Loading...
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/kthread.h>
46#include <linux/module.h>
47#include <linux/random.h>
48#include <linux/ratelimit.h>
49#include <linux/workqueue.h>
50#include <linux/bitops.h>
51#include <linux/jiffies.h>
52
53#include <linux/sunrpc/clnt.h>
54
55#include "nfs4_fs.h"
56#include "callback.h"
57#include "delegation.h"
58#include "internal.h"
59#include "nfs4idmap.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
68const nfs4_stateid zero_stateid = {
69 { .data = { 0 } },
70 .type = NFS4_SPECIAL_STATEID_TYPE,
71};
72const nfs4_stateid invalid_stateid = {
73 {
74 /* Funky initialiser keeps older gcc versions happy */
75 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
76 },
77 .type = NFS4_INVALID_STATEID_TYPE,
78};
79
80static DEFINE_MUTEX(nfs_clid_init_mutex);
81
82int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
83{
84 struct nfs4_setclientid_res clid = {
85 .clientid = clp->cl_clientid,
86 .confirm = clp->cl_confirm,
87 };
88 unsigned short port;
89 int status;
90 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
91
92 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
93 goto do_confirm;
94 port = nn->nfs_callback_tcpport;
95 if (clp->cl_addr.ss_family == AF_INET6)
96 port = nn->nfs_callback_tcpport6;
97
98 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
99 if (status != 0)
100 goto out;
101 clp->cl_clientid = clid.clientid;
102 clp->cl_confirm = clid.confirm;
103 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
104do_confirm:
105 status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
106 if (status != 0)
107 goto out;
108 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
109 nfs4_schedule_state_renewal(clp);
110out:
111 return status;
112}
113
114/**
115 * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
116 *
117 * @clp: nfs_client under test
118 * @result: OUT: found nfs_client, or clp
119 * @cred: credential to use for trunking test
120 *
121 * Returns zero, a negative errno, or a negative NFS4ERR status.
122 * If zero is returned, an nfs_client pointer is planted in
123 * "result".
124 *
125 * Note: The returned client may not yet be marked ready.
126 */
127int nfs40_discover_server_trunking(struct nfs_client *clp,
128 struct nfs_client **result,
129 struct rpc_cred *cred)
130{
131 struct nfs4_setclientid_res clid = {
132 .clientid = clp->cl_clientid,
133 .confirm = clp->cl_confirm,
134 };
135 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
136 unsigned short port;
137 int status;
138
139 port = nn->nfs_callback_tcpport;
140 if (clp->cl_addr.ss_family == AF_INET6)
141 port = nn->nfs_callback_tcpport6;
142
143 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
144 if (status != 0)
145 goto out;
146 clp->cl_clientid = clid.clientid;
147 clp->cl_confirm = clid.confirm;
148
149 status = nfs40_walk_client_list(clp, result, cred);
150 if (status == 0) {
151 /* Sustain the lease, even if it's empty. If the clientid4
152 * goes stale it's of no use for trunking discovery. */
153 nfs4_schedule_state_renewal(*result);
154 }
155out:
156 return status;
157}
158
159struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
160{
161 struct rpc_cred *cred = NULL;
162
163 if (clp->cl_machine_cred != NULL)
164 cred = get_rpccred(clp->cl_machine_cred);
165 return cred;
166}
167
168static void nfs4_root_machine_cred(struct nfs_client *clp)
169{
170 struct rpc_cred *cred, *new;
171
172 new = rpc_lookup_machine_cred(NULL);
173 spin_lock(&clp->cl_lock);
174 cred = clp->cl_machine_cred;
175 clp->cl_machine_cred = new;
176 spin_unlock(&clp->cl_lock);
177 if (cred != NULL)
178 put_rpccred(cred);
179}
180
181static struct rpc_cred *
182nfs4_get_renew_cred_server_locked(struct nfs_server *server)
183{
184 struct rpc_cred *cred = NULL;
185 struct nfs4_state_owner *sp;
186 struct rb_node *pos;
187
188 for (pos = rb_first(&server->state_owners);
189 pos != NULL;
190 pos = rb_next(pos)) {
191 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
192 if (list_empty(&sp->so_states))
193 continue;
194 cred = get_rpccred(sp->so_cred);
195 break;
196 }
197 return cred;
198}
199
200/**
201 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
202 * @clp: client state handle
203 *
204 * Returns an rpc_cred with reference count bumped, or NULL.
205 * Caller must hold clp->cl_lock.
206 */
207struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
208{
209 struct rpc_cred *cred = NULL;
210 struct nfs_server *server;
211
212 /* Use machine credentials if available */
213 cred = nfs4_get_machine_cred_locked(clp);
214 if (cred != NULL)
215 goto out;
216
217 rcu_read_lock();
218 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
219 cred = nfs4_get_renew_cred_server_locked(server);
220 if (cred != NULL)
221 break;
222 }
223 rcu_read_unlock();
224
225out:
226 return cred;
227}
228
229static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
230{
231 if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
232 spin_lock(&tbl->slot_tbl_lock);
233 nfs41_wake_slot_table(tbl);
234 spin_unlock(&tbl->slot_tbl_lock);
235 }
236}
237
238static void nfs4_end_drain_session(struct nfs_client *clp)
239{
240 struct nfs4_session *ses = clp->cl_session;
241
242 if (clp->cl_slot_tbl) {
243 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
244 return;
245 }
246
247 if (ses != NULL) {
248 nfs4_end_drain_slot_table(&ses->bc_slot_table);
249 nfs4_end_drain_slot_table(&ses->fc_slot_table);
250 }
251}
252
253static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
254{
255 set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
256 spin_lock(&tbl->slot_tbl_lock);
257 if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
258 reinit_completion(&tbl->complete);
259 spin_unlock(&tbl->slot_tbl_lock);
260 return wait_for_completion_interruptible(&tbl->complete);
261 }
262 spin_unlock(&tbl->slot_tbl_lock);
263 return 0;
264}
265
266static int nfs4_begin_drain_session(struct nfs_client *clp)
267{
268 struct nfs4_session *ses = clp->cl_session;
269 int ret = 0;
270
271 if (clp->cl_slot_tbl)
272 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
273
274 /* back channel */
275 ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
276 if (ret)
277 return ret;
278 /* fore channel */
279 return nfs4_drain_slot_tbl(&ses->fc_slot_table);
280}
281
282#if defined(CONFIG_NFS_V4_1)
283
284static int nfs41_setup_state_renewal(struct nfs_client *clp)
285{
286 int status;
287 struct nfs_fsinfo fsinfo;
288 unsigned long now;
289
290 if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
291 nfs4_schedule_state_renewal(clp);
292 return 0;
293 }
294
295 now = jiffies;
296 status = nfs4_proc_get_lease_time(clp, &fsinfo);
297 if (status == 0) {
298 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
299 nfs4_schedule_state_renewal(clp);
300 }
301
302 return status;
303}
304
305static void nfs41_finish_session_reset(struct nfs_client *clp)
306{
307 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
308 clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
309 /* create_session negotiated new slot table */
310 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
311 nfs41_setup_state_renewal(clp);
312}
313
314int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
315{
316 int status;
317
318 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
319 goto do_confirm;
320 status = nfs4_proc_exchange_id(clp, cred);
321 if (status != 0)
322 goto out;
323 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
324do_confirm:
325 status = nfs4_proc_create_session(clp, cred);
326 if (status != 0)
327 goto out;
328 nfs41_finish_session_reset(clp);
329 nfs_mark_client_ready(clp, NFS_CS_READY);
330out:
331 return status;
332}
333
334/**
335 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
336 *
337 * @clp: nfs_client under test
338 * @result: OUT: found nfs_client, or clp
339 * @cred: credential to use for trunking test
340 *
341 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
342 * If NFS4_OK is returned, an nfs_client pointer is planted in
343 * "result".
344 *
345 * Note: The returned client may not yet be marked ready.
346 */
347int nfs41_discover_server_trunking(struct nfs_client *clp,
348 struct nfs_client **result,
349 struct rpc_cred *cred)
350{
351 int status;
352
353 status = nfs4_proc_exchange_id(clp, cred);
354 if (status != NFS4_OK)
355 return status;
356
357 status = nfs41_walk_client_list(clp, result, cred);
358 if (status < 0)
359 return status;
360 if (clp != *result)
361 return 0;
362
363 /*
364 * Purge state if the client id was established in a prior
365 * instance and the client id could not have arrived on the
366 * server via Transparent State Migration.
367 */
368 if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
369 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
370 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
371 else
372 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
373 }
374 nfs4_schedule_state_manager(clp);
375 status = nfs_wait_client_init_complete(clp);
376 if (status < 0)
377 nfs_put_client(clp);
378 return status;
379}
380
381#endif /* CONFIG_NFS_V4_1 */
382
383/**
384 * nfs4_get_clid_cred - Acquire credential for a setclientid operation
385 * @clp: client state handle
386 *
387 * Returns an rpc_cred with reference count bumped, or NULL.
388 */
389struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
390{
391 struct rpc_cred *cred;
392
393 spin_lock(&clp->cl_lock);
394 cred = nfs4_get_machine_cred_locked(clp);
395 spin_unlock(&clp->cl_lock);
396 return cred;
397}
398
399static struct nfs4_state_owner *
400nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
401{
402 struct rb_node **p = &server->state_owners.rb_node,
403 *parent = NULL;
404 struct nfs4_state_owner *sp;
405
406 while (*p != NULL) {
407 parent = *p;
408 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
409
410 if (cred < sp->so_cred)
411 p = &parent->rb_left;
412 else if (cred > sp->so_cred)
413 p = &parent->rb_right;
414 else {
415 if (!list_empty(&sp->so_lru))
416 list_del_init(&sp->so_lru);
417 atomic_inc(&sp->so_count);
418 return sp;
419 }
420 }
421 return NULL;
422}
423
424static struct nfs4_state_owner *
425nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
426{
427 struct nfs_server *server = new->so_server;
428 struct rb_node **p = &server->state_owners.rb_node,
429 *parent = NULL;
430 struct nfs4_state_owner *sp;
431
432 while (*p != NULL) {
433 parent = *p;
434 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
435
436 if (new->so_cred < sp->so_cred)
437 p = &parent->rb_left;
438 else if (new->so_cred > sp->so_cred)
439 p = &parent->rb_right;
440 else {
441 if (!list_empty(&sp->so_lru))
442 list_del_init(&sp->so_lru);
443 atomic_inc(&sp->so_count);
444 return sp;
445 }
446 }
447 rb_link_node(&new->so_server_node, parent, p);
448 rb_insert_color(&new->so_server_node, &server->state_owners);
449 return new;
450}
451
452static void
453nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
454{
455 struct nfs_server *server = sp->so_server;
456
457 if (!RB_EMPTY_NODE(&sp->so_server_node))
458 rb_erase(&sp->so_server_node, &server->state_owners);
459}
460
461static void
462nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
463{
464 sc->create_time = ktime_get();
465 sc->flags = 0;
466 sc->counter = 0;
467 spin_lock_init(&sc->lock);
468 INIT_LIST_HEAD(&sc->list);
469 rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
470}
471
472static void
473nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
474{
475 rpc_destroy_wait_queue(&sc->wait);
476}
477
478/*
479 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
480 * create a new state_owner.
481 *
482 */
483static struct nfs4_state_owner *
484nfs4_alloc_state_owner(struct nfs_server *server,
485 struct rpc_cred *cred,
486 gfp_t gfp_flags)
487{
488 struct nfs4_state_owner *sp;
489
490 sp = kzalloc(sizeof(*sp), gfp_flags);
491 if (!sp)
492 return NULL;
493 sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
494 gfp_flags);
495 if (sp->so_seqid.owner_id < 0) {
496 kfree(sp);
497 return NULL;
498 }
499 sp->so_server = server;
500 sp->so_cred = get_rpccred(cred);
501 spin_lock_init(&sp->so_lock);
502 INIT_LIST_HEAD(&sp->so_states);
503 nfs4_init_seqid_counter(&sp->so_seqid);
504 atomic_set(&sp->so_count, 1);
505 INIT_LIST_HEAD(&sp->so_lru);
506 seqcount_init(&sp->so_reclaim_seqcount);
507 mutex_init(&sp->so_delegreturn_mutex);
508 return sp;
509}
510
511static void
512nfs4_reset_state_owner(struct nfs4_state_owner *sp)
513{
514 /* This state_owner is no longer usable, but must
515 * remain in place so that state recovery can find it
516 * and the opens associated with it.
517 * It may also be used for new 'open' request to
518 * return a delegation to the server.
519 * So update the 'create_time' so that it looks like
520 * a new state_owner. This will cause the server to
521 * request an OPEN_CONFIRM to start a new sequence.
522 */
523 sp->so_seqid.create_time = ktime_get();
524}
525
526static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
527{
528 nfs4_destroy_seqid_counter(&sp->so_seqid);
529 put_rpccred(sp->so_cred);
530 ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
531 kfree(sp);
532}
533
534static void nfs4_gc_state_owners(struct nfs_server *server)
535{
536 struct nfs_client *clp = server->nfs_client;
537 struct nfs4_state_owner *sp, *tmp;
538 unsigned long time_min, time_max;
539 LIST_HEAD(doomed);
540
541 spin_lock(&clp->cl_lock);
542 time_max = jiffies;
543 time_min = (long)time_max - (long)clp->cl_lease_time;
544 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
545 /* NB: LRU is sorted so that oldest is at the head */
546 if (time_in_range(sp->so_expires, time_min, time_max))
547 break;
548 list_move(&sp->so_lru, &doomed);
549 nfs4_remove_state_owner_locked(sp);
550 }
551 spin_unlock(&clp->cl_lock);
552
553 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
554 list_del(&sp->so_lru);
555 nfs4_free_state_owner(sp);
556 }
557}
558
559/**
560 * nfs4_get_state_owner - Look up a state owner given a credential
561 * @server: nfs_server to search
562 * @cred: RPC credential to match
563 *
564 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
565 */
566struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
567 struct rpc_cred *cred,
568 gfp_t gfp_flags)
569{
570 struct nfs_client *clp = server->nfs_client;
571 struct nfs4_state_owner *sp, *new;
572
573 spin_lock(&clp->cl_lock);
574 sp = nfs4_find_state_owner_locked(server, cred);
575 spin_unlock(&clp->cl_lock);
576 if (sp != NULL)
577 goto out;
578 new = nfs4_alloc_state_owner(server, cred, gfp_flags);
579 if (new == NULL)
580 goto out;
581 spin_lock(&clp->cl_lock);
582 sp = nfs4_insert_state_owner_locked(new);
583 spin_unlock(&clp->cl_lock);
584 if (sp != new)
585 nfs4_free_state_owner(new);
586out:
587 nfs4_gc_state_owners(server);
588 return sp;
589}
590
591/**
592 * nfs4_put_state_owner - Release a nfs4_state_owner
593 * @sp: state owner data to release
594 *
595 * Note that we keep released state owners on an LRU
596 * list.
597 * This caches valid state owners so that they can be
598 * reused, to avoid the OPEN_CONFIRM on minor version 0.
599 * It also pins the uniquifier of dropped state owners for
600 * a while, to ensure that those state owner names are
601 * never reused.
602 */
603void nfs4_put_state_owner(struct nfs4_state_owner *sp)
604{
605 struct nfs_server *server = sp->so_server;
606 struct nfs_client *clp = server->nfs_client;
607
608 if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
609 return;
610
611 sp->so_expires = jiffies;
612 list_add_tail(&sp->so_lru, &server->state_owners_lru);
613 spin_unlock(&clp->cl_lock);
614}
615
616/**
617 * nfs4_purge_state_owners - Release all cached state owners
618 * @server: nfs_server with cached state owners to release
619 *
620 * Called at umount time. Remaining state owners will be on
621 * the LRU with ref count of zero.
622 */
623void nfs4_purge_state_owners(struct nfs_server *server)
624{
625 struct nfs_client *clp = server->nfs_client;
626 struct nfs4_state_owner *sp, *tmp;
627 LIST_HEAD(doomed);
628
629 spin_lock(&clp->cl_lock);
630 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
631 list_move(&sp->so_lru, &doomed);
632 nfs4_remove_state_owner_locked(sp);
633 }
634 spin_unlock(&clp->cl_lock);
635
636 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
637 list_del(&sp->so_lru);
638 nfs4_free_state_owner(sp);
639 }
640}
641
642static struct nfs4_state *
643nfs4_alloc_open_state(void)
644{
645 struct nfs4_state *state;
646
647 state = kzalloc(sizeof(*state), GFP_NOFS);
648 if (!state)
649 return NULL;
650 atomic_set(&state->count, 1);
651 INIT_LIST_HEAD(&state->lock_states);
652 spin_lock_init(&state->state_lock);
653 seqlock_init(&state->seqlock);
654 init_waitqueue_head(&state->waitq);
655 return state;
656}
657
658void
659nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
660{
661 if (state->state == fmode)
662 return;
663 /* NB! List reordering - see the reclaim code for why. */
664 if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
665 if (fmode & FMODE_WRITE)
666 list_move(&state->open_states, &state->owner->so_states);
667 else
668 list_move_tail(&state->open_states, &state->owner->so_states);
669 }
670 state->state = fmode;
671}
672
673static struct nfs4_state *
674__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
675{
676 struct nfs_inode *nfsi = NFS_I(inode);
677 struct nfs4_state *state;
678
679 list_for_each_entry(state, &nfsi->open_states, inode_states) {
680 if (state->owner != owner)
681 continue;
682 if (!nfs4_valid_open_stateid(state))
683 continue;
684 if (atomic_inc_not_zero(&state->count))
685 return state;
686 }
687 return NULL;
688}
689
690static void
691nfs4_free_open_state(struct nfs4_state *state)
692{
693 kfree(state);
694}
695
696struct nfs4_state *
697nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
698{
699 struct nfs4_state *state, *new;
700 struct nfs_inode *nfsi = NFS_I(inode);
701
702 spin_lock(&inode->i_lock);
703 state = __nfs4_find_state_byowner(inode, owner);
704 spin_unlock(&inode->i_lock);
705 if (state)
706 goto out;
707 new = nfs4_alloc_open_state();
708 spin_lock(&owner->so_lock);
709 spin_lock(&inode->i_lock);
710 state = __nfs4_find_state_byowner(inode, owner);
711 if (state == NULL && new != NULL) {
712 state = new;
713 state->owner = owner;
714 atomic_inc(&owner->so_count);
715 list_add(&state->inode_states, &nfsi->open_states);
716 ihold(inode);
717 state->inode = inode;
718 spin_unlock(&inode->i_lock);
719 /* Note: The reclaim code dictates that we add stateless
720 * and read-only stateids to the end of the list */
721 list_add_tail(&state->open_states, &owner->so_states);
722 spin_unlock(&owner->so_lock);
723 } else {
724 spin_unlock(&inode->i_lock);
725 spin_unlock(&owner->so_lock);
726 if (new)
727 nfs4_free_open_state(new);
728 }
729out:
730 return state;
731}
732
733void nfs4_put_open_state(struct nfs4_state *state)
734{
735 struct inode *inode = state->inode;
736 struct nfs4_state_owner *owner = state->owner;
737
738 if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
739 return;
740 spin_lock(&inode->i_lock);
741 list_del(&state->inode_states);
742 list_del(&state->open_states);
743 spin_unlock(&inode->i_lock);
744 spin_unlock(&owner->so_lock);
745 iput(inode);
746 nfs4_free_open_state(state);
747 nfs4_put_state_owner(owner);
748}
749
750/*
751 * Close the current file.
752 */
753static void __nfs4_close(struct nfs4_state *state,
754 fmode_t fmode, gfp_t gfp_mask, int wait)
755{
756 struct nfs4_state_owner *owner = state->owner;
757 int call_close = 0;
758 fmode_t newstate;
759
760 atomic_inc(&owner->so_count);
761 /* Protect against nfs4_find_state() */
762 spin_lock(&owner->so_lock);
763 switch (fmode & (FMODE_READ | FMODE_WRITE)) {
764 case FMODE_READ:
765 state->n_rdonly--;
766 break;
767 case FMODE_WRITE:
768 state->n_wronly--;
769 break;
770 case FMODE_READ|FMODE_WRITE:
771 state->n_rdwr--;
772 }
773 newstate = FMODE_READ|FMODE_WRITE;
774 if (state->n_rdwr == 0) {
775 if (state->n_rdonly == 0) {
776 newstate &= ~FMODE_READ;
777 call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
778 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
779 }
780 if (state->n_wronly == 0) {
781 newstate &= ~FMODE_WRITE;
782 call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
783 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
784 }
785 if (newstate == 0)
786 clear_bit(NFS_DELEGATED_STATE, &state->flags);
787 }
788 nfs4_state_set_mode_locked(state, newstate);
789 spin_unlock(&owner->so_lock);
790
791 if (!call_close) {
792 nfs4_put_open_state(state);
793 nfs4_put_state_owner(owner);
794 } else
795 nfs4_do_close(state, gfp_mask, wait);
796}
797
798void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
799{
800 __nfs4_close(state, fmode, GFP_NOFS, 0);
801}
802
803void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
804{
805 __nfs4_close(state, fmode, GFP_KERNEL, 1);
806}
807
808/*
809 * Search the state->lock_states for an existing lock_owner
810 * that is compatible with either of the given owners.
811 * If the second is non-zero, then the first refers to a Posix-lock
812 * owner (current->files) and the second refers to a flock/OFD
813 * owner (struct file*). In that case, prefer a match for the first
814 * owner.
815 * If both sorts of locks are held on the one file we cannot know
816 * which stateid was intended to be used, so a "correct" choice cannot
817 * be made. Failing that, a "consistent" choice is preferable. The
818 * consistent choice we make is to prefer the first owner, that of a
819 * Posix lock.
820 */
821static struct nfs4_lock_state *
822__nfs4_find_lock_state(struct nfs4_state *state,
823 fl_owner_t fl_owner, fl_owner_t fl_owner2)
824{
825 struct nfs4_lock_state *pos, *ret = NULL;
826 list_for_each_entry(pos, &state->lock_states, ls_locks) {
827 if (pos->ls_owner == fl_owner) {
828 ret = pos;
829 break;
830 }
831 if (pos->ls_owner == fl_owner2)
832 ret = pos;
833 }
834 if (ret)
835 refcount_inc(&ret->ls_count);
836 return ret;
837}
838
839/*
840 * Return a compatible lock_state. If no initialized lock_state structure
841 * exists, return an uninitialized one.
842 *
843 */
844static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
845{
846 struct nfs4_lock_state *lsp;
847 struct nfs_server *server = state->owner->so_server;
848
849 lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
850 if (lsp == NULL)
851 return NULL;
852 nfs4_init_seqid_counter(&lsp->ls_seqid);
853 refcount_set(&lsp->ls_count, 1);
854 lsp->ls_state = state;
855 lsp->ls_owner = fl_owner;
856 lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
857 if (lsp->ls_seqid.owner_id < 0)
858 goto out_free;
859 INIT_LIST_HEAD(&lsp->ls_locks);
860 return lsp;
861out_free:
862 kfree(lsp);
863 return NULL;
864}
865
866void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
867{
868 ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
869 nfs4_destroy_seqid_counter(&lsp->ls_seqid);
870 kfree(lsp);
871}
872
873/*
874 * Return a compatible lock_state. If no initialized lock_state structure
875 * exists, return an uninitialized one.
876 *
877 */
878static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
879{
880 struct nfs4_lock_state *lsp, *new = NULL;
881
882 for(;;) {
883 spin_lock(&state->state_lock);
884 lsp = __nfs4_find_lock_state(state, owner, NULL);
885 if (lsp != NULL)
886 break;
887 if (new != NULL) {
888 list_add(&new->ls_locks, &state->lock_states);
889 set_bit(LK_STATE_IN_USE, &state->flags);
890 lsp = new;
891 new = NULL;
892 break;
893 }
894 spin_unlock(&state->state_lock);
895 new = nfs4_alloc_lock_state(state, owner);
896 if (new == NULL)
897 return NULL;
898 }
899 spin_unlock(&state->state_lock);
900 if (new != NULL)
901 nfs4_free_lock_state(state->owner->so_server, new);
902 return lsp;
903}
904
905/*
906 * Release reference to lock_state, and free it if we see that
907 * it is no longer in use
908 */
909void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
910{
911 struct nfs_server *server;
912 struct nfs4_state *state;
913
914 if (lsp == NULL)
915 return;
916 state = lsp->ls_state;
917 if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
918 return;
919 list_del(&lsp->ls_locks);
920 if (list_empty(&state->lock_states))
921 clear_bit(LK_STATE_IN_USE, &state->flags);
922 spin_unlock(&state->state_lock);
923 server = state->owner->so_server;
924 if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
925 struct nfs_client *clp = server->nfs_client;
926
927 clp->cl_mvops->free_lock_state(server, lsp);
928 } else
929 nfs4_free_lock_state(server, lsp);
930}
931
932static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
933{
934 struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
935
936 dst->fl_u.nfs4_fl.owner = lsp;
937 refcount_inc(&lsp->ls_count);
938}
939
940static void nfs4_fl_release_lock(struct file_lock *fl)
941{
942 nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
943}
944
945static const struct file_lock_operations nfs4_fl_lock_ops = {
946 .fl_copy_lock = nfs4_fl_copy_lock,
947 .fl_release_private = nfs4_fl_release_lock,
948};
949
950int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
951{
952 struct nfs4_lock_state *lsp;
953
954 if (fl->fl_ops != NULL)
955 return 0;
956 lsp = nfs4_get_lock_state(state, fl->fl_owner);
957 if (lsp == NULL)
958 return -ENOMEM;
959 fl->fl_u.nfs4_fl.owner = lsp;
960 fl->fl_ops = &nfs4_fl_lock_ops;
961 return 0;
962}
963
964static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
965 struct nfs4_state *state,
966 const struct nfs_lock_context *l_ctx)
967{
968 struct nfs4_lock_state *lsp;
969 fl_owner_t fl_owner, fl_flock_owner;
970 int ret = -ENOENT;
971
972 if (l_ctx == NULL)
973 goto out;
974
975 if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
976 goto out;
977
978 fl_owner = l_ctx->lockowner;
979 fl_flock_owner = l_ctx->open_context->flock_owner;
980
981 spin_lock(&state->state_lock);
982 lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
983 if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
984 ret = -EIO;
985 else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
986 nfs4_stateid_copy(dst, &lsp->ls_stateid);
987 ret = 0;
988 }
989 spin_unlock(&state->state_lock);
990 nfs4_put_lock_state(lsp);
991out:
992 return ret;
993}
994
995bool nfs4_refresh_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
996{
997 bool ret;
998 int seq;
999
1000 do {
1001 ret = false;
1002 seq = read_seqbegin(&state->seqlock);
1003 if (nfs4_state_match_open_stateid_other(state, dst)) {
1004 dst->seqid = state->open_stateid.seqid;
1005 ret = true;
1006 }
1007 } while (read_seqretry(&state->seqlock, seq));
1008 return ret;
1009}
1010
1011bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1012{
1013 bool ret;
1014 const nfs4_stateid *src;
1015 int seq;
1016
1017 do {
1018 ret = false;
1019 src = &zero_stateid;
1020 seq = read_seqbegin(&state->seqlock);
1021 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1022 src = &state->open_stateid;
1023 ret = true;
1024 }
1025 nfs4_stateid_copy(dst, src);
1026 } while (read_seqretry(&state->seqlock, seq));
1027 return ret;
1028}
1029
1030/*
1031 * Byte-range lock aware utility to initialize the stateid of read/write
1032 * requests.
1033 */
1034int nfs4_select_rw_stateid(struct nfs4_state *state,
1035 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1036 nfs4_stateid *dst, struct rpc_cred **cred)
1037{
1038 int ret;
1039
1040 if (!nfs4_valid_open_stateid(state))
1041 return -EIO;
1042 if (cred != NULL)
1043 *cred = NULL;
1044 ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1045 if (ret == -EIO)
1046 /* A lost lock - don't even consider delegations */
1047 goto out;
1048 /* returns true if delegation stateid found and copied */
1049 if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1050 ret = 0;
1051 goto out;
1052 }
1053 if (ret != -ENOENT)
1054 /* nfs4_copy_delegation_stateid() didn't over-write
1055 * dst, so it still has the lock stateid which we now
1056 * choose to use.
1057 */
1058 goto out;
1059 nfs4_copy_open_stateid(dst, state);
1060 ret = 0;
1061out:
1062 if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1063 dst->seqid = 0;
1064 return ret;
1065}
1066
1067struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1068{
1069 struct nfs_seqid *new;
1070
1071 new = kmalloc(sizeof(*new), gfp_mask);
1072 if (new == NULL)
1073 return ERR_PTR(-ENOMEM);
1074 new->sequence = counter;
1075 INIT_LIST_HEAD(&new->list);
1076 new->task = NULL;
1077 return new;
1078}
1079
1080void nfs_release_seqid(struct nfs_seqid *seqid)
1081{
1082 struct nfs_seqid_counter *sequence;
1083
1084 if (seqid == NULL || list_empty(&seqid->list))
1085 return;
1086 sequence = seqid->sequence;
1087 spin_lock(&sequence->lock);
1088 list_del_init(&seqid->list);
1089 if (!list_empty(&sequence->list)) {
1090 struct nfs_seqid *next;
1091
1092 next = list_first_entry(&sequence->list,
1093 struct nfs_seqid, list);
1094 rpc_wake_up_queued_task(&sequence->wait, next->task);
1095 }
1096 spin_unlock(&sequence->lock);
1097}
1098
1099void nfs_free_seqid(struct nfs_seqid *seqid)
1100{
1101 nfs_release_seqid(seqid);
1102 kfree(seqid);
1103}
1104
1105/*
1106 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1107 * failed with a seqid incrementing error -
1108 * see comments nfs4.h:seqid_mutating_error()
1109 */
1110static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1111{
1112 switch (status) {
1113 case 0:
1114 break;
1115 case -NFS4ERR_BAD_SEQID:
1116 if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1117 return;
1118 pr_warn_ratelimited("NFS: v4 server returned a bad"
1119 " sequence-id error on an"
1120 " unconfirmed sequence %p!\n",
1121 seqid->sequence);
1122 case -NFS4ERR_STALE_CLIENTID:
1123 case -NFS4ERR_STALE_STATEID:
1124 case -NFS4ERR_BAD_STATEID:
1125 case -NFS4ERR_BADXDR:
1126 case -NFS4ERR_RESOURCE:
1127 case -NFS4ERR_NOFILEHANDLE:
1128 case -NFS4ERR_MOVED:
1129 /* Non-seqid mutating errors */
1130 return;
1131 };
1132 /*
1133 * Note: no locking needed as we are guaranteed to be first
1134 * on the sequence list
1135 */
1136 seqid->sequence->counter++;
1137}
1138
1139void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1140{
1141 struct nfs4_state_owner *sp;
1142
1143 if (seqid == NULL)
1144 return;
1145
1146 sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1147 if (status == -NFS4ERR_BAD_SEQID)
1148 nfs4_reset_state_owner(sp);
1149 if (!nfs4_has_session(sp->so_server->nfs_client))
1150 nfs_increment_seqid(status, seqid);
1151}
1152
1153/*
1154 * Increment the seqid if the LOCK/LOCKU succeeded, or
1155 * failed with a seqid incrementing error -
1156 * see comments nfs4.h:seqid_mutating_error()
1157 */
1158void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1159{
1160 if (seqid != NULL)
1161 nfs_increment_seqid(status, seqid);
1162}
1163
1164int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1165{
1166 struct nfs_seqid_counter *sequence;
1167 int status = 0;
1168
1169 if (seqid == NULL)
1170 goto out;
1171 sequence = seqid->sequence;
1172 spin_lock(&sequence->lock);
1173 seqid->task = task;
1174 if (list_empty(&seqid->list))
1175 list_add_tail(&seqid->list, &sequence->list);
1176 if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1177 goto unlock;
1178 rpc_sleep_on(&sequence->wait, task, NULL);
1179 status = -EAGAIN;
1180unlock:
1181 spin_unlock(&sequence->lock);
1182out:
1183 return status;
1184}
1185
1186static int nfs4_run_state_manager(void *);
1187
1188static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1189{
1190 smp_mb__before_atomic();
1191 clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1192 smp_mb__after_atomic();
1193 wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1194 rpc_wake_up(&clp->cl_rpcwaitq);
1195}
1196
1197/*
1198 * Schedule the nfs_client asynchronous state management routine
1199 */
1200void nfs4_schedule_state_manager(struct nfs_client *clp)
1201{
1202 struct task_struct *task;
1203 char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1204
1205 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1206 return;
1207 __module_get(THIS_MODULE);
1208 refcount_inc(&clp->cl_count);
1209
1210 /* The rcu_read_lock() is not strictly necessary, as the state
1211 * manager is the only thread that ever changes the rpc_xprt
1212 * after it's initialized. At this point, we're single threaded. */
1213 rcu_read_lock();
1214 snprintf(buf, sizeof(buf), "%s-manager",
1215 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1216 rcu_read_unlock();
1217 task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1218 if (IS_ERR(task)) {
1219 printk(KERN_ERR "%s: kthread_run: %ld\n",
1220 __func__, PTR_ERR(task));
1221 nfs4_clear_state_manager_bit(clp);
1222 nfs_put_client(clp);
1223 module_put(THIS_MODULE);
1224 }
1225}
1226
1227/*
1228 * Schedule a lease recovery attempt
1229 */
1230void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1231{
1232 if (!clp)
1233 return;
1234 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1235 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1236 dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1237 clp->cl_hostname);
1238 nfs4_schedule_state_manager(clp);
1239}
1240EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1241
1242/**
1243 * nfs4_schedule_migration_recovery - trigger migration recovery
1244 *
1245 * @server: FSID that is migrating
1246 *
1247 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1248 * value is returned.
1249 */
1250int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1251{
1252 struct nfs_client *clp = server->nfs_client;
1253
1254 if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1255 pr_err("NFS: volatile file handles not supported (server %s)\n",
1256 clp->cl_hostname);
1257 return -NFS4ERR_IO;
1258 }
1259
1260 if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1261 return -NFS4ERR_IO;
1262
1263 dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1264 __func__,
1265 (unsigned long long)server->fsid.major,
1266 (unsigned long long)server->fsid.minor,
1267 clp->cl_hostname);
1268
1269 set_bit(NFS_MIG_IN_TRANSITION,
1270 &((struct nfs_server *)server)->mig_status);
1271 set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1272
1273 nfs4_schedule_state_manager(clp);
1274 return 0;
1275}
1276EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1277
1278/**
1279 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1280 *
1281 * @clp: server to check for moved leases
1282 *
1283 */
1284void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1285{
1286 dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1287 __func__, clp->cl_clientid, clp->cl_hostname);
1288
1289 set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1290 nfs4_schedule_state_manager(clp);
1291}
1292EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1293
1294int nfs4_wait_clnt_recover(struct nfs_client *clp)
1295{
1296 int res;
1297
1298 might_sleep();
1299
1300 refcount_inc(&clp->cl_count);
1301 res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1302 nfs_wait_bit_killable, TASK_KILLABLE);
1303 if (res)
1304 goto out;
1305 if (clp->cl_cons_state < 0)
1306 res = clp->cl_cons_state;
1307out:
1308 nfs_put_client(clp);
1309 return res;
1310}
1311
1312int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1313{
1314 unsigned int loop;
1315 int ret;
1316
1317 for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1318 ret = nfs4_wait_clnt_recover(clp);
1319 if (ret != 0)
1320 break;
1321 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1322 !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1323 break;
1324 nfs4_schedule_state_manager(clp);
1325 ret = -EIO;
1326 }
1327 return ret;
1328}
1329
1330/*
1331 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1332 * @clp: client to process
1333 *
1334 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1335 * resend of the SETCLIENTID and hence re-establish the
1336 * callback channel. Then return all existing delegations.
1337 */
1338static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1339{
1340 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1341 nfs_expire_all_delegations(clp);
1342 dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1343 clp->cl_hostname);
1344}
1345
1346void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1347{
1348 nfs40_handle_cb_pathdown(clp);
1349 nfs4_schedule_state_manager(clp);
1350}
1351
1352static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1353{
1354
1355 if (!nfs4_valid_open_stateid(state))
1356 return 0;
1357 set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1358 /* Don't recover state that expired before the reboot */
1359 if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1360 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1361 return 0;
1362 }
1363 set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1364 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1365 return 1;
1366}
1367
1368int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1369{
1370 if (!nfs4_valid_open_stateid(state))
1371 return 0;
1372 set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1373 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1374 set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1375 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1376 return 1;
1377}
1378
1379int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1380{
1381 struct nfs_client *clp = server->nfs_client;
1382
1383 if (!nfs4_state_mark_reclaim_nograce(clp, state))
1384 return -EBADF;
1385 dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1386 clp->cl_hostname);
1387 nfs4_schedule_state_manager(clp);
1388 return 0;
1389}
1390EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1391
1392static struct nfs4_lock_state *
1393nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1394 const nfs4_stateid *stateid)
1395{
1396 struct nfs4_lock_state *pos;
1397
1398 list_for_each_entry(pos, &state->lock_states, ls_locks) {
1399 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1400 continue;
1401 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1402 return pos;
1403 }
1404 return NULL;
1405}
1406
1407static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1408 const nfs4_stateid *stateid)
1409{
1410 bool found = false;
1411
1412 if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1413 spin_lock(&state->state_lock);
1414 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1415 found = true;
1416 spin_unlock(&state->state_lock);
1417 }
1418 return found;
1419}
1420
1421void nfs_inode_find_state_and_recover(struct inode *inode,
1422 const nfs4_stateid *stateid)
1423{
1424 struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1425 struct nfs_inode *nfsi = NFS_I(inode);
1426 struct nfs_open_context *ctx;
1427 struct nfs4_state *state;
1428 bool found = false;
1429
1430 spin_lock(&inode->i_lock);
1431 list_for_each_entry(ctx, &nfsi->open_files, list) {
1432 state = ctx->state;
1433 if (state == NULL)
1434 continue;
1435 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1436 nfs4_state_mark_reclaim_nograce(clp, state)) {
1437 found = true;
1438 continue;
1439 }
1440 if (nfs4_stateid_match_other(&state->open_stateid, stateid) &&
1441 nfs4_state_mark_reclaim_nograce(clp, state)) {
1442 found = true;
1443 continue;
1444 }
1445 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1446 nfs4_state_mark_reclaim_nograce(clp, state))
1447 found = true;
1448 }
1449 spin_unlock(&inode->i_lock);
1450
1451 nfs_inode_find_delegation_state_and_recover(inode, stateid);
1452 if (found)
1453 nfs4_schedule_state_manager(clp);
1454}
1455
1456static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1457{
1458 struct inode *inode = state->inode;
1459 struct nfs_inode *nfsi = NFS_I(inode);
1460 struct nfs_open_context *ctx;
1461
1462 spin_lock(&inode->i_lock);
1463 list_for_each_entry(ctx, &nfsi->open_files, list) {
1464 if (ctx->state != state)
1465 continue;
1466 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1467 }
1468 spin_unlock(&inode->i_lock);
1469}
1470
1471static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1472{
1473 set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1474 nfs4_state_mark_open_context_bad(state);
1475}
1476
1477
1478static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1479{
1480 struct inode *inode = state->inode;
1481 struct nfs_inode *nfsi = NFS_I(inode);
1482 struct file_lock *fl;
1483 struct nfs4_lock_state *lsp;
1484 int status = 0;
1485 struct file_lock_context *flctx = inode->i_flctx;
1486 struct list_head *list;
1487
1488 if (flctx == NULL)
1489 return 0;
1490
1491 list = &flctx->flc_posix;
1492
1493 /* Guard against delegation returns and new lock/unlock calls */
1494 down_write(&nfsi->rwsem);
1495 spin_lock(&flctx->flc_lock);
1496restart:
1497 list_for_each_entry(fl, list, fl_list) {
1498 if (nfs_file_open_context(fl->fl_file)->state != state)
1499 continue;
1500 spin_unlock(&flctx->flc_lock);
1501 status = ops->recover_lock(state, fl);
1502 switch (status) {
1503 case 0:
1504 break;
1505 case -ESTALE:
1506 case -NFS4ERR_ADMIN_REVOKED:
1507 case -NFS4ERR_STALE_STATEID:
1508 case -NFS4ERR_BAD_STATEID:
1509 case -NFS4ERR_EXPIRED:
1510 case -NFS4ERR_NO_GRACE:
1511 case -NFS4ERR_STALE_CLIENTID:
1512 case -NFS4ERR_BADSESSION:
1513 case -NFS4ERR_BADSLOT:
1514 case -NFS4ERR_BAD_HIGH_SLOT:
1515 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1516 goto out;
1517 default:
1518 pr_err("NFS: %s: unhandled error %d\n",
1519 __func__, status);
1520 case -ENOMEM:
1521 case -NFS4ERR_DENIED:
1522 case -NFS4ERR_RECLAIM_BAD:
1523 case -NFS4ERR_RECLAIM_CONFLICT:
1524 lsp = fl->fl_u.nfs4_fl.owner;
1525 if (lsp)
1526 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1527 status = 0;
1528 }
1529 spin_lock(&flctx->flc_lock);
1530 }
1531 if (list == &flctx->flc_posix) {
1532 list = &flctx->flc_flock;
1533 goto restart;
1534 }
1535 spin_unlock(&flctx->flc_lock);
1536out:
1537 up_write(&nfsi->rwsem);
1538 return status;
1539}
1540
1541static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1542{
1543 struct nfs4_state *state;
1544 struct nfs4_lock_state *lock;
1545 int status = 0;
1546
1547 /* Note: we rely on the sp->so_states list being ordered
1548 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1549 * states first.
1550 * This is needed to ensure that the server won't give us any
1551 * read delegations that we have to return if, say, we are
1552 * recovering after a network partition or a reboot from a
1553 * server that doesn't support a grace period.
1554 */
1555 spin_lock(&sp->so_lock);
1556 raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1557restart:
1558 list_for_each_entry(state, &sp->so_states, open_states) {
1559 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1560 continue;
1561 if (!nfs4_valid_open_stateid(state))
1562 continue;
1563 if (state->state == 0)
1564 continue;
1565 atomic_inc(&state->count);
1566 spin_unlock(&sp->so_lock);
1567 status = ops->recover_open(sp, state);
1568 if (status >= 0) {
1569 status = nfs4_reclaim_locks(state, ops);
1570 if (status >= 0) {
1571 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1572 spin_lock(&state->state_lock);
1573 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1574 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1575 pr_warn_ratelimited("NFS: "
1576 "%s: Lock reclaim "
1577 "failed!\n", __func__);
1578 }
1579 spin_unlock(&state->state_lock);
1580 }
1581 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1582 &state->flags);
1583 nfs4_put_open_state(state);
1584 spin_lock(&sp->so_lock);
1585 goto restart;
1586 }
1587 }
1588 switch (status) {
1589 default:
1590 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1591 __func__, status);
1592 case -ENOENT:
1593 case -ENOMEM:
1594 case -EACCES:
1595 case -EROFS:
1596 case -EIO:
1597 case -ESTALE:
1598 /* Open state on this file cannot be recovered */
1599 nfs4_state_mark_recovery_failed(state, status);
1600 break;
1601 case -EAGAIN:
1602 ssleep(1);
1603 case -NFS4ERR_ADMIN_REVOKED:
1604 case -NFS4ERR_STALE_STATEID:
1605 case -NFS4ERR_OLD_STATEID:
1606 case -NFS4ERR_BAD_STATEID:
1607 case -NFS4ERR_RECLAIM_BAD:
1608 case -NFS4ERR_RECLAIM_CONFLICT:
1609 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1610 break;
1611 case -NFS4ERR_EXPIRED:
1612 case -NFS4ERR_NO_GRACE:
1613 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1614 case -NFS4ERR_STALE_CLIENTID:
1615 case -NFS4ERR_BADSESSION:
1616 case -NFS4ERR_BADSLOT:
1617 case -NFS4ERR_BAD_HIGH_SLOT:
1618 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1619 goto out_err;
1620 }
1621 nfs4_put_open_state(state);
1622 spin_lock(&sp->so_lock);
1623 goto restart;
1624 }
1625 raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1626 spin_unlock(&sp->so_lock);
1627 return 0;
1628out_err:
1629 nfs4_put_open_state(state);
1630 spin_lock(&sp->so_lock);
1631 raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1632 spin_unlock(&sp->so_lock);
1633 return status;
1634}
1635
1636static void nfs4_clear_open_state(struct nfs4_state *state)
1637{
1638 struct nfs4_lock_state *lock;
1639
1640 clear_bit(NFS_DELEGATED_STATE, &state->flags);
1641 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1642 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1643 clear_bit(NFS_O_RDWR_STATE, &state->flags);
1644 spin_lock(&state->state_lock);
1645 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1646 lock->ls_seqid.flags = 0;
1647 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1648 }
1649 spin_unlock(&state->state_lock);
1650}
1651
1652static void nfs4_reset_seqids(struct nfs_server *server,
1653 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1654{
1655 struct nfs_client *clp = server->nfs_client;
1656 struct nfs4_state_owner *sp;
1657 struct rb_node *pos;
1658 struct nfs4_state *state;
1659
1660 spin_lock(&clp->cl_lock);
1661 for (pos = rb_first(&server->state_owners);
1662 pos != NULL;
1663 pos = rb_next(pos)) {
1664 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1665 sp->so_seqid.flags = 0;
1666 spin_lock(&sp->so_lock);
1667 list_for_each_entry(state, &sp->so_states, open_states) {
1668 if (mark_reclaim(clp, state))
1669 nfs4_clear_open_state(state);
1670 }
1671 spin_unlock(&sp->so_lock);
1672 }
1673 spin_unlock(&clp->cl_lock);
1674}
1675
1676static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1677 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1678{
1679 struct nfs_server *server;
1680
1681 rcu_read_lock();
1682 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1683 nfs4_reset_seqids(server, mark_reclaim);
1684 rcu_read_unlock();
1685}
1686
1687static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1688{
1689 /* Mark all delegations for reclaim */
1690 nfs_delegation_mark_reclaim(clp);
1691 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1692}
1693
1694static int nfs4_reclaim_complete(struct nfs_client *clp,
1695 const struct nfs4_state_recovery_ops *ops,
1696 struct rpc_cred *cred)
1697{
1698 /* Notify the server we're done reclaiming our state */
1699 if (ops->reclaim_complete)
1700 return ops->reclaim_complete(clp, cred);
1701 return 0;
1702}
1703
1704static void nfs4_clear_reclaim_server(struct nfs_server *server)
1705{
1706 struct nfs_client *clp = server->nfs_client;
1707 struct nfs4_state_owner *sp;
1708 struct rb_node *pos;
1709 struct nfs4_state *state;
1710
1711 spin_lock(&clp->cl_lock);
1712 for (pos = rb_first(&server->state_owners);
1713 pos != NULL;
1714 pos = rb_next(pos)) {
1715 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1716 spin_lock(&sp->so_lock);
1717 list_for_each_entry(state, &sp->so_states, open_states) {
1718 if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1719 &state->flags))
1720 continue;
1721 nfs4_state_mark_reclaim_nograce(clp, state);
1722 }
1723 spin_unlock(&sp->so_lock);
1724 }
1725 spin_unlock(&clp->cl_lock);
1726}
1727
1728static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1729{
1730 struct nfs_server *server;
1731
1732 if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1733 return 0;
1734
1735 rcu_read_lock();
1736 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1737 nfs4_clear_reclaim_server(server);
1738 rcu_read_unlock();
1739
1740 nfs_delegation_reap_unclaimed(clp);
1741 return 1;
1742}
1743
1744static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1745{
1746 const struct nfs4_state_recovery_ops *ops;
1747 struct rpc_cred *cred;
1748 int err;
1749
1750 if (!nfs4_state_clear_reclaim_reboot(clp))
1751 return;
1752 ops = clp->cl_mvops->reboot_recovery_ops;
1753 cred = nfs4_get_clid_cred(clp);
1754 err = nfs4_reclaim_complete(clp, ops, cred);
1755 put_rpccred(cred);
1756 if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1757 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1758}
1759
1760static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1761{
1762 nfs_mark_test_expired_all_delegations(clp);
1763 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1764}
1765
1766static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1767{
1768 switch (error) {
1769 case 0:
1770 break;
1771 case -NFS4ERR_CB_PATH_DOWN:
1772 nfs40_handle_cb_pathdown(clp);
1773 break;
1774 case -NFS4ERR_NO_GRACE:
1775 nfs4_state_end_reclaim_reboot(clp);
1776 break;
1777 case -NFS4ERR_STALE_CLIENTID:
1778 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1779 nfs4_state_start_reclaim_reboot(clp);
1780 break;
1781 case -NFS4ERR_EXPIRED:
1782 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1783 nfs4_state_start_reclaim_nograce(clp);
1784 break;
1785 case -NFS4ERR_BADSESSION:
1786 case -NFS4ERR_BADSLOT:
1787 case -NFS4ERR_BAD_HIGH_SLOT:
1788 case -NFS4ERR_DEADSESSION:
1789 case -NFS4ERR_SEQ_FALSE_RETRY:
1790 case -NFS4ERR_SEQ_MISORDERED:
1791 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1792 /* Zero session reset errors */
1793 break;
1794 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1795 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1796 break;
1797 default:
1798 dprintk("%s: failed to handle error %d for server %s\n",
1799 __func__, error, clp->cl_hostname);
1800 return error;
1801 }
1802 dprintk("%s: handled error %d for server %s\n", __func__, error,
1803 clp->cl_hostname);
1804 return 0;
1805}
1806
1807static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1808{
1809 struct nfs4_state_owner *sp;
1810 struct nfs_server *server;
1811 struct rb_node *pos;
1812 int status = 0;
1813
1814restart:
1815 rcu_read_lock();
1816 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1817 nfs4_purge_state_owners(server);
1818 spin_lock(&clp->cl_lock);
1819 for (pos = rb_first(&server->state_owners);
1820 pos != NULL;
1821 pos = rb_next(pos)) {
1822 sp = rb_entry(pos,
1823 struct nfs4_state_owner, so_server_node);
1824 if (!test_and_clear_bit(ops->owner_flag_bit,
1825 &sp->so_flags))
1826 continue;
1827 if (!atomic_inc_not_zero(&sp->so_count))
1828 continue;
1829 spin_unlock(&clp->cl_lock);
1830 rcu_read_unlock();
1831
1832 status = nfs4_reclaim_open_state(sp, ops);
1833 if (status < 0) {
1834 set_bit(ops->owner_flag_bit, &sp->so_flags);
1835 nfs4_put_state_owner(sp);
1836 status = nfs4_recovery_handle_error(clp, status);
1837 return (status != 0) ? status : -EAGAIN;
1838 }
1839
1840 nfs4_put_state_owner(sp);
1841 goto restart;
1842 }
1843 spin_unlock(&clp->cl_lock);
1844 }
1845 rcu_read_unlock();
1846 return 0;
1847}
1848
1849static int nfs4_check_lease(struct nfs_client *clp)
1850{
1851 struct rpc_cred *cred;
1852 const struct nfs4_state_maintenance_ops *ops =
1853 clp->cl_mvops->state_renewal_ops;
1854 int status;
1855
1856 /* Is the client already known to have an expired lease? */
1857 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1858 return 0;
1859 spin_lock(&clp->cl_lock);
1860 cred = ops->get_state_renewal_cred_locked(clp);
1861 spin_unlock(&clp->cl_lock);
1862 if (cred == NULL) {
1863 cred = nfs4_get_clid_cred(clp);
1864 status = -ENOKEY;
1865 if (cred == NULL)
1866 goto out;
1867 }
1868 status = ops->renew_lease(clp, cred);
1869 put_rpccred(cred);
1870 if (status == -ETIMEDOUT) {
1871 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1872 return 0;
1873 }
1874out:
1875 return nfs4_recovery_handle_error(clp, status);
1876}
1877
1878/* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1879 * and for recoverable errors on EXCHANGE_ID for v4.1
1880 */
1881static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1882{
1883 switch (status) {
1884 case -NFS4ERR_SEQ_MISORDERED:
1885 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1886 return -ESERVERFAULT;
1887 /* Lease confirmation error: retry after purging the lease */
1888 ssleep(1);
1889 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1890 break;
1891 case -NFS4ERR_STALE_CLIENTID:
1892 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1893 nfs4_state_start_reclaim_reboot(clp);
1894 break;
1895 case -NFS4ERR_CLID_INUSE:
1896 pr_err("NFS: Server %s reports our clientid is in use\n",
1897 clp->cl_hostname);
1898 nfs_mark_client_ready(clp, -EPERM);
1899 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1900 return -EPERM;
1901 case -EACCES:
1902 case -NFS4ERR_DELAY:
1903 case -ETIMEDOUT:
1904 case -EAGAIN:
1905 ssleep(1);
1906 break;
1907
1908 case -NFS4ERR_MINOR_VERS_MISMATCH:
1909 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1910 nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1911 dprintk("%s: exit with error %d for server %s\n",
1912 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1913 return -EPROTONOSUPPORT;
1914 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1915 * in nfs4_exchange_id */
1916 default:
1917 dprintk("%s: exit with error %d for server %s\n", __func__,
1918 status, clp->cl_hostname);
1919 return status;
1920 }
1921 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1922 dprintk("%s: handled error %d for server %s\n", __func__, status,
1923 clp->cl_hostname);
1924 return 0;
1925}
1926
1927static int nfs4_establish_lease(struct nfs_client *clp)
1928{
1929 struct rpc_cred *cred;
1930 const struct nfs4_state_recovery_ops *ops =
1931 clp->cl_mvops->reboot_recovery_ops;
1932 int status;
1933
1934 nfs4_begin_drain_session(clp);
1935 cred = nfs4_get_clid_cred(clp);
1936 if (cred == NULL)
1937 return -ENOENT;
1938 status = ops->establish_clid(clp, cred);
1939 put_rpccred(cred);
1940 if (status != 0)
1941 return status;
1942 pnfs_destroy_all_layouts(clp);
1943 return 0;
1944}
1945
1946/*
1947 * Returns zero or a negative errno. NFS4ERR values are converted
1948 * to local errno values.
1949 */
1950static int nfs4_reclaim_lease(struct nfs_client *clp)
1951{
1952 int status;
1953
1954 status = nfs4_establish_lease(clp);
1955 if (status < 0)
1956 return nfs4_handle_reclaim_lease_error(clp, status);
1957 if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1958 nfs4_state_start_reclaim_nograce(clp);
1959 if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1960 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1961 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1962 clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1963 return 0;
1964}
1965
1966static int nfs4_purge_lease(struct nfs_client *clp)
1967{
1968 int status;
1969
1970 status = nfs4_establish_lease(clp);
1971 if (status < 0)
1972 return nfs4_handle_reclaim_lease_error(clp, status);
1973 clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1974 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1975 nfs4_state_start_reclaim_nograce(clp);
1976 return 0;
1977}
1978
1979/*
1980 * Try remote migration of one FSID from a source server to a
1981 * destination server. The source server provides a list of
1982 * potential destinations.
1983 *
1984 * Returns zero or a negative NFS4ERR status code.
1985 */
1986static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1987{
1988 struct nfs_client *clp = server->nfs_client;
1989 struct nfs4_fs_locations *locations = NULL;
1990 struct inode *inode;
1991 struct page *page;
1992 int status, result;
1993
1994 dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1995 (unsigned long long)server->fsid.major,
1996 (unsigned long long)server->fsid.minor,
1997 clp->cl_hostname);
1998
1999 result = 0;
2000 page = alloc_page(GFP_KERNEL);
2001 locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2002 if (page == NULL || locations == NULL) {
2003 dprintk("<-- %s: no memory\n", __func__);
2004 goto out;
2005 }
2006
2007 inode = d_inode(server->super->s_root);
2008 result = nfs4_proc_get_locations(inode, locations, page, cred);
2009 if (result) {
2010 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2011 __func__, result);
2012 goto out;
2013 }
2014
2015 result = -NFS4ERR_NXIO;
2016 if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2017 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2018 __func__);
2019 goto out;
2020 }
2021
2022 nfs4_begin_drain_session(clp);
2023
2024 status = nfs4_replace_transport(server, locations);
2025 if (status != 0) {
2026 dprintk("<-- %s: failed to replace transport: %d\n",
2027 __func__, status);
2028 goto out;
2029 }
2030
2031 result = 0;
2032 dprintk("<-- %s: migration succeeded\n", __func__);
2033
2034out:
2035 if (page != NULL)
2036 __free_page(page);
2037 kfree(locations);
2038 if (result) {
2039 pr_err("NFS: migration recovery failed (server %s)\n",
2040 clp->cl_hostname);
2041 set_bit(NFS_MIG_FAILED, &server->mig_status);
2042 }
2043 return result;
2044}
2045
2046/*
2047 * Returns zero or a negative NFS4ERR status code.
2048 */
2049static int nfs4_handle_migration(struct nfs_client *clp)
2050{
2051 const struct nfs4_state_maintenance_ops *ops =
2052 clp->cl_mvops->state_renewal_ops;
2053 struct nfs_server *server;
2054 struct rpc_cred *cred;
2055
2056 dprintk("%s: migration reported on \"%s\"\n", __func__,
2057 clp->cl_hostname);
2058
2059 spin_lock(&clp->cl_lock);
2060 cred = ops->get_state_renewal_cred_locked(clp);
2061 spin_unlock(&clp->cl_lock);
2062 if (cred == NULL)
2063 return -NFS4ERR_NOENT;
2064
2065 clp->cl_mig_gen++;
2066restart:
2067 rcu_read_lock();
2068 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2069 int status;
2070
2071 if (server->mig_gen == clp->cl_mig_gen)
2072 continue;
2073 server->mig_gen = clp->cl_mig_gen;
2074
2075 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2076 &server->mig_status))
2077 continue;
2078
2079 rcu_read_unlock();
2080 status = nfs4_try_migration(server, cred);
2081 if (status < 0) {
2082 put_rpccred(cred);
2083 return status;
2084 }
2085 goto restart;
2086 }
2087 rcu_read_unlock();
2088 put_rpccred(cred);
2089 return 0;
2090}
2091
2092/*
2093 * Test each nfs_server on the clp's cl_superblocks list to see
2094 * if it's moved to another server. Stop when the server no longer
2095 * returns NFS4ERR_LEASE_MOVED.
2096 */
2097static int nfs4_handle_lease_moved(struct nfs_client *clp)
2098{
2099 const struct nfs4_state_maintenance_ops *ops =
2100 clp->cl_mvops->state_renewal_ops;
2101 struct nfs_server *server;
2102 struct rpc_cred *cred;
2103
2104 dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2105 clp->cl_hostname);
2106
2107 spin_lock(&clp->cl_lock);
2108 cred = ops->get_state_renewal_cred_locked(clp);
2109 spin_unlock(&clp->cl_lock);
2110 if (cred == NULL)
2111 return -NFS4ERR_NOENT;
2112
2113 clp->cl_mig_gen++;
2114restart:
2115 rcu_read_lock();
2116 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2117 struct inode *inode;
2118 int status;
2119
2120 if (server->mig_gen == clp->cl_mig_gen)
2121 continue;
2122 server->mig_gen = clp->cl_mig_gen;
2123
2124 rcu_read_unlock();
2125
2126 inode = d_inode(server->super->s_root);
2127 status = nfs4_proc_fsid_present(inode, cred);
2128 if (status != -NFS4ERR_MOVED)
2129 goto restart; /* wasn't this one */
2130 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2131 goto restart; /* there are more */
2132 goto out;
2133 }
2134 rcu_read_unlock();
2135
2136out:
2137 put_rpccred(cred);
2138 return 0;
2139}
2140
2141/**
2142 * nfs4_discover_server_trunking - Detect server IP address trunking
2143 *
2144 * @clp: nfs_client under test
2145 * @result: OUT: found nfs_client, or clp
2146 *
2147 * Returns zero or a negative errno. If zero is returned,
2148 * an nfs_client pointer is planted in "result".
2149 *
2150 * Note: since we are invoked in process context, and
2151 * not from inside the state manager, we cannot use
2152 * nfs4_handle_reclaim_lease_error().
2153 */
2154int nfs4_discover_server_trunking(struct nfs_client *clp,
2155 struct nfs_client **result)
2156{
2157 const struct nfs4_state_recovery_ops *ops =
2158 clp->cl_mvops->reboot_recovery_ops;
2159 struct rpc_clnt *clnt;
2160 struct rpc_cred *cred;
2161 int i, status;
2162
2163 dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2164
2165 clnt = clp->cl_rpcclient;
2166 i = 0;
2167
2168 mutex_lock(&nfs_clid_init_mutex);
2169again:
2170 status = -ENOENT;
2171 cred = nfs4_get_clid_cred(clp);
2172 if (cred == NULL)
2173 goto out_unlock;
2174
2175 status = ops->detect_trunking(clp, result, cred);
2176 put_rpccred(cred);
2177 switch (status) {
2178 case 0:
2179 case -EINTR:
2180 case -ERESTARTSYS:
2181 break;
2182 case -ETIMEDOUT:
2183 if (clnt->cl_softrtry)
2184 break;
2185 case -NFS4ERR_DELAY:
2186 case -EAGAIN:
2187 ssleep(1);
2188 case -NFS4ERR_STALE_CLIENTID:
2189 dprintk("NFS: %s after status %d, retrying\n",
2190 __func__, status);
2191 goto again;
2192 case -EACCES:
2193 if (i++ == 0) {
2194 nfs4_root_machine_cred(clp);
2195 goto again;
2196 }
2197 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2198 break;
2199 case -NFS4ERR_CLID_INUSE:
2200 case -NFS4ERR_WRONGSEC:
2201 /* No point in retrying if we already used RPC_AUTH_UNIX */
2202 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2203 status = -EPERM;
2204 break;
2205 }
2206 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2207 if (IS_ERR(clnt)) {
2208 status = PTR_ERR(clnt);
2209 break;
2210 }
2211 /* Note: this is safe because we haven't yet marked the
2212 * client as ready, so we are the only user of
2213 * clp->cl_rpcclient
2214 */
2215 clnt = xchg(&clp->cl_rpcclient, clnt);
2216 rpc_shutdown_client(clnt);
2217 clnt = clp->cl_rpcclient;
2218 goto again;
2219
2220 case -NFS4ERR_MINOR_VERS_MISMATCH:
2221 status = -EPROTONOSUPPORT;
2222 break;
2223
2224 case -EKEYEXPIRED:
2225 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2226 * in nfs4_exchange_id */
2227 status = -EKEYEXPIRED;
2228 break;
2229 default:
2230 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2231 __func__, status);
2232 status = -EIO;
2233 }
2234
2235out_unlock:
2236 mutex_unlock(&nfs_clid_init_mutex);
2237 dprintk("NFS: %s: status = %d\n", __func__, status);
2238 return status;
2239}
2240
2241#ifdef CONFIG_NFS_V4_1
2242void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2243{
2244 struct nfs_client *clp = session->clp;
2245
2246 switch (err) {
2247 default:
2248 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2249 break;
2250 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2251 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2252 }
2253 nfs4_schedule_state_manager(clp);
2254}
2255EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2256
2257void nfs41_notify_server(struct nfs_client *clp)
2258{
2259 /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2260 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2261 nfs4_schedule_state_manager(clp);
2262}
2263
2264static void nfs4_reset_all_state(struct nfs_client *clp)
2265{
2266 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2267 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2268 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2269 nfs4_state_start_reclaim_nograce(clp);
2270 dprintk("%s: scheduling reset of all state for server %s!\n",
2271 __func__, clp->cl_hostname);
2272 nfs4_schedule_state_manager(clp);
2273 }
2274}
2275
2276static void nfs41_handle_server_reboot(struct nfs_client *clp)
2277{
2278 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2279 nfs4_state_start_reclaim_reboot(clp);
2280 dprintk("%s: server %s rebooted!\n", __func__,
2281 clp->cl_hostname);
2282 nfs4_schedule_state_manager(clp);
2283 }
2284}
2285
2286static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2287{
2288 nfs4_reset_all_state(clp);
2289 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2290}
2291
2292static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2293{
2294 nfs4_state_start_reclaim_nograce(clp);
2295 nfs4_schedule_state_manager(clp);
2296
2297 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2298}
2299
2300static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2301{
2302 /* FIXME: For now, we destroy all layouts. */
2303 pnfs_destroy_all_layouts(clp);
2304 /* FIXME: For now, we test all delegations+open state+locks. */
2305 nfs41_handle_some_state_revoked(clp);
2306 dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2307 clp->cl_hostname);
2308}
2309
2310static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2311{
2312 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2313 nfs4_schedule_state_manager(clp);
2314
2315 dprintk("%s: server %s declared a backchannel fault\n", __func__,
2316 clp->cl_hostname);
2317}
2318
2319static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2320{
2321 if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2322 &clp->cl_state) == 0)
2323 nfs4_schedule_state_manager(clp);
2324}
2325
2326void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2327 bool recovery)
2328{
2329 if (!flags)
2330 return;
2331
2332 dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2333 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2334 /*
2335 * If we're called from the state manager thread, then assume we're
2336 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2337 * Those flags are expected to remain set until we're done
2338 * recovering (see RFC5661, section 18.46.3).
2339 */
2340 if (recovery)
2341 goto out_recovery;
2342
2343 if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2344 nfs41_handle_server_reboot(clp);
2345 if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2346 nfs41_handle_all_state_revoked(clp);
2347 if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2348 SEQ4_STATUS_ADMIN_STATE_REVOKED))
2349 nfs41_handle_some_state_revoked(clp);
2350 if (flags & SEQ4_STATUS_LEASE_MOVED)
2351 nfs4_schedule_lease_moved_recovery(clp);
2352 if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2353 nfs41_handle_recallable_state_revoked(clp);
2354out_recovery:
2355 if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2356 nfs41_handle_backchannel_fault(clp);
2357 else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2358 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2359 nfs41_handle_cb_path_down(clp);
2360}
2361
2362static int nfs4_reset_session(struct nfs_client *clp)
2363{
2364 struct rpc_cred *cred;
2365 int status;
2366
2367 if (!nfs4_has_session(clp))
2368 return 0;
2369 nfs4_begin_drain_session(clp);
2370 cred = nfs4_get_clid_cred(clp);
2371 status = nfs4_proc_destroy_session(clp->cl_session, cred);
2372 switch (status) {
2373 case 0:
2374 case -NFS4ERR_BADSESSION:
2375 case -NFS4ERR_DEADSESSION:
2376 break;
2377 case -NFS4ERR_BACK_CHAN_BUSY:
2378 case -NFS4ERR_DELAY:
2379 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2380 status = 0;
2381 ssleep(1);
2382 goto out;
2383 default:
2384 status = nfs4_recovery_handle_error(clp, status);
2385 goto out;
2386 }
2387
2388 memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2389 status = nfs4_proc_create_session(clp, cred);
2390 if (status) {
2391 dprintk("%s: session reset failed with status %d for server %s!\n",
2392 __func__, status, clp->cl_hostname);
2393 status = nfs4_handle_reclaim_lease_error(clp, status);
2394 goto out;
2395 }
2396 nfs41_finish_session_reset(clp);
2397 dprintk("%s: session reset was successful for server %s!\n",
2398 __func__, clp->cl_hostname);
2399out:
2400 if (cred)
2401 put_rpccred(cred);
2402 return status;
2403}
2404
2405static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2406{
2407 struct rpc_cred *cred;
2408 int ret;
2409
2410 if (!nfs4_has_session(clp))
2411 return 0;
2412 nfs4_begin_drain_session(clp);
2413 cred = nfs4_get_clid_cred(clp);
2414 ret = nfs4_proc_bind_conn_to_session(clp, cred);
2415 if (cred)
2416 put_rpccred(cred);
2417 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2418 switch (ret) {
2419 case 0:
2420 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2421 __func__, clp->cl_hostname);
2422 break;
2423 case -NFS4ERR_DELAY:
2424 ssleep(1);
2425 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2426 break;
2427 default:
2428 return nfs4_recovery_handle_error(clp, ret);
2429 }
2430 return 0;
2431}
2432#else /* CONFIG_NFS_V4_1 */
2433static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2434
2435static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2436{
2437 return 0;
2438}
2439#endif /* CONFIG_NFS_V4_1 */
2440
2441static void nfs4_state_manager(struct nfs_client *clp)
2442{
2443 int status = 0;
2444 const char *section = "", *section_sep = "";
2445
2446 /* Ensure exclusive access to NFSv4 state */
2447 do {
2448 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2449 section = "purge state";
2450 status = nfs4_purge_lease(clp);
2451 if (status < 0)
2452 goto out_error;
2453 continue;
2454 }
2455
2456 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2457 section = "lease expired";
2458 /* We're going to have to re-establish a clientid */
2459 status = nfs4_reclaim_lease(clp);
2460 if (status < 0)
2461 goto out_error;
2462 continue;
2463 }
2464
2465 /* Initialize or reset the session */
2466 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2467 section = "reset session";
2468 status = nfs4_reset_session(clp);
2469 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2470 continue;
2471 if (status < 0)
2472 goto out_error;
2473 }
2474
2475 /* Send BIND_CONN_TO_SESSION */
2476 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2477 &clp->cl_state)) {
2478 section = "bind conn to session";
2479 status = nfs4_bind_conn_to_session(clp);
2480 if (status < 0)
2481 goto out_error;
2482 continue;
2483 }
2484
2485 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2486 section = "check lease";
2487 status = nfs4_check_lease(clp);
2488 if (status < 0)
2489 goto out_error;
2490 continue;
2491 }
2492
2493 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2494 section = "migration";
2495 status = nfs4_handle_migration(clp);
2496 if (status < 0)
2497 goto out_error;
2498 }
2499
2500 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2501 section = "lease moved";
2502 status = nfs4_handle_lease_moved(clp);
2503 if (status < 0)
2504 goto out_error;
2505 }
2506
2507 /* First recover reboot state... */
2508 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2509 section = "reclaim reboot";
2510 status = nfs4_do_reclaim(clp,
2511 clp->cl_mvops->reboot_recovery_ops);
2512 if (status == -EAGAIN)
2513 continue;
2514 if (status < 0)
2515 goto out_error;
2516 nfs4_state_end_reclaim_reboot(clp);
2517 }
2518
2519 /* Detect expired delegations... */
2520 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2521 section = "detect expired delegations";
2522 nfs_reap_expired_delegations(clp);
2523 continue;
2524 }
2525
2526 /* Now recover expired state... */
2527 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2528 section = "reclaim nograce";
2529 status = nfs4_do_reclaim(clp,
2530 clp->cl_mvops->nograce_recovery_ops);
2531 if (status == -EAGAIN)
2532 continue;
2533 if (status < 0)
2534 goto out_error;
2535 }
2536
2537 nfs4_end_drain_session(clp);
2538 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2539 nfs_client_return_marked_delegations(clp);
2540 continue;
2541 }
2542
2543 nfs4_clear_state_manager_bit(clp);
2544 /* Did we race with an attempt to give us more work? */
2545 if (clp->cl_state == 0)
2546 break;
2547 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2548 break;
2549 } while (refcount_read(&clp->cl_count) > 1);
2550 return;
2551out_error:
2552 if (strlen(section))
2553 section_sep = ": ";
2554 pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2555 " with error %d\n", section_sep, section,
2556 clp->cl_hostname, -status);
2557 ssleep(1);
2558 nfs4_end_drain_session(clp);
2559 nfs4_clear_state_manager_bit(clp);
2560}
2561
2562static int nfs4_run_state_manager(void *ptr)
2563{
2564 struct nfs_client *clp = ptr;
2565
2566 allow_signal(SIGKILL);
2567 nfs4_state_manager(clp);
2568 nfs_put_client(clp);
2569 module_put_and_exit(0);
2570 return 0;
2571}
2572
2573/*
2574 * Local variables:
2575 * c-basic-offset: 8
2576 * End:
2577 */
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
68const nfs4_stateid zero_stateid;
69static DEFINE_MUTEX(nfs_clid_init_mutex);
70
71int 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);
93do_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);
99out:
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 */
116int 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 }
144out:
145 return status;
146}
147
148struct 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
157static void nfs4_root_machine_cred(struct nfs_client *clp)
158{
159 struct rpc_cred *cred, *new;
160
161 new = rpc_lookup_machine_cred(NULL);
162 spin_lock(&clp->cl_lock);
163 cred = clp->cl_machine_cred;
164 clp->cl_machine_cred = new;
165 spin_unlock(&clp->cl_lock);
166 if (cred != NULL)
167 put_rpccred(cred);
168}
169
170static struct rpc_cred *
171nfs4_get_renew_cred_server_locked(struct nfs_server *server)
172{
173 struct rpc_cred *cred = NULL;
174 struct nfs4_state_owner *sp;
175 struct rb_node *pos;
176
177 for (pos = rb_first(&server->state_owners);
178 pos != NULL;
179 pos = rb_next(pos)) {
180 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
181 if (list_empty(&sp->so_states))
182 continue;
183 cred = get_rpccred(sp->so_cred);
184 break;
185 }
186 return cred;
187}
188
189/**
190 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
191 * @clp: client state handle
192 *
193 * Returns an rpc_cred with reference count bumped, or NULL.
194 * Caller must hold clp->cl_lock.
195 */
196struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
197{
198 struct rpc_cred *cred = NULL;
199 struct nfs_server *server;
200
201 /* Use machine credentials if available */
202 cred = nfs4_get_machine_cred_locked(clp);
203 if (cred != NULL)
204 goto out;
205
206 rcu_read_lock();
207 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
208 cred = nfs4_get_renew_cred_server_locked(server);
209 if (cred != NULL)
210 break;
211 }
212 rcu_read_unlock();
213
214out:
215 return cred;
216}
217
218static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
219{
220 if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
221 spin_lock(&tbl->slot_tbl_lock);
222 nfs41_wake_slot_table(tbl);
223 spin_unlock(&tbl->slot_tbl_lock);
224 }
225}
226
227static void nfs4_end_drain_session(struct nfs_client *clp)
228{
229 struct nfs4_session *ses = clp->cl_session;
230
231 if (clp->cl_slot_tbl) {
232 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
233 return;
234 }
235
236 if (ses != NULL) {
237 nfs4_end_drain_slot_table(&ses->bc_slot_table);
238 nfs4_end_drain_slot_table(&ses->fc_slot_table);
239 }
240}
241
242static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
243{
244 set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
245 spin_lock(&tbl->slot_tbl_lock);
246 if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
247 reinit_completion(&tbl->complete);
248 spin_unlock(&tbl->slot_tbl_lock);
249 return wait_for_completion_interruptible(&tbl->complete);
250 }
251 spin_unlock(&tbl->slot_tbl_lock);
252 return 0;
253}
254
255static int nfs4_begin_drain_session(struct nfs_client *clp)
256{
257 struct nfs4_session *ses = clp->cl_session;
258 int ret = 0;
259
260 if (clp->cl_slot_tbl)
261 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
262
263 /* back channel */
264 ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
265 if (ret)
266 return ret;
267 /* fore channel */
268 return nfs4_drain_slot_tbl(&ses->fc_slot_table);
269}
270
271#if defined(CONFIG_NFS_V4_1)
272
273static int nfs41_setup_state_renewal(struct nfs_client *clp)
274{
275 int status;
276 struct nfs_fsinfo fsinfo;
277
278 if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
279 nfs4_schedule_state_renewal(clp);
280 return 0;
281 }
282
283 status = nfs4_proc_get_lease_time(clp, &fsinfo);
284 if (status == 0) {
285 /* Update lease time and schedule renewal */
286 spin_lock(&clp->cl_lock);
287 clp->cl_lease_time = fsinfo.lease_time * HZ;
288 clp->cl_last_renewal = jiffies;
289 spin_unlock(&clp->cl_lock);
290
291 nfs4_schedule_state_renewal(clp);
292 }
293
294 return status;
295}
296
297static void nfs41_finish_session_reset(struct nfs_client *clp)
298{
299 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300 clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301 /* create_session negotiated new slot table */
302 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303 nfs41_setup_state_renewal(clp);
304}
305
306int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
307{
308 int status;
309
310 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311 goto do_confirm;
312 nfs4_begin_drain_session(clp);
313 status = nfs4_proc_exchange_id(clp, cred);
314 if (status != 0)
315 goto out;
316 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
317do_confirm:
318 status = nfs4_proc_create_session(clp, cred);
319 if (status != 0)
320 goto out;
321 nfs41_finish_session_reset(clp);
322 nfs_mark_client_ready(clp, NFS_CS_READY);
323out:
324 return status;
325}
326
327/**
328 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
329 *
330 * @clp: nfs_client under test
331 * @result: OUT: found nfs_client, or clp
332 * @cred: credential to use for trunking test
333 *
334 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
335 * If NFS4_OK is returned, an nfs_client pointer is planted in
336 * "result".
337 *
338 * Note: The returned client may not yet be marked ready.
339 */
340int nfs41_discover_server_trunking(struct nfs_client *clp,
341 struct nfs_client **result,
342 struct rpc_cred *cred)
343{
344 int status;
345
346 status = nfs4_proc_exchange_id(clp, cred);
347 if (status != NFS4_OK)
348 return status;
349 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
350
351 return nfs41_walk_client_list(clp, result, cred);
352}
353
354#endif /* CONFIG_NFS_V4_1 */
355
356/**
357 * nfs4_get_clid_cred - Acquire credential for a setclientid operation
358 * @clp: client state handle
359 *
360 * Returns an rpc_cred with reference count bumped, or NULL.
361 */
362struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
363{
364 struct rpc_cred *cred;
365
366 spin_lock(&clp->cl_lock);
367 cred = nfs4_get_machine_cred_locked(clp);
368 spin_unlock(&clp->cl_lock);
369 return cred;
370}
371
372static struct nfs4_state_owner *
373nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
374{
375 struct rb_node **p = &server->state_owners.rb_node,
376 *parent = NULL;
377 struct nfs4_state_owner *sp;
378
379 while (*p != NULL) {
380 parent = *p;
381 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
382
383 if (cred < sp->so_cred)
384 p = &parent->rb_left;
385 else if (cred > sp->so_cred)
386 p = &parent->rb_right;
387 else {
388 if (!list_empty(&sp->so_lru))
389 list_del_init(&sp->so_lru);
390 atomic_inc(&sp->so_count);
391 return sp;
392 }
393 }
394 return NULL;
395}
396
397static struct nfs4_state_owner *
398nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
399{
400 struct nfs_server *server = new->so_server;
401 struct rb_node **p = &server->state_owners.rb_node,
402 *parent = NULL;
403 struct nfs4_state_owner *sp;
404 int err;
405
406 while (*p != NULL) {
407 parent = *p;
408 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
409
410 if (new->so_cred < sp->so_cred)
411 p = &parent->rb_left;
412 else if (new->so_cred > sp->so_cred)
413 p = &parent->rb_right;
414 else {
415 if (!list_empty(&sp->so_lru))
416 list_del_init(&sp->so_lru);
417 atomic_inc(&sp->so_count);
418 return sp;
419 }
420 }
421 err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
422 if (err)
423 return ERR_PTR(err);
424 rb_link_node(&new->so_server_node, parent, p);
425 rb_insert_color(&new->so_server_node, &server->state_owners);
426 return new;
427}
428
429static void
430nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
431{
432 struct nfs_server *server = sp->so_server;
433
434 if (!RB_EMPTY_NODE(&sp->so_server_node))
435 rb_erase(&sp->so_server_node, &server->state_owners);
436 ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
437}
438
439static void
440nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
441{
442 sc->create_time = ktime_get();
443 sc->flags = 0;
444 sc->counter = 0;
445 spin_lock_init(&sc->lock);
446 INIT_LIST_HEAD(&sc->list);
447 rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
448}
449
450static void
451nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
452{
453 rpc_destroy_wait_queue(&sc->wait);
454}
455
456/*
457 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
458 * create a new state_owner.
459 *
460 */
461static struct nfs4_state_owner *
462nfs4_alloc_state_owner(struct nfs_server *server,
463 struct rpc_cred *cred,
464 gfp_t gfp_flags)
465{
466 struct nfs4_state_owner *sp;
467
468 sp = kzalloc(sizeof(*sp), gfp_flags);
469 if (!sp)
470 return NULL;
471 sp->so_server = server;
472 sp->so_cred = get_rpccred(cred);
473 spin_lock_init(&sp->so_lock);
474 INIT_LIST_HEAD(&sp->so_states);
475 nfs4_init_seqid_counter(&sp->so_seqid);
476 atomic_set(&sp->so_count, 1);
477 INIT_LIST_HEAD(&sp->so_lru);
478 seqcount_init(&sp->so_reclaim_seqcount);
479 mutex_init(&sp->so_delegreturn_mutex);
480 return sp;
481}
482
483static void
484nfs4_drop_state_owner(struct nfs4_state_owner *sp)
485{
486 struct rb_node *rb_node = &sp->so_server_node;
487
488 if (!RB_EMPTY_NODE(rb_node)) {
489 struct nfs_server *server = sp->so_server;
490 struct nfs_client *clp = server->nfs_client;
491
492 spin_lock(&clp->cl_lock);
493 if (!RB_EMPTY_NODE(rb_node)) {
494 rb_erase(rb_node, &server->state_owners);
495 RB_CLEAR_NODE(rb_node);
496 }
497 spin_unlock(&clp->cl_lock);
498 }
499}
500
501static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
502{
503 nfs4_destroy_seqid_counter(&sp->so_seqid);
504 put_rpccred(sp->so_cred);
505 kfree(sp);
506}
507
508static void nfs4_gc_state_owners(struct nfs_server *server)
509{
510 struct nfs_client *clp = server->nfs_client;
511 struct nfs4_state_owner *sp, *tmp;
512 unsigned long time_min, time_max;
513 LIST_HEAD(doomed);
514
515 spin_lock(&clp->cl_lock);
516 time_max = jiffies;
517 time_min = (long)time_max - (long)clp->cl_lease_time;
518 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
519 /* NB: LRU is sorted so that oldest is at the head */
520 if (time_in_range(sp->so_expires, time_min, time_max))
521 break;
522 list_move(&sp->so_lru, &doomed);
523 nfs4_remove_state_owner_locked(sp);
524 }
525 spin_unlock(&clp->cl_lock);
526
527 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
528 list_del(&sp->so_lru);
529 nfs4_free_state_owner(sp);
530 }
531}
532
533/**
534 * nfs4_get_state_owner - Look up a state owner given a credential
535 * @server: nfs_server to search
536 * @cred: RPC credential to match
537 *
538 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
539 */
540struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
541 struct rpc_cred *cred,
542 gfp_t gfp_flags)
543{
544 struct nfs_client *clp = server->nfs_client;
545 struct nfs4_state_owner *sp, *new;
546
547 spin_lock(&clp->cl_lock);
548 sp = nfs4_find_state_owner_locked(server, cred);
549 spin_unlock(&clp->cl_lock);
550 if (sp != NULL)
551 goto out;
552 new = nfs4_alloc_state_owner(server, cred, gfp_flags);
553 if (new == NULL)
554 goto out;
555 do {
556 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
557 break;
558 spin_lock(&clp->cl_lock);
559 sp = nfs4_insert_state_owner_locked(new);
560 spin_unlock(&clp->cl_lock);
561 } while (sp == ERR_PTR(-EAGAIN));
562 if (sp != new)
563 nfs4_free_state_owner(new);
564out:
565 nfs4_gc_state_owners(server);
566 return sp;
567}
568
569/**
570 * nfs4_put_state_owner - Release a nfs4_state_owner
571 * @sp: state owner data to release
572 *
573 * Note that we keep released state owners on an LRU
574 * list.
575 * This caches valid state owners so that they can be
576 * reused, to avoid the OPEN_CONFIRM on minor version 0.
577 * It also pins the uniquifier of dropped state owners for
578 * a while, to ensure that those state owner names are
579 * never reused.
580 */
581void nfs4_put_state_owner(struct nfs4_state_owner *sp)
582{
583 struct nfs_server *server = sp->so_server;
584 struct nfs_client *clp = server->nfs_client;
585
586 if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
587 return;
588
589 sp->so_expires = jiffies;
590 list_add_tail(&sp->so_lru, &server->state_owners_lru);
591 spin_unlock(&clp->cl_lock);
592}
593
594/**
595 * nfs4_purge_state_owners - Release all cached state owners
596 * @server: nfs_server with cached state owners to release
597 *
598 * Called at umount time. Remaining state owners will be on
599 * the LRU with ref count of zero.
600 */
601void nfs4_purge_state_owners(struct nfs_server *server)
602{
603 struct nfs_client *clp = server->nfs_client;
604 struct nfs4_state_owner *sp, *tmp;
605 LIST_HEAD(doomed);
606
607 spin_lock(&clp->cl_lock);
608 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
609 list_move(&sp->so_lru, &doomed);
610 nfs4_remove_state_owner_locked(sp);
611 }
612 spin_unlock(&clp->cl_lock);
613
614 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
615 list_del(&sp->so_lru);
616 nfs4_free_state_owner(sp);
617 }
618}
619
620static struct nfs4_state *
621nfs4_alloc_open_state(void)
622{
623 struct nfs4_state *state;
624
625 state = kzalloc(sizeof(*state), GFP_NOFS);
626 if (!state)
627 return NULL;
628 atomic_set(&state->count, 1);
629 INIT_LIST_HEAD(&state->lock_states);
630 spin_lock_init(&state->state_lock);
631 seqlock_init(&state->seqlock);
632 return state;
633}
634
635void
636nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
637{
638 if (state->state == fmode)
639 return;
640 /* NB! List reordering - see the reclaim code for why. */
641 if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
642 if (fmode & FMODE_WRITE)
643 list_move(&state->open_states, &state->owner->so_states);
644 else
645 list_move_tail(&state->open_states, &state->owner->so_states);
646 }
647 state->state = fmode;
648}
649
650static struct nfs4_state *
651__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
652{
653 struct nfs_inode *nfsi = NFS_I(inode);
654 struct nfs4_state *state;
655
656 list_for_each_entry(state, &nfsi->open_states, inode_states) {
657 if (state->owner != owner)
658 continue;
659 if (!nfs4_valid_open_stateid(state))
660 continue;
661 if (atomic_inc_not_zero(&state->count))
662 return state;
663 }
664 return NULL;
665}
666
667static void
668nfs4_free_open_state(struct nfs4_state *state)
669{
670 kfree(state);
671}
672
673struct nfs4_state *
674nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
675{
676 struct nfs4_state *state, *new;
677 struct nfs_inode *nfsi = NFS_I(inode);
678
679 spin_lock(&inode->i_lock);
680 state = __nfs4_find_state_byowner(inode, owner);
681 spin_unlock(&inode->i_lock);
682 if (state)
683 goto out;
684 new = nfs4_alloc_open_state();
685 spin_lock(&owner->so_lock);
686 spin_lock(&inode->i_lock);
687 state = __nfs4_find_state_byowner(inode, owner);
688 if (state == NULL && new != NULL) {
689 state = new;
690 state->owner = owner;
691 atomic_inc(&owner->so_count);
692 list_add(&state->inode_states, &nfsi->open_states);
693 ihold(inode);
694 state->inode = inode;
695 spin_unlock(&inode->i_lock);
696 /* Note: The reclaim code dictates that we add stateless
697 * and read-only stateids to the end of the list */
698 list_add_tail(&state->open_states, &owner->so_states);
699 spin_unlock(&owner->so_lock);
700 } else {
701 spin_unlock(&inode->i_lock);
702 spin_unlock(&owner->so_lock);
703 if (new)
704 nfs4_free_open_state(new);
705 }
706out:
707 return state;
708}
709
710void nfs4_put_open_state(struct nfs4_state *state)
711{
712 struct inode *inode = state->inode;
713 struct nfs4_state_owner *owner = state->owner;
714
715 if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
716 return;
717 spin_lock(&inode->i_lock);
718 list_del(&state->inode_states);
719 list_del(&state->open_states);
720 spin_unlock(&inode->i_lock);
721 spin_unlock(&owner->so_lock);
722 iput(inode);
723 nfs4_free_open_state(state);
724 nfs4_put_state_owner(owner);
725}
726
727/*
728 * Close the current file.
729 */
730static void __nfs4_close(struct nfs4_state *state,
731 fmode_t fmode, gfp_t gfp_mask, int wait)
732{
733 struct nfs4_state_owner *owner = state->owner;
734 int call_close = 0;
735 fmode_t newstate;
736
737 atomic_inc(&owner->so_count);
738 /* Protect against nfs4_find_state() */
739 spin_lock(&owner->so_lock);
740 switch (fmode & (FMODE_READ | FMODE_WRITE)) {
741 case FMODE_READ:
742 state->n_rdonly--;
743 break;
744 case FMODE_WRITE:
745 state->n_wronly--;
746 break;
747 case FMODE_READ|FMODE_WRITE:
748 state->n_rdwr--;
749 }
750 newstate = FMODE_READ|FMODE_WRITE;
751 if (state->n_rdwr == 0) {
752 if (state->n_rdonly == 0) {
753 newstate &= ~FMODE_READ;
754 call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
755 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
756 }
757 if (state->n_wronly == 0) {
758 newstate &= ~FMODE_WRITE;
759 call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
760 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
761 }
762 if (newstate == 0)
763 clear_bit(NFS_DELEGATED_STATE, &state->flags);
764 }
765 nfs4_state_set_mode_locked(state, newstate);
766 spin_unlock(&owner->so_lock);
767
768 if (!call_close) {
769 nfs4_put_open_state(state);
770 nfs4_put_state_owner(owner);
771 } else
772 nfs4_do_close(state, gfp_mask, wait);
773}
774
775void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
776{
777 __nfs4_close(state, fmode, GFP_NOFS, 0);
778}
779
780void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
781{
782 __nfs4_close(state, fmode, GFP_KERNEL, 1);
783}
784
785/*
786 * Search the state->lock_states for an existing lock_owner
787 * that is compatible with current->files
788 */
789static struct nfs4_lock_state *
790__nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
791{
792 struct nfs4_lock_state *pos;
793 list_for_each_entry(pos, &state->lock_states, ls_locks) {
794 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
795 continue;
796 switch (pos->ls_owner.lo_type) {
797 case NFS4_POSIX_LOCK_TYPE:
798 if (pos->ls_owner.lo_u.posix_owner != fl_owner)
799 continue;
800 break;
801 case NFS4_FLOCK_LOCK_TYPE:
802 if (pos->ls_owner.lo_u.flock_owner != fl_pid)
803 continue;
804 }
805 atomic_inc(&pos->ls_count);
806 return pos;
807 }
808 return NULL;
809}
810
811/*
812 * Return a compatible lock_state. If no initialized lock_state structure
813 * exists, return an uninitialized one.
814 *
815 */
816static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
817{
818 struct nfs4_lock_state *lsp;
819 struct nfs_server *server = state->owner->so_server;
820
821 lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
822 if (lsp == NULL)
823 return NULL;
824 nfs4_init_seqid_counter(&lsp->ls_seqid);
825 atomic_set(&lsp->ls_count, 1);
826 lsp->ls_state = state;
827 lsp->ls_owner.lo_type = type;
828 switch (lsp->ls_owner.lo_type) {
829 case NFS4_FLOCK_LOCK_TYPE:
830 lsp->ls_owner.lo_u.flock_owner = fl_pid;
831 break;
832 case NFS4_POSIX_LOCK_TYPE:
833 lsp->ls_owner.lo_u.posix_owner = fl_owner;
834 break;
835 default:
836 goto out_free;
837 }
838 lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
839 if (lsp->ls_seqid.owner_id < 0)
840 goto out_free;
841 INIT_LIST_HEAD(&lsp->ls_locks);
842 return lsp;
843out_free:
844 kfree(lsp);
845 return NULL;
846}
847
848void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
849{
850 ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
851 nfs4_destroy_seqid_counter(&lsp->ls_seqid);
852 kfree(lsp);
853}
854
855/*
856 * Return a compatible lock_state. If no initialized lock_state structure
857 * exists, return an uninitialized one.
858 *
859 */
860static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
861{
862 struct nfs4_lock_state *lsp, *new = NULL;
863
864 for(;;) {
865 spin_lock(&state->state_lock);
866 lsp = __nfs4_find_lock_state(state, owner, pid, type);
867 if (lsp != NULL)
868 break;
869 if (new != NULL) {
870 list_add(&new->ls_locks, &state->lock_states);
871 set_bit(LK_STATE_IN_USE, &state->flags);
872 lsp = new;
873 new = NULL;
874 break;
875 }
876 spin_unlock(&state->state_lock);
877 new = nfs4_alloc_lock_state(state, owner, pid, type);
878 if (new == NULL)
879 return NULL;
880 }
881 spin_unlock(&state->state_lock);
882 if (new != NULL)
883 nfs4_free_lock_state(state->owner->so_server, new);
884 return lsp;
885}
886
887/*
888 * Release reference to lock_state, and free it if we see that
889 * it is no longer in use
890 */
891void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
892{
893 struct nfs_server *server;
894 struct nfs4_state *state;
895
896 if (lsp == NULL)
897 return;
898 state = lsp->ls_state;
899 if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
900 return;
901 list_del(&lsp->ls_locks);
902 if (list_empty(&state->lock_states))
903 clear_bit(LK_STATE_IN_USE, &state->flags);
904 spin_unlock(&state->state_lock);
905 server = state->owner->so_server;
906 if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
907 struct nfs_client *clp = server->nfs_client;
908
909 clp->cl_mvops->free_lock_state(server, lsp);
910 } else
911 nfs4_free_lock_state(server, lsp);
912}
913
914static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
915{
916 struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
917
918 dst->fl_u.nfs4_fl.owner = lsp;
919 atomic_inc(&lsp->ls_count);
920}
921
922static void nfs4_fl_release_lock(struct file_lock *fl)
923{
924 nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
925}
926
927static const struct file_lock_operations nfs4_fl_lock_ops = {
928 .fl_copy_lock = nfs4_fl_copy_lock,
929 .fl_release_private = nfs4_fl_release_lock,
930};
931
932int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
933{
934 struct nfs4_lock_state *lsp;
935
936 if (fl->fl_ops != NULL)
937 return 0;
938 if (fl->fl_flags & FL_POSIX)
939 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
940 else if (fl->fl_flags & FL_FLOCK)
941 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
942 NFS4_FLOCK_LOCK_TYPE);
943 else
944 return -EINVAL;
945 if (lsp == NULL)
946 return -ENOMEM;
947 fl->fl_u.nfs4_fl.owner = lsp;
948 fl->fl_ops = &nfs4_fl_lock_ops;
949 return 0;
950}
951
952static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
953 struct nfs4_state *state,
954 const struct nfs_lockowner *lockowner)
955{
956 struct nfs4_lock_state *lsp;
957 fl_owner_t fl_owner;
958 pid_t fl_pid;
959 int ret = -ENOENT;
960
961
962 if (lockowner == NULL)
963 goto out;
964
965 if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
966 goto out;
967
968 fl_owner = lockowner->l_owner;
969 fl_pid = lockowner->l_pid;
970 spin_lock(&state->state_lock);
971 lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
972 if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
973 ret = -EIO;
974 else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
975 nfs4_stateid_copy(dst, &lsp->ls_stateid);
976 ret = 0;
977 }
978 spin_unlock(&state->state_lock);
979 nfs4_put_lock_state(lsp);
980out:
981 return ret;
982}
983
984static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
985{
986 const nfs4_stateid *src;
987 int seq;
988
989 do {
990 src = &zero_stateid;
991 seq = read_seqbegin(&state->seqlock);
992 if (test_bit(NFS_OPEN_STATE, &state->flags))
993 src = &state->open_stateid;
994 nfs4_stateid_copy(dst, src);
995 } while (read_seqretry(&state->seqlock, seq));
996}
997
998/*
999 * Byte-range lock aware utility to initialize the stateid of read/write
1000 * requests.
1001 */
1002int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1003 fmode_t fmode, const struct nfs_lockowner *lockowner)
1004{
1005 int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1006 if (ret == -EIO)
1007 /* A lost lock - don't even consider delegations */
1008 goto out;
1009 /* returns true if delegation stateid found and copied */
1010 if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
1011 ret = 0;
1012 goto out;
1013 }
1014 if (ret != -ENOENT)
1015 /* nfs4_copy_delegation_stateid() didn't over-write
1016 * dst, so it still has the lock stateid which we now
1017 * choose to use.
1018 */
1019 goto out;
1020 nfs4_copy_open_stateid(dst, state);
1021 ret = 0;
1022out:
1023 if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1024 dst->seqid = 0;
1025 return ret;
1026}
1027
1028struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1029{
1030 struct nfs_seqid *new;
1031
1032 new = kmalloc(sizeof(*new), gfp_mask);
1033 if (new != NULL) {
1034 new->sequence = counter;
1035 INIT_LIST_HEAD(&new->list);
1036 new->task = NULL;
1037 }
1038 return new;
1039}
1040
1041void nfs_release_seqid(struct nfs_seqid *seqid)
1042{
1043 struct nfs_seqid_counter *sequence;
1044
1045 if (list_empty(&seqid->list))
1046 return;
1047 sequence = seqid->sequence;
1048 spin_lock(&sequence->lock);
1049 list_del_init(&seqid->list);
1050 if (!list_empty(&sequence->list)) {
1051 struct nfs_seqid *next;
1052
1053 next = list_first_entry(&sequence->list,
1054 struct nfs_seqid, list);
1055 rpc_wake_up_queued_task(&sequence->wait, next->task);
1056 }
1057 spin_unlock(&sequence->lock);
1058}
1059
1060void nfs_free_seqid(struct nfs_seqid *seqid)
1061{
1062 nfs_release_seqid(seqid);
1063 kfree(seqid);
1064}
1065
1066/*
1067 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1068 * failed with a seqid incrementing error -
1069 * see comments nfs4.h:seqid_mutating_error()
1070 */
1071static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1072{
1073 switch (status) {
1074 case 0:
1075 break;
1076 case -NFS4ERR_BAD_SEQID:
1077 if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1078 return;
1079 pr_warn_ratelimited("NFS: v4 server returned a bad"
1080 " sequence-id error on an"
1081 " unconfirmed sequence %p!\n",
1082 seqid->sequence);
1083 case -NFS4ERR_STALE_CLIENTID:
1084 case -NFS4ERR_STALE_STATEID:
1085 case -NFS4ERR_BAD_STATEID:
1086 case -NFS4ERR_BADXDR:
1087 case -NFS4ERR_RESOURCE:
1088 case -NFS4ERR_NOFILEHANDLE:
1089 /* Non-seqid mutating errors */
1090 return;
1091 };
1092 /*
1093 * Note: no locking needed as we are guaranteed to be first
1094 * on the sequence list
1095 */
1096 seqid->sequence->counter++;
1097}
1098
1099void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1100{
1101 struct nfs4_state_owner *sp = container_of(seqid->sequence,
1102 struct nfs4_state_owner, so_seqid);
1103 struct nfs_server *server = sp->so_server;
1104
1105 if (status == -NFS4ERR_BAD_SEQID)
1106 nfs4_drop_state_owner(sp);
1107 if (!nfs4_has_session(server->nfs_client))
1108 nfs_increment_seqid(status, seqid);
1109}
1110
1111/*
1112 * Increment the seqid if the LOCK/LOCKU succeeded, or
1113 * failed with a seqid incrementing error -
1114 * see comments nfs4.h:seqid_mutating_error()
1115 */
1116void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1117{
1118 nfs_increment_seqid(status, seqid);
1119}
1120
1121int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1122{
1123 struct nfs_seqid_counter *sequence = seqid->sequence;
1124 int status = 0;
1125
1126 spin_lock(&sequence->lock);
1127 seqid->task = task;
1128 if (list_empty(&seqid->list))
1129 list_add_tail(&seqid->list, &sequence->list);
1130 if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1131 goto unlock;
1132 rpc_sleep_on(&sequence->wait, task, NULL);
1133 status = -EAGAIN;
1134unlock:
1135 spin_unlock(&sequence->lock);
1136 return status;
1137}
1138
1139static int nfs4_run_state_manager(void *);
1140
1141static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1142{
1143 smp_mb__before_clear_bit();
1144 clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1145 smp_mb__after_clear_bit();
1146 wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1147 rpc_wake_up(&clp->cl_rpcwaitq);
1148}
1149
1150/*
1151 * Schedule the nfs_client asynchronous state management routine
1152 */
1153void nfs4_schedule_state_manager(struct nfs_client *clp)
1154{
1155 struct task_struct *task;
1156 char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1157
1158 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1159 return;
1160 __module_get(THIS_MODULE);
1161 atomic_inc(&clp->cl_count);
1162
1163 /* The rcu_read_lock() is not strictly necessary, as the state
1164 * manager is the only thread that ever changes the rpc_xprt
1165 * after it's initialized. At this point, we're single threaded. */
1166 rcu_read_lock();
1167 snprintf(buf, sizeof(buf), "%s-manager",
1168 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1169 rcu_read_unlock();
1170 task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1171 if (IS_ERR(task)) {
1172 printk(KERN_ERR "%s: kthread_run: %ld\n",
1173 __func__, PTR_ERR(task));
1174 nfs4_clear_state_manager_bit(clp);
1175 nfs_put_client(clp);
1176 module_put(THIS_MODULE);
1177 }
1178}
1179
1180/*
1181 * Schedule a lease recovery attempt
1182 */
1183void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1184{
1185 if (!clp)
1186 return;
1187 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1188 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1189 dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1190 clp->cl_hostname);
1191 nfs4_schedule_state_manager(clp);
1192}
1193EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1194
1195/**
1196 * nfs4_schedule_migration_recovery - trigger migration recovery
1197 *
1198 * @server: FSID that is migrating
1199 *
1200 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1201 * value is returned.
1202 */
1203int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1204{
1205 struct nfs_client *clp = server->nfs_client;
1206
1207 if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1208 pr_err("NFS: volatile file handles not supported (server %s)\n",
1209 clp->cl_hostname);
1210 return -NFS4ERR_IO;
1211 }
1212
1213 if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1214 return -NFS4ERR_IO;
1215
1216 dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1217 __func__,
1218 (unsigned long long)server->fsid.major,
1219 (unsigned long long)server->fsid.minor,
1220 clp->cl_hostname);
1221
1222 set_bit(NFS_MIG_IN_TRANSITION,
1223 &((struct nfs_server *)server)->mig_status);
1224 set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1225
1226 nfs4_schedule_state_manager(clp);
1227 return 0;
1228}
1229EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1230
1231/**
1232 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1233 *
1234 * @clp: server to check for moved leases
1235 *
1236 */
1237void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1238{
1239 dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1240 __func__, clp->cl_clientid, clp->cl_hostname);
1241
1242 set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1243 nfs4_schedule_state_manager(clp);
1244}
1245EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1246
1247int nfs4_wait_clnt_recover(struct nfs_client *clp)
1248{
1249 int res;
1250
1251 might_sleep();
1252
1253 atomic_inc(&clp->cl_count);
1254 res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1255 nfs_wait_bit_killable, TASK_KILLABLE);
1256 if (res)
1257 goto out;
1258 if (clp->cl_cons_state < 0)
1259 res = clp->cl_cons_state;
1260out:
1261 nfs_put_client(clp);
1262 return res;
1263}
1264
1265int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1266{
1267 unsigned int loop;
1268 int ret;
1269
1270 for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1271 ret = nfs4_wait_clnt_recover(clp);
1272 if (ret != 0)
1273 break;
1274 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1275 !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1276 break;
1277 nfs4_schedule_state_manager(clp);
1278 ret = -EIO;
1279 }
1280 return ret;
1281}
1282
1283/*
1284 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1285 * @clp: client to process
1286 *
1287 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1288 * resend of the SETCLIENTID and hence re-establish the
1289 * callback channel. Then return all existing delegations.
1290 */
1291static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1292{
1293 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1294 nfs_expire_all_delegations(clp);
1295 dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1296 clp->cl_hostname);
1297}
1298
1299void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1300{
1301 nfs40_handle_cb_pathdown(clp);
1302 nfs4_schedule_state_manager(clp);
1303}
1304
1305static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1306{
1307
1308 set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1309 /* Don't recover state that expired before the reboot */
1310 if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1311 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1312 return 0;
1313 }
1314 set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1315 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1316 return 1;
1317}
1318
1319int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1320{
1321 set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1322 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1323 set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1324 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1325 return 1;
1326}
1327
1328int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1329{
1330 struct nfs_client *clp = server->nfs_client;
1331
1332 if (!nfs4_valid_open_stateid(state))
1333 return -EBADF;
1334 nfs4_state_mark_reclaim_nograce(clp, state);
1335 dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1336 clp->cl_hostname);
1337 nfs4_schedule_state_manager(clp);
1338 return 0;
1339}
1340EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1341
1342void nfs_inode_find_state_and_recover(struct inode *inode,
1343 const nfs4_stateid *stateid)
1344{
1345 struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1346 struct nfs_inode *nfsi = NFS_I(inode);
1347 struct nfs_open_context *ctx;
1348 struct nfs4_state *state;
1349 bool found = false;
1350
1351 spin_lock(&inode->i_lock);
1352 list_for_each_entry(ctx, &nfsi->open_files, list) {
1353 state = ctx->state;
1354 if (state == NULL)
1355 continue;
1356 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1357 continue;
1358 if (!nfs4_stateid_match(&state->stateid, stateid))
1359 continue;
1360 nfs4_state_mark_reclaim_nograce(clp, state);
1361 found = true;
1362 }
1363 spin_unlock(&inode->i_lock);
1364 if (found)
1365 nfs4_schedule_state_manager(clp);
1366}
1367
1368static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1369{
1370 struct inode *inode = state->inode;
1371 struct nfs_inode *nfsi = NFS_I(inode);
1372 struct nfs_open_context *ctx;
1373
1374 spin_lock(&inode->i_lock);
1375 list_for_each_entry(ctx, &nfsi->open_files, list) {
1376 if (ctx->state != state)
1377 continue;
1378 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1379 }
1380 spin_unlock(&inode->i_lock);
1381}
1382
1383static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1384{
1385 set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1386 nfs4_state_mark_open_context_bad(state);
1387}
1388
1389
1390static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1391{
1392 struct inode *inode = state->inode;
1393 struct nfs_inode *nfsi = NFS_I(inode);
1394 struct file_lock *fl;
1395 int status = 0;
1396
1397 if (inode->i_flock == NULL)
1398 return 0;
1399
1400 /* Guard against delegation returns and new lock/unlock calls */
1401 down_write(&nfsi->rwsem);
1402 /* Protect inode->i_flock using the BKL */
1403 spin_lock(&inode->i_lock);
1404 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1405 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1406 continue;
1407 if (nfs_file_open_context(fl->fl_file)->state != state)
1408 continue;
1409 spin_unlock(&inode->i_lock);
1410 status = ops->recover_lock(state, fl);
1411 switch (status) {
1412 case 0:
1413 break;
1414 case -ESTALE:
1415 case -NFS4ERR_ADMIN_REVOKED:
1416 case -NFS4ERR_STALE_STATEID:
1417 case -NFS4ERR_BAD_STATEID:
1418 case -NFS4ERR_EXPIRED:
1419 case -NFS4ERR_NO_GRACE:
1420 case -NFS4ERR_STALE_CLIENTID:
1421 case -NFS4ERR_BADSESSION:
1422 case -NFS4ERR_BADSLOT:
1423 case -NFS4ERR_BAD_HIGH_SLOT:
1424 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1425 goto out;
1426 default:
1427 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1428 __func__, status);
1429 case -ENOMEM:
1430 case -NFS4ERR_DENIED:
1431 case -NFS4ERR_RECLAIM_BAD:
1432 case -NFS4ERR_RECLAIM_CONFLICT:
1433 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1434 status = 0;
1435 }
1436 spin_lock(&inode->i_lock);
1437 }
1438 spin_unlock(&inode->i_lock);
1439out:
1440 up_write(&nfsi->rwsem);
1441 return status;
1442}
1443
1444static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1445{
1446 struct nfs4_state *state;
1447 struct nfs4_lock_state *lock;
1448 int status = 0;
1449
1450 /* Note: we rely on the sp->so_states list being ordered
1451 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1452 * states first.
1453 * This is needed to ensure that the server won't give us any
1454 * read delegations that we have to return if, say, we are
1455 * recovering after a network partition or a reboot from a
1456 * server that doesn't support a grace period.
1457 */
1458 spin_lock(&sp->so_lock);
1459 write_seqcount_begin(&sp->so_reclaim_seqcount);
1460restart:
1461 list_for_each_entry(state, &sp->so_states, open_states) {
1462 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1463 continue;
1464 if (!nfs4_valid_open_stateid(state))
1465 continue;
1466 if (state->state == 0)
1467 continue;
1468 atomic_inc(&state->count);
1469 spin_unlock(&sp->so_lock);
1470 status = ops->recover_open(sp, state);
1471 if (status >= 0) {
1472 status = nfs4_reclaim_locks(state, ops);
1473 if (status >= 0) {
1474 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1475 spin_lock(&state->state_lock);
1476 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1477 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1478 pr_warn_ratelimited("NFS: "
1479 "%s: Lock reclaim "
1480 "failed!\n", __func__);
1481 }
1482 spin_unlock(&state->state_lock);
1483 }
1484 nfs4_put_open_state(state);
1485 spin_lock(&sp->so_lock);
1486 goto restart;
1487 }
1488 }
1489 switch (status) {
1490 default:
1491 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1492 __func__, status);
1493 case -ENOENT:
1494 case -ENOMEM:
1495 case -ESTALE:
1496 /* Open state on this file cannot be recovered */
1497 nfs4_state_mark_recovery_failed(state, status);
1498 break;
1499 case -EAGAIN:
1500 ssleep(1);
1501 case -NFS4ERR_ADMIN_REVOKED:
1502 case -NFS4ERR_STALE_STATEID:
1503 case -NFS4ERR_BAD_STATEID:
1504 case -NFS4ERR_RECLAIM_BAD:
1505 case -NFS4ERR_RECLAIM_CONFLICT:
1506 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1507 break;
1508 case -NFS4ERR_EXPIRED:
1509 case -NFS4ERR_NO_GRACE:
1510 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1511 case -NFS4ERR_STALE_CLIENTID:
1512 case -NFS4ERR_BADSESSION:
1513 case -NFS4ERR_BADSLOT:
1514 case -NFS4ERR_BAD_HIGH_SLOT:
1515 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1516 goto out_err;
1517 }
1518 nfs4_put_open_state(state);
1519 spin_lock(&sp->so_lock);
1520 goto restart;
1521 }
1522 write_seqcount_end(&sp->so_reclaim_seqcount);
1523 spin_unlock(&sp->so_lock);
1524 return 0;
1525out_err:
1526 nfs4_put_open_state(state);
1527 spin_lock(&sp->so_lock);
1528 write_seqcount_end(&sp->so_reclaim_seqcount);
1529 spin_unlock(&sp->so_lock);
1530 return status;
1531}
1532
1533static void nfs4_clear_open_state(struct nfs4_state *state)
1534{
1535 struct nfs4_lock_state *lock;
1536
1537 clear_bit(NFS_DELEGATED_STATE, &state->flags);
1538 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1539 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1540 clear_bit(NFS_O_RDWR_STATE, &state->flags);
1541 spin_lock(&state->state_lock);
1542 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1543 lock->ls_seqid.flags = 0;
1544 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1545 }
1546 spin_unlock(&state->state_lock);
1547}
1548
1549static void nfs4_reset_seqids(struct nfs_server *server,
1550 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1551{
1552 struct nfs_client *clp = server->nfs_client;
1553 struct nfs4_state_owner *sp;
1554 struct rb_node *pos;
1555 struct nfs4_state *state;
1556
1557 spin_lock(&clp->cl_lock);
1558 for (pos = rb_first(&server->state_owners);
1559 pos != NULL;
1560 pos = rb_next(pos)) {
1561 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1562 sp->so_seqid.flags = 0;
1563 spin_lock(&sp->so_lock);
1564 list_for_each_entry(state, &sp->so_states, open_states) {
1565 if (mark_reclaim(clp, state))
1566 nfs4_clear_open_state(state);
1567 }
1568 spin_unlock(&sp->so_lock);
1569 }
1570 spin_unlock(&clp->cl_lock);
1571}
1572
1573static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1574 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1575{
1576 struct nfs_server *server;
1577
1578 rcu_read_lock();
1579 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1580 nfs4_reset_seqids(server, mark_reclaim);
1581 rcu_read_unlock();
1582}
1583
1584static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1585{
1586 /* Mark all delegations for reclaim */
1587 nfs_delegation_mark_reclaim(clp);
1588 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1589}
1590
1591static void nfs4_reclaim_complete(struct nfs_client *clp,
1592 const struct nfs4_state_recovery_ops *ops,
1593 struct rpc_cred *cred)
1594{
1595 /* Notify the server we're done reclaiming our state */
1596 if (ops->reclaim_complete)
1597 (void)ops->reclaim_complete(clp, cred);
1598}
1599
1600static void nfs4_clear_reclaim_server(struct nfs_server *server)
1601{
1602 struct nfs_client *clp = server->nfs_client;
1603 struct nfs4_state_owner *sp;
1604 struct rb_node *pos;
1605 struct nfs4_state *state;
1606
1607 spin_lock(&clp->cl_lock);
1608 for (pos = rb_first(&server->state_owners);
1609 pos != NULL;
1610 pos = rb_next(pos)) {
1611 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1612 spin_lock(&sp->so_lock);
1613 list_for_each_entry(state, &sp->so_states, open_states) {
1614 if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1615 &state->flags))
1616 continue;
1617 nfs4_state_mark_reclaim_nograce(clp, state);
1618 }
1619 spin_unlock(&sp->so_lock);
1620 }
1621 spin_unlock(&clp->cl_lock);
1622}
1623
1624static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1625{
1626 struct nfs_server *server;
1627
1628 if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1629 return 0;
1630
1631 rcu_read_lock();
1632 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1633 nfs4_clear_reclaim_server(server);
1634 rcu_read_unlock();
1635
1636 nfs_delegation_reap_unclaimed(clp);
1637 return 1;
1638}
1639
1640static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1641{
1642 const struct nfs4_state_recovery_ops *ops;
1643 struct rpc_cred *cred;
1644
1645 if (!nfs4_state_clear_reclaim_reboot(clp))
1646 return;
1647 ops = clp->cl_mvops->reboot_recovery_ops;
1648 cred = nfs4_get_clid_cred(clp);
1649 nfs4_reclaim_complete(clp, ops, cred);
1650 put_rpccred(cred);
1651}
1652
1653static void nfs_delegation_clear_all(struct nfs_client *clp)
1654{
1655 nfs_delegation_mark_reclaim(clp);
1656 nfs_delegation_reap_unclaimed(clp);
1657}
1658
1659static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1660{
1661 nfs_delegation_clear_all(clp);
1662 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1663}
1664
1665static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1666{
1667 switch (error) {
1668 case 0:
1669 break;
1670 case -NFS4ERR_CB_PATH_DOWN:
1671 nfs40_handle_cb_pathdown(clp);
1672 break;
1673 case -NFS4ERR_NO_GRACE:
1674 nfs4_state_end_reclaim_reboot(clp);
1675 break;
1676 case -NFS4ERR_STALE_CLIENTID:
1677 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1678 nfs4_state_clear_reclaim_reboot(clp);
1679 nfs4_state_start_reclaim_reboot(clp);
1680 break;
1681 case -NFS4ERR_EXPIRED:
1682 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1683 nfs4_state_start_reclaim_nograce(clp);
1684 break;
1685 case -NFS4ERR_BADSESSION:
1686 case -NFS4ERR_BADSLOT:
1687 case -NFS4ERR_BAD_HIGH_SLOT:
1688 case -NFS4ERR_DEADSESSION:
1689 case -NFS4ERR_SEQ_FALSE_RETRY:
1690 case -NFS4ERR_SEQ_MISORDERED:
1691 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1692 /* Zero session reset errors */
1693 break;
1694 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1695 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1696 break;
1697 default:
1698 dprintk("%s: failed to handle error %d for server %s\n",
1699 __func__, error, clp->cl_hostname);
1700 return error;
1701 }
1702 dprintk("%s: handled error %d for server %s\n", __func__, error,
1703 clp->cl_hostname);
1704 return 0;
1705}
1706
1707static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1708{
1709 struct nfs4_state_owner *sp;
1710 struct nfs_server *server;
1711 struct rb_node *pos;
1712 int status = 0;
1713
1714restart:
1715 rcu_read_lock();
1716 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1717 nfs4_purge_state_owners(server);
1718 spin_lock(&clp->cl_lock);
1719 for (pos = rb_first(&server->state_owners);
1720 pos != NULL;
1721 pos = rb_next(pos)) {
1722 sp = rb_entry(pos,
1723 struct nfs4_state_owner, so_server_node);
1724 if (!test_and_clear_bit(ops->owner_flag_bit,
1725 &sp->so_flags))
1726 continue;
1727 atomic_inc(&sp->so_count);
1728 spin_unlock(&clp->cl_lock);
1729 rcu_read_unlock();
1730
1731 status = nfs4_reclaim_open_state(sp, ops);
1732 if (status < 0) {
1733 set_bit(ops->owner_flag_bit, &sp->so_flags);
1734 nfs4_put_state_owner(sp);
1735 return nfs4_recovery_handle_error(clp, status);
1736 }
1737
1738 nfs4_put_state_owner(sp);
1739 goto restart;
1740 }
1741 spin_unlock(&clp->cl_lock);
1742 }
1743 rcu_read_unlock();
1744 return status;
1745}
1746
1747static int nfs4_check_lease(struct nfs_client *clp)
1748{
1749 struct rpc_cred *cred;
1750 const struct nfs4_state_maintenance_ops *ops =
1751 clp->cl_mvops->state_renewal_ops;
1752 int status;
1753
1754 /* Is the client already known to have an expired lease? */
1755 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1756 return 0;
1757 spin_lock(&clp->cl_lock);
1758 cred = ops->get_state_renewal_cred_locked(clp);
1759 spin_unlock(&clp->cl_lock);
1760 if (cred == NULL) {
1761 cred = nfs4_get_clid_cred(clp);
1762 status = -ENOKEY;
1763 if (cred == NULL)
1764 goto out;
1765 }
1766 status = ops->renew_lease(clp, cred);
1767 put_rpccred(cred);
1768 if (status == -ETIMEDOUT) {
1769 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1770 return 0;
1771 }
1772out:
1773 return nfs4_recovery_handle_error(clp, status);
1774}
1775
1776/* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1777 * and for recoverable errors on EXCHANGE_ID for v4.1
1778 */
1779static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1780{
1781 switch (status) {
1782 case -NFS4ERR_SEQ_MISORDERED:
1783 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1784 return -ESERVERFAULT;
1785 /* Lease confirmation error: retry after purging the lease */
1786 ssleep(1);
1787 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1788 break;
1789 case -NFS4ERR_STALE_CLIENTID:
1790 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1791 nfs4_state_clear_reclaim_reboot(clp);
1792 nfs4_state_start_reclaim_reboot(clp);
1793 break;
1794 case -NFS4ERR_CLID_INUSE:
1795 pr_err("NFS: Server %s reports our clientid is in use\n",
1796 clp->cl_hostname);
1797 nfs_mark_client_ready(clp, -EPERM);
1798 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1799 return -EPERM;
1800 case -EACCES:
1801 case -NFS4ERR_DELAY:
1802 case -ETIMEDOUT:
1803 case -EAGAIN:
1804 ssleep(1);
1805 break;
1806
1807 case -NFS4ERR_MINOR_VERS_MISMATCH:
1808 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1809 nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1810 dprintk("%s: exit with error %d for server %s\n",
1811 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1812 return -EPROTONOSUPPORT;
1813 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1814 * in nfs4_exchange_id */
1815 default:
1816 dprintk("%s: exit with error %d for server %s\n", __func__,
1817 status, clp->cl_hostname);
1818 return status;
1819 }
1820 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1821 dprintk("%s: handled error %d for server %s\n", __func__, status,
1822 clp->cl_hostname);
1823 return 0;
1824}
1825
1826static int nfs4_establish_lease(struct nfs_client *clp)
1827{
1828 struct rpc_cred *cred;
1829 const struct nfs4_state_recovery_ops *ops =
1830 clp->cl_mvops->reboot_recovery_ops;
1831 int status;
1832
1833 cred = nfs4_get_clid_cred(clp);
1834 if (cred == NULL)
1835 return -ENOENT;
1836 status = ops->establish_clid(clp, cred);
1837 put_rpccred(cred);
1838 if (status != 0)
1839 return status;
1840 pnfs_destroy_all_layouts(clp);
1841 return 0;
1842}
1843
1844/*
1845 * Returns zero or a negative errno. NFS4ERR values are converted
1846 * to local errno values.
1847 */
1848static int nfs4_reclaim_lease(struct nfs_client *clp)
1849{
1850 int status;
1851
1852 status = nfs4_establish_lease(clp);
1853 if (status < 0)
1854 return nfs4_handle_reclaim_lease_error(clp, status);
1855 if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1856 nfs4_state_start_reclaim_nograce(clp);
1857 if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1858 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1859 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1860 clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1861 return 0;
1862}
1863
1864static int nfs4_purge_lease(struct nfs_client *clp)
1865{
1866 int status;
1867
1868 status = nfs4_establish_lease(clp);
1869 if (status < 0)
1870 return nfs4_handle_reclaim_lease_error(clp, status);
1871 clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1872 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1873 nfs4_state_start_reclaim_nograce(clp);
1874 return 0;
1875}
1876
1877/*
1878 * Try remote migration of one FSID from a source server to a
1879 * destination server. The source server provides a list of
1880 * potential destinations.
1881 *
1882 * Returns zero or a negative NFS4ERR status code.
1883 */
1884static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1885{
1886 struct nfs_client *clp = server->nfs_client;
1887 struct nfs4_fs_locations *locations = NULL;
1888 struct inode *inode;
1889 struct page *page;
1890 int status, result;
1891
1892 dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1893 (unsigned long long)server->fsid.major,
1894 (unsigned long long)server->fsid.minor,
1895 clp->cl_hostname);
1896
1897 result = 0;
1898 page = alloc_page(GFP_KERNEL);
1899 locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1900 if (page == NULL || locations == NULL) {
1901 dprintk("<-- %s: no memory\n", __func__);
1902 goto out;
1903 }
1904
1905 inode = server->super->s_root->d_inode;
1906 result = nfs4_proc_get_locations(inode, locations, page, cred);
1907 if (result) {
1908 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1909 __func__, result);
1910 goto out;
1911 }
1912
1913 result = -NFS4ERR_NXIO;
1914 if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1915 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1916 __func__);
1917 goto out;
1918 }
1919
1920 nfs4_begin_drain_session(clp);
1921
1922 status = nfs4_replace_transport(server, locations);
1923 if (status != 0) {
1924 dprintk("<-- %s: failed to replace transport: %d\n",
1925 __func__, status);
1926 goto out;
1927 }
1928
1929 result = 0;
1930 dprintk("<-- %s: migration succeeded\n", __func__);
1931
1932out:
1933 if (page != NULL)
1934 __free_page(page);
1935 kfree(locations);
1936 if (result) {
1937 pr_err("NFS: migration recovery failed (server %s)\n",
1938 clp->cl_hostname);
1939 set_bit(NFS_MIG_FAILED, &server->mig_status);
1940 }
1941 return result;
1942}
1943
1944/*
1945 * Returns zero or a negative NFS4ERR status code.
1946 */
1947static int nfs4_handle_migration(struct nfs_client *clp)
1948{
1949 const struct nfs4_state_maintenance_ops *ops =
1950 clp->cl_mvops->state_renewal_ops;
1951 struct nfs_server *server;
1952 struct rpc_cred *cred;
1953
1954 dprintk("%s: migration reported on \"%s\"\n", __func__,
1955 clp->cl_hostname);
1956
1957 spin_lock(&clp->cl_lock);
1958 cred = ops->get_state_renewal_cred_locked(clp);
1959 spin_unlock(&clp->cl_lock);
1960 if (cred == NULL)
1961 return -NFS4ERR_NOENT;
1962
1963 clp->cl_mig_gen++;
1964restart:
1965 rcu_read_lock();
1966 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1967 int status;
1968
1969 if (server->mig_gen == clp->cl_mig_gen)
1970 continue;
1971 server->mig_gen = clp->cl_mig_gen;
1972
1973 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1974 &server->mig_status))
1975 continue;
1976
1977 rcu_read_unlock();
1978 status = nfs4_try_migration(server, cred);
1979 if (status < 0) {
1980 put_rpccred(cred);
1981 return status;
1982 }
1983 goto restart;
1984 }
1985 rcu_read_unlock();
1986 put_rpccred(cred);
1987 return 0;
1988}
1989
1990/*
1991 * Test each nfs_server on the clp's cl_superblocks list to see
1992 * if it's moved to another server. Stop when the server no longer
1993 * returns NFS4ERR_LEASE_MOVED.
1994 */
1995static int nfs4_handle_lease_moved(struct nfs_client *clp)
1996{
1997 const struct nfs4_state_maintenance_ops *ops =
1998 clp->cl_mvops->state_renewal_ops;
1999 struct nfs_server *server;
2000 struct rpc_cred *cred;
2001
2002 dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2003 clp->cl_hostname);
2004
2005 spin_lock(&clp->cl_lock);
2006 cred = ops->get_state_renewal_cred_locked(clp);
2007 spin_unlock(&clp->cl_lock);
2008 if (cred == NULL)
2009 return -NFS4ERR_NOENT;
2010
2011 clp->cl_mig_gen++;
2012restart:
2013 rcu_read_lock();
2014 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2015 struct inode *inode;
2016 int status;
2017
2018 if (server->mig_gen == clp->cl_mig_gen)
2019 continue;
2020 server->mig_gen = clp->cl_mig_gen;
2021
2022 rcu_read_unlock();
2023
2024 inode = server->super->s_root->d_inode;
2025 status = nfs4_proc_fsid_present(inode, cred);
2026 if (status != -NFS4ERR_MOVED)
2027 goto restart; /* wasn't this one */
2028 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2029 goto restart; /* there are more */
2030 goto out;
2031 }
2032 rcu_read_unlock();
2033
2034out:
2035 put_rpccred(cred);
2036 return 0;
2037}
2038
2039/**
2040 * nfs4_discover_server_trunking - Detect server IP address trunking
2041 *
2042 * @clp: nfs_client under test
2043 * @result: OUT: found nfs_client, or clp
2044 *
2045 * Returns zero or a negative errno. If zero is returned,
2046 * an nfs_client pointer is planted in "result".
2047 *
2048 * Note: since we are invoked in process context, and
2049 * not from inside the state manager, we cannot use
2050 * nfs4_handle_reclaim_lease_error().
2051 */
2052int nfs4_discover_server_trunking(struct nfs_client *clp,
2053 struct nfs_client **result)
2054{
2055 const struct nfs4_state_recovery_ops *ops =
2056 clp->cl_mvops->reboot_recovery_ops;
2057 struct rpc_clnt *clnt;
2058 struct rpc_cred *cred;
2059 int i, status;
2060
2061 dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2062
2063 clnt = clp->cl_rpcclient;
2064 i = 0;
2065
2066 mutex_lock(&nfs_clid_init_mutex);
2067again:
2068 status = -ENOENT;
2069 cred = nfs4_get_clid_cred(clp);
2070 if (cred == NULL)
2071 goto out_unlock;
2072
2073 status = ops->detect_trunking(clp, result, cred);
2074 put_rpccred(cred);
2075 switch (status) {
2076 case 0:
2077 break;
2078 case -ETIMEDOUT:
2079 if (clnt->cl_softrtry)
2080 break;
2081 case -NFS4ERR_DELAY:
2082 case -EAGAIN:
2083 ssleep(1);
2084 case -NFS4ERR_STALE_CLIENTID:
2085 dprintk("NFS: %s after status %d, retrying\n",
2086 __func__, status);
2087 goto again;
2088 case -EACCES:
2089 if (i++ == 0) {
2090 nfs4_root_machine_cred(clp);
2091 goto again;
2092 }
2093 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2094 break;
2095 case -NFS4ERR_CLID_INUSE:
2096 case -NFS4ERR_WRONGSEC:
2097 /* No point in retrying if we already used RPC_AUTH_UNIX */
2098 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2099 status = -EPERM;
2100 break;
2101 }
2102 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2103 if (IS_ERR(clnt)) {
2104 status = PTR_ERR(clnt);
2105 break;
2106 }
2107 /* Note: this is safe because we haven't yet marked the
2108 * client as ready, so we are the only user of
2109 * clp->cl_rpcclient
2110 */
2111 clnt = xchg(&clp->cl_rpcclient, clnt);
2112 rpc_shutdown_client(clnt);
2113 clnt = clp->cl_rpcclient;
2114 goto again;
2115
2116 case -NFS4ERR_MINOR_VERS_MISMATCH:
2117 status = -EPROTONOSUPPORT;
2118 break;
2119
2120 case -EKEYEXPIRED:
2121 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2122 * in nfs4_exchange_id */
2123 status = -EKEYEXPIRED;
2124 break;
2125 default:
2126 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2127 __func__, status);
2128 status = -EIO;
2129 }
2130
2131out_unlock:
2132 mutex_unlock(&nfs_clid_init_mutex);
2133 dprintk("NFS: %s: status = %d\n", __func__, status);
2134 return status;
2135}
2136
2137#ifdef CONFIG_NFS_V4_1
2138void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2139{
2140 struct nfs_client *clp = session->clp;
2141
2142 switch (err) {
2143 default:
2144 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2145 break;
2146 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2147 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2148 }
2149 nfs4_schedule_lease_recovery(clp);
2150}
2151EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2152
2153static void nfs41_ping_server(struct nfs_client *clp)
2154{
2155 /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2156 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2157 nfs4_schedule_state_manager(clp);
2158}
2159
2160void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2161{
2162 nfs41_ping_server(clp);
2163}
2164
2165void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2166{
2167 nfs41_ping_server(clp);
2168}
2169
2170static void nfs4_reset_all_state(struct nfs_client *clp)
2171{
2172 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2173 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2174 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2175 nfs4_state_start_reclaim_nograce(clp);
2176 dprintk("%s: scheduling reset of all state for server %s!\n",
2177 __func__, clp->cl_hostname);
2178 nfs4_schedule_state_manager(clp);
2179 }
2180}
2181
2182static void nfs41_handle_server_reboot(struct nfs_client *clp)
2183{
2184 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2185 nfs4_state_start_reclaim_reboot(clp);
2186 dprintk("%s: server %s rebooted!\n", __func__,
2187 clp->cl_hostname);
2188 nfs4_schedule_state_manager(clp);
2189 }
2190}
2191
2192static void nfs41_handle_state_revoked(struct nfs_client *clp)
2193{
2194 nfs4_reset_all_state(clp);
2195 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2196}
2197
2198static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2199{
2200 /* This will need to handle layouts too */
2201 nfs_expire_all_delegations(clp);
2202 dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2203 clp->cl_hostname);
2204}
2205
2206static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2207{
2208 nfs_expire_all_delegations(clp);
2209 if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2210 nfs4_schedule_state_manager(clp);
2211 dprintk("%s: server %s declared a backchannel fault\n", __func__,
2212 clp->cl_hostname);
2213}
2214
2215static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2216{
2217 if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2218 &clp->cl_state) == 0)
2219 nfs4_schedule_state_manager(clp);
2220}
2221
2222void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2223{
2224 if (!flags)
2225 return;
2226
2227 dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2228 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2229
2230 if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2231 nfs41_handle_server_reboot(clp);
2232 if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2233 SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2234 SEQ4_STATUS_ADMIN_STATE_REVOKED))
2235 nfs41_handle_state_revoked(clp);
2236 if (flags & SEQ4_STATUS_LEASE_MOVED)
2237 nfs4_schedule_lease_moved_recovery(clp);
2238 if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2239 nfs41_handle_recallable_state_revoked(clp);
2240 if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2241 nfs41_handle_backchannel_fault(clp);
2242 else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2243 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2244 nfs41_handle_cb_path_down(clp);
2245}
2246
2247static int nfs4_reset_session(struct nfs_client *clp)
2248{
2249 struct rpc_cred *cred;
2250 int status;
2251
2252 if (!nfs4_has_session(clp))
2253 return 0;
2254 nfs4_begin_drain_session(clp);
2255 cred = nfs4_get_clid_cred(clp);
2256 status = nfs4_proc_destroy_session(clp->cl_session, cred);
2257 switch (status) {
2258 case 0:
2259 case -NFS4ERR_BADSESSION:
2260 case -NFS4ERR_DEADSESSION:
2261 break;
2262 case -NFS4ERR_BACK_CHAN_BUSY:
2263 case -NFS4ERR_DELAY:
2264 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2265 status = 0;
2266 ssleep(1);
2267 goto out;
2268 default:
2269 status = nfs4_recovery_handle_error(clp, status);
2270 goto out;
2271 }
2272
2273 memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2274 status = nfs4_proc_create_session(clp, cred);
2275 if (status) {
2276 dprintk("%s: session reset failed with status %d for server %s!\n",
2277 __func__, status, clp->cl_hostname);
2278 status = nfs4_handle_reclaim_lease_error(clp, status);
2279 goto out;
2280 }
2281 nfs41_finish_session_reset(clp);
2282 dprintk("%s: session reset was successful for server %s!\n",
2283 __func__, clp->cl_hostname);
2284out:
2285 if (cred)
2286 put_rpccred(cred);
2287 return status;
2288}
2289
2290static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2291{
2292 struct rpc_cred *cred;
2293 int ret;
2294
2295 if (!nfs4_has_session(clp))
2296 return 0;
2297 nfs4_begin_drain_session(clp);
2298 cred = nfs4_get_clid_cred(clp);
2299 ret = nfs4_proc_bind_conn_to_session(clp, cred);
2300 if (cred)
2301 put_rpccred(cred);
2302 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2303 switch (ret) {
2304 case 0:
2305 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2306 __func__, clp->cl_hostname);
2307 break;
2308 case -NFS4ERR_DELAY:
2309 ssleep(1);
2310 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2311 break;
2312 default:
2313 return nfs4_recovery_handle_error(clp, ret);
2314 }
2315 return 0;
2316}
2317#else /* CONFIG_NFS_V4_1 */
2318static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2319
2320static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2321{
2322 return 0;
2323}
2324#endif /* CONFIG_NFS_V4_1 */
2325
2326static void nfs4_state_manager(struct nfs_client *clp)
2327{
2328 int status = 0;
2329 const char *section = "", *section_sep = "";
2330
2331 /* Ensure exclusive access to NFSv4 state */
2332 do {
2333 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2334 section = "purge state";
2335 status = nfs4_purge_lease(clp);
2336 if (status < 0)
2337 goto out_error;
2338 continue;
2339 }
2340
2341 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2342 section = "lease expired";
2343 /* We're going to have to re-establish a clientid */
2344 status = nfs4_reclaim_lease(clp);
2345 if (status < 0)
2346 goto out_error;
2347 continue;
2348 }
2349
2350 /* Initialize or reset the session */
2351 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2352 section = "reset session";
2353 status = nfs4_reset_session(clp);
2354 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2355 continue;
2356 if (status < 0)
2357 goto out_error;
2358 }
2359
2360 /* Send BIND_CONN_TO_SESSION */
2361 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2362 &clp->cl_state)) {
2363 section = "bind conn to session";
2364 status = nfs4_bind_conn_to_session(clp);
2365 if (status < 0)
2366 goto out_error;
2367 continue;
2368 }
2369
2370 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2371 section = "check lease";
2372 status = nfs4_check_lease(clp);
2373 if (status < 0)
2374 goto out_error;
2375 }
2376
2377 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2378 section = "migration";
2379 status = nfs4_handle_migration(clp);
2380 if (status < 0)
2381 goto out_error;
2382 }
2383
2384 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2385 section = "lease moved";
2386 status = nfs4_handle_lease_moved(clp);
2387 if (status < 0)
2388 goto out_error;
2389 }
2390
2391 /* First recover reboot state... */
2392 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2393 section = "reclaim reboot";
2394 status = nfs4_do_reclaim(clp,
2395 clp->cl_mvops->reboot_recovery_ops);
2396 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2397 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2398 continue;
2399 nfs4_state_end_reclaim_reboot(clp);
2400 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2401 continue;
2402 if (status < 0)
2403 goto out_error;
2404 }
2405
2406 /* Now recover expired state... */
2407 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2408 section = "reclaim nograce";
2409 status = nfs4_do_reclaim(clp,
2410 clp->cl_mvops->nograce_recovery_ops);
2411 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2412 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2413 test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2414 continue;
2415 if (status < 0)
2416 goto out_error;
2417 }
2418
2419 nfs4_end_drain_session(clp);
2420 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2421 nfs_client_return_marked_delegations(clp);
2422 continue;
2423 }
2424
2425 nfs4_clear_state_manager_bit(clp);
2426 /* Did we race with an attempt to give us more work? */
2427 if (clp->cl_state == 0)
2428 break;
2429 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2430 break;
2431 } while (atomic_read(&clp->cl_count) > 1);
2432 return;
2433out_error:
2434 if (strlen(section))
2435 section_sep = ": ";
2436 pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2437 " with error %d\n", section_sep, section,
2438 clp->cl_hostname, -status);
2439 ssleep(1);
2440 nfs4_end_drain_session(clp);
2441 nfs4_clear_state_manager_bit(clp);
2442}
2443
2444static int nfs4_run_state_manager(void *ptr)
2445{
2446 struct nfs_client *clp = ptr;
2447
2448 allow_signal(SIGKILL);
2449 nfs4_state_manager(clp);
2450 nfs_put_client(clp);
2451 module_put_and_exit(0);
2452 return 0;
2453}
2454
2455/*
2456 * Local variables:
2457 * c-basic-offset: 8
2458 * End:
2459 */