 {
1352         struct nfs_client *clp = server->nfs_client;
1353         struct nfs4_state_owner *sp;
1354         struct rb_node *pos;
1355         struct nfs4_state *state;
1356
1357         spin_lock(&clp->cl_lock);
1358         for (pos = rb_first(&server->state_owners);
1359              pos != NULL;
1360              pos = rb_next(pos)) {
1361                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1362                 spin_lock(&sp->so_lock);
1363                 list_for_each_entry(state, &sp->so_states, open_states) {
1364                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1365                                                 &state->flags))
1366                                 continue;
1367                         nfs4_state_mark_reclaim_nograce(clp, state);
1368                 }
1369                 spin_unlock(&sp->so_lock);
1370         }
1371         spin_unlock(&clp->cl_lock);
1372 }
1373
1374 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1375 {
1376         struct nfs_server *server;
1377
1378         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1379                 return 0;
1380
1381         rcu_read_lock();
1382         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1383                 nfs4_clear_reclaim_server(server);
1384         rcu_read_unlock();
1385
1386         nfs_delegation_reap_unclaimed(clp);
1387         return 1;
1388 }
1389
1390 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1391 {
1392         if (!nfs4_state_clear_reclaim_reboot(clp))
1393                 return;
1394         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1395 }
1396
1397 static void nfs_delegation_clear_all(struct nfs_client *clp)
1398 {
1399         nfs_delegation_mark_reclaim(clp);
1400         nfs_delegation_reap_unclaimed(clp);
1401 }
1402
1403 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1404 {
1405         nfs_delegation_clear_all(clp);
1406         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1407 }
1408
1409 static void nfs4_warn_keyexpired(const char *s)
1410 {
1411         printk_ratelimited(KERN_WARNING "Error: state manager"
1412                         " encountered RPCSEC_GSS session"
1413                         " expired against NFSv4 server %s.\n",
1414                         s);
1415 }
1416
1417 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1418 {
1419         switch (error) {
1420                 case 0:
1421                         break;
1422                 case -NFS4ERR_CB_PATH_DOWN:
1423                         nfs_handle_cb_pathdown(clp);
1424                         break;
1425                 case -NFS4ERR_NO_GRACE:
1426                         nfs4_state_end_reclaim_reboot(clp);
1427                         break;
1428                 case -NFS4ERR_STALE_CLIENTID:
1429                 case -NFS4ERR_LEASE_MOVED:
1430                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1431                         nfs4_state_clear_reclaim_reboot(clp);
1432                         nfs4_state_start_reclaim_reboot(clp);
1433                         break;
1434                 case -NFS4ERR_EXPIRED:
1435                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1436                         nfs4_state_start_reclaim_nograce(clp);
1437                         break;
1438                 case -NFS4ERR_BADSESSION:
1439                 case -NFS4ERR_BADSLOT:
1440                 case -NFS4ERR_BAD_HIGH_SLOT:
1441                 case -NFS4ERR_DEADSESSION:
1442                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1443                 case -NFS4ERR_SEQ_FALSE_RETRY:
1444                 case -NFS4ERR_SEQ_MISORDERED:
1445                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1446                         /* Zero session reset errors */
1447                         break;
1448                 case -EKEYEXPIRED:
1449                         /* Nothing we can do */
1450                         nfs4_warn_keyexpired(clp->cl_hostname);
1451                         break;
1452                 default:
1453                         return error;
1454         }
1455         return 0;
1456 }
1457
1458 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1459 {
1460         struct nfs4_state_owner *sp;
1461         struct nfs_server *server;
1462         struct rb_node *pos;
1463         int status = 0;
1464
1465 restart:
1466         rcu_read_lock();
1467         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1468                 nfs4_purge_state_owners(server);
1469                 spin_lock(&clp->cl_lock);
1470                 for (pos = rb_first(&server->state_owners);
1471                      pos != NULL;
1472                      pos = rb_next(pos)) {
1473                         sp = rb_entry(pos,
1474                                 struct nfs4_state_owner, so_server_node);
1475                         if (!test_and_clear_bit(ops->owner_flag_bit,
1476                                                         &sp->so_flags))
1477                                 continue;
1478                         atomic_inc(&sp->so_count);
1479                         spin_unlock(&clp->cl_lock);
1480                         rcu_read_unlock();
1481
1482                         status = nfs4_reclaim_open_state(sp, ops);
1483                         if (status < 0) {
1484                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1485                                 nfs4_put_state_owner(sp);
1486                                 return nfs4_recovery_handle_error(clp, status);
1487                         }
1488
1489                         nfs4_put_state_owner(sp);
1490                         goto restart;
1491                 }
1492                 spin_unlock(&clp->cl_lock);
1493         }
1494         rcu_read_unlock();
1495         return status;
1496 }
1497
1498 static int nfs4_check_lease(struct nfs_client *clp)
1499 {
1500         struct rpc_cred *cred;
1501         const struct nfs4_state_maintenance_ops *ops =
1502                 clp->cl_mvops->state_renewal_ops;
1503         int status;
1504
1505         /* Is the client already known to have an expired lease? */
1506         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1507                 return 0;
1508         spin_lock(&clp->cl_lock);
1509         cred = ops->get_state_renewal_cred_locked(clp);
1510         spin_unlock(&clp->cl_lock);
1511         if (cred == NULL) {
1512                 cred = nfs4_get_setclientid_cred(clp);
1513                 status = -ENOKEY;
1514                 if (cred == NULL)
1515                         goto out;
1516         }
1517         status = ops->renew_lease(clp, cred);
1518         put_rpccred(cred);
1519 out:
1520         return nfs4_recovery_handle_error(clp, status);
1521 }
1522
1523 static int nfs4_reclaim_lease(struct nfs_client *clp)
1524 {
1525         struct rpc_cred *cred;
1526         const struct nfs4_state_recovery_ops *ops =
1527                 clp->cl_mvops->reboot_recovery_ops;
1528         int status = -ENOENT;
1529
1530         cred = ops->get_clid_cred(clp);
1531         if (cred != NULL) {
1532                 status = ops->establish_clid(clp, cred);
1533                 put_rpccred(cred);
1534                 /* Handle case where the user hasn't set up machine creds */
1535                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1536                         nfs4_clear_machine_cred(clp);
1537                         status = -EAGAIN;
1538                 }
1539                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1540                         status = -EPROTONOSUPPORT;
1541         }
1542         return status;
1543 }
1544
1545 #ifdef CONFIG_NFS_V4_1
1546 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1547 {
1548         struct nfs_client *clp = session->clp;
1549
1550         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1551         nfs4_schedule_lease_recovery(clp);
1552 }
1553 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1554
1555 void nfs41_handle_recall_slot(struct nfs_client *clp)
1556 {
1557         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1558         nfs4_schedule_state_manager(clp);
1559 }
1560
1561 static void nfs4_reset_all_state(struct nfs_client *clp)
1562 {
1563         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1564                 clp->cl_boot_time = CURRENT_TIME;
1565                 nfs4_state_start_reclaim_nograce(clp);
1566                 nfs4_schedule_state_manager(clp);
1567         }
1568 }
1569
1570 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1571 {
1572         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1573                 nfs4_state_start_reclaim_reboot(clp);
1574                 nfs4_schedule_state_manager(clp);
1575         }
1576 }
1577
1578 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1579 {
1580         /* Temporary */
1581         nfs4_reset_all_state(clp);
1582 }
1583
1584 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1585 {
1586         /* This will need to handle layouts too */
1587         nfs_expire_all_delegations(clp);
1588 }
1589
1590 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1591 {
1592         nfs_expire_all_delegations(clp);
1593         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1594                 nfs4_schedule_state_manager(clp);
1595 }
1596
1597 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1598 {
1599         if (!flags)
1600                 return;
1601         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1602                 nfs41_handle_server_reboot(clp);
1603         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1604                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1605                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1606                             SEQ4_STATUS_LEASE_MOVED))
1607                 nfs41_handle_state_revoked(clp);
1608         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1609                 nfs41_handle_recallable_state_revoked(clp);
1610         if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1611                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1612                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1613                 nfs41_handle_cb_path_down(clp);
1614 }
1615
1616 static int nfs4_reset_session(struct nfs_client *clp)
1617 {
1618         int status;
1619
1620         nfs4_begin_drain_session(clp);
1621         status = nfs4_proc_destroy_session(clp->cl_session);
1622         if (status && status != -NFS4ERR_BADSESSION &&
1623             status != -NFS4ERR_DEADSESSION) {
1624                 status = nfs4_recovery_handle_error(clp, status);
1625                 goto out;
1626         }
1627
1628         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1629         status = nfs4_proc_create_session(clp);
1630         if (status) {
1631                 status = nfs4_recovery_handle_error(clp, status);
1632                 goto out;
1633         }
1634         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1635         /* create_session negotiated new slot table */
1636         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1637
1638          /* Let the state manager reestablish state */
1639         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1640                 nfs41_setup_state_renewal(clp);
1641 out:
1642         return status;
1643 }
1644
1645 static int nfs4_recall_slot(struct nfs_client *clp)
1646 {
1647         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1648         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1649         struct nfs4_slot *new, *old;
1650         int i;
1651
1652         nfs4_begin_drain_session(clp);
1653         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1654                       GFP_NOFS);
1655         if (!new)
1656                 return -ENOMEM;
1657
1658         spin_lock(&fc_tbl->slot_tbl_lock);
1659         for (i = 0; i < fc_tbl->target_max_slots; i++)
1660                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1661         old = fc_tbl->slots;
1662         fc_tbl->slots = new;
1663         fc_tbl->max_slots = fc_tbl->target_max_slots;
1664         fc_tbl->target_max_slots = 0;
1665         fc_attrs->max_reqs = fc_tbl->max_slots;
1666         spin_unlock(&fc_tbl->slot_tbl_lock);
1667
1668         kfree(old);
1669         nfs4_end_drain_session(clp);
1670         return 0;
1671 }
1672
1673 #else /* CONFIG_NFS_V4_1 */
1674 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1675 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1676 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1677 #endif /* CONFIG_NFS_V4_1 */
1678
1679 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1680  * on EXCHANGE_ID for v4.1
1681  */
1682 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1683 {
1684         switch (status) {
1685         case -NFS4ERR_CLID_INUSE:
1686         case -NFS4ERR_STALE_CLIENTID:
1687                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1688                 break;
1689         case -NFS4ERR_DELAY:
1690         case -ETIMEDOUT:
1691         case -EAGAIN:
1692                 ssleep(1);
1693                 break;
1694
1695         case -EKEYEXPIRED:
1696                 nfs4_warn_keyexpired(clp->cl_hostname);
1697         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1698                                  * in nfs4_exchange_id */
1699         default:
1700                 return;
1701         }
1702         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1703 }
1704
1705 static void nfs4_state_manager(struct nfs_client *clp)
1706 {
1707         int status = 0;
1708
1709         /* Ensure exclusive access to NFSv4 state */
1710         do {
1711                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1712                         /* We're going to have to re-establish a clientid */
1713                         status = nfs4_reclaim_lease(clp);
1714                         if (status) {
1715                                 nfs4_set_lease_expired(clp, status);
1716                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1717                                                         &clp->cl_state))
1718                                         continue;
1719                                 if (clp->cl_cons_state ==
1720                                                         NFS_CS_SESSION_INITING)
1721                                         nfs_mark_client_ready(clp, status);
1722                                 goto out_error;
1723                         }
1724                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1725
1726                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1727                                                &clp->cl_state))
1728                                 nfs4_state_start_reclaim_nograce(clp);
1729                         else
1730                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1731                                         &clp->cl_state);
1732
1733                         pnfs_destroy_all_layouts(clp);
1734                 }
1735
1736                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1737                         status = nfs4_check_lease(clp);
1738                         if (status < 0)
1739                                 goto out_error;
1740                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1741                                 continue;
1742                 }
1743
1744                 /* Initialize or reset the session */
1745                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1746                    && nfs4_has_session(clp)) {
1747                         status = nfs4_reset_session(clp);
1748                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1749                                 continue;
1750                         if (status < 0)
1751                                 goto out_error;
1752                 }
1753
1754                 /* First recover reboot state... */
1755                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1756                         status = nfs4_do_reclaim(clp,
1757                                 clp->cl_mvops->reboot_recovery_ops);
1758                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1759                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1760                                 continue;
1761                         nfs4_state_end_reclaim_reboot(clp);
1762                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1763                                 continue;
1764                         if (status < 0)
1765                                 goto out_error;
1766                 }
1767
1768                 /* Now recover expired state... */
1769                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1770                         status = nfs4_do_reclaim(clp,
1771                                 clp->cl_mvops->nograce_recovery_ops);
1772                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1773                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1774                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1775                                 continue;
1776                         if (status < 0)
1777                                 goto out_error;
1778                 }
1779
1780                 nfs4_end_drain_session(clp);
1781                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1782                         nfs_client_return_marked_delegations(clp);
1783                         continue;
1784                 }
1785                 /* Recall session slots */
1786                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1787                    && nfs4_has_session(clp)) {
1788                         status = nfs4_recall_slot(clp);
1789                         if (status < 0)
1790                                 goto out_error;
1791                         continue;
1792                 }
1793
1794
1795                 nfs4_clear_state_manager_bit(clp);
1796                 /* Did we race with an attempt to give us more work? */
1797                 if (clp->cl_state == 0)
1798                         break;
1799                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1800                         break;
1801         } while (atomic_read(&clp->cl_count) > 1);
1802         return;
1803 out_error:
1804         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1805                         " with error %d\n", clp->cl_hostname, -status);
1806         nfs4_end_drain_session(clp);
1807         nfs4_clear_state_manager_bit(clp);
1808 }
1809
1810 static int nfs4_run_state_manager(void *ptr)
1811 {
1812         struct nfs_client *clp = ptr;
1813
1814         allow_signal(SIGKILL);
1815         nfs4_state_manager(clp);
1816         nfs_put_client(clp);
1817         module_put_and_exit(0);
1818         return 0;
1819 }
1820
1821 /*
1822  * Local variables:
1823  *  c-basic-offset: 8
1824  * End:
1825  */