]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/nfs/nfs4state.c
72ab97ef3d617fdee9d2f538908dcac245f42093
[~shefty/rdma-dev.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
53 #include "nfs4_fs.h"
54 #include "callback.h"
55 #include "delegation.h"
56 #include "internal.h"
57 #include "pnfs.h"
58
59 #define OPENOWNER_POOL_SIZE     8
60
61 const nfs4_stateid zero_stateid;
62
63 static LIST_HEAD(nfs4_clientid_list);
64
65 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
66 {
67         struct nfs4_setclientid_res clid = {
68                 .clientid = clp->cl_clientid,
69                 .confirm = clp->cl_confirm,
70         };
71         unsigned short port;
72         int status;
73
74         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
75                 goto do_confirm;
76         port = nfs_callback_tcpport;
77         if (clp->cl_addr.ss_family == AF_INET6)
78                 port = nfs_callback_tcpport6;
79
80         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
81         if (status != 0)
82                 goto out;
83         clp->cl_clientid = clid.clientid;
84         clp->cl_confirm = clid.confirm;
85         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
86 do_confirm:
87         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
88         if (status != 0)
89                 goto out;
90         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
91         nfs4_schedule_state_renewal(clp);
92 out:
93         return status;
94 }
95
96 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
97 {
98         struct rpc_cred *cred = NULL;
99
100         if (clp->cl_machine_cred != NULL)
101                 cred = get_rpccred(clp->cl_machine_cred);
102         return cred;
103 }
104
105 static void nfs4_clear_machine_cred(struct nfs_client *clp)
106 {
107         struct rpc_cred *cred;
108
109         spin_lock(&clp->cl_lock);
110         cred = clp->cl_machine_cred;
111         clp->cl_machine_cred = NULL;
112         spin_unlock(&clp->cl_lock);
113         if (cred != NULL)
114                 put_rpccred(cred);
115 }
116
117 static struct rpc_cred *
118 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
119 {
120         struct rpc_cred *cred = NULL;
121         struct nfs4_state_owner *sp;
122         struct rb_node *pos;
123
124         for (pos = rb_first(&server->state_owners);
125              pos != NULL;
126              pos = rb_next(pos)) {
127                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
128                 if (list_empty(&sp->so_states))
129                         continue;
130                 cred = get_rpccred(sp->so_cred);
131                 break;
132         }
133         return cred;
134 }
135
136 /**
137  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
138  * @clp: client state handle
139  *
140  * Returns an rpc_cred with reference count bumped, or NULL.
141  * Caller must hold clp->cl_lock.
142  */
143 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
144 {
145         struct rpc_cred *cred = NULL;
146         struct nfs_server *server;
147
148         rcu_read_lock();
149         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
150                 cred = nfs4_get_renew_cred_server_locked(server);
151                 if (cred != NULL)
152                         break;
153         }
154         rcu_read_unlock();
155         return cred;
156 }
157
158 #if defined(CONFIG_NFS_V4_1)
159
160 static int nfs41_setup_state_renewal(struct nfs_client *clp)
161 {
162         int status;
163         struct nfs_fsinfo fsinfo;
164
165         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
166                 nfs4_schedule_state_renewal(clp);
167                 return 0;
168         }
169
170         status = nfs4_proc_get_lease_time(clp, &fsinfo);
171         if (status == 0) {
172                 /* Update lease time and schedule renewal */
173                 spin_lock(&clp->cl_lock);
174                 clp->cl_lease_time = fsinfo.lease_time * HZ;
175                 clp->cl_last_renewal = jiffies;
176                 spin_unlock(&clp->cl_lock);
177
178                 nfs4_schedule_state_renewal(clp);
179         }
180
181         return status;
182 }
183
184 /*
185  * Back channel returns NFS4ERR_DELAY for new requests when
186  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
187  * is ended.
188  */
189 static void nfs4_end_drain_session(struct nfs_client *clp)
190 {
191         struct nfs4_session *ses = clp->cl_session;
192         int max_slots;
193
194         if (ses == NULL)
195                 return;
196         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
197                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
198                 max_slots = ses->fc_slot_table.max_slots;
199                 while (max_slots--) {
200                         struct rpc_task *task;
201
202                         task = rpc_wake_up_next(&ses->fc_slot_table.
203                                                 slot_tbl_waitq);
204                         if (!task)
205                                 break;
206                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
207                 }
208                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
209         }
210 }
211
212 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
213 {
214         spin_lock(&tbl->slot_tbl_lock);
215         if (tbl->highest_used_slotid != -1) {
216                 INIT_COMPLETION(tbl->complete);
217                 spin_unlock(&tbl->slot_tbl_lock);
218                 return wait_for_completion_interruptible(&tbl->complete);
219         }
220         spin_unlock(&tbl->slot_tbl_lock);
221         return 0;
222 }
223
224 static int nfs4_begin_drain_session(struct nfs_client *clp)
225 {
226         struct nfs4_session *ses = clp->cl_session;
227         int ret = 0;
228
229         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
230         /* back channel */
231         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
232         if (ret)
233                 return ret;
234         /* fore channel */
235         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
236 }
237
238 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
239 {
240         int status;
241
242         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
243                 goto do_confirm;
244         nfs4_begin_drain_session(clp);
245         status = nfs4_proc_exchange_id(clp, cred);
246         if (status != 0)
247                 goto out;
248         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
249 do_confirm:
250         status = nfs4_proc_create_session(clp);
251         if (status != 0)
252                 goto out;
253         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
254         nfs41_setup_state_renewal(clp);
255         nfs_mark_client_ready(clp, NFS_CS_READY);
256 out:
257         return status;
258 }
259
260 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
261 {
262         struct rpc_cred *cred;
263
264         spin_lock(&clp->cl_lock);
265         cred = nfs4_get_machine_cred_locked(clp);
266         spin_unlock(&clp->cl_lock);
267         return cred;
268 }
269
270 #endif /* CONFIG_NFS_V4_1 */
271
272 static struct rpc_cred *
273 nfs4_get_setclientid_cred_server(struct nfs_server *server)
274 {
275         struct nfs_client *clp = server->nfs_client;
276         struct rpc_cred *cred = NULL;
277         struct nfs4_state_owner *sp;
278         struct rb_node *pos;
279
280         spin_lock(&clp->cl_lock);
281         pos = rb_first(&server->state_owners);
282         if (pos != NULL) {
283                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
284                 cred = get_rpccred(sp->so_cred);
285         }
286         spin_unlock(&clp->cl_lock);
287         return cred;
288 }
289
290 /**
291  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
292  * @clp: client state handle
293  *
294  * Returns an rpc_cred with reference count bumped, or NULL.
295  */
296 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
297 {
298         struct nfs_server *server;
299         struct rpc_cred *cred;
300
301         spin_lock(&clp->cl_lock);
302         cred = nfs4_get_machine_cred_locked(clp);
303         spin_unlock(&clp->cl_lock);
304         if (cred != NULL)
305                 goto out;
306
307         rcu_read_lock();
308         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
309                 cred = nfs4_get_setclientid_cred_server(server);
310                 if (cred != NULL)
311                         break;
312         }
313         rcu_read_unlock();
314
315 out:
316         return cred;
317 }
318
319 static void nfs_alloc_unique_id_locked(struct rb_root *root,
320                                        struct nfs_unique_id *new,
321                                        __u64 minval, int maxbits)
322 {
323         struct rb_node **p, *parent;
324         struct nfs_unique_id *pos;
325         __u64 mask = ~0ULL;
326
327         if (maxbits < 64)
328                 mask = (1ULL << maxbits) - 1ULL;
329
330         /* Ensure distribution is more or less flat */
331         get_random_bytes(&new->id, sizeof(new->id));
332         new->id &= mask;
333         if (new->id < minval)
334                 new->id += minval;
335 retry:
336         p = &root->rb_node;
337         parent = NULL;
338
339         while (*p != NULL) {
340                 parent = *p;
341                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
342
343                 if (new->id < pos->id)
344                         p = &(*p)->rb_left;
345                 else if (new->id > pos->id)
346                         p = &(*p)->rb_right;
347                 else
348                         goto id_exists;
349         }
350         rb_link_node(&new->rb_node, parent, p);
351         rb_insert_color(&new->rb_node, root);
352         return;
353 id_exists:
354         for (;;) {
355                 new->id++;
356                 if (new->id < minval || (new->id & mask) != new->id) {
357                         new->id = minval;
358                         break;
359                 }
360                 parent = rb_next(parent);
361                 if (parent == NULL)
362                         break;
363                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
364                 if (new->id < pos->id)
365                         break;
366         }
367         goto retry;
368 }
369
370 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
371 {
372         rb_erase(&id->rb_node, root);
373 }
374
375 static struct nfs4_state_owner *
376 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
377 {
378         struct rb_node **p = &server->state_owners.rb_node,
379                        *parent = NULL;
380         struct nfs4_state_owner *sp, *res = NULL;
381
382         while (*p != NULL) {
383                 parent = *p;
384                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
385
386                 if (server < sp->so_server) {
387                         p = &parent->rb_left;
388                         continue;
389                 }
390                 if (server > sp->so_server) {
391                         p = &parent->rb_right;
392                         continue;
393                 }
394                 if (cred < sp->so_cred)
395                         p = &parent->rb_left;
396                 else if (cred > sp->so_cred)
397                         p = &parent->rb_right;
398                 else {
399                         atomic_inc(&sp->so_count);
400                         res = sp;
401                         break;
402                 }
403         }
404         return res;
405 }
406
407 static struct nfs4_state_owner *
408 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
409 {
410         struct nfs_server *server = new->so_server;
411         struct rb_node **p = &server->state_owners.rb_node,
412                        *parent = NULL;
413         struct nfs4_state_owner *sp;
414
415         while (*p != NULL) {
416                 parent = *p;
417                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
418
419                 if (new->so_cred < sp->so_cred)
420                         p = &parent->rb_left;
421                 else if (new->so_cred > sp->so_cred)
422                         p = &parent->rb_right;
423                 else {
424                         atomic_inc(&sp->so_count);
425                         return sp;
426                 }
427         }
428         nfs_alloc_unique_id_locked(&server->openowner_id,
429                                         &new->so_owner_id, 1, 64);
430         rb_link_node(&new->so_server_node, parent, p);
431         rb_insert_color(&new->so_server_node, &server->state_owners);
432         return new;
433 }
434
435 static void
436 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
437 {
438         struct nfs_server *server = sp->so_server;
439
440         if (!RB_EMPTY_NODE(&sp->so_server_node))
441                 rb_erase(&sp->so_server_node, &server->state_owners);
442         nfs_free_unique_id(&server->openowner_id, &sp->so_owner_id);
443 }
444
445 /*
446  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
447  * create a new state_owner.
448  *
449  */
450 static struct nfs4_state_owner *
451 nfs4_alloc_state_owner(void)
452 {
453         struct nfs4_state_owner *sp;
454
455         sp = kzalloc(sizeof(*sp),GFP_NOFS);
456         if (!sp)
457                 return NULL;
458         spin_lock_init(&sp->so_lock);
459         INIT_LIST_HEAD(&sp->so_states);
460         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
461         sp->so_seqid.sequence = &sp->so_sequence;
462         spin_lock_init(&sp->so_sequence.lock);
463         INIT_LIST_HEAD(&sp->so_sequence.list);
464         atomic_set(&sp->so_count, 1);
465         return sp;
466 }
467
468 static void
469 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
470 {
471         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
472                 struct nfs_server *server = sp->so_server;
473                 struct nfs_client *clp = server->nfs_client;
474
475                 spin_lock(&clp->cl_lock);
476                 rb_erase(&sp->so_server_node, &server->state_owners);
477                 RB_CLEAR_NODE(&sp->so_server_node);
478                 spin_unlock(&clp->cl_lock);
479         }
480 }
481
482 /**
483  * nfs4_get_state_owner - Look up a state owner given a credential
484  * @server: nfs_server to search
485  * @cred: RPC credential to match
486  *
487  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
488  */
489 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
490                                               struct rpc_cred *cred)
491 {
492         struct nfs_client *clp = server->nfs_client;
493         struct nfs4_state_owner *sp, *new;
494
495         spin_lock(&clp->cl_lock);
496         sp = nfs4_find_state_owner_locked(server, cred);
497         spin_unlock(&clp->cl_lock);
498         if (sp != NULL)
499                 return sp;
500         new = nfs4_alloc_state_owner();
501         if (new == NULL)
502                 return NULL;
503         new->so_server = server;
504         new->so_cred = cred;
505         spin_lock(&clp->cl_lock);
506         sp = nfs4_insert_state_owner_locked(new);
507         spin_unlock(&clp->cl_lock);
508         if (sp == new)
509                 get_rpccred(cred);
510         else {
511                 rpc_destroy_wait_queue(&new->so_sequence.wait);
512                 kfree(new);
513         }
514         return sp;
515 }
516
517 /**
518  * nfs4_put_state_owner - Release a nfs4_state_owner
519  * @sp: state owner data to release
520  *
521  */
522 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
523 {
524         struct nfs_client *clp = sp->so_server->nfs_client;
525         struct rpc_cred *cred = sp->so_cred;
526
527         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
528                 return;
529         nfs4_remove_state_owner_locked(sp);
530         spin_unlock(&clp->cl_lock);
531         rpc_destroy_wait_queue(&sp->so_sequence.wait);
532         put_rpccred(cred);
533         kfree(sp);
534 }
535
536 static struct nfs4_state *
537 nfs4_alloc_open_state(void)
538 {
539         struct nfs4_state *state;
540
541         state = kzalloc(sizeof(*state), GFP_NOFS);
542         if (!state)
543                 return NULL;
544         atomic_set(&state->count, 1);
545         INIT_LIST_HEAD(&state->lock_states);
546         spin_lock_init(&state->state_lock);
547         seqlock_init(&state->seqlock);
548         return state;
549 }
550
551 void
552 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
553 {
554         if (state->state == fmode)
555                 return;
556         /* NB! List reordering - see the reclaim code for why.  */
557         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
558                 if (fmode & FMODE_WRITE)
559                         list_move(&state->open_states, &state->owner->so_states);
560                 else
561                         list_move_tail(&state->open_states, &state->owner->so_states);
562         }
563         state->state = fmode;
564 }
565
566 static struct nfs4_state *
567 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
568 {
569         struct nfs_inode *nfsi = NFS_I(inode);
570         struct nfs4_state *state;
571
572         list_for_each_entry(state, &nfsi->open_states, inode_states) {
573                 if (state->owner != owner)
574                         continue;
575                 if (atomic_inc_not_zero(&state->count))
576                         return state;
577         }
578         return NULL;
579 }
580
581 static void
582 nfs4_free_open_state(struct nfs4_state *state)
583 {
584         kfree(state);
585 }
586
587 struct nfs4_state *
588 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
589 {
590         struct nfs4_state *state, *new;
591         struct nfs_inode *nfsi = NFS_I(inode);
592
593         spin_lock(&inode->i_lock);
594         state = __nfs4_find_state_byowner(inode, owner);
595         spin_unlock(&inode->i_lock);
596         if (state)
597                 goto out;
598         new = nfs4_alloc_open_state();
599         spin_lock(&owner->so_lock);
600         spin_lock(&inode->i_lock);
601         state = __nfs4_find_state_byowner(inode, owner);
602         if (state == NULL && new != NULL) {
603                 state = new;
604                 state->owner = owner;
605                 atomic_inc(&owner->so_count);
606                 list_add(&state->inode_states, &nfsi->open_states);
607                 ihold(inode);
608                 state->inode = inode;
609                 spin_unlock(&inode->i_lock);
610                 /* Note: The reclaim code dictates that we add stateless
611                  * and read-only stateids to the end of the list */
612                 list_add_tail(&state->open_states, &owner->so_states);
613                 spin_unlock(&owner->so_lock);
614         } else {
615                 spin_unlock(&inode->i_lock);
616                 spin_unlock(&owner->so_lock);
617                 if (new)
618                         nfs4_free_open_state(new);
619         }
620 out:
621         return state;
622 }
623
624 void nfs4_put_open_state(struct nfs4_state *state)
625 {
626         struct inode *inode = state->inode;
627         struct nfs4_state_owner *owner = state->owner;
628
629         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
630                 return;
631         spin_lock(&inode->i_lock);
632         list_del(&state->inode_states);
633         list_del(&state->open_states);
634         spin_unlock(&inode->i_lock);
635         spin_unlock(&owner->so_lock);
636         iput(inode);
637         nfs4_free_open_state(state);
638         nfs4_put_state_owner(owner);
639 }
640
641 /*
642  * Close the current file.
643  */
644 static void __nfs4_close(struct nfs4_state *state,
645                 fmode_t fmode, gfp_t gfp_mask, int wait)
646 {
647         struct nfs4_state_owner *owner = state->owner;
648         int call_close = 0;
649         fmode_t newstate;
650
651         atomic_inc(&owner->so_count);
652         /* Protect against nfs4_find_state() */
653         spin_lock(&owner->so_lock);
654         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
655                 case FMODE_READ:
656                         state->n_rdonly--;
657                         break;
658                 case FMODE_WRITE:
659                         state->n_wronly--;
660                         break;
661                 case FMODE_READ|FMODE_WRITE:
662                         state->n_rdwr--;
663         }
664         newstate = FMODE_READ|FMODE_WRITE;
665         if (state->n_rdwr == 0) {
666                 if (state->n_rdonly == 0) {
667                         newstate &= ~FMODE_READ;
668                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
669                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
670                 }
671                 if (state->n_wronly == 0) {
672                         newstate &= ~FMODE_WRITE;
673                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
674                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
675                 }
676                 if (newstate == 0)
677                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
678         }
679         nfs4_state_set_mode_locked(state, newstate);
680         spin_unlock(&owner->so_lock);
681
682         if (!call_close) {
683                 nfs4_put_open_state(state);
684                 nfs4_put_state_owner(owner);
685         } else {
686                 bool roc = pnfs_roc(state->inode);
687
688                 nfs4_do_close(state, gfp_mask, wait, roc);
689         }
690 }
691
692 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
693 {
694         __nfs4_close(state, fmode, GFP_NOFS, 0);
695 }
696
697 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
698 {
699         __nfs4_close(state, fmode, GFP_KERNEL, 1);
700 }
701
702 /*
703  * Search the state->lock_states for an existing lock_owner
704  * that is compatible with current->files
705  */
706 static struct nfs4_lock_state *
707 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
708 {
709         struct nfs4_lock_state *pos;
710         list_for_each_entry(pos, &state->lock_states, ls_locks) {
711                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
712                         continue;
713                 switch (pos->ls_owner.lo_type) {
714                 case NFS4_POSIX_LOCK_TYPE:
715                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
716                                 continue;
717                         break;
718                 case NFS4_FLOCK_LOCK_TYPE:
719                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
720                                 continue;
721                 }
722                 atomic_inc(&pos->ls_count);
723                 return pos;
724         }
725         return NULL;
726 }
727
728 /*
729  * Return a compatible lock_state. If no initialized lock_state structure
730  * exists, return an uninitialized one.
731  *
732  */
733 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)
734 {
735         struct nfs4_lock_state *lsp;
736         struct nfs_server *server = state->owner->so_server;
737         struct nfs_client *clp = server->nfs_client;
738
739         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
740         if (lsp == NULL)
741                 return NULL;
742         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
743         spin_lock_init(&lsp->ls_sequence.lock);
744         INIT_LIST_HEAD(&lsp->ls_sequence.list);
745         lsp->ls_seqid.sequence = &lsp->ls_sequence;
746         atomic_set(&lsp->ls_count, 1);
747         lsp->ls_state = state;
748         lsp->ls_owner.lo_type = type;
749         switch (lsp->ls_owner.lo_type) {
750         case NFS4_FLOCK_LOCK_TYPE:
751                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
752                 break;
753         case NFS4_POSIX_LOCK_TYPE:
754                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
755                 break;
756         default:
757                 kfree(lsp);
758                 return NULL;
759         }
760         spin_lock(&clp->cl_lock);
761         nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
762         spin_unlock(&clp->cl_lock);
763         INIT_LIST_HEAD(&lsp->ls_locks);
764         return lsp;
765 }
766
767 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
768 {
769         struct nfs_server *server = lsp->ls_state->owner->so_server;
770         struct nfs_client *clp = server->nfs_client;
771
772         spin_lock(&clp->cl_lock);
773         nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
774         spin_unlock(&clp->cl_lock);
775         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
776         kfree(lsp);
777 }
778
779 /*
780  * Return a compatible lock_state. If no initialized lock_state structure
781  * exists, return an uninitialized one.
782  *
783  */
784 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
785 {
786         struct nfs4_lock_state *lsp, *new = NULL;
787         
788         for(;;) {
789                 spin_lock(&state->state_lock);
790                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
791                 if (lsp != NULL)
792                         break;
793                 if (new != NULL) {
794                         list_add(&new->ls_locks, &state->lock_states);
795                         set_bit(LK_STATE_IN_USE, &state->flags);
796                         lsp = new;
797                         new = NULL;
798                         break;
799                 }
800                 spin_unlock(&state->state_lock);
801                 new = nfs4_alloc_lock_state(state, owner, pid, type);
802                 if (new == NULL)
803                         return NULL;
804         }
805         spin_unlock(&state->state_lock);
806         if (new != NULL)
807                 nfs4_free_lock_state(new);
808         return lsp;
809 }
810
811 /*
812  * Release reference to lock_state, and free it if we see that
813  * it is no longer in use
814  */
815 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
816 {
817         struct nfs4_state *state;
818
819         if (lsp == NULL)
820                 return;
821         state = lsp->ls_state;
822         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
823                 return;
824         list_del(&lsp->ls_locks);
825         if (list_empty(&state->lock_states))
826                 clear_bit(LK_STATE_IN_USE, &state->flags);
827         spin_unlock(&state->state_lock);
828         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
829                 nfs4_release_lockowner(lsp);
830         nfs4_free_lock_state(lsp);
831 }
832
833 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
834 {
835         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
836
837         dst->fl_u.nfs4_fl.owner = lsp;
838         atomic_inc(&lsp->ls_count);
839 }
840
841 static void nfs4_fl_release_lock(struct file_lock *fl)
842 {
843         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
844 }
845
846 static const struct file_lock_operations nfs4_fl_lock_ops = {
847         .fl_copy_lock = nfs4_fl_copy_lock,
848         .fl_release_private = nfs4_fl_release_lock,
849 };
850
851 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
852 {
853         struct nfs4_lock_state *lsp;
854
855         if (fl->fl_ops != NULL)
856                 return 0;
857         if (fl->fl_flags & FL_POSIX)
858                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
859         else if (fl->fl_flags & FL_FLOCK)
860                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
861         else
862                 return -EINVAL;
863         if (lsp == NULL)
864                 return -ENOMEM;
865         fl->fl_u.nfs4_fl.owner = lsp;
866         fl->fl_ops = &nfs4_fl_lock_ops;
867         return 0;
868 }
869
870 /*
871  * Byte-range lock aware utility to initialize the stateid of read/write
872  * requests.
873  */
874 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
875 {
876         struct nfs4_lock_state *lsp;
877         int seq;
878
879         do {
880                 seq = read_seqbegin(&state->seqlock);
881                 memcpy(dst, &state->stateid, sizeof(*dst));
882         } while (read_seqretry(&state->seqlock, seq));
883         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
884                 return;
885
886         spin_lock(&state->state_lock);
887         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
888         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
889                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
890         spin_unlock(&state->state_lock);
891         nfs4_put_lock_state(lsp);
892 }
893
894 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
895 {
896         struct nfs_seqid *new;
897
898         new = kmalloc(sizeof(*new), gfp_mask);
899         if (new != NULL) {
900                 new->sequence = counter;
901                 INIT_LIST_HEAD(&new->list);
902         }
903         return new;
904 }
905
906 void nfs_release_seqid(struct nfs_seqid *seqid)
907 {
908         if (!list_empty(&seqid->list)) {
909                 struct rpc_sequence *sequence = seqid->sequence->sequence;
910
911                 spin_lock(&sequence->lock);
912                 list_del_init(&seqid->list);
913                 spin_unlock(&sequence->lock);
914                 rpc_wake_up(&sequence->wait);
915         }
916 }
917
918 void nfs_free_seqid(struct nfs_seqid *seqid)
919 {
920         nfs_release_seqid(seqid);
921         kfree(seqid);
922 }
923
924 /*
925  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
926  * failed with a seqid incrementing error -
927  * see comments nfs_fs.h:seqid_mutating_error()
928  */
929 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
930 {
931         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
932         switch (status) {
933                 case 0:
934                         break;
935                 case -NFS4ERR_BAD_SEQID:
936                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
937                                 return;
938                         printk(KERN_WARNING "NFS: v4 server returned a bad"
939                                         " sequence-id error on an"
940                                         " unconfirmed sequence %p!\n",
941                                         seqid->sequence);
942                 case -NFS4ERR_STALE_CLIENTID:
943                 case -NFS4ERR_STALE_STATEID:
944                 case -NFS4ERR_BAD_STATEID:
945                 case -NFS4ERR_BADXDR:
946                 case -NFS4ERR_RESOURCE:
947                 case -NFS4ERR_NOFILEHANDLE:
948                         /* Non-seqid mutating errors */
949                         return;
950         };
951         /*
952          * Note: no locking needed as we are guaranteed to be first
953          * on the sequence list
954          */
955         seqid->sequence->counter++;
956 }
957
958 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
959 {
960         struct nfs4_state_owner *sp = container_of(seqid->sequence,
961                                         struct nfs4_state_owner, so_seqid);
962         struct nfs_server *server = sp->so_server;
963
964         if (status == -NFS4ERR_BAD_SEQID)
965                 nfs4_drop_state_owner(sp);
966         if (!nfs4_has_session(server->nfs_client))
967                 nfs_increment_seqid(status, seqid);
968 }
969
970 /*
971  * Increment the seqid if the LOCK/LOCKU succeeded, or
972  * failed with a seqid incrementing error -
973  * see comments nfs_fs.h:seqid_mutating_error()
974  */
975 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
976 {
977         nfs_increment_seqid(status, seqid);
978 }
979
980 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
981 {
982         struct rpc_sequence *sequence = seqid->sequence->sequence;
983         int status = 0;
984
985         spin_lock(&sequence->lock);
986         if (list_empty(&seqid->list))
987                 list_add_tail(&seqid->list, &sequence->list);
988         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
989                 goto unlock;
990         rpc_sleep_on(&sequence->wait, task, NULL);
991         status = -EAGAIN;
992 unlock:
993         spin_unlock(&sequence->lock);
994         return status;
995 }
996
997 static int nfs4_run_state_manager(void *);
998
999 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1000 {
1001         smp_mb__before_clear_bit();
1002         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1003         smp_mb__after_clear_bit();
1004         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1005         rpc_wake_up(&clp->cl_rpcwaitq);
1006 }
1007
1008 /*
1009  * Schedule the nfs_client asynchronous state management routine
1010  */
1011 void nfs4_schedule_state_manager(struct nfs_client *clp)
1012 {
1013         struct task_struct *task;
1014
1015         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1016                 return;
1017         __module_get(THIS_MODULE);
1018         atomic_inc(&clp->cl_count);
1019         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1020                                 rpc_peeraddr2str(clp->cl_rpcclient,
1021                                                         RPC_DISPLAY_ADDR));
1022         if (!IS_ERR(task))
1023                 return;
1024         nfs4_clear_state_manager_bit(clp);
1025         nfs_put_client(clp);
1026         module_put(THIS_MODULE);
1027 }
1028
1029 /*
1030  * Schedule a lease recovery attempt
1031  */
1032 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1033 {
1034         if (!clp)
1035                 return;
1036         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1037                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1038         nfs4_schedule_state_manager(clp);
1039 }
1040
1041 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1042 {
1043
1044         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1045         /* Don't recover state that expired before the reboot */
1046         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1047                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1048                 return 0;
1049         }
1050         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1051         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1052         return 1;
1053 }
1054
1055 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1056 {
1057         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1058         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1059         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1060         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1061         return 1;
1062 }
1063
1064 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1065 {
1066         struct nfs_client *clp = server->nfs_client;
1067
1068         nfs4_state_mark_reclaim_nograce(clp, state);
1069         nfs4_schedule_state_manager(clp);
1070 }
1071
1072 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1073 {
1074         struct inode *inode = state->inode;
1075         struct nfs_inode *nfsi = NFS_I(inode);
1076         struct file_lock *fl;
1077         int status = 0;
1078
1079         if (inode->i_flock == NULL)
1080                 return 0;
1081
1082         /* Guard against delegation returns and new lock/unlock calls */
1083         down_write(&nfsi->rwsem);
1084         /* Protect inode->i_flock using the BKL */
1085         lock_flocks();
1086         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1087                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1088                         continue;
1089                 if (nfs_file_open_context(fl->fl_file)->state != state)
1090                         continue;
1091                 unlock_flocks();
1092                 status = ops->recover_lock(state, fl);
1093                 switch (status) {
1094                         case 0:
1095                                 break;
1096                         case -ESTALE:
1097                         case -NFS4ERR_ADMIN_REVOKED:
1098                         case -NFS4ERR_STALE_STATEID:
1099                         case -NFS4ERR_BAD_STATEID:
1100                         case -NFS4ERR_EXPIRED:
1101                         case -NFS4ERR_NO_GRACE:
1102                         case -NFS4ERR_STALE_CLIENTID:
1103                         case -NFS4ERR_BADSESSION:
1104                         case -NFS4ERR_BADSLOT:
1105                         case -NFS4ERR_BAD_HIGH_SLOT:
1106                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1107                                 goto out;
1108                         default:
1109                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1110                                                 __func__, status);
1111                         case -ENOMEM:
1112                         case -NFS4ERR_DENIED:
1113                         case -NFS4ERR_RECLAIM_BAD:
1114                         case -NFS4ERR_RECLAIM_CONFLICT:
1115                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1116                                 status = 0;
1117                 }
1118                 lock_flocks();
1119         }
1120         unlock_flocks();
1121 out:
1122         up_write(&nfsi->rwsem);
1123         return status;
1124 }
1125
1126 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1127 {
1128         struct nfs4_state *state;
1129         struct nfs4_lock_state *lock;
1130         int status = 0;
1131
1132         /* Note: we rely on the sp->so_states list being ordered 
1133          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1134          * states first.
1135          * This is needed to ensure that the server won't give us any
1136          * read delegations that we have to return if, say, we are
1137          * recovering after a network partition or a reboot from a
1138          * server that doesn't support a grace period.
1139          */
1140 restart:
1141         spin_lock(&sp->so_lock);
1142         list_for_each_entry(state, &sp->so_states, open_states) {
1143                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1144                         continue;
1145                 if (state->state == 0)
1146                         continue;
1147                 atomic_inc(&state->count);
1148                 spin_unlock(&sp->so_lock);
1149                 status = ops->recover_open(sp, state);
1150                 if (status >= 0) {
1151                         status = nfs4_reclaim_locks(state, ops);
1152                         if (status >= 0) {
1153                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1154                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1155                                                 printk("%s: Lock reclaim failed!\n",
1156                                                         __func__);
1157                                 }
1158                                 nfs4_put_open_state(state);
1159                                 goto restart;
1160                         }
1161                 }
1162                 switch (status) {
1163                         default:
1164                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1165                                                 __func__, status);
1166                         case -ENOENT:
1167                         case -ENOMEM:
1168                         case -ESTALE:
1169                                 /*
1170                                  * Open state on this file cannot be recovered
1171                                  * All we can do is revert to using the zero stateid.
1172                                  */
1173                                 memset(state->stateid.data, 0,
1174                                         sizeof(state->stateid.data));
1175                                 /* Mark the file as being 'closed' */
1176                                 state->state = 0;
1177                                 break;
1178                         case -EKEYEXPIRED:
1179                                 /*
1180                                  * User RPCSEC_GSS context has expired.
1181                                  * We cannot recover this stateid now, so
1182                                  * skip it and allow recovery thread to
1183                                  * proceed.
1184                                  */
1185                                 break;
1186                         case -NFS4ERR_ADMIN_REVOKED:
1187                         case -NFS4ERR_STALE_STATEID:
1188                         case -NFS4ERR_BAD_STATEID:
1189                         case -NFS4ERR_RECLAIM_BAD:
1190                         case -NFS4ERR_RECLAIM_CONFLICT:
1191                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1192                                 break;
1193                         case -NFS4ERR_EXPIRED:
1194                         case -NFS4ERR_NO_GRACE:
1195                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1196                         case -NFS4ERR_STALE_CLIENTID:
1197                         case -NFS4ERR_BADSESSION:
1198                         case -NFS4ERR_BADSLOT:
1199                         case -NFS4ERR_BAD_HIGH_SLOT:
1200                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1201                                 goto out_err;
1202                 }
1203                 nfs4_put_open_state(state);
1204                 goto restart;
1205         }
1206         spin_unlock(&sp->so_lock);
1207         return 0;
1208 out_err:
1209         nfs4_put_open_state(state);
1210         return status;
1211 }
1212
1213 static void nfs4_clear_open_state(struct nfs4_state *state)
1214 {
1215         struct nfs4_lock_state *lock;
1216
1217         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1218         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1219         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1220         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1221         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1222                 lock->ls_seqid.flags = 0;
1223                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1224         }
1225 }
1226
1227 static void nfs4_reset_seqids(struct nfs_server *server,
1228         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1229 {
1230         struct nfs_client *clp = server->nfs_client;
1231         struct nfs4_state_owner *sp;
1232         struct rb_node *pos;
1233         struct nfs4_state *state;
1234
1235         spin_lock(&clp->cl_lock);
1236         for (pos = rb_first(&server->state_owners);
1237              pos != NULL;
1238              pos = rb_next(pos)) {
1239                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1240                 sp->so_seqid.flags = 0;
1241                 spin_lock(&sp->so_lock);
1242                 list_for_each_entry(state, &sp->so_states, open_states) {
1243                         if (mark_reclaim(clp, state))
1244                                 nfs4_clear_open_state(state);
1245                 }
1246                 spin_unlock(&sp->so_lock);
1247         }
1248         spin_unlock(&clp->cl_lock);
1249 }
1250
1251 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1252         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1253 {
1254         struct nfs_server *server;
1255
1256         rcu_read_lock();
1257         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1258                 nfs4_reset_seqids(server, mark_reclaim);
1259         rcu_read_unlock();
1260 }
1261
1262 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1263 {
1264         /* Mark all delegations for reclaim */
1265         nfs_delegation_mark_reclaim(clp);
1266         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1267 }
1268
1269 static void nfs4_reclaim_complete(struct nfs_client *clp,
1270                                  const struct nfs4_state_recovery_ops *ops)
1271 {
1272         /* Notify the server we're done reclaiming our state */
1273         if (ops->reclaim_complete)
1274                 (void)ops->reclaim_complete(clp);
1275 }
1276
1277 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1278 {
1279         struct nfs_client *clp = server->nfs_client;
1280         struct nfs4_state_owner *sp;
1281         struct rb_node *pos;
1282         struct nfs4_state *state;
1283
1284         spin_lock(&clp->cl_lock);
1285         for (pos = rb_first(&server->state_owners);
1286              pos != NULL;
1287              pos = rb_next(pos)) {
1288                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1289                 spin_lock(&sp->so_lock);
1290                 list_for_each_entry(state, &sp->so_states, open_states) {
1291                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1292                                                 &state->flags))
1293                                 continue;
1294                         nfs4_state_mark_reclaim_nograce(clp, state);
1295                 }
1296                 spin_unlock(&sp->so_lock);
1297         }
1298         spin_unlock(&clp->cl_lock);
1299 }
1300
1301 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1302 {
1303         struct nfs_server *server;
1304
1305         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1306                 return 0;
1307
1308         rcu_read_lock();
1309         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1310                 nfs4_clear_reclaim_server(server);
1311         rcu_read_unlock();
1312
1313         nfs_delegation_reap_unclaimed(clp);
1314         return 1;
1315 }
1316
1317 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1318 {
1319         if (!nfs4_state_clear_reclaim_reboot(clp))
1320                 return;
1321         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1322 }
1323
1324 static void nfs_delegation_clear_all(struct nfs_client *clp)
1325 {
1326         nfs_delegation_mark_reclaim(clp);
1327         nfs_delegation_reap_unclaimed(clp);
1328 }
1329
1330 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1331 {
1332         nfs_delegation_clear_all(clp);
1333         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1334 }
1335
1336 static void nfs4_warn_keyexpired(const char *s)
1337 {
1338         printk_ratelimited(KERN_WARNING "Error: state manager"
1339                         " encountered RPCSEC_GSS session"
1340                         " expired against NFSv4 server %s.\n",
1341                         s);
1342 }
1343
1344 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1345 {
1346         switch (error) {
1347                 case -NFS4ERR_CB_PATH_DOWN:
1348                         nfs_handle_cb_pathdown(clp);
1349                         return 0;
1350                 case -NFS4ERR_NO_GRACE:
1351                         nfs4_state_end_reclaim_reboot(clp);
1352                         return 0;
1353                 case -NFS4ERR_STALE_CLIENTID:
1354                 case -NFS4ERR_LEASE_MOVED:
1355                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1356                         nfs4_state_clear_reclaim_reboot(clp);
1357                         nfs4_state_start_reclaim_reboot(clp);
1358                         break;
1359                 case -NFS4ERR_EXPIRED:
1360                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1361                         nfs4_state_start_reclaim_nograce(clp);
1362                         break;
1363                 case -NFS4ERR_BADSESSION:
1364                 case -NFS4ERR_BADSLOT:
1365                 case -NFS4ERR_BAD_HIGH_SLOT:
1366                 case -NFS4ERR_DEADSESSION:
1367                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1368                 case -NFS4ERR_SEQ_FALSE_RETRY:
1369                 case -NFS4ERR_SEQ_MISORDERED:
1370                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1371                         /* Zero session reset errors */
1372                         return 0;
1373                 case -EKEYEXPIRED:
1374                         /* Nothing we can do */
1375                         nfs4_warn_keyexpired(clp->cl_hostname);
1376                         return 0;
1377         }
1378         return error;
1379 }
1380
1381 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1382 {
1383         struct nfs4_state_owner *sp;
1384         struct nfs_server *server;
1385         struct rb_node *pos;
1386         int status = 0;
1387
1388 restart:
1389         rcu_read_lock();
1390         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1391                 spin_lock(&clp->cl_lock);
1392                 for (pos = rb_first(&server->state_owners);
1393                      pos != NULL;
1394                      pos = rb_next(pos)) {
1395                         sp = rb_entry(pos,
1396                                 struct nfs4_state_owner, so_server_node);
1397                         if (!test_and_clear_bit(ops->owner_flag_bit,
1398                                                         &sp->so_flags))
1399                                 continue;
1400                         atomic_inc(&sp->so_count);
1401                         spin_unlock(&clp->cl_lock);
1402                         rcu_read_unlock();
1403
1404                         status = nfs4_reclaim_open_state(sp, ops);
1405                         if (status < 0) {
1406                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1407                                 nfs4_put_state_owner(sp);
1408                                 return nfs4_recovery_handle_error(clp, status);
1409                         }
1410
1411                         nfs4_put_state_owner(sp);
1412                         goto restart;
1413                 }
1414                 spin_unlock(&clp->cl_lock);
1415         }
1416         rcu_read_unlock();
1417         return status;
1418 }
1419
1420 static int nfs4_check_lease(struct nfs_client *clp)
1421 {
1422         struct rpc_cred *cred;
1423         const struct nfs4_state_maintenance_ops *ops =
1424                 clp->cl_mvops->state_renewal_ops;
1425         int status = -NFS4ERR_EXPIRED;
1426
1427         /* Is the client already known to have an expired lease? */
1428         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1429                 return 0;
1430         spin_lock(&clp->cl_lock);
1431         cred = ops->get_state_renewal_cred_locked(clp);
1432         spin_unlock(&clp->cl_lock);
1433         if (cred == NULL) {
1434                 cred = nfs4_get_setclientid_cred(clp);
1435                 if (cred == NULL)
1436                         goto out;
1437         }
1438         status = ops->renew_lease(clp, cred);
1439         put_rpccred(cred);
1440 out:
1441         return nfs4_recovery_handle_error(clp, status);
1442 }
1443
1444 static int nfs4_reclaim_lease(struct nfs_client *clp)
1445 {
1446         struct rpc_cred *cred;
1447         const struct nfs4_state_recovery_ops *ops =
1448                 clp->cl_mvops->reboot_recovery_ops;
1449         int status = -ENOENT;
1450
1451         cred = ops->get_clid_cred(clp);
1452         if (cred != NULL) {
1453                 status = ops->establish_clid(clp, cred);
1454                 put_rpccred(cred);
1455                 /* Handle case where the user hasn't set up machine creds */
1456                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1457                         nfs4_clear_machine_cred(clp);
1458                         status = -EAGAIN;
1459                 }
1460                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1461                         status = -EPROTONOSUPPORT;
1462         }
1463         return status;
1464 }
1465
1466 #ifdef CONFIG_NFS_V4_1
1467 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1468 {
1469         struct nfs_client *clp = session->clp;
1470
1471         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1472         nfs4_schedule_lease_recovery(clp);
1473 }
1474 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1475
1476 void nfs41_handle_recall_slot(struct nfs_client *clp)
1477 {
1478         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1479         nfs4_schedule_state_manager(clp);
1480 }
1481
1482 static void nfs4_reset_all_state(struct nfs_client *clp)
1483 {
1484         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1485                 clp->cl_boot_time = CURRENT_TIME;
1486                 nfs4_state_start_reclaim_nograce(clp);
1487                 nfs4_schedule_state_manager(clp);
1488         }
1489 }
1490
1491 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1492 {
1493         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1494                 nfs4_state_start_reclaim_reboot(clp);
1495                 nfs4_schedule_state_manager(clp);
1496         }
1497 }
1498
1499 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1500 {
1501         /* Temporary */
1502         nfs4_reset_all_state(clp);
1503 }
1504
1505 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1506 {
1507         /* This will need to handle layouts too */
1508         nfs_expire_all_delegations(clp);
1509 }
1510
1511 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1512 {
1513         nfs_expire_all_delegations(clp);
1514         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1515                 nfs4_schedule_state_manager(clp);
1516 }
1517
1518 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1519 {
1520         if (!flags)
1521                 return;
1522         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1523                 nfs41_handle_server_reboot(clp);
1524         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1525                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1526                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1527                             SEQ4_STATUS_LEASE_MOVED))
1528                 nfs41_handle_state_revoked(clp);
1529         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1530                 nfs41_handle_recallable_state_revoked(clp);
1531         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1532                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1533                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1534                 nfs41_handle_cb_path_down(clp);
1535 }
1536
1537 static int nfs4_reset_session(struct nfs_client *clp)
1538 {
1539         int status;
1540
1541         nfs4_begin_drain_session(clp);
1542         status = nfs4_proc_destroy_session(clp->cl_session);
1543         if (status && status != -NFS4ERR_BADSESSION &&
1544             status != -NFS4ERR_DEADSESSION) {
1545                 status = nfs4_recovery_handle_error(clp, status);
1546                 goto out;
1547         }
1548
1549         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1550         status = nfs4_proc_create_session(clp);
1551         if (status) {
1552                 status = nfs4_recovery_handle_error(clp, status);
1553                 goto out;
1554         }
1555         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1556         /* create_session negotiated new slot table */
1557         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1558
1559          /* Let the state manager reestablish state */
1560         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1561                 nfs41_setup_state_renewal(clp);
1562 out:
1563         return status;
1564 }
1565
1566 static int nfs4_recall_slot(struct nfs_client *clp)
1567 {
1568         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1569         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1570         struct nfs4_slot *new, *old;
1571         int i;
1572
1573         nfs4_begin_drain_session(clp);
1574         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1575                       GFP_NOFS);
1576         if (!new)
1577                 return -ENOMEM;
1578
1579         spin_lock(&fc_tbl->slot_tbl_lock);
1580         for (i = 0; i < fc_tbl->target_max_slots; i++)
1581                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1582         old = fc_tbl->slots;
1583         fc_tbl->slots = new;
1584         fc_tbl->max_slots = fc_tbl->target_max_slots;
1585         fc_tbl->target_max_slots = 0;
1586         fc_attrs->max_reqs = fc_tbl->max_slots;
1587         spin_unlock(&fc_tbl->slot_tbl_lock);
1588
1589         kfree(old);
1590         nfs4_end_drain_session(clp);
1591         return 0;
1592 }
1593
1594 #else /* CONFIG_NFS_V4_1 */
1595 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1596 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1597 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1598 #endif /* CONFIG_NFS_V4_1 */
1599
1600 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1601  * on EXCHANGE_ID for v4.1
1602  */
1603 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1604 {
1605         switch (status) {
1606         case -NFS4ERR_CLID_INUSE:
1607         case -NFS4ERR_STALE_CLIENTID:
1608                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1609                 break;
1610         case -NFS4ERR_DELAY:
1611         case -ETIMEDOUT:
1612         case -EAGAIN:
1613                 ssleep(1);
1614                 break;
1615
1616         case -EKEYEXPIRED:
1617                 nfs4_warn_keyexpired(clp->cl_hostname);
1618         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1619                                  * in nfs4_exchange_id */
1620         default:
1621                 return;
1622         }
1623         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1624 }
1625
1626 static void nfs4_state_manager(struct nfs_client *clp)
1627 {
1628         int status = 0;
1629
1630         /* Ensure exclusive access to NFSv4 state */
1631         do {
1632                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1633                         /* We're going to have to re-establish a clientid */
1634                         status = nfs4_reclaim_lease(clp);
1635                         if (status) {
1636                                 nfs4_set_lease_expired(clp, status);
1637                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1638                                                         &clp->cl_state))
1639                                         continue;
1640                                 if (clp->cl_cons_state ==
1641                                                         NFS_CS_SESSION_INITING)
1642                                         nfs_mark_client_ready(clp, status);
1643                                 goto out_error;
1644                         }
1645                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1646
1647                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1648                                                &clp->cl_state))
1649                                 nfs4_state_start_reclaim_nograce(clp);
1650                         else
1651                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1652                                         &clp->cl_state);
1653
1654                         pnfs_destroy_all_layouts(clp);
1655                 }
1656
1657                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1658                         status = nfs4_check_lease(clp);
1659                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1660                                 continue;
1661                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1662                                 goto out_error;
1663                 }
1664
1665                 /* Initialize or reset the session */
1666                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1667                    && nfs4_has_session(clp)) {
1668                         status = nfs4_reset_session(clp);
1669                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1670                                 continue;
1671                         if (status < 0)
1672                                 goto out_error;
1673                 }
1674
1675                 /* First recover reboot state... */
1676                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1677                         status = nfs4_do_reclaim(clp,
1678                                 clp->cl_mvops->reboot_recovery_ops);
1679                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1680                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1681                                 continue;
1682                         nfs4_state_end_reclaim_reboot(clp);
1683                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1684                                 continue;
1685                         if (status < 0)
1686                                 goto out_error;
1687                 }
1688
1689                 /* Now recover expired state... */
1690                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1691                         status = nfs4_do_reclaim(clp,
1692                                 clp->cl_mvops->nograce_recovery_ops);
1693                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1694                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1695                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1696                                 continue;
1697                         if (status < 0)
1698                                 goto out_error;
1699                 }
1700
1701                 nfs4_end_drain_session(clp);
1702                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1703                         nfs_client_return_marked_delegations(clp);
1704                         continue;
1705                 }
1706                 /* Recall session slots */
1707                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1708                    && nfs4_has_session(clp)) {
1709                         status = nfs4_recall_slot(clp);
1710                         if (status < 0)
1711                                 goto out_error;
1712                         continue;
1713                 }
1714
1715
1716                 nfs4_clear_state_manager_bit(clp);
1717                 /* Did we race with an attempt to give us more work? */
1718                 if (clp->cl_state == 0)
1719                         break;
1720                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1721                         break;
1722         } while (atomic_read(&clp->cl_count) > 1);
1723         return;
1724 out_error:
1725         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1726                         " with error %d\n", clp->cl_hostname, -status);
1727         nfs4_end_drain_session(clp);
1728         nfs4_clear_state_manager_bit(clp);
1729 }
1730
1731 static int nfs4_run_state_manager(void *ptr)
1732 {
1733         struct nfs_client *clp = ptr;
1734
1735         allow_signal(SIGKILL);
1736         nfs4_state_manager(clp);
1737         nfs_put_client(clp);
1738         module_put_and_exit(0);
1739         return 0;
1740 }
1741
1742 /*
1743  * Local variables:
1744  *  c-basic-offset: 8
1745  * End:
1746  */