xref: /linux/fs/nfs/nfs4state.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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 	switch (status) {
140 	case -NFS4ERR_STALE_CLIENTID:
141 		set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
142 	case 0:
143 		/* Sustain the lease, even if it's empty.  If the clientid4
144 		 * goes stale it's of no use for trunking discovery. */
145 		nfs4_schedule_state_renewal(*result);
146 		break;
147 	}
148 
149 out:
150 	return status;
151 }
152 
153 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
154 {
155 	struct rpc_cred *cred = NULL;
156 
157 	if (clp->cl_machine_cred != NULL)
158 		cred = get_rpccred(clp->cl_machine_cred);
159 	return cred;
160 }
161 
162 static void nfs4_clear_machine_cred(struct nfs_client *clp)
163 {
164 	struct rpc_cred *cred;
165 
166 	spin_lock(&clp->cl_lock);
167 	cred = clp->cl_machine_cred;
168 	clp->cl_machine_cred = NULL;
169 	spin_unlock(&clp->cl_lock);
170 	if (cred != NULL)
171 		put_rpccred(cred);
172 }
173 
174 static struct rpc_cred *
175 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
176 {
177 	struct rpc_cred *cred = NULL;
178 	struct nfs4_state_owner *sp;
179 	struct rb_node *pos;
180 
181 	for (pos = rb_first(&server->state_owners);
182 	     pos != NULL;
183 	     pos = rb_next(pos)) {
184 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
185 		if (list_empty(&sp->so_states))
186 			continue;
187 		cred = get_rpccred(sp->so_cred);
188 		break;
189 	}
190 	return cred;
191 }
192 
193 /**
194  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
195  * @clp: client state handle
196  *
197  * Returns an rpc_cred with reference count bumped, or NULL.
198  * Caller must hold clp->cl_lock.
199  */
200 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
201 {
202 	struct rpc_cred *cred = NULL;
203 	struct nfs_server *server;
204 
205 	/* Use machine credentials if available */
206 	cred = nfs4_get_machine_cred_locked(clp);
207 	if (cred != NULL)
208 		goto out;
209 
210 	rcu_read_lock();
211 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
212 		cred = nfs4_get_renew_cred_server_locked(server);
213 		if (cred != NULL)
214 			break;
215 	}
216 	rcu_read_unlock();
217 
218 out:
219 	return cred;
220 }
221 
222 #if defined(CONFIG_NFS_V4_1)
223 
224 static int nfs41_setup_state_renewal(struct nfs_client *clp)
225 {
226 	int status;
227 	struct nfs_fsinfo fsinfo;
228 
229 	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
230 		nfs4_schedule_state_renewal(clp);
231 		return 0;
232 	}
233 
234 	status = nfs4_proc_get_lease_time(clp, &fsinfo);
235 	if (status == 0) {
236 		/* Update lease time and schedule renewal */
237 		spin_lock(&clp->cl_lock);
238 		clp->cl_lease_time = fsinfo.lease_time * HZ;
239 		clp->cl_last_renewal = jiffies;
240 		spin_unlock(&clp->cl_lock);
241 
242 		nfs4_schedule_state_renewal(clp);
243 	}
244 
245 	return status;
246 }
247 
248 /*
249  * Back channel returns NFS4ERR_DELAY for new requests when
250  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
251  * is ended.
252  */
253 static void nfs4_end_drain_session(struct nfs_client *clp)
254 {
255 	struct nfs4_session *ses = clp->cl_session;
256 	struct nfs4_slot_table *tbl;
257 
258 	if (ses == NULL)
259 		return;
260 	tbl = &ses->fc_slot_table;
261 	if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
262 		spin_lock(&tbl->slot_tbl_lock);
263 		nfs41_wake_slot_table(tbl);
264 		spin_unlock(&tbl->slot_tbl_lock);
265 	}
266 }
267 
268 /*
269  * Signal state manager thread if session fore channel is drained
270  */
271 void nfs4_session_drain_complete(struct nfs4_session *session,
272 		struct nfs4_slot_table *tbl)
273 {
274 	if (nfs4_session_draining(session))
275 		complete(&tbl->complete);
276 }
277 
278 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
279 {
280 	spin_lock(&tbl->slot_tbl_lock);
281 	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
282 		INIT_COMPLETION(tbl->complete);
283 		spin_unlock(&tbl->slot_tbl_lock);
284 		return wait_for_completion_interruptible(&tbl->complete);
285 	}
286 	spin_unlock(&tbl->slot_tbl_lock);
287 	return 0;
288 }
289 
290 static int nfs4_begin_drain_session(struct nfs_client *clp)
291 {
292 	struct nfs4_session *ses = clp->cl_session;
293 	int ret = 0;
294 
295 	set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
296 	/* back channel */
297 	ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
298 	if (ret)
299 		return ret;
300 	/* fore channel */
301 	return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
302 }
303 
304 static void nfs41_finish_session_reset(struct nfs_client *clp)
305 {
306 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
307 	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
308 	/* create_session negotiated new slot table */
309 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
310 	nfs41_setup_state_renewal(clp);
311 }
312 
313 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
314 {
315 	int status;
316 
317 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
318 		goto do_confirm;
319 	nfs4_begin_drain_session(clp);
320 	status = nfs4_proc_exchange_id(clp, cred);
321 	if (status != 0)
322 		goto out;
323 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
324 do_confirm:
325 	status = nfs4_proc_create_session(clp, cred);
326 	if (status != 0)
327 		goto out;
328 	nfs41_finish_session_reset(clp);
329 	nfs_mark_client_ready(clp, NFS_CS_READY);
330 out:
331 	return status;
332 }
333 
334 /**
335  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
336  *
337  * @clp: nfs_client under test
338  * @result: OUT: found nfs_client, or clp
339  * @cred: credential to use for trunking test
340  *
341  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
342  * If NFS4_OK is returned, an nfs_client pointer is planted in
343  * "result".
344  *
345  * Note: The returned client may not yet be marked ready.
346  */
347 int nfs41_discover_server_trunking(struct nfs_client *clp,
348 				   struct nfs_client **result,
349 				   struct rpc_cred *cred)
350 {
351 	int status;
352 
353 	status = nfs4_proc_exchange_id(clp, cred);
354 	if (status != NFS4_OK)
355 		return status;
356 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
357 
358 	return nfs41_walk_client_list(clp, result, cred);
359 }
360 
361 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
362 {
363 	struct rpc_cred *cred;
364 
365 	spin_lock(&clp->cl_lock);
366 	cred = nfs4_get_machine_cred_locked(clp);
367 	spin_unlock(&clp->cl_lock);
368 	return cred;
369 }
370 
371 #endif /* CONFIG_NFS_V4_1 */
372 
373 static struct rpc_cred *
374 nfs4_get_setclientid_cred_server(struct nfs_server *server)
375 {
376 	struct nfs_client *clp = server->nfs_client;
377 	struct rpc_cred *cred = NULL;
378 	struct nfs4_state_owner *sp;
379 	struct rb_node *pos;
380 
381 	spin_lock(&clp->cl_lock);
382 	pos = rb_first(&server->state_owners);
383 	if (pos != NULL) {
384 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
385 		cred = get_rpccred(sp->so_cred);
386 	}
387 	spin_unlock(&clp->cl_lock);
388 	return cred;
389 }
390 
391 /**
392  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
393  * @clp: client state handle
394  *
395  * Returns an rpc_cred with reference count bumped, or NULL.
396  */
397 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
398 {
399 	struct nfs_server *server;
400 	struct rpc_cred *cred;
401 
402 	spin_lock(&clp->cl_lock);
403 	cred = nfs4_get_machine_cred_locked(clp);
404 	spin_unlock(&clp->cl_lock);
405 	if (cred != NULL)
406 		goto out;
407 
408 	rcu_read_lock();
409 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
410 		cred = nfs4_get_setclientid_cred_server(server);
411 		if (cred != NULL)
412 			break;
413 	}
414 	rcu_read_unlock();
415 
416 out:
417 	return cred;
418 }
419 
420 static struct nfs4_state_owner *
421 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
422 {
423 	struct rb_node **p = &server->state_owners.rb_node,
424 		       *parent = NULL;
425 	struct nfs4_state_owner *sp;
426 
427 	while (*p != NULL) {
428 		parent = *p;
429 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
430 
431 		if (cred < sp->so_cred)
432 			p = &parent->rb_left;
433 		else if (cred > sp->so_cred)
434 			p = &parent->rb_right;
435 		else {
436 			if (!list_empty(&sp->so_lru))
437 				list_del_init(&sp->so_lru);
438 			atomic_inc(&sp->so_count);
439 			return sp;
440 		}
441 	}
442 	return NULL;
443 }
444 
445 static struct nfs4_state_owner *
446 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
447 {
448 	struct nfs_server *server = new->so_server;
449 	struct rb_node **p = &server->state_owners.rb_node,
450 		       *parent = NULL;
451 	struct nfs4_state_owner *sp;
452 	int err;
453 
454 	while (*p != NULL) {
455 		parent = *p;
456 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
457 
458 		if (new->so_cred < sp->so_cred)
459 			p = &parent->rb_left;
460 		else if (new->so_cred > sp->so_cred)
461 			p = &parent->rb_right;
462 		else {
463 			if (!list_empty(&sp->so_lru))
464 				list_del_init(&sp->so_lru);
465 			atomic_inc(&sp->so_count);
466 			return sp;
467 		}
468 	}
469 	err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
470 	if (err)
471 		return ERR_PTR(err);
472 	rb_link_node(&new->so_server_node, parent, p);
473 	rb_insert_color(&new->so_server_node, &server->state_owners);
474 	return new;
475 }
476 
477 static void
478 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
479 {
480 	struct nfs_server *server = sp->so_server;
481 
482 	if (!RB_EMPTY_NODE(&sp->so_server_node))
483 		rb_erase(&sp->so_server_node, &server->state_owners);
484 	ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
485 }
486 
487 static void
488 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
489 {
490 	sc->create_time = ktime_get();
491 	sc->flags = 0;
492 	sc->counter = 0;
493 	spin_lock_init(&sc->lock);
494 	INIT_LIST_HEAD(&sc->list);
495 	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
496 }
497 
498 static void
499 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
500 {
501 	rpc_destroy_wait_queue(&sc->wait);
502 }
503 
504 /*
505  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
506  * create a new state_owner.
507  *
508  */
509 static struct nfs4_state_owner *
510 nfs4_alloc_state_owner(struct nfs_server *server,
511 		struct rpc_cred *cred,
512 		gfp_t gfp_flags)
513 {
514 	struct nfs4_state_owner *sp;
515 
516 	sp = kzalloc(sizeof(*sp), gfp_flags);
517 	if (!sp)
518 		return NULL;
519 	sp->so_server = server;
520 	sp->so_cred = get_rpccred(cred);
521 	spin_lock_init(&sp->so_lock);
522 	INIT_LIST_HEAD(&sp->so_states);
523 	nfs4_init_seqid_counter(&sp->so_seqid);
524 	atomic_set(&sp->so_count, 1);
525 	INIT_LIST_HEAD(&sp->so_lru);
526 	return sp;
527 }
528 
529 static void
530 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
531 {
532 	struct rb_node *rb_node = &sp->so_server_node;
533 
534 	if (!RB_EMPTY_NODE(rb_node)) {
535 		struct nfs_server *server = sp->so_server;
536 		struct nfs_client *clp = server->nfs_client;
537 
538 		spin_lock(&clp->cl_lock);
539 		if (!RB_EMPTY_NODE(rb_node)) {
540 			rb_erase(rb_node, &server->state_owners);
541 			RB_CLEAR_NODE(rb_node);
542 		}
543 		spin_unlock(&clp->cl_lock);
544 	}
545 }
546 
547 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
548 {
549 	nfs4_destroy_seqid_counter(&sp->so_seqid);
550 	put_rpccred(sp->so_cred);
551 	kfree(sp);
552 }
553 
554 static void nfs4_gc_state_owners(struct nfs_server *server)
555 {
556 	struct nfs_client *clp = server->nfs_client;
557 	struct nfs4_state_owner *sp, *tmp;
558 	unsigned long time_min, time_max;
559 	LIST_HEAD(doomed);
560 
561 	spin_lock(&clp->cl_lock);
562 	time_max = jiffies;
563 	time_min = (long)time_max - (long)clp->cl_lease_time;
564 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
565 		/* NB: LRU is sorted so that oldest is at the head */
566 		if (time_in_range(sp->so_expires, time_min, time_max))
567 			break;
568 		list_move(&sp->so_lru, &doomed);
569 		nfs4_remove_state_owner_locked(sp);
570 	}
571 	spin_unlock(&clp->cl_lock);
572 
573 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
574 		list_del(&sp->so_lru);
575 		nfs4_free_state_owner(sp);
576 	}
577 }
578 
579 /**
580  * nfs4_get_state_owner - Look up a state owner given a credential
581  * @server: nfs_server to search
582  * @cred: RPC credential to match
583  *
584  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
585  */
586 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
587 					      struct rpc_cred *cred,
588 					      gfp_t gfp_flags)
589 {
590 	struct nfs_client *clp = server->nfs_client;
591 	struct nfs4_state_owner *sp, *new;
592 
593 	spin_lock(&clp->cl_lock);
594 	sp = nfs4_find_state_owner_locked(server, cred);
595 	spin_unlock(&clp->cl_lock);
596 	if (sp != NULL)
597 		goto out;
598 	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
599 	if (new == NULL)
600 		goto out;
601 	do {
602 		if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
603 			break;
604 		spin_lock(&clp->cl_lock);
605 		sp = nfs4_insert_state_owner_locked(new);
606 		spin_unlock(&clp->cl_lock);
607 	} while (sp == ERR_PTR(-EAGAIN));
608 	if (sp != new)
609 		nfs4_free_state_owner(new);
610 out:
611 	nfs4_gc_state_owners(server);
612 	return sp;
613 }
614 
615 /**
616  * nfs4_put_state_owner - Release a nfs4_state_owner
617  * @sp: state owner data to release
618  *
619  * Note that we keep released state owners on an LRU
620  * list.
621  * This caches valid state owners so that they can be
622  * reused, to avoid the OPEN_CONFIRM on minor version 0.
623  * It also pins the uniquifier of dropped state owners for
624  * a while, to ensure that those state owner names are
625  * never reused.
626  */
627 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
628 {
629 	struct nfs_server *server = sp->so_server;
630 	struct nfs_client *clp = server->nfs_client;
631 
632 	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
633 		return;
634 
635 	sp->so_expires = jiffies;
636 	list_add_tail(&sp->so_lru, &server->state_owners_lru);
637 	spin_unlock(&clp->cl_lock);
638 }
639 
640 /**
641  * nfs4_purge_state_owners - Release all cached state owners
642  * @server: nfs_server with cached state owners to release
643  *
644  * Called at umount time.  Remaining state owners will be on
645  * the LRU with ref count of zero.
646  */
647 void nfs4_purge_state_owners(struct nfs_server *server)
648 {
649 	struct nfs_client *clp = server->nfs_client;
650 	struct nfs4_state_owner *sp, *tmp;
651 	LIST_HEAD(doomed);
652 
653 	spin_lock(&clp->cl_lock);
654 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
655 		list_move(&sp->so_lru, &doomed);
656 		nfs4_remove_state_owner_locked(sp);
657 	}
658 	spin_unlock(&clp->cl_lock);
659 
660 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
661 		list_del(&sp->so_lru);
662 		nfs4_free_state_owner(sp);
663 	}
664 }
665 
666 static struct nfs4_state *
667 nfs4_alloc_open_state(void)
668 {
669 	struct nfs4_state *state;
670 
671 	state = kzalloc(sizeof(*state), GFP_NOFS);
672 	if (!state)
673 		return NULL;
674 	atomic_set(&state->count, 1);
675 	INIT_LIST_HEAD(&state->lock_states);
676 	spin_lock_init(&state->state_lock);
677 	seqlock_init(&state->seqlock);
678 	return state;
679 }
680 
681 void
682 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
683 {
684 	if (state->state == fmode)
685 		return;
686 	/* NB! List reordering - see the reclaim code for why.  */
687 	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
688 		if (fmode & FMODE_WRITE)
689 			list_move(&state->open_states, &state->owner->so_states);
690 		else
691 			list_move_tail(&state->open_states, &state->owner->so_states);
692 	}
693 	state->state = fmode;
694 }
695 
696 static struct nfs4_state *
697 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
698 {
699 	struct nfs_inode *nfsi = NFS_I(inode);
700 	struct nfs4_state *state;
701 
702 	list_for_each_entry(state, &nfsi->open_states, inode_states) {
703 		if (state->owner != owner)
704 			continue;
705 		if (atomic_inc_not_zero(&state->count))
706 			return state;
707 	}
708 	return NULL;
709 }
710 
711 static void
712 nfs4_free_open_state(struct nfs4_state *state)
713 {
714 	kfree(state);
715 }
716 
717 struct nfs4_state *
718 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
719 {
720 	struct nfs4_state *state, *new;
721 	struct nfs_inode *nfsi = NFS_I(inode);
722 
723 	spin_lock(&inode->i_lock);
724 	state = __nfs4_find_state_byowner(inode, owner);
725 	spin_unlock(&inode->i_lock);
726 	if (state)
727 		goto out;
728 	new = nfs4_alloc_open_state();
729 	spin_lock(&owner->so_lock);
730 	spin_lock(&inode->i_lock);
731 	state = __nfs4_find_state_byowner(inode, owner);
732 	if (state == NULL && new != NULL) {
733 		state = new;
734 		state->owner = owner;
735 		atomic_inc(&owner->so_count);
736 		list_add(&state->inode_states, &nfsi->open_states);
737 		ihold(inode);
738 		state->inode = inode;
739 		spin_unlock(&inode->i_lock);
740 		/* Note: The reclaim code dictates that we add stateless
741 		 * and read-only stateids to the end of the list */
742 		list_add_tail(&state->open_states, &owner->so_states);
743 		spin_unlock(&owner->so_lock);
744 	} else {
745 		spin_unlock(&inode->i_lock);
746 		spin_unlock(&owner->so_lock);
747 		if (new)
748 			nfs4_free_open_state(new);
749 	}
750 out:
751 	return state;
752 }
753 
754 void nfs4_put_open_state(struct nfs4_state *state)
755 {
756 	struct inode *inode = state->inode;
757 	struct nfs4_state_owner *owner = state->owner;
758 
759 	if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
760 		return;
761 	spin_lock(&inode->i_lock);
762 	list_del(&state->inode_states);
763 	list_del(&state->open_states);
764 	spin_unlock(&inode->i_lock);
765 	spin_unlock(&owner->so_lock);
766 	iput(inode);
767 	nfs4_free_open_state(state);
768 	nfs4_put_state_owner(owner);
769 }
770 
771 /*
772  * Close the current file.
773  */
774 static void __nfs4_close(struct nfs4_state *state,
775 		fmode_t fmode, gfp_t gfp_mask, int wait)
776 {
777 	struct nfs4_state_owner *owner = state->owner;
778 	int call_close = 0;
779 	fmode_t newstate;
780 
781 	atomic_inc(&owner->so_count);
782 	/* Protect against nfs4_find_state() */
783 	spin_lock(&owner->so_lock);
784 	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
785 		case FMODE_READ:
786 			state->n_rdonly--;
787 			break;
788 		case FMODE_WRITE:
789 			state->n_wronly--;
790 			break;
791 		case FMODE_READ|FMODE_WRITE:
792 			state->n_rdwr--;
793 	}
794 	newstate = FMODE_READ|FMODE_WRITE;
795 	if (state->n_rdwr == 0) {
796 		if (state->n_rdonly == 0) {
797 			newstate &= ~FMODE_READ;
798 			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
799 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
800 		}
801 		if (state->n_wronly == 0) {
802 			newstate &= ~FMODE_WRITE;
803 			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
804 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
805 		}
806 		if (newstate == 0)
807 			clear_bit(NFS_DELEGATED_STATE, &state->flags);
808 	}
809 	nfs4_state_set_mode_locked(state, newstate);
810 	spin_unlock(&owner->so_lock);
811 
812 	if (!call_close) {
813 		nfs4_put_open_state(state);
814 		nfs4_put_state_owner(owner);
815 	} else
816 		nfs4_do_close(state, gfp_mask, wait);
817 }
818 
819 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
820 {
821 	__nfs4_close(state, fmode, GFP_NOFS, 0);
822 }
823 
824 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
825 {
826 	__nfs4_close(state, fmode, GFP_KERNEL, 1);
827 }
828 
829 /*
830  * Search the state->lock_states for an existing lock_owner
831  * that is compatible with current->files
832  */
833 static struct nfs4_lock_state *
834 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
835 {
836 	struct nfs4_lock_state *pos;
837 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
838 		if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
839 			continue;
840 		switch (pos->ls_owner.lo_type) {
841 		case NFS4_POSIX_LOCK_TYPE:
842 			if (pos->ls_owner.lo_u.posix_owner != fl_owner)
843 				continue;
844 			break;
845 		case NFS4_FLOCK_LOCK_TYPE:
846 			if (pos->ls_owner.lo_u.flock_owner != fl_pid)
847 				continue;
848 		}
849 		atomic_inc(&pos->ls_count);
850 		return pos;
851 	}
852 	return NULL;
853 }
854 
855 /*
856  * Return a compatible lock_state. If no initialized lock_state structure
857  * exists, return an uninitialized one.
858  *
859  */
860 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)
861 {
862 	struct nfs4_lock_state *lsp;
863 	struct nfs_server *server = state->owner->so_server;
864 
865 	lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
866 	if (lsp == NULL)
867 		return NULL;
868 	nfs4_init_seqid_counter(&lsp->ls_seqid);
869 	atomic_set(&lsp->ls_count, 1);
870 	lsp->ls_state = state;
871 	lsp->ls_owner.lo_type = type;
872 	switch (lsp->ls_owner.lo_type) {
873 	case NFS4_FLOCK_LOCK_TYPE:
874 		lsp->ls_owner.lo_u.flock_owner = fl_pid;
875 		break;
876 	case NFS4_POSIX_LOCK_TYPE:
877 		lsp->ls_owner.lo_u.posix_owner = fl_owner;
878 		break;
879 	default:
880 		goto out_free;
881 	}
882 	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
883 	if (lsp->ls_seqid.owner_id < 0)
884 		goto out_free;
885 	INIT_LIST_HEAD(&lsp->ls_locks);
886 	return lsp;
887 out_free:
888 	kfree(lsp);
889 	return NULL;
890 }
891 
892 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
893 {
894 	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
895 	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
896 	kfree(lsp);
897 }
898 
899 /*
900  * Return a compatible lock_state. If no initialized lock_state structure
901  * exists, return an uninitialized one.
902  *
903  */
904 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
905 {
906 	struct nfs4_lock_state *lsp, *new = NULL;
907 
908 	for(;;) {
909 		spin_lock(&state->state_lock);
910 		lsp = __nfs4_find_lock_state(state, owner, pid, type);
911 		if (lsp != NULL)
912 			break;
913 		if (new != NULL) {
914 			list_add(&new->ls_locks, &state->lock_states);
915 			set_bit(LK_STATE_IN_USE, &state->flags);
916 			lsp = new;
917 			new = NULL;
918 			break;
919 		}
920 		spin_unlock(&state->state_lock);
921 		new = nfs4_alloc_lock_state(state, owner, pid, type);
922 		if (new == NULL)
923 			return NULL;
924 	}
925 	spin_unlock(&state->state_lock);
926 	if (new != NULL)
927 		nfs4_free_lock_state(state->owner->so_server, new);
928 	return lsp;
929 }
930 
931 /*
932  * Release reference to lock_state, and free it if we see that
933  * it is no longer in use
934  */
935 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
936 {
937 	struct nfs4_state *state;
938 
939 	if (lsp == NULL)
940 		return;
941 	state = lsp->ls_state;
942 	if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
943 		return;
944 	list_del(&lsp->ls_locks);
945 	if (list_empty(&state->lock_states))
946 		clear_bit(LK_STATE_IN_USE, &state->flags);
947 	spin_unlock(&state->state_lock);
948 	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
949 		if (nfs4_release_lockowner(lsp) == 0)
950 			return;
951 	}
952 	nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
953 }
954 
955 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
956 {
957 	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
958 
959 	dst->fl_u.nfs4_fl.owner = lsp;
960 	atomic_inc(&lsp->ls_count);
961 }
962 
963 static void nfs4_fl_release_lock(struct file_lock *fl)
964 {
965 	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
966 }
967 
968 static const struct file_lock_operations nfs4_fl_lock_ops = {
969 	.fl_copy_lock = nfs4_fl_copy_lock,
970 	.fl_release_private = nfs4_fl_release_lock,
971 };
972 
973 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
974 {
975 	struct nfs4_lock_state *lsp;
976 
977 	if (fl->fl_ops != NULL)
978 		return 0;
979 	if (fl->fl_flags & FL_POSIX)
980 		lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
981 	else if (fl->fl_flags & FL_FLOCK)
982 		lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
983 				NFS4_FLOCK_LOCK_TYPE);
984 	else
985 		return -EINVAL;
986 	if (lsp == NULL)
987 		return -ENOMEM;
988 	fl->fl_u.nfs4_fl.owner = lsp;
989 	fl->fl_ops = &nfs4_fl_lock_ops;
990 	return 0;
991 }
992 
993 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
994 		const struct nfs_lockowner *lockowner)
995 {
996 	struct nfs4_lock_state *lsp;
997 	fl_owner_t fl_owner;
998 	pid_t fl_pid;
999 	bool ret = false;
1000 
1001 
1002 	if (lockowner == NULL)
1003 		goto out;
1004 
1005 	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1006 		goto out;
1007 
1008 	fl_owner = lockowner->l_owner;
1009 	fl_pid = lockowner->l_pid;
1010 	spin_lock(&state->state_lock);
1011 	lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1012 	if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1013 		nfs4_stateid_copy(dst, &lsp->ls_stateid);
1014 		ret = true;
1015 	}
1016 	spin_unlock(&state->state_lock);
1017 	nfs4_put_lock_state(lsp);
1018 out:
1019 	return ret;
1020 }
1021 
1022 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1023 {
1024 	int seq;
1025 
1026 	do {
1027 		seq = read_seqbegin(&state->seqlock);
1028 		nfs4_stateid_copy(dst, &state->stateid);
1029 	} while (read_seqretry(&state->seqlock, seq));
1030 }
1031 
1032 /*
1033  * Byte-range lock aware utility to initialize the stateid of read/write
1034  * requests.
1035  */
1036 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1037 		fmode_t fmode, const struct nfs_lockowner *lockowner)
1038 {
1039 	if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1040 		return;
1041 	if (nfs4_copy_lock_stateid(dst, state, lockowner))
1042 		return;
1043 	nfs4_copy_open_stateid(dst, state);
1044 }
1045 
1046 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1047 {
1048 	struct nfs_seqid *new;
1049 
1050 	new = kmalloc(sizeof(*new), gfp_mask);
1051 	if (new != NULL) {
1052 		new->sequence = counter;
1053 		INIT_LIST_HEAD(&new->list);
1054 		new->task = NULL;
1055 	}
1056 	return new;
1057 }
1058 
1059 void nfs_release_seqid(struct nfs_seqid *seqid)
1060 {
1061 	struct nfs_seqid_counter *sequence;
1062 
1063 	if (list_empty(&seqid->list))
1064 		return;
1065 	sequence = seqid->sequence;
1066 	spin_lock(&sequence->lock);
1067 	list_del_init(&seqid->list);
1068 	if (!list_empty(&sequence->list)) {
1069 		struct nfs_seqid *next;
1070 
1071 		next = list_first_entry(&sequence->list,
1072 				struct nfs_seqid, list);
1073 		rpc_wake_up_queued_task(&sequence->wait, next->task);
1074 	}
1075 	spin_unlock(&sequence->lock);
1076 }
1077 
1078 void nfs_free_seqid(struct nfs_seqid *seqid)
1079 {
1080 	nfs_release_seqid(seqid);
1081 	kfree(seqid);
1082 }
1083 
1084 /*
1085  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1086  * failed with a seqid incrementing error -
1087  * see comments nfs_fs.h:seqid_mutating_error()
1088  */
1089 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1090 {
1091 	switch (status) {
1092 		case 0:
1093 			break;
1094 		case -NFS4ERR_BAD_SEQID:
1095 			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1096 				return;
1097 			pr_warn_ratelimited("NFS: v4 server returned a bad"
1098 					" sequence-id error on an"
1099 					" unconfirmed sequence %p!\n",
1100 					seqid->sequence);
1101 		case -NFS4ERR_STALE_CLIENTID:
1102 		case -NFS4ERR_STALE_STATEID:
1103 		case -NFS4ERR_BAD_STATEID:
1104 		case -NFS4ERR_BADXDR:
1105 		case -NFS4ERR_RESOURCE:
1106 		case -NFS4ERR_NOFILEHANDLE:
1107 			/* Non-seqid mutating errors */
1108 			return;
1109 	};
1110 	/*
1111 	 * Note: no locking needed as we are guaranteed to be first
1112 	 * on the sequence list
1113 	 */
1114 	seqid->sequence->counter++;
1115 }
1116 
1117 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1118 {
1119 	struct nfs4_state_owner *sp = container_of(seqid->sequence,
1120 					struct nfs4_state_owner, so_seqid);
1121 	struct nfs_server *server = sp->so_server;
1122 
1123 	if (status == -NFS4ERR_BAD_SEQID)
1124 		nfs4_drop_state_owner(sp);
1125 	if (!nfs4_has_session(server->nfs_client))
1126 		nfs_increment_seqid(status, seqid);
1127 }
1128 
1129 /*
1130  * Increment the seqid if the LOCK/LOCKU succeeded, or
1131  * failed with a seqid incrementing error -
1132  * see comments nfs_fs.h:seqid_mutating_error()
1133  */
1134 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1135 {
1136 	nfs_increment_seqid(status, seqid);
1137 }
1138 
1139 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1140 {
1141 	struct nfs_seqid_counter *sequence = seqid->sequence;
1142 	int status = 0;
1143 
1144 	spin_lock(&sequence->lock);
1145 	seqid->task = task;
1146 	if (list_empty(&seqid->list))
1147 		list_add_tail(&seqid->list, &sequence->list);
1148 	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1149 		goto unlock;
1150 	rpc_sleep_on(&sequence->wait, task, NULL);
1151 	status = -EAGAIN;
1152 unlock:
1153 	spin_unlock(&sequence->lock);
1154 	return status;
1155 }
1156 
1157 static int nfs4_run_state_manager(void *);
1158 
1159 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1160 {
1161 	smp_mb__before_clear_bit();
1162 	clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1163 	smp_mb__after_clear_bit();
1164 	wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1165 	rpc_wake_up(&clp->cl_rpcwaitq);
1166 }
1167 
1168 /*
1169  * Schedule the nfs_client asynchronous state management routine
1170  */
1171 void nfs4_schedule_state_manager(struct nfs_client *clp)
1172 {
1173 	struct task_struct *task;
1174 	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1175 
1176 	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1177 		return;
1178 	__module_get(THIS_MODULE);
1179 	atomic_inc(&clp->cl_count);
1180 
1181 	/* The rcu_read_lock() is not strictly necessary, as the state
1182 	 * manager is the only thread that ever changes the rpc_xprt
1183 	 * after it's initialized.  At this point, we're single threaded. */
1184 	rcu_read_lock();
1185 	snprintf(buf, sizeof(buf), "%s-manager",
1186 			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1187 	rcu_read_unlock();
1188 	task = kthread_run(nfs4_run_state_manager, clp, buf);
1189 	if (IS_ERR(task)) {
1190 		printk(KERN_ERR "%s: kthread_run: %ld\n",
1191 			__func__, PTR_ERR(task));
1192 		nfs4_clear_state_manager_bit(clp);
1193 		nfs_put_client(clp);
1194 		module_put(THIS_MODULE);
1195 	}
1196 }
1197 
1198 /*
1199  * Schedule a lease recovery attempt
1200  */
1201 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1202 {
1203 	if (!clp)
1204 		return;
1205 	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1206 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1207 	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1208 			clp->cl_hostname);
1209 	nfs4_schedule_state_manager(clp);
1210 }
1211 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1212 
1213 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1214 {
1215 	int res;
1216 
1217 	might_sleep();
1218 
1219 	res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1220 			nfs_wait_bit_killable, TASK_KILLABLE);
1221 	if (res)
1222 		return res;
1223 
1224 	if (clp->cl_cons_state < 0)
1225 		return clp->cl_cons_state;
1226 	return 0;
1227 }
1228 
1229 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1230 {
1231 	unsigned int loop;
1232 	int ret;
1233 
1234 	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1235 		ret = nfs4_wait_clnt_recover(clp);
1236 		if (ret != 0)
1237 			break;
1238 		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1239 		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1240 			break;
1241 		nfs4_schedule_state_manager(clp);
1242 		ret = -EIO;
1243 	}
1244 	return ret;
1245 }
1246 
1247 /*
1248  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1249  * @clp: client to process
1250  *
1251  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1252  * resend of the SETCLIENTID and hence re-establish the
1253  * callback channel. Then return all existing delegations.
1254  */
1255 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1256 {
1257 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1258 	nfs_expire_all_delegations(clp);
1259 	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1260 			clp->cl_hostname);
1261 }
1262 
1263 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1264 {
1265 	nfs40_handle_cb_pathdown(clp);
1266 	nfs4_schedule_state_manager(clp);
1267 }
1268 
1269 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1270 {
1271 
1272 	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1273 	/* Don't recover state that expired before the reboot */
1274 	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1275 		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1276 		return 0;
1277 	}
1278 	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1279 	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1280 	return 1;
1281 }
1282 
1283 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1284 {
1285 	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1286 	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1287 	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1288 	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1289 	return 1;
1290 }
1291 
1292 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1293 {
1294 	struct nfs_client *clp = server->nfs_client;
1295 
1296 	nfs4_state_mark_reclaim_nograce(clp, state);
1297 	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1298 			clp->cl_hostname);
1299 	nfs4_schedule_state_manager(clp);
1300 }
1301 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1302 
1303 void nfs_inode_find_state_and_recover(struct inode *inode,
1304 		const nfs4_stateid *stateid)
1305 {
1306 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1307 	struct nfs_inode *nfsi = NFS_I(inode);
1308 	struct nfs_open_context *ctx;
1309 	struct nfs4_state *state;
1310 	bool found = false;
1311 
1312 	spin_lock(&inode->i_lock);
1313 	list_for_each_entry(ctx, &nfsi->open_files, list) {
1314 		state = ctx->state;
1315 		if (state == NULL)
1316 			continue;
1317 		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1318 			continue;
1319 		if (!nfs4_stateid_match(&state->stateid, stateid))
1320 			continue;
1321 		nfs4_state_mark_reclaim_nograce(clp, state);
1322 		found = true;
1323 	}
1324 	spin_unlock(&inode->i_lock);
1325 	if (found)
1326 		nfs4_schedule_state_manager(clp);
1327 }
1328 
1329 
1330 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1331 {
1332 	struct inode *inode = state->inode;
1333 	struct nfs_inode *nfsi = NFS_I(inode);
1334 	struct file_lock *fl;
1335 	int status = 0;
1336 
1337 	if (inode->i_flock == NULL)
1338 		return 0;
1339 
1340 	/* Guard against delegation returns and new lock/unlock calls */
1341 	down_write(&nfsi->rwsem);
1342 	/* Protect inode->i_flock using the BKL */
1343 	lock_flocks();
1344 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1345 		if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1346 			continue;
1347 		if (nfs_file_open_context(fl->fl_file)->state != state)
1348 			continue;
1349 		unlock_flocks();
1350 		status = ops->recover_lock(state, fl);
1351 		switch (status) {
1352 			case 0:
1353 				break;
1354 			case -ESTALE:
1355 			case -NFS4ERR_ADMIN_REVOKED:
1356 			case -NFS4ERR_STALE_STATEID:
1357 			case -NFS4ERR_BAD_STATEID:
1358 			case -NFS4ERR_EXPIRED:
1359 			case -NFS4ERR_NO_GRACE:
1360 			case -NFS4ERR_STALE_CLIENTID:
1361 			case -NFS4ERR_BADSESSION:
1362 			case -NFS4ERR_BADSLOT:
1363 			case -NFS4ERR_BAD_HIGH_SLOT:
1364 			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1365 				goto out;
1366 			default:
1367 				printk(KERN_ERR "NFS: %s: unhandled error %d. "
1368 					"Zeroing state\n", __func__, status);
1369 			case -ENOMEM:
1370 			case -NFS4ERR_DENIED:
1371 			case -NFS4ERR_RECLAIM_BAD:
1372 			case -NFS4ERR_RECLAIM_CONFLICT:
1373 				/* kill_proc(fl->fl_pid, SIGLOST, 1); */
1374 				status = 0;
1375 		}
1376 		lock_flocks();
1377 	}
1378 	unlock_flocks();
1379 out:
1380 	up_write(&nfsi->rwsem);
1381 	return status;
1382 }
1383 
1384 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1385 {
1386 	struct nfs4_state *state;
1387 	struct nfs4_lock_state *lock;
1388 	int status = 0;
1389 
1390 	/* Note: we rely on the sp->so_states list being ordered
1391 	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1392 	 * states first.
1393 	 * This is needed to ensure that the server won't give us any
1394 	 * read delegations that we have to return if, say, we are
1395 	 * recovering after a network partition or a reboot from a
1396 	 * server that doesn't support a grace period.
1397 	 */
1398 restart:
1399 	spin_lock(&sp->so_lock);
1400 	list_for_each_entry(state, &sp->so_states, open_states) {
1401 		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1402 			continue;
1403 		if (state->state == 0)
1404 			continue;
1405 		atomic_inc(&state->count);
1406 		spin_unlock(&sp->so_lock);
1407 		status = ops->recover_open(sp, state);
1408 		if (status >= 0) {
1409 			status = nfs4_reclaim_locks(state, ops);
1410 			if (status >= 0) {
1411 				spin_lock(&state->state_lock);
1412 				list_for_each_entry(lock, &state->lock_states, ls_locks) {
1413 					if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1414 						pr_warn_ratelimited("NFS: "
1415 							"%s: Lock reclaim "
1416 							"failed!\n", __func__);
1417 				}
1418 				spin_unlock(&state->state_lock);
1419 				nfs4_put_open_state(state);
1420 				goto restart;
1421 			}
1422 		}
1423 		switch (status) {
1424 			default:
1425 				printk(KERN_ERR "NFS: %s: unhandled error %d. "
1426 					"Zeroing state\n", __func__, status);
1427 			case -ENOENT:
1428 			case -ENOMEM:
1429 			case -ESTALE:
1430 				/*
1431 				 * Open state on this file cannot be recovered
1432 				 * All we can do is revert to using the zero stateid.
1433 				 */
1434 				memset(&state->stateid, 0,
1435 					sizeof(state->stateid));
1436 				/* Mark the file as being 'closed' */
1437 				state->state = 0;
1438 				break;
1439 			case -NFS4ERR_ADMIN_REVOKED:
1440 			case -NFS4ERR_STALE_STATEID:
1441 			case -NFS4ERR_BAD_STATEID:
1442 			case -NFS4ERR_RECLAIM_BAD:
1443 			case -NFS4ERR_RECLAIM_CONFLICT:
1444 				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1445 				break;
1446 			case -NFS4ERR_EXPIRED:
1447 			case -NFS4ERR_NO_GRACE:
1448 				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1449 			case -NFS4ERR_STALE_CLIENTID:
1450 			case -NFS4ERR_BADSESSION:
1451 			case -NFS4ERR_BADSLOT:
1452 			case -NFS4ERR_BAD_HIGH_SLOT:
1453 			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1454 				goto out_err;
1455 		}
1456 		nfs4_put_open_state(state);
1457 		goto restart;
1458 	}
1459 	spin_unlock(&sp->so_lock);
1460 	return 0;
1461 out_err:
1462 	nfs4_put_open_state(state);
1463 	return status;
1464 }
1465 
1466 static void nfs4_clear_open_state(struct nfs4_state *state)
1467 {
1468 	struct nfs4_lock_state *lock;
1469 
1470 	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1471 	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1472 	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1473 	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1474 	spin_lock(&state->state_lock);
1475 	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1476 		lock->ls_seqid.flags = 0;
1477 		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1478 	}
1479 	spin_unlock(&state->state_lock);
1480 }
1481 
1482 static void nfs4_reset_seqids(struct nfs_server *server,
1483 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1484 {
1485 	struct nfs_client *clp = server->nfs_client;
1486 	struct nfs4_state_owner *sp;
1487 	struct rb_node *pos;
1488 	struct nfs4_state *state;
1489 
1490 	spin_lock(&clp->cl_lock);
1491 	for (pos = rb_first(&server->state_owners);
1492 	     pos != NULL;
1493 	     pos = rb_next(pos)) {
1494 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1495 		sp->so_seqid.flags = 0;
1496 		spin_lock(&sp->so_lock);
1497 		list_for_each_entry(state, &sp->so_states, open_states) {
1498 			if (mark_reclaim(clp, state))
1499 				nfs4_clear_open_state(state);
1500 		}
1501 		spin_unlock(&sp->so_lock);
1502 	}
1503 	spin_unlock(&clp->cl_lock);
1504 }
1505 
1506 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1507 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1508 {
1509 	struct nfs_server *server;
1510 
1511 	rcu_read_lock();
1512 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1513 		nfs4_reset_seqids(server, mark_reclaim);
1514 	rcu_read_unlock();
1515 }
1516 
1517 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1518 {
1519 	/* Mark all delegations for reclaim */
1520 	nfs_delegation_mark_reclaim(clp);
1521 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1522 }
1523 
1524 static void nfs4_reclaim_complete(struct nfs_client *clp,
1525 				 const struct nfs4_state_recovery_ops *ops)
1526 {
1527 	/* Notify the server we're done reclaiming our state */
1528 	if (ops->reclaim_complete)
1529 		(void)ops->reclaim_complete(clp);
1530 }
1531 
1532 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1533 {
1534 	struct nfs_client *clp = server->nfs_client;
1535 	struct nfs4_state_owner *sp;
1536 	struct rb_node *pos;
1537 	struct nfs4_state *state;
1538 
1539 	spin_lock(&clp->cl_lock);
1540 	for (pos = rb_first(&server->state_owners);
1541 	     pos != NULL;
1542 	     pos = rb_next(pos)) {
1543 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1544 		spin_lock(&sp->so_lock);
1545 		list_for_each_entry(state, &sp->so_states, open_states) {
1546 			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1547 						&state->flags))
1548 				continue;
1549 			nfs4_state_mark_reclaim_nograce(clp, state);
1550 		}
1551 		spin_unlock(&sp->so_lock);
1552 	}
1553 	spin_unlock(&clp->cl_lock);
1554 }
1555 
1556 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1557 {
1558 	struct nfs_server *server;
1559 
1560 	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1561 		return 0;
1562 
1563 	rcu_read_lock();
1564 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1565 		nfs4_clear_reclaim_server(server);
1566 	rcu_read_unlock();
1567 
1568 	nfs_delegation_reap_unclaimed(clp);
1569 	return 1;
1570 }
1571 
1572 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1573 {
1574 	if (!nfs4_state_clear_reclaim_reboot(clp))
1575 		return;
1576 	nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1577 }
1578 
1579 static void nfs_delegation_clear_all(struct nfs_client *clp)
1580 {
1581 	nfs_delegation_mark_reclaim(clp);
1582 	nfs_delegation_reap_unclaimed(clp);
1583 }
1584 
1585 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1586 {
1587 	nfs_delegation_clear_all(clp);
1588 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1589 }
1590 
1591 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1592 {
1593 	switch (error) {
1594 		case 0:
1595 			break;
1596 		case -NFS4ERR_CB_PATH_DOWN:
1597 			nfs40_handle_cb_pathdown(clp);
1598 			break;
1599 		case -NFS4ERR_NO_GRACE:
1600 			nfs4_state_end_reclaim_reboot(clp);
1601 			break;
1602 		case -NFS4ERR_STALE_CLIENTID:
1603 		case -NFS4ERR_LEASE_MOVED:
1604 			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1605 			nfs4_state_clear_reclaim_reboot(clp);
1606 			nfs4_state_start_reclaim_reboot(clp);
1607 			break;
1608 		case -NFS4ERR_EXPIRED:
1609 			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1610 			nfs4_state_start_reclaim_nograce(clp);
1611 			break;
1612 		case -NFS4ERR_BADSESSION:
1613 		case -NFS4ERR_BADSLOT:
1614 		case -NFS4ERR_BAD_HIGH_SLOT:
1615 		case -NFS4ERR_DEADSESSION:
1616 		case -NFS4ERR_SEQ_FALSE_RETRY:
1617 		case -NFS4ERR_SEQ_MISORDERED:
1618 			set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1619 			/* Zero session reset errors */
1620 			break;
1621 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1622 			set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1623 			break;
1624 		default:
1625 			dprintk("%s: failed to handle error %d for server %s\n",
1626 					__func__, error, clp->cl_hostname);
1627 			return error;
1628 	}
1629 	dprintk("%s: handled error %d for server %s\n", __func__, error,
1630 			clp->cl_hostname);
1631 	return 0;
1632 }
1633 
1634 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1635 {
1636 	struct nfs4_state_owner *sp;
1637 	struct nfs_server *server;
1638 	struct rb_node *pos;
1639 	int status = 0;
1640 
1641 restart:
1642 	rcu_read_lock();
1643 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1644 		nfs4_purge_state_owners(server);
1645 		spin_lock(&clp->cl_lock);
1646 		for (pos = rb_first(&server->state_owners);
1647 		     pos != NULL;
1648 		     pos = rb_next(pos)) {
1649 			sp = rb_entry(pos,
1650 				struct nfs4_state_owner, so_server_node);
1651 			if (!test_and_clear_bit(ops->owner_flag_bit,
1652 							&sp->so_flags))
1653 				continue;
1654 			atomic_inc(&sp->so_count);
1655 			spin_unlock(&clp->cl_lock);
1656 			rcu_read_unlock();
1657 
1658 			status = nfs4_reclaim_open_state(sp, ops);
1659 			if (status < 0) {
1660 				set_bit(ops->owner_flag_bit, &sp->so_flags);
1661 				nfs4_put_state_owner(sp);
1662 				return nfs4_recovery_handle_error(clp, status);
1663 			}
1664 
1665 			nfs4_put_state_owner(sp);
1666 			goto restart;
1667 		}
1668 		spin_unlock(&clp->cl_lock);
1669 	}
1670 	rcu_read_unlock();
1671 	return status;
1672 }
1673 
1674 static int nfs4_check_lease(struct nfs_client *clp)
1675 {
1676 	struct rpc_cred *cred;
1677 	const struct nfs4_state_maintenance_ops *ops =
1678 		clp->cl_mvops->state_renewal_ops;
1679 	int status;
1680 
1681 	/* Is the client already known to have an expired lease? */
1682 	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1683 		return 0;
1684 	spin_lock(&clp->cl_lock);
1685 	cred = ops->get_state_renewal_cred_locked(clp);
1686 	spin_unlock(&clp->cl_lock);
1687 	if (cred == NULL) {
1688 		cred = nfs4_get_setclientid_cred(clp);
1689 		status = -ENOKEY;
1690 		if (cred == NULL)
1691 			goto out;
1692 	}
1693 	status = ops->renew_lease(clp, cred);
1694 	put_rpccred(cred);
1695 out:
1696 	return nfs4_recovery_handle_error(clp, status);
1697 }
1698 
1699 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1700  * and for recoverable errors on EXCHANGE_ID for v4.1
1701  */
1702 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1703 {
1704 	switch (status) {
1705 	case -NFS4ERR_SEQ_MISORDERED:
1706 		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1707 			return -ESERVERFAULT;
1708 		/* Lease confirmation error: retry after purging the lease */
1709 		ssleep(1);
1710 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1711 		break;
1712 	case -NFS4ERR_STALE_CLIENTID:
1713 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1714 		nfs4_state_clear_reclaim_reboot(clp);
1715 		nfs4_state_start_reclaim_reboot(clp);
1716 		break;
1717 	case -NFS4ERR_CLID_INUSE:
1718 		pr_err("NFS: Server %s reports our clientid is in use\n",
1719 			clp->cl_hostname);
1720 		nfs_mark_client_ready(clp, -EPERM);
1721 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1722 		return -EPERM;
1723 	case -EACCES:
1724 		if (clp->cl_machine_cred == NULL)
1725 			return -EACCES;
1726 		/* Handle case where the user hasn't set up machine creds */
1727 		nfs4_clear_machine_cred(clp);
1728 	case -NFS4ERR_DELAY:
1729 	case -ETIMEDOUT:
1730 	case -EAGAIN:
1731 		ssleep(1);
1732 		break;
1733 
1734 	case -NFS4ERR_MINOR_VERS_MISMATCH:
1735 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1736 			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1737 		dprintk("%s: exit with error %d for server %s\n",
1738 				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
1739 		return -EPROTONOSUPPORT;
1740 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1741 				 * in nfs4_exchange_id */
1742 	default:
1743 		dprintk("%s: exit with error %d for server %s\n", __func__,
1744 				status, clp->cl_hostname);
1745 		return status;
1746 	}
1747 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1748 	dprintk("%s: handled error %d for server %s\n", __func__, status,
1749 			clp->cl_hostname);
1750 	return 0;
1751 }
1752 
1753 static int nfs4_establish_lease(struct nfs_client *clp)
1754 {
1755 	struct rpc_cred *cred;
1756 	const struct nfs4_state_recovery_ops *ops =
1757 		clp->cl_mvops->reboot_recovery_ops;
1758 	int status;
1759 
1760 	cred = ops->get_clid_cred(clp);
1761 	if (cred == NULL)
1762 		return -ENOENT;
1763 	status = ops->establish_clid(clp, cred);
1764 	put_rpccred(cred);
1765 	if (status != 0)
1766 		return status;
1767 	pnfs_destroy_all_layouts(clp);
1768 	return 0;
1769 }
1770 
1771 /*
1772  * Returns zero or a negative errno.  NFS4ERR values are converted
1773  * to local errno values.
1774  */
1775 static int nfs4_reclaim_lease(struct nfs_client *clp)
1776 {
1777 	int status;
1778 
1779 	status = nfs4_establish_lease(clp);
1780 	if (status < 0)
1781 		return nfs4_handle_reclaim_lease_error(clp, status);
1782 	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1783 		nfs4_state_start_reclaim_nograce(clp);
1784 	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1785 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1786 	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1787 	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1788 	return 0;
1789 }
1790 
1791 static int nfs4_purge_lease(struct nfs_client *clp)
1792 {
1793 	int status;
1794 
1795 	status = nfs4_establish_lease(clp);
1796 	if (status < 0)
1797 		return nfs4_handle_reclaim_lease_error(clp, status);
1798 	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1799 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1800 	nfs4_state_start_reclaim_nograce(clp);
1801 	return 0;
1802 }
1803 
1804 /**
1805  * nfs4_discover_server_trunking - Detect server IP address trunking
1806  *
1807  * @clp: nfs_client under test
1808  * @result: OUT: found nfs_client, or clp
1809  *
1810  * Returns zero or a negative errno.  If zero is returned,
1811  * an nfs_client pointer is planted in "result".
1812  *
1813  * Note: since we are invoked in process context, and
1814  * not from inside the state manager, we cannot use
1815  * nfs4_handle_reclaim_lease_error().
1816  */
1817 int nfs4_discover_server_trunking(struct nfs_client *clp,
1818 				  struct nfs_client **result)
1819 {
1820 	const struct nfs4_state_recovery_ops *ops =
1821 				clp->cl_mvops->reboot_recovery_ops;
1822 	rpc_authflavor_t *flavors, flav, save;
1823 	struct rpc_clnt *clnt;
1824 	struct rpc_cred *cred;
1825 	int i, len, status;
1826 
1827 	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1828 
1829 	len = NFS_MAX_SECFLAVORS;
1830 	flavors = kcalloc(len, sizeof(*flavors), GFP_KERNEL);
1831 	if (flavors == NULL) {
1832 		status = -ENOMEM;
1833 		goto out;
1834 	}
1835 	len = rpcauth_list_flavors(flavors, len);
1836 	if (len < 0) {
1837 		status = len;
1838 		goto out_free;
1839 	}
1840 	clnt = clp->cl_rpcclient;
1841 	save = clnt->cl_auth->au_flavor;
1842 	i = 0;
1843 
1844 	mutex_lock(&nfs_clid_init_mutex);
1845 	status  = -ENOENT;
1846 again:
1847 	cred = ops->get_clid_cred(clp);
1848 	if (cred == NULL)
1849 		goto out_unlock;
1850 
1851 	status = ops->detect_trunking(clp, result, cred);
1852 	put_rpccred(cred);
1853 	switch (status) {
1854 	case 0:
1855 		break;
1856 
1857 	case -EACCES:
1858 		if (clp->cl_machine_cred == NULL)
1859 			break;
1860 		/* Handle case where the user hasn't set up machine creds */
1861 		nfs4_clear_machine_cred(clp);
1862 	case -NFS4ERR_DELAY:
1863 	case -ETIMEDOUT:
1864 	case -EAGAIN:
1865 		ssleep(1);
1866 		dprintk("NFS: %s after status %d, retrying\n",
1867 			__func__, status);
1868 		goto again;
1869 
1870 	case -NFS4ERR_CLID_INUSE:
1871 	case -NFS4ERR_WRONGSEC:
1872 		status = -EPERM;
1873 		if (i >= len)
1874 			break;
1875 
1876 		flav = flavors[i++];
1877 		if (flav == save)
1878 			flav = flavors[i++];
1879 		clnt = rpc_clone_client_set_auth(clnt, flav);
1880 		if (IS_ERR(clnt)) {
1881 			status = PTR_ERR(clnt);
1882 			break;
1883 		}
1884 		clp->cl_rpcclient = clnt;
1885 		goto again;
1886 
1887 	case -NFS4ERR_MINOR_VERS_MISMATCH:
1888 		status = -EPROTONOSUPPORT;
1889 		break;
1890 
1891 	case -EKEYEXPIRED:
1892 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1893 				 * in nfs4_exchange_id */
1894 		status = -EKEYEXPIRED;
1895 	}
1896 
1897 out_unlock:
1898 	mutex_unlock(&nfs_clid_init_mutex);
1899 out_free:
1900 	kfree(flavors);
1901 out:
1902 	dprintk("NFS: %s: status = %d\n", __func__, status);
1903 	return status;
1904 }
1905 
1906 #ifdef CONFIG_NFS_V4_1
1907 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1908 {
1909 	struct nfs_client *clp = session->clp;
1910 
1911 	switch (err) {
1912 	default:
1913 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1914 		break;
1915 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1916 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1917 	}
1918 	nfs4_schedule_lease_recovery(clp);
1919 }
1920 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1921 
1922 static void nfs41_ping_server(struct nfs_client *clp)
1923 {
1924 	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
1925 	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1926 	nfs4_schedule_state_manager(clp);
1927 }
1928 
1929 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
1930 {
1931 	nfs41_ping_server(clp);
1932 }
1933 
1934 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
1935 {
1936 	nfs41_ping_server(clp);
1937 }
1938 
1939 static void nfs4_reset_all_state(struct nfs_client *clp)
1940 {
1941 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1942 		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1943 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1944 		nfs4_state_start_reclaim_nograce(clp);
1945 		dprintk("%s: scheduling reset of all state for server %s!\n",
1946 				__func__, clp->cl_hostname);
1947 		nfs4_schedule_state_manager(clp);
1948 	}
1949 }
1950 
1951 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1952 {
1953 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1954 		nfs4_state_start_reclaim_reboot(clp);
1955 		dprintk("%s: server %s rebooted!\n", __func__,
1956 				clp->cl_hostname);
1957 		nfs4_schedule_state_manager(clp);
1958 	}
1959 }
1960 
1961 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1962 {
1963 	nfs4_reset_all_state(clp);
1964 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1965 }
1966 
1967 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1968 {
1969 	/* This will need to handle layouts too */
1970 	nfs_expire_all_delegations(clp);
1971 	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1972 			clp->cl_hostname);
1973 }
1974 
1975 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1976 {
1977 	nfs_expire_all_delegations(clp);
1978 	if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1979 		nfs4_schedule_state_manager(clp);
1980 	dprintk("%s: server %s declared a backchannel fault\n", __func__,
1981 			clp->cl_hostname);
1982 }
1983 
1984 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1985 {
1986 	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1987 		&clp->cl_state) == 0)
1988 		nfs4_schedule_state_manager(clp);
1989 }
1990 
1991 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1992 {
1993 	if (!flags)
1994 		return;
1995 
1996 	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
1997 		__func__, clp->cl_hostname, clp->cl_clientid, flags);
1998 
1999 	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2000 		nfs41_handle_server_reboot(clp);
2001 	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2002 			    SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2003 			    SEQ4_STATUS_ADMIN_STATE_REVOKED |
2004 			    SEQ4_STATUS_LEASE_MOVED))
2005 		nfs41_handle_state_revoked(clp);
2006 	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2007 		nfs41_handle_recallable_state_revoked(clp);
2008 	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2009 		nfs41_handle_backchannel_fault(clp);
2010 	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2011 				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2012 		nfs41_handle_cb_path_down(clp);
2013 }
2014 
2015 static int nfs4_reset_session(struct nfs_client *clp)
2016 {
2017 	struct rpc_cred *cred;
2018 	int status;
2019 
2020 	if (!nfs4_has_session(clp))
2021 		return 0;
2022 	nfs4_begin_drain_session(clp);
2023 	cred = nfs4_get_exchange_id_cred(clp);
2024 	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2025 	if (status && status != -NFS4ERR_BADSESSION &&
2026 	    status != -NFS4ERR_DEADSESSION) {
2027 		status = nfs4_recovery_handle_error(clp, status);
2028 		goto out;
2029 	}
2030 
2031 	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2032 	status = nfs4_proc_create_session(clp, cred);
2033 	if (status) {
2034 		dprintk("%s: session reset failed with status %d for server %s!\n",
2035 			__func__, status, clp->cl_hostname);
2036 		status = nfs4_handle_reclaim_lease_error(clp, status);
2037 		goto out;
2038 	}
2039 	nfs41_finish_session_reset(clp);
2040 	dprintk("%s: session reset was successful for server %s!\n",
2041 			__func__, clp->cl_hostname);
2042 out:
2043 	if (cred)
2044 		put_rpccred(cred);
2045 	return status;
2046 }
2047 
2048 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2049 {
2050 	struct rpc_cred *cred;
2051 	int ret;
2052 
2053 	if (!nfs4_has_session(clp))
2054 		return 0;
2055 	nfs4_begin_drain_session(clp);
2056 	cred = nfs4_get_exchange_id_cred(clp);
2057 	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2058 	if (cred)
2059 		put_rpccred(cred);
2060 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2061 	switch (ret) {
2062 	case 0:
2063 		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2064 			__func__, clp->cl_hostname);
2065 		break;
2066 	case -NFS4ERR_DELAY:
2067 		ssleep(1);
2068 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2069 		break;
2070 	default:
2071 		return nfs4_recovery_handle_error(clp, ret);
2072 	}
2073 	return 0;
2074 }
2075 #else /* CONFIG_NFS_V4_1 */
2076 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2077 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2078 
2079 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2080 {
2081 	return 0;
2082 }
2083 #endif /* CONFIG_NFS_V4_1 */
2084 
2085 static void nfs4_state_manager(struct nfs_client *clp)
2086 {
2087 	int status = 0;
2088 	const char *section = "", *section_sep = "";
2089 
2090 	/* Ensure exclusive access to NFSv4 state */
2091 	do {
2092 		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2093 			section = "purge state";
2094 			status = nfs4_purge_lease(clp);
2095 			if (status < 0)
2096 				goto out_error;
2097 			continue;
2098 		}
2099 
2100 		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2101 			section = "lease expired";
2102 			/* We're going to have to re-establish a clientid */
2103 			status = nfs4_reclaim_lease(clp);
2104 			if (status < 0)
2105 				goto out_error;
2106 			continue;
2107 		}
2108 
2109 		/* Initialize or reset the session */
2110 		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2111 			section = "reset session";
2112 			status = nfs4_reset_session(clp);
2113 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2114 				continue;
2115 			if (status < 0)
2116 				goto out_error;
2117 		}
2118 
2119 		/* Send BIND_CONN_TO_SESSION */
2120 		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2121 				&clp->cl_state)) {
2122 			section = "bind conn to session";
2123 			status = nfs4_bind_conn_to_session(clp);
2124 			if (status < 0)
2125 				goto out_error;
2126 			continue;
2127 		}
2128 
2129 		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2130 			section = "check lease";
2131 			status = nfs4_check_lease(clp);
2132 			if (status < 0)
2133 				goto out_error;
2134 			continue;
2135 		}
2136 
2137 		/* First recover reboot state... */
2138 		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2139 			section = "reclaim reboot";
2140 			status = nfs4_do_reclaim(clp,
2141 				clp->cl_mvops->reboot_recovery_ops);
2142 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2143 			    test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2144 				continue;
2145 			nfs4_state_end_reclaim_reboot(clp);
2146 			if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2147 				continue;
2148 			if (status < 0)
2149 				goto out_error;
2150 		}
2151 
2152 		/* Now recover expired state... */
2153 		if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2154 			section = "reclaim nograce";
2155 			status = nfs4_do_reclaim(clp,
2156 				clp->cl_mvops->nograce_recovery_ops);
2157 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2158 			    test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2159 			    test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2160 				continue;
2161 			if (status < 0)
2162 				goto out_error;
2163 		}
2164 
2165 		nfs4_end_drain_session(clp);
2166 		if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2167 			nfs_client_return_marked_delegations(clp);
2168 			continue;
2169 		}
2170 
2171 		nfs4_clear_state_manager_bit(clp);
2172 		/* Did we race with an attempt to give us more work? */
2173 		if (clp->cl_state == 0)
2174 			break;
2175 		if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2176 			break;
2177 	} while (atomic_read(&clp->cl_count) > 1);
2178 	return;
2179 out_error:
2180 	if (strlen(section))
2181 		section_sep = ": ";
2182 	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2183 			" with error %d\n", section_sep, section,
2184 			clp->cl_hostname, -status);
2185 	ssleep(1);
2186 	nfs4_end_drain_session(clp);
2187 	nfs4_clear_state_manager_bit(clp);
2188 }
2189 
2190 static int nfs4_run_state_manager(void *ptr)
2191 {
2192 	struct nfs_client *clp = ptr;
2193 
2194 	allow_signal(SIGKILL);
2195 	nfs4_state_manager(clp);
2196 	nfs_put_client(clp);
2197 	module_put_and_exit(0);
2198 	return 0;
2199 }
2200 
2201 /*
2202  * Local variables:
2203  *  c-basic-offset: 8
2204  * End:
2205  */
2206