xref: /openbmc/linux/fs/nfs/nfs4state.c (revision ff4a7481c3898ffc3cc271d6aca431d190c37247)
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 
68 const nfs4_stateid zero_stateid = {
69 	{ .data = { 0 } },
70 	.type = NFS4_SPECIAL_STATEID_TYPE,
71 };
72 const 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 
80 const nfs4_stateid current_stateid = {
81 	{
82 		/* Funky initialiser keeps older gcc versions happy */
83 		.data = { 0x0, 0x0, 0x0, 0x1, 0 },
84 	},
85 	.type = NFS4_SPECIAL_STATEID_TYPE,
86 };
87 
88 static DEFINE_MUTEX(nfs_clid_init_mutex);
89 
90 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
91 {
92 	struct nfs4_setclientid_res clid = {
93 		.clientid = clp->cl_clientid,
94 		.confirm = clp->cl_confirm,
95 	};
96 	unsigned short port;
97 	int status;
98 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
99 
100 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
101 		goto do_confirm;
102 	port = nn->nfs_callback_tcpport;
103 	if (clp->cl_addr.ss_family == AF_INET6)
104 		port = nn->nfs_callback_tcpport6;
105 
106 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
107 	if (status != 0)
108 		goto out;
109 	clp->cl_clientid = clid.clientid;
110 	clp->cl_confirm = clid.confirm;
111 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
112 do_confirm:
113 	status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
114 	if (status != 0)
115 		goto out;
116 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
117 	nfs4_schedule_state_renewal(clp);
118 out:
119 	return status;
120 }
121 
122 /**
123  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
124  *
125  * @clp: nfs_client under test
126  * @result: OUT: found nfs_client, or clp
127  * @cred: credential to use for trunking test
128  *
129  * Returns zero, a negative errno, or a negative NFS4ERR status.
130  * If zero is returned, an nfs_client pointer is planted in
131  * "result".
132  *
133  * Note: The returned client may not yet be marked ready.
134  */
135 int nfs40_discover_server_trunking(struct nfs_client *clp,
136 				   struct nfs_client **result,
137 				   struct rpc_cred *cred)
138 {
139 	struct nfs4_setclientid_res clid = {
140 		.clientid = clp->cl_clientid,
141 		.confirm = clp->cl_confirm,
142 	};
143 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
144 	unsigned short port;
145 	int status;
146 
147 	port = nn->nfs_callback_tcpport;
148 	if (clp->cl_addr.ss_family == AF_INET6)
149 		port = nn->nfs_callback_tcpport6;
150 
151 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
152 	if (status != 0)
153 		goto out;
154 	clp->cl_clientid = clid.clientid;
155 	clp->cl_confirm = clid.confirm;
156 
157 	status = nfs40_walk_client_list(clp, result, cred);
158 	if (status == 0) {
159 		/* Sustain the lease, even if it's empty.  If the clientid4
160 		 * goes stale it's of no use for trunking discovery. */
161 		nfs4_schedule_state_renewal(*result);
162 	}
163 out:
164 	return status;
165 }
166 
167 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
168 {
169 	struct rpc_cred *cred = NULL;
170 
171 	if (clp->cl_machine_cred != NULL)
172 		cred = get_rpccred(clp->cl_machine_cred);
173 	return cred;
174 }
175 
176 static void nfs4_root_machine_cred(struct nfs_client *clp)
177 {
178 	struct rpc_cred *cred, *new;
179 
180 	new = rpc_lookup_machine_cred(NULL);
181 	spin_lock(&clp->cl_lock);
182 	cred = clp->cl_machine_cred;
183 	clp->cl_machine_cred = new;
184 	spin_unlock(&clp->cl_lock);
185 	if (cred != NULL)
186 		put_rpccred(cred);
187 }
188 
189 static struct rpc_cred *
190 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
191 {
192 	struct rpc_cred *cred = NULL;
193 	struct nfs4_state_owner *sp;
194 	struct rb_node *pos;
195 
196 	for (pos = rb_first(&server->state_owners);
197 	     pos != NULL;
198 	     pos = rb_next(pos)) {
199 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
200 		if (list_empty(&sp->so_states))
201 			continue;
202 		cred = get_rpccred(sp->so_cred);
203 		break;
204 	}
205 	return cred;
206 }
207 
208 /**
209  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
210  * @clp: client state handle
211  *
212  * Returns an rpc_cred with reference count bumped, or NULL.
213  * Caller must hold clp->cl_lock.
214  */
215 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
216 {
217 	struct rpc_cred *cred = NULL;
218 	struct nfs_server *server;
219 
220 	/* Use machine credentials if available */
221 	cred = nfs4_get_machine_cred_locked(clp);
222 	if (cred != NULL)
223 		goto out;
224 
225 	rcu_read_lock();
226 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
227 		cred = nfs4_get_renew_cred_server_locked(server);
228 		if (cred != NULL)
229 			break;
230 	}
231 	rcu_read_unlock();
232 
233 out:
234 	return cred;
235 }
236 
237 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
238 {
239 	if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
240 		spin_lock(&tbl->slot_tbl_lock);
241 		nfs41_wake_slot_table(tbl);
242 		spin_unlock(&tbl->slot_tbl_lock);
243 	}
244 }
245 
246 static void nfs4_end_drain_session(struct nfs_client *clp)
247 {
248 	struct nfs4_session *ses = clp->cl_session;
249 
250 	if (clp->cl_slot_tbl) {
251 		nfs4_end_drain_slot_table(clp->cl_slot_tbl);
252 		return;
253 	}
254 
255 	if (ses != NULL) {
256 		nfs4_end_drain_slot_table(&ses->bc_slot_table);
257 		nfs4_end_drain_slot_table(&ses->fc_slot_table);
258 	}
259 }
260 
261 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
262 {
263 	set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
264 	spin_lock(&tbl->slot_tbl_lock);
265 	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
266 		reinit_completion(&tbl->complete);
267 		spin_unlock(&tbl->slot_tbl_lock);
268 		return wait_for_completion_interruptible(&tbl->complete);
269 	}
270 	spin_unlock(&tbl->slot_tbl_lock);
271 	return 0;
272 }
273 
274 static int nfs4_begin_drain_session(struct nfs_client *clp)
275 {
276 	struct nfs4_session *ses = clp->cl_session;
277 	int ret;
278 
279 	if (clp->cl_slot_tbl)
280 		return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
281 
282 	/* back channel */
283 	ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
284 	if (ret)
285 		return ret;
286 	/* fore channel */
287 	return nfs4_drain_slot_tbl(&ses->fc_slot_table);
288 }
289 
290 #if defined(CONFIG_NFS_V4_1)
291 
292 static int nfs41_setup_state_renewal(struct nfs_client *clp)
293 {
294 	int status;
295 	struct nfs_fsinfo fsinfo;
296 	unsigned long now;
297 
298 	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
299 		nfs4_schedule_state_renewal(clp);
300 		return 0;
301 	}
302 
303 	now = jiffies;
304 	status = nfs4_proc_get_lease_time(clp, &fsinfo);
305 	if (status == 0) {
306 		nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
307 		nfs4_schedule_state_renewal(clp);
308 	}
309 
310 	return status;
311 }
312 
313 static void nfs41_finish_session_reset(struct nfs_client *clp)
314 {
315 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
316 	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
317 	/* create_session negotiated new slot table */
318 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
319 	nfs41_setup_state_renewal(clp);
320 }
321 
322 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
323 {
324 	int status;
325 
326 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
327 		goto do_confirm;
328 	status = nfs4_proc_exchange_id(clp, cred);
329 	if (status != 0)
330 		goto out;
331 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
332 do_confirm:
333 	status = nfs4_proc_create_session(clp, cred);
334 	if (status != 0)
335 		goto out;
336 	nfs41_finish_session_reset(clp);
337 	nfs_mark_client_ready(clp, NFS_CS_READY);
338 out:
339 	return status;
340 }
341 
342 /**
343  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
344  *
345  * @clp: nfs_client under test
346  * @result: OUT: found nfs_client, or clp
347  * @cred: credential to use for trunking test
348  *
349  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
350  * If NFS4_OK is returned, an nfs_client pointer is planted in
351  * "result".
352  *
353  * Note: The returned client may not yet be marked ready.
354  */
355 int nfs41_discover_server_trunking(struct nfs_client *clp,
356 				   struct nfs_client **result,
357 				   struct rpc_cred *cred)
358 {
359 	int status;
360 
361 	status = nfs4_proc_exchange_id(clp, cred);
362 	if (status != NFS4_OK)
363 		return status;
364 
365 	status = nfs41_walk_client_list(clp, result, cred);
366 	if (status < 0)
367 		return status;
368 	if (clp != *result)
369 		return 0;
370 
371 	/*
372 	 * Purge state if the client id was established in a prior
373 	 * instance and the client id could not have arrived on the
374 	 * server via Transparent State Migration.
375 	 */
376 	if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
377 		if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
378 			set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
379 		else
380 			set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
381 	}
382 	nfs4_schedule_state_manager(clp);
383 	status = nfs_wait_client_init_complete(clp);
384 	if (status < 0)
385 		nfs_put_client(clp);
386 	return status;
387 }
388 
389 #endif /* CONFIG_NFS_V4_1 */
390 
391 /**
392  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
393  * @clp: client state handle
394  *
395  * Returns an rpc_cred with reference count bumped, or NULL.
396  */
397 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
398 {
399 	struct rpc_cred *cred;
400 
401 	spin_lock(&clp->cl_lock);
402 	cred = nfs4_get_machine_cred_locked(clp);
403 	spin_unlock(&clp->cl_lock);
404 	return cred;
405 }
406 
407 static struct nfs4_state_owner *
408 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
409 {
410 	struct rb_node **p = &server->state_owners.rb_node,
411 		       *parent = NULL;
412 	struct nfs4_state_owner *sp;
413 
414 	while (*p != NULL) {
415 		parent = *p;
416 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
417 
418 		if (cred < sp->so_cred)
419 			p = &parent->rb_left;
420 		else if (cred > sp->so_cred)
421 			p = &parent->rb_right;
422 		else {
423 			if (!list_empty(&sp->so_lru))
424 				list_del_init(&sp->so_lru);
425 			atomic_inc(&sp->so_count);
426 			return sp;
427 		}
428 	}
429 	return NULL;
430 }
431 
432 static struct nfs4_state_owner *
433 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
434 {
435 	struct nfs_server *server = new->so_server;
436 	struct rb_node **p = &server->state_owners.rb_node,
437 		       *parent = NULL;
438 	struct nfs4_state_owner *sp;
439 
440 	while (*p != NULL) {
441 		parent = *p;
442 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
443 
444 		if (new->so_cred < sp->so_cred)
445 			p = &parent->rb_left;
446 		else if (new->so_cred > sp->so_cred)
447 			p = &parent->rb_right;
448 		else {
449 			if (!list_empty(&sp->so_lru))
450 				list_del_init(&sp->so_lru);
451 			atomic_inc(&sp->so_count);
452 			return sp;
453 		}
454 	}
455 	rb_link_node(&new->so_server_node, parent, p);
456 	rb_insert_color(&new->so_server_node, &server->state_owners);
457 	return new;
458 }
459 
460 static void
461 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
462 {
463 	struct nfs_server *server = sp->so_server;
464 
465 	if (!RB_EMPTY_NODE(&sp->so_server_node))
466 		rb_erase(&sp->so_server_node, &server->state_owners);
467 }
468 
469 static void
470 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
471 {
472 	sc->create_time = ktime_get();
473 	sc->flags = 0;
474 	sc->counter = 0;
475 	spin_lock_init(&sc->lock);
476 	INIT_LIST_HEAD(&sc->list);
477 	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
478 }
479 
480 static void
481 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
482 {
483 	rpc_destroy_wait_queue(&sc->wait);
484 }
485 
486 /*
487  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
488  * create a new state_owner.
489  *
490  */
491 static struct nfs4_state_owner *
492 nfs4_alloc_state_owner(struct nfs_server *server,
493 		struct rpc_cred *cred,
494 		gfp_t gfp_flags)
495 {
496 	struct nfs4_state_owner *sp;
497 
498 	sp = kzalloc(sizeof(*sp), gfp_flags);
499 	if (!sp)
500 		return NULL;
501 	sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
502 						gfp_flags);
503 	if (sp->so_seqid.owner_id < 0) {
504 		kfree(sp);
505 		return NULL;
506 	}
507 	sp->so_server = server;
508 	sp->so_cred = get_rpccred(cred);
509 	spin_lock_init(&sp->so_lock);
510 	INIT_LIST_HEAD(&sp->so_states);
511 	nfs4_init_seqid_counter(&sp->so_seqid);
512 	atomic_set(&sp->so_count, 1);
513 	INIT_LIST_HEAD(&sp->so_lru);
514 	seqcount_init(&sp->so_reclaim_seqcount);
515 	mutex_init(&sp->so_delegreturn_mutex);
516 	return sp;
517 }
518 
519 static void
520 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
521 {
522 	/* This state_owner is no longer usable, but must
523 	 * remain in place so that state recovery can find it
524 	 * and the opens associated with it.
525 	 * It may also be used for new 'open' request to
526 	 * return a delegation to the server.
527 	 * So update the 'create_time' so that it looks like
528 	 * a new state_owner.  This will cause the server to
529 	 * request an OPEN_CONFIRM to start a new sequence.
530 	 */
531 	sp->so_seqid.create_time = ktime_get();
532 }
533 
534 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
535 {
536 	nfs4_destroy_seqid_counter(&sp->so_seqid);
537 	put_rpccred(sp->so_cred);
538 	ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
539 	kfree(sp);
540 }
541 
542 static void nfs4_gc_state_owners(struct nfs_server *server)
543 {
544 	struct nfs_client *clp = server->nfs_client;
545 	struct nfs4_state_owner *sp, *tmp;
546 	unsigned long time_min, time_max;
547 	LIST_HEAD(doomed);
548 
549 	spin_lock(&clp->cl_lock);
550 	time_max = jiffies;
551 	time_min = (long)time_max - (long)clp->cl_lease_time;
552 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
553 		/* NB: LRU is sorted so that oldest is at the head */
554 		if (time_in_range(sp->so_expires, time_min, time_max))
555 			break;
556 		list_move(&sp->so_lru, &doomed);
557 		nfs4_remove_state_owner_locked(sp);
558 	}
559 	spin_unlock(&clp->cl_lock);
560 
561 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
562 		list_del(&sp->so_lru);
563 		nfs4_free_state_owner(sp);
564 	}
565 }
566 
567 /**
568  * nfs4_get_state_owner - Look up a state owner given a credential
569  * @server: nfs_server to search
570  * @cred: RPC credential to match
571  *
572  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
573  */
574 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
575 					      struct rpc_cred *cred,
576 					      gfp_t gfp_flags)
577 {
578 	struct nfs_client *clp = server->nfs_client;
579 	struct nfs4_state_owner *sp, *new;
580 
581 	spin_lock(&clp->cl_lock);
582 	sp = nfs4_find_state_owner_locked(server, cred);
583 	spin_unlock(&clp->cl_lock);
584 	if (sp != NULL)
585 		goto out;
586 	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
587 	if (new == NULL)
588 		goto out;
589 	spin_lock(&clp->cl_lock);
590 	sp = nfs4_insert_state_owner_locked(new);
591 	spin_unlock(&clp->cl_lock);
592 	if (sp != new)
593 		nfs4_free_state_owner(new);
594 out:
595 	nfs4_gc_state_owners(server);
596 	return sp;
597 }
598 
599 /**
600  * nfs4_put_state_owner - Release a nfs4_state_owner
601  * @sp: state owner data to release
602  *
603  * Note that we keep released state owners on an LRU
604  * list.
605  * This caches valid state owners so that they can be
606  * reused, to avoid the OPEN_CONFIRM on minor version 0.
607  * It also pins the uniquifier of dropped state owners for
608  * a while, to ensure that those state owner names are
609  * never reused.
610  */
611 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
612 {
613 	struct nfs_server *server = sp->so_server;
614 	struct nfs_client *clp = server->nfs_client;
615 
616 	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
617 		return;
618 
619 	sp->so_expires = jiffies;
620 	list_add_tail(&sp->so_lru, &server->state_owners_lru);
621 	spin_unlock(&clp->cl_lock);
622 }
623 
624 /**
625  * nfs4_purge_state_owners - Release all cached state owners
626  * @server: nfs_server with cached state owners to release
627  *
628  * Called at umount time.  Remaining state owners will be on
629  * the LRU with ref count of zero.
630  */
631 void nfs4_purge_state_owners(struct nfs_server *server)
632 {
633 	struct nfs_client *clp = server->nfs_client;
634 	struct nfs4_state_owner *sp, *tmp;
635 	LIST_HEAD(doomed);
636 
637 	spin_lock(&clp->cl_lock);
638 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
639 		list_move(&sp->so_lru, &doomed);
640 		nfs4_remove_state_owner_locked(sp);
641 	}
642 	spin_unlock(&clp->cl_lock);
643 
644 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
645 		list_del(&sp->so_lru);
646 		nfs4_free_state_owner(sp);
647 	}
648 }
649 
650 static struct nfs4_state *
651 nfs4_alloc_open_state(void)
652 {
653 	struct nfs4_state *state;
654 
655 	state = kzalloc(sizeof(*state), GFP_NOFS);
656 	if (!state)
657 		return NULL;
658 	refcount_set(&state->count, 1);
659 	INIT_LIST_HEAD(&state->lock_states);
660 	spin_lock_init(&state->state_lock);
661 	seqlock_init(&state->seqlock);
662 	init_waitqueue_head(&state->waitq);
663 	return state;
664 }
665 
666 void
667 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
668 {
669 	if (state->state == fmode)
670 		return;
671 	/* NB! List reordering - see the reclaim code for why.  */
672 	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
673 		if (fmode & FMODE_WRITE)
674 			list_move(&state->open_states, &state->owner->so_states);
675 		else
676 			list_move_tail(&state->open_states, &state->owner->so_states);
677 	}
678 	state->state = fmode;
679 }
680 
681 static struct nfs4_state *
682 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
683 {
684 	struct nfs_inode *nfsi = NFS_I(inode);
685 	struct nfs4_state *state;
686 
687 	list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
688 		if (state->owner != owner)
689 			continue;
690 		if (!nfs4_valid_open_stateid(state))
691 			continue;
692 		if (refcount_inc_not_zero(&state->count))
693 			return state;
694 	}
695 	return NULL;
696 }
697 
698 static void
699 nfs4_free_open_state(struct nfs4_state *state)
700 {
701 	kfree_rcu(state, rcu_head);
702 }
703 
704 struct nfs4_state *
705 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
706 {
707 	struct nfs4_state *state, *new;
708 	struct nfs_inode *nfsi = NFS_I(inode);
709 
710 	rcu_read_lock();
711 	state = __nfs4_find_state_byowner(inode, owner);
712 	rcu_read_unlock();
713 	if (state)
714 		goto out;
715 	new = nfs4_alloc_open_state();
716 	spin_lock(&owner->so_lock);
717 	spin_lock(&inode->i_lock);
718 	state = __nfs4_find_state_byowner(inode, owner);
719 	if (state == NULL && new != NULL) {
720 		state = new;
721 		state->owner = owner;
722 		atomic_inc(&owner->so_count);
723 		list_add_rcu(&state->inode_states, &nfsi->open_states);
724 		ihold(inode);
725 		state->inode = inode;
726 		spin_unlock(&inode->i_lock);
727 		/* Note: The reclaim code dictates that we add stateless
728 		 * and read-only stateids to the end of the list */
729 		list_add_tail(&state->open_states, &owner->so_states);
730 		spin_unlock(&owner->so_lock);
731 	} else {
732 		spin_unlock(&inode->i_lock);
733 		spin_unlock(&owner->so_lock);
734 		if (new)
735 			nfs4_free_open_state(new);
736 	}
737 out:
738 	return state;
739 }
740 
741 void nfs4_put_open_state(struct nfs4_state *state)
742 {
743 	struct inode *inode = state->inode;
744 	struct nfs4_state_owner *owner = state->owner;
745 
746 	if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
747 		return;
748 	spin_lock(&inode->i_lock);
749 	list_del_rcu(&state->inode_states);
750 	list_del(&state->open_states);
751 	spin_unlock(&inode->i_lock);
752 	spin_unlock(&owner->so_lock);
753 	iput(inode);
754 	nfs4_free_open_state(state);
755 	nfs4_put_state_owner(owner);
756 }
757 
758 /*
759  * Close the current file.
760  */
761 static void __nfs4_close(struct nfs4_state *state,
762 		fmode_t fmode, gfp_t gfp_mask, int wait)
763 {
764 	struct nfs4_state_owner *owner = state->owner;
765 	int call_close = 0;
766 	fmode_t newstate;
767 
768 	atomic_inc(&owner->so_count);
769 	/* Protect against nfs4_find_state() */
770 	spin_lock(&owner->so_lock);
771 	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
772 		case FMODE_READ:
773 			state->n_rdonly--;
774 			break;
775 		case FMODE_WRITE:
776 			state->n_wronly--;
777 			break;
778 		case FMODE_READ|FMODE_WRITE:
779 			state->n_rdwr--;
780 	}
781 	newstate = FMODE_READ|FMODE_WRITE;
782 	if (state->n_rdwr == 0) {
783 		if (state->n_rdonly == 0) {
784 			newstate &= ~FMODE_READ;
785 			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
786 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
787 		}
788 		if (state->n_wronly == 0) {
789 			newstate &= ~FMODE_WRITE;
790 			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
791 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
792 		}
793 		if (newstate == 0)
794 			clear_bit(NFS_DELEGATED_STATE, &state->flags);
795 	}
796 	nfs4_state_set_mode_locked(state, newstate);
797 	spin_unlock(&owner->so_lock);
798 
799 	if (!call_close) {
800 		nfs4_put_open_state(state);
801 		nfs4_put_state_owner(owner);
802 	} else
803 		nfs4_do_close(state, gfp_mask, wait);
804 }
805 
806 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
807 {
808 	__nfs4_close(state, fmode, GFP_NOFS, 0);
809 }
810 
811 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
812 {
813 	__nfs4_close(state, fmode, GFP_KERNEL, 1);
814 }
815 
816 /*
817  * Search the state->lock_states for an existing lock_owner
818  * that is compatible with either of the given owners.
819  * If the second is non-zero, then the first refers to a Posix-lock
820  * owner (current->files) and the second refers to a flock/OFD
821  * owner (struct file*).  In that case, prefer a match for the first
822  * owner.
823  * If both sorts of locks are held on the one file we cannot know
824  * which stateid was intended to be used, so a "correct" choice cannot
825  * be made.  Failing that, a "consistent" choice is preferable.  The
826  * consistent choice we make is to prefer the first owner, that of a
827  * Posix lock.
828  */
829 static struct nfs4_lock_state *
830 __nfs4_find_lock_state(struct nfs4_state *state,
831 		       fl_owner_t fl_owner, fl_owner_t fl_owner2)
832 {
833 	struct nfs4_lock_state *pos, *ret = NULL;
834 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
835 		if (pos->ls_owner == fl_owner) {
836 			ret = pos;
837 			break;
838 		}
839 		if (pos->ls_owner == fl_owner2)
840 			ret = pos;
841 	}
842 	if (ret)
843 		refcount_inc(&ret->ls_count);
844 	return ret;
845 }
846 
847 /*
848  * Return a compatible lock_state. If no initialized lock_state structure
849  * exists, return an uninitialized one.
850  *
851  */
852 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
853 {
854 	struct nfs4_lock_state *lsp;
855 	struct nfs_server *server = state->owner->so_server;
856 
857 	lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
858 	if (lsp == NULL)
859 		return NULL;
860 	nfs4_init_seqid_counter(&lsp->ls_seqid);
861 	refcount_set(&lsp->ls_count, 1);
862 	lsp->ls_state = state;
863 	lsp->ls_owner = fl_owner;
864 	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
865 	if (lsp->ls_seqid.owner_id < 0)
866 		goto out_free;
867 	INIT_LIST_HEAD(&lsp->ls_locks);
868 	return lsp;
869 out_free:
870 	kfree(lsp);
871 	return NULL;
872 }
873 
874 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
875 {
876 	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
877 	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
878 	kfree(lsp);
879 }
880 
881 /*
882  * Return a compatible lock_state. If no initialized lock_state structure
883  * exists, return an uninitialized one.
884  *
885  */
886 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
887 {
888 	struct nfs4_lock_state *lsp, *new = NULL;
889 
890 	for(;;) {
891 		spin_lock(&state->state_lock);
892 		lsp = __nfs4_find_lock_state(state, owner, NULL);
893 		if (lsp != NULL)
894 			break;
895 		if (new != NULL) {
896 			list_add(&new->ls_locks, &state->lock_states);
897 			set_bit(LK_STATE_IN_USE, &state->flags);
898 			lsp = new;
899 			new = NULL;
900 			break;
901 		}
902 		spin_unlock(&state->state_lock);
903 		new = nfs4_alloc_lock_state(state, owner);
904 		if (new == NULL)
905 			return NULL;
906 	}
907 	spin_unlock(&state->state_lock);
908 	if (new != NULL)
909 		nfs4_free_lock_state(state->owner->so_server, new);
910 	return lsp;
911 }
912 
913 /*
914  * Release reference to lock_state, and free it if we see that
915  * it is no longer in use
916  */
917 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
918 {
919 	struct nfs_server *server;
920 	struct nfs4_state *state;
921 
922 	if (lsp == NULL)
923 		return;
924 	state = lsp->ls_state;
925 	if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
926 		return;
927 	list_del(&lsp->ls_locks);
928 	if (list_empty(&state->lock_states))
929 		clear_bit(LK_STATE_IN_USE, &state->flags);
930 	spin_unlock(&state->state_lock);
931 	server = state->owner->so_server;
932 	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
933 		struct nfs_client *clp = server->nfs_client;
934 
935 		clp->cl_mvops->free_lock_state(server, lsp);
936 	} else
937 		nfs4_free_lock_state(server, lsp);
938 }
939 
940 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
941 {
942 	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
943 
944 	dst->fl_u.nfs4_fl.owner = lsp;
945 	refcount_inc(&lsp->ls_count);
946 }
947 
948 static void nfs4_fl_release_lock(struct file_lock *fl)
949 {
950 	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
951 }
952 
953 static const struct file_lock_operations nfs4_fl_lock_ops = {
954 	.fl_copy_lock = nfs4_fl_copy_lock,
955 	.fl_release_private = nfs4_fl_release_lock,
956 };
957 
958 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
959 {
960 	struct nfs4_lock_state *lsp;
961 
962 	if (fl->fl_ops != NULL)
963 		return 0;
964 	lsp = nfs4_get_lock_state(state, fl->fl_owner);
965 	if (lsp == NULL)
966 		return -ENOMEM;
967 	fl->fl_u.nfs4_fl.owner = lsp;
968 	fl->fl_ops = &nfs4_fl_lock_ops;
969 	return 0;
970 }
971 
972 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
973 		struct nfs4_state *state,
974 		const struct nfs_lock_context *l_ctx)
975 {
976 	struct nfs4_lock_state *lsp;
977 	fl_owner_t fl_owner, fl_flock_owner;
978 	int ret = -ENOENT;
979 
980 	if (l_ctx == NULL)
981 		goto out;
982 
983 	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
984 		goto out;
985 
986 	fl_owner = l_ctx->lockowner;
987 	fl_flock_owner = l_ctx->open_context->flock_owner;
988 
989 	spin_lock(&state->state_lock);
990 	lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
991 	if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
992 		ret = -EIO;
993 	else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
994 		nfs4_stateid_copy(dst, &lsp->ls_stateid);
995 		ret = 0;
996 	}
997 	spin_unlock(&state->state_lock);
998 	nfs4_put_lock_state(lsp);
999 out:
1000 	return ret;
1001 }
1002 
1003 bool nfs4_refresh_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1004 {
1005 	bool ret;
1006 	int seq;
1007 
1008 	do {
1009 		ret = false;
1010 		seq = read_seqbegin(&state->seqlock);
1011 		if (nfs4_state_match_open_stateid_other(state, dst)) {
1012 			dst->seqid = state->open_stateid.seqid;
1013 			ret = true;
1014 		}
1015 	} while (read_seqretry(&state->seqlock, seq));
1016 	return ret;
1017 }
1018 
1019 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1020 {
1021 	bool ret;
1022 	const nfs4_stateid *src;
1023 	int seq;
1024 
1025 	do {
1026 		ret = false;
1027 		src = &zero_stateid;
1028 		seq = read_seqbegin(&state->seqlock);
1029 		if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1030 			src = &state->open_stateid;
1031 			ret = true;
1032 		}
1033 		nfs4_stateid_copy(dst, src);
1034 	} while (read_seqretry(&state->seqlock, seq));
1035 	return ret;
1036 }
1037 
1038 /*
1039  * Byte-range lock aware utility to initialize the stateid of read/write
1040  * requests.
1041  */
1042 int nfs4_select_rw_stateid(struct nfs4_state *state,
1043 		fmode_t fmode, const struct nfs_lock_context *l_ctx,
1044 		nfs4_stateid *dst, struct rpc_cred **cred)
1045 {
1046 	int ret;
1047 
1048 	if (!nfs4_valid_open_stateid(state))
1049 		return -EIO;
1050 	if (cred != NULL)
1051 		*cred = NULL;
1052 	ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1053 	if (ret == -EIO)
1054 		/* A lost lock - don't even consider delegations */
1055 		goto out;
1056 	/* returns true if delegation stateid found and copied */
1057 	if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1058 		ret = 0;
1059 		goto out;
1060 	}
1061 	if (ret != -ENOENT)
1062 		/* nfs4_copy_delegation_stateid() didn't over-write
1063 		 * dst, so it still has the lock stateid which we now
1064 		 * choose to use.
1065 		 */
1066 		goto out;
1067 	nfs4_copy_open_stateid(dst, state);
1068 	ret = 0;
1069 out:
1070 	if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1071 		dst->seqid = 0;
1072 	return ret;
1073 }
1074 
1075 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1076 {
1077 	struct nfs_seqid *new;
1078 
1079 	new = kmalloc(sizeof(*new), gfp_mask);
1080 	if (new == NULL)
1081 		return ERR_PTR(-ENOMEM);
1082 	new->sequence = counter;
1083 	INIT_LIST_HEAD(&new->list);
1084 	new->task = NULL;
1085 	return new;
1086 }
1087 
1088 void nfs_release_seqid(struct nfs_seqid *seqid)
1089 {
1090 	struct nfs_seqid_counter *sequence;
1091 
1092 	if (seqid == NULL || list_empty(&seqid->list))
1093 		return;
1094 	sequence = seqid->sequence;
1095 	spin_lock(&sequence->lock);
1096 	list_del_init(&seqid->list);
1097 	if (!list_empty(&sequence->list)) {
1098 		struct nfs_seqid *next;
1099 
1100 		next = list_first_entry(&sequence->list,
1101 				struct nfs_seqid, list);
1102 		rpc_wake_up_queued_task(&sequence->wait, next->task);
1103 	}
1104 	spin_unlock(&sequence->lock);
1105 }
1106 
1107 void nfs_free_seqid(struct nfs_seqid *seqid)
1108 {
1109 	nfs_release_seqid(seqid);
1110 	kfree(seqid);
1111 }
1112 
1113 /*
1114  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1115  * failed with a seqid incrementing error -
1116  * see comments nfs4.h:seqid_mutating_error()
1117  */
1118 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1119 {
1120 	switch (status) {
1121 		case 0:
1122 			break;
1123 		case -NFS4ERR_BAD_SEQID:
1124 			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1125 				return;
1126 			pr_warn_ratelimited("NFS: v4 server returned a bad"
1127 					" sequence-id error on an"
1128 					" unconfirmed sequence %p!\n",
1129 					seqid->sequence);
1130 		case -NFS4ERR_STALE_CLIENTID:
1131 		case -NFS4ERR_STALE_STATEID:
1132 		case -NFS4ERR_BAD_STATEID:
1133 		case -NFS4ERR_BADXDR:
1134 		case -NFS4ERR_RESOURCE:
1135 		case -NFS4ERR_NOFILEHANDLE:
1136 		case -NFS4ERR_MOVED:
1137 			/* Non-seqid mutating errors */
1138 			return;
1139 	};
1140 	/*
1141 	 * Note: no locking needed as we are guaranteed to be first
1142 	 * on the sequence list
1143 	 */
1144 	seqid->sequence->counter++;
1145 }
1146 
1147 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1148 {
1149 	struct nfs4_state_owner *sp;
1150 
1151 	if (seqid == NULL)
1152 		return;
1153 
1154 	sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1155 	if (status == -NFS4ERR_BAD_SEQID)
1156 		nfs4_reset_state_owner(sp);
1157 	if (!nfs4_has_session(sp->so_server->nfs_client))
1158 		nfs_increment_seqid(status, seqid);
1159 }
1160 
1161 /*
1162  * Increment the seqid if the LOCK/LOCKU succeeded, or
1163  * failed with a seqid incrementing error -
1164  * see comments nfs4.h:seqid_mutating_error()
1165  */
1166 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1167 {
1168 	if (seqid != NULL)
1169 		nfs_increment_seqid(status, seqid);
1170 }
1171 
1172 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1173 {
1174 	struct nfs_seqid_counter *sequence;
1175 	int status = 0;
1176 
1177 	if (seqid == NULL)
1178 		goto out;
1179 	sequence = seqid->sequence;
1180 	spin_lock(&sequence->lock);
1181 	seqid->task = task;
1182 	if (list_empty(&seqid->list))
1183 		list_add_tail(&seqid->list, &sequence->list);
1184 	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1185 		goto unlock;
1186 	rpc_sleep_on(&sequence->wait, task, NULL);
1187 	status = -EAGAIN;
1188 unlock:
1189 	spin_unlock(&sequence->lock);
1190 out:
1191 	return status;
1192 }
1193 
1194 static int nfs4_run_state_manager(void *);
1195 
1196 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1197 {
1198 	smp_mb__before_atomic();
1199 	clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1200 	smp_mb__after_atomic();
1201 	wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1202 	rpc_wake_up(&clp->cl_rpcwaitq);
1203 }
1204 
1205 /*
1206  * Schedule the nfs_client asynchronous state management routine
1207  */
1208 void nfs4_schedule_state_manager(struct nfs_client *clp)
1209 {
1210 	struct task_struct *task;
1211 	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1212 
1213 	set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1214 	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1215 		return;
1216 	__module_get(THIS_MODULE);
1217 	refcount_inc(&clp->cl_count);
1218 
1219 	/* The rcu_read_lock() is not strictly necessary, as the state
1220 	 * manager is the only thread that ever changes the rpc_xprt
1221 	 * after it's initialized.  At this point, we're single threaded. */
1222 	rcu_read_lock();
1223 	snprintf(buf, sizeof(buf), "%s-manager",
1224 			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1225 	rcu_read_unlock();
1226 	task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1227 	if (IS_ERR(task)) {
1228 		printk(KERN_ERR "%s: kthread_run: %ld\n",
1229 			__func__, PTR_ERR(task));
1230 		nfs4_clear_state_manager_bit(clp);
1231 		nfs_put_client(clp);
1232 		module_put(THIS_MODULE);
1233 	}
1234 }
1235 
1236 /*
1237  * Schedule a lease recovery attempt
1238  */
1239 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1240 {
1241 	if (!clp)
1242 		return;
1243 	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1244 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1245 	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1246 			clp->cl_hostname);
1247 	nfs4_schedule_state_manager(clp);
1248 }
1249 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1250 
1251 /**
1252  * nfs4_schedule_migration_recovery - trigger migration recovery
1253  *
1254  * @server: FSID that is migrating
1255  *
1256  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1257  * value is returned.
1258  */
1259 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1260 {
1261 	struct nfs_client *clp = server->nfs_client;
1262 
1263 	if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1264 		pr_err("NFS: volatile file handles not supported (server %s)\n",
1265 				clp->cl_hostname);
1266 		return -NFS4ERR_IO;
1267 	}
1268 
1269 	if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1270 		return -NFS4ERR_IO;
1271 
1272 	dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1273 			__func__,
1274 			(unsigned long long)server->fsid.major,
1275 			(unsigned long long)server->fsid.minor,
1276 			clp->cl_hostname);
1277 
1278 	set_bit(NFS_MIG_IN_TRANSITION,
1279 			&((struct nfs_server *)server)->mig_status);
1280 	set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1281 
1282 	nfs4_schedule_state_manager(clp);
1283 	return 0;
1284 }
1285 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1286 
1287 /**
1288  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1289  *
1290  * @clp: server to check for moved leases
1291  *
1292  */
1293 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1294 {
1295 	dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1296 		__func__, clp->cl_clientid, clp->cl_hostname);
1297 
1298 	set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1299 	nfs4_schedule_state_manager(clp);
1300 }
1301 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1302 
1303 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1304 {
1305 	int res;
1306 
1307 	might_sleep();
1308 
1309 	refcount_inc(&clp->cl_count);
1310 	res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1311 				 nfs_wait_bit_killable, TASK_KILLABLE);
1312 	if (res)
1313 		goto out;
1314 	if (clp->cl_cons_state < 0)
1315 		res = clp->cl_cons_state;
1316 out:
1317 	nfs_put_client(clp);
1318 	return res;
1319 }
1320 
1321 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1322 {
1323 	unsigned int loop;
1324 	int ret;
1325 
1326 	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1327 		ret = nfs4_wait_clnt_recover(clp);
1328 		if (ret != 0)
1329 			break;
1330 		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1331 		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1332 			break;
1333 		nfs4_schedule_state_manager(clp);
1334 		ret = -EIO;
1335 	}
1336 	return ret;
1337 }
1338 
1339 /*
1340  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1341  * @clp: client to process
1342  *
1343  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1344  * resend of the SETCLIENTID and hence re-establish the
1345  * callback channel. Then return all existing delegations.
1346  */
1347 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1348 {
1349 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1350 	nfs_expire_all_delegations(clp);
1351 	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1352 			clp->cl_hostname);
1353 }
1354 
1355 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1356 {
1357 	nfs40_handle_cb_pathdown(clp);
1358 	nfs4_schedule_state_manager(clp);
1359 }
1360 
1361 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1362 {
1363 
1364 	if (!nfs4_valid_open_stateid(state))
1365 		return 0;
1366 	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1367 	/* Don't recover state that expired before the reboot */
1368 	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1369 		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1370 		return 0;
1371 	}
1372 	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1373 	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1374 	return 1;
1375 }
1376 
1377 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1378 {
1379 	if (!nfs4_valid_open_stateid(state))
1380 		return 0;
1381 	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1382 	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1383 	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1384 	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1385 	return 1;
1386 }
1387 
1388 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1389 {
1390 	struct nfs_client *clp = server->nfs_client;
1391 
1392 	if (!nfs4_state_mark_reclaim_nograce(clp, state))
1393 		return -EBADF;
1394 	nfs_inode_find_delegation_state_and_recover(state->inode,
1395 			&state->stateid);
1396 	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1397 			clp->cl_hostname);
1398 	nfs4_schedule_state_manager(clp);
1399 	return 0;
1400 }
1401 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1402 
1403 static struct nfs4_lock_state *
1404 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1405 		const nfs4_stateid *stateid)
1406 {
1407 	struct nfs4_lock_state *pos;
1408 
1409 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
1410 		if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1411 			continue;
1412 		if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1413 			return pos;
1414 	}
1415 	return NULL;
1416 }
1417 
1418 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1419 		const nfs4_stateid *stateid)
1420 {
1421 	bool found = false;
1422 
1423 	if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1424 		spin_lock(&state->state_lock);
1425 		if (nfs_state_find_lock_state_by_stateid(state, stateid))
1426 			found = true;
1427 		spin_unlock(&state->state_lock);
1428 	}
1429 	return found;
1430 }
1431 
1432 void nfs_inode_find_state_and_recover(struct inode *inode,
1433 		const nfs4_stateid *stateid)
1434 {
1435 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1436 	struct nfs_inode *nfsi = NFS_I(inode);
1437 	struct nfs_open_context *ctx;
1438 	struct nfs4_state *state;
1439 	bool found = false;
1440 
1441 	rcu_read_lock();
1442 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1443 		state = ctx->state;
1444 		if (state == NULL)
1445 			continue;
1446 		if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1447 		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1448 			found = true;
1449 			continue;
1450 		}
1451 		if (nfs4_stateid_match_other(&state->open_stateid, stateid) &&
1452 		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1453 			found = true;
1454 			continue;
1455 		}
1456 		if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1457 		    nfs4_state_mark_reclaim_nograce(clp, state))
1458 			found = true;
1459 	}
1460 	rcu_read_unlock();
1461 
1462 	nfs_inode_find_delegation_state_and_recover(inode, stateid);
1463 	if (found)
1464 		nfs4_schedule_state_manager(clp);
1465 }
1466 
1467 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1468 {
1469 	struct inode *inode = state->inode;
1470 	struct nfs_inode *nfsi = NFS_I(inode);
1471 	struct nfs_open_context *ctx;
1472 
1473 	rcu_read_lock();
1474 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1475 		if (ctx->state != state)
1476 			continue;
1477 		set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1478 	}
1479 	rcu_read_unlock();
1480 }
1481 
1482 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1483 {
1484 	set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1485 	nfs4_state_mark_open_context_bad(state);
1486 }
1487 
1488 
1489 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1490 {
1491 	struct inode *inode = state->inode;
1492 	struct nfs_inode *nfsi = NFS_I(inode);
1493 	struct file_lock *fl;
1494 	struct nfs4_lock_state *lsp;
1495 	int status = 0;
1496 	struct file_lock_context *flctx = inode->i_flctx;
1497 	struct list_head *list;
1498 
1499 	if (flctx == NULL)
1500 		return 0;
1501 
1502 	list = &flctx->flc_posix;
1503 
1504 	/* Guard against delegation returns and new lock/unlock calls */
1505 	down_write(&nfsi->rwsem);
1506 	spin_lock(&flctx->flc_lock);
1507 restart:
1508 	list_for_each_entry(fl, list, fl_list) {
1509 		if (nfs_file_open_context(fl->fl_file)->state != state)
1510 			continue;
1511 		spin_unlock(&flctx->flc_lock);
1512 		status = ops->recover_lock(state, fl);
1513 		switch (status) {
1514 		case 0:
1515 			break;
1516 		case -ESTALE:
1517 		case -NFS4ERR_ADMIN_REVOKED:
1518 		case -NFS4ERR_STALE_STATEID:
1519 		case -NFS4ERR_BAD_STATEID:
1520 		case -NFS4ERR_EXPIRED:
1521 		case -NFS4ERR_NO_GRACE:
1522 		case -NFS4ERR_STALE_CLIENTID:
1523 		case -NFS4ERR_BADSESSION:
1524 		case -NFS4ERR_BADSLOT:
1525 		case -NFS4ERR_BAD_HIGH_SLOT:
1526 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1527 			goto out;
1528 		default:
1529 			pr_err("NFS: %s: unhandled error %d\n",
1530 					__func__, status);
1531 			/* Fall through */
1532 		case -ENOMEM:
1533 		case -NFS4ERR_DENIED:
1534 		case -NFS4ERR_RECLAIM_BAD:
1535 		case -NFS4ERR_RECLAIM_CONFLICT:
1536 			lsp = fl->fl_u.nfs4_fl.owner;
1537 			if (lsp)
1538 				set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1539 			status = 0;
1540 		}
1541 		spin_lock(&flctx->flc_lock);
1542 	}
1543 	if (list == &flctx->flc_posix) {
1544 		list = &flctx->flc_flock;
1545 		goto restart;
1546 	}
1547 	spin_unlock(&flctx->flc_lock);
1548 out:
1549 	up_write(&nfsi->rwsem);
1550 	return status;
1551 }
1552 
1553 #ifdef CONFIG_NFS_V4_2
1554 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1555 {
1556 	struct nfs4_copy_state *copy;
1557 
1558 	if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags))
1559 		return;
1560 
1561 	spin_lock(&sp->so_server->nfs_client->cl_lock);
1562 	list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1563 		if (nfs4_stateid_match_other(&state->stateid, &copy->parent_state->stateid))
1564 			continue;
1565 		copy->flags = 1;
1566 		complete(&copy->completion);
1567 		break;
1568 	}
1569 	spin_unlock(&sp->so_server->nfs_client->cl_lock);
1570 }
1571 #else /* !CONFIG_NFS_V4_2 */
1572 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1573 					 struct nfs4_state *state)
1574 {
1575 }
1576 #endif /* CONFIG_NFS_V4_2 */
1577 
1578 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1579 				     const struct nfs4_state_recovery_ops *ops)
1580 {
1581 	struct nfs4_lock_state *lock;
1582 	int status;
1583 
1584 	status = ops->recover_open(sp, state);
1585 	if (status < 0)
1586 		return status;
1587 
1588 	status = nfs4_reclaim_locks(state, ops);
1589 	if (status < 0)
1590 		return status;
1591 
1592 	if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1593 		spin_lock(&state->state_lock);
1594 		list_for_each_entry(lock, &state->lock_states, ls_locks) {
1595 			if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1596 				pr_warn_ratelimited("NFS: %s: Lock reclaim failed!\n", __func__);
1597 		}
1598 		spin_unlock(&state->state_lock);
1599 	}
1600 
1601 	nfs42_complete_copies(sp, state);
1602 	clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1603 	return status;
1604 }
1605 
1606 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1607 {
1608 	struct nfs4_state *state;
1609 	int status = 0;
1610 
1611 	/* Note: we rely on the sp->so_states list being ordered
1612 	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1613 	 * states first.
1614 	 * This is needed to ensure that the server won't give us any
1615 	 * read delegations that we have to return if, say, we are
1616 	 * recovering after a network partition or a reboot from a
1617 	 * server that doesn't support a grace period.
1618 	 */
1619 	spin_lock(&sp->so_lock);
1620 	raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1621 restart:
1622 	list_for_each_entry(state, &sp->so_states, open_states) {
1623 		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1624 			continue;
1625 		if (!nfs4_valid_open_stateid(state))
1626 			continue;
1627 		if (state->state == 0)
1628 			continue;
1629 		refcount_inc(&state->count);
1630 		spin_unlock(&sp->so_lock);
1631 		status = __nfs4_reclaim_open_state(sp, state, ops);
1632 
1633 		switch (status) {
1634 		default:
1635 			if (status >= 0)
1636 				break;
1637 			printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1638 			/* Fall through */
1639 		case -ENOENT:
1640 		case -ENOMEM:
1641 		case -EACCES:
1642 		case -EROFS:
1643 		case -EIO:
1644 		case -ESTALE:
1645 			/* Open state on this file cannot be recovered */
1646 			nfs4_state_mark_recovery_failed(state, status);
1647 			break;
1648 		case -EAGAIN:
1649 			ssleep(1);
1650 			/* Fall through */
1651 		case -NFS4ERR_ADMIN_REVOKED:
1652 		case -NFS4ERR_STALE_STATEID:
1653 		case -NFS4ERR_OLD_STATEID:
1654 		case -NFS4ERR_BAD_STATEID:
1655 		case -NFS4ERR_RECLAIM_BAD:
1656 		case -NFS4ERR_RECLAIM_CONFLICT:
1657 			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1658 			break;
1659 		case -NFS4ERR_EXPIRED:
1660 		case -NFS4ERR_NO_GRACE:
1661 			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1662 		case -NFS4ERR_STALE_CLIENTID:
1663 		case -NFS4ERR_BADSESSION:
1664 		case -NFS4ERR_BADSLOT:
1665 		case -NFS4ERR_BAD_HIGH_SLOT:
1666 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1667 			goto out_err;
1668 		}
1669 		nfs4_put_open_state(state);
1670 		spin_lock(&sp->so_lock);
1671 		goto restart;
1672 	}
1673 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1674 	spin_unlock(&sp->so_lock);
1675 	return 0;
1676 out_err:
1677 	nfs4_put_open_state(state);
1678 	spin_lock(&sp->so_lock);
1679 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1680 	spin_unlock(&sp->so_lock);
1681 	return status;
1682 }
1683 
1684 static void nfs4_clear_open_state(struct nfs4_state *state)
1685 {
1686 	struct nfs4_lock_state *lock;
1687 
1688 	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1689 	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1690 	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1691 	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1692 	spin_lock(&state->state_lock);
1693 	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1694 		lock->ls_seqid.flags = 0;
1695 		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1696 	}
1697 	spin_unlock(&state->state_lock);
1698 }
1699 
1700 static void nfs4_reset_seqids(struct nfs_server *server,
1701 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1702 {
1703 	struct nfs_client *clp = server->nfs_client;
1704 	struct nfs4_state_owner *sp;
1705 	struct rb_node *pos;
1706 	struct nfs4_state *state;
1707 
1708 	spin_lock(&clp->cl_lock);
1709 	for (pos = rb_first(&server->state_owners);
1710 	     pos != NULL;
1711 	     pos = rb_next(pos)) {
1712 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1713 		sp->so_seqid.flags = 0;
1714 		spin_lock(&sp->so_lock);
1715 		list_for_each_entry(state, &sp->so_states, open_states) {
1716 			if (mark_reclaim(clp, state))
1717 				nfs4_clear_open_state(state);
1718 		}
1719 		spin_unlock(&sp->so_lock);
1720 	}
1721 	spin_unlock(&clp->cl_lock);
1722 }
1723 
1724 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1725 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1726 {
1727 	struct nfs_server *server;
1728 
1729 	rcu_read_lock();
1730 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1731 		nfs4_reset_seqids(server, mark_reclaim);
1732 	rcu_read_unlock();
1733 }
1734 
1735 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1736 {
1737 	/* Mark all delegations for reclaim */
1738 	nfs_delegation_mark_reclaim(clp);
1739 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1740 }
1741 
1742 static int nfs4_reclaim_complete(struct nfs_client *clp,
1743 				 const struct nfs4_state_recovery_ops *ops,
1744 				 struct rpc_cred *cred)
1745 {
1746 	/* Notify the server we're done reclaiming our state */
1747 	if (ops->reclaim_complete)
1748 		return ops->reclaim_complete(clp, cred);
1749 	return 0;
1750 }
1751 
1752 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1753 {
1754 	struct nfs_client *clp = server->nfs_client;
1755 	struct nfs4_state_owner *sp;
1756 	struct rb_node *pos;
1757 	struct nfs4_state *state;
1758 
1759 	spin_lock(&clp->cl_lock);
1760 	for (pos = rb_first(&server->state_owners);
1761 	     pos != NULL;
1762 	     pos = rb_next(pos)) {
1763 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1764 		spin_lock(&sp->so_lock);
1765 		list_for_each_entry(state, &sp->so_states, open_states) {
1766 			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1767 						&state->flags))
1768 				continue;
1769 			nfs4_state_mark_reclaim_nograce(clp, state);
1770 		}
1771 		spin_unlock(&sp->so_lock);
1772 	}
1773 	spin_unlock(&clp->cl_lock);
1774 }
1775 
1776 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1777 {
1778 	struct nfs_server *server;
1779 
1780 	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1781 		return 0;
1782 
1783 	rcu_read_lock();
1784 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1785 		nfs4_clear_reclaim_server(server);
1786 	rcu_read_unlock();
1787 
1788 	nfs_delegation_reap_unclaimed(clp);
1789 	return 1;
1790 }
1791 
1792 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1793 {
1794 	const struct nfs4_state_recovery_ops *ops;
1795 	struct rpc_cred *cred;
1796 	int err;
1797 
1798 	if (!nfs4_state_clear_reclaim_reboot(clp))
1799 		return;
1800 	ops = clp->cl_mvops->reboot_recovery_ops;
1801 	cred = nfs4_get_clid_cred(clp);
1802 	err = nfs4_reclaim_complete(clp, ops, cred);
1803 	put_rpccred(cred);
1804 	if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1805 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1806 }
1807 
1808 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1809 {
1810 	nfs_mark_test_expired_all_delegations(clp);
1811 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1812 }
1813 
1814 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1815 {
1816 	switch (error) {
1817 	case 0:
1818 		break;
1819 	case -NFS4ERR_CB_PATH_DOWN:
1820 		nfs40_handle_cb_pathdown(clp);
1821 		break;
1822 	case -NFS4ERR_NO_GRACE:
1823 		nfs4_state_end_reclaim_reboot(clp);
1824 		break;
1825 	case -NFS4ERR_STALE_CLIENTID:
1826 		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1827 		nfs4_state_start_reclaim_reboot(clp);
1828 		break;
1829 	case -NFS4ERR_EXPIRED:
1830 		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1831 		nfs4_state_start_reclaim_nograce(clp);
1832 		break;
1833 	case -NFS4ERR_BADSESSION:
1834 	case -NFS4ERR_BADSLOT:
1835 	case -NFS4ERR_BAD_HIGH_SLOT:
1836 	case -NFS4ERR_DEADSESSION:
1837 	case -NFS4ERR_SEQ_FALSE_RETRY:
1838 	case -NFS4ERR_SEQ_MISORDERED:
1839 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1840 		/* Zero session reset errors */
1841 		break;
1842 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1843 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1844 		break;
1845 	default:
1846 		dprintk("%s: failed to handle error %d for server %s\n",
1847 				__func__, error, clp->cl_hostname);
1848 		return error;
1849 	}
1850 	dprintk("%s: handled error %d for server %s\n", __func__, error,
1851 			clp->cl_hostname);
1852 	return 0;
1853 }
1854 
1855 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1856 {
1857 	struct nfs4_state_owner *sp;
1858 	struct nfs_server *server;
1859 	struct rb_node *pos;
1860 	int status = 0;
1861 
1862 restart:
1863 	rcu_read_lock();
1864 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1865 		nfs4_purge_state_owners(server);
1866 		spin_lock(&clp->cl_lock);
1867 		for (pos = rb_first(&server->state_owners);
1868 		     pos != NULL;
1869 		     pos = rb_next(pos)) {
1870 			sp = rb_entry(pos,
1871 				struct nfs4_state_owner, so_server_node);
1872 			if (!test_and_clear_bit(ops->owner_flag_bit,
1873 							&sp->so_flags))
1874 				continue;
1875 			if (!atomic_inc_not_zero(&sp->so_count))
1876 				continue;
1877 			spin_unlock(&clp->cl_lock);
1878 			rcu_read_unlock();
1879 
1880 			status = nfs4_reclaim_open_state(sp, ops);
1881 			if (status < 0) {
1882 				set_bit(ops->owner_flag_bit, &sp->so_flags);
1883 				nfs4_put_state_owner(sp);
1884 				status = nfs4_recovery_handle_error(clp, status);
1885 				return (status != 0) ? status : -EAGAIN;
1886 			}
1887 
1888 			nfs4_put_state_owner(sp);
1889 			goto restart;
1890 		}
1891 		spin_unlock(&clp->cl_lock);
1892 	}
1893 	rcu_read_unlock();
1894 	return 0;
1895 }
1896 
1897 static int nfs4_check_lease(struct nfs_client *clp)
1898 {
1899 	struct rpc_cred *cred;
1900 	const struct nfs4_state_maintenance_ops *ops =
1901 		clp->cl_mvops->state_renewal_ops;
1902 	int status;
1903 
1904 	/* Is the client already known to have an expired lease? */
1905 	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1906 		return 0;
1907 	spin_lock(&clp->cl_lock);
1908 	cred = ops->get_state_renewal_cred_locked(clp);
1909 	spin_unlock(&clp->cl_lock);
1910 	if (cred == NULL) {
1911 		cred = nfs4_get_clid_cred(clp);
1912 		status = -ENOKEY;
1913 		if (cred == NULL)
1914 			goto out;
1915 	}
1916 	status = ops->renew_lease(clp, cred);
1917 	put_rpccred(cred);
1918 	if (status == -ETIMEDOUT) {
1919 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1920 		return 0;
1921 	}
1922 out:
1923 	return nfs4_recovery_handle_error(clp, status);
1924 }
1925 
1926 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1927  * and for recoverable errors on EXCHANGE_ID for v4.1
1928  */
1929 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1930 {
1931 	switch (status) {
1932 	case -NFS4ERR_SEQ_MISORDERED:
1933 		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1934 			return -ESERVERFAULT;
1935 		/* Lease confirmation error: retry after purging the lease */
1936 		ssleep(1);
1937 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1938 		break;
1939 	case -NFS4ERR_STALE_CLIENTID:
1940 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1941 		nfs4_state_start_reclaim_reboot(clp);
1942 		break;
1943 	case -NFS4ERR_CLID_INUSE:
1944 		pr_err("NFS: Server %s reports our clientid is in use\n",
1945 			clp->cl_hostname);
1946 		nfs_mark_client_ready(clp, -EPERM);
1947 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1948 		return -EPERM;
1949 	case -EACCES:
1950 	case -NFS4ERR_DELAY:
1951 	case -ETIMEDOUT:
1952 	case -EAGAIN:
1953 		ssleep(1);
1954 		break;
1955 
1956 	case -NFS4ERR_MINOR_VERS_MISMATCH:
1957 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1958 			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1959 		dprintk("%s: exit with error %d for server %s\n",
1960 				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
1961 		return -EPROTONOSUPPORT;
1962 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1963 				 * in nfs4_exchange_id */
1964 	default:
1965 		dprintk("%s: exit with error %d for server %s\n", __func__,
1966 				status, clp->cl_hostname);
1967 		return status;
1968 	}
1969 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1970 	dprintk("%s: handled error %d for server %s\n", __func__, status,
1971 			clp->cl_hostname);
1972 	return 0;
1973 }
1974 
1975 static int nfs4_establish_lease(struct nfs_client *clp)
1976 {
1977 	struct rpc_cred *cred;
1978 	const struct nfs4_state_recovery_ops *ops =
1979 		clp->cl_mvops->reboot_recovery_ops;
1980 	int status;
1981 
1982 	status = nfs4_begin_drain_session(clp);
1983 	if (status != 0)
1984 		return status;
1985 	cred = nfs4_get_clid_cred(clp);
1986 	if (cred == NULL)
1987 		return -ENOENT;
1988 	status = ops->establish_clid(clp, cred);
1989 	put_rpccred(cred);
1990 	if (status != 0)
1991 		return status;
1992 	pnfs_destroy_all_layouts(clp);
1993 	return 0;
1994 }
1995 
1996 /*
1997  * Returns zero or a negative errno.  NFS4ERR values are converted
1998  * to local errno values.
1999  */
2000 static int nfs4_reclaim_lease(struct nfs_client *clp)
2001 {
2002 	int status;
2003 
2004 	status = nfs4_establish_lease(clp);
2005 	if (status < 0)
2006 		return nfs4_handle_reclaim_lease_error(clp, status);
2007 	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2008 		nfs4_state_start_reclaim_nograce(clp);
2009 	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2010 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2011 	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2012 	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2013 	return 0;
2014 }
2015 
2016 static int nfs4_purge_lease(struct nfs_client *clp)
2017 {
2018 	int status;
2019 
2020 	status = nfs4_establish_lease(clp);
2021 	if (status < 0)
2022 		return nfs4_handle_reclaim_lease_error(clp, status);
2023 	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2024 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2025 	nfs4_state_start_reclaim_nograce(clp);
2026 	return 0;
2027 }
2028 
2029 /*
2030  * Try remote migration of one FSID from a source server to a
2031  * destination server.  The source server provides a list of
2032  * potential destinations.
2033  *
2034  * Returns zero or a negative NFS4ERR status code.
2035  */
2036 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
2037 {
2038 	struct nfs_client *clp = server->nfs_client;
2039 	struct nfs4_fs_locations *locations = NULL;
2040 	struct inode *inode;
2041 	struct page *page;
2042 	int status, result;
2043 
2044 	dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2045 			(unsigned long long)server->fsid.major,
2046 			(unsigned long long)server->fsid.minor,
2047 			clp->cl_hostname);
2048 
2049 	result = 0;
2050 	page = alloc_page(GFP_KERNEL);
2051 	locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2052 	if (page == NULL || locations == NULL) {
2053 		dprintk("<-- %s: no memory\n", __func__);
2054 		goto out;
2055 	}
2056 
2057 	inode = d_inode(server->super->s_root);
2058 	result = nfs4_proc_get_locations(inode, locations, page, cred);
2059 	if (result) {
2060 		dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2061 			__func__, result);
2062 		goto out;
2063 	}
2064 
2065 	result = -NFS4ERR_NXIO;
2066 	if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2067 		dprintk("<-- %s: No fs_locations data, migration skipped\n",
2068 			__func__);
2069 		goto out;
2070 	}
2071 
2072 	status = nfs4_begin_drain_session(clp);
2073 	if (status != 0)
2074 		return status;
2075 
2076 	status = nfs4_replace_transport(server, locations);
2077 	if (status != 0) {
2078 		dprintk("<-- %s: failed to replace transport: %d\n",
2079 			__func__, status);
2080 		goto out;
2081 	}
2082 
2083 	result = 0;
2084 	dprintk("<-- %s: migration succeeded\n", __func__);
2085 
2086 out:
2087 	if (page != NULL)
2088 		__free_page(page);
2089 	kfree(locations);
2090 	if (result) {
2091 		pr_err("NFS: migration recovery failed (server %s)\n",
2092 				clp->cl_hostname);
2093 		set_bit(NFS_MIG_FAILED, &server->mig_status);
2094 	}
2095 	return result;
2096 }
2097 
2098 /*
2099  * Returns zero or a negative NFS4ERR status code.
2100  */
2101 static int nfs4_handle_migration(struct nfs_client *clp)
2102 {
2103 	const struct nfs4_state_maintenance_ops *ops =
2104 				clp->cl_mvops->state_renewal_ops;
2105 	struct nfs_server *server;
2106 	struct rpc_cred *cred;
2107 
2108 	dprintk("%s: migration reported on \"%s\"\n", __func__,
2109 			clp->cl_hostname);
2110 
2111 	spin_lock(&clp->cl_lock);
2112 	cred = ops->get_state_renewal_cred_locked(clp);
2113 	spin_unlock(&clp->cl_lock);
2114 	if (cred == NULL)
2115 		return -NFS4ERR_NOENT;
2116 
2117 	clp->cl_mig_gen++;
2118 restart:
2119 	rcu_read_lock();
2120 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2121 		int status;
2122 
2123 		if (server->mig_gen == clp->cl_mig_gen)
2124 			continue;
2125 		server->mig_gen = clp->cl_mig_gen;
2126 
2127 		if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2128 						&server->mig_status))
2129 			continue;
2130 
2131 		rcu_read_unlock();
2132 		status = nfs4_try_migration(server, cred);
2133 		if (status < 0) {
2134 			put_rpccred(cred);
2135 			return status;
2136 		}
2137 		goto restart;
2138 	}
2139 	rcu_read_unlock();
2140 	put_rpccred(cred);
2141 	return 0;
2142 }
2143 
2144 /*
2145  * Test each nfs_server on the clp's cl_superblocks list to see
2146  * if it's moved to another server.  Stop when the server no longer
2147  * returns NFS4ERR_LEASE_MOVED.
2148  */
2149 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2150 {
2151 	const struct nfs4_state_maintenance_ops *ops =
2152 				clp->cl_mvops->state_renewal_ops;
2153 	struct nfs_server *server;
2154 	struct rpc_cred *cred;
2155 
2156 	dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2157 			clp->cl_hostname);
2158 
2159 	spin_lock(&clp->cl_lock);
2160 	cred = ops->get_state_renewal_cred_locked(clp);
2161 	spin_unlock(&clp->cl_lock);
2162 	if (cred == NULL)
2163 		return -NFS4ERR_NOENT;
2164 
2165 	clp->cl_mig_gen++;
2166 restart:
2167 	rcu_read_lock();
2168 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2169 		struct inode *inode;
2170 		int status;
2171 
2172 		if (server->mig_gen == clp->cl_mig_gen)
2173 			continue;
2174 		server->mig_gen = clp->cl_mig_gen;
2175 
2176 		rcu_read_unlock();
2177 
2178 		inode = d_inode(server->super->s_root);
2179 		status = nfs4_proc_fsid_present(inode, cred);
2180 		if (status != -NFS4ERR_MOVED)
2181 			goto restart;	/* wasn't this one */
2182 		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2183 			goto restart;	/* there are more */
2184 		goto out;
2185 	}
2186 	rcu_read_unlock();
2187 
2188 out:
2189 	put_rpccred(cred);
2190 	return 0;
2191 }
2192 
2193 /**
2194  * nfs4_discover_server_trunking - Detect server IP address trunking
2195  *
2196  * @clp: nfs_client under test
2197  * @result: OUT: found nfs_client, or clp
2198  *
2199  * Returns zero or a negative errno.  If zero is returned,
2200  * an nfs_client pointer is planted in "result".
2201  *
2202  * Note: since we are invoked in process context, and
2203  * not from inside the state manager, we cannot use
2204  * nfs4_handle_reclaim_lease_error().
2205  */
2206 int nfs4_discover_server_trunking(struct nfs_client *clp,
2207 				  struct nfs_client **result)
2208 {
2209 	const struct nfs4_state_recovery_ops *ops =
2210 				clp->cl_mvops->reboot_recovery_ops;
2211 	struct rpc_clnt *clnt;
2212 	struct rpc_cred *cred;
2213 	int i, status;
2214 
2215 	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2216 
2217 	clnt = clp->cl_rpcclient;
2218 	i = 0;
2219 
2220 	mutex_lock(&nfs_clid_init_mutex);
2221 again:
2222 	status  = -ENOENT;
2223 	cred = nfs4_get_clid_cred(clp);
2224 	if (cred == NULL)
2225 		goto out_unlock;
2226 
2227 	status = ops->detect_trunking(clp, result, cred);
2228 	put_rpccred(cred);
2229 	switch (status) {
2230 	case 0:
2231 	case -EINTR:
2232 	case -ERESTARTSYS:
2233 		break;
2234 	case -ETIMEDOUT:
2235 		if (clnt->cl_softrtry)
2236 			break;
2237 		/* Fall through */
2238 	case -NFS4ERR_DELAY:
2239 	case -EAGAIN:
2240 		ssleep(1);
2241 		/* Fall through */
2242 	case -NFS4ERR_STALE_CLIENTID:
2243 		dprintk("NFS: %s after status %d, retrying\n",
2244 			__func__, status);
2245 		goto again;
2246 	case -EACCES:
2247 		if (i++ == 0) {
2248 			nfs4_root_machine_cred(clp);
2249 			goto again;
2250 		}
2251 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2252 			break;
2253 		/* Fall through */
2254 	case -NFS4ERR_CLID_INUSE:
2255 	case -NFS4ERR_WRONGSEC:
2256 		/* No point in retrying if we already used RPC_AUTH_UNIX */
2257 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2258 			status = -EPERM;
2259 			break;
2260 		}
2261 		clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2262 		if (IS_ERR(clnt)) {
2263 			status = PTR_ERR(clnt);
2264 			break;
2265 		}
2266 		/* Note: this is safe because we haven't yet marked the
2267 		 * client as ready, so we are the only user of
2268 		 * clp->cl_rpcclient
2269 		 */
2270 		clnt = xchg(&clp->cl_rpcclient, clnt);
2271 		rpc_shutdown_client(clnt);
2272 		clnt = clp->cl_rpcclient;
2273 		goto again;
2274 
2275 	case -NFS4ERR_MINOR_VERS_MISMATCH:
2276 		status = -EPROTONOSUPPORT;
2277 		break;
2278 
2279 	case -EKEYEXPIRED:
2280 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2281 				 * in nfs4_exchange_id */
2282 		status = -EKEYEXPIRED;
2283 		break;
2284 	default:
2285 		pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2286 				__func__, status);
2287 		status = -EIO;
2288 	}
2289 
2290 out_unlock:
2291 	mutex_unlock(&nfs_clid_init_mutex);
2292 	dprintk("NFS: %s: status = %d\n", __func__, status);
2293 	return status;
2294 }
2295 
2296 #ifdef CONFIG_NFS_V4_1
2297 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2298 {
2299 	struct nfs_client *clp = session->clp;
2300 
2301 	switch (err) {
2302 	default:
2303 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2304 		break;
2305 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2306 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2307 	}
2308 	nfs4_schedule_state_manager(clp);
2309 }
2310 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2311 
2312 void nfs41_notify_server(struct nfs_client *clp)
2313 {
2314 	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
2315 	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2316 	nfs4_schedule_state_manager(clp);
2317 }
2318 
2319 static void nfs4_reset_all_state(struct nfs_client *clp)
2320 {
2321 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2322 		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2323 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2324 		nfs4_state_start_reclaim_nograce(clp);
2325 		dprintk("%s: scheduling reset of all state for server %s!\n",
2326 				__func__, clp->cl_hostname);
2327 		nfs4_schedule_state_manager(clp);
2328 	}
2329 }
2330 
2331 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2332 {
2333 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2334 		nfs4_state_start_reclaim_reboot(clp);
2335 		dprintk("%s: server %s rebooted!\n", __func__,
2336 				clp->cl_hostname);
2337 		nfs4_schedule_state_manager(clp);
2338 	}
2339 }
2340 
2341 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2342 {
2343 	nfs4_reset_all_state(clp);
2344 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2345 }
2346 
2347 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2348 {
2349 	nfs4_state_start_reclaim_nograce(clp);
2350 	nfs4_schedule_state_manager(clp);
2351 
2352 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2353 }
2354 
2355 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2356 {
2357 	/* FIXME: For now, we destroy all layouts. */
2358 	pnfs_destroy_all_layouts(clp);
2359 	/* FIXME: For now, we test all delegations+open state+locks. */
2360 	nfs41_handle_some_state_revoked(clp);
2361 	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2362 			clp->cl_hostname);
2363 }
2364 
2365 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2366 {
2367 	set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2368 	nfs4_schedule_state_manager(clp);
2369 
2370 	dprintk("%s: server %s declared a backchannel fault\n", __func__,
2371 			clp->cl_hostname);
2372 }
2373 
2374 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2375 {
2376 	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2377 		&clp->cl_state) == 0)
2378 		nfs4_schedule_state_manager(clp);
2379 }
2380 
2381 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2382 		bool recovery)
2383 {
2384 	if (!flags)
2385 		return;
2386 
2387 	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2388 		__func__, clp->cl_hostname, clp->cl_clientid, flags);
2389 	/*
2390 	 * If we're called from the state manager thread, then assume we're
2391 	 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2392 	 * Those flags are expected to remain set until we're done
2393 	 * recovering (see RFC5661, section 18.46.3).
2394 	 */
2395 	if (recovery)
2396 		goto out_recovery;
2397 
2398 	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2399 		nfs41_handle_server_reboot(clp);
2400 	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2401 		nfs41_handle_all_state_revoked(clp);
2402 	if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2403 			    SEQ4_STATUS_ADMIN_STATE_REVOKED))
2404 		nfs41_handle_some_state_revoked(clp);
2405 	if (flags & SEQ4_STATUS_LEASE_MOVED)
2406 		nfs4_schedule_lease_moved_recovery(clp);
2407 	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2408 		nfs41_handle_recallable_state_revoked(clp);
2409 out_recovery:
2410 	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2411 		nfs41_handle_backchannel_fault(clp);
2412 	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2413 				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2414 		nfs41_handle_cb_path_down(clp);
2415 }
2416 
2417 static int nfs4_reset_session(struct nfs_client *clp)
2418 {
2419 	struct rpc_cred *cred;
2420 	int status;
2421 
2422 	if (!nfs4_has_session(clp))
2423 		return 0;
2424 	status = nfs4_begin_drain_session(clp);
2425 	if (status != 0)
2426 		return status;
2427 	cred = nfs4_get_clid_cred(clp);
2428 	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2429 	switch (status) {
2430 	case 0:
2431 	case -NFS4ERR_BADSESSION:
2432 	case -NFS4ERR_DEADSESSION:
2433 		break;
2434 	case -NFS4ERR_BACK_CHAN_BUSY:
2435 	case -NFS4ERR_DELAY:
2436 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2437 		status = 0;
2438 		ssleep(1);
2439 		goto out;
2440 	default:
2441 		status = nfs4_recovery_handle_error(clp, status);
2442 		goto out;
2443 	}
2444 
2445 	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2446 	status = nfs4_proc_create_session(clp, cred);
2447 	if (status) {
2448 		dprintk("%s: session reset failed with status %d for server %s!\n",
2449 			__func__, status, clp->cl_hostname);
2450 		status = nfs4_handle_reclaim_lease_error(clp, status);
2451 		goto out;
2452 	}
2453 	nfs41_finish_session_reset(clp);
2454 	dprintk("%s: session reset was successful for server %s!\n",
2455 			__func__, clp->cl_hostname);
2456 out:
2457 	if (cred)
2458 		put_rpccred(cred);
2459 	return status;
2460 }
2461 
2462 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2463 {
2464 	struct rpc_cred *cred;
2465 	int ret;
2466 
2467 	if (!nfs4_has_session(clp))
2468 		return 0;
2469 	ret = nfs4_begin_drain_session(clp);
2470 	if (ret != 0)
2471 		return ret;
2472 	cred = nfs4_get_clid_cred(clp);
2473 	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2474 	if (cred)
2475 		put_rpccred(cred);
2476 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2477 	switch (ret) {
2478 	case 0:
2479 		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2480 			__func__, clp->cl_hostname);
2481 		break;
2482 	case -NFS4ERR_DELAY:
2483 		ssleep(1);
2484 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2485 		break;
2486 	default:
2487 		return nfs4_recovery_handle_error(clp, ret);
2488 	}
2489 	return 0;
2490 }
2491 #else /* CONFIG_NFS_V4_1 */
2492 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2493 
2494 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2495 {
2496 	return 0;
2497 }
2498 #endif /* CONFIG_NFS_V4_1 */
2499 
2500 static void nfs4_state_manager(struct nfs_client *clp)
2501 {
2502 	int status = 0;
2503 	const char *section = "", *section_sep = "";
2504 
2505 	/* Ensure exclusive access to NFSv4 state */
2506 	do {
2507 		clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2508 		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2509 			section = "purge state";
2510 			status = nfs4_purge_lease(clp);
2511 			if (status < 0)
2512 				goto out_error;
2513 			continue;
2514 		}
2515 
2516 		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2517 			section = "lease expired";
2518 			/* We're going to have to re-establish a clientid */
2519 			status = nfs4_reclaim_lease(clp);
2520 			if (status < 0)
2521 				goto out_error;
2522 			continue;
2523 		}
2524 
2525 		/* Initialize or reset the session */
2526 		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2527 			section = "reset session";
2528 			status = nfs4_reset_session(clp);
2529 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2530 				continue;
2531 			if (status < 0)
2532 				goto out_error;
2533 		}
2534 
2535 		/* Send BIND_CONN_TO_SESSION */
2536 		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2537 				&clp->cl_state)) {
2538 			section = "bind conn to session";
2539 			status = nfs4_bind_conn_to_session(clp);
2540 			if (status < 0)
2541 				goto out_error;
2542 			continue;
2543 		}
2544 
2545 		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2546 			section = "check lease";
2547 			status = nfs4_check_lease(clp);
2548 			if (status < 0)
2549 				goto out_error;
2550 			continue;
2551 		}
2552 
2553 		if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2554 			section = "migration";
2555 			status = nfs4_handle_migration(clp);
2556 			if (status < 0)
2557 				goto out_error;
2558 		}
2559 
2560 		if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2561 			section = "lease moved";
2562 			status = nfs4_handle_lease_moved(clp);
2563 			if (status < 0)
2564 				goto out_error;
2565 		}
2566 
2567 		/* First recover reboot state... */
2568 		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2569 			section = "reclaim reboot";
2570 			status = nfs4_do_reclaim(clp,
2571 				clp->cl_mvops->reboot_recovery_ops);
2572 			if (status == -EAGAIN)
2573 				continue;
2574 			if (status < 0)
2575 				goto out_error;
2576 			nfs4_state_end_reclaim_reboot(clp);
2577 		}
2578 
2579 		/* Detect expired delegations... */
2580 		if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2581 			section = "detect expired delegations";
2582 			nfs_reap_expired_delegations(clp);
2583 			continue;
2584 		}
2585 
2586 		/* Now recover expired state... */
2587 		if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2588 			section = "reclaim nograce";
2589 			status = nfs4_do_reclaim(clp,
2590 				clp->cl_mvops->nograce_recovery_ops);
2591 			if (status == -EAGAIN)
2592 				continue;
2593 			if (status < 0)
2594 				goto out_error;
2595 		}
2596 
2597 		nfs4_end_drain_session(clp);
2598 		nfs4_clear_state_manager_bit(clp);
2599 
2600 		if (!test_and_set_bit(NFS4CLNT_DELEGRETURN_RUNNING, &clp->cl_state)) {
2601 			if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2602 				nfs_client_return_marked_delegations(clp);
2603 				set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2604 			}
2605 			clear_bit(NFS4CLNT_DELEGRETURN_RUNNING, &clp->cl_state);
2606 		}
2607 
2608 		/* Did we race with an attempt to give us more work? */
2609 		if (!test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state))
2610 			return;
2611 		if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2612 			return;
2613 	} while (refcount_read(&clp->cl_count) > 1 && !signalled());
2614 	goto out_drain;
2615 
2616 out_error:
2617 	if (strlen(section))
2618 		section_sep = ": ";
2619 	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2620 			" with error %d\n", section_sep, section,
2621 			clp->cl_hostname, -status);
2622 	ssleep(1);
2623 out_drain:
2624 	nfs4_end_drain_session(clp);
2625 	nfs4_clear_state_manager_bit(clp);
2626 }
2627 
2628 static int nfs4_run_state_manager(void *ptr)
2629 {
2630 	struct nfs_client *clp = ptr;
2631 
2632 	allow_signal(SIGKILL);
2633 	nfs4_state_manager(clp);
2634 	nfs_put_client(clp);
2635 	module_put_and_exit(0);
2636 	return 0;
2637 }
2638 
2639 /*
2640  * Local variables:
2641  *  c-basic-offset: 8
2642  * End:
2643  */
2644