NFSv4: Fix lock recovery when CREATE_SESSION/SETCLIENTID_CONFIRM fails
[linux-3.10.git] / fs / nfs / nfs4state.c
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/ratelimit.h>
50 #include <linux/workqueue.h>
51 #include <linux/bitops.h>
52 #include <linux/jiffies.h>
53
54 #include <linux/sunrpc/clnt.h>
55
56 #include "nfs4_fs.h"
57 #include "callback.h"
58 #include "delegation.h"
59 #include "internal.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
63
64 #define NFSDBG_FACILITY         NFSDBG_STATE
65
66 #define OPENOWNER_POOL_SIZE     8
67
68 const nfs4_stateid zero_stateid;
69 static DEFINE_MUTEX(nfs_clid_init_mutex);
70
71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
72 {
73         struct nfs4_setclientid_res clid = {
74                 .clientid = clp->cl_clientid,
75                 .confirm = clp->cl_confirm,
76         };
77         unsigned short port;
78         int status;
79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
80
81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
82                 goto do_confirm;
83         port = nn->nfs_callback_tcpport;
84         if (clp->cl_addr.ss_family == AF_INET6)
85                 port = nn->nfs_callback_tcpport6;
86
87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
88         if (status != 0)
89                 goto out;
90         clp->cl_clientid = clid.clientid;
91         clp->cl_confirm = clid.confirm;
92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
93 do_confirm:
94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
95         if (status != 0)
96                 goto out;
97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
98         nfs4_schedule_state_renewal(clp);
99 out:
100         return status;
101 }
102
103 /**
104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
105  *
106  * @clp: nfs_client under test
107  * @result: OUT: found nfs_client, or clp
108  * @cred: credential to use for trunking test
109  *
110  * Returns zero, a negative errno, or a negative NFS4ERR status.
111  * If zero is returned, an nfs_client pointer is planted in
112  * "result".
113  *
114  * Note: The returned client may not yet be marked ready.
115  */
116 int nfs40_discover_server_trunking(struct nfs_client *clp,
117                                    struct nfs_client **result,
118                                    struct rpc_cred *cred)
119 {
120         struct nfs4_setclientid_res clid = {
121                 .clientid = clp->cl_clientid,
122                 .confirm = clp->cl_confirm,
123         };
124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
125         unsigned short port;
126         int status;
127
128         port = nn->nfs_callback_tcpport;
129         if (clp->cl_addr.ss_family == AF_INET6)
130                 port = nn->nfs_callback_tcpport6;
131
132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
133         if (status != 0)
134                 goto out;
135         clp->cl_clientid = clid.clientid;
136         clp->cl_confirm = clid.confirm;
137
138         status = nfs40_walk_client_list(clp, result, cred);
139         if (status == 0) {
140                 /* Sustain the lease, even if it's empty.  If the clientid4
141                  * goes stale it's of no use for trunking discovery. */
142                 nfs4_schedule_state_renewal(*result);
143         }
144 out:
145         return status;
146 }
147
148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
149 {
150         struct rpc_cred *cred = NULL;
151
152         if (clp->cl_machine_cred != NULL)
153                 cred = get_rpccred(clp->cl_machine_cred);
154         return cred;
155 }
156
157 static struct rpc_cred *
158 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
159 {
160         struct rpc_cred *cred = NULL;
161         struct nfs4_state_owner *sp;
162         struct rb_node *pos;
163
164         for (pos = rb_first(&server->state_owners);
165              pos != NULL;
166              pos = rb_next(pos)) {
167                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
168                 if (list_empty(&sp->so_states))
169                         continue;
170                 cred = get_rpccred(sp->so_cred);
171                 break;
172         }
173         return cred;
174 }
175
176 /**
177  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
178  * @clp: client state handle
179  *
180  * Returns an rpc_cred with reference count bumped, or NULL.
181  * Caller must hold clp->cl_lock.
182  */
183 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
184 {
185         struct rpc_cred *cred = NULL;
186         struct nfs_server *server;
187
188         /* Use machine credentials if available */
189         cred = nfs4_get_machine_cred_locked(clp);
190         if (cred != NULL)
191                 goto out;
192
193         rcu_read_lock();
194         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
195                 cred = nfs4_get_renew_cred_server_locked(server);
196                 if (cred != NULL)
197                         break;
198         }
199         rcu_read_unlock();
200
201 out:
202         return cred;
203 }
204
205 #if defined(CONFIG_NFS_V4_1)
206
207 static int nfs41_setup_state_renewal(struct nfs_client *clp)
208 {
209         int status;
210         struct nfs_fsinfo fsinfo;
211
212         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
213                 nfs4_schedule_state_renewal(clp);
214                 return 0;
215         }
216
217         status = nfs4_proc_get_lease_time(clp, &fsinfo);
218         if (status == 0) {
219                 /* Update lease time and schedule renewal */
220                 spin_lock(&clp->cl_lock);
221                 clp->cl_lease_time = fsinfo.lease_time * HZ;
222                 clp->cl_last_renewal = jiffies;
223                 spin_unlock(&clp->cl_lock);
224
225                 nfs4_schedule_state_renewal(clp);
226         }
227
228         return status;
229 }
230
231 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
232 {
233         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
234                 spin_lock(&tbl->slot_tbl_lock);
235                 nfs41_wake_slot_table(tbl);
236                 spin_unlock(&tbl->slot_tbl_lock);
237         }
238 }
239
240 static void nfs4_end_drain_session(struct nfs_client *clp)
241 {
242         struct nfs4_session *ses = clp->cl_session;
243
244         if (ses != NULL) {
245                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
246                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
247         }
248 }
249
250 /*
251  * Signal state manager thread if session fore channel is drained
252  */
253 void nfs4_slot_tbl_drain_complete(struct nfs4_slot_table *tbl)
254 {
255         if (nfs4_slot_tbl_draining(tbl))
256                 complete(&tbl->complete);
257 }
258
259 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
260 {
261         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
262         spin_lock(&tbl->slot_tbl_lock);
263         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
264                 INIT_COMPLETION(tbl->complete);
265                 spin_unlock(&tbl->slot_tbl_lock);
266                 return wait_for_completion_interruptible(&tbl->complete);
267         }
268         spin_unlock(&tbl->slot_tbl_lock);
269         return 0;
270 }
271
272 static int nfs4_begin_drain_session(struct nfs_client *clp)
273 {
274         struct nfs4_session *ses = clp->cl_session;
275         int ret = 0;
276
277         /* back channel */
278         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
279         if (ret)
280                 return ret;
281         /* fore channel */
282         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
283 }
284
285 static void nfs41_finish_session_reset(struct nfs_client *clp)
286 {
287         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
288         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
289         /* create_session negotiated new slot table */
290         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
291         nfs41_setup_state_renewal(clp);
292 }
293
294 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
295 {
296         int status;
297
298         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
299                 goto do_confirm;
300         nfs4_begin_drain_session(clp);
301         status = nfs4_proc_exchange_id(clp, cred);
302         if (status != 0)
303                 goto out;
304         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
305 do_confirm:
306         status = nfs4_proc_create_session(clp, cred);
307         if (status != 0)
308                 goto out;
309         nfs41_finish_session_reset(clp);
310         nfs_mark_client_ready(clp, NFS_CS_READY);
311 out:
312         return status;
313 }
314
315 /**
316  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
317  *
318  * @clp: nfs_client under test
319  * @result: OUT: found nfs_client, or clp
320  * @cred: credential to use for trunking test
321  *
322  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
323  * If NFS4_OK is returned, an nfs_client pointer is planted in
324  * "result".
325  *
326  * Note: The returned client may not yet be marked ready.
327  */
328 int nfs41_discover_server_trunking(struct nfs_client *clp,
329                                    struct nfs_client **result,
330                                    struct rpc_cred *cred)
331 {
332         int status;
333
334         status = nfs4_proc_exchange_id(clp, cred);
335         if (status != NFS4_OK)
336                 return status;
337         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
338
339         return nfs41_walk_client_list(clp, result, cred);
340 }
341
342 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
343 {
344         struct rpc_cred *cred;
345
346         spin_lock(&clp->cl_lock);
347         cred = nfs4_get_machine_cred_locked(clp);
348         spin_unlock(&clp->cl_lock);
349         return cred;
350 }
351
352 #endif /* CONFIG_NFS_V4_1 */
353
354 static struct rpc_cred *
355 nfs4_get_setclientid_cred_server(struct nfs_server *server)
356 {
357         struct nfs_client *clp = server->nfs_client;
358         struct rpc_cred *cred = NULL;
359         struct nfs4_state_owner *sp;
360         struct rb_node *pos;
361
362         spin_lock(&clp->cl_lock);
363         pos = rb_first(&server->state_owners);
364         if (pos != NULL) {
365                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
366                 cred = get_rpccred(sp->so_cred);
367         }
368         spin_unlock(&clp->cl_lock);
369         return cred;
370 }
371
372 /**
373  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
374  * @clp: client state handle
375  *
376  * Returns an rpc_cred with reference count bumped, or NULL.
377  */
378 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
379 {
380         struct nfs_server *server;
381         struct rpc_cred *cred;
382
383         spin_lock(&clp->cl_lock);
384         cred = nfs4_get_machine_cred_locked(clp);
385         spin_unlock(&clp->cl_lock);
386         if (cred != NULL)
387                 goto out;
388
389         rcu_read_lock();
390         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
391                 cred = nfs4_get_setclientid_cred_server(server);
392                 if (cred != NULL)
393                         break;
394         }
395         rcu_read_unlock();
396
397 out:
398         return cred;
399 }
400
401 static struct nfs4_state_owner *
402 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
403 {
404         struct rb_node **p = &server->state_owners.rb_node,
405                        *parent = NULL;
406         struct nfs4_state_owner *sp;
407
408         while (*p != NULL) {
409                 parent = *p;
410                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
411
412                 if (cred < sp->so_cred)
413                         p = &parent->rb_left;
414                 else if (cred > sp->so_cred)
415                         p = &parent->rb_right;
416                 else {
417                         if (!list_empty(&sp->so_lru))
418                                 list_del_init(&sp->so_lru);
419                         atomic_inc(&sp->so_count);
420                         return sp;
421                 }
422         }
423         return NULL;
424 }
425
426 static struct nfs4_state_owner *
427 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
428 {
429         struct nfs_server *server = new->so_server;
430         struct rb_node **p = &server->state_owners.rb_node,
431                        *parent = NULL;
432         struct nfs4_state_owner *sp;
433         int err;
434
435         while (*p != NULL) {
436                 parent = *p;
437                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
438
439                 if (new->so_cred < sp->so_cred)
440                         p = &parent->rb_left;
441                 else if (new->so_cred > sp->so_cred)
442                         p = &parent->rb_right;
443                 else {
444                         if (!list_empty(&sp->so_lru))
445                                 list_del_init(&sp->so_lru);
446                         atomic_inc(&sp->so_count);
447                         return sp;
448                 }
449         }
450         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
451         if (err)
452                 return ERR_PTR(err);
453         rb_link_node(&new->so_server_node, parent, p);
454         rb_insert_color(&new->so_server_node, &server->state_owners);
455         return new;
456 }
457
458 static void
459 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
460 {
461         struct nfs_server *server = sp->so_server;
462
463         if (!RB_EMPTY_NODE(&sp->so_server_node))
464                 rb_erase(&sp->so_server_node, &server->state_owners);
465         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
466 }
467
468 static void
469 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
470 {
471         sc->create_time = ktime_get();
472         sc->flags = 0;
473         sc->counter = 0;
474         spin_lock_init(&sc->lock);
475         INIT_LIST_HEAD(&sc->list);
476         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
477 }
478
479 static void
480 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
481 {
482         rpc_destroy_wait_queue(&sc->wait);
483 }
484
485 /*
486  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
487  * create a new state_owner.
488  *
489  */
490 static struct nfs4_state_owner *
491 nfs4_alloc_state_owner(struct nfs_server *server,
492                 struct rpc_cred *cred,
493                 gfp_t gfp_flags)
494 {
495         struct nfs4_state_owner *sp;
496
497         sp = kzalloc(sizeof(*sp), gfp_flags);
498         if (!sp)
499                 return NULL;
500         sp->so_server = server;
501         sp->so_cred = get_rpccred(cred);
502         spin_lock_init(&sp->so_lock);
503         INIT_LIST_HEAD(&sp->so_states);
504         nfs4_init_seqid_counter(&sp->so_seqid);
505         atomic_set(&sp->so_count, 1);
506         INIT_LIST_HEAD(&sp->so_lru);
507         seqcount_init(&sp->so_reclaim_seqcount);
508         mutex_init(&sp->so_delegreturn_mutex);
509         return sp;
510 }
511
512 static void
513 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
514 {
515         struct rb_node *rb_node = &sp->so_server_node;
516
517         if (!RB_EMPTY_NODE(rb_node)) {
518                 struct nfs_server *server = sp->so_server;
519                 struct nfs_client *clp = server->nfs_client;
520
521                 spin_lock(&clp->cl_lock);
522                 if (!RB_EMPTY_NODE(rb_node)) {
523                         rb_erase(rb_node, &server->state_owners);
524                         RB_CLEAR_NODE(rb_node);
525                 }
526                 spin_unlock(&clp->cl_lock);
527         }
528 }
529
530 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
531 {
532         nfs4_destroy_seqid_counter(&sp->so_seqid);
533         put_rpccred(sp->so_cred);
534         kfree(sp);
535 }
536
537 static void nfs4_gc_state_owners(struct nfs_server *server)
538 {
539         struct nfs_client *clp = server->nfs_client;
540         struct nfs4_state_owner *sp, *tmp;
541         unsigned long time_min, time_max;
542         LIST_HEAD(doomed);
543
544         spin_lock(&clp->cl_lock);
545         time_max = jiffies;
546         time_min = (long)time_max - (long)clp->cl_lease_time;
547         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
548                 /* NB: LRU is sorted so that oldest is at the head */
549                 if (time_in_range(sp->so_expires, time_min, time_max))
550                         break;
551                 list_move(&sp->so_lru, &doomed);
552                 nfs4_remove_state_owner_locked(sp);
553         }
554         spin_unlock(&clp->cl_lock);
555
556         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
557                 list_del(&sp->so_lru);
558                 nfs4_free_state_owner(sp);
559         }
560 }
561
562 /**
563  * nfs4_get_state_owner - Look up a state owner given a credential
564  * @server: nfs_server to search
565  * @cred: RPC credential to match
566  *
567  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
568  */
569 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
570                                               struct rpc_cred *cred,
571                                               gfp_t gfp_flags)
572 {
573         struct nfs_client *clp = server->nfs_client;
574         struct nfs4_state_owner *sp, *new;
575
576         spin_lock(&clp->cl_lock);
577         sp = nfs4_find_state_owner_locked(server, cred);
578         spin_unlock(&clp->cl_lock);
579         if (sp != NULL)
580                 goto out;
581         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
582         if (new == NULL)
583                 goto out;
584         do {
585                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
586                         break;
587                 spin_lock(&clp->cl_lock);
588                 sp = nfs4_insert_state_owner_locked(new);
589                 spin_unlock(&clp->cl_lock);
590         } while (sp == ERR_PTR(-EAGAIN));
591         if (sp != new)
592                 nfs4_free_state_owner(new);
593 out:
594         nfs4_gc_state_owners(server);
595         return sp;
596 }
597
598 /**
599  * nfs4_put_state_owner - Release a nfs4_state_owner
600  * @sp: state owner data to release
601  *
602  * Note that we keep released state owners on an LRU
603  * list.
604  * This caches valid state owners so that they can be
605  * reused, to avoid the OPEN_CONFIRM on minor version 0.
606  * It also pins the uniquifier of dropped state owners for
607  * a while, to ensure that those state owner names are
608  * never reused.
609  */
610 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
611 {
612         struct nfs_server *server = sp->so_server;
613         struct nfs_client *clp = server->nfs_client;
614
615         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
616                 return;
617
618         sp->so_expires = jiffies;
619         list_add_tail(&sp->so_lru, &server->state_owners_lru);
620         spin_unlock(&clp->cl_lock);
621 }
622
623 /**
624  * nfs4_purge_state_owners - Release all cached state owners
625  * @server: nfs_server with cached state owners to release
626  *
627  * Called at umount time.  Remaining state owners will be on
628  * the LRU with ref count of zero.
629  */
630 void nfs4_purge_state_owners(struct nfs_server *server)
631 {
632         struct nfs_client *clp = server->nfs_client;
633         struct nfs4_state_owner *sp, *tmp;
634         LIST_HEAD(doomed);
635
636         spin_lock(&clp->cl_lock);
637         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
638                 list_move(&sp->so_lru, &doomed);
639                 nfs4_remove_state_owner_locked(sp);
640         }
641         spin_unlock(&clp->cl_lock);
642
643         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
644                 list_del(&sp->so_lru);
645                 nfs4_free_state_owner(sp);
646         }
647 }
648
649 static struct nfs4_state *
650 nfs4_alloc_open_state(void)
651 {
652         struct nfs4_state *state;
653
654         state = kzalloc(sizeof(*state), GFP_NOFS);
655         if (!state)
656                 return NULL;
657         atomic_set(&state->count, 1);
658         INIT_LIST_HEAD(&state->lock_states);
659         spin_lock_init(&state->state_lock);
660         seqlock_init(&state->seqlock);
661         return state;
662 }
663
664 void
665 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
666 {
667         if (state->state == fmode)
668                 return;
669         /* NB! List reordering - see the reclaim code for why.  */
670         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
671                 if (fmode & FMODE_WRITE)
672                         list_move(&state->open_states, &state->owner->so_states);
673                 else
674                         list_move_tail(&state->open_states, &state->owner->so_states);
675         }
676         state->state = fmode;
677 }
678
679 static struct nfs4_state *
680 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
681 {
682         struct nfs_inode *nfsi = NFS_I(inode);
683         struct nfs4_state *state;
684
685         list_for_each_entry(state, &nfsi->open_states, inode_states) {
686                 if (state->owner != owner)
687                         continue;
688                 if (!nfs4_valid_open_stateid(state))
689                         continue;
690                 if (atomic_inc_not_zero(&state->count))
691                         return state;
692         }
693         return NULL;
694 }
695
696 static void
697 nfs4_free_open_state(struct nfs4_state *state)
698 {
699         kfree(state);
700 }
701
702 struct nfs4_state *
703 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
704 {
705         struct nfs4_state *state, *new;
706         struct nfs_inode *nfsi = NFS_I(inode);
707
708         spin_lock(&inode->i_lock);
709         state = __nfs4_find_state_byowner(inode, owner);
710         spin_unlock(&inode->i_lock);
711         if (state)
712                 goto out;
713         new = nfs4_alloc_open_state();
714         spin_lock(&owner->so_lock);
715         spin_lock(&inode->i_lock);
716         state = __nfs4_find_state_byowner(inode, owner);
717         if (state == NULL && new != NULL) {
718                 state = new;
719                 state->owner = owner;
720                 atomic_inc(&owner->so_count);
721                 list_add(&state->inode_states, &nfsi->open_states);
722                 ihold(inode);
723                 state->inode = inode;
724                 spin_unlock(&inode->i_lock);
725                 /* Note: The reclaim code dictates that we add stateless
726                  * and read-only stateids to the end of the list */
727                 list_add_tail(&state->open_states, &owner->so_states);
728                 spin_unlock(&owner->so_lock);
729         } else {
730                 spin_unlock(&inode->i_lock);
731                 spin_unlock(&owner->so_lock);
732                 if (new)
733                         nfs4_free_open_state(new);
734         }
735 out:
736         return state;
737 }
738
739 void nfs4_put_open_state(struct nfs4_state *state)
740 {
741         struct inode *inode = state->inode;
742         struct nfs4_state_owner *owner = state->owner;
743
744         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
745                 return;
746         spin_lock(&inode->i_lock);
747         list_del(&state->inode_states);
748         list_del(&state->open_states);
749         spin_unlock(&inode->i_lock);
750         spin_unlock(&owner->so_lock);
751         iput(inode);
752         nfs4_free_open_state(state);
753         nfs4_put_state_owner(owner);
754 }
755
756 /*
757  * Close the current file.
758  */
759 static void __nfs4_close(struct nfs4_state *state,
760                 fmode_t fmode, gfp_t gfp_mask, int wait)
761 {
762         struct nfs4_state_owner *owner = state->owner;
763         int call_close = 0;
764         fmode_t newstate;
765
766         atomic_inc(&owner->so_count);
767         /* Protect against nfs4_find_state() */
768         spin_lock(&owner->so_lock);
769         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
770                 case FMODE_READ:
771                         state->n_rdonly--;
772                         break;
773                 case FMODE_WRITE:
774                         state->n_wronly--;
775                         break;
776                 case FMODE_READ|FMODE_WRITE:
777                         state->n_rdwr--;
778         }
779         newstate = FMODE_READ|FMODE_WRITE;
780         if (state->n_rdwr == 0) {
781                 if (state->n_rdonly == 0) {
782                         newstate &= ~FMODE_READ;
783                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
784                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
785                 }
786                 if (state->n_wronly == 0) {
787                         newstate &= ~FMODE_WRITE;
788                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
789                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
790                 }
791                 if (newstate == 0)
792                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
793         }
794         nfs4_state_set_mode_locked(state, newstate);
795         spin_unlock(&owner->so_lock);
796
797         if (!call_close) {
798                 nfs4_put_open_state(state);
799                 nfs4_put_state_owner(owner);
800         } else
801                 nfs4_do_close(state, gfp_mask, wait);
802 }
803
804 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
805 {
806         __nfs4_close(state, fmode, GFP_NOFS, 0);
807 }
808
809 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
810 {
811         __nfs4_close(state, fmode, GFP_KERNEL, 1);
812 }
813
814 /*
815  * Search the state->lock_states for an existing lock_owner
816  * that is compatible with current->files
817  */
818 static struct nfs4_lock_state *
819 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
820 {
821         struct nfs4_lock_state *pos;
822         list_for_each_entry(pos, &state->lock_states, ls_locks) {
823                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
824                         continue;
825                 switch (pos->ls_owner.lo_type) {
826                 case NFS4_POSIX_LOCK_TYPE:
827                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
828                                 continue;
829                         break;
830                 case NFS4_FLOCK_LOCK_TYPE:
831                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
832                                 continue;
833                 }
834                 atomic_inc(&pos->ls_count);
835                 return pos;
836         }
837         return NULL;
838 }
839
840 /*
841  * Return a compatible lock_state. If no initialized lock_state structure
842  * exists, return an uninitialized one.
843  *
844  */
845 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
846 {
847         struct nfs4_lock_state *lsp;
848         struct nfs_server *server = state->owner->so_server;
849
850         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
851         if (lsp == NULL)
852                 return NULL;
853         nfs4_init_seqid_counter(&lsp->ls_seqid);
854         atomic_set(&lsp->ls_count, 1);
855         lsp->ls_state = state;
856         lsp->ls_owner.lo_type = type;
857         switch (lsp->ls_owner.lo_type) {
858         case NFS4_FLOCK_LOCK_TYPE:
859                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
860                 break;
861         case NFS4_POSIX_LOCK_TYPE:
862                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
863                 break;
864         default:
865                 goto out_free;
866         }
867         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
868         if (lsp->ls_seqid.owner_id < 0)
869                 goto out_free;
870         INIT_LIST_HEAD(&lsp->ls_locks);
871         return lsp;
872 out_free:
873         kfree(lsp);
874         return NULL;
875 }
876
877 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
878 {
879         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
880         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
881         kfree(lsp);
882 }
883
884 /*
885  * Return a compatible lock_state. If no initialized lock_state structure
886  * exists, return an uninitialized one.
887  *
888  */
889 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
890 {
891         struct nfs4_lock_state *lsp, *new = NULL;
892         
893         for(;;) {
894                 spin_lock(&state->state_lock);
895                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
896                 if (lsp != NULL)
897                         break;
898                 if (new != NULL) {
899                         list_add(&new->ls_locks, &state->lock_states);
900                         set_bit(LK_STATE_IN_USE, &state->flags);
901                         lsp = new;
902                         new = NULL;
903                         break;
904                 }
905                 spin_unlock(&state->state_lock);
906                 new = nfs4_alloc_lock_state(state, owner, pid, type);
907                 if (new == NULL)
908                         return NULL;
909         }
910         spin_unlock(&state->state_lock);
911         if (new != NULL)
912                 nfs4_free_lock_state(state->owner->so_server, new);
913         return lsp;
914 }
915
916 /*
917  * Release reference to lock_state, and free it if we see that
918  * it is no longer in use
919  */
920 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
921 {
922         struct nfs_server *server;
923         struct nfs4_state *state;
924
925         if (lsp == NULL)
926                 return;
927         state = lsp->ls_state;
928         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
929                 return;
930         list_del(&lsp->ls_locks);
931         if (list_empty(&state->lock_states))
932                 clear_bit(LK_STATE_IN_USE, &state->flags);
933         spin_unlock(&state->state_lock);
934         server = state->owner->so_server;
935         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
936                 struct nfs_client *clp = server->nfs_client;
937
938                 clp->cl_mvops->free_lock_state(server, lsp);
939         } else
940                 nfs4_free_lock_state(server, lsp);
941 }
942
943 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
944 {
945         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
946
947         dst->fl_u.nfs4_fl.owner = lsp;
948         atomic_inc(&lsp->ls_count);
949 }
950
951 static void nfs4_fl_release_lock(struct file_lock *fl)
952 {
953         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
954 }
955
956 static const struct file_lock_operations nfs4_fl_lock_ops = {
957         .fl_copy_lock = nfs4_fl_copy_lock,
958         .fl_release_private = nfs4_fl_release_lock,
959 };
960
961 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
962 {
963         struct nfs4_lock_state *lsp;
964
965         if (fl->fl_ops != NULL)
966                 return 0;
967         if (fl->fl_flags & FL_POSIX)
968                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
969         else if (fl->fl_flags & FL_FLOCK)
970                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
971                                 NFS4_FLOCK_LOCK_TYPE);
972         else
973                 return -EINVAL;
974         if (lsp == NULL)
975                 return -ENOMEM;
976         fl->fl_u.nfs4_fl.owner = lsp;
977         fl->fl_ops = &nfs4_fl_lock_ops;
978         return 0;
979 }
980
981 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
982                 struct nfs4_state *state,
983                 const struct nfs_lockowner *lockowner)
984 {
985         struct nfs4_lock_state *lsp;
986         fl_owner_t fl_owner;
987         pid_t fl_pid;
988         int ret = -ENOENT;
989
990
991         if (lockowner == NULL)
992                 goto out;
993
994         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
995                 goto out;
996
997         fl_owner = lockowner->l_owner;
998         fl_pid = lockowner->l_pid;
999         spin_lock(&state->state_lock);
1000         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1001         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1002                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1003                 ret = 0;
1004                 smp_rmb();
1005                 if (!list_empty(&lsp->ls_seqid.list))
1006                         ret = -EWOULDBLOCK;
1007         }
1008         spin_unlock(&state->state_lock);
1009         nfs4_put_lock_state(lsp);
1010 out:
1011         return ret;
1012 }
1013
1014 static int nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1015 {
1016         const nfs4_stateid *src;
1017         int ret;
1018         int seq;
1019
1020         do {
1021                 src = &zero_stateid;
1022                 seq = read_seqbegin(&state->seqlock);
1023                 if (test_bit(NFS_OPEN_STATE, &state->flags))
1024                         src = &state->open_stateid;
1025                 nfs4_stateid_copy(dst, src);
1026                 ret = 0;
1027                 smp_rmb();
1028                 if (!list_empty(&state->owner->so_seqid.list))
1029                         ret = -EWOULDBLOCK;
1030         } while (read_seqretry(&state->seqlock, seq));
1031         return ret;
1032 }
1033
1034 /*
1035  * Byte-range lock aware utility to initialize the stateid of read/write
1036  * requests.
1037  */
1038 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1039                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1040 {
1041         int ret = 0;
1042         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1043                 goto out;
1044         ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1045         if (ret != -ENOENT)
1046                 goto out;
1047         ret = nfs4_copy_open_stateid(dst, state);
1048 out:
1049         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1050                 dst->seqid = 0;
1051         return ret;
1052 }
1053
1054 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1055 {
1056         struct nfs_seqid *new;
1057
1058         new = kmalloc(sizeof(*new), gfp_mask);
1059         if (new != NULL) {
1060                 new->sequence = counter;
1061                 INIT_LIST_HEAD(&new->list);
1062                 new->task = NULL;
1063         }
1064         return new;
1065 }
1066
1067 void nfs_release_seqid(struct nfs_seqid *seqid)
1068 {
1069         struct nfs_seqid_counter *sequence;
1070
1071         if (list_empty(&seqid->list))
1072                 return;
1073         sequence = seqid->sequence;
1074         spin_lock(&sequence->lock);
1075         list_del_init(&seqid->list);
1076         if (!list_empty(&sequence->list)) {
1077                 struct nfs_seqid *next;
1078
1079                 next = list_first_entry(&sequence->list,
1080                                 struct nfs_seqid, list);
1081                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1082         }
1083         spin_unlock(&sequence->lock);
1084 }
1085
1086 void nfs_free_seqid(struct nfs_seqid *seqid)
1087 {
1088         nfs_release_seqid(seqid);
1089         kfree(seqid);
1090 }
1091
1092 /*
1093  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1094  * failed with a seqid incrementing error -
1095  * see comments nfs_fs.h:seqid_mutating_error()
1096  */
1097 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1098 {
1099         switch (status) {
1100                 case 0:
1101                         break;
1102                 case -NFS4ERR_BAD_SEQID:
1103                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1104                                 return;
1105                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1106                                         " sequence-id error on an"
1107                                         " unconfirmed sequence %p!\n",
1108                                         seqid->sequence);
1109                 case -NFS4ERR_STALE_CLIENTID:
1110                 case -NFS4ERR_STALE_STATEID:
1111                 case -NFS4ERR_BAD_STATEID:
1112                 case -NFS4ERR_BADXDR:
1113                 case -NFS4ERR_RESOURCE:
1114                 case -NFS4ERR_NOFILEHANDLE:
1115                         /* Non-seqid mutating errors */
1116                         return;
1117         };
1118         /*
1119          * Note: no locking needed as we are guaranteed to be first
1120          * on the sequence list
1121          */
1122         seqid->sequence->counter++;
1123 }
1124
1125 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1126 {
1127         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1128                                         struct nfs4_state_owner, so_seqid);
1129         struct nfs_server *server = sp->so_server;
1130
1131         if (status == -NFS4ERR_BAD_SEQID)
1132                 nfs4_drop_state_owner(sp);
1133         if (!nfs4_has_session(server->nfs_client))
1134                 nfs_increment_seqid(status, seqid);
1135 }
1136
1137 /*
1138  * Increment the seqid if the LOCK/LOCKU succeeded, or
1139  * failed with a seqid incrementing error -
1140  * see comments nfs_fs.h:seqid_mutating_error()
1141  */
1142 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1143 {
1144         nfs_increment_seqid(status, seqid);
1145 }
1146
1147 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1148 {
1149         struct nfs_seqid_counter *sequence = seqid->sequence;
1150         int status = 0;
1151
1152         spin_lock(&sequence->lock);
1153         seqid->task = task;
1154         if (list_empty(&seqid->list))
1155                 list_add_tail(&seqid->list, &sequence->list);
1156         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1157                 goto unlock;
1158         rpc_sleep_on(&sequence->wait, task, NULL);
1159         status = -EAGAIN;
1160 unlock:
1161         spin_unlock(&sequence->lock);
1162         return status;
1163 }
1164
1165 static int nfs4_run_state_manager(void *);
1166
1167 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1168 {
1169         smp_mb__before_clear_bit();
1170         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1171         smp_mb__after_clear_bit();
1172         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1173         rpc_wake_up(&clp->cl_rpcwaitq);
1174 }
1175
1176 /*
1177  * Schedule the nfs_client asynchronous state management routine
1178  */
1179 void nfs4_schedule_state_manager(struct nfs_client *clp)
1180 {
1181         struct task_struct *task;
1182         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1183
1184         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1185                 return;
1186         __module_get(THIS_MODULE);
1187         atomic_inc(&clp->cl_count);
1188
1189         /* The rcu_read_lock() is not strictly necessary, as the state
1190          * manager is the only thread that ever changes the rpc_xprt
1191          * after it's initialized.  At this point, we're single threaded. */
1192         rcu_read_lock();
1193         snprintf(buf, sizeof(buf), "%s-manager",
1194                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1195         rcu_read_unlock();
1196         task = kthread_run(nfs4_run_state_manager, clp, buf);
1197         if (IS_ERR(task)) {
1198                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1199                         __func__, PTR_ERR(task));
1200                 nfs4_clear_state_manager_bit(clp);
1201                 nfs_put_client(clp);
1202                 module_put(THIS_MODULE);
1203         }
1204 }
1205
1206 /*
1207  * Schedule a lease recovery attempt
1208  */
1209 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1210 {
1211         if (!clp)
1212                 return;
1213         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1214                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1215         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1216                         clp->cl_hostname);
1217         nfs4_schedule_state_manager(clp);
1218 }
1219 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1220
1221 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1222 {
1223         int res;
1224
1225         might_sleep();
1226
1227         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1228                         nfs_wait_bit_killable, TASK_KILLABLE);
1229         if (res)
1230                 return res;
1231
1232         if (clp->cl_cons_state < 0)
1233                 return clp->cl_cons_state;
1234         return 0;
1235 }
1236
1237 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1238 {
1239         unsigned int loop;
1240         int ret;
1241
1242         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1243                 ret = nfs4_wait_clnt_recover(clp);
1244                 if (ret != 0)
1245                         break;
1246                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1247                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1248                         break;
1249                 nfs4_schedule_state_manager(clp);
1250                 ret = -EIO;
1251         }
1252         return ret;
1253 }
1254
1255 /*
1256  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1257  * @clp: client to process
1258  *
1259  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1260  * resend of the SETCLIENTID and hence re-establish the
1261  * callback channel. Then return all existing delegations.
1262  */
1263 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1264 {
1265         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1266         nfs_expire_all_delegations(clp);
1267         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1268                         clp->cl_hostname);
1269 }
1270
1271 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1272 {
1273         nfs40_handle_cb_pathdown(clp);
1274         nfs4_schedule_state_manager(clp);
1275 }
1276
1277 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1278 {
1279
1280         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1281         /* Don't recover state that expired before the reboot */
1282         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1283                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1284                 return 0;
1285         }
1286         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1287         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1288         return 1;
1289 }
1290
1291 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1292 {
1293         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1294         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1295         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1296         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1297         return 1;
1298 }
1299
1300 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1301 {
1302         struct nfs_client *clp = server->nfs_client;
1303
1304         if (!nfs4_valid_open_stateid(state))
1305                 return -EBADF;
1306         nfs4_state_mark_reclaim_nograce(clp, state);
1307         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1308                         clp->cl_hostname);
1309         nfs4_schedule_state_manager(clp);
1310         return 0;
1311 }
1312 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1313
1314 void nfs_inode_find_state_and_recover(struct inode *inode,
1315                 const nfs4_stateid *stateid)
1316 {
1317         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1318         struct nfs_inode *nfsi = NFS_I(inode);
1319         struct nfs_open_context *ctx;
1320         struct nfs4_state *state;
1321         bool found = false;
1322
1323         spin_lock(&inode->i_lock);
1324         list_for_each_entry(ctx, &nfsi->open_files, list) {
1325                 state = ctx->state;
1326                 if (state == NULL)
1327                         continue;
1328                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1329                         continue;
1330                 if (!nfs4_stateid_match(&state->stateid, stateid))
1331                         continue;
1332                 nfs4_state_mark_reclaim_nograce(clp, state);
1333                 found = true;
1334         }
1335         spin_unlock(&inode->i_lock);
1336         if (found)
1337                 nfs4_schedule_state_manager(clp);
1338 }
1339
1340 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1341 {
1342         struct inode *inode = state->inode;
1343         struct nfs_inode *nfsi = NFS_I(inode);
1344         struct nfs_open_context *ctx;
1345
1346         spin_lock(&inode->i_lock);
1347         list_for_each_entry(ctx, &nfsi->open_files, list) {
1348                 if (ctx->state != state)
1349                         continue;
1350                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1351         }
1352         spin_unlock(&inode->i_lock);
1353 }
1354
1355 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1356 {
1357         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1358         nfs4_state_mark_open_context_bad(state);
1359 }
1360
1361
1362 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1363 {
1364         struct inode *inode = state->inode;
1365         struct nfs_inode *nfsi = NFS_I(inode);
1366         struct file_lock *fl;
1367         int status = 0;
1368
1369         if (inode->i_flock == NULL)
1370                 return 0;
1371
1372         /* Guard against delegation returns and new lock/unlock calls */
1373         down_write(&nfsi->rwsem);
1374         /* Protect inode->i_flock using the BKL */
1375         lock_flocks();
1376         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1377                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1378                         continue;
1379                 if (nfs_file_open_context(fl->fl_file)->state != state)
1380                         continue;
1381                 unlock_flocks();
1382                 status = ops->recover_lock(state, fl);
1383                 switch (status) {
1384                         case 0:
1385                                 break;
1386                         case -ESTALE:
1387                         case -NFS4ERR_ADMIN_REVOKED:
1388                         case -NFS4ERR_STALE_STATEID:
1389                         case -NFS4ERR_BAD_STATEID:
1390                         case -NFS4ERR_EXPIRED:
1391                         case -NFS4ERR_NO_GRACE:
1392                         case -NFS4ERR_STALE_CLIENTID:
1393                         case -NFS4ERR_BADSESSION:
1394                         case -NFS4ERR_BADSLOT:
1395                         case -NFS4ERR_BAD_HIGH_SLOT:
1396                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1397                                 goto out;
1398                         default:
1399                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1400                                         "Zeroing state\n", __func__, status);
1401                         case -ENOMEM:
1402                         case -NFS4ERR_DENIED:
1403                         case -NFS4ERR_RECLAIM_BAD:
1404                         case -NFS4ERR_RECLAIM_CONFLICT:
1405                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1406                                 status = 0;
1407                 }
1408                 lock_flocks();
1409         }
1410         unlock_flocks();
1411 out:
1412         up_write(&nfsi->rwsem);
1413         return status;
1414 }
1415
1416 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1417 {
1418         struct nfs4_state *state;
1419         struct nfs4_lock_state *lock;
1420         int status = 0;
1421
1422         /* Note: we rely on the sp->so_states list being ordered 
1423          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1424          * states first.
1425          * This is needed to ensure that the server won't give us any
1426          * read delegations that we have to return if, say, we are
1427          * recovering after a network partition or a reboot from a
1428          * server that doesn't support a grace period.
1429          */
1430         spin_lock(&sp->so_lock);
1431         write_seqcount_begin(&sp->so_reclaim_seqcount);
1432 restart:
1433         list_for_each_entry(state, &sp->so_states, open_states) {
1434                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1435                         continue;
1436                 if (!nfs4_valid_open_stateid(state))
1437                         continue;
1438                 if (state->state == 0)
1439                         continue;
1440                 atomic_inc(&state->count);
1441                 spin_unlock(&sp->so_lock);
1442                 status = ops->recover_open(sp, state);
1443                 if (status >= 0) {
1444                         status = nfs4_reclaim_locks(state, ops);
1445                         if (status >= 0) {
1446                                 spin_lock(&state->state_lock);
1447                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1448                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1449                                                 pr_warn_ratelimited("NFS: "
1450                                                         "%s: Lock reclaim "
1451                                                         "failed!\n", __func__);
1452                                 }
1453                                 spin_unlock(&state->state_lock);
1454                                 nfs4_put_open_state(state);
1455                                 spin_lock(&sp->so_lock);
1456                                 goto restart;
1457                         }
1458                 }
1459                 switch (status) {
1460                         default:
1461                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1462                                         "Zeroing state\n", __func__, status);
1463                         case -ENOENT:
1464                         case -ENOMEM:
1465                         case -ESTALE:
1466                                 /*
1467                                  * Open state on this file cannot be recovered
1468                                  * All we can do is revert to using the zero stateid.
1469                                  */
1470                                 nfs4_state_mark_recovery_failed(state, status);
1471                                 break;
1472                         case -EAGAIN:
1473                                 ssleep(1);
1474                         case -NFS4ERR_ADMIN_REVOKED:
1475                         case -NFS4ERR_STALE_STATEID:
1476                         case -NFS4ERR_BAD_STATEID:
1477                         case -NFS4ERR_RECLAIM_BAD:
1478                         case -NFS4ERR_RECLAIM_CONFLICT:
1479                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1480                                 break;
1481                         case -NFS4ERR_EXPIRED:
1482                         case -NFS4ERR_NO_GRACE:
1483                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1484                         case -NFS4ERR_STALE_CLIENTID:
1485                         case -NFS4ERR_BADSESSION:
1486                         case -NFS4ERR_BADSLOT:
1487                         case -NFS4ERR_BAD_HIGH_SLOT:
1488                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1489                                 goto out_err;
1490                 }
1491                 nfs4_put_open_state(state);
1492                 spin_lock(&sp->so_lock);
1493                 goto restart;
1494         }
1495         write_seqcount_end(&sp->so_reclaim_seqcount);
1496         spin_unlock(&sp->so_lock);
1497         return 0;
1498 out_err:
1499         nfs4_put_open_state(state);
1500         spin_lock(&sp->so_lock);
1501         write_seqcount_end(&sp->so_reclaim_seqcount);
1502         spin_unlock(&sp->so_lock);
1503         return status;
1504 }
1505
1506 static void nfs4_clear_open_state(struct nfs4_state *state)
1507 {
1508         struct nfs4_lock_state *lock;
1509
1510         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1511         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1512         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1513         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1514         spin_lock(&state->state_lock);
1515         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1516                 lock->ls_seqid.flags = 0;
1517                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1518         }
1519         spin_unlock(&state->state_lock);
1520 }
1521
1522 static void nfs4_reset_seqids(struct nfs_server *server,
1523         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1524 {
1525         struct nfs_client *clp = server->nfs_client;
1526         struct nfs4_state_owner *sp;
1527         struct rb_node *pos;
1528         struct nfs4_state *state;
1529
1530         spin_lock(&clp->cl_lock);
1531         for (pos = rb_first(&server->state_owners);
1532              pos != NULL;
1533              pos = rb_next(pos)) {
1534                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1535                 sp->so_seqid.flags = 0;
1536                 spin_lock(&sp->so_lock);
1537                 list_for_each_entry(state, &sp->so_states, open_states) {
1538                         if (mark_reclaim(clp, state))
1539                                 nfs4_clear_open_state(state);
1540                 }
1541                 spin_unlock(&sp->so_lock);
1542         }
1543         spin_unlock(&clp->cl_lock);
1544 }
1545
1546 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1547         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1548 {
1549         struct nfs_server *server;
1550
1551         rcu_read_lock();
1552         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1553                 nfs4_reset_seqids(server, mark_reclaim);
1554         rcu_read_unlock();
1555 }
1556
1557 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1558 {
1559         /* Mark all delegations for reclaim */
1560         nfs_delegation_mark_reclaim(clp);
1561         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1562 }
1563
1564 static void nfs4_reclaim_complete(struct nfs_client *clp,
1565                                  const struct nfs4_state_recovery_ops *ops)
1566 {
1567         /* Notify the server we're done reclaiming our state */
1568         if (ops->reclaim_complete)
1569                 (void)ops->reclaim_complete(clp);
1570 }
1571
1572 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1573 {
1574         struct nfs_client *clp = server->nfs_client;
1575         struct nfs4_state_owner *sp;
1576         struct rb_node *pos;
1577         struct nfs4_state *state;
1578
1579         spin_lock(&clp->cl_lock);
1580         for (pos = rb_first(&server->state_owners);
1581              pos != NULL;
1582              pos = rb_next(pos)) {
1583                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1584                 spin_lock(&sp->so_lock);
1585                 list_for_each_entry(state, &sp->so_states, open_states) {
1586                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1587                                                 &state->flags))
1588                                 continue;
1589                         nfs4_state_mark_reclaim_nograce(clp, state);
1590                 }
1591                 spin_unlock(&sp->so_lock);
1592         }
1593         spin_unlock(&clp->cl_lock);
1594 }
1595
1596 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1597 {
1598         struct nfs_server *server;
1599
1600         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1601                 return 0;
1602
1603         rcu_read_lock();
1604         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1605                 nfs4_clear_reclaim_server(server);
1606         rcu_read_unlock();
1607
1608         nfs_delegation_reap_unclaimed(clp);
1609         return 1;
1610 }
1611
1612 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1613 {
1614         if (!nfs4_state_clear_reclaim_reboot(clp))
1615                 return;
1616         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1617 }
1618
1619 static void nfs_delegation_clear_all(struct nfs_client *clp)
1620 {
1621         nfs_delegation_mark_reclaim(clp);
1622         nfs_delegation_reap_unclaimed(clp);
1623 }
1624
1625 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1626 {
1627         nfs_delegation_clear_all(clp);
1628         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1629 }
1630
1631 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1632 {
1633         switch (error) {
1634                 case 0:
1635                         break;
1636                 case -NFS4ERR_CB_PATH_DOWN:
1637                         nfs40_handle_cb_pathdown(clp);
1638                         break;
1639                 case -NFS4ERR_NO_GRACE:
1640                         nfs4_state_end_reclaim_reboot(clp);
1641                         break;
1642                 case -NFS4ERR_STALE_CLIENTID:
1643                 case -NFS4ERR_LEASE_MOVED:
1644                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1645                         nfs4_state_clear_reclaim_reboot(clp);
1646                         nfs4_state_start_reclaim_reboot(clp);
1647                         break;
1648                 case -NFS4ERR_EXPIRED:
1649                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1650                         nfs4_state_start_reclaim_nograce(clp);
1651                         break;
1652                 case -NFS4ERR_BADSESSION:
1653                 case -NFS4ERR_BADSLOT:
1654                 case -NFS4ERR_BAD_HIGH_SLOT:
1655                 case -NFS4ERR_DEADSESSION:
1656                 case -NFS4ERR_SEQ_FALSE_RETRY:
1657                 case -NFS4ERR_SEQ_MISORDERED:
1658                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1659                         /* Zero session reset errors */
1660                         break;
1661                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1662                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1663                         break;
1664                 default:
1665                         dprintk("%s: failed to handle error %d for server %s\n",
1666                                         __func__, error, clp->cl_hostname);
1667                         return error;
1668         }
1669         dprintk("%s: handled error %d for server %s\n", __func__, error,
1670                         clp->cl_hostname);
1671         return 0;
1672 }
1673
1674 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1675 {
1676         struct nfs4_state_owner *sp;
1677         struct nfs_server *server;
1678         struct rb_node *pos;
1679         int status = 0;
1680
1681 restart:
1682         rcu_read_lock();
1683         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1684                 nfs4_purge_state_owners(server);
1685                 spin_lock(&clp->cl_lock);
1686                 for (pos = rb_first(&server->state_owners);
1687                      pos != NULL;
1688                      pos = rb_next(pos)) {
1689                         sp = rb_entry(pos,
1690                                 struct nfs4_state_owner, so_server_node);
1691                         if (!test_and_clear_bit(ops->owner_flag_bit,
1692                                                         &sp->so_flags))
1693                                 continue;
1694                         atomic_inc(&sp->so_count);
1695                         spin_unlock(&clp->cl_lock);
1696                         rcu_read_unlock();
1697
1698                         status = nfs4_reclaim_open_state(sp, ops);
1699                         if (status < 0) {
1700                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1701                                 nfs4_put_state_owner(sp);
1702                                 return nfs4_recovery_handle_error(clp, status);
1703                         }
1704
1705                         nfs4_put_state_owner(sp);
1706                         goto restart;
1707                 }
1708                 spin_unlock(&clp->cl_lock);
1709         }
1710         rcu_read_unlock();
1711         return status;
1712 }
1713
1714 static int nfs4_check_lease(struct nfs_client *clp)
1715 {
1716         struct rpc_cred *cred;
1717         const struct nfs4_state_maintenance_ops *ops =
1718                 clp->cl_mvops->state_renewal_ops;
1719         int status;
1720
1721         /* Is the client already known to have an expired lease? */
1722         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1723                 return 0;
1724         spin_lock(&clp->cl_lock);
1725         cred = ops->get_state_renewal_cred_locked(clp);
1726         spin_unlock(&clp->cl_lock);
1727         if (cred == NULL) {
1728                 cred = nfs4_get_setclientid_cred(clp);
1729                 status = -ENOKEY;
1730                 if (cred == NULL)
1731                         goto out;
1732         }
1733         status = ops->renew_lease(clp, cred);
1734         put_rpccred(cred);
1735         if (status == -ETIMEDOUT) {
1736                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1737                 return 0;
1738         }
1739 out:
1740         return nfs4_recovery_handle_error(clp, status);
1741 }
1742
1743 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1744  * and for recoverable errors on EXCHANGE_ID for v4.1
1745  */
1746 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1747 {
1748         switch (status) {
1749         case -NFS4ERR_SEQ_MISORDERED:
1750                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1751                         return -ESERVERFAULT;
1752                 /* Lease confirmation error: retry after purging the lease */
1753                 ssleep(1);
1754                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1755                 break;
1756         case -NFS4ERR_STALE_CLIENTID:
1757                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1758                 nfs4_state_start_reclaim_reboot(clp);
1759                 break;
1760         case -NFS4ERR_CLID_INUSE:
1761                 pr_err("NFS: Server %s reports our clientid is in use\n",
1762                         clp->cl_hostname);
1763                 nfs_mark_client_ready(clp, -EPERM);
1764                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1765                 return -EPERM;
1766         case -EACCES:
1767         case -NFS4ERR_DELAY:
1768         case -ETIMEDOUT:
1769         case -EAGAIN:
1770                 ssleep(1);
1771                 break;
1772
1773         case -NFS4ERR_MINOR_VERS_MISMATCH:
1774                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1775                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1776                 dprintk("%s: exit with error %d for server %s\n",
1777                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1778                 return -EPROTONOSUPPORT;
1779         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1780                                  * in nfs4_exchange_id */
1781         default:
1782                 dprintk("%s: exit with error %d for server %s\n", __func__,
1783                                 status, clp->cl_hostname);
1784                 return status;
1785         }
1786         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1787         dprintk("%s: handled error %d for server %s\n", __func__, status,
1788                         clp->cl_hostname);
1789         return 0;
1790 }
1791
1792 static int nfs4_establish_lease(struct nfs_client *clp)
1793 {
1794         struct rpc_cred *cred;
1795         const struct nfs4_state_recovery_ops *ops =
1796                 clp->cl_mvops->reboot_recovery_ops;
1797         int status;
1798
1799         cred = ops->get_clid_cred(clp);
1800         if (cred == NULL)
1801                 return -ENOENT;
1802         status = ops->establish_clid(clp, cred);
1803         put_rpccred(cred);
1804         if (status != 0)
1805                 return status;
1806         pnfs_destroy_all_layouts(clp);
1807         return 0;
1808 }
1809
1810 /*
1811  * Returns zero or a negative errno.  NFS4ERR values are converted
1812  * to local errno values.
1813  */
1814 static int nfs4_reclaim_lease(struct nfs_client *clp)
1815 {
1816         int status;
1817
1818         status = nfs4_establish_lease(clp);
1819         if (status < 0)
1820                 return nfs4_handle_reclaim_lease_error(clp, status);
1821         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1822                 nfs4_state_start_reclaim_nograce(clp);
1823         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1824                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1825         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1826         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1827         return 0;
1828 }
1829
1830 static int nfs4_purge_lease(struct nfs_client *clp)
1831 {
1832         int status;
1833
1834         status = nfs4_establish_lease(clp);
1835         if (status < 0)
1836                 return nfs4_handle_reclaim_lease_error(clp, status);
1837         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1838         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1839         nfs4_state_start_reclaim_nograce(clp);
1840         return 0;
1841 }
1842
1843 /**
1844  * nfs4_discover_server_trunking - Detect server IP address trunking
1845  *
1846  * @clp: nfs_client under test
1847  * @result: OUT: found nfs_client, or clp
1848  *
1849  * Returns zero or a negative errno.  If zero is returned,
1850  * an nfs_client pointer is planted in "result".
1851  *
1852  * Note: since we are invoked in process context, and
1853  * not from inside the state manager, we cannot use
1854  * nfs4_handle_reclaim_lease_error().
1855  */
1856 int nfs4_discover_server_trunking(struct nfs_client *clp,
1857                                   struct nfs_client **result)
1858 {
1859         const struct nfs4_state_recovery_ops *ops =
1860                                 clp->cl_mvops->reboot_recovery_ops;
1861         struct rpc_clnt *clnt;
1862         struct rpc_cred *cred;
1863         int i, status;
1864
1865         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1866
1867         clnt = clp->cl_rpcclient;
1868         i = 0;
1869
1870         mutex_lock(&nfs_clid_init_mutex);
1871 again:
1872         status  = -ENOENT;
1873         cred = ops->get_clid_cred(clp);
1874         if (cred == NULL)
1875                 goto out_unlock;
1876
1877         status = ops->detect_trunking(clp, result, cred);
1878         put_rpccred(cred);
1879         switch (status) {
1880         case 0:
1881                 break;
1882         case -NFS4ERR_DELAY:
1883         case -ETIMEDOUT:
1884         case -EAGAIN:
1885                 ssleep(1);
1886         case -NFS4ERR_STALE_CLIENTID:
1887                 dprintk("NFS: %s after status %d, retrying\n",
1888                         __func__, status);
1889                 goto again;
1890         case -EACCES:
1891                 if (i++)
1892                         break;
1893         case -NFS4ERR_CLID_INUSE:
1894         case -NFS4ERR_WRONGSEC:
1895                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
1896                 if (IS_ERR(clnt)) {
1897                         status = PTR_ERR(clnt);
1898                         break;
1899                 }
1900                 /* Note: this is safe because we haven't yet marked the
1901                  * client as ready, so we are the only user of
1902                  * clp->cl_rpcclient
1903                  */
1904                 clnt = xchg(&clp->cl_rpcclient, clnt);
1905                 rpc_shutdown_client(clnt);
1906                 clnt = clp->cl_rpcclient;
1907                 goto again;
1908
1909         case -NFS4ERR_MINOR_VERS_MISMATCH:
1910                 status = -EPROTONOSUPPORT;
1911                 break;
1912
1913         case -EKEYEXPIRED:
1914         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1915                                  * in nfs4_exchange_id */
1916                 status = -EKEYEXPIRED;
1917                 break;
1918         default:
1919                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
1920                                 __func__, status);
1921                 status = -EIO;
1922         }
1923
1924 out_unlock:
1925         mutex_unlock(&nfs_clid_init_mutex);
1926         dprintk("NFS: %s: status = %d\n", __func__, status);
1927         return status;
1928 }
1929
1930 #ifdef CONFIG_NFS_V4_1
1931 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1932 {
1933         struct nfs_client *clp = session->clp;
1934
1935         switch (err) {
1936         default:
1937                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1938                 break;
1939         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1940                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1941         }
1942         nfs4_schedule_lease_recovery(clp);
1943 }
1944 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1945
1946 static void nfs41_ping_server(struct nfs_client *clp)
1947 {
1948         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
1949         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1950         nfs4_schedule_state_manager(clp);
1951 }
1952
1953 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
1954 {
1955         nfs41_ping_server(clp);
1956 }
1957
1958 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
1959 {
1960         nfs41_ping_server(clp);
1961 }
1962
1963 static void nfs4_reset_all_state(struct nfs_client *clp)
1964 {
1965         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1966                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1967                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1968                 nfs4_state_start_reclaim_nograce(clp);
1969                 dprintk("%s: scheduling reset of all state for server %s!\n",
1970                                 __func__, clp->cl_hostname);
1971                 nfs4_schedule_state_manager(clp);
1972         }
1973 }
1974
1975 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1976 {
1977         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1978                 nfs4_state_start_reclaim_reboot(clp);
1979                 dprintk("%s: server %s rebooted!\n", __func__,
1980                                 clp->cl_hostname);
1981                 nfs4_schedule_state_manager(clp);
1982         }
1983 }
1984
1985 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1986 {
1987         nfs4_reset_all_state(clp);
1988         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1989 }
1990
1991 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1992 {
1993         /* This will need to handle layouts too */
1994         nfs_expire_all_delegations(clp);
1995         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1996                         clp->cl_hostname);
1997 }
1998
1999 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2000 {
2001         nfs_expire_all_delegations(clp);
2002         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2003                 nfs4_schedule_state_manager(clp);
2004         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2005                         clp->cl_hostname);
2006 }
2007
2008 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2009 {
2010         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2011                 &clp->cl_state) == 0)
2012                 nfs4_schedule_state_manager(clp);
2013 }
2014
2015 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2016 {
2017         if (!flags)
2018                 return;
2019
2020         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2021                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2022
2023         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2024                 nfs41_handle_server_reboot(clp);
2025         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2026                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2027                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
2028                             SEQ4_STATUS_LEASE_MOVED))
2029                 nfs41_handle_state_revoked(clp);
2030         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2031                 nfs41_handle_recallable_state_revoked(clp);
2032         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2033                 nfs41_handle_backchannel_fault(clp);
2034         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2035                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2036                 nfs41_handle_cb_path_down(clp);
2037 }
2038
2039 static int nfs4_reset_session(struct nfs_client *clp)
2040 {
2041         struct rpc_cred *cred;
2042         int status;
2043
2044         if (!nfs4_has_session(clp))
2045                 return 0;
2046         nfs4_begin_drain_session(clp);
2047         cred = nfs4_get_exchange_id_cred(clp);
2048         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2049         switch (status) {
2050         case 0:
2051         case -NFS4ERR_BADSESSION:
2052         case -NFS4ERR_DEADSESSION:
2053                 break;
2054         case -NFS4ERR_BACK_CHAN_BUSY:
2055         case -NFS4ERR_DELAY:
2056                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2057                 status = 0;
2058                 ssleep(1);
2059                 goto out;
2060         default:
2061                 status = nfs4_recovery_handle_error(clp, status);
2062                 goto out;
2063         }
2064
2065         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2066         status = nfs4_proc_create_session(clp, cred);
2067         if (status) {
2068                 dprintk("%s: session reset failed with status %d for server %s!\n",
2069                         __func__, status, clp->cl_hostname);
2070                 status = nfs4_handle_reclaim_lease_error(clp, status);
2071                 goto out;
2072         }
2073         nfs41_finish_session_reset(clp);
2074         dprintk("%s: session reset was successful for server %s!\n",
2075                         __func__, clp->cl_hostname);
2076 out:
2077         if (cred)
2078                 put_rpccred(cred);
2079         return status;
2080 }
2081
2082 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2083 {
2084         struct rpc_cred *cred;
2085         int ret;
2086
2087         if (!nfs4_has_session(clp))
2088                 return 0;
2089         nfs4_begin_drain_session(clp);
2090         cred = nfs4_get_exchange_id_cred(clp);
2091         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2092         if (cred)
2093                 put_rpccred(cred);
2094         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2095         switch (ret) {
2096         case 0:
2097                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2098                         __func__, clp->cl_hostname);
2099                 break;
2100         case -NFS4ERR_DELAY:
2101                 ssleep(1);
2102                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2103                 break;
2104         default:
2105                 return nfs4_recovery_handle_error(clp, ret);
2106         }
2107         return 0;
2108 }
2109 #else /* CONFIG_NFS_V4_1 */
2110 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2111 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2112
2113 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2114 {
2115         return 0;
2116 }
2117 #endif /* CONFIG_NFS_V4_1 */
2118
2119 static void nfs4_state_manager(struct nfs_client *clp)
2120 {
2121         int status = 0;
2122         const char *section = "", *section_sep = "";
2123
2124         /* Ensure exclusive access to NFSv4 state */
2125         do {
2126                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2127                         section = "purge state";
2128                         status = nfs4_purge_lease(clp);
2129                         if (status < 0)
2130                                 goto out_error;
2131                         continue;
2132                 }
2133
2134                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2135                         section = "lease expired";
2136                         /* We're going to have to re-establish a clientid */
2137                         status = nfs4_reclaim_lease(clp);
2138                         if (status < 0)
2139                                 goto out_error;
2140                         continue;
2141                 }
2142
2143                 /* Initialize or reset the session */
2144                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2145                         section = "reset session";
2146                         status = nfs4_reset_session(clp);
2147                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2148                                 continue;
2149                         if (status < 0)
2150                                 goto out_error;
2151                 }
2152
2153                 /* Send BIND_CONN_TO_SESSION */
2154                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2155                                 &clp->cl_state)) {
2156                         section = "bind conn to session";
2157                         status = nfs4_bind_conn_to_session(clp);
2158                         if (status < 0)
2159                                 goto out_error;
2160                         continue;
2161                 }
2162
2163                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2164                         section = "check lease";
2165                         status = nfs4_check_lease(clp);
2166                         if (status < 0)
2167                                 goto out_error;
2168                         continue;
2169                 }
2170
2171                 /* First recover reboot state... */
2172                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2173                         section = "reclaim reboot";
2174                         status = nfs4_do_reclaim(clp,
2175                                 clp->cl_mvops->reboot_recovery_ops);
2176                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2177                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2178                                 continue;
2179                         nfs4_state_end_reclaim_reboot(clp);
2180                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2181                                 continue;
2182                         if (status < 0)
2183                                 goto out_error;
2184                 }
2185
2186                 /* Now recover expired state... */
2187                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2188                         section = "reclaim nograce";
2189                         status = nfs4_do_reclaim(clp,
2190                                 clp->cl_mvops->nograce_recovery_ops);
2191                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2192                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2193                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2194                                 continue;
2195                         if (status < 0)
2196                                 goto out_error;
2197                 }
2198
2199                 nfs4_end_drain_session(clp);
2200                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2201                         nfs_client_return_marked_delegations(clp);
2202                         continue;
2203                 }
2204
2205                 nfs4_clear_state_manager_bit(clp);
2206                 /* Did we race with an attempt to give us more work? */
2207                 if (clp->cl_state == 0)
2208                         break;
2209                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2210                         break;
2211         } while (atomic_read(&clp->cl_count) > 1);
2212         return;
2213 out_error:
2214         if (strlen(section))
2215                 section_sep = ": ";
2216         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2217                         " with error %d\n", section_sep, section,
2218                         clp->cl_hostname, -status);
2219         ssleep(1);
2220         nfs4_end_drain_session(clp);
2221         nfs4_clear_state_manager_bit(clp);
2222 }
2223
2224 static int nfs4_run_state_manager(void *ptr)
2225 {
2226         struct nfs_client *clp = ptr;
2227
2228         allow_signal(SIGKILL);
2229         nfs4_state_manager(clp);
2230         nfs_put_client(clp);
2231         module_put_and_exit(0);
2232         return 0;
2233 }
2234
2235 /*
2236  * Local variables:
2237  *  c-basic-offset: 8
2238  * End:
2239  */