xref: /linux/fs/nfs/nfs4state.c (revision d670b479586e457c7c36604cea08ae236fb933ac)
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 "nfs4_fs.h"
55 #include "callback.h"
56 #include "delegation.h"
57 #include "internal.h"
58 #include "pnfs.h"
59 
60 #define NFSDBG_FACILITY		NFSDBG_STATE
61 
62 #define OPENOWNER_POOL_SIZE	8
63 
64 const nfs4_stateid zero_stateid;
65 
66 static LIST_HEAD(nfs4_clientid_list);
67 
68 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
69 {
70 	struct nfs4_setclientid_res clid = {
71 		.clientid = clp->cl_clientid,
72 		.confirm = clp->cl_confirm,
73 	};
74 	unsigned short port;
75 	int status;
76 
77 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
78 		goto do_confirm;
79 	port = nfs_callback_tcpport;
80 	if (clp->cl_addr.ss_family == AF_INET6)
81 		port = nfs_callback_tcpport6;
82 
83 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
84 	if (status != 0)
85 		goto out;
86 	clp->cl_clientid = clid.clientid;
87 	clp->cl_confirm = clid.confirm;
88 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
89 do_confirm:
90 	status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
91 	if (status != 0)
92 		goto out;
93 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
94 	nfs4_schedule_state_renewal(clp);
95 out:
96 	return status;
97 }
98 
99 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
100 {
101 	struct rpc_cred *cred = NULL;
102 
103 	if (clp->cl_machine_cred != NULL)
104 		cred = get_rpccred(clp->cl_machine_cred);
105 	return cred;
106 }
107 
108 static void nfs4_clear_machine_cred(struct nfs_client *clp)
109 {
110 	struct rpc_cred *cred;
111 
112 	spin_lock(&clp->cl_lock);
113 	cred = clp->cl_machine_cred;
114 	clp->cl_machine_cred = NULL;
115 	spin_unlock(&clp->cl_lock);
116 	if (cred != NULL)
117 		put_rpccred(cred);
118 }
119 
120 static struct rpc_cred *
121 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
122 {
123 	struct rpc_cred *cred = NULL;
124 	struct nfs4_state_owner *sp;
125 	struct rb_node *pos;
126 
127 	for (pos = rb_first(&server->state_owners);
128 	     pos != NULL;
129 	     pos = rb_next(pos)) {
130 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
131 		if (list_empty(&sp->so_states))
132 			continue;
133 		cred = get_rpccred(sp->so_cred);
134 		break;
135 	}
136 	return cred;
137 }
138 
139 /**
140  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
141  * @clp: client state handle
142  *
143  * Returns an rpc_cred with reference count bumped, or NULL.
144  * Caller must hold clp->cl_lock.
145  */
146 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
147 {
148 	struct rpc_cred *cred = NULL;
149 	struct nfs_server *server;
150 
151 	/* Use machine credentials if available */
152 	cred = nfs4_get_machine_cred_locked(clp);
153 	if (cred != NULL)
154 		goto out;
155 
156 	rcu_read_lock();
157 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
158 		cred = nfs4_get_renew_cred_server_locked(server);
159 		if (cred != NULL)
160 			break;
161 	}
162 	rcu_read_unlock();
163 
164 out:
165 	return cred;
166 }
167 
168 #if defined(CONFIG_NFS_V4_1)
169 
170 static int nfs41_setup_state_renewal(struct nfs_client *clp)
171 {
172 	int status;
173 	struct nfs_fsinfo fsinfo;
174 
175 	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
176 		nfs4_schedule_state_renewal(clp);
177 		return 0;
178 	}
179 
180 	status = nfs4_proc_get_lease_time(clp, &fsinfo);
181 	if (status == 0) {
182 		/* Update lease time and schedule renewal */
183 		spin_lock(&clp->cl_lock);
184 		clp->cl_lease_time = fsinfo.lease_time * HZ;
185 		clp->cl_last_renewal = jiffies;
186 		spin_unlock(&clp->cl_lock);
187 
188 		nfs4_schedule_state_renewal(clp);
189 	}
190 
191 	return status;
192 }
193 
194 /*
195  * Back channel returns NFS4ERR_DELAY for new requests when
196  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
197  * is ended.
198  */
199 static void nfs4_end_drain_session(struct nfs_client *clp)
200 {
201 	struct nfs4_session *ses = clp->cl_session;
202 	struct nfs4_slot_table *tbl;
203 	int max_slots;
204 
205 	if (ses == NULL)
206 		return;
207 	tbl = &ses->fc_slot_table;
208 	if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
209 		spin_lock(&tbl->slot_tbl_lock);
210 		max_slots = tbl->max_slots;
211 		while (max_slots--) {
212 			if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
213 						nfs4_set_task_privileged,
214 						NULL) == NULL)
215 				break;
216 		}
217 		spin_unlock(&tbl->slot_tbl_lock);
218 	}
219 }
220 
221 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
222 {
223 	spin_lock(&tbl->slot_tbl_lock);
224 	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
225 		INIT_COMPLETION(tbl->complete);
226 		spin_unlock(&tbl->slot_tbl_lock);
227 		return wait_for_completion_interruptible(&tbl->complete);
228 	}
229 	spin_unlock(&tbl->slot_tbl_lock);
230 	return 0;
231 }
232 
233 static int nfs4_begin_drain_session(struct nfs_client *clp)
234 {
235 	struct nfs4_session *ses = clp->cl_session;
236 	int ret = 0;
237 
238 	set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
239 	/* back channel */
240 	ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
241 	if (ret)
242 		return ret;
243 	/* fore channel */
244 	return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
245 }
246 
247 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
248 {
249 	int status;
250 
251 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
252 		goto do_confirm;
253 	nfs4_begin_drain_session(clp);
254 	status = nfs4_proc_exchange_id(clp, cred);
255 	if (status != 0)
256 		goto out;
257 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
258 do_confirm:
259 	status = nfs4_proc_create_session(clp, cred);
260 	if (status != 0)
261 		goto out;
262 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
263 	nfs41_setup_state_renewal(clp);
264 	nfs_mark_client_ready(clp, NFS_CS_READY);
265 out:
266 	return status;
267 }
268 
269 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
270 {
271 	struct rpc_cred *cred;
272 
273 	spin_lock(&clp->cl_lock);
274 	cred = nfs4_get_machine_cred_locked(clp);
275 	spin_unlock(&clp->cl_lock);
276 	return cred;
277 }
278 
279 #endif /* CONFIG_NFS_V4_1 */
280 
281 static struct rpc_cred *
282 nfs4_get_setclientid_cred_server(struct nfs_server *server)
283 {
284 	struct nfs_client *clp = server->nfs_client;
285 	struct rpc_cred *cred = NULL;
286 	struct nfs4_state_owner *sp;
287 	struct rb_node *pos;
288 
289 	spin_lock(&clp->cl_lock);
290 	pos = rb_first(&server->state_owners);
291 	if (pos != NULL) {
292 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
293 		cred = get_rpccred(sp->so_cred);
294 	}
295 	spin_unlock(&clp->cl_lock);
296 	return cred;
297 }
298 
299 /**
300  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
301  * @clp: client state handle
302  *
303  * Returns an rpc_cred with reference count bumped, or NULL.
304  */
305 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
306 {
307 	struct nfs_server *server;
308 	struct rpc_cred *cred;
309 
310 	spin_lock(&clp->cl_lock);
311 	cred = nfs4_get_machine_cred_locked(clp);
312 	spin_unlock(&clp->cl_lock);
313 	if (cred != NULL)
314 		goto out;
315 
316 	rcu_read_lock();
317 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
318 		cred = nfs4_get_setclientid_cred_server(server);
319 		if (cred != NULL)
320 			break;
321 	}
322 	rcu_read_unlock();
323 
324 out:
325 	return cred;
326 }
327 
328 static struct nfs4_state_owner *
329 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
330 {
331 	struct rb_node **p = &server->state_owners.rb_node,
332 		       *parent = NULL;
333 	struct nfs4_state_owner *sp;
334 
335 	while (*p != NULL) {
336 		parent = *p;
337 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
338 
339 		if (cred < sp->so_cred)
340 			p = &parent->rb_left;
341 		else if (cred > sp->so_cred)
342 			p = &parent->rb_right;
343 		else {
344 			if (!list_empty(&sp->so_lru))
345 				list_del_init(&sp->so_lru);
346 			atomic_inc(&sp->so_count);
347 			return sp;
348 		}
349 	}
350 	return NULL;
351 }
352 
353 static struct nfs4_state_owner *
354 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
355 {
356 	struct nfs_server *server = new->so_server;
357 	struct rb_node **p = &server->state_owners.rb_node,
358 		       *parent = NULL;
359 	struct nfs4_state_owner *sp;
360 	int err;
361 
362 	while (*p != NULL) {
363 		parent = *p;
364 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
365 
366 		if (new->so_cred < sp->so_cred)
367 			p = &parent->rb_left;
368 		else if (new->so_cred > sp->so_cred)
369 			p = &parent->rb_right;
370 		else {
371 			if (!list_empty(&sp->so_lru))
372 				list_del_init(&sp->so_lru);
373 			atomic_inc(&sp->so_count);
374 			return sp;
375 		}
376 	}
377 	err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
378 	if (err)
379 		return ERR_PTR(err);
380 	rb_link_node(&new->so_server_node, parent, p);
381 	rb_insert_color(&new->so_server_node, &server->state_owners);
382 	return new;
383 }
384 
385 static void
386 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
387 {
388 	struct nfs_server *server = sp->so_server;
389 
390 	if (!RB_EMPTY_NODE(&sp->so_server_node))
391 		rb_erase(&sp->so_server_node, &server->state_owners);
392 	ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
393 }
394 
395 static void
396 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
397 {
398 	sc->create_time = ktime_get();
399 	sc->flags = 0;
400 	sc->counter = 0;
401 	spin_lock_init(&sc->lock);
402 	INIT_LIST_HEAD(&sc->list);
403 	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
404 }
405 
406 static void
407 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
408 {
409 	rpc_destroy_wait_queue(&sc->wait);
410 }
411 
412 /*
413  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
414  * create a new state_owner.
415  *
416  */
417 static struct nfs4_state_owner *
418 nfs4_alloc_state_owner(struct nfs_server *server,
419 		struct rpc_cred *cred,
420 		gfp_t gfp_flags)
421 {
422 	struct nfs4_state_owner *sp;
423 
424 	sp = kzalloc(sizeof(*sp), gfp_flags);
425 	if (!sp)
426 		return NULL;
427 	sp->so_server = server;
428 	sp->so_cred = get_rpccred(cred);
429 	spin_lock_init(&sp->so_lock);
430 	INIT_LIST_HEAD(&sp->so_states);
431 	nfs4_init_seqid_counter(&sp->so_seqid);
432 	atomic_set(&sp->so_count, 1);
433 	INIT_LIST_HEAD(&sp->so_lru);
434 	return sp;
435 }
436 
437 static void
438 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
439 {
440 	struct rb_node *rb_node = &sp->so_server_node;
441 
442 	if (!RB_EMPTY_NODE(rb_node)) {
443 		struct nfs_server *server = sp->so_server;
444 		struct nfs_client *clp = server->nfs_client;
445 
446 		spin_lock(&clp->cl_lock);
447 		if (!RB_EMPTY_NODE(rb_node)) {
448 			rb_erase(rb_node, &server->state_owners);
449 			RB_CLEAR_NODE(rb_node);
450 		}
451 		spin_unlock(&clp->cl_lock);
452 	}
453 }
454 
455 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
456 {
457 	nfs4_destroy_seqid_counter(&sp->so_seqid);
458 	put_rpccred(sp->so_cred);
459 	kfree(sp);
460 }
461 
462 static void nfs4_gc_state_owners(struct nfs_server *server)
463 {
464 	struct nfs_client *clp = server->nfs_client;
465 	struct nfs4_state_owner *sp, *tmp;
466 	unsigned long time_min, time_max;
467 	LIST_HEAD(doomed);
468 
469 	spin_lock(&clp->cl_lock);
470 	time_max = jiffies;
471 	time_min = (long)time_max - (long)clp->cl_lease_time;
472 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
473 		/* NB: LRU is sorted so that oldest is at the head */
474 		if (time_in_range(sp->so_expires, time_min, time_max))
475 			break;
476 		list_move(&sp->so_lru, &doomed);
477 		nfs4_remove_state_owner_locked(sp);
478 	}
479 	spin_unlock(&clp->cl_lock);
480 
481 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
482 		list_del(&sp->so_lru);
483 		nfs4_free_state_owner(sp);
484 	}
485 }
486 
487 /**
488  * nfs4_get_state_owner - Look up a state owner given a credential
489  * @server: nfs_server to search
490  * @cred: RPC credential to match
491  *
492  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
493  */
494 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
495 					      struct rpc_cred *cred,
496 					      gfp_t gfp_flags)
497 {
498 	struct nfs_client *clp = server->nfs_client;
499 	struct nfs4_state_owner *sp, *new;
500 
501 	spin_lock(&clp->cl_lock);
502 	sp = nfs4_find_state_owner_locked(server, cred);
503 	spin_unlock(&clp->cl_lock);
504 	if (sp != NULL)
505 		goto out;
506 	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
507 	if (new == NULL)
508 		goto out;
509 	do {
510 		if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
511 			break;
512 		spin_lock(&clp->cl_lock);
513 		sp = nfs4_insert_state_owner_locked(new);
514 		spin_unlock(&clp->cl_lock);
515 	} while (sp == ERR_PTR(-EAGAIN));
516 	if (sp != new)
517 		nfs4_free_state_owner(new);
518 out:
519 	nfs4_gc_state_owners(server);
520 	return sp;
521 }
522 
523 /**
524  * nfs4_put_state_owner - Release a nfs4_state_owner
525  * @sp: state owner data to release
526  *
527  * Note that we keep released state owners on an LRU
528  * list.
529  * This caches valid state owners so that they can be
530  * reused, to avoid the OPEN_CONFIRM on minor version 0.
531  * It also pins the uniquifier of dropped state owners for
532  * a while, to ensure that those state owner names are
533  * never reused.
534  */
535 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
536 {
537 	struct nfs_server *server = sp->so_server;
538 	struct nfs_client *clp = server->nfs_client;
539 
540 	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
541 		return;
542 
543 	sp->so_expires = jiffies;
544 	list_add_tail(&sp->so_lru, &server->state_owners_lru);
545 	spin_unlock(&clp->cl_lock);
546 }
547 
548 /**
549  * nfs4_purge_state_owners - Release all cached state owners
550  * @server: nfs_server with cached state owners to release
551  *
552  * Called at umount time.  Remaining state owners will be on
553  * the LRU with ref count of zero.
554  */
555 void nfs4_purge_state_owners(struct nfs_server *server)
556 {
557 	struct nfs_client *clp = server->nfs_client;
558 	struct nfs4_state_owner *sp, *tmp;
559 	LIST_HEAD(doomed);
560 
561 	spin_lock(&clp->cl_lock);
562 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
563 		list_move(&sp->so_lru, &doomed);
564 		nfs4_remove_state_owner_locked(sp);
565 	}
566 	spin_unlock(&clp->cl_lock);
567 
568 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
569 		list_del(&sp->so_lru);
570 		nfs4_free_state_owner(sp);
571 	}
572 }
573 
574 static struct nfs4_state *
575 nfs4_alloc_open_state(void)
576 {
577 	struct nfs4_state *state;
578 
579 	state = kzalloc(sizeof(*state), GFP_NOFS);
580 	if (!state)
581 		return NULL;
582 	atomic_set(&state->count, 1);
583 	INIT_LIST_HEAD(&state->lock_states);
584 	spin_lock_init(&state->state_lock);
585 	seqlock_init(&state->seqlock);
586 	return state;
587 }
588 
589 void
590 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
591 {
592 	if (state->state == fmode)
593 		return;
594 	/* NB! List reordering - see the reclaim code for why.  */
595 	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
596 		if (fmode & FMODE_WRITE)
597 			list_move(&state->open_states, &state->owner->so_states);
598 		else
599 			list_move_tail(&state->open_states, &state->owner->so_states);
600 	}
601 	state->state = fmode;
602 }
603 
604 static struct nfs4_state *
605 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
606 {
607 	struct nfs_inode *nfsi = NFS_I(inode);
608 	struct nfs4_state *state;
609 
610 	list_for_each_entry(state, &nfsi->open_states, inode_states) {
611 		if (state->owner != owner)
612 			continue;
613 		if (atomic_inc_not_zero(&state->count))
614 			return state;
615 	}
616 	return NULL;
617 }
618 
619 static void
620 nfs4_free_open_state(struct nfs4_state *state)
621 {
622 	kfree(state);
623 }
624 
625 struct nfs4_state *
626 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
627 {
628 	struct nfs4_state *state, *new;
629 	struct nfs_inode *nfsi = NFS_I(inode);
630 
631 	spin_lock(&inode->i_lock);
632 	state = __nfs4_find_state_byowner(inode, owner);
633 	spin_unlock(&inode->i_lock);
634 	if (state)
635 		goto out;
636 	new = nfs4_alloc_open_state();
637 	spin_lock(&owner->so_lock);
638 	spin_lock(&inode->i_lock);
639 	state = __nfs4_find_state_byowner(inode, owner);
640 	if (state == NULL && new != NULL) {
641 		state = new;
642 		state->owner = owner;
643 		atomic_inc(&owner->so_count);
644 		list_add(&state->inode_states, &nfsi->open_states);
645 		ihold(inode);
646 		state->inode = inode;
647 		spin_unlock(&inode->i_lock);
648 		/* Note: The reclaim code dictates that we add stateless
649 		 * and read-only stateids to the end of the list */
650 		list_add_tail(&state->open_states, &owner->so_states);
651 		spin_unlock(&owner->so_lock);
652 	} else {
653 		spin_unlock(&inode->i_lock);
654 		spin_unlock(&owner->so_lock);
655 		if (new)
656 			nfs4_free_open_state(new);
657 	}
658 out:
659 	return state;
660 }
661 
662 void nfs4_put_open_state(struct nfs4_state *state)
663 {
664 	struct inode *inode = state->inode;
665 	struct nfs4_state_owner *owner = state->owner;
666 
667 	if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
668 		return;
669 	spin_lock(&inode->i_lock);
670 	list_del(&state->inode_states);
671 	list_del(&state->open_states);
672 	spin_unlock(&inode->i_lock);
673 	spin_unlock(&owner->so_lock);
674 	iput(inode);
675 	nfs4_free_open_state(state);
676 	nfs4_put_state_owner(owner);
677 }
678 
679 /*
680  * Close the current file.
681  */
682 static void __nfs4_close(struct nfs4_state *state,
683 		fmode_t fmode, gfp_t gfp_mask, int wait)
684 {
685 	struct nfs4_state_owner *owner = state->owner;
686 	int call_close = 0;
687 	fmode_t newstate;
688 
689 	atomic_inc(&owner->so_count);
690 	/* Protect against nfs4_find_state() */
691 	spin_lock(&owner->so_lock);
692 	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
693 		case FMODE_READ:
694 			state->n_rdonly--;
695 			break;
696 		case FMODE_WRITE:
697 			state->n_wronly--;
698 			break;
699 		case FMODE_READ|FMODE_WRITE:
700 			state->n_rdwr--;
701 	}
702 	newstate = FMODE_READ|FMODE_WRITE;
703 	if (state->n_rdwr == 0) {
704 		if (state->n_rdonly == 0) {
705 			newstate &= ~FMODE_READ;
706 			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
707 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
708 		}
709 		if (state->n_wronly == 0) {
710 			newstate &= ~FMODE_WRITE;
711 			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
712 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
713 		}
714 		if (newstate == 0)
715 			clear_bit(NFS_DELEGATED_STATE, &state->flags);
716 	}
717 	nfs4_state_set_mode_locked(state, newstate);
718 	spin_unlock(&owner->so_lock);
719 
720 	if (!call_close) {
721 		nfs4_put_open_state(state);
722 		nfs4_put_state_owner(owner);
723 	} else {
724 		bool roc = pnfs_roc(state->inode);
725 
726 		nfs4_do_close(state, gfp_mask, wait, roc);
727 	}
728 }
729 
730 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
731 {
732 	__nfs4_close(state, fmode, GFP_NOFS, 0);
733 }
734 
735 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
736 {
737 	__nfs4_close(state, fmode, GFP_KERNEL, 1);
738 }
739 
740 /*
741  * Search the state->lock_states for an existing lock_owner
742  * that is compatible with current->files
743  */
744 static struct nfs4_lock_state *
745 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
746 {
747 	struct nfs4_lock_state *pos;
748 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
749 		if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
750 			continue;
751 		switch (pos->ls_owner.lo_type) {
752 		case NFS4_POSIX_LOCK_TYPE:
753 			if (pos->ls_owner.lo_u.posix_owner != fl_owner)
754 				continue;
755 			break;
756 		case NFS4_FLOCK_LOCK_TYPE:
757 			if (pos->ls_owner.lo_u.flock_owner != fl_pid)
758 				continue;
759 		}
760 		atomic_inc(&pos->ls_count);
761 		return pos;
762 	}
763 	return NULL;
764 }
765 
766 /*
767  * Return a compatible lock_state. If no initialized lock_state structure
768  * exists, return an uninitialized one.
769  *
770  */
771 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
772 {
773 	struct nfs4_lock_state *lsp;
774 	struct nfs_server *server = state->owner->so_server;
775 
776 	lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
777 	if (lsp == NULL)
778 		return NULL;
779 	nfs4_init_seqid_counter(&lsp->ls_seqid);
780 	atomic_set(&lsp->ls_count, 1);
781 	lsp->ls_state = state;
782 	lsp->ls_owner.lo_type = type;
783 	switch (lsp->ls_owner.lo_type) {
784 	case NFS4_FLOCK_LOCK_TYPE:
785 		lsp->ls_owner.lo_u.flock_owner = fl_pid;
786 		break;
787 	case NFS4_POSIX_LOCK_TYPE:
788 		lsp->ls_owner.lo_u.posix_owner = fl_owner;
789 		break;
790 	default:
791 		goto out_free;
792 	}
793 	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
794 	if (lsp->ls_seqid.owner_id < 0)
795 		goto out_free;
796 	INIT_LIST_HEAD(&lsp->ls_locks);
797 	return lsp;
798 out_free:
799 	kfree(lsp);
800 	return NULL;
801 }
802 
803 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
804 {
805 	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
806 	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
807 	kfree(lsp);
808 }
809 
810 /*
811  * Return a compatible lock_state. If no initialized lock_state structure
812  * exists, return an uninitialized one.
813  *
814  */
815 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
816 {
817 	struct nfs4_lock_state *lsp, *new = NULL;
818 
819 	for(;;) {
820 		spin_lock(&state->state_lock);
821 		lsp = __nfs4_find_lock_state(state, owner, pid, type);
822 		if (lsp != NULL)
823 			break;
824 		if (new != NULL) {
825 			list_add(&new->ls_locks, &state->lock_states);
826 			set_bit(LK_STATE_IN_USE, &state->flags);
827 			lsp = new;
828 			new = NULL;
829 			break;
830 		}
831 		spin_unlock(&state->state_lock);
832 		new = nfs4_alloc_lock_state(state, owner, pid, type);
833 		if (new == NULL)
834 			return NULL;
835 	}
836 	spin_unlock(&state->state_lock);
837 	if (new != NULL)
838 		nfs4_free_lock_state(state->owner->so_server, new);
839 	return lsp;
840 }
841 
842 /*
843  * Release reference to lock_state, and free it if we see that
844  * it is no longer in use
845  */
846 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
847 {
848 	struct nfs4_state *state;
849 
850 	if (lsp == NULL)
851 		return;
852 	state = lsp->ls_state;
853 	if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
854 		return;
855 	list_del(&lsp->ls_locks);
856 	if (list_empty(&state->lock_states))
857 		clear_bit(LK_STATE_IN_USE, &state->flags);
858 	spin_unlock(&state->state_lock);
859 	if (lsp->ls_flags & NFS_LOCK_INITIALIZED) {
860 		if (nfs4_release_lockowner(lsp) == 0)
861 			return;
862 	}
863 	nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
864 }
865 
866 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
867 {
868 	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
869 
870 	dst->fl_u.nfs4_fl.owner = lsp;
871 	atomic_inc(&lsp->ls_count);
872 }
873 
874 static void nfs4_fl_release_lock(struct file_lock *fl)
875 {
876 	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
877 }
878 
879 static const struct file_lock_operations nfs4_fl_lock_ops = {
880 	.fl_copy_lock = nfs4_fl_copy_lock,
881 	.fl_release_private = nfs4_fl_release_lock,
882 };
883 
884 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
885 {
886 	struct nfs4_lock_state *lsp;
887 
888 	if (fl->fl_ops != NULL)
889 		return 0;
890 	if (fl->fl_flags & FL_POSIX)
891 		lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
892 	else if (fl->fl_flags & FL_FLOCK)
893 		lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
894 				NFS4_FLOCK_LOCK_TYPE);
895 	else
896 		return -EINVAL;
897 	if (lsp == NULL)
898 		return -ENOMEM;
899 	fl->fl_u.nfs4_fl.owner = lsp;
900 	fl->fl_ops = &nfs4_fl_lock_ops;
901 	return 0;
902 }
903 
904 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
905 		fl_owner_t fl_owner, pid_t fl_pid)
906 {
907 	struct nfs4_lock_state *lsp;
908 	bool ret = false;
909 
910 	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
911 		goto out;
912 
913 	spin_lock(&state->state_lock);
914 	lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
915 	if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0) {
916 		nfs4_stateid_copy(dst, &lsp->ls_stateid);
917 		ret = true;
918 	}
919 	spin_unlock(&state->state_lock);
920 	nfs4_put_lock_state(lsp);
921 out:
922 	return ret;
923 }
924 
925 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
926 {
927 	int seq;
928 
929 	do {
930 		seq = read_seqbegin(&state->seqlock);
931 		nfs4_stateid_copy(dst, &state->stateid);
932 	} while (read_seqretry(&state->seqlock, seq));
933 }
934 
935 /*
936  * Byte-range lock aware utility to initialize the stateid of read/write
937  * requests.
938  */
939 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
940 		fmode_t fmode, fl_owner_t fl_owner, pid_t fl_pid)
941 {
942 	if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
943 		return;
944 	if (nfs4_copy_lock_stateid(dst, state, fl_owner, fl_pid))
945 		return;
946 	nfs4_copy_open_stateid(dst, state);
947 }
948 
949 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
950 {
951 	struct nfs_seqid *new;
952 
953 	new = kmalloc(sizeof(*new), gfp_mask);
954 	if (new != NULL) {
955 		new->sequence = counter;
956 		INIT_LIST_HEAD(&new->list);
957 		new->task = NULL;
958 	}
959 	return new;
960 }
961 
962 void nfs_release_seqid(struct nfs_seqid *seqid)
963 {
964 	struct nfs_seqid_counter *sequence;
965 
966 	if (list_empty(&seqid->list))
967 		return;
968 	sequence = seqid->sequence;
969 	spin_lock(&sequence->lock);
970 	list_del_init(&seqid->list);
971 	if (!list_empty(&sequence->list)) {
972 		struct nfs_seqid *next;
973 
974 		next = list_first_entry(&sequence->list,
975 				struct nfs_seqid, list);
976 		rpc_wake_up_queued_task(&sequence->wait, next->task);
977 	}
978 	spin_unlock(&sequence->lock);
979 }
980 
981 void nfs_free_seqid(struct nfs_seqid *seqid)
982 {
983 	nfs_release_seqid(seqid);
984 	kfree(seqid);
985 }
986 
987 /*
988  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
989  * failed with a seqid incrementing error -
990  * see comments nfs_fs.h:seqid_mutating_error()
991  */
992 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
993 {
994 	BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
995 	switch (status) {
996 		case 0:
997 			break;
998 		case -NFS4ERR_BAD_SEQID:
999 			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1000 				return;
1001 			pr_warn_ratelimited("NFS: v4 server returned a bad"
1002 					" sequence-id error on an"
1003 					" unconfirmed sequence %p!\n",
1004 					seqid->sequence);
1005 		case -NFS4ERR_STALE_CLIENTID:
1006 		case -NFS4ERR_STALE_STATEID:
1007 		case -NFS4ERR_BAD_STATEID:
1008 		case -NFS4ERR_BADXDR:
1009 		case -NFS4ERR_RESOURCE:
1010 		case -NFS4ERR_NOFILEHANDLE:
1011 			/* Non-seqid mutating errors */
1012 			return;
1013 	};
1014 	/*
1015 	 * Note: no locking needed as we are guaranteed to be first
1016 	 * on the sequence list
1017 	 */
1018 	seqid->sequence->counter++;
1019 }
1020 
1021 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1022 {
1023 	struct nfs4_state_owner *sp = container_of(seqid->sequence,
1024 					struct nfs4_state_owner, so_seqid);
1025 	struct nfs_server *server = sp->so_server;
1026 
1027 	if (status == -NFS4ERR_BAD_SEQID)
1028 		nfs4_drop_state_owner(sp);
1029 	if (!nfs4_has_session(server->nfs_client))
1030 		nfs_increment_seqid(status, seqid);
1031 }
1032 
1033 /*
1034  * Increment the seqid if the LOCK/LOCKU succeeded, or
1035  * failed with a seqid incrementing error -
1036  * see comments nfs_fs.h:seqid_mutating_error()
1037  */
1038 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1039 {
1040 	nfs_increment_seqid(status, seqid);
1041 }
1042 
1043 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1044 {
1045 	struct nfs_seqid_counter *sequence = seqid->sequence;
1046 	int status = 0;
1047 
1048 	spin_lock(&sequence->lock);
1049 	seqid->task = task;
1050 	if (list_empty(&seqid->list))
1051 		list_add_tail(&seqid->list, &sequence->list);
1052 	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1053 		goto unlock;
1054 	rpc_sleep_on(&sequence->wait, task, NULL);
1055 	status = -EAGAIN;
1056 unlock:
1057 	spin_unlock(&sequence->lock);
1058 	return status;
1059 }
1060 
1061 static int nfs4_run_state_manager(void *);
1062 
1063 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1064 {
1065 	smp_mb__before_clear_bit();
1066 	clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1067 	smp_mb__after_clear_bit();
1068 	wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1069 	rpc_wake_up(&clp->cl_rpcwaitq);
1070 }
1071 
1072 /*
1073  * Schedule the nfs_client asynchronous state management routine
1074  */
1075 void nfs4_schedule_state_manager(struct nfs_client *clp)
1076 {
1077 	struct task_struct *task;
1078 	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1079 
1080 	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1081 		return;
1082 	__module_get(THIS_MODULE);
1083 	atomic_inc(&clp->cl_count);
1084 
1085 	/* The rcu_read_lock() is not strictly necessary, as the state
1086 	 * manager is the only thread that ever changes the rpc_xprt
1087 	 * after it's initialized.  At this point, we're single threaded. */
1088 	rcu_read_lock();
1089 	snprintf(buf, sizeof(buf), "%s-manager",
1090 			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1091 	rcu_read_unlock();
1092 	task = kthread_run(nfs4_run_state_manager, clp, buf);
1093 	if (IS_ERR(task)) {
1094 		printk(KERN_ERR "%s: kthread_run: %ld\n",
1095 			__func__, PTR_ERR(task));
1096 		nfs4_clear_state_manager_bit(clp);
1097 		nfs_put_client(clp);
1098 		module_put(THIS_MODULE);
1099 	}
1100 }
1101 
1102 /*
1103  * Schedule a lease recovery attempt
1104  */
1105 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1106 {
1107 	if (!clp)
1108 		return;
1109 	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1110 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1111 	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1112 			clp->cl_hostname);
1113 	nfs4_schedule_state_manager(clp);
1114 }
1115 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1116 
1117 /*
1118  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1119  * @clp: client to process
1120  *
1121  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1122  * resend of the SETCLIENTID and hence re-establish the
1123  * callback channel. Then return all existing delegations.
1124  */
1125 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1126 {
1127 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1128 	nfs_expire_all_delegations(clp);
1129 	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1130 			clp->cl_hostname);
1131 }
1132 
1133 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1134 {
1135 	nfs40_handle_cb_pathdown(clp);
1136 	nfs4_schedule_state_manager(clp);
1137 }
1138 
1139 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1140 {
1141 
1142 	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1143 	/* Don't recover state that expired before the reboot */
1144 	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1145 		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1146 		return 0;
1147 	}
1148 	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1149 	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1150 	return 1;
1151 }
1152 
1153 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1154 {
1155 	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1156 	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1157 	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1158 	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1159 	return 1;
1160 }
1161 
1162 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1163 {
1164 	struct nfs_client *clp = server->nfs_client;
1165 
1166 	nfs4_state_mark_reclaim_nograce(clp, state);
1167 	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1168 			clp->cl_hostname);
1169 	nfs4_schedule_state_manager(clp);
1170 }
1171 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1172 
1173 void nfs_inode_find_state_and_recover(struct inode *inode,
1174 		const nfs4_stateid *stateid)
1175 {
1176 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1177 	struct nfs_inode *nfsi = NFS_I(inode);
1178 	struct nfs_open_context *ctx;
1179 	struct nfs4_state *state;
1180 	bool found = false;
1181 
1182 	spin_lock(&inode->i_lock);
1183 	list_for_each_entry(ctx, &nfsi->open_files, list) {
1184 		state = ctx->state;
1185 		if (state == NULL)
1186 			continue;
1187 		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1188 			continue;
1189 		if (!nfs4_stateid_match(&state->stateid, stateid))
1190 			continue;
1191 		nfs4_state_mark_reclaim_nograce(clp, state);
1192 		found = true;
1193 	}
1194 	spin_unlock(&inode->i_lock);
1195 	if (found)
1196 		nfs4_schedule_state_manager(clp);
1197 }
1198 
1199 
1200 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1201 {
1202 	struct inode *inode = state->inode;
1203 	struct nfs_inode *nfsi = NFS_I(inode);
1204 	struct file_lock *fl;
1205 	int status = 0;
1206 
1207 	if (inode->i_flock == NULL)
1208 		return 0;
1209 
1210 	/* Guard against delegation returns and new lock/unlock calls */
1211 	down_write(&nfsi->rwsem);
1212 	/* Protect inode->i_flock using the BKL */
1213 	lock_flocks();
1214 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1215 		if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1216 			continue;
1217 		if (nfs_file_open_context(fl->fl_file)->state != state)
1218 			continue;
1219 		unlock_flocks();
1220 		status = ops->recover_lock(state, fl);
1221 		switch (status) {
1222 			case 0:
1223 				break;
1224 			case -ESTALE:
1225 			case -NFS4ERR_ADMIN_REVOKED:
1226 			case -NFS4ERR_STALE_STATEID:
1227 			case -NFS4ERR_BAD_STATEID:
1228 			case -NFS4ERR_EXPIRED:
1229 			case -NFS4ERR_NO_GRACE:
1230 			case -NFS4ERR_STALE_CLIENTID:
1231 			case -NFS4ERR_BADSESSION:
1232 			case -NFS4ERR_BADSLOT:
1233 			case -NFS4ERR_BAD_HIGH_SLOT:
1234 			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1235 				goto out;
1236 			default:
1237 				printk(KERN_ERR "NFS: %s: unhandled error %d. "
1238 					"Zeroing state\n", __func__, status);
1239 			case -ENOMEM:
1240 			case -NFS4ERR_DENIED:
1241 			case -NFS4ERR_RECLAIM_BAD:
1242 			case -NFS4ERR_RECLAIM_CONFLICT:
1243 				/* kill_proc(fl->fl_pid, SIGLOST, 1); */
1244 				status = 0;
1245 		}
1246 		lock_flocks();
1247 	}
1248 	unlock_flocks();
1249 out:
1250 	up_write(&nfsi->rwsem);
1251 	return status;
1252 }
1253 
1254 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1255 {
1256 	struct nfs4_state *state;
1257 	struct nfs4_lock_state *lock;
1258 	int status = 0;
1259 
1260 	/* Note: we rely on the sp->so_states list being ordered
1261 	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1262 	 * states first.
1263 	 * This is needed to ensure that the server won't give us any
1264 	 * read delegations that we have to return if, say, we are
1265 	 * recovering after a network partition or a reboot from a
1266 	 * server that doesn't support a grace period.
1267 	 */
1268 restart:
1269 	spin_lock(&sp->so_lock);
1270 	list_for_each_entry(state, &sp->so_states, open_states) {
1271 		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1272 			continue;
1273 		if (state->state == 0)
1274 			continue;
1275 		atomic_inc(&state->count);
1276 		spin_unlock(&sp->so_lock);
1277 		status = ops->recover_open(sp, state);
1278 		if (status >= 0) {
1279 			status = nfs4_reclaim_locks(state, ops);
1280 			if (status >= 0) {
1281 				spin_lock(&state->state_lock);
1282 				list_for_each_entry(lock, &state->lock_states, ls_locks) {
1283 					if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1284 						pr_warn_ratelimited("NFS: "
1285 							"%s: Lock reclaim "
1286 							"failed!\n", __func__);
1287 				}
1288 				spin_unlock(&state->state_lock);
1289 				nfs4_put_open_state(state);
1290 				goto restart;
1291 			}
1292 		}
1293 		switch (status) {
1294 			default:
1295 				printk(KERN_ERR "NFS: %s: unhandled error %d. "
1296 					"Zeroing state\n", __func__, status);
1297 			case -ENOENT:
1298 			case -ENOMEM:
1299 			case -ESTALE:
1300 				/*
1301 				 * Open state on this file cannot be recovered
1302 				 * All we can do is revert to using the zero stateid.
1303 				 */
1304 				memset(&state->stateid, 0,
1305 					sizeof(state->stateid));
1306 				/* Mark the file as being 'closed' */
1307 				state->state = 0;
1308 				break;
1309 			case -EKEYEXPIRED:
1310 				/*
1311 				 * User RPCSEC_GSS context has expired.
1312 				 * We cannot recover this stateid now, so
1313 				 * skip it and allow recovery thread to
1314 				 * proceed.
1315 				 */
1316 				break;
1317 			case -NFS4ERR_ADMIN_REVOKED:
1318 			case -NFS4ERR_STALE_STATEID:
1319 			case -NFS4ERR_BAD_STATEID:
1320 			case -NFS4ERR_RECLAIM_BAD:
1321 			case -NFS4ERR_RECLAIM_CONFLICT:
1322 				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1323 				break;
1324 			case -NFS4ERR_EXPIRED:
1325 			case -NFS4ERR_NO_GRACE:
1326 				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1327 			case -NFS4ERR_STALE_CLIENTID:
1328 			case -NFS4ERR_BADSESSION:
1329 			case -NFS4ERR_BADSLOT:
1330 			case -NFS4ERR_BAD_HIGH_SLOT:
1331 			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1332 				goto out_err;
1333 		}
1334 		nfs4_put_open_state(state);
1335 		goto restart;
1336 	}
1337 	spin_unlock(&sp->so_lock);
1338 	return 0;
1339 out_err:
1340 	nfs4_put_open_state(state);
1341 	return status;
1342 }
1343 
1344 static void nfs4_clear_open_state(struct nfs4_state *state)
1345 {
1346 	struct nfs4_lock_state *lock;
1347 
1348 	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1349 	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1350 	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1351 	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1352 	spin_lock(&state->state_lock);
1353 	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1354 		lock->ls_seqid.flags = 0;
1355 		lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1356 	}
1357 	spin_unlock(&state->state_lock);
1358 }
1359 
1360 static void nfs4_reset_seqids(struct nfs_server *server,
1361 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1362 {
1363 	struct nfs_client *clp = server->nfs_client;
1364 	struct nfs4_state_owner *sp;
1365 	struct rb_node *pos;
1366 	struct nfs4_state *state;
1367 
1368 	spin_lock(&clp->cl_lock);
1369 	for (pos = rb_first(&server->state_owners);
1370 	     pos != NULL;
1371 	     pos = rb_next(pos)) {
1372 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1373 		sp->so_seqid.flags = 0;
1374 		spin_lock(&sp->so_lock);
1375 		list_for_each_entry(state, &sp->so_states, open_states) {
1376 			if (mark_reclaim(clp, state))
1377 				nfs4_clear_open_state(state);
1378 		}
1379 		spin_unlock(&sp->so_lock);
1380 	}
1381 	spin_unlock(&clp->cl_lock);
1382 }
1383 
1384 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1385 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1386 {
1387 	struct nfs_server *server;
1388 
1389 	rcu_read_lock();
1390 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1391 		nfs4_reset_seqids(server, mark_reclaim);
1392 	rcu_read_unlock();
1393 }
1394 
1395 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1396 {
1397 	/* Mark all delegations for reclaim */
1398 	nfs_delegation_mark_reclaim(clp);
1399 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1400 }
1401 
1402 static void nfs4_reclaim_complete(struct nfs_client *clp,
1403 				 const struct nfs4_state_recovery_ops *ops)
1404 {
1405 	/* Notify the server we're done reclaiming our state */
1406 	if (ops->reclaim_complete)
1407 		(void)ops->reclaim_complete(clp);
1408 }
1409 
1410 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1411 {
1412 	struct nfs_client *clp = server->nfs_client;
1413 	struct nfs4_state_owner *sp;
1414 	struct rb_node *pos;
1415 	struct nfs4_state *state;
1416 
1417 	spin_lock(&clp->cl_lock);
1418 	for (pos = rb_first(&server->state_owners);
1419 	     pos != NULL;
1420 	     pos = rb_next(pos)) {
1421 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1422 		spin_lock(&sp->so_lock);
1423 		list_for_each_entry(state, &sp->so_states, open_states) {
1424 			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1425 						&state->flags))
1426 				continue;
1427 			nfs4_state_mark_reclaim_nograce(clp, state);
1428 		}
1429 		spin_unlock(&sp->so_lock);
1430 	}
1431 	spin_unlock(&clp->cl_lock);
1432 }
1433 
1434 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1435 {
1436 	struct nfs_server *server;
1437 
1438 	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1439 		return 0;
1440 
1441 	rcu_read_lock();
1442 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1443 		nfs4_clear_reclaim_server(server);
1444 	rcu_read_unlock();
1445 
1446 	nfs_delegation_reap_unclaimed(clp);
1447 	return 1;
1448 }
1449 
1450 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1451 {
1452 	if (!nfs4_state_clear_reclaim_reboot(clp))
1453 		return;
1454 	nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1455 }
1456 
1457 static void nfs_delegation_clear_all(struct nfs_client *clp)
1458 {
1459 	nfs_delegation_mark_reclaim(clp);
1460 	nfs_delegation_reap_unclaimed(clp);
1461 }
1462 
1463 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1464 {
1465 	nfs_delegation_clear_all(clp);
1466 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1467 }
1468 
1469 static void nfs4_warn_keyexpired(const char *s)
1470 {
1471 	printk_ratelimited(KERN_WARNING "Error: state manager"
1472 			" encountered RPCSEC_GSS session"
1473 			" expired against NFSv4 server %s.\n",
1474 			s);
1475 }
1476 
1477 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1478 {
1479 	switch (error) {
1480 		case 0:
1481 			break;
1482 		case -NFS4ERR_CB_PATH_DOWN:
1483 			nfs40_handle_cb_pathdown(clp);
1484 			break;
1485 		case -NFS4ERR_NO_GRACE:
1486 			nfs4_state_end_reclaim_reboot(clp);
1487 			break;
1488 		case -NFS4ERR_STALE_CLIENTID:
1489 		case -NFS4ERR_LEASE_MOVED:
1490 			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1491 			nfs4_state_clear_reclaim_reboot(clp);
1492 			nfs4_state_start_reclaim_reboot(clp);
1493 			break;
1494 		case -NFS4ERR_EXPIRED:
1495 			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1496 			nfs4_state_start_reclaim_nograce(clp);
1497 			break;
1498 		case -NFS4ERR_BADSESSION:
1499 		case -NFS4ERR_BADSLOT:
1500 		case -NFS4ERR_BAD_HIGH_SLOT:
1501 		case -NFS4ERR_DEADSESSION:
1502 		case -NFS4ERR_SEQ_FALSE_RETRY:
1503 		case -NFS4ERR_SEQ_MISORDERED:
1504 			set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1505 			/* Zero session reset errors */
1506 			break;
1507 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1508 			set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1509 			break;
1510 		case -EKEYEXPIRED:
1511 			/* Nothing we can do */
1512 			nfs4_warn_keyexpired(clp->cl_hostname);
1513 			break;
1514 		default:
1515 			dprintk("%s: failed to handle error %d for server %s\n",
1516 					__func__, error, clp->cl_hostname);
1517 			return error;
1518 	}
1519 	dprintk("%s: handled error %d for server %s\n", __func__, error,
1520 			clp->cl_hostname);
1521 	return 0;
1522 }
1523 
1524 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1525 {
1526 	struct nfs4_state_owner *sp;
1527 	struct nfs_server *server;
1528 	struct rb_node *pos;
1529 	int status = 0;
1530 
1531 restart:
1532 	rcu_read_lock();
1533 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1534 		nfs4_purge_state_owners(server);
1535 		spin_lock(&clp->cl_lock);
1536 		for (pos = rb_first(&server->state_owners);
1537 		     pos != NULL;
1538 		     pos = rb_next(pos)) {
1539 			sp = rb_entry(pos,
1540 				struct nfs4_state_owner, so_server_node);
1541 			if (!test_and_clear_bit(ops->owner_flag_bit,
1542 							&sp->so_flags))
1543 				continue;
1544 			atomic_inc(&sp->so_count);
1545 			spin_unlock(&clp->cl_lock);
1546 			rcu_read_unlock();
1547 
1548 			status = nfs4_reclaim_open_state(sp, ops);
1549 			if (status < 0) {
1550 				set_bit(ops->owner_flag_bit, &sp->so_flags);
1551 				nfs4_put_state_owner(sp);
1552 				return nfs4_recovery_handle_error(clp, status);
1553 			}
1554 
1555 			nfs4_put_state_owner(sp);
1556 			goto restart;
1557 		}
1558 		spin_unlock(&clp->cl_lock);
1559 	}
1560 	rcu_read_unlock();
1561 	return status;
1562 }
1563 
1564 static int nfs4_check_lease(struct nfs_client *clp)
1565 {
1566 	struct rpc_cred *cred;
1567 	const struct nfs4_state_maintenance_ops *ops =
1568 		clp->cl_mvops->state_renewal_ops;
1569 	int status;
1570 
1571 	/* Is the client already known to have an expired lease? */
1572 	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1573 		return 0;
1574 	spin_lock(&clp->cl_lock);
1575 	cred = ops->get_state_renewal_cred_locked(clp);
1576 	spin_unlock(&clp->cl_lock);
1577 	if (cred == NULL) {
1578 		cred = nfs4_get_setclientid_cred(clp);
1579 		status = -ENOKEY;
1580 		if (cred == NULL)
1581 			goto out;
1582 	}
1583 	status = ops->renew_lease(clp, cred);
1584 	put_rpccred(cred);
1585 out:
1586 	return nfs4_recovery_handle_error(clp, status);
1587 }
1588 
1589 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1590  * on EXCHANGE_ID for v4.1
1591  */
1592 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1593 {
1594 	switch (status) {
1595 	case -NFS4ERR_SEQ_MISORDERED:
1596 		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1597 			return -ESERVERFAULT;
1598 		/* Lease confirmation error: retry after purging the lease */
1599 		ssleep(1);
1600 	case -NFS4ERR_CLID_INUSE:
1601 	case -NFS4ERR_STALE_CLIENTID:
1602 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1603 		break;
1604 	case -EACCES:
1605 		if (clp->cl_machine_cred == NULL)
1606 			return -EACCES;
1607 		/* Handle case where the user hasn't set up machine creds */
1608 		nfs4_clear_machine_cred(clp);
1609 	case -NFS4ERR_DELAY:
1610 	case -ETIMEDOUT:
1611 	case -EAGAIN:
1612 		ssleep(1);
1613 		break;
1614 
1615 	case -NFS4ERR_MINOR_VERS_MISMATCH:
1616 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1617 			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1618 		dprintk("%s: exit with error %d for server %s\n",
1619 				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
1620 		return -EPROTONOSUPPORT;
1621 	case -EKEYEXPIRED:
1622 		nfs4_warn_keyexpired(clp->cl_hostname);
1623 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1624 				 * in nfs4_exchange_id */
1625 	default:
1626 		dprintk("%s: exit with error %d for server %s\n", __func__,
1627 				status, clp->cl_hostname);
1628 		return status;
1629 	}
1630 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1631 	dprintk("%s: handled error %d for server %s\n", __func__, status,
1632 			clp->cl_hostname);
1633 	return 0;
1634 }
1635 
1636 static int nfs4_reclaim_lease(struct nfs_client *clp)
1637 {
1638 	struct rpc_cred *cred;
1639 	const struct nfs4_state_recovery_ops *ops =
1640 		clp->cl_mvops->reboot_recovery_ops;
1641 	int status;
1642 
1643 	cred = ops->get_clid_cred(clp);
1644 	if (cred == NULL)
1645 		return -ENOENT;
1646 	status = ops->establish_clid(clp, cred);
1647 	put_rpccred(cred);
1648 	if (status != 0)
1649 		return nfs4_handle_reclaim_lease_error(clp, status);
1650 	return 0;
1651 }
1652 
1653 #ifdef CONFIG_NFS_V4_1
1654 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1655 {
1656 	struct nfs_client *clp = session->clp;
1657 
1658 	switch (err) {
1659 	default:
1660 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1661 		break;
1662 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1663 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1664 	}
1665 	nfs4_schedule_lease_recovery(clp);
1666 }
1667 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1668 
1669 void nfs41_handle_recall_slot(struct nfs_client *clp)
1670 {
1671 	set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1672 	dprintk("%s: scheduling slot recall for server %s\n", __func__,
1673 			clp->cl_hostname);
1674 	nfs4_schedule_state_manager(clp);
1675 }
1676 
1677 static void nfs4_reset_all_state(struct nfs_client *clp)
1678 {
1679 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1680 		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1681 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1682 		nfs4_state_start_reclaim_nograce(clp);
1683 		dprintk("%s: scheduling reset of all state for server %s!\n",
1684 				__func__, clp->cl_hostname);
1685 		nfs4_schedule_state_manager(clp);
1686 	}
1687 }
1688 
1689 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1690 {
1691 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1692 		nfs4_state_start_reclaim_reboot(clp);
1693 		dprintk("%s: server %s rebooted!\n", __func__,
1694 				clp->cl_hostname);
1695 		nfs4_schedule_state_manager(clp);
1696 	}
1697 }
1698 
1699 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1700 {
1701 	nfs4_reset_all_state(clp);
1702 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1703 }
1704 
1705 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1706 {
1707 	/* This will need to handle layouts too */
1708 	nfs_expire_all_delegations(clp);
1709 	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1710 			clp->cl_hostname);
1711 }
1712 
1713 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1714 {
1715 	nfs_expire_all_delegations(clp);
1716 	if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1717 		nfs4_schedule_state_manager(clp);
1718 	dprintk("%s: server %s declared a backchannel fault\n", __func__,
1719 			clp->cl_hostname);
1720 }
1721 
1722 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1723 {
1724 	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1725 		&clp->cl_state) == 0)
1726 		nfs4_schedule_state_manager(clp);
1727 }
1728 
1729 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1730 {
1731 	if (!flags)
1732 		return;
1733 
1734 	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
1735 		__func__, clp->cl_hostname, clp->cl_clientid, flags);
1736 
1737 	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1738 		nfs41_handle_server_reboot(clp);
1739 	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1740 			    SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1741 			    SEQ4_STATUS_ADMIN_STATE_REVOKED |
1742 			    SEQ4_STATUS_LEASE_MOVED))
1743 		nfs41_handle_state_revoked(clp);
1744 	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1745 		nfs41_handle_recallable_state_revoked(clp);
1746 	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
1747 		nfs41_handle_backchannel_fault(clp);
1748 	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1749 				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1750 		nfs41_handle_cb_path_down(clp);
1751 }
1752 
1753 static int nfs4_reset_session(struct nfs_client *clp)
1754 {
1755 	struct rpc_cred *cred;
1756 	int status;
1757 
1758 	nfs4_begin_drain_session(clp);
1759 	cred = nfs4_get_exchange_id_cred(clp);
1760 	status = nfs4_proc_destroy_session(clp->cl_session, cred);
1761 	if (status && status != -NFS4ERR_BADSESSION &&
1762 	    status != -NFS4ERR_DEADSESSION) {
1763 		status = nfs4_recovery_handle_error(clp, status);
1764 		goto out;
1765 	}
1766 
1767 	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1768 	status = nfs4_proc_create_session(clp, cred);
1769 	if (status) {
1770 		dprintk("%s: session reset failed with status %d for server %s!\n",
1771 			__func__, status, clp->cl_hostname);
1772 		status = nfs4_handle_reclaim_lease_error(clp, status);
1773 		goto out;
1774 	}
1775 	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1776 	/* create_session negotiated new slot table */
1777 	clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1778 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1779 	dprintk("%s: session reset was successful for server %s!\n",
1780 			__func__, clp->cl_hostname);
1781 
1782 	 /* Let the state manager reestablish state */
1783 	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1784 		nfs41_setup_state_renewal(clp);
1785 out:
1786 	if (cred)
1787 		put_rpccred(cred);
1788 	return status;
1789 }
1790 
1791 static int nfs4_recall_slot(struct nfs_client *clp)
1792 {
1793 	struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1794 	struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1795 	struct nfs4_slot *new, *old;
1796 	int i;
1797 
1798 	nfs4_begin_drain_session(clp);
1799 	new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1800 		      GFP_NOFS);
1801         if (!new)
1802 		return -ENOMEM;
1803 
1804 	spin_lock(&fc_tbl->slot_tbl_lock);
1805 	for (i = 0; i < fc_tbl->target_max_slots; i++)
1806 		new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1807 	old = fc_tbl->slots;
1808 	fc_tbl->slots = new;
1809 	fc_tbl->max_slots = fc_tbl->target_max_slots;
1810 	fc_tbl->target_max_slots = 0;
1811 	fc_attrs->max_reqs = fc_tbl->max_slots;
1812 	spin_unlock(&fc_tbl->slot_tbl_lock);
1813 
1814 	kfree(old);
1815 	nfs4_end_drain_session(clp);
1816 	return 0;
1817 }
1818 
1819 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1820 {
1821 	struct rpc_cred *cred;
1822 	int ret;
1823 
1824 	nfs4_begin_drain_session(clp);
1825 	cred = nfs4_get_exchange_id_cred(clp);
1826 	ret = nfs4_proc_bind_conn_to_session(clp, cred);
1827 	if (cred)
1828 		put_rpccred(cred);
1829 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1830 	switch (ret) {
1831 	case 0:
1832 		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
1833 			__func__, clp->cl_hostname);
1834 		break;
1835 	case -NFS4ERR_DELAY:
1836 		ssleep(1);
1837 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1838 		break;
1839 	default:
1840 		return nfs4_recovery_handle_error(clp, ret);
1841 	}
1842 	return 0;
1843 }
1844 #else /* CONFIG_NFS_V4_1 */
1845 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1846 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1847 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1848 
1849 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1850 {
1851 	return 0;
1852 }
1853 #endif /* CONFIG_NFS_V4_1 */
1854 
1855 static void nfs4_state_manager(struct nfs_client *clp)
1856 {
1857 	int status = 0;
1858 
1859 	/* Ensure exclusive access to NFSv4 state */
1860 	do {
1861 		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
1862 			status = nfs4_reclaim_lease(clp);
1863 			if (status < 0)
1864 				goto out_error;
1865 			clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1866 			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1867 		}
1868 
1869 		if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1870 			/* We're going to have to re-establish a clientid */
1871 			status = nfs4_reclaim_lease(clp);
1872 			if (status < 0)
1873 				goto out_error;
1874 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1875 				continue;
1876 			clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1877 
1878 			if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1879 					       &clp->cl_state))
1880 				nfs4_state_start_reclaim_nograce(clp);
1881 			else
1882 				set_bit(NFS4CLNT_RECLAIM_REBOOT,
1883 					&clp->cl_state);
1884 
1885 			pnfs_destroy_all_layouts(clp);
1886 		}
1887 
1888 		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1889 			status = nfs4_check_lease(clp);
1890 			if (status < 0)
1891 				goto out_error;
1892 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1893 				continue;
1894 		}
1895 
1896 		/* Initialize or reset the session */
1897 		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1898 		   && nfs4_has_session(clp)) {
1899 			status = nfs4_reset_session(clp);
1900 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1901 				continue;
1902 			if (status < 0)
1903 				goto out_error;
1904 		}
1905 
1906 		/* Send BIND_CONN_TO_SESSION */
1907 		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1908 				&clp->cl_state) && nfs4_has_session(clp)) {
1909 			status = nfs4_bind_conn_to_session(clp);
1910 			if (status < 0)
1911 				goto out_error;
1912 			continue;
1913 		}
1914 
1915 		/* First recover reboot state... */
1916 		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1917 			status = nfs4_do_reclaim(clp,
1918 				clp->cl_mvops->reboot_recovery_ops);
1919 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1920 			    test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1921 				continue;
1922 			nfs4_state_end_reclaim_reboot(clp);
1923 			if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1924 				continue;
1925 			if (status < 0)
1926 				goto out_error;
1927 		}
1928 
1929 		/* Now recover expired state... */
1930 		if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1931 			status = nfs4_do_reclaim(clp,
1932 				clp->cl_mvops->nograce_recovery_ops);
1933 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1934 			    test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1935 			    test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1936 				continue;
1937 			if (status < 0)
1938 				goto out_error;
1939 		}
1940 
1941 		nfs4_end_drain_session(clp);
1942 		if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1943 			nfs_client_return_marked_delegations(clp);
1944 			continue;
1945 		}
1946 		/* Recall session slots */
1947 		if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1948 		   && nfs4_has_session(clp)) {
1949 			status = nfs4_recall_slot(clp);
1950 			if (status < 0)
1951 				goto out_error;
1952 			continue;
1953 		}
1954 
1955 
1956 		nfs4_clear_state_manager_bit(clp);
1957 		/* Did we race with an attempt to give us more work? */
1958 		if (clp->cl_state == 0)
1959 			break;
1960 		if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1961 			break;
1962 	} while (atomic_read(&clp->cl_count) > 1);
1963 	return;
1964 out_error:
1965 	pr_warn_ratelimited("NFS: state manager failed on NFSv4 server %s"
1966 			" with error %d\n", clp->cl_hostname, -status);
1967 	nfs4_end_drain_session(clp);
1968 	nfs4_clear_state_manager_bit(clp);
1969 }
1970 
1971 static int nfs4_run_state_manager(void *ptr)
1972 {
1973 	struct nfs_client *clp = ptr;
1974 
1975 	allow_signal(SIGKILL);
1976 	nfs4_state_manager(clp);
1977 	nfs_put_client(clp);
1978 	module_put_and_exit(0);
1979 	return 0;
1980 }
1981 
1982 /*
1983  * Local variables:
1984  *  c-basic-offset: 8
1985  * End:
1986  */
1987