xref: /openbmc/linux/fs/nfs/nfs4state.c (revision 6774def6)
1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40 
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/nfs_idmap.h>
46 #include <linux/kthread.h>
47 #include <linux/module.h>
48 #include <linux/random.h>
49 #include <linux/ratelimit.h>
50 #include <linux/workqueue.h>
51 #include <linux/bitops.h>
52 #include <linux/jiffies.h>
53 
54 #include <linux/sunrpc/clnt.h>
55 
56 #include "nfs4_fs.h"
57 #include "callback.h"
58 #include "delegation.h"
59 #include "internal.h"
60 #include "nfs4session.h"
61 #include "pnfs.h"
62 #include "netns.h"
63 
64 #define NFSDBG_FACILITY		NFSDBG_STATE
65 
66 #define OPENOWNER_POOL_SIZE	8
67 
68 const nfs4_stateid zero_stateid;
69 static DEFINE_MUTEX(nfs_clid_init_mutex);
70 
71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
72 {
73 	struct nfs4_setclientid_res clid = {
74 		.clientid = clp->cl_clientid,
75 		.confirm = clp->cl_confirm,
76 	};
77 	unsigned short port;
78 	int status;
79 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
80 
81 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
82 		goto do_confirm;
83 	port = nn->nfs_callback_tcpport;
84 	if (clp->cl_addr.ss_family == AF_INET6)
85 		port = nn->nfs_callback_tcpport6;
86 
87 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
88 	if (status != 0)
89 		goto out;
90 	clp->cl_clientid = clid.clientid;
91 	clp->cl_confirm = clid.confirm;
92 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
93 do_confirm:
94 	status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
95 	if (status != 0)
96 		goto out;
97 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
98 	nfs4_schedule_state_renewal(clp);
99 out:
100 	return status;
101 }
102 
103 /**
104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
105  *
106  * @clp: nfs_client under test
107  * @result: OUT: found nfs_client, or clp
108  * @cred: credential to use for trunking test
109  *
110  * Returns zero, a negative errno, or a negative NFS4ERR status.
111  * If zero is returned, an nfs_client pointer is planted in
112  * "result".
113  *
114  * Note: The returned client may not yet be marked ready.
115  */
116 int nfs40_discover_server_trunking(struct nfs_client *clp,
117 				   struct nfs_client **result,
118 				   struct rpc_cred *cred)
119 {
120 	struct nfs4_setclientid_res clid = {
121 		.clientid = clp->cl_clientid,
122 		.confirm = clp->cl_confirm,
123 	};
124 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
125 	unsigned short port;
126 	int status;
127 
128 	port = nn->nfs_callback_tcpport;
129 	if (clp->cl_addr.ss_family == AF_INET6)
130 		port = nn->nfs_callback_tcpport6;
131 
132 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
133 	if (status != 0)
134 		goto out;
135 	clp->cl_clientid = clid.clientid;
136 	clp->cl_confirm = clid.confirm;
137 
138 	status = nfs40_walk_client_list(clp, result, cred);
139 	if (status == 0) {
140 		/* Sustain the lease, even if it's empty.  If the clientid4
141 		 * goes stale it's of no use for trunking discovery. */
142 		nfs4_schedule_state_renewal(*result);
143 	}
144 out:
145 	return status;
146 }
147 
148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
149 {
150 	struct rpc_cred *cred = NULL;
151 
152 	if (clp->cl_machine_cred != NULL)
153 		cred = get_rpccred(clp->cl_machine_cred);
154 	return cred;
155 }
156 
157 static void nfs4_root_machine_cred(struct nfs_client *clp)
158 {
159 	struct rpc_cred *cred, *new;
160 
161 	new = rpc_lookup_machine_cred(NULL);
162 	spin_lock(&clp->cl_lock);
163 	cred = clp->cl_machine_cred;
164 	clp->cl_machine_cred = new;
165 	spin_unlock(&clp->cl_lock);
166 	if (cred != NULL)
167 		put_rpccred(cred);
168 }
169 
170 static struct rpc_cred *
171 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
172 {
173 	struct rpc_cred *cred = NULL;
174 	struct nfs4_state_owner *sp;
175 	struct rb_node *pos;
176 
177 	for (pos = rb_first(&server->state_owners);
178 	     pos != NULL;
179 	     pos = rb_next(pos)) {
180 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
181 		if (list_empty(&sp->so_states))
182 			continue;
183 		cred = get_rpccred(sp->so_cred);
184 		break;
185 	}
186 	return cred;
187 }
188 
189 /**
190  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
191  * @clp: client state handle
192  *
193  * Returns an rpc_cred with reference count bumped, or NULL.
194  * Caller must hold clp->cl_lock.
195  */
196 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
197 {
198 	struct rpc_cred *cred = NULL;
199 	struct nfs_server *server;
200 
201 	/* Use machine credentials if available */
202 	cred = nfs4_get_machine_cred_locked(clp);
203 	if (cred != NULL)
204 		goto out;
205 
206 	rcu_read_lock();
207 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
208 		cred = nfs4_get_renew_cred_server_locked(server);
209 		if (cred != NULL)
210 			break;
211 	}
212 	rcu_read_unlock();
213 
214 out:
215 	return cred;
216 }
217 
218 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
219 {
220 	if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
221 		spin_lock(&tbl->slot_tbl_lock);
222 		nfs41_wake_slot_table(tbl);
223 		spin_unlock(&tbl->slot_tbl_lock);
224 	}
225 }
226 
227 static void nfs4_end_drain_session(struct nfs_client *clp)
228 {
229 	struct nfs4_session *ses = clp->cl_session;
230 
231 	if (clp->cl_slot_tbl) {
232 		nfs4_end_drain_slot_table(clp->cl_slot_tbl);
233 		return;
234 	}
235 
236 	if (ses != NULL) {
237 		nfs4_end_drain_slot_table(&ses->bc_slot_table);
238 		nfs4_end_drain_slot_table(&ses->fc_slot_table);
239 	}
240 }
241 
242 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
243 {
244 	set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
245 	spin_lock(&tbl->slot_tbl_lock);
246 	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
247 		reinit_completion(&tbl->complete);
248 		spin_unlock(&tbl->slot_tbl_lock);
249 		return wait_for_completion_interruptible(&tbl->complete);
250 	}
251 	spin_unlock(&tbl->slot_tbl_lock);
252 	return 0;
253 }
254 
255 static int nfs4_begin_drain_session(struct nfs_client *clp)
256 {
257 	struct nfs4_session *ses = clp->cl_session;
258 	int ret = 0;
259 
260 	if (clp->cl_slot_tbl)
261 		return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
262 
263 	/* back channel */
264 	ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
265 	if (ret)
266 		return ret;
267 	/* fore channel */
268 	return nfs4_drain_slot_tbl(&ses->fc_slot_table);
269 }
270 
271 #if defined(CONFIG_NFS_V4_1)
272 
273 static int nfs41_setup_state_renewal(struct nfs_client *clp)
274 {
275 	int status;
276 	struct nfs_fsinfo fsinfo;
277 
278 	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
279 		nfs4_schedule_state_renewal(clp);
280 		return 0;
281 	}
282 
283 	status = nfs4_proc_get_lease_time(clp, &fsinfo);
284 	if (status == 0) {
285 		/* Update lease time and schedule renewal */
286 		spin_lock(&clp->cl_lock);
287 		clp->cl_lease_time = fsinfo.lease_time * HZ;
288 		clp->cl_last_renewal = jiffies;
289 		spin_unlock(&clp->cl_lock);
290 
291 		nfs4_schedule_state_renewal(clp);
292 	}
293 
294 	return status;
295 }
296 
297 static void nfs41_finish_session_reset(struct nfs_client *clp)
298 {
299 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300 	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301 	/* create_session negotiated new slot table */
302 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303 	nfs41_setup_state_renewal(clp);
304 }
305 
306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
307 {
308 	int status;
309 
310 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311 		goto do_confirm;
312 	nfs4_begin_drain_session(clp);
313 	status = nfs4_proc_exchange_id(clp, cred);
314 	if (status != 0)
315 		goto out;
316 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
317 do_confirm:
318 	status = nfs4_proc_create_session(clp, cred);
319 	if (status != 0)
320 		goto out;
321 	nfs41_finish_session_reset(clp);
322 	nfs_mark_client_ready(clp, NFS_CS_READY);
323 out:
324 	return status;
325 }
326 
327 /**
328  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
329  *
330  * @clp: nfs_client under test
331  * @result: OUT: found nfs_client, or clp
332  * @cred: credential to use for trunking test
333  *
334  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
335  * If NFS4_OK is returned, an nfs_client pointer is planted in
336  * "result".
337  *
338  * Note: The returned client may not yet be marked ready.
339  */
340 int nfs41_discover_server_trunking(struct nfs_client *clp,
341 				   struct nfs_client **result,
342 				   struct rpc_cred *cred)
343 {
344 	int status;
345 
346 	status = nfs4_proc_exchange_id(clp, cred);
347 	if (status != NFS4_OK)
348 		return status;
349 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
350 
351 	return nfs41_walk_client_list(clp, result, cred);
352 }
353 
354 #endif /* CONFIG_NFS_V4_1 */
355 
356 /**
357  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
358  * @clp: client state handle
359  *
360  * Returns an rpc_cred with reference count bumped, or NULL.
361  */
362 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
363 {
364 	struct rpc_cred *cred;
365 
366 	spin_lock(&clp->cl_lock);
367 	cred = nfs4_get_machine_cred_locked(clp);
368 	spin_unlock(&clp->cl_lock);
369 	return cred;
370 }
371 
372 static struct nfs4_state_owner *
373 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
374 {
375 	struct rb_node **p = &server->state_owners.rb_node,
376 		       *parent = NULL;
377 	struct nfs4_state_owner *sp;
378 
379 	while (*p != NULL) {
380 		parent = *p;
381 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
382 
383 		if (cred < sp->so_cred)
384 			p = &parent->rb_left;
385 		else if (cred > sp->so_cred)
386 			p = &parent->rb_right;
387 		else {
388 			if (!list_empty(&sp->so_lru))
389 				list_del_init(&sp->so_lru);
390 			atomic_inc(&sp->so_count);
391 			return sp;
392 		}
393 	}
394 	return NULL;
395 }
396 
397 static struct nfs4_state_owner *
398 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
399 {
400 	struct nfs_server *server = new->so_server;
401 	struct rb_node **p = &server->state_owners.rb_node,
402 		       *parent = NULL;
403 	struct nfs4_state_owner *sp;
404 	int err;
405 
406 	while (*p != NULL) {
407 		parent = *p;
408 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
409 
410 		if (new->so_cred < sp->so_cred)
411 			p = &parent->rb_left;
412 		else if (new->so_cred > sp->so_cred)
413 			p = &parent->rb_right;
414 		else {
415 			if (!list_empty(&sp->so_lru))
416 				list_del_init(&sp->so_lru);
417 			atomic_inc(&sp->so_count);
418 			return sp;
419 		}
420 	}
421 	err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
422 	if (err)
423 		return ERR_PTR(err);
424 	rb_link_node(&new->so_server_node, parent, p);
425 	rb_insert_color(&new->so_server_node, &server->state_owners);
426 	return new;
427 }
428 
429 static void
430 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
431 {
432 	struct nfs_server *server = sp->so_server;
433 
434 	if (!RB_EMPTY_NODE(&sp->so_server_node))
435 		rb_erase(&sp->so_server_node, &server->state_owners);
436 	ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
437 }
438 
439 static void
440 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
441 {
442 	sc->create_time = ktime_get();
443 	sc->flags = 0;
444 	sc->counter = 0;
445 	spin_lock_init(&sc->lock);
446 	INIT_LIST_HEAD(&sc->list);
447 	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
448 }
449 
450 static void
451 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
452 {
453 	rpc_destroy_wait_queue(&sc->wait);
454 }
455 
456 /*
457  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
458  * create a new state_owner.
459  *
460  */
461 static struct nfs4_state_owner *
462 nfs4_alloc_state_owner(struct nfs_server *server,
463 		struct rpc_cred *cred,
464 		gfp_t gfp_flags)
465 {
466 	struct nfs4_state_owner *sp;
467 
468 	sp = kzalloc(sizeof(*sp), gfp_flags);
469 	if (!sp)
470 		return NULL;
471 	sp->so_server = server;
472 	sp->so_cred = get_rpccred(cred);
473 	spin_lock_init(&sp->so_lock);
474 	INIT_LIST_HEAD(&sp->so_states);
475 	nfs4_init_seqid_counter(&sp->so_seqid);
476 	atomic_set(&sp->so_count, 1);
477 	INIT_LIST_HEAD(&sp->so_lru);
478 	seqcount_init(&sp->so_reclaim_seqcount);
479 	mutex_init(&sp->so_delegreturn_mutex);
480 	return sp;
481 }
482 
483 static void
484 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
485 {
486 	struct rb_node *rb_node = &sp->so_server_node;
487 
488 	if (!RB_EMPTY_NODE(rb_node)) {
489 		struct nfs_server *server = sp->so_server;
490 		struct nfs_client *clp = server->nfs_client;
491 
492 		spin_lock(&clp->cl_lock);
493 		if (!RB_EMPTY_NODE(rb_node)) {
494 			rb_erase(rb_node, &server->state_owners);
495 			RB_CLEAR_NODE(rb_node);
496 		}
497 		spin_unlock(&clp->cl_lock);
498 	}
499 }
500 
501 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
502 {
503 	nfs4_destroy_seqid_counter(&sp->so_seqid);
504 	put_rpccred(sp->so_cred);
505 	kfree(sp);
506 }
507 
508 static void nfs4_gc_state_owners(struct nfs_server *server)
509 {
510 	struct nfs_client *clp = server->nfs_client;
511 	struct nfs4_state_owner *sp, *tmp;
512 	unsigned long time_min, time_max;
513 	LIST_HEAD(doomed);
514 
515 	spin_lock(&clp->cl_lock);
516 	time_max = jiffies;
517 	time_min = (long)time_max - (long)clp->cl_lease_time;
518 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
519 		/* NB: LRU is sorted so that oldest is at the head */
520 		if (time_in_range(sp->so_expires, time_min, time_max))
521 			break;
522 		list_move(&sp->so_lru, &doomed);
523 		nfs4_remove_state_owner_locked(sp);
524 	}
525 	spin_unlock(&clp->cl_lock);
526 
527 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
528 		list_del(&sp->so_lru);
529 		nfs4_free_state_owner(sp);
530 	}
531 }
532 
533 /**
534  * nfs4_get_state_owner - Look up a state owner given a credential
535  * @server: nfs_server to search
536  * @cred: RPC credential to match
537  *
538  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
539  */
540 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
541 					      struct rpc_cred *cred,
542 					      gfp_t gfp_flags)
543 {
544 	struct nfs_client *clp = server->nfs_client;
545 	struct nfs4_state_owner *sp, *new;
546 
547 	spin_lock(&clp->cl_lock);
548 	sp = nfs4_find_state_owner_locked(server, cred);
549 	spin_unlock(&clp->cl_lock);
550 	if (sp != NULL)
551 		goto out;
552 	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
553 	if (new == NULL)
554 		goto out;
555 	do {
556 		if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
557 			break;
558 		spin_lock(&clp->cl_lock);
559 		sp = nfs4_insert_state_owner_locked(new);
560 		spin_unlock(&clp->cl_lock);
561 	} while (sp == ERR_PTR(-EAGAIN));
562 	if (sp != new)
563 		nfs4_free_state_owner(new);
564 out:
565 	nfs4_gc_state_owners(server);
566 	return sp;
567 }
568 
569 /**
570  * nfs4_put_state_owner - Release a nfs4_state_owner
571  * @sp: state owner data to release
572  *
573  * Note that we keep released state owners on an LRU
574  * list.
575  * This caches valid state owners so that they can be
576  * reused, to avoid the OPEN_CONFIRM on minor version 0.
577  * It also pins the uniquifier of dropped state owners for
578  * a while, to ensure that those state owner names are
579  * never reused.
580  */
581 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
582 {
583 	struct nfs_server *server = sp->so_server;
584 	struct nfs_client *clp = server->nfs_client;
585 
586 	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
587 		return;
588 
589 	sp->so_expires = jiffies;
590 	list_add_tail(&sp->so_lru, &server->state_owners_lru);
591 	spin_unlock(&clp->cl_lock);
592 }
593 
594 /**
595  * nfs4_purge_state_owners - Release all cached state owners
596  * @server: nfs_server with cached state owners to release
597  *
598  * Called at umount time.  Remaining state owners will be on
599  * the LRU with ref count of zero.
600  */
601 void nfs4_purge_state_owners(struct nfs_server *server)
602 {
603 	struct nfs_client *clp = server->nfs_client;
604 	struct nfs4_state_owner *sp, *tmp;
605 	LIST_HEAD(doomed);
606 
607 	spin_lock(&clp->cl_lock);
608 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
609 		list_move(&sp->so_lru, &doomed);
610 		nfs4_remove_state_owner_locked(sp);
611 	}
612 	spin_unlock(&clp->cl_lock);
613 
614 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
615 		list_del(&sp->so_lru);
616 		nfs4_free_state_owner(sp);
617 	}
618 }
619 
620 static struct nfs4_state *
621 nfs4_alloc_open_state(void)
622 {
623 	struct nfs4_state *state;
624 
625 	state = kzalloc(sizeof(*state), GFP_NOFS);
626 	if (!state)
627 		return NULL;
628 	atomic_set(&state->count, 1);
629 	INIT_LIST_HEAD(&state->lock_states);
630 	spin_lock_init(&state->state_lock);
631 	seqlock_init(&state->seqlock);
632 	return state;
633 }
634 
635 void
636 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
637 {
638 	if (state->state == fmode)
639 		return;
640 	/* NB! List reordering - see the reclaim code for why.  */
641 	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
642 		if (fmode & FMODE_WRITE)
643 			list_move(&state->open_states, &state->owner->so_states);
644 		else
645 			list_move_tail(&state->open_states, &state->owner->so_states);
646 	}
647 	state->state = fmode;
648 }
649 
650 static struct nfs4_state *
651 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
652 {
653 	struct nfs_inode *nfsi = NFS_I(inode);
654 	struct nfs4_state *state;
655 
656 	list_for_each_entry(state, &nfsi->open_states, inode_states) {
657 		if (state->owner != owner)
658 			continue;
659 		if (!nfs4_valid_open_stateid(state))
660 			continue;
661 		if (atomic_inc_not_zero(&state->count))
662 			return state;
663 	}
664 	return NULL;
665 }
666 
667 static void
668 nfs4_free_open_state(struct nfs4_state *state)
669 {
670 	kfree(state);
671 }
672 
673 struct nfs4_state *
674 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
675 {
676 	struct nfs4_state *state, *new;
677 	struct nfs_inode *nfsi = NFS_I(inode);
678 
679 	spin_lock(&inode->i_lock);
680 	state = __nfs4_find_state_byowner(inode, owner);
681 	spin_unlock(&inode->i_lock);
682 	if (state)
683 		goto out;
684 	new = nfs4_alloc_open_state();
685 	spin_lock(&owner->so_lock);
686 	spin_lock(&inode->i_lock);
687 	state = __nfs4_find_state_byowner(inode, owner);
688 	if (state == NULL && new != NULL) {
689 		state = new;
690 		state->owner = owner;
691 		atomic_inc(&owner->so_count);
692 		list_add(&state->inode_states, &nfsi->open_states);
693 		ihold(inode);
694 		state->inode = inode;
695 		spin_unlock(&inode->i_lock);
696 		/* Note: The reclaim code dictates that we add stateless
697 		 * and read-only stateids to the end of the list */
698 		list_add_tail(&state->open_states, &owner->so_states);
699 		spin_unlock(&owner->so_lock);
700 	} else {
701 		spin_unlock(&inode->i_lock);
702 		spin_unlock(&owner->so_lock);
703 		if (new)
704 			nfs4_free_open_state(new);
705 	}
706 out:
707 	return state;
708 }
709 
710 void nfs4_put_open_state(struct nfs4_state *state)
711 {
712 	struct inode *inode = state->inode;
713 	struct nfs4_state_owner *owner = state->owner;
714 
715 	if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
716 		return;
717 	spin_lock(&inode->i_lock);
718 	list_del(&state->inode_states);
719 	list_del(&state->open_states);
720 	spin_unlock(&inode->i_lock);
721 	spin_unlock(&owner->so_lock);
722 	iput(inode);
723 	nfs4_free_open_state(state);
724 	nfs4_put_state_owner(owner);
725 }
726 
727 /*
728  * Close the current file.
729  */
730 static void __nfs4_close(struct nfs4_state *state,
731 		fmode_t fmode, gfp_t gfp_mask, int wait)
732 {
733 	struct nfs4_state_owner *owner = state->owner;
734 	int call_close = 0;
735 	fmode_t newstate;
736 
737 	atomic_inc(&owner->so_count);
738 	/* Protect against nfs4_find_state() */
739 	spin_lock(&owner->so_lock);
740 	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
741 		case FMODE_READ:
742 			state->n_rdonly--;
743 			break;
744 		case FMODE_WRITE:
745 			state->n_wronly--;
746 			break;
747 		case FMODE_READ|FMODE_WRITE:
748 			state->n_rdwr--;
749 	}
750 	newstate = FMODE_READ|FMODE_WRITE;
751 	if (state->n_rdwr == 0) {
752 		if (state->n_rdonly == 0) {
753 			newstate &= ~FMODE_READ;
754 			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
755 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
756 		}
757 		if (state->n_wronly == 0) {
758 			newstate &= ~FMODE_WRITE;
759 			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
760 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
761 		}
762 		if (newstate == 0)
763 			clear_bit(NFS_DELEGATED_STATE, &state->flags);
764 	}
765 	nfs4_state_set_mode_locked(state, newstate);
766 	spin_unlock(&owner->so_lock);
767 
768 	if (!call_close) {
769 		nfs4_put_open_state(state);
770 		nfs4_put_state_owner(owner);
771 	} else
772 		nfs4_do_close(state, gfp_mask, wait);
773 }
774 
775 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
776 {
777 	__nfs4_close(state, fmode, GFP_NOFS, 0);
778 }
779 
780 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
781 {
782 	__nfs4_close(state, fmode, GFP_KERNEL, 1);
783 }
784 
785 /*
786  * Search the state->lock_states for an existing lock_owner
787  * that is compatible with current->files
788  */
789 static struct nfs4_lock_state *
790 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
791 {
792 	struct nfs4_lock_state *pos;
793 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
794 		if (pos->ls_owner != fl_owner)
795 			continue;
796 		atomic_inc(&pos->ls_count);
797 		return pos;
798 	}
799 	return NULL;
800 }
801 
802 /*
803  * Return a compatible lock_state. If no initialized lock_state structure
804  * exists, return an uninitialized one.
805  *
806  */
807 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
808 {
809 	struct nfs4_lock_state *lsp;
810 	struct nfs_server *server = state->owner->so_server;
811 
812 	lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
813 	if (lsp == NULL)
814 		return NULL;
815 	nfs4_init_seqid_counter(&lsp->ls_seqid);
816 	atomic_set(&lsp->ls_count, 1);
817 	lsp->ls_state = state;
818 	lsp->ls_owner = fl_owner;
819 	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
820 	if (lsp->ls_seqid.owner_id < 0)
821 		goto out_free;
822 	INIT_LIST_HEAD(&lsp->ls_locks);
823 	return lsp;
824 out_free:
825 	kfree(lsp);
826 	return NULL;
827 }
828 
829 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
830 {
831 	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
832 	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
833 	kfree(lsp);
834 }
835 
836 /*
837  * Return a compatible lock_state. If no initialized lock_state structure
838  * exists, return an uninitialized one.
839  *
840  */
841 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
842 {
843 	struct nfs4_lock_state *lsp, *new = NULL;
844 
845 	for(;;) {
846 		spin_lock(&state->state_lock);
847 		lsp = __nfs4_find_lock_state(state, owner);
848 		if (lsp != NULL)
849 			break;
850 		if (new != NULL) {
851 			list_add(&new->ls_locks, &state->lock_states);
852 			set_bit(LK_STATE_IN_USE, &state->flags);
853 			lsp = new;
854 			new = NULL;
855 			break;
856 		}
857 		spin_unlock(&state->state_lock);
858 		new = nfs4_alloc_lock_state(state, owner);
859 		if (new == NULL)
860 			return NULL;
861 	}
862 	spin_unlock(&state->state_lock);
863 	if (new != NULL)
864 		nfs4_free_lock_state(state->owner->so_server, new);
865 	return lsp;
866 }
867 
868 /*
869  * Release reference to lock_state, and free it if we see that
870  * it is no longer in use
871  */
872 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
873 {
874 	struct nfs_server *server;
875 	struct nfs4_state *state;
876 
877 	if (lsp == NULL)
878 		return;
879 	state = lsp->ls_state;
880 	if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
881 		return;
882 	list_del(&lsp->ls_locks);
883 	if (list_empty(&state->lock_states))
884 		clear_bit(LK_STATE_IN_USE, &state->flags);
885 	spin_unlock(&state->state_lock);
886 	server = state->owner->so_server;
887 	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
888 		struct nfs_client *clp = server->nfs_client;
889 
890 		clp->cl_mvops->free_lock_state(server, lsp);
891 	} else
892 		nfs4_free_lock_state(server, lsp);
893 }
894 
895 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
896 {
897 	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
898 
899 	dst->fl_u.nfs4_fl.owner = lsp;
900 	atomic_inc(&lsp->ls_count);
901 }
902 
903 static void nfs4_fl_release_lock(struct file_lock *fl)
904 {
905 	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
906 }
907 
908 static const struct file_lock_operations nfs4_fl_lock_ops = {
909 	.fl_copy_lock = nfs4_fl_copy_lock,
910 	.fl_release_private = nfs4_fl_release_lock,
911 };
912 
913 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
914 {
915 	struct nfs4_lock_state *lsp;
916 
917 	if (fl->fl_ops != NULL)
918 		return 0;
919 	lsp = nfs4_get_lock_state(state, fl->fl_owner);
920 	if (lsp == NULL)
921 		return -ENOMEM;
922 	fl->fl_u.nfs4_fl.owner = lsp;
923 	fl->fl_ops = &nfs4_fl_lock_ops;
924 	return 0;
925 }
926 
927 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
928 		struct nfs4_state *state,
929 		const struct nfs_lockowner *lockowner)
930 {
931 	struct nfs4_lock_state *lsp;
932 	fl_owner_t fl_owner;
933 	int ret = -ENOENT;
934 
935 
936 	if (lockowner == NULL)
937 		goto out;
938 
939 	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
940 		goto out;
941 
942 	fl_owner = lockowner->l_owner;
943 	spin_lock(&state->state_lock);
944 	lsp = __nfs4_find_lock_state(state, fl_owner);
945 	if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
946 		ret = -EIO;
947 	else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
948 		nfs4_stateid_copy(dst, &lsp->ls_stateid);
949 		ret = 0;
950 	}
951 	spin_unlock(&state->state_lock);
952 	nfs4_put_lock_state(lsp);
953 out:
954 	return ret;
955 }
956 
957 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
958 {
959 	const nfs4_stateid *src;
960 	int seq;
961 
962 	do {
963 		src = &zero_stateid;
964 		seq = read_seqbegin(&state->seqlock);
965 		if (test_bit(NFS_OPEN_STATE, &state->flags))
966 			src = &state->open_stateid;
967 		nfs4_stateid_copy(dst, src);
968 	} while (read_seqretry(&state->seqlock, seq));
969 }
970 
971 /*
972  * Byte-range lock aware utility to initialize the stateid of read/write
973  * requests.
974  */
975 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
976 		fmode_t fmode, const struct nfs_lockowner *lockowner)
977 {
978 	int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
979 	if (ret == -EIO)
980 		/* A lost lock - don't even consider delegations */
981 		goto out;
982 	/* returns true if delegation stateid found and copied */
983 	if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
984 		ret = 0;
985 		goto out;
986 	}
987 	if (ret != -ENOENT)
988 		/* nfs4_copy_delegation_stateid() didn't over-write
989 		 * dst, so it still has the lock stateid which we now
990 		 * choose to use.
991 		 */
992 		goto out;
993 	nfs4_copy_open_stateid(dst, state);
994 	ret = 0;
995 out:
996 	if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
997 		dst->seqid = 0;
998 	return ret;
999 }
1000 
1001 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1002 {
1003 	struct nfs_seqid *new;
1004 
1005 	new = kmalloc(sizeof(*new), gfp_mask);
1006 	if (new != NULL) {
1007 		new->sequence = counter;
1008 		INIT_LIST_HEAD(&new->list);
1009 		new->task = NULL;
1010 	}
1011 	return new;
1012 }
1013 
1014 void nfs_release_seqid(struct nfs_seqid *seqid)
1015 {
1016 	struct nfs_seqid_counter *sequence;
1017 
1018 	if (list_empty(&seqid->list))
1019 		return;
1020 	sequence = seqid->sequence;
1021 	spin_lock(&sequence->lock);
1022 	list_del_init(&seqid->list);
1023 	if (!list_empty(&sequence->list)) {
1024 		struct nfs_seqid *next;
1025 
1026 		next = list_first_entry(&sequence->list,
1027 				struct nfs_seqid, list);
1028 		rpc_wake_up_queued_task(&sequence->wait, next->task);
1029 	}
1030 	spin_unlock(&sequence->lock);
1031 }
1032 
1033 void nfs_free_seqid(struct nfs_seqid *seqid)
1034 {
1035 	nfs_release_seqid(seqid);
1036 	kfree(seqid);
1037 }
1038 
1039 /*
1040  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1041  * failed with a seqid incrementing error -
1042  * see comments nfs4.h:seqid_mutating_error()
1043  */
1044 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1045 {
1046 	switch (status) {
1047 		case 0:
1048 			break;
1049 		case -NFS4ERR_BAD_SEQID:
1050 			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1051 				return;
1052 			pr_warn_ratelimited("NFS: v4 server returned a bad"
1053 					" sequence-id error on an"
1054 					" unconfirmed sequence %p!\n",
1055 					seqid->sequence);
1056 		case -NFS4ERR_STALE_CLIENTID:
1057 		case -NFS4ERR_STALE_STATEID:
1058 		case -NFS4ERR_BAD_STATEID:
1059 		case -NFS4ERR_BADXDR:
1060 		case -NFS4ERR_RESOURCE:
1061 		case -NFS4ERR_NOFILEHANDLE:
1062 			/* Non-seqid mutating errors */
1063 			return;
1064 	};
1065 	/*
1066 	 * Note: no locking needed as we are guaranteed to be first
1067 	 * on the sequence list
1068 	 */
1069 	seqid->sequence->counter++;
1070 }
1071 
1072 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1073 {
1074 	struct nfs4_state_owner *sp = container_of(seqid->sequence,
1075 					struct nfs4_state_owner, so_seqid);
1076 	struct nfs_server *server = sp->so_server;
1077 
1078 	if (status == -NFS4ERR_BAD_SEQID)
1079 		nfs4_drop_state_owner(sp);
1080 	if (!nfs4_has_session(server->nfs_client))
1081 		nfs_increment_seqid(status, seqid);
1082 }
1083 
1084 /*
1085  * Increment the seqid if the LOCK/LOCKU succeeded, or
1086  * failed with a seqid incrementing error -
1087  * see comments nfs4.h:seqid_mutating_error()
1088  */
1089 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1090 {
1091 	nfs_increment_seqid(status, seqid);
1092 }
1093 
1094 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1095 {
1096 	struct nfs_seqid_counter *sequence = seqid->sequence;
1097 	int status = 0;
1098 
1099 	spin_lock(&sequence->lock);
1100 	seqid->task = task;
1101 	if (list_empty(&seqid->list))
1102 		list_add_tail(&seqid->list, &sequence->list);
1103 	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1104 		goto unlock;
1105 	rpc_sleep_on(&sequence->wait, task, NULL);
1106 	status = -EAGAIN;
1107 unlock:
1108 	spin_unlock(&sequence->lock);
1109 	return status;
1110 }
1111 
1112 static int nfs4_run_state_manager(void *);
1113 
1114 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1115 {
1116 	smp_mb__before_atomic();
1117 	clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1118 	smp_mb__after_atomic();
1119 	wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1120 	rpc_wake_up(&clp->cl_rpcwaitq);
1121 }
1122 
1123 /*
1124  * Schedule the nfs_client asynchronous state management routine
1125  */
1126 void nfs4_schedule_state_manager(struct nfs_client *clp)
1127 {
1128 	struct task_struct *task;
1129 	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1130 
1131 	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1132 		return;
1133 	__module_get(THIS_MODULE);
1134 	atomic_inc(&clp->cl_count);
1135 
1136 	/* The rcu_read_lock() is not strictly necessary, as the state
1137 	 * manager is the only thread that ever changes the rpc_xprt
1138 	 * after it's initialized.  At this point, we're single threaded. */
1139 	rcu_read_lock();
1140 	snprintf(buf, sizeof(buf), "%s-manager",
1141 			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1142 	rcu_read_unlock();
1143 	task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1144 	if (IS_ERR(task)) {
1145 		printk(KERN_ERR "%s: kthread_run: %ld\n",
1146 			__func__, PTR_ERR(task));
1147 		nfs4_clear_state_manager_bit(clp);
1148 		nfs_put_client(clp);
1149 		module_put(THIS_MODULE);
1150 	}
1151 }
1152 
1153 /*
1154  * Schedule a lease recovery attempt
1155  */
1156 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1157 {
1158 	if (!clp)
1159 		return;
1160 	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1161 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1162 	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1163 			clp->cl_hostname);
1164 	nfs4_schedule_state_manager(clp);
1165 }
1166 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1167 
1168 /**
1169  * nfs4_schedule_migration_recovery - trigger migration recovery
1170  *
1171  * @server: FSID that is migrating
1172  *
1173  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1174  * value is returned.
1175  */
1176 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1177 {
1178 	struct nfs_client *clp = server->nfs_client;
1179 
1180 	if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1181 		pr_err("NFS: volatile file handles not supported (server %s)\n",
1182 				clp->cl_hostname);
1183 		return -NFS4ERR_IO;
1184 	}
1185 
1186 	if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1187 		return -NFS4ERR_IO;
1188 
1189 	dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1190 			__func__,
1191 			(unsigned long long)server->fsid.major,
1192 			(unsigned long long)server->fsid.minor,
1193 			clp->cl_hostname);
1194 
1195 	set_bit(NFS_MIG_IN_TRANSITION,
1196 			&((struct nfs_server *)server)->mig_status);
1197 	set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1198 
1199 	nfs4_schedule_state_manager(clp);
1200 	return 0;
1201 }
1202 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1203 
1204 /**
1205  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1206  *
1207  * @clp: server to check for moved leases
1208  *
1209  */
1210 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1211 {
1212 	dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1213 		__func__, clp->cl_clientid, clp->cl_hostname);
1214 
1215 	set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1216 	nfs4_schedule_state_manager(clp);
1217 }
1218 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1219 
1220 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1221 {
1222 	int res;
1223 
1224 	might_sleep();
1225 
1226 	atomic_inc(&clp->cl_count);
1227 	res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1228 				 nfs_wait_bit_killable, TASK_KILLABLE);
1229 	if (res)
1230 		goto out;
1231 	if (clp->cl_cons_state < 0)
1232 		res = clp->cl_cons_state;
1233 out:
1234 	nfs_put_client(clp);
1235 	return res;
1236 }
1237 
1238 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1239 {
1240 	unsigned int loop;
1241 	int ret;
1242 
1243 	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1244 		ret = nfs4_wait_clnt_recover(clp);
1245 		if (ret != 0)
1246 			break;
1247 		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1248 		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1249 			break;
1250 		nfs4_schedule_state_manager(clp);
1251 		ret = -EIO;
1252 	}
1253 	return ret;
1254 }
1255 
1256 /*
1257  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1258  * @clp: client to process
1259  *
1260  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1261  * resend of the SETCLIENTID and hence re-establish the
1262  * callback channel. Then return all existing delegations.
1263  */
1264 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1265 {
1266 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1267 	nfs_expire_all_delegations(clp);
1268 	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1269 			clp->cl_hostname);
1270 }
1271 
1272 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1273 {
1274 	nfs40_handle_cb_pathdown(clp);
1275 	nfs4_schedule_state_manager(clp);
1276 }
1277 
1278 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1279 {
1280 
1281 	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1282 	/* Don't recover state that expired before the reboot */
1283 	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1284 		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1285 		return 0;
1286 	}
1287 	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1288 	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1289 	return 1;
1290 }
1291 
1292 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1293 {
1294 	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1295 	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1296 	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1297 	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1298 	return 1;
1299 }
1300 
1301 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1302 {
1303 	struct nfs_client *clp = server->nfs_client;
1304 
1305 	if (!nfs4_valid_open_stateid(state))
1306 		return -EBADF;
1307 	nfs4_state_mark_reclaim_nograce(clp, state);
1308 	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1309 			clp->cl_hostname);
1310 	nfs4_schedule_state_manager(clp);
1311 	return 0;
1312 }
1313 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1314 
1315 void nfs_inode_find_state_and_recover(struct inode *inode,
1316 		const nfs4_stateid *stateid)
1317 {
1318 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1319 	struct nfs_inode *nfsi = NFS_I(inode);
1320 	struct nfs_open_context *ctx;
1321 	struct nfs4_state *state;
1322 	bool found = false;
1323 
1324 	spin_lock(&inode->i_lock);
1325 	list_for_each_entry(ctx, &nfsi->open_files, list) {
1326 		state = ctx->state;
1327 		if (state == NULL)
1328 			continue;
1329 		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1330 			continue;
1331 		if (!nfs4_stateid_match(&state->stateid, stateid))
1332 			continue;
1333 		nfs4_state_mark_reclaim_nograce(clp, state);
1334 		found = true;
1335 	}
1336 	spin_unlock(&inode->i_lock);
1337 	if (found)
1338 		nfs4_schedule_state_manager(clp);
1339 }
1340 
1341 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1342 {
1343 	struct inode *inode = state->inode;
1344 	struct nfs_inode *nfsi = NFS_I(inode);
1345 	struct nfs_open_context *ctx;
1346 
1347 	spin_lock(&inode->i_lock);
1348 	list_for_each_entry(ctx, &nfsi->open_files, list) {
1349 		if (ctx->state != state)
1350 			continue;
1351 		set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1352 	}
1353 	spin_unlock(&inode->i_lock);
1354 }
1355 
1356 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1357 {
1358 	set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1359 	nfs4_state_mark_open_context_bad(state);
1360 }
1361 
1362 
1363 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1364 {
1365 	struct inode *inode = state->inode;
1366 	struct nfs_inode *nfsi = NFS_I(inode);
1367 	struct file_lock *fl;
1368 	int status = 0;
1369 
1370 	if (inode->i_flock == NULL)
1371 		return 0;
1372 
1373 	/* Guard against delegation returns and new lock/unlock calls */
1374 	down_write(&nfsi->rwsem);
1375 	/* Protect inode->i_flock using the BKL */
1376 	spin_lock(&inode->i_lock);
1377 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1378 		if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1379 			continue;
1380 		if (nfs_file_open_context(fl->fl_file)->state != state)
1381 			continue;
1382 		spin_unlock(&inode->i_lock);
1383 		status = ops->recover_lock(state, fl);
1384 		switch (status) {
1385 			case 0:
1386 				break;
1387 			case -ESTALE:
1388 			case -NFS4ERR_ADMIN_REVOKED:
1389 			case -NFS4ERR_STALE_STATEID:
1390 			case -NFS4ERR_BAD_STATEID:
1391 			case -NFS4ERR_EXPIRED:
1392 			case -NFS4ERR_NO_GRACE:
1393 			case -NFS4ERR_STALE_CLIENTID:
1394 			case -NFS4ERR_BADSESSION:
1395 			case -NFS4ERR_BADSLOT:
1396 			case -NFS4ERR_BAD_HIGH_SLOT:
1397 			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1398 				goto out;
1399 			default:
1400 				printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1401 					 __func__, status);
1402 			case -ENOMEM:
1403 			case -NFS4ERR_DENIED:
1404 			case -NFS4ERR_RECLAIM_BAD:
1405 			case -NFS4ERR_RECLAIM_CONFLICT:
1406 				/* kill_proc(fl->fl_pid, SIGLOST, 1); */
1407 				status = 0;
1408 		}
1409 		spin_lock(&inode->i_lock);
1410 	}
1411 	spin_unlock(&inode->i_lock);
1412 out:
1413 	up_write(&nfsi->rwsem);
1414 	return status;
1415 }
1416 
1417 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1418 {
1419 	struct nfs4_state *state;
1420 	struct nfs4_lock_state *lock;
1421 	int status = 0;
1422 
1423 	/* Note: we rely on the sp->so_states list being ordered
1424 	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1425 	 * states first.
1426 	 * This is needed to ensure that the server won't give us any
1427 	 * read delegations that we have to return if, say, we are
1428 	 * recovering after a network partition or a reboot from a
1429 	 * server that doesn't support a grace period.
1430 	 */
1431 	spin_lock(&sp->so_lock);
1432 	raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1433 restart:
1434 	list_for_each_entry(state, &sp->so_states, open_states) {
1435 		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1436 			continue;
1437 		if (!nfs4_valid_open_stateid(state))
1438 			continue;
1439 		if (state->state == 0)
1440 			continue;
1441 		atomic_inc(&state->count);
1442 		spin_unlock(&sp->so_lock);
1443 		status = ops->recover_open(sp, state);
1444 		if (status >= 0) {
1445 			status = nfs4_reclaim_locks(state, ops);
1446 			if (status >= 0) {
1447 				if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1448 					spin_lock(&state->state_lock);
1449 					list_for_each_entry(lock, &state->lock_states, ls_locks) {
1450 						if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1451 							pr_warn_ratelimited("NFS: "
1452 									    "%s: Lock reclaim "
1453 									    "failed!\n", __func__);
1454 					}
1455 					spin_unlock(&state->state_lock);
1456 				}
1457 				nfs4_put_open_state(state);
1458 				spin_lock(&sp->so_lock);
1459 				goto restart;
1460 			}
1461 		}
1462 		switch (status) {
1463 			default:
1464 				printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1465 					__func__, status);
1466 			case -ENOENT:
1467 			case -ENOMEM:
1468 			case -ESTALE:
1469 				/* Open state on this file cannot be recovered */
1470 				nfs4_state_mark_recovery_failed(state, status);
1471 				break;
1472 			case -EAGAIN:
1473 				ssleep(1);
1474 			case -NFS4ERR_ADMIN_REVOKED:
1475 			case -NFS4ERR_STALE_STATEID:
1476 			case -NFS4ERR_BAD_STATEID:
1477 			case -NFS4ERR_RECLAIM_BAD:
1478 			case -NFS4ERR_RECLAIM_CONFLICT:
1479 				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1480 				break;
1481 			case -NFS4ERR_EXPIRED:
1482 			case -NFS4ERR_NO_GRACE:
1483 				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1484 			case -NFS4ERR_STALE_CLIENTID:
1485 			case -NFS4ERR_BADSESSION:
1486 			case -NFS4ERR_BADSLOT:
1487 			case -NFS4ERR_BAD_HIGH_SLOT:
1488 			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1489 				goto out_err;
1490 		}
1491 		nfs4_put_open_state(state);
1492 		spin_lock(&sp->so_lock);
1493 		goto restart;
1494 	}
1495 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1496 	spin_unlock(&sp->so_lock);
1497 	return 0;
1498 out_err:
1499 	nfs4_put_open_state(state);
1500 	spin_lock(&sp->so_lock);
1501 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1502 	spin_unlock(&sp->so_lock);
1503 	return status;
1504 }
1505 
1506 static void nfs4_clear_open_state(struct nfs4_state *state)
1507 {
1508 	struct nfs4_lock_state *lock;
1509 
1510 	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1511 	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1512 	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1513 	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1514 	spin_lock(&state->state_lock);
1515 	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1516 		lock->ls_seqid.flags = 0;
1517 		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1518 	}
1519 	spin_unlock(&state->state_lock);
1520 }
1521 
1522 static void nfs4_reset_seqids(struct nfs_server *server,
1523 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1524 {
1525 	struct nfs_client *clp = server->nfs_client;
1526 	struct nfs4_state_owner *sp;
1527 	struct rb_node *pos;
1528 	struct nfs4_state *state;
1529 
1530 	spin_lock(&clp->cl_lock);
1531 	for (pos = rb_first(&server->state_owners);
1532 	     pos != NULL;
1533 	     pos = rb_next(pos)) {
1534 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1535 		sp->so_seqid.flags = 0;
1536 		spin_lock(&sp->so_lock);
1537 		list_for_each_entry(state, &sp->so_states, open_states) {
1538 			if (mark_reclaim(clp, state))
1539 				nfs4_clear_open_state(state);
1540 		}
1541 		spin_unlock(&sp->so_lock);
1542 	}
1543 	spin_unlock(&clp->cl_lock);
1544 }
1545 
1546 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1547 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1548 {
1549 	struct nfs_server *server;
1550 
1551 	rcu_read_lock();
1552 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1553 		nfs4_reset_seqids(server, mark_reclaim);
1554 	rcu_read_unlock();
1555 }
1556 
1557 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1558 {
1559 	/* Mark all delegations for reclaim */
1560 	nfs_delegation_mark_reclaim(clp);
1561 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1562 }
1563 
1564 static void nfs4_reclaim_complete(struct nfs_client *clp,
1565 				 const struct nfs4_state_recovery_ops *ops,
1566 				 struct rpc_cred *cred)
1567 {
1568 	/* Notify the server we're done reclaiming our state */
1569 	if (ops->reclaim_complete)
1570 		(void)ops->reclaim_complete(clp, cred);
1571 }
1572 
1573 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1574 {
1575 	struct nfs_client *clp = server->nfs_client;
1576 	struct nfs4_state_owner *sp;
1577 	struct rb_node *pos;
1578 	struct nfs4_state *state;
1579 
1580 	spin_lock(&clp->cl_lock);
1581 	for (pos = rb_first(&server->state_owners);
1582 	     pos != NULL;
1583 	     pos = rb_next(pos)) {
1584 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1585 		spin_lock(&sp->so_lock);
1586 		list_for_each_entry(state, &sp->so_states, open_states) {
1587 			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1588 						&state->flags))
1589 				continue;
1590 			nfs4_state_mark_reclaim_nograce(clp, state);
1591 		}
1592 		spin_unlock(&sp->so_lock);
1593 	}
1594 	spin_unlock(&clp->cl_lock);
1595 }
1596 
1597 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1598 {
1599 	struct nfs_server *server;
1600 
1601 	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1602 		return 0;
1603 
1604 	rcu_read_lock();
1605 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1606 		nfs4_clear_reclaim_server(server);
1607 	rcu_read_unlock();
1608 
1609 	nfs_delegation_reap_unclaimed(clp);
1610 	return 1;
1611 }
1612 
1613 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1614 {
1615 	const struct nfs4_state_recovery_ops *ops;
1616 	struct rpc_cred *cred;
1617 
1618 	if (!nfs4_state_clear_reclaim_reboot(clp))
1619 		return;
1620 	ops = clp->cl_mvops->reboot_recovery_ops;
1621 	cred = nfs4_get_clid_cred(clp);
1622 	nfs4_reclaim_complete(clp, ops, cred);
1623 	put_rpccred(cred);
1624 }
1625 
1626 static void nfs_delegation_clear_all(struct nfs_client *clp)
1627 {
1628 	nfs_delegation_mark_reclaim(clp);
1629 	nfs_delegation_reap_unclaimed(clp);
1630 }
1631 
1632 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1633 {
1634 	nfs_delegation_clear_all(clp);
1635 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1636 }
1637 
1638 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1639 {
1640 	switch (error) {
1641 		case 0:
1642 			break;
1643 		case -NFS4ERR_CB_PATH_DOWN:
1644 			nfs40_handle_cb_pathdown(clp);
1645 			break;
1646 		case -NFS4ERR_NO_GRACE:
1647 			nfs4_state_end_reclaim_reboot(clp);
1648 			break;
1649 		case -NFS4ERR_STALE_CLIENTID:
1650 			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1651 			nfs4_state_clear_reclaim_reboot(clp);
1652 			nfs4_state_start_reclaim_reboot(clp);
1653 			break;
1654 		case -NFS4ERR_EXPIRED:
1655 			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1656 			nfs4_state_start_reclaim_nograce(clp);
1657 			break;
1658 		case -NFS4ERR_BADSESSION:
1659 		case -NFS4ERR_BADSLOT:
1660 		case -NFS4ERR_BAD_HIGH_SLOT:
1661 		case -NFS4ERR_DEADSESSION:
1662 		case -NFS4ERR_SEQ_FALSE_RETRY:
1663 		case -NFS4ERR_SEQ_MISORDERED:
1664 			set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1665 			/* Zero session reset errors */
1666 			break;
1667 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1668 			set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1669 			break;
1670 		default:
1671 			dprintk("%s: failed to handle error %d for server %s\n",
1672 					__func__, error, clp->cl_hostname);
1673 			return error;
1674 	}
1675 	dprintk("%s: handled error %d for server %s\n", __func__, error,
1676 			clp->cl_hostname);
1677 	return 0;
1678 }
1679 
1680 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1681 {
1682 	struct nfs4_state_owner *sp;
1683 	struct nfs_server *server;
1684 	struct rb_node *pos;
1685 	int status = 0;
1686 
1687 restart:
1688 	rcu_read_lock();
1689 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1690 		nfs4_purge_state_owners(server);
1691 		spin_lock(&clp->cl_lock);
1692 		for (pos = rb_first(&server->state_owners);
1693 		     pos != NULL;
1694 		     pos = rb_next(pos)) {
1695 			sp = rb_entry(pos,
1696 				struct nfs4_state_owner, so_server_node);
1697 			if (!test_and_clear_bit(ops->owner_flag_bit,
1698 							&sp->so_flags))
1699 				continue;
1700 			atomic_inc(&sp->so_count);
1701 			spin_unlock(&clp->cl_lock);
1702 			rcu_read_unlock();
1703 
1704 			status = nfs4_reclaim_open_state(sp, ops);
1705 			if (status < 0) {
1706 				set_bit(ops->owner_flag_bit, &sp->so_flags);
1707 				nfs4_put_state_owner(sp);
1708 				status = nfs4_recovery_handle_error(clp, status);
1709 				return (status != 0) ? status : -EAGAIN;
1710 			}
1711 
1712 			nfs4_put_state_owner(sp);
1713 			goto restart;
1714 		}
1715 		spin_unlock(&clp->cl_lock);
1716 	}
1717 	rcu_read_unlock();
1718 	return 0;
1719 }
1720 
1721 static int nfs4_check_lease(struct nfs_client *clp)
1722 {
1723 	struct rpc_cred *cred;
1724 	const struct nfs4_state_maintenance_ops *ops =
1725 		clp->cl_mvops->state_renewal_ops;
1726 	int status;
1727 
1728 	/* Is the client already known to have an expired lease? */
1729 	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1730 		return 0;
1731 	spin_lock(&clp->cl_lock);
1732 	cred = ops->get_state_renewal_cred_locked(clp);
1733 	spin_unlock(&clp->cl_lock);
1734 	if (cred == NULL) {
1735 		cred = nfs4_get_clid_cred(clp);
1736 		status = -ENOKEY;
1737 		if (cred == NULL)
1738 			goto out;
1739 	}
1740 	status = ops->renew_lease(clp, cred);
1741 	put_rpccred(cred);
1742 	if (status == -ETIMEDOUT) {
1743 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1744 		return 0;
1745 	}
1746 out:
1747 	return nfs4_recovery_handle_error(clp, status);
1748 }
1749 
1750 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1751  * and for recoverable errors on EXCHANGE_ID for v4.1
1752  */
1753 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1754 {
1755 	switch (status) {
1756 	case -NFS4ERR_SEQ_MISORDERED:
1757 		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1758 			return -ESERVERFAULT;
1759 		/* Lease confirmation error: retry after purging the lease */
1760 		ssleep(1);
1761 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1762 		break;
1763 	case -NFS4ERR_STALE_CLIENTID:
1764 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1765 		nfs4_state_start_reclaim_reboot(clp);
1766 		break;
1767 	case -NFS4ERR_CLID_INUSE:
1768 		pr_err("NFS: Server %s reports our clientid is in use\n",
1769 			clp->cl_hostname);
1770 		nfs_mark_client_ready(clp, -EPERM);
1771 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1772 		return -EPERM;
1773 	case -EACCES:
1774 	case -NFS4ERR_DELAY:
1775 	case -ETIMEDOUT:
1776 	case -EAGAIN:
1777 		ssleep(1);
1778 		break;
1779 
1780 	case -NFS4ERR_MINOR_VERS_MISMATCH:
1781 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1782 			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1783 		dprintk("%s: exit with error %d for server %s\n",
1784 				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
1785 		return -EPROTONOSUPPORT;
1786 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1787 				 * in nfs4_exchange_id */
1788 	default:
1789 		dprintk("%s: exit with error %d for server %s\n", __func__,
1790 				status, clp->cl_hostname);
1791 		return status;
1792 	}
1793 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1794 	dprintk("%s: handled error %d for server %s\n", __func__, status,
1795 			clp->cl_hostname);
1796 	return 0;
1797 }
1798 
1799 static int nfs4_establish_lease(struct nfs_client *clp)
1800 {
1801 	struct rpc_cred *cred;
1802 	const struct nfs4_state_recovery_ops *ops =
1803 		clp->cl_mvops->reboot_recovery_ops;
1804 	int status;
1805 
1806 	cred = nfs4_get_clid_cred(clp);
1807 	if (cred == NULL)
1808 		return -ENOENT;
1809 	status = ops->establish_clid(clp, cred);
1810 	put_rpccred(cred);
1811 	if (status != 0)
1812 		return status;
1813 	pnfs_destroy_all_layouts(clp);
1814 	return 0;
1815 }
1816 
1817 /*
1818  * Returns zero or a negative errno.  NFS4ERR values are converted
1819  * to local errno values.
1820  */
1821 static int nfs4_reclaim_lease(struct nfs_client *clp)
1822 {
1823 	int status;
1824 
1825 	status = nfs4_establish_lease(clp);
1826 	if (status < 0)
1827 		return nfs4_handle_reclaim_lease_error(clp, status);
1828 	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1829 		nfs4_state_start_reclaim_nograce(clp);
1830 	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1831 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1832 	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1833 	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1834 	return 0;
1835 }
1836 
1837 static int nfs4_purge_lease(struct nfs_client *clp)
1838 {
1839 	int status;
1840 
1841 	status = nfs4_establish_lease(clp);
1842 	if (status < 0)
1843 		return nfs4_handle_reclaim_lease_error(clp, status);
1844 	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1845 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1846 	nfs4_state_start_reclaim_nograce(clp);
1847 	return 0;
1848 }
1849 
1850 /*
1851  * Try remote migration of one FSID from a source server to a
1852  * destination server.  The source server provides a list of
1853  * potential destinations.
1854  *
1855  * Returns zero or a negative NFS4ERR status code.
1856  */
1857 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1858 {
1859 	struct nfs_client *clp = server->nfs_client;
1860 	struct nfs4_fs_locations *locations = NULL;
1861 	struct inode *inode;
1862 	struct page *page;
1863 	int status, result;
1864 
1865 	dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1866 			(unsigned long long)server->fsid.major,
1867 			(unsigned long long)server->fsid.minor,
1868 			clp->cl_hostname);
1869 
1870 	result = 0;
1871 	page = alloc_page(GFP_KERNEL);
1872 	locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1873 	if (page == NULL || locations == NULL) {
1874 		dprintk("<-- %s: no memory\n", __func__);
1875 		goto out;
1876 	}
1877 
1878 	inode = server->super->s_root->d_inode;
1879 	result = nfs4_proc_get_locations(inode, locations, page, cred);
1880 	if (result) {
1881 		dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1882 			__func__, result);
1883 		goto out;
1884 	}
1885 
1886 	result = -NFS4ERR_NXIO;
1887 	if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1888 		dprintk("<-- %s: No fs_locations data, migration skipped\n",
1889 			__func__);
1890 		goto out;
1891 	}
1892 
1893 	nfs4_begin_drain_session(clp);
1894 
1895 	status = nfs4_replace_transport(server, locations);
1896 	if (status != 0) {
1897 		dprintk("<-- %s: failed to replace transport: %d\n",
1898 			__func__, status);
1899 		goto out;
1900 	}
1901 
1902 	result = 0;
1903 	dprintk("<-- %s: migration succeeded\n", __func__);
1904 
1905 out:
1906 	if (page != NULL)
1907 		__free_page(page);
1908 	kfree(locations);
1909 	if (result) {
1910 		pr_err("NFS: migration recovery failed (server %s)\n",
1911 				clp->cl_hostname);
1912 		set_bit(NFS_MIG_FAILED, &server->mig_status);
1913 	}
1914 	return result;
1915 }
1916 
1917 /*
1918  * Returns zero or a negative NFS4ERR status code.
1919  */
1920 static int nfs4_handle_migration(struct nfs_client *clp)
1921 {
1922 	const struct nfs4_state_maintenance_ops *ops =
1923 				clp->cl_mvops->state_renewal_ops;
1924 	struct nfs_server *server;
1925 	struct rpc_cred *cred;
1926 
1927 	dprintk("%s: migration reported on \"%s\"\n", __func__,
1928 			clp->cl_hostname);
1929 
1930 	spin_lock(&clp->cl_lock);
1931 	cred = ops->get_state_renewal_cred_locked(clp);
1932 	spin_unlock(&clp->cl_lock);
1933 	if (cred == NULL)
1934 		return -NFS4ERR_NOENT;
1935 
1936 	clp->cl_mig_gen++;
1937 restart:
1938 	rcu_read_lock();
1939 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1940 		int status;
1941 
1942 		if (server->mig_gen == clp->cl_mig_gen)
1943 			continue;
1944 		server->mig_gen = clp->cl_mig_gen;
1945 
1946 		if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1947 						&server->mig_status))
1948 			continue;
1949 
1950 		rcu_read_unlock();
1951 		status = nfs4_try_migration(server, cred);
1952 		if (status < 0) {
1953 			put_rpccred(cred);
1954 			return status;
1955 		}
1956 		goto restart;
1957 	}
1958 	rcu_read_unlock();
1959 	put_rpccred(cred);
1960 	return 0;
1961 }
1962 
1963 /*
1964  * Test each nfs_server on the clp's cl_superblocks list to see
1965  * if it's moved to another server.  Stop when the server no longer
1966  * returns NFS4ERR_LEASE_MOVED.
1967  */
1968 static int nfs4_handle_lease_moved(struct nfs_client *clp)
1969 {
1970 	const struct nfs4_state_maintenance_ops *ops =
1971 				clp->cl_mvops->state_renewal_ops;
1972 	struct nfs_server *server;
1973 	struct rpc_cred *cred;
1974 
1975 	dprintk("%s: lease moved reported on \"%s\"\n", __func__,
1976 			clp->cl_hostname);
1977 
1978 	spin_lock(&clp->cl_lock);
1979 	cred = ops->get_state_renewal_cred_locked(clp);
1980 	spin_unlock(&clp->cl_lock);
1981 	if (cred == NULL)
1982 		return -NFS4ERR_NOENT;
1983 
1984 	clp->cl_mig_gen++;
1985 restart:
1986 	rcu_read_lock();
1987 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1988 		struct inode *inode;
1989 		int status;
1990 
1991 		if (server->mig_gen == clp->cl_mig_gen)
1992 			continue;
1993 		server->mig_gen = clp->cl_mig_gen;
1994 
1995 		rcu_read_unlock();
1996 
1997 		inode = server->super->s_root->d_inode;
1998 		status = nfs4_proc_fsid_present(inode, cred);
1999 		if (status != -NFS4ERR_MOVED)
2000 			goto restart;	/* wasn't this one */
2001 		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2002 			goto restart;	/* there are more */
2003 		goto out;
2004 	}
2005 	rcu_read_unlock();
2006 
2007 out:
2008 	put_rpccred(cred);
2009 	return 0;
2010 }
2011 
2012 /**
2013  * nfs4_discover_server_trunking - Detect server IP address trunking
2014  *
2015  * @clp: nfs_client under test
2016  * @result: OUT: found nfs_client, or clp
2017  *
2018  * Returns zero or a negative errno.  If zero is returned,
2019  * an nfs_client pointer is planted in "result".
2020  *
2021  * Note: since we are invoked in process context, and
2022  * not from inside the state manager, we cannot use
2023  * nfs4_handle_reclaim_lease_error().
2024  */
2025 int nfs4_discover_server_trunking(struct nfs_client *clp,
2026 				  struct nfs_client **result)
2027 {
2028 	const struct nfs4_state_recovery_ops *ops =
2029 				clp->cl_mvops->reboot_recovery_ops;
2030 	struct rpc_clnt *clnt;
2031 	struct rpc_cred *cred;
2032 	int i, status;
2033 
2034 	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2035 
2036 	clnt = clp->cl_rpcclient;
2037 	i = 0;
2038 
2039 	mutex_lock(&nfs_clid_init_mutex);
2040 again:
2041 	status  = -ENOENT;
2042 	cred = nfs4_get_clid_cred(clp);
2043 	if (cred == NULL)
2044 		goto out_unlock;
2045 
2046 	status = ops->detect_trunking(clp, result, cred);
2047 	put_rpccred(cred);
2048 	switch (status) {
2049 	case 0:
2050 		break;
2051 	case -ETIMEDOUT:
2052 		if (clnt->cl_softrtry)
2053 			break;
2054 	case -NFS4ERR_DELAY:
2055 	case -EAGAIN:
2056 		ssleep(1);
2057 	case -NFS4ERR_STALE_CLIENTID:
2058 		dprintk("NFS: %s after status %d, retrying\n",
2059 			__func__, status);
2060 		goto again;
2061 	case -EACCES:
2062 		if (i++ == 0) {
2063 			nfs4_root_machine_cred(clp);
2064 			goto again;
2065 		}
2066 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2067 			break;
2068 	case -NFS4ERR_CLID_INUSE:
2069 	case -NFS4ERR_WRONGSEC:
2070 		/* No point in retrying if we already used RPC_AUTH_UNIX */
2071 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2072 			status = -EPERM;
2073 			break;
2074 		}
2075 		clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2076 		if (IS_ERR(clnt)) {
2077 			status = PTR_ERR(clnt);
2078 			break;
2079 		}
2080 		/* Note: this is safe because we haven't yet marked the
2081 		 * client as ready, so we are the only user of
2082 		 * clp->cl_rpcclient
2083 		 */
2084 		clnt = xchg(&clp->cl_rpcclient, clnt);
2085 		rpc_shutdown_client(clnt);
2086 		clnt = clp->cl_rpcclient;
2087 		goto again;
2088 
2089 	case -NFS4ERR_MINOR_VERS_MISMATCH:
2090 		status = -EPROTONOSUPPORT;
2091 		break;
2092 
2093 	case -EKEYEXPIRED:
2094 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2095 				 * in nfs4_exchange_id */
2096 		status = -EKEYEXPIRED;
2097 		break;
2098 	default:
2099 		pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2100 				__func__, status);
2101 		status = -EIO;
2102 	}
2103 
2104 out_unlock:
2105 	mutex_unlock(&nfs_clid_init_mutex);
2106 	dprintk("NFS: %s: status = %d\n", __func__, status);
2107 	return status;
2108 }
2109 
2110 #ifdef CONFIG_NFS_V4_1
2111 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2112 {
2113 	struct nfs_client *clp = session->clp;
2114 
2115 	switch (err) {
2116 	default:
2117 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2118 		break;
2119 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2120 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2121 	}
2122 	nfs4_schedule_lease_recovery(clp);
2123 }
2124 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2125 
2126 static void nfs41_ping_server(struct nfs_client *clp)
2127 {
2128 	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
2129 	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2130 	nfs4_schedule_state_manager(clp);
2131 }
2132 
2133 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2134 {
2135 	nfs41_ping_server(clp);
2136 }
2137 
2138 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2139 {
2140 	nfs41_ping_server(clp);
2141 }
2142 
2143 static void nfs4_reset_all_state(struct nfs_client *clp)
2144 {
2145 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2146 		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2147 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2148 		nfs4_state_start_reclaim_nograce(clp);
2149 		dprintk("%s: scheduling reset of all state for server %s!\n",
2150 				__func__, clp->cl_hostname);
2151 		nfs4_schedule_state_manager(clp);
2152 	}
2153 }
2154 
2155 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2156 {
2157 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2158 		nfs4_state_start_reclaim_reboot(clp);
2159 		dprintk("%s: server %s rebooted!\n", __func__,
2160 				clp->cl_hostname);
2161 		nfs4_schedule_state_manager(clp);
2162 	}
2163 }
2164 
2165 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2166 {
2167 	nfs4_reset_all_state(clp);
2168 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2169 }
2170 
2171 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2172 {
2173 	/* This will need to handle layouts too */
2174 	nfs_expire_all_delegations(clp);
2175 	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2176 			clp->cl_hostname);
2177 }
2178 
2179 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2180 {
2181 	nfs_expire_all_delegations(clp);
2182 	if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2183 		nfs4_schedule_state_manager(clp);
2184 	dprintk("%s: server %s declared a backchannel fault\n", __func__,
2185 			clp->cl_hostname);
2186 }
2187 
2188 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2189 {
2190 	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2191 		&clp->cl_state) == 0)
2192 		nfs4_schedule_state_manager(clp);
2193 }
2194 
2195 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2196 {
2197 	if (!flags)
2198 		return;
2199 
2200 	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2201 		__func__, clp->cl_hostname, clp->cl_clientid, flags);
2202 
2203 	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2204 		nfs41_handle_server_reboot(clp);
2205 	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2206 			    SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2207 			    SEQ4_STATUS_ADMIN_STATE_REVOKED))
2208 		nfs41_handle_state_revoked(clp);
2209 	if (flags & SEQ4_STATUS_LEASE_MOVED)
2210 		nfs4_schedule_lease_moved_recovery(clp);
2211 	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2212 		nfs41_handle_recallable_state_revoked(clp);
2213 	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2214 		nfs41_handle_backchannel_fault(clp);
2215 	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2216 				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2217 		nfs41_handle_cb_path_down(clp);
2218 }
2219 
2220 static int nfs4_reset_session(struct nfs_client *clp)
2221 {
2222 	struct rpc_cred *cred;
2223 	int status;
2224 
2225 	if (!nfs4_has_session(clp))
2226 		return 0;
2227 	nfs4_begin_drain_session(clp);
2228 	cred = nfs4_get_clid_cred(clp);
2229 	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2230 	switch (status) {
2231 	case 0:
2232 	case -NFS4ERR_BADSESSION:
2233 	case -NFS4ERR_DEADSESSION:
2234 		break;
2235 	case -NFS4ERR_BACK_CHAN_BUSY:
2236 	case -NFS4ERR_DELAY:
2237 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2238 		status = 0;
2239 		ssleep(1);
2240 		goto out;
2241 	default:
2242 		status = nfs4_recovery_handle_error(clp, status);
2243 		goto out;
2244 	}
2245 
2246 	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2247 	status = nfs4_proc_create_session(clp, cred);
2248 	if (status) {
2249 		dprintk("%s: session reset failed with status %d for server %s!\n",
2250 			__func__, status, clp->cl_hostname);
2251 		status = nfs4_handle_reclaim_lease_error(clp, status);
2252 		goto out;
2253 	}
2254 	nfs41_finish_session_reset(clp);
2255 	dprintk("%s: session reset was successful for server %s!\n",
2256 			__func__, clp->cl_hostname);
2257 out:
2258 	if (cred)
2259 		put_rpccred(cred);
2260 	return status;
2261 }
2262 
2263 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2264 {
2265 	struct rpc_cred *cred;
2266 	int ret;
2267 
2268 	if (!nfs4_has_session(clp))
2269 		return 0;
2270 	nfs4_begin_drain_session(clp);
2271 	cred = nfs4_get_clid_cred(clp);
2272 	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2273 	if (cred)
2274 		put_rpccred(cred);
2275 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2276 	switch (ret) {
2277 	case 0:
2278 		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2279 			__func__, clp->cl_hostname);
2280 		break;
2281 	case -NFS4ERR_DELAY:
2282 		ssleep(1);
2283 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2284 		break;
2285 	default:
2286 		return nfs4_recovery_handle_error(clp, ret);
2287 	}
2288 	return 0;
2289 }
2290 #else /* CONFIG_NFS_V4_1 */
2291 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2292 
2293 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2294 {
2295 	return 0;
2296 }
2297 #endif /* CONFIG_NFS_V4_1 */
2298 
2299 static void nfs4_state_manager(struct nfs_client *clp)
2300 {
2301 	int status = 0;
2302 	const char *section = "", *section_sep = "";
2303 
2304 	/* Ensure exclusive access to NFSv4 state */
2305 	do {
2306 		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2307 			section = "purge state";
2308 			status = nfs4_purge_lease(clp);
2309 			if (status < 0)
2310 				goto out_error;
2311 			continue;
2312 		}
2313 
2314 		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2315 			section = "lease expired";
2316 			/* We're going to have to re-establish a clientid */
2317 			status = nfs4_reclaim_lease(clp);
2318 			if (status < 0)
2319 				goto out_error;
2320 			continue;
2321 		}
2322 
2323 		/* Initialize or reset the session */
2324 		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2325 			section = "reset session";
2326 			status = nfs4_reset_session(clp);
2327 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2328 				continue;
2329 			if (status < 0)
2330 				goto out_error;
2331 		}
2332 
2333 		/* Send BIND_CONN_TO_SESSION */
2334 		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2335 				&clp->cl_state)) {
2336 			section = "bind conn to session";
2337 			status = nfs4_bind_conn_to_session(clp);
2338 			if (status < 0)
2339 				goto out_error;
2340 			continue;
2341 		}
2342 
2343 		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2344 			section = "check lease";
2345 			status = nfs4_check_lease(clp);
2346 			if (status < 0)
2347 				goto out_error;
2348 			continue;
2349 		}
2350 
2351 		if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2352 			section = "migration";
2353 			status = nfs4_handle_migration(clp);
2354 			if (status < 0)
2355 				goto out_error;
2356 		}
2357 
2358 		if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2359 			section = "lease moved";
2360 			status = nfs4_handle_lease_moved(clp);
2361 			if (status < 0)
2362 				goto out_error;
2363 		}
2364 
2365 		/* First recover reboot state... */
2366 		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2367 			section = "reclaim reboot";
2368 			status = nfs4_do_reclaim(clp,
2369 				clp->cl_mvops->reboot_recovery_ops);
2370 			if (status == -EAGAIN)
2371 				continue;
2372 			if (status < 0)
2373 				goto out_error;
2374 			nfs4_state_end_reclaim_reboot(clp);
2375 		}
2376 
2377 		/* Now recover expired state... */
2378 		if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2379 			section = "reclaim nograce";
2380 			status = nfs4_do_reclaim(clp,
2381 				clp->cl_mvops->nograce_recovery_ops);
2382 			if (status == -EAGAIN)
2383 				continue;
2384 			if (status < 0)
2385 				goto out_error;
2386 		}
2387 
2388 		nfs4_end_drain_session(clp);
2389 		if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2390 			nfs_client_return_marked_delegations(clp);
2391 			continue;
2392 		}
2393 
2394 		nfs4_clear_state_manager_bit(clp);
2395 		/* Did we race with an attempt to give us more work? */
2396 		if (clp->cl_state == 0)
2397 			break;
2398 		if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2399 			break;
2400 	} while (atomic_read(&clp->cl_count) > 1);
2401 	return;
2402 out_error:
2403 	if (strlen(section))
2404 		section_sep = ": ";
2405 	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2406 			" with error %d\n", section_sep, section,
2407 			clp->cl_hostname, -status);
2408 	ssleep(1);
2409 	nfs4_end_drain_session(clp);
2410 	nfs4_clear_state_manager_bit(clp);
2411 }
2412 
2413 static int nfs4_run_state_manager(void *ptr)
2414 {
2415 	struct nfs_client *clp = ptr;
2416 
2417 	allow_signal(SIGKILL);
2418 	nfs4_state_manager(clp);
2419 	nfs_put_client(clp);
2420 	module_put_and_exit(0);
2421 	return 0;
2422 }
2423 
2424 /*
2425  * Local variables:
2426  *  c-basic-offset: 8
2427  * End:
2428  */
2429