xref: /linux/fs/nfs/nfs4state.c (revision f6e0a4984c2e7244689ea87b62b433bed9d07e94)
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 #include <linux/sched/mm.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 "nfs4idmap.h"
61 #include "nfs4session.h"
62 #include "pnfs.h"
63 #include "netns.h"
64 #include "nfs4trace.h"
65 
66 #define NFSDBG_FACILITY		NFSDBG_STATE
67 
68 #define OPENOWNER_POOL_SIZE	8
69 
70 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp);
71 
72 const nfs4_stateid zero_stateid = {
73 	{ .data = { 0 } },
74 	.type = NFS4_SPECIAL_STATEID_TYPE,
75 };
76 const nfs4_stateid invalid_stateid = {
77 	{
78 		/* Funky initialiser keeps older gcc versions happy */
79 		.data = { 0xff, 0xff, 0xff, 0xff, 0 },
80 	},
81 	.type = NFS4_INVALID_STATEID_TYPE,
82 };
83 
84 const nfs4_stateid current_stateid = {
85 	{
86 		/* Funky initialiser keeps older gcc versions happy */
87 		.data = { 0x0, 0x0, 0x0, 0x1, 0 },
88 	},
89 	.type = NFS4_SPECIAL_STATEID_TYPE,
90 };
91 
92 static DEFINE_MUTEX(nfs_clid_init_mutex);
93 
94 static int nfs4_setup_state_renewal(struct nfs_client *clp)
95 {
96 	int status;
97 	struct nfs_fsinfo fsinfo;
98 
99 	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
100 		nfs4_schedule_state_renewal(clp);
101 		return 0;
102 	}
103 
104 	status = nfs4_proc_get_lease_time(clp, &fsinfo);
105 	if (status == 0) {
106 		nfs4_set_lease_period(clp, fsinfo.lease_time * HZ);
107 		nfs4_schedule_state_renewal(clp);
108 	}
109 
110 	return status;
111 }
112 
113 int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
114 {
115 	struct nfs4_setclientid_res clid = {
116 		.clientid = clp->cl_clientid,
117 		.confirm = clp->cl_confirm,
118 	};
119 	unsigned short port;
120 	int status;
121 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
122 
123 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
124 		goto do_confirm;
125 	port = nn->nfs_callback_tcpport;
126 	if (clp->cl_addr.ss_family == AF_INET6)
127 		port = nn->nfs_callback_tcpport6;
128 
129 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
130 	if (status != 0)
131 		goto out;
132 	clp->cl_clientid = clid.clientid;
133 	clp->cl_confirm = clid.confirm;
134 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
135 do_confirm:
136 	status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
137 	if (status != 0)
138 		goto out;
139 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
140 	nfs4_setup_state_renewal(clp);
141 out:
142 	return status;
143 }
144 
145 /**
146  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
147  *
148  * @clp: nfs_client under test
149  * @result: OUT: found nfs_client, or clp
150  * @cred: credential to use for trunking test
151  *
152  * Returns zero, a negative errno, or a negative NFS4ERR status.
153  * If zero is returned, an nfs_client pointer is planted in
154  * "result".
155  *
156  * Note: The returned client may not yet be marked ready.
157  */
158 int nfs40_discover_server_trunking(struct nfs_client *clp,
159 				   struct nfs_client **result,
160 				   const struct cred *cred)
161 {
162 	struct nfs4_setclientid_res clid = {
163 		.clientid = clp->cl_clientid,
164 		.confirm = clp->cl_confirm,
165 	};
166 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
167 	unsigned short port;
168 	int status;
169 
170 	port = nn->nfs_callback_tcpport;
171 	if (clp->cl_addr.ss_family == AF_INET6)
172 		port = nn->nfs_callback_tcpport6;
173 
174 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
175 	if (status != 0)
176 		goto out;
177 	clp->cl_clientid = clid.clientid;
178 	clp->cl_confirm = clid.confirm;
179 
180 	status = nfs40_walk_client_list(clp, result, cred);
181 	if (status == 0) {
182 		/* Sustain the lease, even if it's empty.  If the clientid4
183 		 * goes stale it's of no use for trunking discovery. */
184 		nfs4_schedule_state_renewal(*result);
185 
186 		/* If the client state need to recover, do it. */
187 		if (clp->cl_state)
188 			nfs4_schedule_state_manager(clp);
189 	}
190 out:
191 	return status;
192 }
193 
194 const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
195 {
196 	return get_cred(rpc_machine_cred());
197 }
198 
199 static void nfs4_root_machine_cred(struct nfs_client *clp)
200 {
201 
202 	/* Force root creds instead of machine */
203 	clp->cl_principal = NULL;
204 	clp->cl_rpcclient->cl_principal = NULL;
205 }
206 
207 static const struct cred *
208 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
209 {
210 	const struct cred *cred = NULL;
211 	struct nfs4_state_owner *sp;
212 	struct rb_node *pos;
213 
214 	for (pos = rb_first(&server->state_owners);
215 	     pos != NULL;
216 	     pos = rb_next(pos)) {
217 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
218 		if (list_empty(&sp->so_states))
219 			continue;
220 		cred = get_cred(sp->so_cred);
221 		break;
222 	}
223 	return cred;
224 }
225 
226 /**
227  * nfs4_get_renew_cred - Acquire credential for a renew operation
228  * @clp: client state handle
229  *
230  * Returns an rpc_cred with reference count bumped, or NULL.
231  * Caller must hold clp->cl_lock.
232  */
233 const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
234 {
235 	const struct cred *cred = NULL;
236 	struct nfs_server *server;
237 
238 	/* Use machine credentials if available */
239 	cred = nfs4_get_machine_cred(clp);
240 	if (cred != NULL)
241 		goto out;
242 
243 	spin_lock(&clp->cl_lock);
244 	rcu_read_lock();
245 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
246 		cred = nfs4_get_renew_cred_server_locked(server);
247 		if (cred != NULL)
248 			break;
249 	}
250 	rcu_read_unlock();
251 	spin_unlock(&clp->cl_lock);
252 
253 out:
254 	return cred;
255 }
256 
257 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
258 {
259 	if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
260 		spin_lock(&tbl->slot_tbl_lock);
261 		nfs41_wake_slot_table(tbl);
262 		spin_unlock(&tbl->slot_tbl_lock);
263 	}
264 }
265 
266 static void nfs4_end_drain_session(struct nfs_client *clp)
267 {
268 	struct nfs4_session *ses = clp->cl_session;
269 
270 	if (clp->cl_slot_tbl) {
271 		nfs4_end_drain_slot_table(clp->cl_slot_tbl);
272 		return;
273 	}
274 
275 	if (ses != NULL) {
276 		nfs4_end_drain_slot_table(&ses->bc_slot_table);
277 		nfs4_end_drain_slot_table(&ses->fc_slot_table);
278 	}
279 }
280 
281 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
282 {
283 	set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
284 	spin_lock(&tbl->slot_tbl_lock);
285 	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
286 		reinit_completion(&tbl->complete);
287 		spin_unlock(&tbl->slot_tbl_lock);
288 		return wait_for_completion_interruptible(&tbl->complete);
289 	}
290 	spin_unlock(&tbl->slot_tbl_lock);
291 	return 0;
292 }
293 
294 static int nfs4_begin_drain_session(struct nfs_client *clp)
295 {
296 	struct nfs4_session *ses = clp->cl_session;
297 	int ret;
298 
299 	if (clp->cl_slot_tbl)
300 		return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
301 
302 	/* back channel */
303 	ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
304 	if (ret)
305 		return ret;
306 	/* fore channel */
307 	return nfs4_drain_slot_tbl(&ses->fc_slot_table);
308 }
309 
310 #if defined(CONFIG_NFS_V4_1)
311 
312 static void nfs41_finish_session_reset(struct nfs_client *clp)
313 {
314 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
315 	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
316 	/* create_session negotiated new slot table */
317 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
318 	nfs4_setup_state_renewal(clp);
319 }
320 
321 int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
322 {
323 	int status;
324 
325 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
326 		goto do_confirm;
327 	status = nfs4_proc_exchange_id(clp, cred);
328 	if (status != 0)
329 		goto out;
330 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
331 do_confirm:
332 	status = nfs4_proc_create_session(clp, cred);
333 	if (status != 0)
334 		goto out;
335 	if (!(clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R))
336 		nfs4_state_start_reclaim_reboot(clp);
337 	nfs41_finish_session_reset(clp);
338 	nfs_mark_client_ready(clp, NFS_CS_READY);
339 out:
340 	return status;
341 }
342 
343 /**
344  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
345  *
346  * @clp: nfs_client under test
347  * @result: OUT: found nfs_client, or clp
348  * @cred: credential to use for trunking test
349  *
350  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
351  * If NFS4_OK is returned, an nfs_client pointer is planted in
352  * "result".
353  *
354  * Note: The returned client may not yet be marked ready.
355  */
356 int nfs41_discover_server_trunking(struct nfs_client *clp,
357 				   struct nfs_client **result,
358 				   const struct cred *cred)
359 {
360 	int status;
361 
362 	status = nfs4_proc_exchange_id(clp, cred);
363 	if (status != NFS4_OK)
364 		return status;
365 
366 	status = nfs41_walk_client_list(clp, result, cred);
367 	if (status < 0)
368 		return status;
369 	if (clp != *result)
370 		return 0;
371 
372 	/*
373 	 * Purge state if the client id was established in a prior
374 	 * instance and the client id could not have arrived on the
375 	 * server via Transparent State Migration.
376 	 */
377 	if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
378 		if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
379 			set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
380 		else
381 			set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
382 	}
383 	nfs4_schedule_state_manager(clp);
384 	status = nfs_wait_client_init_complete(clp);
385 	if (status < 0)
386 		nfs_put_client(clp);
387 	return status;
388 }
389 
390 #endif /* CONFIG_NFS_V4_1 */
391 
392 /**
393  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
394  * @clp: client state handle
395  *
396  * Returns a cred with reference count bumped, or NULL.
397  */
398 const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
399 {
400 	const struct cred *cred;
401 
402 	cred = nfs4_get_machine_cred(clp);
403 	return cred;
404 }
405 
406 static struct nfs4_state_owner *
407 nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
408 {
409 	struct rb_node **p = &server->state_owners.rb_node,
410 		       *parent = NULL;
411 	struct nfs4_state_owner *sp;
412 	int cmp;
413 
414 	while (*p != NULL) {
415 		parent = *p;
416 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
417 		cmp = cred_fscmp(cred, sp->so_cred);
418 
419 		if (cmp < 0)
420 			p = &parent->rb_left;
421 		else if (cmp > 0)
422 			p = &parent->rb_right;
423 		else {
424 			if (!list_empty(&sp->so_lru))
425 				list_del_init(&sp->so_lru);
426 			atomic_inc(&sp->so_count);
427 			return sp;
428 		}
429 	}
430 	return NULL;
431 }
432 
433 static struct nfs4_state_owner *
434 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
435 {
436 	struct nfs_server *server = new->so_server;
437 	struct rb_node **p = &server->state_owners.rb_node,
438 		       *parent = NULL;
439 	struct nfs4_state_owner *sp;
440 	int cmp;
441 
442 	while (*p != NULL) {
443 		parent = *p;
444 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
445 		cmp = cred_fscmp(new->so_cred, sp->so_cred);
446 
447 		if (cmp < 0)
448 			p = &parent->rb_left;
449 		else if (cmp > 0)
450 			p = &parent->rb_right;
451 		else {
452 			if (!list_empty(&sp->so_lru))
453 				list_del_init(&sp->so_lru);
454 			atomic_inc(&sp->so_count);
455 			return sp;
456 		}
457 	}
458 	rb_link_node(&new->so_server_node, parent, p);
459 	rb_insert_color(&new->so_server_node, &server->state_owners);
460 	return new;
461 }
462 
463 static void
464 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
465 {
466 	struct nfs_server *server = sp->so_server;
467 
468 	if (!RB_EMPTY_NODE(&sp->so_server_node))
469 		rb_erase(&sp->so_server_node, &server->state_owners);
470 }
471 
472 static void
473 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
474 {
475 	sc->create_time = ktime_get();
476 	sc->flags = 0;
477 	sc->counter = 0;
478 	spin_lock_init(&sc->lock);
479 	INIT_LIST_HEAD(&sc->list);
480 	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
481 }
482 
483 static void
484 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
485 {
486 	rpc_destroy_wait_queue(&sc->wait);
487 }
488 
489 /*
490  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
491  * create a new state_owner.
492  *
493  */
494 static struct nfs4_state_owner *
495 nfs4_alloc_state_owner(struct nfs_server *server,
496 		const struct cred *cred,
497 		gfp_t gfp_flags)
498 {
499 	struct nfs4_state_owner *sp;
500 
501 	sp = kzalloc(sizeof(*sp), gfp_flags);
502 	if (!sp)
503 		return NULL;
504 	sp->so_seqid.owner_id = ida_alloc(&server->openowner_id, gfp_flags);
505 	if (sp->so_seqid.owner_id < 0) {
506 		kfree(sp);
507 		return NULL;
508 	}
509 	sp->so_server = server;
510 	sp->so_cred = get_cred(cred);
511 	spin_lock_init(&sp->so_lock);
512 	INIT_LIST_HEAD(&sp->so_states);
513 	nfs4_init_seqid_counter(&sp->so_seqid);
514 	atomic_set(&sp->so_count, 1);
515 	INIT_LIST_HEAD(&sp->so_lru);
516 	seqcount_spinlock_init(&sp->so_reclaim_seqcount, &sp->so_lock);
517 	mutex_init(&sp->so_delegreturn_mutex);
518 	return sp;
519 }
520 
521 static void
522 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
523 {
524 	/* This state_owner is no longer usable, but must
525 	 * remain in place so that state recovery can find it
526 	 * and the opens associated with it.
527 	 * It may also be used for new 'open' request to
528 	 * return a delegation to the server.
529 	 * So update the 'create_time' so that it looks like
530 	 * a new state_owner.  This will cause the server to
531 	 * request an OPEN_CONFIRM to start a new sequence.
532 	 */
533 	sp->so_seqid.create_time = ktime_get();
534 }
535 
536 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
537 {
538 	nfs4_destroy_seqid_counter(&sp->so_seqid);
539 	put_cred(sp->so_cred);
540 	ida_free(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
541 	kfree(sp);
542 }
543 
544 static void nfs4_gc_state_owners(struct nfs_server *server)
545 {
546 	struct nfs_client *clp = server->nfs_client;
547 	struct nfs4_state_owner *sp, *tmp;
548 	unsigned long time_min, time_max;
549 	LIST_HEAD(doomed);
550 
551 	spin_lock(&clp->cl_lock);
552 	time_max = jiffies;
553 	time_min = (long)time_max - (long)clp->cl_lease_time;
554 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
555 		/* NB: LRU is sorted so that oldest is at the head */
556 		if (time_in_range(sp->so_expires, time_min, time_max))
557 			break;
558 		list_move(&sp->so_lru, &doomed);
559 		nfs4_remove_state_owner_locked(sp);
560 	}
561 	spin_unlock(&clp->cl_lock);
562 
563 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
564 		list_del(&sp->so_lru);
565 		nfs4_free_state_owner(sp);
566 	}
567 }
568 
569 /**
570  * nfs4_get_state_owner - Look up a state owner given a credential
571  * @server: nfs_server to search
572  * @cred: RPC credential to match
573  * @gfp_flags: allocation mode
574  *
575  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
576  */
577 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
578 					      const struct cred *cred,
579 					      gfp_t gfp_flags)
580 {
581 	struct nfs_client *clp = server->nfs_client;
582 	struct nfs4_state_owner *sp, *new;
583 
584 	spin_lock(&clp->cl_lock);
585 	sp = nfs4_find_state_owner_locked(server, cred);
586 	spin_unlock(&clp->cl_lock);
587 	if (sp != NULL)
588 		goto out;
589 	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
590 	if (new == NULL)
591 		goto out;
592 	spin_lock(&clp->cl_lock);
593 	sp = nfs4_insert_state_owner_locked(new);
594 	spin_unlock(&clp->cl_lock);
595 	if (sp != new)
596 		nfs4_free_state_owner(new);
597 out:
598 	nfs4_gc_state_owners(server);
599 	return sp;
600 }
601 
602 /**
603  * nfs4_put_state_owner - Release a nfs4_state_owner
604  * @sp: state owner data to release
605  *
606  * Note that we keep released state owners on an LRU
607  * list.
608  * This caches valid state owners so that they can be
609  * reused, to avoid the OPEN_CONFIRM on minor version 0.
610  * It also pins the uniquifier of dropped state owners for
611  * a while, to ensure that those state owner names are
612  * never reused.
613  */
614 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
615 {
616 	struct nfs_server *server = sp->so_server;
617 	struct nfs_client *clp = server->nfs_client;
618 
619 	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
620 		return;
621 
622 	sp->so_expires = jiffies;
623 	list_add_tail(&sp->so_lru, &server->state_owners_lru);
624 	spin_unlock(&clp->cl_lock);
625 }
626 
627 /**
628  * nfs4_purge_state_owners - Release all cached state owners
629  * @server: nfs_server with cached state owners to release
630  * @head: resulting list of state owners
631  *
632  * Called at umount time.  Remaining state owners will be on
633  * the LRU with ref count of zero.
634  * Note that the state owners are not freed, but are added
635  * to the list @head, which can later be used as an argument
636  * to nfs4_free_state_owners.
637  */
638 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
639 {
640 	struct nfs_client *clp = server->nfs_client;
641 	struct nfs4_state_owner *sp, *tmp;
642 
643 	spin_lock(&clp->cl_lock);
644 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
645 		list_move(&sp->so_lru, head);
646 		nfs4_remove_state_owner_locked(sp);
647 	}
648 	spin_unlock(&clp->cl_lock);
649 }
650 
651 /**
652  * nfs4_free_state_owners - Release all cached state owners
653  * @head: resulting list of state owners
654  *
655  * Frees a list of state owners that was generated by
656  * nfs4_purge_state_owners
657  */
658 void nfs4_free_state_owners(struct list_head *head)
659 {
660 	struct nfs4_state_owner *sp, *tmp;
661 
662 	list_for_each_entry_safe(sp, tmp, head, so_lru) {
663 		list_del(&sp->so_lru);
664 		nfs4_free_state_owner(sp);
665 	}
666 }
667 
668 static struct nfs4_state *
669 nfs4_alloc_open_state(void)
670 {
671 	struct nfs4_state *state;
672 
673 	state = kzalloc(sizeof(*state), GFP_KERNEL_ACCOUNT);
674 	if (!state)
675 		return NULL;
676 	refcount_set(&state->count, 1);
677 	INIT_LIST_HEAD(&state->lock_states);
678 	spin_lock_init(&state->state_lock);
679 	seqlock_init(&state->seqlock);
680 	init_waitqueue_head(&state->waitq);
681 	return state;
682 }
683 
684 void
685 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
686 {
687 	if (state->state == fmode)
688 		return;
689 	/* NB! List reordering - see the reclaim code for why.  */
690 	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
691 		if (fmode & FMODE_WRITE)
692 			list_move(&state->open_states, &state->owner->so_states);
693 		else
694 			list_move_tail(&state->open_states, &state->owner->so_states);
695 	}
696 	state->state = fmode;
697 }
698 
699 static struct nfs4_state *
700 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
701 {
702 	struct nfs_inode *nfsi = NFS_I(inode);
703 	struct nfs4_state *state;
704 
705 	list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
706 		if (state->owner != owner)
707 			continue;
708 		if (!nfs4_valid_open_stateid(state))
709 			continue;
710 		if (refcount_inc_not_zero(&state->count))
711 			return state;
712 	}
713 	return NULL;
714 }
715 
716 static void
717 nfs4_free_open_state(struct nfs4_state *state)
718 {
719 	kfree_rcu(state, rcu_head);
720 }
721 
722 struct nfs4_state *
723 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
724 {
725 	struct nfs4_state *state, *new;
726 	struct nfs_inode *nfsi = NFS_I(inode);
727 
728 	rcu_read_lock();
729 	state = __nfs4_find_state_byowner(inode, owner);
730 	rcu_read_unlock();
731 	if (state)
732 		goto out;
733 	new = nfs4_alloc_open_state();
734 	spin_lock(&owner->so_lock);
735 	spin_lock(&inode->i_lock);
736 	state = __nfs4_find_state_byowner(inode, owner);
737 	if (state == NULL && new != NULL) {
738 		state = new;
739 		state->owner = owner;
740 		atomic_inc(&owner->so_count);
741 		ihold(inode);
742 		state->inode = inode;
743 		list_add_rcu(&state->inode_states, &nfsi->open_states);
744 		spin_unlock(&inode->i_lock);
745 		/* Note: The reclaim code dictates that we add stateless
746 		 * and read-only stateids to the end of the list */
747 		list_add_tail(&state->open_states, &owner->so_states);
748 		spin_unlock(&owner->so_lock);
749 	} else {
750 		spin_unlock(&inode->i_lock);
751 		spin_unlock(&owner->so_lock);
752 		if (new)
753 			nfs4_free_open_state(new);
754 	}
755 out:
756 	return state;
757 }
758 
759 void nfs4_put_open_state(struct nfs4_state *state)
760 {
761 	struct inode *inode = state->inode;
762 	struct nfs4_state_owner *owner = state->owner;
763 
764 	if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
765 		return;
766 	spin_lock(&inode->i_lock);
767 	list_del_rcu(&state->inode_states);
768 	list_del(&state->open_states);
769 	spin_unlock(&inode->i_lock);
770 	spin_unlock(&owner->so_lock);
771 	nfs4_inode_return_delegation_on_close(inode);
772 	iput(inode);
773 	nfs4_free_open_state(state);
774 	nfs4_put_state_owner(owner);
775 }
776 
777 /*
778  * Close the current file.
779  */
780 static void __nfs4_close(struct nfs4_state *state,
781 		fmode_t fmode, gfp_t gfp_mask, int wait)
782 {
783 	struct nfs4_state_owner *owner = state->owner;
784 	int call_close = 0;
785 	fmode_t newstate;
786 
787 	atomic_inc(&owner->so_count);
788 	/* Protect against nfs4_find_state() */
789 	spin_lock(&owner->so_lock);
790 	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
791 		case FMODE_READ:
792 			state->n_rdonly--;
793 			break;
794 		case FMODE_WRITE:
795 			state->n_wronly--;
796 			break;
797 		case FMODE_READ|FMODE_WRITE:
798 			state->n_rdwr--;
799 	}
800 	newstate = FMODE_READ|FMODE_WRITE;
801 	if (state->n_rdwr == 0) {
802 		if (state->n_rdonly == 0) {
803 			newstate &= ~FMODE_READ;
804 			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
805 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
806 		}
807 		if (state->n_wronly == 0) {
808 			newstate &= ~FMODE_WRITE;
809 			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
810 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
811 		}
812 		if (newstate == 0)
813 			clear_bit(NFS_DELEGATED_STATE, &state->flags);
814 	}
815 	nfs4_state_set_mode_locked(state, newstate);
816 	spin_unlock(&owner->so_lock);
817 
818 	if (!call_close) {
819 		nfs4_put_open_state(state);
820 		nfs4_put_state_owner(owner);
821 	} else
822 		nfs4_do_close(state, gfp_mask, wait);
823 }
824 
825 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
826 {
827 	__nfs4_close(state, fmode, GFP_KERNEL, 0);
828 }
829 
830 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
831 {
832 	__nfs4_close(state, fmode, GFP_KERNEL, 1);
833 }
834 
835 /*
836  * Search the state->lock_states for an existing lock_owner
837  * that is compatible with either of the given owners.
838  * If the second is non-zero, then the first refers to a Posix-lock
839  * owner (current->files) and the second refers to a flock/OFD
840  * owner (struct file*).  In that case, prefer a match for the first
841  * owner.
842  * If both sorts of locks are held on the one file we cannot know
843  * which stateid was intended to be used, so a "correct" choice cannot
844  * be made.  Failing that, a "consistent" choice is preferable.  The
845  * consistent choice we make is to prefer the first owner, that of a
846  * Posix lock.
847  */
848 static struct nfs4_lock_state *
849 __nfs4_find_lock_state(struct nfs4_state *state,
850 		       fl_owner_t owner, fl_owner_t owner2)
851 {
852 	struct nfs4_lock_state *pos, *ret = NULL;
853 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
854 		if (pos->ls_owner == owner) {
855 			ret = pos;
856 			break;
857 		}
858 		if (pos->ls_owner == owner2)
859 			ret = pos;
860 	}
861 	if (ret)
862 		refcount_inc(&ret->ls_count);
863 	return ret;
864 }
865 
866 /*
867  * Return a compatible lock_state. If no initialized lock_state structure
868  * exists, return an uninitialized one.
869  *
870  */
871 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t owner)
872 {
873 	struct nfs4_lock_state *lsp;
874 	struct nfs_server *server = state->owner->so_server;
875 
876 	lsp = kzalloc(sizeof(*lsp), GFP_KERNEL_ACCOUNT);
877 	if (lsp == NULL)
878 		return NULL;
879 	nfs4_init_seqid_counter(&lsp->ls_seqid);
880 	refcount_set(&lsp->ls_count, 1);
881 	lsp->ls_state = state;
882 	lsp->ls_owner = owner;
883 	lsp->ls_seqid.owner_id = ida_alloc(&server->lockowner_id, GFP_KERNEL_ACCOUNT);
884 	if (lsp->ls_seqid.owner_id < 0)
885 		goto out_free;
886 	INIT_LIST_HEAD(&lsp->ls_locks);
887 	return lsp;
888 out_free:
889 	kfree(lsp);
890 	return NULL;
891 }
892 
893 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
894 {
895 	ida_free(&server->lockowner_id, lsp->ls_seqid.owner_id);
896 	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
897 	kfree(lsp);
898 }
899 
900 /*
901  * Return a compatible lock_state. If no initialized lock_state structure
902  * exists, return an uninitialized one.
903  *
904  */
905 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
906 {
907 	struct nfs4_lock_state *lsp, *new = NULL;
908 
909 	for(;;) {
910 		spin_lock(&state->state_lock);
911 		lsp = __nfs4_find_lock_state(state, owner, NULL);
912 		if (lsp != NULL)
913 			break;
914 		if (new != NULL) {
915 			list_add(&new->ls_locks, &state->lock_states);
916 			set_bit(LK_STATE_IN_USE, &state->flags);
917 			lsp = new;
918 			new = NULL;
919 			break;
920 		}
921 		spin_unlock(&state->state_lock);
922 		new = nfs4_alloc_lock_state(state, owner);
923 		if (new == NULL)
924 			return NULL;
925 	}
926 	spin_unlock(&state->state_lock);
927 	if (new != NULL)
928 		nfs4_free_lock_state(state->owner->so_server, new);
929 	return lsp;
930 }
931 
932 /*
933  * Release reference to lock_state, and free it if we see that
934  * it is no longer in use
935  */
936 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
937 {
938 	struct nfs_server *server;
939 	struct nfs4_state *state;
940 
941 	if (lsp == NULL)
942 		return;
943 	state = lsp->ls_state;
944 	if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
945 		return;
946 	list_del(&lsp->ls_locks);
947 	if (list_empty(&state->lock_states))
948 		clear_bit(LK_STATE_IN_USE, &state->flags);
949 	spin_unlock(&state->state_lock);
950 	server = state->owner->so_server;
951 	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
952 		struct nfs_client *clp = server->nfs_client;
953 
954 		clp->cl_mvops->free_lock_state(server, lsp);
955 	} else
956 		nfs4_free_lock_state(server, lsp);
957 }
958 
959 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
960 {
961 	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
962 
963 	dst->fl_u.nfs4_fl.owner = lsp;
964 	refcount_inc(&lsp->ls_count);
965 }
966 
967 static void nfs4_fl_release_lock(struct file_lock *fl)
968 {
969 	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
970 }
971 
972 static const struct file_lock_operations nfs4_fl_lock_ops = {
973 	.fl_copy_lock = nfs4_fl_copy_lock,
974 	.fl_release_private = nfs4_fl_release_lock,
975 };
976 
977 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
978 {
979 	struct nfs4_lock_state *lsp;
980 
981 	if (fl->fl_ops != NULL)
982 		return 0;
983 	lsp = nfs4_get_lock_state(state, fl->c.flc_owner);
984 	if (lsp == NULL)
985 		return -ENOMEM;
986 	fl->fl_u.nfs4_fl.owner = lsp;
987 	fl->fl_ops = &nfs4_fl_lock_ops;
988 	return 0;
989 }
990 
991 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
992 		struct nfs4_state *state,
993 		const struct nfs_lock_context *l_ctx)
994 {
995 	struct nfs4_lock_state *lsp;
996 	fl_owner_t owner, fl_flock_owner;
997 	int ret = -ENOENT;
998 
999 	if (l_ctx == NULL)
1000 		goto out;
1001 
1002 	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1003 		goto out;
1004 
1005 	owner = l_ctx->lockowner;
1006 	fl_flock_owner = l_ctx->open_context->flock_owner;
1007 
1008 	spin_lock(&state->state_lock);
1009 	lsp = __nfs4_find_lock_state(state, owner, fl_flock_owner);
1010 	if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1011 		ret = -EIO;
1012 	else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1013 		nfs4_stateid_copy(dst, &lsp->ls_stateid);
1014 		ret = 0;
1015 	}
1016 	spin_unlock(&state->state_lock);
1017 	nfs4_put_lock_state(lsp);
1018 out:
1019 	return ret;
1020 }
1021 
1022 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1023 {
1024 	bool ret;
1025 	const nfs4_stateid *src;
1026 	int seq;
1027 
1028 	do {
1029 		ret = false;
1030 		src = &zero_stateid;
1031 		seq = read_seqbegin(&state->seqlock);
1032 		if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1033 			src = &state->open_stateid;
1034 			ret = true;
1035 		}
1036 		nfs4_stateid_copy(dst, src);
1037 	} while (read_seqretry(&state->seqlock, seq));
1038 	return ret;
1039 }
1040 
1041 /*
1042  * Byte-range lock aware utility to initialize the stateid of read/write
1043  * requests.
1044  */
1045 int nfs4_select_rw_stateid(struct nfs4_state *state,
1046 		fmode_t fmode, const struct nfs_lock_context *l_ctx,
1047 		nfs4_stateid *dst, const struct cred **cred)
1048 {
1049 	int ret;
1050 
1051 	if (!nfs4_valid_open_stateid(state))
1052 		return -EIO;
1053 	if (cred != NULL)
1054 		*cred = NULL;
1055 	ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1056 	if (ret == -EIO)
1057 		/* A lost lock - don't even consider delegations */
1058 		goto out;
1059 	/* returns true if delegation stateid found and copied */
1060 	if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1061 		ret = 0;
1062 		goto out;
1063 	}
1064 	if (ret != -ENOENT)
1065 		/* nfs4_copy_delegation_stateid() didn't over-write
1066 		 * dst, so it still has the lock stateid which we now
1067 		 * choose to use.
1068 		 */
1069 		goto out;
1070 	ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1071 out:
1072 	if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1073 		dst->seqid = 0;
1074 	return ret;
1075 }
1076 
1077 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1078 {
1079 	struct nfs_seqid *new;
1080 
1081 	new = kmalloc(sizeof(*new), gfp_mask);
1082 	if (new == NULL)
1083 		return ERR_PTR(-ENOMEM);
1084 	new->sequence = counter;
1085 	INIT_LIST_HEAD(&new->list);
1086 	new->task = NULL;
1087 	return new;
1088 }
1089 
1090 void nfs_release_seqid(struct nfs_seqid *seqid)
1091 {
1092 	struct nfs_seqid_counter *sequence;
1093 
1094 	if (seqid == NULL || list_empty(&seqid->list))
1095 		return;
1096 	sequence = seqid->sequence;
1097 	spin_lock(&sequence->lock);
1098 	list_del_init(&seqid->list);
1099 	if (!list_empty(&sequence->list)) {
1100 		struct nfs_seqid *next;
1101 
1102 		next = list_first_entry(&sequence->list,
1103 				struct nfs_seqid, list);
1104 		rpc_wake_up_queued_task(&sequence->wait, next->task);
1105 	}
1106 	spin_unlock(&sequence->lock);
1107 }
1108 
1109 void nfs_free_seqid(struct nfs_seqid *seqid)
1110 {
1111 	nfs_release_seqid(seqid);
1112 	kfree(seqid);
1113 }
1114 
1115 /*
1116  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1117  * failed with a seqid incrementing error -
1118  * see comments nfs4.h:seqid_mutating_error()
1119  */
1120 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1121 {
1122 	switch (status) {
1123 		case 0:
1124 			break;
1125 		case -NFS4ERR_BAD_SEQID:
1126 			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1127 				return;
1128 			pr_warn_ratelimited("NFS: v4 server returned a bad"
1129 					" sequence-id error on an"
1130 					" unconfirmed sequence %p!\n",
1131 					seqid->sequence);
1132 			return;
1133 		case -NFS4ERR_STALE_CLIENTID:
1134 		case -NFS4ERR_STALE_STATEID:
1135 		case -NFS4ERR_BAD_STATEID:
1136 		case -NFS4ERR_BADXDR:
1137 		case -NFS4ERR_RESOURCE:
1138 		case -NFS4ERR_NOFILEHANDLE:
1139 		case -NFS4ERR_MOVED:
1140 			/* Non-seqid mutating errors */
1141 			return;
1142 	}
1143 	/*
1144 	 * Note: no locking needed as we are guaranteed to be first
1145 	 * on the sequence list
1146 	 */
1147 	seqid->sequence->counter++;
1148 }
1149 
1150 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1151 {
1152 	struct nfs4_state_owner *sp;
1153 
1154 	if (seqid == NULL)
1155 		return;
1156 
1157 	sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1158 	if (status == -NFS4ERR_BAD_SEQID)
1159 		nfs4_reset_state_owner(sp);
1160 	if (!nfs4_has_session(sp->so_server->nfs_client))
1161 		nfs_increment_seqid(status, seqid);
1162 }
1163 
1164 /*
1165  * Increment the seqid if the LOCK/LOCKU succeeded, or
1166  * failed with a seqid incrementing error -
1167  * see comments nfs4.h:seqid_mutating_error()
1168  */
1169 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1170 {
1171 	if (seqid != NULL)
1172 		nfs_increment_seqid(status, seqid);
1173 }
1174 
1175 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1176 {
1177 	struct nfs_seqid_counter *sequence;
1178 	int status = 0;
1179 
1180 	if (seqid == NULL)
1181 		goto out;
1182 	sequence = seqid->sequence;
1183 	spin_lock(&sequence->lock);
1184 	seqid->task = task;
1185 	if (list_empty(&seqid->list))
1186 		list_add_tail(&seqid->list, &sequence->list);
1187 	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1188 		goto unlock;
1189 	rpc_sleep_on(&sequence->wait, task, NULL);
1190 	status = -EAGAIN;
1191 unlock:
1192 	spin_unlock(&sequence->lock);
1193 out:
1194 	return status;
1195 }
1196 
1197 static int nfs4_run_state_manager(void *);
1198 
1199 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1200 {
1201 	clear_and_wake_up_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
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 	struct rpc_clnt *clnt = clp->cl_rpcclient;
1213 	bool swapon = false;
1214 
1215 	if (clnt->cl_shutdown)
1216 		return;
1217 
1218 	set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1219 
1220 	if (atomic_read(&clnt->cl_swapper)) {
1221 		swapon = !test_and_set_bit(NFS4CLNT_MANAGER_AVAILABLE,
1222 					   &clp->cl_state);
1223 		if (!swapon) {
1224 			wake_up_var(&clp->cl_state);
1225 			return;
1226 		}
1227 	}
1228 
1229 	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1230 		return;
1231 
1232 	__module_get(THIS_MODULE);
1233 	refcount_inc(&clp->cl_count);
1234 
1235 	/* The rcu_read_lock() is not strictly necessary, as the state
1236 	 * manager is the only thread that ever changes the rpc_xprt
1237 	 * after it's initialized.  At this point, we're single threaded. */
1238 	rcu_read_lock();
1239 	snprintf(buf, sizeof(buf), "%s-manager",
1240 			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1241 	rcu_read_unlock();
1242 	task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1243 	if (IS_ERR(task)) {
1244 		printk(KERN_ERR "%s: kthread_run: %ld\n",
1245 			__func__, PTR_ERR(task));
1246 		if (!nfs_client_init_is_complete(clp))
1247 			nfs_mark_client_ready(clp, PTR_ERR(task));
1248 		if (swapon)
1249 			clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
1250 		nfs4_clear_state_manager_bit(clp);
1251 		nfs_put_client(clp);
1252 		module_put(THIS_MODULE);
1253 	}
1254 }
1255 
1256 /*
1257  * Schedule a lease recovery attempt
1258  */
1259 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1260 {
1261 	if (!clp)
1262 		return;
1263 	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1264 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1265 	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1266 			clp->cl_hostname);
1267 	nfs4_schedule_state_manager(clp);
1268 }
1269 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1270 
1271 /**
1272  * nfs4_schedule_migration_recovery - trigger migration recovery
1273  *
1274  * @server: FSID that is migrating
1275  *
1276  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1277  * value is returned.
1278  */
1279 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1280 {
1281 	struct nfs_client *clp = server->nfs_client;
1282 
1283 	if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1284 		pr_err("NFS: volatile file handles not supported (server %s)\n",
1285 				clp->cl_hostname);
1286 		return -NFS4ERR_IO;
1287 	}
1288 
1289 	if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1290 		return -NFS4ERR_IO;
1291 
1292 	dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1293 			__func__,
1294 			(unsigned long long)server->fsid.major,
1295 			(unsigned long long)server->fsid.minor,
1296 			clp->cl_hostname);
1297 
1298 	set_bit(NFS_MIG_IN_TRANSITION,
1299 			&((struct nfs_server *)server)->mig_status);
1300 	set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1301 
1302 	nfs4_schedule_state_manager(clp);
1303 	return 0;
1304 }
1305 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1306 
1307 /**
1308  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1309  *
1310  * @clp: server to check for moved leases
1311  *
1312  */
1313 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1314 {
1315 	dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1316 		__func__, clp->cl_clientid, clp->cl_hostname);
1317 
1318 	set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1319 	nfs4_schedule_state_manager(clp);
1320 }
1321 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1322 
1323 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1324 {
1325 	int res;
1326 
1327 	might_sleep();
1328 
1329 	refcount_inc(&clp->cl_count);
1330 	res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1331 				 nfs_wait_bit_killable,
1332 				 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1333 	if (res)
1334 		goto out;
1335 	if (clp->cl_cons_state < 0)
1336 		res = clp->cl_cons_state;
1337 out:
1338 	nfs_put_client(clp);
1339 	return res;
1340 }
1341 
1342 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1343 {
1344 	unsigned int loop;
1345 	int ret;
1346 
1347 	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1348 		ret = nfs4_wait_clnt_recover(clp);
1349 		if (ret != 0)
1350 			break;
1351 		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1352 		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1353 			break;
1354 		nfs4_schedule_state_manager(clp);
1355 		ret = -EIO;
1356 	}
1357 	return ret;
1358 }
1359 
1360 /*
1361  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1362  * @clp: client to process
1363  *
1364  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1365  * resend of the SETCLIENTID and hence re-establish the
1366  * callback channel. Then return all existing delegations.
1367  */
1368 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1369 {
1370 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1371 	nfs_expire_all_delegations(clp);
1372 	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1373 			clp->cl_hostname);
1374 }
1375 
1376 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1377 {
1378 	nfs40_handle_cb_pathdown(clp);
1379 	nfs4_schedule_state_manager(clp);
1380 }
1381 
1382 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1383 {
1384 
1385 	if (!nfs4_valid_open_stateid(state))
1386 		return 0;
1387 	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1388 	/* Don't recover state that expired before the reboot */
1389 	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1390 		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1391 		return 0;
1392 	}
1393 	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1394 	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1395 	return 1;
1396 }
1397 
1398 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1399 {
1400 	if (!nfs4_valid_open_stateid(state))
1401 		return 0;
1402 	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1403 	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1404 	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1405 	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1406 	return 1;
1407 }
1408 
1409 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1410 {
1411 	struct nfs_client *clp = server->nfs_client;
1412 
1413 	if (!nfs4_state_mark_reclaim_nograce(clp, state))
1414 		return -EBADF;
1415 	nfs_inode_find_delegation_state_and_recover(state->inode,
1416 			&state->stateid);
1417 	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1418 			clp->cl_hostname);
1419 	nfs4_schedule_state_manager(clp);
1420 	return 0;
1421 }
1422 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1423 
1424 static struct nfs4_lock_state *
1425 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1426 		const nfs4_stateid *stateid)
1427 {
1428 	struct nfs4_lock_state *pos;
1429 
1430 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
1431 		if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1432 			continue;
1433 		if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1434 			return pos;
1435 	}
1436 	return NULL;
1437 }
1438 
1439 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1440 		const nfs4_stateid *stateid)
1441 {
1442 	bool found = false;
1443 
1444 	if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1445 		spin_lock(&state->state_lock);
1446 		if (nfs_state_find_lock_state_by_stateid(state, stateid))
1447 			found = true;
1448 		spin_unlock(&state->state_lock);
1449 	}
1450 	return found;
1451 }
1452 
1453 void nfs_inode_find_state_and_recover(struct inode *inode,
1454 		const nfs4_stateid *stateid)
1455 {
1456 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1457 	struct nfs_inode *nfsi = NFS_I(inode);
1458 	struct nfs_open_context *ctx;
1459 	struct nfs4_state *state;
1460 	bool found = false;
1461 
1462 	rcu_read_lock();
1463 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1464 		state = ctx->state;
1465 		if (state == NULL)
1466 			continue;
1467 		if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1468 		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1469 			found = true;
1470 			continue;
1471 		}
1472 		if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1473 		    nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1474 		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1475 			found = true;
1476 			continue;
1477 		}
1478 		if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1479 		    nfs4_state_mark_reclaim_nograce(clp, state))
1480 			found = true;
1481 	}
1482 	rcu_read_unlock();
1483 
1484 	nfs_inode_find_delegation_state_and_recover(inode, stateid);
1485 	if (found)
1486 		nfs4_schedule_state_manager(clp);
1487 }
1488 
1489 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1490 {
1491 	struct inode *inode = state->inode;
1492 	struct nfs_inode *nfsi = NFS_I(inode);
1493 	struct nfs_open_context *ctx;
1494 
1495 	rcu_read_lock();
1496 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1497 		if (ctx->state != state)
1498 			continue;
1499 		set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1500 		pr_warn("NFSv4: state recovery failed for open file %pd2, "
1501 				"error = %d\n", ctx->dentry, err);
1502 	}
1503 	rcu_read_unlock();
1504 }
1505 
1506 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1507 {
1508 	set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1509 	nfs4_state_mark_open_context_bad(state, error);
1510 }
1511 
1512 
1513 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1514 {
1515 	struct inode *inode = state->inode;
1516 	struct nfs_inode *nfsi = NFS_I(inode);
1517 	struct file_lock *fl;
1518 	struct nfs4_lock_state *lsp;
1519 	int status = 0;
1520 	struct file_lock_context *flctx = locks_inode_context(inode);
1521 	struct list_head *list;
1522 
1523 	if (flctx == NULL)
1524 		return 0;
1525 
1526 	list = &flctx->flc_posix;
1527 
1528 	/* Guard against delegation returns and new lock/unlock calls */
1529 	down_write(&nfsi->rwsem);
1530 	spin_lock(&flctx->flc_lock);
1531 restart:
1532 	for_each_file_lock(fl, list) {
1533 		if (nfs_file_open_context(fl->c.flc_file)->state != state)
1534 			continue;
1535 		spin_unlock(&flctx->flc_lock);
1536 		status = ops->recover_lock(state, fl);
1537 		switch (status) {
1538 		case 0:
1539 			break;
1540 		case -ETIMEDOUT:
1541 		case -ESTALE:
1542 		case -NFS4ERR_ADMIN_REVOKED:
1543 		case -NFS4ERR_STALE_STATEID:
1544 		case -NFS4ERR_BAD_STATEID:
1545 		case -NFS4ERR_EXPIRED:
1546 		case -NFS4ERR_NO_GRACE:
1547 		case -NFS4ERR_STALE_CLIENTID:
1548 		case -NFS4ERR_BADSESSION:
1549 		case -NFS4ERR_BADSLOT:
1550 		case -NFS4ERR_BAD_HIGH_SLOT:
1551 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1552 			goto out;
1553 		default:
1554 			pr_err("NFS: %s: unhandled error %d\n",
1555 					__func__, status);
1556 			fallthrough;
1557 		case -ENOMEM:
1558 		case -NFS4ERR_DENIED:
1559 		case -NFS4ERR_RECLAIM_BAD:
1560 		case -NFS4ERR_RECLAIM_CONFLICT:
1561 			lsp = fl->fl_u.nfs4_fl.owner;
1562 			if (lsp)
1563 				set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1564 			status = 0;
1565 		}
1566 		spin_lock(&flctx->flc_lock);
1567 	}
1568 	if (list == &flctx->flc_posix) {
1569 		list = &flctx->flc_flock;
1570 		goto restart;
1571 	}
1572 	spin_unlock(&flctx->flc_lock);
1573 out:
1574 	up_write(&nfsi->rwsem);
1575 	return status;
1576 }
1577 
1578 #ifdef CONFIG_NFS_V4_2
1579 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1580 {
1581 	struct nfs4_copy_state *copy;
1582 
1583 	if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1584 		!test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1585 		return;
1586 
1587 	spin_lock(&sp->so_server->nfs_client->cl_lock);
1588 	list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1589 		if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1590 				!nfs4_stateid_match_other(&state->stateid,
1591 				&copy->parent_dst_state->stateid)))
1592 				continue;
1593 		copy->flags = 1;
1594 		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1595 				&state->flags)) {
1596 			clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1597 			complete(&copy->completion);
1598 		}
1599 	}
1600 	list_for_each_entry(copy, &sp->so_server->ss_copies, src_copies) {
1601 		if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1602 				!nfs4_stateid_match_other(&state->stateid,
1603 				&copy->parent_src_state->stateid)))
1604 				continue;
1605 		copy->flags = 1;
1606 		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1607 				&state->flags))
1608 			complete(&copy->completion);
1609 	}
1610 	spin_unlock(&sp->so_server->nfs_client->cl_lock);
1611 }
1612 #else /* !CONFIG_NFS_V4_2 */
1613 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1614 					 struct nfs4_state *state)
1615 {
1616 }
1617 #endif /* CONFIG_NFS_V4_2 */
1618 
1619 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1620 				     const struct nfs4_state_recovery_ops *ops,
1621 				     int *lost_locks)
1622 {
1623 	struct nfs4_lock_state *lock;
1624 	int status;
1625 
1626 	status = ops->recover_open(sp, state);
1627 	if (status < 0)
1628 		return status;
1629 
1630 	status = nfs4_reclaim_locks(state, ops);
1631 	if (status < 0)
1632 		return status;
1633 
1634 	if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1635 		spin_lock(&state->state_lock);
1636 		list_for_each_entry(lock, &state->lock_states, ls_locks) {
1637 			trace_nfs4_state_lock_reclaim(state, lock);
1638 			if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags) &&
1639 			    !test_bit(NFS_LOCK_UNLOCKING, &lock->ls_flags))
1640 				*lost_locks += 1;
1641 		}
1642 		spin_unlock(&state->state_lock);
1643 	}
1644 
1645 	nfs42_complete_copies(sp, state);
1646 	clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1647 	return status;
1648 }
1649 
1650 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp,
1651 				   const struct nfs4_state_recovery_ops *ops,
1652 				   int *lost_locks)
1653 {
1654 	struct nfs4_state *state;
1655 	unsigned int loop = 0;
1656 	int status = 0;
1657 #ifdef CONFIG_NFS_V4_2
1658 	bool found_ssc_copy_state = false;
1659 #endif /* CONFIG_NFS_V4_2 */
1660 
1661 	/* Note: we rely on the sp->so_states list being ordered
1662 	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1663 	 * states first.
1664 	 * This is needed to ensure that the server won't give us any
1665 	 * read delegations that we have to return if, say, we are
1666 	 * recovering after a network partition or a reboot from a
1667 	 * server that doesn't support a grace period.
1668 	 */
1669 	spin_lock(&sp->so_lock);
1670 	raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1671 restart:
1672 	list_for_each_entry(state, &sp->so_states, open_states) {
1673 		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1674 			continue;
1675 		if (!nfs4_valid_open_stateid(state))
1676 			continue;
1677 		if (state->state == 0)
1678 			continue;
1679 #ifdef CONFIG_NFS_V4_2
1680 		if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1681 			nfs4_state_mark_recovery_failed(state, -EIO);
1682 			found_ssc_copy_state = true;
1683 			continue;
1684 		}
1685 #endif /* CONFIG_NFS_V4_2 */
1686 		refcount_inc(&state->count);
1687 		spin_unlock(&sp->so_lock);
1688 		status = __nfs4_reclaim_open_state(sp, state, ops, lost_locks);
1689 
1690 		switch (status) {
1691 		default:
1692 			if (status >= 0) {
1693 				loop = 0;
1694 				break;
1695 			}
1696 			printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1697 			fallthrough;
1698 		case -ENOENT:
1699 		case -ENOMEM:
1700 		case -EACCES:
1701 		case -EROFS:
1702 		case -EIO:
1703 		case -ESTALE:
1704 			/* Open state on this file cannot be recovered */
1705 			nfs4_state_mark_recovery_failed(state, status);
1706 			break;
1707 		case -EAGAIN:
1708 			ssleep(1);
1709 			if (loop++ < 10) {
1710 				set_bit(ops->state_flag_bit, &state->flags);
1711 				break;
1712 			}
1713 			fallthrough;
1714 		case -NFS4ERR_ADMIN_REVOKED:
1715 		case -NFS4ERR_STALE_STATEID:
1716 		case -NFS4ERR_OLD_STATEID:
1717 		case -NFS4ERR_BAD_STATEID:
1718 		case -NFS4ERR_RECLAIM_BAD:
1719 		case -NFS4ERR_RECLAIM_CONFLICT:
1720 			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1721 			break;
1722 		case -NFS4ERR_EXPIRED:
1723 		case -NFS4ERR_NO_GRACE:
1724 			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1725 			fallthrough;
1726 		case -NFS4ERR_STALE_CLIENTID:
1727 		case -NFS4ERR_BADSESSION:
1728 		case -NFS4ERR_BADSLOT:
1729 		case -NFS4ERR_BAD_HIGH_SLOT:
1730 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1731 		case -ETIMEDOUT:
1732 			goto out_err;
1733 		}
1734 		nfs4_put_open_state(state);
1735 		spin_lock(&sp->so_lock);
1736 		goto restart;
1737 	}
1738 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1739 	spin_unlock(&sp->so_lock);
1740 #ifdef CONFIG_NFS_V4_2
1741 	if (found_ssc_copy_state)
1742 		return -EIO;
1743 #endif /* CONFIG_NFS_V4_2 */
1744 	return 0;
1745 out_err:
1746 	nfs4_put_open_state(state);
1747 	spin_lock(&sp->so_lock);
1748 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1749 	spin_unlock(&sp->so_lock);
1750 	return status;
1751 }
1752 
1753 static void nfs4_clear_open_state(struct nfs4_state *state)
1754 {
1755 	struct nfs4_lock_state *lock;
1756 
1757 	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1758 	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1759 	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1760 	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1761 	spin_lock(&state->state_lock);
1762 	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1763 		lock->ls_seqid.flags = 0;
1764 		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1765 	}
1766 	spin_unlock(&state->state_lock);
1767 }
1768 
1769 static void nfs4_reset_seqids(struct nfs_server *server,
1770 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1771 {
1772 	struct nfs_client *clp = server->nfs_client;
1773 	struct nfs4_state_owner *sp;
1774 	struct rb_node *pos;
1775 	struct nfs4_state *state;
1776 
1777 	spin_lock(&clp->cl_lock);
1778 	for (pos = rb_first(&server->state_owners);
1779 	     pos != NULL;
1780 	     pos = rb_next(pos)) {
1781 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1782 		sp->so_seqid.flags = 0;
1783 		spin_lock(&sp->so_lock);
1784 		list_for_each_entry(state, &sp->so_states, open_states) {
1785 			if (mark_reclaim(clp, state))
1786 				nfs4_clear_open_state(state);
1787 		}
1788 		spin_unlock(&sp->so_lock);
1789 	}
1790 	spin_unlock(&clp->cl_lock);
1791 }
1792 
1793 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1794 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1795 {
1796 	struct nfs_server *server;
1797 
1798 	rcu_read_lock();
1799 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1800 		nfs4_reset_seqids(server, mark_reclaim);
1801 	rcu_read_unlock();
1802 }
1803 
1804 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1805 {
1806 	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1807 	/* Mark all delegations for reclaim */
1808 	nfs_delegation_mark_reclaim(clp);
1809 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1810 }
1811 
1812 static int nfs4_reclaim_complete(struct nfs_client *clp,
1813 				 const struct nfs4_state_recovery_ops *ops,
1814 				 const struct cred *cred)
1815 {
1816 	/* Notify the server we're done reclaiming our state */
1817 	if (ops->reclaim_complete)
1818 		return ops->reclaim_complete(clp, cred);
1819 	return 0;
1820 }
1821 
1822 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1823 {
1824 	struct nfs_client *clp = server->nfs_client;
1825 	struct nfs4_state_owner *sp;
1826 	struct rb_node *pos;
1827 	struct nfs4_state *state;
1828 
1829 	spin_lock(&clp->cl_lock);
1830 	for (pos = rb_first(&server->state_owners);
1831 	     pos != NULL;
1832 	     pos = rb_next(pos)) {
1833 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1834 		spin_lock(&sp->so_lock);
1835 		list_for_each_entry(state, &sp->so_states, open_states) {
1836 			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1837 						&state->flags))
1838 				continue;
1839 			nfs4_state_mark_reclaim_nograce(clp, state);
1840 		}
1841 		spin_unlock(&sp->so_lock);
1842 	}
1843 	spin_unlock(&clp->cl_lock);
1844 }
1845 
1846 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1847 {
1848 	struct nfs_server *server;
1849 
1850 	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1851 		return 0;
1852 
1853 	rcu_read_lock();
1854 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1855 		nfs4_clear_reclaim_server(server);
1856 	rcu_read_unlock();
1857 
1858 	nfs_delegation_reap_unclaimed(clp);
1859 	return 1;
1860 }
1861 
1862 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1863 {
1864 	const struct nfs4_state_recovery_ops *ops;
1865 	const struct cred *cred;
1866 	int err;
1867 
1868 	if (!nfs4_state_clear_reclaim_reboot(clp))
1869 		return;
1870 	ops = clp->cl_mvops->reboot_recovery_ops;
1871 	cred = nfs4_get_clid_cred(clp);
1872 	err = nfs4_reclaim_complete(clp, ops, cred);
1873 	put_cred(cred);
1874 	if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1875 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1876 }
1877 
1878 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1879 {
1880 	nfs_mark_test_expired_all_delegations(clp);
1881 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1882 }
1883 
1884 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1885 {
1886 	switch (error) {
1887 	case 0:
1888 		break;
1889 	case -NFS4ERR_CB_PATH_DOWN:
1890 		nfs40_handle_cb_pathdown(clp);
1891 		break;
1892 	case -NFS4ERR_NO_GRACE:
1893 		nfs4_state_end_reclaim_reboot(clp);
1894 		break;
1895 	case -NFS4ERR_STALE_CLIENTID:
1896 		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1897 		nfs4_state_start_reclaim_reboot(clp);
1898 		break;
1899 	case -NFS4ERR_EXPIRED:
1900 		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1901 		nfs4_state_start_reclaim_nograce(clp);
1902 		break;
1903 	case -NFS4ERR_BADSESSION:
1904 	case -NFS4ERR_BADSLOT:
1905 	case -NFS4ERR_BAD_HIGH_SLOT:
1906 	case -NFS4ERR_DEADSESSION:
1907 	case -NFS4ERR_SEQ_FALSE_RETRY:
1908 	case -NFS4ERR_SEQ_MISORDERED:
1909 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1910 		/* Zero session reset errors */
1911 		break;
1912 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1913 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1914 		break;
1915 	default:
1916 		dprintk("%s: failed to handle error %d for server %s\n",
1917 				__func__, error, clp->cl_hostname);
1918 		return error;
1919 	}
1920 	dprintk("%s: handled error %d for server %s\n", __func__, error,
1921 			clp->cl_hostname);
1922 	return 0;
1923 }
1924 
1925 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1926 {
1927 	struct nfs4_state_owner *sp;
1928 	struct nfs_server *server;
1929 	struct rb_node *pos;
1930 	LIST_HEAD(freeme);
1931 	int status = 0;
1932 	int lost_locks = 0;
1933 
1934 restart:
1935 	rcu_read_lock();
1936 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1937 		nfs4_purge_state_owners(server, &freeme);
1938 		spin_lock(&clp->cl_lock);
1939 		for (pos = rb_first(&server->state_owners);
1940 		     pos != NULL;
1941 		     pos = rb_next(pos)) {
1942 			sp = rb_entry(pos,
1943 				struct nfs4_state_owner, so_server_node);
1944 			if (!test_and_clear_bit(ops->owner_flag_bit,
1945 							&sp->so_flags))
1946 				continue;
1947 			if (!atomic_inc_not_zero(&sp->so_count))
1948 				continue;
1949 			spin_unlock(&clp->cl_lock);
1950 			rcu_read_unlock();
1951 
1952 			status = nfs4_reclaim_open_state(sp, ops, &lost_locks);
1953 			if (status < 0) {
1954 				if (lost_locks)
1955 					pr_warn("NFS: %s: lost %d locks\n",
1956 						clp->cl_hostname, lost_locks);
1957 				set_bit(ops->owner_flag_bit, &sp->so_flags);
1958 				nfs4_put_state_owner(sp);
1959 				status = nfs4_recovery_handle_error(clp, status);
1960 				return (status != 0) ? status : -EAGAIN;
1961 			}
1962 
1963 			nfs4_put_state_owner(sp);
1964 			goto restart;
1965 		}
1966 		spin_unlock(&clp->cl_lock);
1967 	}
1968 	rcu_read_unlock();
1969 	nfs4_free_state_owners(&freeme);
1970 	if (lost_locks)
1971 		pr_warn("NFS: %s: lost %d locks\n",
1972 			clp->cl_hostname, lost_locks);
1973 	return 0;
1974 }
1975 
1976 static int nfs4_check_lease(struct nfs_client *clp)
1977 {
1978 	const struct cred *cred;
1979 	const struct nfs4_state_maintenance_ops *ops =
1980 		clp->cl_mvops->state_renewal_ops;
1981 	int status;
1982 
1983 	/* Is the client already known to have an expired lease? */
1984 	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1985 		return 0;
1986 	cred = ops->get_state_renewal_cred(clp);
1987 	if (cred == NULL) {
1988 		cred = nfs4_get_clid_cred(clp);
1989 		status = -ENOKEY;
1990 		if (cred == NULL)
1991 			goto out;
1992 	}
1993 	status = ops->renew_lease(clp, cred);
1994 	put_cred(cred);
1995 	if (status == -ETIMEDOUT) {
1996 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1997 		return 0;
1998 	}
1999 out:
2000 	return nfs4_recovery_handle_error(clp, status);
2001 }
2002 
2003 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
2004  * and for recoverable errors on EXCHANGE_ID for v4.1
2005  */
2006 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
2007 {
2008 	switch (status) {
2009 	case -NFS4ERR_SEQ_MISORDERED:
2010 		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
2011 			return -ESERVERFAULT;
2012 		/* Lease confirmation error: retry after purging the lease */
2013 		ssleep(1);
2014 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2015 		break;
2016 	case -NFS4ERR_STALE_CLIENTID:
2017 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2018 		nfs4_state_start_reclaim_reboot(clp);
2019 		break;
2020 	case -NFS4ERR_CLID_INUSE:
2021 		pr_err("NFS: Server %s reports our clientid is in use\n",
2022 			clp->cl_hostname);
2023 		nfs_mark_client_ready(clp, -EPERM);
2024 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2025 		return -EPERM;
2026 	case -EACCES:
2027 	case -NFS4ERR_DELAY:
2028 	case -EAGAIN:
2029 		ssleep(1);
2030 		break;
2031 
2032 	case -NFS4ERR_MINOR_VERS_MISMATCH:
2033 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2034 			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2035 		dprintk("%s: exit with error %d for server %s\n",
2036 				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
2037 		return -EPROTONOSUPPORT;
2038 	case -ENOSPC:
2039 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2040 			nfs_mark_client_ready(clp, -EIO);
2041 		return -EIO;
2042 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2043 				 * in nfs4_exchange_id */
2044 	default:
2045 		dprintk("%s: exit with error %d for server %s\n", __func__,
2046 				status, clp->cl_hostname);
2047 		return status;
2048 	}
2049 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2050 	dprintk("%s: handled error %d for server %s\n", __func__, status,
2051 			clp->cl_hostname);
2052 	return 0;
2053 }
2054 
2055 static int nfs4_establish_lease(struct nfs_client *clp)
2056 {
2057 	const struct cred *cred;
2058 	const struct nfs4_state_recovery_ops *ops =
2059 		clp->cl_mvops->reboot_recovery_ops;
2060 	int status;
2061 
2062 	status = nfs4_begin_drain_session(clp);
2063 	if (status != 0)
2064 		return status;
2065 	cred = nfs4_get_clid_cred(clp);
2066 	if (cred == NULL)
2067 		return -ENOENT;
2068 	status = ops->establish_clid(clp, cred);
2069 	put_cred(cred);
2070 	if (status != 0)
2071 		return status;
2072 	pnfs_destroy_all_layouts(clp);
2073 	return 0;
2074 }
2075 
2076 /*
2077  * Returns zero or a negative errno.  NFS4ERR values are converted
2078  * to local errno values.
2079  */
2080 static int nfs4_reclaim_lease(struct nfs_client *clp)
2081 {
2082 	int status;
2083 
2084 	status = nfs4_establish_lease(clp);
2085 	if (status < 0)
2086 		return nfs4_handle_reclaim_lease_error(clp, status);
2087 	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2088 		nfs4_state_start_reclaim_nograce(clp);
2089 	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2090 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2091 	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2092 	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2093 	return 0;
2094 }
2095 
2096 static int nfs4_purge_lease(struct nfs_client *clp)
2097 {
2098 	int status;
2099 
2100 	status = nfs4_establish_lease(clp);
2101 	if (status < 0)
2102 		return nfs4_handle_reclaim_lease_error(clp, status);
2103 	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2104 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2105 	nfs4_state_start_reclaim_nograce(clp);
2106 	return 0;
2107 }
2108 
2109 /*
2110  * Try remote migration of one FSID from a source server to a
2111  * destination server.  The source server provides a list of
2112  * potential destinations.
2113  *
2114  * Returns zero or a negative NFS4ERR status code.
2115  */
2116 static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2117 {
2118 	struct nfs_client *clp = server->nfs_client;
2119 	struct nfs4_fs_locations *locations = NULL;
2120 	struct inode *inode;
2121 	struct page *page;
2122 	int status, result;
2123 
2124 	dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2125 			(unsigned long long)server->fsid.major,
2126 			(unsigned long long)server->fsid.minor,
2127 			clp->cl_hostname);
2128 
2129 	result = 0;
2130 	page = alloc_page(GFP_KERNEL);
2131 	locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2132 	if (page == NULL || locations == NULL) {
2133 		dprintk("<-- %s: no memory\n", __func__);
2134 		goto out;
2135 	}
2136 	locations->fattr = nfs_alloc_fattr();
2137 	if (locations->fattr == NULL) {
2138 		dprintk("<-- %s: no memory\n", __func__);
2139 		goto out;
2140 	}
2141 
2142 	inode = d_inode(server->super->s_root);
2143 	result = nfs4_proc_get_locations(server, NFS_FH(inode), locations,
2144 					 page, cred);
2145 	if (result) {
2146 		dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2147 			__func__, result);
2148 		goto out;
2149 	}
2150 
2151 	result = -NFS4ERR_NXIO;
2152 	if (!locations->nlocations)
2153 		goto out;
2154 
2155 	if (!(locations->fattr->valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2156 		dprintk("<-- %s: No fs_locations data, migration skipped\n",
2157 			__func__);
2158 		goto out;
2159 	}
2160 
2161 	status = nfs4_begin_drain_session(clp);
2162 	if (status != 0) {
2163 		result = status;
2164 		goto out;
2165 	}
2166 
2167 	status = nfs4_replace_transport(server, locations);
2168 	if (status != 0) {
2169 		dprintk("<-- %s: failed to replace transport: %d\n",
2170 			__func__, status);
2171 		goto out;
2172 	}
2173 
2174 	result = 0;
2175 	dprintk("<-- %s: migration succeeded\n", __func__);
2176 
2177 out:
2178 	if (page != NULL)
2179 		__free_page(page);
2180 	if (locations != NULL)
2181 		kfree(locations->fattr);
2182 	kfree(locations);
2183 	if (result) {
2184 		pr_err("NFS: migration recovery failed (server %s)\n",
2185 				clp->cl_hostname);
2186 		set_bit(NFS_MIG_FAILED, &server->mig_status);
2187 	}
2188 	return result;
2189 }
2190 
2191 /*
2192  * Returns zero or a negative NFS4ERR status code.
2193  */
2194 static int nfs4_handle_migration(struct nfs_client *clp)
2195 {
2196 	const struct nfs4_state_maintenance_ops *ops =
2197 				clp->cl_mvops->state_renewal_ops;
2198 	struct nfs_server *server;
2199 	const struct cred *cred;
2200 
2201 	dprintk("%s: migration reported on \"%s\"\n", __func__,
2202 			clp->cl_hostname);
2203 
2204 	cred = ops->get_state_renewal_cred(clp);
2205 	if (cred == NULL)
2206 		return -NFS4ERR_NOENT;
2207 
2208 	clp->cl_mig_gen++;
2209 restart:
2210 	rcu_read_lock();
2211 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2212 		int status;
2213 
2214 		if (server->mig_gen == clp->cl_mig_gen)
2215 			continue;
2216 		server->mig_gen = clp->cl_mig_gen;
2217 
2218 		if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2219 						&server->mig_status))
2220 			continue;
2221 
2222 		rcu_read_unlock();
2223 		status = nfs4_try_migration(server, cred);
2224 		if (status < 0) {
2225 			put_cred(cred);
2226 			return status;
2227 		}
2228 		goto restart;
2229 	}
2230 	rcu_read_unlock();
2231 	put_cred(cred);
2232 	return 0;
2233 }
2234 
2235 /*
2236  * Test each nfs_server on the clp's cl_superblocks list to see
2237  * if it's moved to another server.  Stop when the server no longer
2238  * returns NFS4ERR_LEASE_MOVED.
2239  */
2240 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2241 {
2242 	const struct nfs4_state_maintenance_ops *ops =
2243 				clp->cl_mvops->state_renewal_ops;
2244 	struct nfs_server *server;
2245 	const struct cred *cred;
2246 
2247 	dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2248 			clp->cl_hostname);
2249 
2250 	cred = ops->get_state_renewal_cred(clp);
2251 	if (cred == NULL)
2252 		return -NFS4ERR_NOENT;
2253 
2254 	clp->cl_mig_gen++;
2255 restart:
2256 	rcu_read_lock();
2257 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2258 		struct inode *inode;
2259 		int status;
2260 
2261 		if (server->mig_gen == clp->cl_mig_gen)
2262 			continue;
2263 		server->mig_gen = clp->cl_mig_gen;
2264 
2265 		rcu_read_unlock();
2266 
2267 		inode = d_inode(server->super->s_root);
2268 		status = nfs4_proc_fsid_present(inode, cred);
2269 		if (status != -NFS4ERR_MOVED)
2270 			goto restart;	/* wasn't this one */
2271 		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2272 			goto restart;	/* there are more */
2273 		goto out;
2274 	}
2275 	rcu_read_unlock();
2276 
2277 out:
2278 	put_cred(cred);
2279 	return 0;
2280 }
2281 
2282 /**
2283  * nfs4_discover_server_trunking - Detect server IP address trunking
2284  *
2285  * @clp: nfs_client under test
2286  * @result: OUT: found nfs_client, or clp
2287  *
2288  * Returns zero or a negative errno.  If zero is returned,
2289  * an nfs_client pointer is planted in "result".
2290  *
2291  * Note: since we are invoked in process context, and
2292  * not from inside the state manager, we cannot use
2293  * nfs4_handle_reclaim_lease_error().
2294  */
2295 int nfs4_discover_server_trunking(struct nfs_client *clp,
2296 				  struct nfs_client **result)
2297 {
2298 	const struct nfs4_state_recovery_ops *ops =
2299 				clp->cl_mvops->reboot_recovery_ops;
2300 	struct rpc_clnt *clnt;
2301 	const struct cred *cred;
2302 	int i, status;
2303 
2304 	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2305 
2306 	clnt = clp->cl_rpcclient;
2307 	i = 0;
2308 
2309 	mutex_lock(&nfs_clid_init_mutex);
2310 again:
2311 	status  = -ENOENT;
2312 	cred = nfs4_get_clid_cred(clp);
2313 	if (cred == NULL)
2314 		goto out_unlock;
2315 
2316 	status = ops->detect_trunking(clp, result, cred);
2317 	put_cred(cred);
2318 	switch (status) {
2319 	case 0:
2320 	case -EINTR:
2321 	case -ERESTARTSYS:
2322 		break;
2323 	case -ETIMEDOUT:
2324 		if (clnt->cl_softrtry)
2325 			break;
2326 		fallthrough;
2327 	case -NFS4ERR_DELAY:
2328 	case -EAGAIN:
2329 		ssleep(1);
2330 		fallthrough;
2331 	case -NFS4ERR_STALE_CLIENTID:
2332 		dprintk("NFS: %s after status %d, retrying\n",
2333 			__func__, status);
2334 		goto again;
2335 	case -EACCES:
2336 		if (i++ == 0) {
2337 			nfs4_root_machine_cred(clp);
2338 			goto again;
2339 		}
2340 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2341 			break;
2342 		fallthrough;
2343 	case -NFS4ERR_CLID_INUSE:
2344 	case -NFS4ERR_WRONGSEC:
2345 		/* No point in retrying if we already used RPC_AUTH_UNIX */
2346 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2347 			status = -EPERM;
2348 			break;
2349 		}
2350 		clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2351 		if (IS_ERR(clnt)) {
2352 			status = PTR_ERR(clnt);
2353 			break;
2354 		}
2355 		/* Note: this is safe because we haven't yet marked the
2356 		 * client as ready, so we are the only user of
2357 		 * clp->cl_rpcclient
2358 		 */
2359 		clnt = xchg(&clp->cl_rpcclient, clnt);
2360 		rpc_shutdown_client(clnt);
2361 		clnt = clp->cl_rpcclient;
2362 		goto again;
2363 
2364 	case -NFS4ERR_MINOR_VERS_MISMATCH:
2365 		status = -EPROTONOSUPPORT;
2366 		break;
2367 
2368 	case -EKEYEXPIRED:
2369 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2370 				 * in nfs4_exchange_id */
2371 		status = -EKEYEXPIRED;
2372 		break;
2373 	default:
2374 		pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2375 				__func__, status);
2376 		status = -EIO;
2377 	}
2378 
2379 out_unlock:
2380 	mutex_unlock(&nfs_clid_init_mutex);
2381 	dprintk("NFS: %s: status = %d\n", __func__, status);
2382 	return status;
2383 }
2384 
2385 #ifdef CONFIG_NFS_V4_1
2386 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2387 {
2388 	struct nfs_client *clp = session->clp;
2389 
2390 	switch (err) {
2391 	default:
2392 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2393 		break;
2394 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2395 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2396 	}
2397 	nfs4_schedule_state_manager(clp);
2398 }
2399 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2400 
2401 void nfs41_notify_server(struct nfs_client *clp)
2402 {
2403 	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
2404 	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2405 	nfs4_schedule_state_manager(clp);
2406 }
2407 
2408 static void nfs4_reset_all_state(struct nfs_client *clp)
2409 {
2410 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2411 		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2412 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2413 		nfs4_state_start_reclaim_nograce(clp);
2414 		dprintk("%s: scheduling reset of all state for server %s!\n",
2415 				__func__, clp->cl_hostname);
2416 		nfs4_schedule_state_manager(clp);
2417 	}
2418 }
2419 
2420 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2421 {
2422 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2423 		nfs4_state_start_reclaim_reboot(clp);
2424 		dprintk("%s: server %s rebooted!\n", __func__,
2425 				clp->cl_hostname);
2426 		nfs4_schedule_state_manager(clp);
2427 	}
2428 }
2429 
2430 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2431 {
2432 	nfs4_reset_all_state(clp);
2433 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2434 }
2435 
2436 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2437 {
2438 	nfs4_state_start_reclaim_nograce(clp);
2439 	nfs4_schedule_state_manager(clp);
2440 
2441 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2442 }
2443 
2444 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2445 {
2446 	/* FIXME: For now, we destroy all layouts. */
2447 	pnfs_destroy_all_layouts(clp);
2448 	nfs_test_expired_all_delegations(clp);
2449 	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2450 			clp->cl_hostname);
2451 }
2452 
2453 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2454 {
2455 	set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2456 	nfs4_schedule_state_manager(clp);
2457 
2458 	dprintk("%s: server %s declared a backchannel fault\n", __func__,
2459 			clp->cl_hostname);
2460 }
2461 
2462 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2463 {
2464 	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2465 		&clp->cl_state) == 0)
2466 		nfs4_schedule_state_manager(clp);
2467 }
2468 
2469 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2470 		bool recovery)
2471 {
2472 	if (!flags)
2473 		return;
2474 
2475 	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2476 		__func__, clp->cl_hostname, clp->cl_clientid, flags);
2477 	/*
2478 	 * If we're called from the state manager thread, then assume we're
2479 	 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2480 	 * Those flags are expected to remain set until we're done
2481 	 * recovering (see RFC5661, section 18.46.3).
2482 	 */
2483 	if (recovery)
2484 		goto out_recovery;
2485 
2486 	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2487 		nfs41_handle_server_reboot(clp);
2488 	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2489 		nfs41_handle_all_state_revoked(clp);
2490 	if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2491 			    SEQ4_STATUS_ADMIN_STATE_REVOKED))
2492 		nfs41_handle_some_state_revoked(clp);
2493 	if (flags & SEQ4_STATUS_LEASE_MOVED)
2494 		nfs4_schedule_lease_moved_recovery(clp);
2495 	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2496 		nfs41_handle_recallable_state_revoked(clp);
2497 out_recovery:
2498 	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2499 		nfs41_handle_backchannel_fault(clp);
2500 	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2501 				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2502 		nfs41_handle_cb_path_down(clp);
2503 }
2504 
2505 static int nfs4_reset_session(struct nfs_client *clp)
2506 {
2507 	const struct cred *cred;
2508 	int status;
2509 
2510 	if (!nfs4_has_session(clp))
2511 		return 0;
2512 	status = nfs4_begin_drain_session(clp);
2513 	if (status != 0)
2514 		return status;
2515 	cred = nfs4_get_clid_cred(clp);
2516 	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2517 	switch (status) {
2518 	case 0:
2519 	case -NFS4ERR_BADSESSION:
2520 	case -NFS4ERR_DEADSESSION:
2521 		break;
2522 	case -NFS4ERR_BACK_CHAN_BUSY:
2523 	case -NFS4ERR_DELAY:
2524 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2525 		status = 0;
2526 		ssleep(1);
2527 		goto out;
2528 	default:
2529 		status = nfs4_recovery_handle_error(clp, status);
2530 		goto out;
2531 	}
2532 
2533 	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2534 	status = nfs4_proc_create_session(clp, cred);
2535 	if (status) {
2536 		dprintk("%s: session reset failed with status %d for server %s!\n",
2537 			__func__, status, clp->cl_hostname);
2538 		status = nfs4_handle_reclaim_lease_error(clp, status);
2539 		goto out;
2540 	}
2541 	nfs41_finish_session_reset(clp);
2542 	dprintk("%s: session reset was successful for server %s!\n",
2543 			__func__, clp->cl_hostname);
2544 out:
2545 	put_cred(cred);
2546 	return status;
2547 }
2548 
2549 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2550 {
2551 	const struct cred *cred;
2552 	int ret;
2553 
2554 	if (!nfs4_has_session(clp))
2555 		return 0;
2556 	ret = nfs4_begin_drain_session(clp);
2557 	if (ret != 0)
2558 		return ret;
2559 	cred = nfs4_get_clid_cred(clp);
2560 	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2561 	put_cred(cred);
2562 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2563 	switch (ret) {
2564 	case 0:
2565 		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2566 			__func__, clp->cl_hostname);
2567 		break;
2568 	case -NFS4ERR_DELAY:
2569 		ssleep(1);
2570 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2571 		break;
2572 	default:
2573 		return nfs4_recovery_handle_error(clp, ret);
2574 	}
2575 	return 0;
2576 }
2577 
2578 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2579 {
2580 	int iomode = 0;
2581 
2582 	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &clp->cl_state))
2583 		iomode += IOMODE_READ;
2584 	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &clp->cl_state))
2585 		iomode += IOMODE_RW;
2586 	/* Note: IOMODE_READ + IOMODE_RW == IOMODE_ANY */
2587 	if (iomode) {
2588 		pnfs_layout_return_unused_byclid(clp, iomode);
2589 		set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2590 	}
2591 }
2592 #else /* CONFIG_NFS_V4_1 */
2593 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2594 
2595 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2596 {
2597 	return 0;
2598 }
2599 
2600 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2601 {
2602 }
2603 #endif /* CONFIG_NFS_V4_1 */
2604 
2605 static void nfs4_state_manager(struct nfs_client *clp)
2606 {
2607 	unsigned int memflags;
2608 	int status = 0;
2609 	const char *section = "", *section_sep = "";
2610 
2611 	/*
2612 	 * State recovery can deadlock if the direct reclaim code tries
2613 	 * start NFS writeback. So ensure memory allocations are all
2614 	 * GFP_NOFS.
2615 	 */
2616 	memflags = memalloc_nofs_save();
2617 
2618 	/* Ensure exclusive access to NFSv4 state */
2619 	do {
2620 		trace_nfs4_state_mgr(clp);
2621 		clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2622 		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2623 			section = "purge state";
2624 			status = nfs4_purge_lease(clp);
2625 			if (status < 0)
2626 				goto out_error;
2627 			continue;
2628 		}
2629 
2630 		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2631 			section = "lease expired";
2632 			/* We're going to have to re-establish a clientid */
2633 			status = nfs4_reclaim_lease(clp);
2634 			if (status < 0)
2635 				goto out_error;
2636 			continue;
2637 		}
2638 
2639 		/* Initialize or reset the session */
2640 		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2641 			section = "reset session";
2642 			status = nfs4_reset_session(clp);
2643 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2644 				continue;
2645 			if (status < 0)
2646 				goto out_error;
2647 		}
2648 
2649 		/* Send BIND_CONN_TO_SESSION */
2650 		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2651 				&clp->cl_state)) {
2652 			section = "bind conn to session";
2653 			status = nfs4_bind_conn_to_session(clp);
2654 			if (status < 0)
2655 				goto out_error;
2656 			continue;
2657 		}
2658 
2659 		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2660 			section = "check lease";
2661 			status = nfs4_check_lease(clp);
2662 			if (status < 0)
2663 				goto out_error;
2664 			continue;
2665 		}
2666 
2667 		if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2668 			section = "migration";
2669 			status = nfs4_handle_migration(clp);
2670 			if (status < 0)
2671 				goto out_error;
2672 		}
2673 
2674 		if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2675 			section = "lease moved";
2676 			status = nfs4_handle_lease_moved(clp);
2677 			if (status < 0)
2678 				goto out_error;
2679 		}
2680 
2681 		/* First recover reboot state... */
2682 		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2683 			section = "reclaim reboot";
2684 			status = nfs4_do_reclaim(clp,
2685 				clp->cl_mvops->reboot_recovery_ops);
2686 			if (status == -EAGAIN)
2687 				continue;
2688 			if (status < 0)
2689 				goto out_error;
2690 			nfs4_state_end_reclaim_reboot(clp);
2691 			continue;
2692 		}
2693 
2694 		/* Detect expired delegations... */
2695 		if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2696 			section = "detect expired delegations";
2697 			nfs_reap_expired_delegations(clp);
2698 			continue;
2699 		}
2700 
2701 		/* Now recover expired state... */
2702 		if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2703 			section = "reclaim nograce";
2704 			status = nfs4_do_reclaim(clp,
2705 				clp->cl_mvops->nograce_recovery_ops);
2706 			if (status == -EAGAIN)
2707 				continue;
2708 			if (status < 0)
2709 				goto out_error;
2710 			clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2711 		}
2712 
2713 		memalloc_nofs_restore(memflags);
2714 		nfs4_end_drain_session(clp);
2715 		nfs4_clear_state_manager_bit(clp);
2716 
2717 		if (test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2718 		    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING,
2719 				      &clp->cl_state)) {
2720 			memflags = memalloc_nofs_save();
2721 			continue;
2722 		}
2723 
2724 		if (!test_and_set_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state)) {
2725 			if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2726 				nfs_client_return_marked_delegations(clp);
2727 				set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2728 			}
2729 			nfs4_layoutreturn_any_run(clp);
2730 			clear_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state);
2731 		}
2732 
2733 		return;
2734 
2735 	} while (refcount_read(&clp->cl_count) > 1 && !signalled());
2736 	goto out_drain;
2737 
2738 out_error:
2739 	if (strlen(section))
2740 		section_sep = ": ";
2741 	trace_nfs4_state_mgr_failed(clp, section, status);
2742 	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2743 			" with error %d\n", section_sep, section,
2744 			clp->cl_hostname, -status);
2745 	ssleep(1);
2746 out_drain:
2747 	memalloc_nofs_restore(memflags);
2748 	nfs4_end_drain_session(clp);
2749 	nfs4_clear_state_manager_bit(clp);
2750 }
2751 
2752 static int nfs4_run_state_manager(void *ptr)
2753 {
2754 	struct nfs_client *clp = ptr;
2755 	struct rpc_clnt *cl = clp->cl_rpcclient;
2756 
2757 	while (cl != cl->cl_parent)
2758 		cl = cl->cl_parent;
2759 
2760 	allow_signal(SIGKILL);
2761 again:
2762 	nfs4_state_manager(clp);
2763 
2764 	if (test_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state) &&
2765 	    !test_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state)) {
2766 		wait_var_event_interruptible(&clp->cl_state,
2767 					     test_bit(NFS4CLNT_RUN_MANAGER,
2768 						      &clp->cl_state));
2769 		if (!atomic_read(&cl->cl_swapper))
2770 			clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2771 		if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2772 		    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2773 			goto again;
2774 		/* Either no longer a swapper, or were signalled */
2775 		clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2776 	}
2777 
2778 	if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2779 	    test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2780 	    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2781 		goto again;
2782 
2783 	nfs_put_client(clp);
2784 	module_put_and_kthread_exit(0);
2785 	return 0;
2786 }
2787