]> Pileus Git - ~andy/linux/blob - fs/nfs/nfs4state.c
NFSv4: Replace state_owner->so_owner_id with an ida based allocator
[~andy/linux] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/ratelimit.h>
50 #include <linux/workqueue.h>
51 #include <linux/bitops.h>
52 #include <linux/jiffies.h>
53
54 #include "nfs4_fs.h"
55 #include "callback.h"
56 #include "delegation.h"
57 #include "internal.h"
58 #include "pnfs.h"
59
60 #define OPENOWNER_POOL_SIZE     8
61
62 const nfs4_stateid zero_stateid;
63
64 static LIST_HEAD(nfs4_clientid_list);
65
66 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
67 {
68         struct nfs4_setclientid_res clid = {
69                 .clientid = clp->cl_clientid,
70                 .confirm = clp->cl_confirm,
71         };
72         unsigned short port;
73         int status;
74
75         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
76                 goto do_confirm;
77         port = nfs_callback_tcpport;
78         if (clp->cl_addr.ss_family == AF_INET6)
79                 port = nfs_callback_tcpport6;
80
81         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
82         if (status != 0)
83                 goto out;
84         clp->cl_clientid = clid.clientid;
85         clp->cl_confirm = clid.confirm;
86         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
87 do_confirm:
88         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
89         if (status != 0)
90                 goto out;
91         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
92         nfs4_schedule_state_renewal(clp);
93 out:
94         return status;
95 }
96
97 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
98 {
99         struct rpc_cred *cred = NULL;
100
101         if (clp->cl_machine_cred != NULL)
102                 cred = get_rpccred(clp->cl_machine_cred);
103         return cred;
104 }
105
106 static void nfs4_clear_machine_cred(struct nfs_client *clp)
107 {
108         struct rpc_cred *cred;
109
110         spin_lock(&clp->cl_lock);
111         cred = clp->cl_machine_cred;
112         clp->cl_machine_cred = NULL;
113         spin_unlock(&clp->cl_lock);
114         if (cred != NULL)
115                 put_rpccred(cred);
116 }
117
118 static struct rpc_cred *
119 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
120 {
121         struct rpc_cred *cred = NULL;
122         struct nfs4_state_owner *sp;
123         struct rb_node *pos;
124
125         for (pos = rb_first(&server->state_owners);
126              pos != NULL;
127              pos = rb_next(pos)) {
128                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
129                 if (list_empty(&sp->so_states))
130                         continue;
131                 cred = get_rpccred(sp->so_cred);
132                 break;
133         }
134         return cred;
135 }
136
137 /**
138  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
139  * @clp: client state handle
140  *
141  * Returns an rpc_cred with reference count bumped, or NULL.
142  * Caller must hold clp->cl_lock.
143  */
144 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
145 {
146         struct rpc_cred *cred = NULL;
147         struct nfs_server *server;
148
149         rcu_read_lock();
150         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
151                 cred = nfs4_get_renew_cred_server_locked(server);
152                 if (cred != NULL)
153                         break;
154         }
155         rcu_read_unlock();
156         return cred;
157 }
158
159 #if defined(CONFIG_NFS_V4_1)
160
161 static int nfs41_setup_state_renewal(struct nfs_client *clp)
162 {
163         int status;
164         struct nfs_fsinfo fsinfo;
165
166         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
167                 nfs4_schedule_state_renewal(clp);
168                 return 0;
169         }
170
171         status = nfs4_proc_get_lease_time(clp, &fsinfo);
172         if (status == 0) {
173                 /* Update lease time and schedule renewal */
174                 spin_lock(&clp->cl_lock);
175                 clp->cl_lease_time = fsinfo.lease_time * HZ;
176                 clp->cl_last_renewal = jiffies;
177                 spin_unlock(&clp->cl_lock);
178
179                 nfs4_schedule_state_renewal(clp);
180         }
181
182         return status;
183 }
184
185 /*
186  * Back channel returns NFS4ERR_DELAY for new requests when
187  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
188  * is ended.
189  */
190 static void nfs4_end_drain_session(struct nfs_client *clp)
191 {
192         struct nfs4_session *ses = clp->cl_session;
193         int max_slots;
194
195         if (ses == NULL)
196                 return;
197         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
198                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
199                 max_slots = ses->fc_slot_table.max_slots;
200                 while (max_slots--) {
201                         struct rpc_task *task;
202
203                         task = rpc_wake_up_next(&ses->fc_slot_table.
204                                                 slot_tbl_waitq);
205                         if (!task)
206                                 break;
207                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
208                 }
209                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
210         }
211 }
212
213 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
214 {
215         spin_lock(&tbl->slot_tbl_lock);
216         if (tbl->highest_used_slotid != -1) {
217                 INIT_COMPLETION(tbl->complete);
218                 spin_unlock(&tbl->slot_tbl_lock);
219                 return wait_for_completion_interruptible(&tbl->complete);
220         }
221         spin_unlock(&tbl->slot_tbl_lock);
222         return 0;
223 }
224
225 static int nfs4_begin_drain_session(struct nfs_client *clp)
226 {
227         struct nfs4_session *ses = clp->cl_session;
228         int ret = 0;
229
230         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
231         /* back channel */
232         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
233         if (ret)
234                 return ret;
235         /* fore channel */
236         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
237 }
238
239 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
240 {
241         int status;
242
243         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
244                 goto do_confirm;
245         nfs4_begin_drain_session(clp);
246         status = nfs4_proc_exchange_id(clp, cred);
247         if (status != 0)
248                 goto out;
249         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
250 do_confirm:
251         status = nfs4_proc_create_session(clp);
252         if (status != 0)
253                 goto out;
254         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
255         nfs41_setup_state_renewal(clp);
256         nfs_mark_client_ready(clp, NFS_CS_READY);
257 out:
258         return status;
259 }
260
261 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
262 {
263         struct rpc_cred *cred;
264
265         spin_lock(&clp->cl_lock);
266         cred = nfs4_get_machine_cred_locked(clp);
267         spin_unlock(&clp->cl_lock);
268         return cred;
269 }
270
271 #endif /* CONFIG_NFS_V4_1 */
272
273 static struct rpc_cred *
274 nfs4_get_setclientid_cred_server(struct nfs_server *server)
275 {
276         struct nfs_client *clp = server->nfs_client;
277         struct rpc_cred *cred = NULL;
278         struct nfs4_state_owner *sp;
279         struct rb_node *pos;
280
281         spin_lock(&clp->cl_lock);
282         pos = rb_first(&server->state_owners);
283         if (pos != NULL) {
284                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
285                 cred = get_rpccred(sp->so_cred);
286         }
287         spin_unlock(&clp->cl_lock);
288         return cred;
289 }
290
291 /**
292  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
293  * @clp: client state handle
294  *
295  * Returns an rpc_cred with reference count bumped, or NULL.
296  */
297 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
298 {
299         struct nfs_server *server;
300         struct rpc_cred *cred;
301
302         spin_lock(&clp->cl_lock);
303         cred = nfs4_get_machine_cred_locked(clp);
304         spin_unlock(&clp->cl_lock);
305         if (cred != NULL)
306                 goto out;
307
308         rcu_read_lock();
309         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
310                 cred = nfs4_get_setclientid_cred_server(server);
311                 if (cred != NULL)
312                         break;
313         }
314         rcu_read_unlock();
315
316 out:
317         return cred;
318 }
319
320 static void nfs_alloc_unique_id_locked(struct rb_root *root,
321                                        struct nfs_unique_id *new,
322                                        __u64 minval, int maxbits)
323 {
324         struct rb_node **p, *parent;
325         struct nfs_unique_id *pos;
326         __u64 mask = ~0ULL;
327
328         if (maxbits < 64)
329                 mask = (1ULL << maxbits) - 1ULL;
330
331         /* Ensure distribution is more or less flat */
332         get_random_bytes(&new->id, sizeof(new->id));
333         new->id &= mask;
334         if (new->id < minval)
335                 new->id += minval;
336 retry:
337         p = &root->rb_node;
338         parent = NULL;
339
340         while (*p != NULL) {
341                 parent = *p;
342                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
343
344                 if (new->id < pos->id)
345                         p = &(*p)->rb_left;
346                 else if (new->id > pos->id)
347                         p = &(*p)->rb_right;
348                 else
349                         goto id_exists;
350         }
351         rb_link_node(&new->rb_node, parent, p);
352         rb_insert_color(&new->rb_node, root);
353         return;
354 id_exists:
355         for (;;) {
356                 new->id++;
357                 if (new->id < minval || (new->id & mask) != new->id) {
358                         new->id = minval;
359                         break;
360                 }
361                 parent = rb_next(parent);
362                 if (parent == NULL)
363                         break;
364                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
365                 if (new->id < pos->id)
366                         break;
367         }
368         goto retry;
369 }
370
371 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
372 {
373         rb_erase(&id->rb_node, root);
374 }
375
376 static struct nfs4_state_owner *
377 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
378 {
379         struct rb_node **p = &server->state_owners.rb_node,
380                        *parent = NULL;
381         struct nfs4_state_owner *sp;
382
383         while (*p != NULL) {
384                 parent = *p;
385                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
386
387                 if (cred < sp->so_cred)
388                         p = &parent->rb_left;
389                 else if (cred > sp->so_cred)
390                         p = &parent->rb_right;
391                 else {
392                         if (!list_empty(&sp->so_lru))
393                                 list_del_init(&sp->so_lru);
394                         atomic_inc(&sp->so_count);
395                         return sp;
396                 }
397         }
398         return NULL;
399 }
400
401 static struct nfs4_state_owner *
402 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
403 {
404         struct nfs_server *server = new->so_server;
405         struct rb_node **p = &server->state_owners.rb_node,
406                        *parent = NULL;
407         struct nfs4_state_owner *sp;
408         int err;
409
410         while (*p != NULL) {
411                 parent = *p;
412                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
413
414                 if (new->so_cred < sp->so_cred)
415                         p = &parent->rb_left;
416                 else if (new->so_cred > sp->so_cred)
417                         p = &parent->rb_right;
418                 else {
419                         if (!list_empty(&sp->so_lru))
420                                 list_del_init(&sp->so_lru);
421                         atomic_inc(&sp->so_count);
422                         return sp;
423                 }
424         }
425         err = ida_get_new(&server->openowner_id, &new->so_owner_id);
426         if (err)
427                 return ERR_PTR(err);
428         rb_link_node(&new->so_server_node, parent, p);
429         rb_insert_color(&new->so_server_node, &server->state_owners);
430         return new;
431 }
432
433 static void
434 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
435 {
436         struct nfs_server *server = sp->so_server;
437
438         if (!RB_EMPTY_NODE(&sp->so_server_node))
439                 rb_erase(&sp->so_server_node, &server->state_owners);
440         ida_remove(&server->openowner_id, sp->so_owner_id);
441 }
442
443 /*
444  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
445  * create a new state_owner.
446  *
447  */
448 static struct nfs4_state_owner *
449 nfs4_alloc_state_owner(struct nfs_server *server,
450                 struct rpc_cred *cred,
451                 gfp_t gfp_flags)
452 {
453         struct nfs4_state_owner *sp;
454
455         sp = kzalloc(sizeof(*sp), gfp_flags);
456         if (!sp)
457                 return NULL;
458         sp->so_server = server;
459         sp->so_cred = get_rpccred(cred);
460         spin_lock_init(&sp->so_lock);
461         INIT_LIST_HEAD(&sp->so_states);
462         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
463         sp->so_seqid.sequence = &sp->so_sequence;
464         spin_lock_init(&sp->so_sequence.lock);
465         INIT_LIST_HEAD(&sp->so_sequence.list);
466         atomic_set(&sp->so_count, 1);
467         INIT_LIST_HEAD(&sp->so_lru);
468         return sp;
469 }
470
471 static void
472 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
473 {
474         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
475                 struct nfs_server *server = sp->so_server;
476                 struct nfs_client *clp = server->nfs_client;
477
478                 spin_lock(&clp->cl_lock);
479                 rb_erase(&sp->so_server_node, &server->state_owners);
480                 RB_CLEAR_NODE(&sp->so_server_node);
481                 spin_unlock(&clp->cl_lock);
482         }
483 }
484
485 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
486 {
487         rpc_destroy_wait_queue(&sp->so_sequence.wait);
488         put_rpccred(sp->so_cred);
489         kfree(sp);
490 }
491
492 static void nfs4_gc_state_owners(struct nfs_server *server)
493 {
494         struct nfs_client *clp = server->nfs_client;
495         struct nfs4_state_owner *sp, *tmp;
496         unsigned long time_min, time_max;
497         LIST_HEAD(doomed);
498
499         spin_lock(&clp->cl_lock);
500         time_max = jiffies;
501         time_min = (long)time_max - (long)clp->cl_lease_time;
502         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
503                 /* NB: LRU is sorted so that oldest is at the head */
504                 if (time_in_range(sp->so_expires, time_min, time_max))
505                         break;
506                 list_move(&sp->so_lru, &doomed);
507                 nfs4_remove_state_owner_locked(sp);
508         }
509         spin_unlock(&clp->cl_lock);
510
511         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
512                 list_del(&sp->so_lru);
513                 nfs4_free_state_owner(sp);
514         }
515 }
516
517 /**
518  * nfs4_get_state_owner - Look up a state owner given a credential
519  * @server: nfs_server to search
520  * @cred: RPC credential to match
521  *
522  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
523  */
524 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
525                                               struct rpc_cred *cred,
526                                               gfp_t gfp_flags)
527 {
528         struct nfs_client *clp = server->nfs_client;
529         struct nfs4_state_owner *sp, *new;
530
531         spin_lock(&clp->cl_lock);
532         sp = nfs4_find_state_owner_locked(server, cred);
533         spin_unlock(&clp->cl_lock);
534         if (sp != NULL)
535                 goto out;
536         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
537         if (new == NULL)
538                 goto out;
539         do {
540                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
541                         break;
542                 spin_lock(&clp->cl_lock);
543                 sp = nfs4_insert_state_owner_locked(new);
544                 spin_unlock(&clp->cl_lock);
545         } while (sp == ERR_PTR(-EAGAIN));
546         if (sp != new)
547                 nfs4_free_state_owner(new);
548 out:
549         nfs4_gc_state_owners(server);
550         return sp;
551 }
552
553 /**
554  * nfs4_put_state_owner - Release a nfs4_state_owner
555  * @sp: state owner data to release
556  */
557 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
558 {
559         struct nfs_server *server = sp->so_server;
560         struct nfs_client *clp = server->nfs_client;
561
562         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
563                 return;
564
565         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
566                 sp->so_expires = jiffies;
567                 list_add_tail(&sp->so_lru, &server->state_owners_lru);
568                 spin_unlock(&clp->cl_lock);
569         } else {
570                 nfs4_remove_state_owner_locked(sp);
571                 spin_unlock(&clp->cl_lock);
572                 nfs4_free_state_owner(sp);
573         }
574 }
575
576 /**
577  * nfs4_purge_state_owners - Release all cached state owners
578  * @server: nfs_server with cached state owners to release
579  *
580  * Called at umount time.  Remaining state owners will be on
581  * the LRU with ref count of zero.
582  */
583 void nfs4_purge_state_owners(struct nfs_server *server)
584 {
585         struct nfs_client *clp = server->nfs_client;
586         struct nfs4_state_owner *sp, *tmp;
587         LIST_HEAD(doomed);
588
589         spin_lock(&clp->cl_lock);
590         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
591                 list_move(&sp->so_lru, &doomed);
592                 nfs4_remove_state_owner_locked(sp);
593         }
594         spin_unlock(&clp->cl_lock);
595
596         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
597                 list_del(&sp->so_lru);
598                 nfs4_free_state_owner(sp);
599         }
600 }
601
602 static struct nfs4_state *
603 nfs4_alloc_open_state(void)
604 {
605         struct nfs4_state *state;
606
607         state = kzalloc(sizeof(*state), GFP_NOFS);
608         if (!state)
609                 return NULL;
610         atomic_set(&state->count, 1);
611         INIT_LIST_HEAD(&state->lock_states);
612         spin_lock_init(&state->state_lock);
613         seqlock_init(&state->seqlock);
614         return state;
615 }
616
617 void
618 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
619 {
620         if (state->state == fmode)
621                 return;
622         /* NB! List reordering - see the reclaim code for why.  */
623         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
624                 if (fmode & FMODE_WRITE)
625                         list_move(&state->open_states, &state->owner->so_states);
626                 else
627                         list_move_tail(&state->open_states, &state->owner->so_states);
628         }
629         state->state = fmode;
630 }
631
632 static struct nfs4_state *
633 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
634 {
635         struct nfs_inode *nfsi = NFS_I(inode);
636         struct nfs4_state *state;
637
638         list_for_each_entry(state, &nfsi->open_states, inode_states) {
639                 if (state->owner != owner)
640                         continue;
641                 if (atomic_inc_not_zero(&state->count))
642                         return state;
643         }
644         return NULL;
645 }
646
647 static void
648 nfs4_free_open_state(struct nfs4_state *state)
649 {
650         kfree(state);
651 }
652
653 struct nfs4_state *
654 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
655 {
656         struct nfs4_state *state, *new;
657         struct nfs_inode *nfsi = NFS_I(inode);
658
659         spin_lock(&inode->i_lock);
660         state = __nfs4_find_state_byowner(inode, owner);
661         spin_unlock(&inode->i_lock);
662         if (state)
663                 goto out;
664         new = nfs4_alloc_open_state();
665         spin_lock(&owner->so_lock);
666         spin_lock(&inode->i_lock);
667         state = __nfs4_find_state_byowner(inode, owner);
668         if (state == NULL && new != NULL) {
669                 state = new;
670                 state->owner = owner;
671                 atomic_inc(&owner->so_count);
672                 list_add(&state->inode_states, &nfsi->open_states);
673                 ihold(inode);
674                 state->inode = inode;
675                 spin_unlock(&inode->i_lock);
676                 /* Note: The reclaim code dictates that we add stateless
677                  * and read-only stateids to the end of the list */
678                 list_add_tail(&state->open_states, &owner->so_states);
679                 spin_unlock(&owner->so_lock);
680         } else {
681                 spin_unlock(&inode->i_lock);
682                 spin_unlock(&owner->so_lock);
683                 if (new)
684                         nfs4_free_open_state(new);
685         }
686 out:
687         return state;
688 }
689
690 void nfs4_put_open_state(struct nfs4_state *state)
691 {
692         struct inode *inode = state->inode;
693         struct nfs4_state_owner *owner = state->owner;
694
695         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
696                 return;
697         spin_lock(&inode->i_lock);
698         list_del(&state->inode_states);
699         list_del(&state->open_states);
700         spin_unlock(&inode->i_lock);
701         spin_unlock(&owner->so_lock);
702         iput(inode);
703         nfs4_free_open_state(state);
704         nfs4_put_state_owner(owner);
705 }
706
707 /*
708  * Close the current file.
709  */
710 static void __nfs4_close(struct nfs4_state *state,
711                 fmode_t fmode, gfp_t gfp_mask, int wait)
712 {
713         struct nfs4_state_owner *owner = state->owner;
714         int call_close = 0;
715         fmode_t newstate;
716
717         atomic_inc(&owner->so_count);
718         /* Protect against nfs4_find_state() */
719         spin_lock(&owner->so_lock);
720         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
721                 case FMODE_READ:
722                         state->n_rdonly--;
723                         break;
724                 case FMODE_WRITE:
725                         state->n_wronly--;
726                         break;
727                 case FMODE_READ|FMODE_WRITE:
728                         state->n_rdwr--;
729         }
730         newstate = FMODE_READ|FMODE_WRITE;
731         if (state->n_rdwr == 0) {
732                 if (state->n_rdonly == 0) {
733                         newstate &= ~FMODE_READ;
734                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
735                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
736                 }
737                 if (state->n_wronly == 0) {
738                         newstate &= ~FMODE_WRITE;
739                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
740                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
741                 }
742                 if (newstate == 0)
743                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
744         }
745         nfs4_state_set_mode_locked(state, newstate);
746         spin_unlock(&owner->so_lock);
747
748         if (!call_close) {
749                 nfs4_put_open_state(state);
750                 nfs4_put_state_owner(owner);
751         } else {
752                 bool roc = pnfs_roc(state->inode);
753
754                 nfs4_do_close(state, gfp_mask, wait, roc);
755         }
756 }
757
758 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
759 {
760         __nfs4_close(state, fmode, GFP_NOFS, 0);
761 }
762
763 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
764 {
765         __nfs4_close(state, fmode, GFP_KERNEL, 1);
766 }
767
768 /*
769  * Search the state->lock_states for an existing lock_owner
770  * that is compatible with current->files
771  */
772 static struct nfs4_lock_state *
773 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
774 {
775         struct nfs4_lock_state *pos;
776         list_for_each_entry(pos, &state->lock_states, ls_locks) {
777                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
778                         continue;
779                 switch (pos->ls_owner.lo_type) {
780                 case NFS4_POSIX_LOCK_TYPE:
781                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
782                                 continue;
783                         break;
784                 case NFS4_FLOCK_LOCK_TYPE:
785                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
786                                 continue;
787                 }
788                 atomic_inc(&pos->ls_count);
789                 return pos;
790         }
791         return NULL;
792 }
793
794 /*
795  * Return a compatible lock_state. If no initialized lock_state structure
796  * exists, return an uninitialized one.
797  *
798  */
799 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)
800 {
801         struct nfs4_lock_state *lsp;
802         struct nfs_server *server = state->owner->so_server;
803         struct nfs_client *clp = server->nfs_client;
804
805         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
806         if (lsp == NULL)
807                 return NULL;
808         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
809         spin_lock_init(&lsp->ls_sequence.lock);
810         INIT_LIST_HEAD(&lsp->ls_sequence.list);
811         lsp->ls_seqid.sequence = &lsp->ls_sequence;
812         atomic_set(&lsp->ls_count, 1);
813         lsp->ls_state = state;
814         lsp->ls_owner.lo_type = type;
815         switch (lsp->ls_owner.lo_type) {
816         case NFS4_FLOCK_LOCK_TYPE:
817                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
818                 break;
819         case NFS4_POSIX_LOCK_TYPE:
820                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
821                 break;
822         default:
823                 kfree(lsp);
824                 return NULL;
825         }
826         spin_lock(&clp->cl_lock);
827         nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
828         spin_unlock(&clp->cl_lock);
829         INIT_LIST_HEAD(&lsp->ls_locks);
830         return lsp;
831 }
832
833 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
834 {
835         struct nfs_server *server = lsp->ls_state->owner->so_server;
836         struct nfs_client *clp = server->nfs_client;
837
838         spin_lock(&clp->cl_lock);
839         nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
840         spin_unlock(&clp->cl_lock);
841         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
842         kfree(lsp);
843 }
844
845 /*
846  * Return a compatible lock_state. If no initialized lock_state structure
847  * exists, return an uninitialized one.
848  *
849  */
850 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
851 {
852         struct nfs4_lock_state *lsp, *new = NULL;
853         
854         for(;;) {
855                 spin_lock(&state->state_lock);
856                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
857                 if (lsp != NULL)
858                         break;
859                 if (new != NULL) {
860                         list_add(&new->ls_locks, &state->lock_states);
861                         set_bit(LK_STATE_IN_USE, &state->flags);
862                         lsp = new;
863                         new = NULL;
864                         break;
865                 }
866                 spin_unlock(&state->state_lock);
867                 new = nfs4_alloc_lock_state(state, owner, pid, type);
868                 if (new == NULL)
869                         return NULL;
870         }
871         spin_unlock(&state->state_lock);
872         if (new != NULL)
873                 nfs4_free_lock_state(new);
874         return lsp;
875 }
876
877 /*
878  * Release reference to lock_state, and free it if we see that
879  * it is no longer in use
880  */
881 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
882 {
883         struct nfs4_state *state;
884
885         if (lsp == NULL)
886                 return;
887         state = lsp->ls_state;
888         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
889                 return;
890         list_del(&lsp->ls_locks);
891         if (list_empty(&state->lock_states))
892                 clear_bit(LK_STATE_IN_USE, &state->flags);
893         spin_unlock(&state->state_lock);
894         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
895                 nfs4_release_lockowner(lsp);
896         nfs4_free_lock_state(lsp);
897 }
898
899 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
900 {
901         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
902
903         dst->fl_u.nfs4_fl.owner = lsp;
904         atomic_inc(&lsp->ls_count);
905 }
906
907 static void nfs4_fl_release_lock(struct file_lock *fl)
908 {
909         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
910 }
911
912 static const struct file_lock_operations nfs4_fl_lock_ops = {
913         .fl_copy_lock = nfs4_fl_copy_lock,
914         .fl_release_private = nfs4_fl_release_lock,
915 };
916
917 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
918 {
919         struct nfs4_lock_state *lsp;
920
921         if (fl->fl_ops != NULL)
922                 return 0;
923         if (fl->fl_flags & FL_POSIX)
924                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
925         else if (fl->fl_flags & FL_FLOCK)
926                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
927         else
928                 return -EINVAL;
929         if (lsp == NULL)
930                 return -ENOMEM;
931         fl->fl_u.nfs4_fl.owner = lsp;
932         fl->fl_ops = &nfs4_fl_lock_ops;
933         return 0;
934 }
935
936 /*
937  * Byte-range lock aware utility to initialize the stateid of read/write
938  * requests.
939  */
940 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
941 {
942         struct nfs4_lock_state *lsp;
943         int seq;
944
945         do {
946                 seq = read_seqbegin(&state->seqlock);
947                 memcpy(dst, &state->stateid, sizeof(*dst));
948         } while (read_seqretry(&state->seqlock, seq));
949         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
950                 return;
951
952         spin_lock(&state->state_lock);
953         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
954         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
955                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
956         spin_unlock(&state->state_lock);
957         nfs4_put_lock_state(lsp);
958 }
959
960 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
961 {
962         struct nfs_seqid *new;
963
964         new = kmalloc(sizeof(*new), gfp_mask);
965         if (new != NULL) {
966                 new->sequence = counter;
967                 INIT_LIST_HEAD(&new->list);
968         }
969         return new;
970 }
971
972 void nfs_release_seqid(struct nfs_seqid *seqid)
973 {
974         if (!list_empty(&seqid->list)) {
975                 struct rpc_sequence *sequence = seqid->sequence->sequence;
976
977                 spin_lock(&sequence->lock);
978                 list_del_init(&seqid->list);
979                 spin_unlock(&sequence->lock);
980                 rpc_wake_up(&sequence->wait);
981         }
982 }
983
984 void nfs_free_seqid(struct nfs_seqid *seqid)
985 {
986         nfs_release_seqid(seqid);
987         kfree(seqid);
988 }
989
990 /*
991  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
992  * failed with a seqid incrementing error -
993  * see comments nfs_fs.h:seqid_mutating_error()
994  */
995 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
996 {
997         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
998         switch (status) {
999                 case 0:
1000                         break;
1001                 case -NFS4ERR_BAD_SEQID:
1002                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1003                                 return;
1004                         printk(KERN_WARNING "NFS: v4 server returned a bad"
1005                                         " sequence-id error on an"
1006                                         " unconfirmed sequence %p!\n",
1007                                         seqid->sequence);
1008                 case -NFS4ERR_STALE_CLIENTID:
1009                 case -NFS4ERR_STALE_STATEID:
1010                 case -NFS4ERR_BAD_STATEID:
1011                 case -NFS4ERR_BADXDR:
1012                 case -NFS4ERR_RESOURCE:
1013                 case -NFS4ERR_NOFILEHANDLE:
1014                         /* Non-seqid mutating errors */
1015                         return;
1016         };
1017         /*
1018          * Note: no locking needed as we are guaranteed to be first
1019          * on the sequence list
1020          */
1021         seqid->sequence->counter++;
1022 }
1023
1024 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1025 {
1026         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1027                                         struct nfs4_state_owner, so_seqid);
1028         struct nfs_server *server = sp->so_server;
1029
1030         if (status == -NFS4ERR_BAD_SEQID)
1031                 nfs4_drop_state_owner(sp);
1032         if (!nfs4_has_session(server->nfs_client))
1033                 nfs_increment_seqid(status, seqid);
1034 }
1035
1036 /*
1037  * Increment the seqid if the LOCK/LOCKU succeeded, or
1038  * failed with a seqid incrementing error -
1039  * see comments nfs_fs.h:seqid_mutating_error()
1040  */
1041 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1042 {
1043         nfs_increment_seqid(status, seqid);
1044 }
1045
1046 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1047 {
1048         struct rpc_sequence *sequence = seqid->sequence->sequence;
1049         int status = 0;
1050
1051         spin_lock(&sequence->lock);
1052         if (list_empty(&seqid->list))
1053                 list_add_tail(&seqid->list, &sequence->list);
1054         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1055                 goto unlock;
1056         rpc_sleep_on(&sequence->wait, task, NULL);
1057         status = -EAGAIN;
1058 unlock:
1059         spin_unlock(&sequence->lock);
1060         return status;
1061 }
1062
1063 static int nfs4_run_state_manager(void *);
1064
1065 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1066 {
1067         smp_mb__before_clear_bit();
1068         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1069         smp_mb__after_clear_bit();
1070         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1071         rpc_wake_up(&clp->cl_rpcwaitq);
1072 }
1073
1074 /*
1075  * Schedule the nfs_client asynchronous state management routine
1076  */
1077 void nfs4_schedule_state_manager(struct nfs_client *clp)
1078 {
1079         struct task_struct *task;
1080
1081         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1082                 return;
1083         __module_get(THIS_MODULE);
1084         atomic_inc(&clp->cl_count);
1085         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1086                                 rpc_peeraddr2str(clp->cl_rpcclient,
1087                                                         RPC_DISPLAY_ADDR));
1088         if (!IS_ERR(task))
1089                 return;
1090         nfs4_clear_state_manager_bit(clp);
1091         nfs_put_client(clp);
1092         module_put(THIS_MODULE);
1093 }
1094
1095 /*
1096  * Schedule a lease recovery attempt
1097  */
1098 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1099 {
1100         if (!clp)
1101                 return;
1102         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1103                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1104         nfs4_schedule_state_manager(clp);
1105 }
1106
1107 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1108 {
1109         nfs_handle_cb_pathdown(clp);
1110         nfs4_schedule_state_manager(clp);
1111 }
1112
1113 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1114 {
1115
1116         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1117         /* Don't recover state that expired before the reboot */
1118         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1119                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1120                 return 0;
1121         }
1122         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1123         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1124         return 1;
1125 }
1126
1127 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1128 {
1129         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1130         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1131         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1132         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1133         return 1;
1134 }
1135
1136 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1137 {
1138         struct nfs_client *clp = server->nfs_client;
1139
1140         nfs4_state_mark_reclaim_nograce(clp, state);
1141         nfs4_schedule_state_manager(clp);
1142 }
1143
1144 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1145 {
1146         struct inode *inode = state->inode;
1147         struct nfs_inode *nfsi = NFS_I(inode);
1148         struct file_lock *fl;
1149         int status = 0;
1150
1151         if (inode->i_flock == NULL)
1152                 return 0;
1153
1154         /* Guard against delegation returns and new lock/unlock calls */
1155         down_write(&nfsi->rwsem);
1156         /* Protect inode->i_flock using the BKL */
1157         lock_flocks();
1158         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1159                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1160                         continue;
1161                 if (nfs_file_open_context(fl->fl_file)->state != state)
1162                         continue;
1163                 unlock_flocks();
1164                 status = ops->recover_lock(state, fl);
1165                 switch (status) {
1166                         case 0:
1167                                 break;
1168                         case -ESTALE:
1169                         case -NFS4ERR_ADMIN_REVOKED:
1170                         case -NFS4ERR_STALE_STATEID:
1171                         case -NFS4ERR_BAD_STATEID:
1172                         case -NFS4ERR_EXPIRED:
1173                         case -NFS4ERR_NO_GRACE:
1174                         case -NFS4ERR_STALE_CLIENTID:
1175                         case -NFS4ERR_BADSESSION:
1176                         case -NFS4ERR_BADSLOT:
1177                         case -NFS4ERR_BAD_HIGH_SLOT:
1178                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1179                                 goto out;
1180                         default:
1181                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1182                                                 __func__, status);
1183                         case -ENOMEM:
1184                         case -NFS4ERR_DENIED:
1185                         case -NFS4ERR_RECLAIM_BAD:
1186                         case -NFS4ERR_RECLAIM_CONFLICT:
1187                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1188                                 status = 0;
1189                 }
1190                 lock_flocks();
1191         }
1192         unlock_flocks();
1193 out:
1194         up_write(&nfsi->rwsem);
1195         return status;
1196 }
1197
1198 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1199 {
1200         struct nfs4_state *state;
1201         struct nfs4_lock_state *lock;
1202         int status = 0;
1203
1204         /* Note: we rely on the sp->so_states list being ordered 
1205          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1206          * states first.
1207          * This is needed to ensure that the server won't give us any
1208          * read delegations that we have to return if, say, we are
1209          * recovering after a network partition or a reboot from a
1210          * server that doesn't support a grace period.
1211          */
1212 restart:
1213         spin_lock(&sp->so_lock);
1214         list_for_each_entry(state, &sp->so_states, open_states) {
1215                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1216                         continue;
1217                 if (state->state == 0)
1218                         continue;
1219                 atomic_inc(&state->count);
1220                 spin_unlock(&sp->so_lock);
1221                 status = ops->recover_open(sp, state);
1222                 if (status >= 0) {
1223                         status = nfs4_reclaim_locks(state, ops);
1224                         if (status >= 0) {
1225                                 spin_lock(&state->state_lock);
1226                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1227                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1228                                                 printk("%s: Lock reclaim failed!\n",
1229                                                         __func__);
1230                                 }
1231                                 spin_unlock(&state->state_lock);
1232                                 nfs4_put_open_state(state);
1233                                 goto restart;
1234                         }
1235                 }
1236                 switch (status) {
1237                         default:
1238                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1239                                                 __func__, status);
1240                         case -ENOENT:
1241                         case -ENOMEM:
1242                         case -ESTALE:
1243                                 /*
1244                                  * Open state on this file cannot be recovered
1245                                  * All we can do is revert to using the zero stateid.
1246                                  */
1247                                 memset(state->stateid.data, 0,
1248                                         sizeof(state->stateid.data));
1249                                 /* Mark the file as being 'closed' */
1250                                 state->state = 0;
1251                                 break;
1252                         case -EKEYEXPIRED:
1253                                 /*
1254                                  * User RPCSEC_GSS context has expired.
1255                                  * We cannot recover this stateid now, so
1256                                  * skip it and allow recovery thread to
1257                                  * proceed.
1258                                  */
1259                                 break;
1260                         case -NFS4ERR_ADMIN_REVOKED:
1261                         case -NFS4ERR_STALE_STATEID:
1262                         case -NFS4ERR_BAD_STATEID:
1263                         case -NFS4ERR_RECLAIM_BAD:
1264                         case -NFS4ERR_RECLAIM_CONFLICT:
1265                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1266                                 break;
1267                         case -NFS4ERR_EXPIRED:
1268                         case -NFS4ERR_NO_GRACE:
1269                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1270                         case -NFS4ERR_STALE_CLIENTID:
1271                         case -NFS4ERR_BADSESSION:
1272                         case -NFS4ERR_BADSLOT:
1273                         case -NFS4ERR_BAD_HIGH_SLOT:
1274                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1275                                 goto out_err;
1276                 }
1277                 nfs4_put_open_state(state);
1278                 goto restart;
1279         }
1280         spin_unlock(&sp->so_lock);
1281         return 0;
1282 out_err:
1283         nfs4_put_open_state(state);
1284         return status;
1285 }
1286
1287 static void nfs4_clear_open_state(struct nfs4_state *state)
1288 {
1289         struct nfs4_lock_state *lock;
1290
1291         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1292         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1293         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1294         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1295         spin_lock(&state->state_lock);
1296         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1297                 lock->ls_seqid.flags = 0;
1298                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1299         }
1300         spin_unlock(&state->state_lock);
1301 }
1302
1303 static void nfs4_reset_seqids(struct nfs_server *server,
1304         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1305 {
1306         struct nfs_client *clp = server->nfs_client;
1307         struct nfs4_state_owner *sp;
1308         struct rb_node *pos;
1309         struct nfs4_state *state;
1310
1311         spin_lock(&clp->cl_lock);
1312         for (pos = rb_first(&server->state_owners);
1313              pos != NULL;
1314              pos = rb_next(pos)) {
1315                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1316                 sp->so_seqid.flags = 0;
1317                 spin_lock(&sp->so_lock);
1318                 list_for_each_entry(state, &sp->so_states, open_states) {
1319                         if (mark_reclaim(clp, state))
1320                                 nfs4_clear_open_state(state);
1321                 }
1322                 spin_unlock(&sp->so_lock);
1323         }
1324         spin_unlock(&clp->cl_lock);
1325 }
1326
1327 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1328         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1329 {
1330         struct nfs_server *server;
1331
1332         rcu_read_lock();
1333         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1334                 nfs4_reset_seqids(server, mark_reclaim);
1335         rcu_read_unlock();
1336 }
1337
1338 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1339 {
1340         /* Mark all delegations for reclaim */
1341         nfs_delegation_mark_reclaim(clp);
1342         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1343 }
1344
1345 static void nfs4_reclaim_complete(struct nfs_client *clp,
1346                                  const struct nfs4_state_recovery_ops *ops)
1347 {
1348         /* Notify the server we're done reclaiming our state */
1349         if (ops->reclaim_complete)
1350                 (void)ops->reclaim_complete(clp);
1351 }
1352
1353 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1354 {
1355         struct nfs_client *clp = server->nfs_client;
1356         struct nfs4_state_owner *sp;
1357         struct rb_node *pos;
1358         struct nfs4_state *state;
1359
1360         spin_lock(&clp->cl_lock);
1361         for (pos = rb_first(&server->state_owners);
1362              pos != NULL;
1363              pos = rb_next(pos)) {
1364                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1365                 spin_lock(&sp->so_lock);
1366                 list_for_each_entry(state, &sp->so_states, open_states) {
1367                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1368                                                 &state->flags))
1369                                 continue;
1370                         nfs4_state_mark_reclaim_nograce(clp, state);
1371                 }
1372                 spin_unlock(&sp->so_lock);
1373         }
1374         spin_unlock(&clp->cl_lock);
1375 }
1376
1377 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1378 {
1379         struct nfs_server *server;
1380
1381         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1382                 return 0;
1383
1384         rcu_read_lock();
1385         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1386                 nfs4_clear_reclaim_server(server);
1387         rcu_read_unlock();
1388
1389         nfs_delegation_reap_unclaimed(clp);
1390         return 1;
1391 }
1392
1393 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1394 {
1395         if (!nfs4_state_clear_reclaim_reboot(clp))
1396                 return;
1397         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1398 }
1399
1400 static void nfs_delegation_clear_all(struct nfs_client *clp)
1401 {
1402         nfs_delegation_mark_reclaim(clp);
1403         nfs_delegation_reap_unclaimed(clp);
1404 }
1405
1406 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1407 {
1408         nfs_delegation_clear_all(clp);
1409         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1410 }
1411
1412 static void nfs4_warn_keyexpired(const char *s)
1413 {
1414         printk_ratelimited(KERN_WARNING "Error: state manager"
1415                         " encountered RPCSEC_GSS session"
1416                         " expired against NFSv4 server %s.\n",
1417                         s);
1418 }
1419
1420 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1421 {
1422         switch (error) {
1423                 case 0:
1424                         break;
1425                 case -NFS4ERR_CB_PATH_DOWN:
1426                         nfs_handle_cb_pathdown(clp);
1427                         break;
1428                 case -NFS4ERR_NO_GRACE:
1429                         nfs4_state_end_reclaim_reboot(clp);
1430                         break;
1431                 case -NFS4ERR_STALE_CLIENTID:
1432                 case -NFS4ERR_LEASE_MOVED:
1433                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1434                         nfs4_state_clear_reclaim_reboot(clp);
1435                         nfs4_state_start_reclaim_reboot(clp);
1436                         break;
1437                 case -NFS4ERR_EXPIRED:
1438                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1439                         nfs4_state_start_reclaim_nograce(clp);
1440                         break;
1441                 case -NFS4ERR_BADSESSION:
1442                 case -NFS4ERR_BADSLOT:
1443                 case -NFS4ERR_BAD_HIGH_SLOT:
1444                 case -NFS4ERR_DEADSESSION:
1445                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1446                 case -NFS4ERR_SEQ_FALSE_RETRY:
1447                 case -NFS4ERR_SEQ_MISORDERED:
1448                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1449                         /* Zero session reset errors */
1450                         break;
1451                 case -EKEYEXPIRED:
1452                         /* Nothing we can do */
1453                         nfs4_warn_keyexpired(clp->cl_hostname);
1454                         break;
1455                 default:
1456                         return error;
1457         }
1458         return 0;
1459 }
1460
1461 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1462 {
1463         struct nfs4_state_owner *sp;
1464         struct nfs_server *server;
1465         struct rb_node *pos;
1466         int status = 0;
1467
1468 restart:
1469         rcu_read_lock();
1470         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1471                 nfs4_purge_state_owners(server);
1472                 spin_lock(&clp->cl_lock);
1473                 for (pos = rb_first(&server->state_owners);
1474                      pos != NULL;
1475                      pos = rb_next(pos)) {
1476                         sp = rb_entry(pos,
1477                                 struct nfs4_state_owner, so_server_node);
1478                         if (!test_and_clear_bit(ops->owner_flag_bit,
1479                                                         &sp->so_flags))
1480                                 continue;
1481                         atomic_inc(&sp->so_count);
1482                         spin_unlock(&clp->cl_lock);
1483                         rcu_read_unlock();
1484
1485                         status = nfs4_reclaim_open_state(sp, ops);
1486                         if (status < 0) {
1487                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1488                                 nfs4_put_state_owner(sp);
1489                                 return nfs4_recovery_handle_error(clp, status);
1490                         }
1491
1492                         nfs4_put_state_owner(sp);
1493                         goto restart;
1494                 }
1495                 spin_unlock(&clp->cl_lock);
1496         }
1497         rcu_read_unlock();
1498         return status;
1499 }
1500
1501 static int nfs4_check_lease(struct nfs_client *clp)
1502 {
1503         struct rpc_cred *cred;
1504         const struct nfs4_state_maintenance_ops *ops =
1505                 clp->cl_mvops->state_renewal_ops;
1506         int status;
1507
1508         /* Is the client already known to have an expired lease? */
1509         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1510                 return 0;
1511         spin_lock(&clp->cl_lock);
1512         cred = ops->get_state_renewal_cred_locked(clp);
1513         spin_unlock(&clp->cl_lock);
1514         if (cred == NULL) {
1515                 cred = nfs4_get_setclientid_cred(clp);
1516                 status = -ENOKEY;
1517                 if (cred == NULL)
1518                         goto out;
1519         }
1520         status = ops->renew_lease(clp, cred);
1521         put_rpccred(cred);
1522 out:
1523         return nfs4_recovery_handle_error(clp, status);
1524 }
1525
1526 static int nfs4_reclaim_lease(struct nfs_client *clp)
1527 {
1528         struct rpc_cred *cred;
1529         const struct nfs4_state_recovery_ops *ops =
1530                 clp->cl_mvops->reboot_recovery_ops;
1531         int status = -ENOENT;
1532
1533         cred = ops->get_clid_cred(clp);
1534         if (cred != NULL) {
1535                 status = ops->establish_clid(clp, cred);
1536                 put_rpccred(cred);
1537                 /* Handle case where the user hasn't set up machine creds */
1538                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1539                         nfs4_clear_machine_cred(clp);
1540                         status = -EAGAIN;
1541                 }
1542                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1543                         status = -EPROTONOSUPPORT;
1544         }
1545         return status;
1546 }
1547
1548 #ifdef CONFIG_NFS_V4_1
1549 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1550 {
1551         struct nfs_client *clp = session->clp;
1552
1553         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1554         nfs4_schedule_lease_recovery(clp);
1555 }
1556 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1557
1558 void nfs41_handle_recall_slot(struct nfs_client *clp)
1559 {
1560         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1561         nfs4_schedule_state_manager(clp);
1562 }
1563
1564 static void nfs4_reset_all_state(struct nfs_client *clp)
1565 {
1566         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1567                 clp->cl_boot_time = CURRENT_TIME;
1568                 nfs4_state_start_reclaim_nograce(clp);
1569                 nfs4_schedule_state_manager(clp);
1570         }
1571 }
1572
1573 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1574 {
1575         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1576                 nfs4_state_start_reclaim_reboot(clp);
1577                 nfs4_schedule_state_manager(clp);
1578         }
1579 }
1580
1581 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1582 {
1583         /* Temporary */
1584         nfs4_reset_all_state(clp);
1585 }
1586
1587 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1588 {
1589         /* This will need to handle layouts too */
1590         nfs_expire_all_delegations(clp);
1591 }
1592
1593 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1594 {
1595         nfs_expire_all_delegations(clp);
1596         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1597                 nfs4_schedule_state_manager(clp);
1598 }
1599
1600 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1601 {
1602         if (!flags)
1603                 return;
1604         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1605                 nfs41_handle_server_reboot(clp);
1606         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1607                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1608                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1609                             SEQ4_STATUS_LEASE_MOVED))
1610                 nfs41_handle_state_revoked(clp);
1611         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1612                 nfs41_handle_recallable_state_revoked(clp);
1613         if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1614                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1615                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1616                 nfs41_handle_cb_path_down(clp);
1617 }
1618
1619 static int nfs4_reset_session(struct nfs_client *clp)
1620 {
1621         int status;
1622
1623         nfs4_begin_drain_session(clp);
1624         status = nfs4_proc_destroy_session(clp->cl_session);
1625         if (status && status != -NFS4ERR_BADSESSION &&
1626             status != -NFS4ERR_DEADSESSION) {
1627                 status = nfs4_recovery_handle_error(clp, status);
1628                 goto out;
1629         }
1630
1631         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1632         status = nfs4_proc_create_session(clp);
1633         if (status) {
1634                 status = nfs4_recovery_handle_error(clp, status);
1635                 goto out;
1636         }
1637         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1638         /* create_session negotiated new slot table */
1639         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1640
1641          /* Let the state manager reestablish state */
1642         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1643                 nfs41_setup_state_renewal(clp);
1644 out:
1645         return status;
1646 }
1647
1648 static int nfs4_recall_slot(struct nfs_client *clp)
1649 {
1650         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1651         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1652         struct nfs4_slot *new, *old;
1653         int i;
1654
1655         nfs4_begin_drain_session(clp);
1656         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1657                       GFP_NOFS);
1658         if (!new)
1659                 return -ENOMEM;
1660
1661         spin_lock(&fc_tbl->slot_tbl_lock);
1662         for (i = 0; i < fc_tbl->target_max_slots; i++)
1663                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1664         old = fc_tbl->slots;
1665         fc_tbl->slots = new;
1666         fc_tbl->max_slots = fc_tbl->target_max_slots;
1667         fc_tbl->target_max_slots = 0;
1668         fc_attrs->max_reqs = fc_tbl->max_slots;
1669         spin_unlock(&fc_tbl->slot_tbl_lock);
1670
1671         kfree(old);
1672         nfs4_end_drain_session(clp);
1673         return 0;
1674 }
1675
1676 #else /* CONFIG_NFS_V4_1 */
1677 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1678 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1679 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1680 #endif /* CONFIG_NFS_V4_1 */
1681
1682 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1683  * on EXCHANGE_ID for v4.1
1684  */
1685 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1686 {
1687         switch (status) {
1688         case -NFS4ERR_CLID_INUSE:
1689         case -NFS4ERR_STALE_CLIENTID:
1690                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1691                 break;
1692         case -NFS4ERR_DELAY:
1693         case -ETIMEDOUT:
1694         case -EAGAIN:
1695                 ssleep(1);
1696                 break;
1697
1698         case -EKEYEXPIRED:
1699                 nfs4_warn_keyexpired(clp->cl_hostname);
1700         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1701                                  * in nfs4_exchange_id */
1702         default:
1703                 return;
1704         }
1705         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1706 }
1707
1708 static void nfs4_state_manager(struct nfs_client *clp)
1709 {
1710         int status = 0;
1711
1712         /* Ensure exclusive access to NFSv4 state */
1713         do {
1714                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1715                         /* We're going to have to re-establish a clientid */
1716                         status = nfs4_reclaim_lease(clp);
1717                         if (status) {
1718                                 nfs4_set_lease_expired(clp, status);
1719                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1720                                                         &clp->cl_state))
1721                                         continue;
1722                                 if (clp->cl_cons_state ==
1723                                                         NFS_CS_SESSION_INITING)
1724                                         nfs_mark_client_ready(clp, status);
1725                                 goto out_error;
1726                         }
1727                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1728
1729                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1730                                                &clp->cl_state))
1731                                 nfs4_state_start_reclaim_nograce(clp);
1732                         else
1733                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1734                                         &clp->cl_state);
1735
1736                         pnfs_destroy_all_layouts(clp);
1737                 }
1738
1739                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1740                         status = nfs4_check_lease(clp);
1741                         if (status < 0)
1742                                 goto out_error;
1743                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1744                                 continue;
1745                 }
1746
1747                 /* Initialize or reset the session */
1748                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1749                    && nfs4_has_session(clp)) {
1750                         status = nfs4_reset_session(clp);
1751                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1752                                 continue;
1753                         if (status < 0)
1754                                 goto out_error;
1755                 }
1756
1757                 /* First recover reboot state... */
1758                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1759                         status = nfs4_do_reclaim(clp,
1760                                 clp->cl_mvops->reboot_recovery_ops);
1761                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1762                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1763                                 continue;
1764                         nfs4_state_end_reclaim_reboot(clp);
1765                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1766                                 continue;
1767                         if (status < 0)
1768                                 goto out_error;
1769                 }
1770
1771                 /* Now recover expired state... */
1772                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1773                         status = nfs4_do_reclaim(clp,
1774                                 clp->cl_mvops->nograce_recovery_ops);
1775                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1776                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1777                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1778                                 continue;
1779                         if (status < 0)
1780                                 goto out_error;
1781                 }
1782
1783                 nfs4_end_drain_session(clp);
1784                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1785                         nfs_client_return_marked_delegations(clp);
1786                         continue;
1787                 }
1788                 /* Recall session slots */
1789                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1790                    && nfs4_has_session(clp)) {
1791                         status = nfs4_recall_slot(clp);
1792                         if (status < 0)
1793                                 goto out_error;
1794                         continue;
1795                 }
1796
1797
1798                 nfs4_clear_state_manager_bit(clp);
1799                 /* Did we race with an attempt to give us more work? */
1800                 if (clp->cl_state == 0)
1801                         break;
1802                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1803                         break;
1804         } while (atomic_read(&clp->cl_count) > 1);
1805         return;
1806 out_error:
1807         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1808                         " with error %d\n", clp->cl_hostname, -status);
1809         nfs4_end_drain_session(clp);
1810         nfs4_clear_state_manager_bit(clp);
1811 }
1812
1813 static int nfs4_run_state_manager(void *ptr)
1814 {
1815         struct nfs_client *clp = ptr;
1816
1817         allow_signal(SIGKILL);
1818         nfs4_state_manager(clp);
1819         nfs_put_client(clp);
1820         module_put_and_exit(0);
1821         return 0;
1822 }
1823
1824 /*
1825  * Local variables:
1826  *  c-basic-offset: 8
1827  * End:
1828  */