xref: /kernel/linux/linux-5.10/fs/nfs/nfs4state.c (revision 8c2ecf20)
1/*
2 *  fs/nfs/nfs4state.c
3 *
4 *  Client-side XDR for NFSv4.
5 *
6 *  Copyright (c) 2002 The Regents of the University of Michigan.
7 *  All rights reserved.
8 *
9 *  Kendrick Smith <kmsmith@umich.edu>
10 *
11 *  Redistribution and use in source and binary forms, with or without
12 *  modification, are permitted provided that the following conditions
13 *  are met:
14 *
15 *  1. Redistributions of source code must retain the above copyright
16 *     notice, this list of conditions and the following disclaimer.
17 *  2. Redistributions in binary form must reproduce the above copyright
18 *     notice, this list of conditions and the following disclaimer in the
19 *     documentation and/or other materials provided with the distribution.
20 *  3. Neither the name of the University nor the names of its
21 *     contributors may be used to endorse or promote products derived
22 *     from this software without specific prior written permission.
23 *
24 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 *
36 * Implementation of the NFSv4 state model.  For the time being,
37 * this is minimal, but will be made much more complex in a
38 * subsequent patch.
39 */
40
41#include <linux/kernel.h>
42#include <linux/slab.h>
43#include <linux/fs.h>
44#include <linux/nfs_fs.h>
45#include <linux/kthread.h>
46#include <linux/module.h>
47#include <linux/random.h>
48#include <linux/ratelimit.h>
49#include <linux/workqueue.h>
50#include <linux/bitops.h>
51#include <linux/jiffies.h>
52#include <linux/sched/mm.h>
53
54#include <linux/sunrpc/clnt.h>
55
56#include "nfs4_fs.h"
57#include "callback.h"
58#include "delegation.h"
59#include "internal.h"
60#include "nfs4idmap.h"
61#include "nfs4session.h"
62#include "pnfs.h"
63#include "netns.h"
64#include "nfs4trace.h"
65
66#define NFSDBG_FACILITY		NFSDBG_STATE
67
68#define OPENOWNER_POOL_SIZE	8
69
70static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp);
71
72const nfs4_stateid zero_stateid = {
73	{ .data = { 0 } },
74	.type = NFS4_SPECIAL_STATEID_TYPE,
75};
76const nfs4_stateid invalid_stateid = {
77	{
78		/* Funky initialiser keeps older gcc versions happy */
79		.data = { 0xff, 0xff, 0xff, 0xff, 0 },
80	},
81	.type = NFS4_INVALID_STATEID_TYPE,
82};
83
84const nfs4_stateid current_stateid = {
85	{
86		/* Funky initialiser keeps older gcc versions happy */
87		.data = { 0x0, 0x0, 0x0, 0x1, 0 },
88	},
89	.type = NFS4_SPECIAL_STATEID_TYPE,
90};
91
92static DEFINE_MUTEX(nfs_clid_init_mutex);
93
94static int nfs4_setup_state_renewal(struct nfs_client *clp)
95{
96	int status;
97	struct nfs_fsinfo fsinfo;
98
99	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
100		nfs4_schedule_state_renewal(clp);
101		return 0;
102	}
103
104	status = nfs4_proc_get_lease_time(clp, &fsinfo);
105	if (status == 0) {
106		nfs4_set_lease_period(clp, fsinfo.lease_time * HZ);
107		nfs4_schedule_state_renewal(clp);
108	}
109
110	return status;
111}
112
113int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
114{
115	struct nfs4_setclientid_res clid = {
116		.clientid = clp->cl_clientid,
117		.confirm = clp->cl_confirm,
118	};
119	unsigned short port;
120	int status;
121	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
122
123	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
124		goto do_confirm;
125	port = nn->nfs_callback_tcpport;
126	if (clp->cl_addr.ss_family == AF_INET6)
127		port = nn->nfs_callback_tcpport6;
128
129	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
130	if (status != 0)
131		goto out;
132	clp->cl_clientid = clid.clientid;
133	clp->cl_confirm = clid.confirm;
134	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
135do_confirm:
136	status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
137	if (status != 0)
138		goto out;
139	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
140	nfs4_setup_state_renewal(clp);
141out:
142	return status;
143}
144
145/**
146 * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
147 *
148 * @clp: nfs_client under test
149 * @result: OUT: found nfs_client, or clp
150 * @cred: credential to use for trunking test
151 *
152 * Returns zero, a negative errno, or a negative NFS4ERR status.
153 * If zero is returned, an nfs_client pointer is planted in
154 * "result".
155 *
156 * Note: The returned client may not yet be marked ready.
157 */
158int nfs40_discover_server_trunking(struct nfs_client *clp,
159				   struct nfs_client **result,
160				   const struct cred *cred)
161{
162	struct nfs4_setclientid_res clid = {
163		.clientid = clp->cl_clientid,
164		.confirm = clp->cl_confirm,
165	};
166	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
167	unsigned short port;
168	int status;
169
170	port = nn->nfs_callback_tcpport;
171	if (clp->cl_addr.ss_family == AF_INET6)
172		port = nn->nfs_callback_tcpport6;
173
174	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
175	if (status != 0)
176		goto out;
177	clp->cl_clientid = clid.clientid;
178	clp->cl_confirm = clid.confirm;
179
180	status = nfs40_walk_client_list(clp, result, cred);
181	if (status == 0) {
182		/* Sustain the lease, even if it's empty.  If the clientid4
183		 * goes stale it's of no use for trunking discovery. */
184		nfs4_schedule_state_renewal(*result);
185
186		/* If the client state need to recover, do it. */
187		if (clp->cl_state)
188			nfs4_schedule_state_manager(clp);
189	}
190out:
191	return status;
192}
193
194const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
195{
196	return get_cred(rpc_machine_cred());
197}
198
199static void nfs4_root_machine_cred(struct nfs_client *clp)
200{
201
202	/* Force root creds instead of machine */
203	clp->cl_principal = NULL;
204	clp->cl_rpcclient->cl_principal = NULL;
205}
206
207static const struct cred *
208nfs4_get_renew_cred_server_locked(struct nfs_server *server)
209{
210	const struct cred *cred = NULL;
211	struct nfs4_state_owner *sp;
212	struct rb_node *pos;
213
214	for (pos = rb_first(&server->state_owners);
215	     pos != NULL;
216	     pos = rb_next(pos)) {
217		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
218		if (list_empty(&sp->so_states))
219			continue;
220		cred = get_cred(sp->so_cred);
221		break;
222	}
223	return cred;
224}
225
226/**
227 * nfs4_get_renew_cred - Acquire credential for a renew operation
228 * @clp: client state handle
229 *
230 * Returns an rpc_cred with reference count bumped, or NULL.
231 * Caller must hold clp->cl_lock.
232 */
233const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
234{
235	const struct cred *cred = NULL;
236	struct nfs_server *server;
237
238	/* Use machine credentials if available */
239	cred = nfs4_get_machine_cred(clp);
240	if (cred != NULL)
241		goto out;
242
243	spin_lock(&clp->cl_lock);
244	rcu_read_lock();
245	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
246		cred = nfs4_get_renew_cred_server_locked(server);
247		if (cred != NULL)
248			break;
249	}
250	rcu_read_unlock();
251	spin_unlock(&clp->cl_lock);
252
253out:
254	return cred;
255}
256
257static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
258{
259	if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
260		spin_lock(&tbl->slot_tbl_lock);
261		nfs41_wake_slot_table(tbl);
262		spin_unlock(&tbl->slot_tbl_lock);
263	}
264}
265
266static void nfs4_end_drain_session(struct nfs_client *clp)
267{
268	struct nfs4_session *ses = clp->cl_session;
269
270	if (clp->cl_slot_tbl) {
271		nfs4_end_drain_slot_table(clp->cl_slot_tbl);
272		return;
273	}
274
275	if (ses != NULL) {
276		nfs4_end_drain_slot_table(&ses->bc_slot_table);
277		nfs4_end_drain_slot_table(&ses->fc_slot_table);
278	}
279}
280
281static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
282{
283	set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
284	spin_lock(&tbl->slot_tbl_lock);
285	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
286		reinit_completion(&tbl->complete);
287		spin_unlock(&tbl->slot_tbl_lock);
288		return wait_for_completion_interruptible(&tbl->complete);
289	}
290	spin_unlock(&tbl->slot_tbl_lock);
291	return 0;
292}
293
294static int nfs4_begin_drain_session(struct nfs_client *clp)
295{
296	struct nfs4_session *ses = clp->cl_session;
297	int ret;
298
299	if (clp->cl_slot_tbl)
300		return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
301
302	/* back channel */
303	ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
304	if (ret)
305		return ret;
306	/* fore channel */
307	return nfs4_drain_slot_tbl(&ses->fc_slot_table);
308}
309
310#if defined(CONFIG_NFS_V4_1)
311
312static void nfs41_finish_session_reset(struct nfs_client *clp)
313{
314	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
315	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
316	/* create_session negotiated new slot table */
317	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
318	nfs4_setup_state_renewal(clp);
319}
320
321int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
322{
323	int status;
324
325	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
326		goto do_confirm;
327	status = nfs4_proc_exchange_id(clp, cred);
328	if (status != 0)
329		goto out;
330	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
331do_confirm:
332	status = nfs4_proc_create_session(clp, cred);
333	if (status != 0)
334		goto out;
335	if (!(clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R))
336		nfs4_state_start_reclaim_reboot(clp);
337	nfs41_finish_session_reset(clp);
338	nfs_mark_client_ready(clp, NFS_CS_READY);
339out:
340	return status;
341}
342
343/**
344 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
345 *
346 * @clp: nfs_client under test
347 * @result: OUT: found nfs_client, or clp
348 * @cred: credential to use for trunking test
349 *
350 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
351 * If NFS4_OK is returned, an nfs_client pointer is planted in
352 * "result".
353 *
354 * Note: The returned client may not yet be marked ready.
355 */
356int nfs41_discover_server_trunking(struct nfs_client *clp,
357				   struct nfs_client **result,
358				   const struct cred *cred)
359{
360	int status;
361
362	status = nfs4_proc_exchange_id(clp, cred);
363	if (status != NFS4_OK)
364		return status;
365
366	status = nfs41_walk_client_list(clp, result, cred);
367	if (status < 0)
368		return status;
369	if (clp != *result)
370		return 0;
371
372	/*
373	 * Purge state if the client id was established in a prior
374	 * instance and the client id could not have arrived on the
375	 * server via Transparent State Migration.
376	 */
377	if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
378		if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
379			set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
380		else
381			set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
382	}
383	nfs4_schedule_state_manager(clp);
384	status = nfs_wait_client_init_complete(clp);
385	if (status < 0)
386		nfs_put_client(clp);
387	return status;
388}
389
390#endif /* CONFIG_NFS_V4_1 */
391
392/**
393 * nfs4_get_clid_cred - Acquire credential for a setclientid operation
394 * @clp: client state handle
395 *
396 * Returns a cred with reference count bumped, or NULL.
397 */
398const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
399{
400	const struct cred *cred;
401
402	cred = nfs4_get_machine_cred(clp);
403	return cred;
404}
405
406static struct nfs4_state_owner *
407nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
408{
409	struct rb_node **p = &server->state_owners.rb_node,
410		       *parent = NULL;
411	struct nfs4_state_owner *sp;
412	int cmp;
413
414	while (*p != NULL) {
415		parent = *p;
416		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
417		cmp = cred_fscmp(cred, sp->so_cred);
418
419		if (cmp < 0)
420			p = &parent->rb_left;
421		else if (cmp > 0)
422			p = &parent->rb_right;
423		else {
424			if (!list_empty(&sp->so_lru))
425				list_del_init(&sp->so_lru);
426			atomic_inc(&sp->so_count);
427			return sp;
428		}
429	}
430	return NULL;
431}
432
433static struct nfs4_state_owner *
434nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
435{
436	struct nfs_server *server = new->so_server;
437	struct rb_node **p = &server->state_owners.rb_node,
438		       *parent = NULL;
439	struct nfs4_state_owner *sp;
440	int cmp;
441
442	while (*p != NULL) {
443		parent = *p;
444		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
445		cmp = cred_fscmp(new->so_cred, sp->so_cred);
446
447		if (cmp < 0)
448			p = &parent->rb_left;
449		else if (cmp > 0)
450			p = &parent->rb_right;
451		else {
452			if (!list_empty(&sp->so_lru))
453				list_del_init(&sp->so_lru);
454			atomic_inc(&sp->so_count);
455			return sp;
456		}
457	}
458	rb_link_node(&new->so_server_node, parent, p);
459	rb_insert_color(&new->so_server_node, &server->state_owners);
460	return new;
461}
462
463static void
464nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
465{
466	struct nfs_server *server = sp->so_server;
467
468	if (!RB_EMPTY_NODE(&sp->so_server_node))
469		rb_erase(&sp->so_server_node, &server->state_owners);
470}
471
472static void
473nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
474{
475	sc->create_time = ktime_get();
476	sc->flags = 0;
477	sc->counter = 0;
478	spin_lock_init(&sc->lock);
479	INIT_LIST_HEAD(&sc->list);
480	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
481}
482
483static void
484nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
485{
486	rpc_destroy_wait_queue(&sc->wait);
487}
488
489/*
490 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
491 * create a new state_owner.
492 *
493 */
494static struct nfs4_state_owner *
495nfs4_alloc_state_owner(struct nfs_server *server,
496		const struct cred *cred,
497		gfp_t gfp_flags)
498{
499	struct nfs4_state_owner *sp;
500
501	sp = kzalloc(sizeof(*sp), gfp_flags);
502	if (!sp)
503		return NULL;
504	sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
505						gfp_flags);
506	if (sp->so_seqid.owner_id < 0) {
507		kfree(sp);
508		return NULL;
509	}
510	sp->so_server = server;
511	sp->so_cred = get_cred(cred);
512	spin_lock_init(&sp->so_lock);
513	INIT_LIST_HEAD(&sp->so_states);
514	nfs4_init_seqid_counter(&sp->so_seqid);
515	atomic_set(&sp->so_count, 1);
516	INIT_LIST_HEAD(&sp->so_lru);
517	seqcount_spinlock_init(&sp->so_reclaim_seqcount, &sp->so_lock);
518	mutex_init(&sp->so_delegreturn_mutex);
519	return sp;
520}
521
522static void
523nfs4_reset_state_owner(struct nfs4_state_owner *sp)
524{
525	/* This state_owner is no longer usable, but must
526	 * remain in place so that state recovery can find it
527	 * and the opens associated with it.
528	 * It may also be used for new 'open' request to
529	 * return a delegation to the server.
530	 * So update the 'create_time' so that it looks like
531	 * a new state_owner.  This will cause the server to
532	 * request an OPEN_CONFIRM to start a new sequence.
533	 */
534	sp->so_seqid.create_time = ktime_get();
535}
536
537static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
538{
539	nfs4_destroy_seqid_counter(&sp->so_seqid);
540	put_cred(sp->so_cred);
541	ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
542	kfree(sp);
543}
544
545static void nfs4_gc_state_owners(struct nfs_server *server)
546{
547	struct nfs_client *clp = server->nfs_client;
548	struct nfs4_state_owner *sp, *tmp;
549	unsigned long time_min, time_max;
550	LIST_HEAD(doomed);
551
552	spin_lock(&clp->cl_lock);
553	time_max = jiffies;
554	time_min = (long)time_max - (long)clp->cl_lease_time;
555	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
556		/* NB: LRU is sorted so that oldest is at the head */
557		if (time_in_range(sp->so_expires, time_min, time_max))
558			break;
559		list_move(&sp->so_lru, &doomed);
560		nfs4_remove_state_owner_locked(sp);
561	}
562	spin_unlock(&clp->cl_lock);
563
564	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
565		list_del(&sp->so_lru);
566		nfs4_free_state_owner(sp);
567	}
568}
569
570/**
571 * nfs4_get_state_owner - Look up a state owner given a credential
572 * @server: nfs_server to search
573 * @cred: RPC credential to match
574 * @gfp_flags: allocation mode
575 *
576 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
577 */
578struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
579					      const struct cred *cred,
580					      gfp_t gfp_flags)
581{
582	struct nfs_client *clp = server->nfs_client;
583	struct nfs4_state_owner *sp, *new;
584
585	spin_lock(&clp->cl_lock);
586	sp = nfs4_find_state_owner_locked(server, cred);
587	spin_unlock(&clp->cl_lock);
588	if (sp != NULL)
589		goto out;
590	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
591	if (new == NULL)
592		goto out;
593	spin_lock(&clp->cl_lock);
594	sp = nfs4_insert_state_owner_locked(new);
595	spin_unlock(&clp->cl_lock);
596	if (sp != new)
597		nfs4_free_state_owner(new);
598out:
599	nfs4_gc_state_owners(server);
600	return sp;
601}
602
603/**
604 * nfs4_put_state_owner - Release a nfs4_state_owner
605 * @sp: state owner data to release
606 *
607 * Note that we keep released state owners on an LRU
608 * list.
609 * This caches valid state owners so that they can be
610 * reused, to avoid the OPEN_CONFIRM on minor version 0.
611 * It also pins the uniquifier of dropped state owners for
612 * a while, to ensure that those state owner names are
613 * never reused.
614 */
615void nfs4_put_state_owner(struct nfs4_state_owner *sp)
616{
617	struct nfs_server *server = sp->so_server;
618	struct nfs_client *clp = server->nfs_client;
619
620	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
621		return;
622
623	sp->so_expires = jiffies;
624	list_add_tail(&sp->so_lru, &server->state_owners_lru);
625	spin_unlock(&clp->cl_lock);
626}
627
628/**
629 * nfs4_purge_state_owners - Release all cached state owners
630 * @server: nfs_server with cached state owners to release
631 * @head: resulting list of state owners
632 *
633 * Called at umount time.  Remaining state owners will be on
634 * the LRU with ref count of zero.
635 * Note that the state owners are not freed, but are added
636 * to the list @head, which can later be used as an argument
637 * to nfs4_free_state_owners.
638 */
639void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
640{
641	struct nfs_client *clp = server->nfs_client;
642	struct nfs4_state_owner *sp, *tmp;
643
644	spin_lock(&clp->cl_lock);
645	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
646		list_move(&sp->so_lru, head);
647		nfs4_remove_state_owner_locked(sp);
648	}
649	spin_unlock(&clp->cl_lock);
650}
651
652/**
653 * nfs4_purge_state_owners - Release all cached state owners
654 * @head: resulting list of state owners
655 *
656 * Frees a list of state owners that was generated by
657 * nfs4_purge_state_owners
658 */
659void nfs4_free_state_owners(struct list_head *head)
660{
661	struct nfs4_state_owner *sp, *tmp;
662
663	list_for_each_entry_safe(sp, tmp, head, so_lru) {
664		list_del(&sp->so_lru);
665		nfs4_free_state_owner(sp);
666	}
667}
668
669static struct nfs4_state *
670nfs4_alloc_open_state(void)
671{
672	struct nfs4_state *state;
673
674	state = kzalloc(sizeof(*state), GFP_NOFS);
675	if (!state)
676		return NULL;
677	refcount_set(&state->count, 1);
678	INIT_LIST_HEAD(&state->lock_states);
679	spin_lock_init(&state->state_lock);
680	seqlock_init(&state->seqlock);
681	init_waitqueue_head(&state->waitq);
682	return state;
683}
684
685void
686nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
687{
688	if (state->state == fmode)
689		return;
690	/* NB! List reordering - see the reclaim code for why.  */
691	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
692		if (fmode & FMODE_WRITE)
693			list_move(&state->open_states, &state->owner->so_states);
694		else
695			list_move_tail(&state->open_states, &state->owner->so_states);
696	}
697	state->state = fmode;
698}
699
700static struct nfs4_state *
701__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
702{
703	struct nfs_inode *nfsi = NFS_I(inode);
704	struct nfs4_state *state;
705
706	list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
707		if (state->owner != owner)
708			continue;
709		if (!nfs4_valid_open_stateid(state))
710			continue;
711		if (refcount_inc_not_zero(&state->count))
712			return state;
713	}
714	return NULL;
715}
716
717static void
718nfs4_free_open_state(struct nfs4_state *state)
719{
720	kfree_rcu(state, rcu_head);
721}
722
723struct nfs4_state *
724nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
725{
726	struct nfs4_state *state, *new;
727	struct nfs_inode *nfsi = NFS_I(inode);
728
729	rcu_read_lock();
730	state = __nfs4_find_state_byowner(inode, owner);
731	rcu_read_unlock();
732	if (state)
733		goto out;
734	new = nfs4_alloc_open_state();
735	spin_lock(&owner->so_lock);
736	spin_lock(&inode->i_lock);
737	state = __nfs4_find_state_byowner(inode, owner);
738	if (state == NULL && new != NULL) {
739		state = new;
740		state->owner = owner;
741		atomic_inc(&owner->so_count);
742		ihold(inode);
743		state->inode = inode;
744		list_add_rcu(&state->inode_states, &nfsi->open_states);
745		spin_unlock(&inode->i_lock);
746		/* Note: The reclaim code dictates that we add stateless
747		 * and read-only stateids to the end of the list */
748		list_add_tail(&state->open_states, &owner->so_states);
749		spin_unlock(&owner->so_lock);
750	} else {
751		spin_unlock(&inode->i_lock);
752		spin_unlock(&owner->so_lock);
753		if (new)
754			nfs4_free_open_state(new);
755	}
756out:
757	return state;
758}
759
760void nfs4_put_open_state(struct nfs4_state *state)
761{
762	struct inode *inode = state->inode;
763	struct nfs4_state_owner *owner = state->owner;
764
765	if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
766		return;
767	spin_lock(&inode->i_lock);
768	list_del_rcu(&state->inode_states);
769	list_del(&state->open_states);
770	spin_unlock(&inode->i_lock);
771	spin_unlock(&owner->so_lock);
772	nfs4_inode_return_delegation_on_close(inode);
773	iput(inode);
774	nfs4_free_open_state(state);
775	nfs4_put_state_owner(owner);
776}
777
778/*
779 * Close the current file.
780 */
781static void __nfs4_close(struct nfs4_state *state,
782		fmode_t fmode, gfp_t gfp_mask, int wait)
783{
784	struct nfs4_state_owner *owner = state->owner;
785	int call_close = 0;
786	fmode_t newstate;
787
788	atomic_inc(&owner->so_count);
789	/* Protect against nfs4_find_state() */
790	spin_lock(&owner->so_lock);
791	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
792		case FMODE_READ:
793			state->n_rdonly--;
794			break;
795		case FMODE_WRITE:
796			state->n_wronly--;
797			break;
798		case FMODE_READ|FMODE_WRITE:
799			state->n_rdwr--;
800	}
801	newstate = FMODE_READ|FMODE_WRITE;
802	if (state->n_rdwr == 0) {
803		if (state->n_rdonly == 0) {
804			newstate &= ~FMODE_READ;
805			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
806			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
807		}
808		if (state->n_wronly == 0) {
809			newstate &= ~FMODE_WRITE;
810			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
811			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
812		}
813		if (newstate == 0)
814			clear_bit(NFS_DELEGATED_STATE, &state->flags);
815	}
816	nfs4_state_set_mode_locked(state, newstate);
817	spin_unlock(&owner->so_lock);
818
819	if (!call_close) {
820		nfs4_put_open_state(state);
821		nfs4_put_state_owner(owner);
822	} else
823		nfs4_do_close(state, gfp_mask, wait);
824}
825
826void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
827{
828	__nfs4_close(state, fmode, GFP_NOFS, 0);
829}
830
831void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
832{
833	__nfs4_close(state, fmode, GFP_KERNEL, 1);
834}
835
836/*
837 * Search the state->lock_states for an existing lock_owner
838 * that is compatible with either of the given owners.
839 * If the second is non-zero, then the first refers to a Posix-lock
840 * owner (current->files) and the second refers to a flock/OFD
841 * owner (struct file*).  In that case, prefer a match for the first
842 * owner.
843 * If both sorts of locks are held on the one file we cannot know
844 * which stateid was intended to be used, so a "correct" choice cannot
845 * be made.  Failing that, a "consistent" choice is preferable.  The
846 * consistent choice we make is to prefer the first owner, that of a
847 * Posix lock.
848 */
849static struct nfs4_lock_state *
850__nfs4_find_lock_state(struct nfs4_state *state,
851		       fl_owner_t fl_owner, fl_owner_t fl_owner2)
852{
853	struct nfs4_lock_state *pos, *ret = NULL;
854	list_for_each_entry(pos, &state->lock_states, ls_locks) {
855		if (pos->ls_owner == fl_owner) {
856			ret = pos;
857			break;
858		}
859		if (pos->ls_owner == fl_owner2)
860			ret = pos;
861	}
862	if (ret)
863		refcount_inc(&ret->ls_count);
864	return ret;
865}
866
867/*
868 * Return a compatible lock_state. If no initialized lock_state structure
869 * exists, return an uninitialized one.
870 *
871 */
872static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
873{
874	struct nfs4_lock_state *lsp;
875	struct nfs_server *server = state->owner->so_server;
876
877	lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
878	if (lsp == NULL)
879		return NULL;
880	nfs4_init_seqid_counter(&lsp->ls_seqid);
881	refcount_set(&lsp->ls_count, 1);
882	lsp->ls_state = state;
883	lsp->ls_owner = fl_owner;
884	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
885	if (lsp->ls_seqid.owner_id < 0)
886		goto out_free;
887	INIT_LIST_HEAD(&lsp->ls_locks);
888	return lsp;
889out_free:
890	kfree(lsp);
891	return NULL;
892}
893
894void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
895{
896	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
897	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
898	kfree(lsp);
899}
900
901/*
902 * Return a compatible lock_state. If no initialized lock_state structure
903 * exists, return an uninitialized one.
904 *
905 */
906static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
907{
908	struct nfs4_lock_state *lsp, *new = NULL;
909
910	for(;;) {
911		spin_lock(&state->state_lock);
912		lsp = __nfs4_find_lock_state(state, owner, NULL);
913		if (lsp != NULL)
914			break;
915		if (new != NULL) {
916			list_add(&new->ls_locks, &state->lock_states);
917			set_bit(LK_STATE_IN_USE, &state->flags);
918			lsp = new;
919			new = NULL;
920			break;
921		}
922		spin_unlock(&state->state_lock);
923		new = nfs4_alloc_lock_state(state, owner);
924		if (new == NULL)
925			return NULL;
926	}
927	spin_unlock(&state->state_lock);
928	if (new != NULL)
929		nfs4_free_lock_state(state->owner->so_server, new);
930	return lsp;
931}
932
933/*
934 * Release reference to lock_state, and free it if we see that
935 * it is no longer in use
936 */
937void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
938{
939	struct nfs_server *server;
940	struct nfs4_state *state;
941
942	if (lsp == NULL)
943		return;
944	state = lsp->ls_state;
945	if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
946		return;
947	list_del(&lsp->ls_locks);
948	if (list_empty(&state->lock_states))
949		clear_bit(LK_STATE_IN_USE, &state->flags);
950	spin_unlock(&state->state_lock);
951	server = state->owner->so_server;
952	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
953		struct nfs_client *clp = server->nfs_client;
954
955		clp->cl_mvops->free_lock_state(server, lsp);
956	} else
957		nfs4_free_lock_state(server, lsp);
958}
959
960static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
961{
962	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
963
964	dst->fl_u.nfs4_fl.owner = lsp;
965	refcount_inc(&lsp->ls_count);
966}
967
968static void nfs4_fl_release_lock(struct file_lock *fl)
969{
970	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
971}
972
973static const struct file_lock_operations nfs4_fl_lock_ops = {
974	.fl_copy_lock = nfs4_fl_copy_lock,
975	.fl_release_private = nfs4_fl_release_lock,
976};
977
978int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
979{
980	struct nfs4_lock_state *lsp;
981
982	if (fl->fl_ops != NULL)
983		return 0;
984	lsp = nfs4_get_lock_state(state, fl->fl_owner);
985	if (lsp == NULL)
986		return -ENOMEM;
987	fl->fl_u.nfs4_fl.owner = lsp;
988	fl->fl_ops = &nfs4_fl_lock_ops;
989	return 0;
990}
991
992static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
993		struct nfs4_state *state,
994		const struct nfs_lock_context *l_ctx)
995{
996	struct nfs4_lock_state *lsp;
997	fl_owner_t fl_owner, fl_flock_owner;
998	int ret = -ENOENT;
999
1000	if (l_ctx == NULL)
1001		goto out;
1002
1003	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1004		goto out;
1005
1006	fl_owner = l_ctx->lockowner;
1007	fl_flock_owner = l_ctx->open_context->flock_owner;
1008
1009	spin_lock(&state->state_lock);
1010	lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
1011	if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1012		ret = -EIO;
1013	else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1014		nfs4_stateid_copy(dst, &lsp->ls_stateid);
1015		ret = 0;
1016	}
1017	spin_unlock(&state->state_lock);
1018	nfs4_put_lock_state(lsp);
1019out:
1020	return ret;
1021}
1022
1023bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1024{
1025	bool ret;
1026	const nfs4_stateid *src;
1027	int seq;
1028
1029	do {
1030		ret = false;
1031		src = &zero_stateid;
1032		seq = read_seqbegin(&state->seqlock);
1033		if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1034			src = &state->open_stateid;
1035			ret = true;
1036		}
1037		nfs4_stateid_copy(dst, src);
1038	} while (read_seqretry(&state->seqlock, seq));
1039	return ret;
1040}
1041
1042/*
1043 * Byte-range lock aware utility to initialize the stateid of read/write
1044 * requests.
1045 */
1046int nfs4_select_rw_stateid(struct nfs4_state *state,
1047		fmode_t fmode, const struct nfs_lock_context *l_ctx,
1048		nfs4_stateid *dst, const struct cred **cred)
1049{
1050	int ret;
1051
1052	if (!nfs4_valid_open_stateid(state))
1053		return -EIO;
1054	if (cred != NULL)
1055		*cred = NULL;
1056	ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1057	if (ret == -EIO)
1058		/* A lost lock - don't even consider delegations */
1059		goto out;
1060	/* returns true if delegation stateid found and copied */
1061	if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1062		ret = 0;
1063		goto out;
1064	}
1065	if (ret != -ENOENT)
1066		/* nfs4_copy_delegation_stateid() didn't over-write
1067		 * dst, so it still has the lock stateid which we now
1068		 * choose to use.
1069		 */
1070		goto out;
1071	ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1072out:
1073	if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1074		dst->seqid = 0;
1075	return ret;
1076}
1077
1078struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1079{
1080	struct nfs_seqid *new;
1081
1082	new = kmalloc(sizeof(*new), gfp_mask);
1083	if (new == NULL)
1084		return ERR_PTR(-ENOMEM);
1085	new->sequence = counter;
1086	INIT_LIST_HEAD(&new->list);
1087	new->task = NULL;
1088	return new;
1089}
1090
1091void nfs_release_seqid(struct nfs_seqid *seqid)
1092{
1093	struct nfs_seqid_counter *sequence;
1094
1095	if (seqid == NULL || list_empty(&seqid->list))
1096		return;
1097	sequence = seqid->sequence;
1098	spin_lock(&sequence->lock);
1099	list_del_init(&seqid->list);
1100	if (!list_empty(&sequence->list)) {
1101		struct nfs_seqid *next;
1102
1103		next = list_first_entry(&sequence->list,
1104				struct nfs_seqid, list);
1105		rpc_wake_up_queued_task(&sequence->wait, next->task);
1106	}
1107	spin_unlock(&sequence->lock);
1108}
1109
1110void nfs_free_seqid(struct nfs_seqid *seqid)
1111{
1112	nfs_release_seqid(seqid);
1113	kfree(seqid);
1114}
1115
1116/*
1117 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1118 * failed with a seqid incrementing error -
1119 * see comments nfs4.h:seqid_mutating_error()
1120 */
1121static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1122{
1123	switch (status) {
1124		case 0:
1125			break;
1126		case -NFS4ERR_BAD_SEQID:
1127			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1128				return;
1129			pr_warn_ratelimited("NFS: v4 server returned a bad"
1130					" sequence-id error on an"
1131					" unconfirmed sequence %p!\n",
1132					seqid->sequence);
1133		case -NFS4ERR_STALE_CLIENTID:
1134		case -NFS4ERR_STALE_STATEID:
1135		case -NFS4ERR_BAD_STATEID:
1136		case -NFS4ERR_BADXDR:
1137		case -NFS4ERR_RESOURCE:
1138		case -NFS4ERR_NOFILEHANDLE:
1139		case -NFS4ERR_MOVED:
1140			/* Non-seqid mutating errors */
1141			return;
1142	}
1143	/*
1144	 * Note: no locking needed as we are guaranteed to be first
1145	 * on the sequence list
1146	 */
1147	seqid->sequence->counter++;
1148}
1149
1150void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1151{
1152	struct nfs4_state_owner *sp;
1153
1154	if (seqid == NULL)
1155		return;
1156
1157	sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1158	if (status == -NFS4ERR_BAD_SEQID)
1159		nfs4_reset_state_owner(sp);
1160	if (!nfs4_has_session(sp->so_server->nfs_client))
1161		nfs_increment_seqid(status, seqid);
1162}
1163
1164/*
1165 * Increment the seqid if the LOCK/LOCKU succeeded, or
1166 * failed with a seqid incrementing error -
1167 * see comments nfs4.h:seqid_mutating_error()
1168 */
1169void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1170{
1171	if (seqid != NULL)
1172		nfs_increment_seqid(status, seqid);
1173}
1174
1175int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1176{
1177	struct nfs_seqid_counter *sequence;
1178	int status = 0;
1179
1180	if (seqid == NULL)
1181		goto out;
1182	sequence = seqid->sequence;
1183	spin_lock(&sequence->lock);
1184	seqid->task = task;
1185	if (list_empty(&seqid->list))
1186		list_add_tail(&seqid->list, &sequence->list);
1187	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1188		goto unlock;
1189	rpc_sleep_on(&sequence->wait, task, NULL);
1190	status = -EAGAIN;
1191unlock:
1192	spin_unlock(&sequence->lock);
1193out:
1194	return status;
1195}
1196
1197static int nfs4_run_state_manager(void *);
1198
1199static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1200{
1201	smp_mb__before_atomic();
1202	clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1203	smp_mb__after_atomic();
1204	wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1205	rpc_wake_up(&clp->cl_rpcwaitq);
1206}
1207
1208/*
1209 * Schedule the nfs_client asynchronous state management routine
1210 */
1211void nfs4_schedule_state_manager(struct nfs_client *clp)
1212{
1213	struct task_struct *task;
1214	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1215	struct rpc_clnt *clnt = clp->cl_rpcclient;
1216	bool swapon = false;
1217
1218	set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1219
1220	if (atomic_read(&clnt->cl_swapper)) {
1221		swapon = !test_and_set_bit(NFS4CLNT_MANAGER_AVAILABLE,
1222					   &clp->cl_state);
1223		if (!swapon) {
1224			wake_up_var(&clp->cl_state);
1225			return;
1226		}
1227	}
1228
1229	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1230		return;
1231
1232	__module_get(THIS_MODULE);
1233	refcount_inc(&clp->cl_count);
1234
1235	/* The rcu_read_lock() is not strictly necessary, as the state
1236	 * manager is the only thread that ever changes the rpc_xprt
1237	 * after it's initialized.  At this point, we're single threaded. */
1238	rcu_read_lock();
1239	snprintf(buf, sizeof(buf), "%s-manager",
1240			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1241	rcu_read_unlock();
1242	task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1243	if (IS_ERR(task)) {
1244		printk(KERN_ERR "%s: kthread_run: %ld\n",
1245			__func__, PTR_ERR(task));
1246		if (!nfs_client_init_is_complete(clp))
1247			nfs_mark_client_ready(clp, PTR_ERR(task));
1248		if (swapon)
1249			clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
1250		nfs4_clear_state_manager_bit(clp);
1251		nfs_put_client(clp);
1252		module_put(THIS_MODULE);
1253	}
1254}
1255
1256/*
1257 * Schedule a lease recovery attempt
1258 */
1259void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1260{
1261	if (!clp)
1262		return;
1263	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1264		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1265	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1266			clp->cl_hostname);
1267	nfs4_schedule_state_manager(clp);
1268}
1269EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1270
1271/**
1272 * nfs4_schedule_migration_recovery - trigger migration recovery
1273 *
1274 * @server: FSID that is migrating
1275 *
1276 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1277 * value is returned.
1278 */
1279int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1280{
1281	struct nfs_client *clp = server->nfs_client;
1282
1283	if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1284		pr_err("NFS: volatile file handles not supported (server %s)\n",
1285				clp->cl_hostname);
1286		return -NFS4ERR_IO;
1287	}
1288
1289	if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1290		return -NFS4ERR_IO;
1291
1292	dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1293			__func__,
1294			(unsigned long long)server->fsid.major,
1295			(unsigned long long)server->fsid.minor,
1296			clp->cl_hostname);
1297
1298	set_bit(NFS_MIG_IN_TRANSITION,
1299			&((struct nfs_server *)server)->mig_status);
1300	set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1301
1302	nfs4_schedule_state_manager(clp);
1303	return 0;
1304}
1305EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1306
1307/**
1308 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1309 *
1310 * @clp: server to check for moved leases
1311 *
1312 */
1313void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1314{
1315	dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1316		__func__, clp->cl_clientid, clp->cl_hostname);
1317
1318	set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1319	nfs4_schedule_state_manager(clp);
1320}
1321EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1322
1323int nfs4_wait_clnt_recover(struct nfs_client *clp)
1324{
1325	int res;
1326
1327	might_sleep();
1328
1329	refcount_inc(&clp->cl_count);
1330	res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1331				 nfs_wait_bit_killable, TASK_KILLABLE);
1332	if (res)
1333		goto out;
1334	if (clp->cl_cons_state < 0)
1335		res = clp->cl_cons_state;
1336out:
1337	nfs_put_client(clp);
1338	return res;
1339}
1340
1341int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1342{
1343	unsigned int loop;
1344	int ret;
1345
1346	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1347		ret = nfs4_wait_clnt_recover(clp);
1348		if (ret != 0)
1349			break;
1350		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1351		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1352			break;
1353		nfs4_schedule_state_manager(clp);
1354		ret = -EIO;
1355	}
1356	return ret;
1357}
1358
1359/*
1360 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1361 * @clp: client to process
1362 *
1363 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1364 * resend of the SETCLIENTID and hence re-establish the
1365 * callback channel. Then return all existing delegations.
1366 */
1367static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1368{
1369	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1370	nfs_expire_all_delegations(clp);
1371	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1372			clp->cl_hostname);
1373}
1374
1375void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1376{
1377	nfs40_handle_cb_pathdown(clp);
1378	nfs4_schedule_state_manager(clp);
1379}
1380
1381static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1382{
1383
1384	if (!nfs4_valid_open_stateid(state))
1385		return 0;
1386	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1387	/* Don't recover state that expired before the reboot */
1388	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1389		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1390		return 0;
1391	}
1392	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1393	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1394	return 1;
1395}
1396
1397int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1398{
1399	if (!nfs4_valid_open_stateid(state))
1400		return 0;
1401	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1402	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1403	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1404	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1405	return 1;
1406}
1407
1408int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1409{
1410	struct nfs_client *clp = server->nfs_client;
1411
1412	if (!nfs4_state_mark_reclaim_nograce(clp, state))
1413		return -EBADF;
1414	nfs_inode_find_delegation_state_and_recover(state->inode,
1415			&state->stateid);
1416	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1417			clp->cl_hostname);
1418	nfs4_schedule_state_manager(clp);
1419	return 0;
1420}
1421EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1422
1423static struct nfs4_lock_state *
1424nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1425		const nfs4_stateid *stateid)
1426{
1427	struct nfs4_lock_state *pos;
1428
1429	list_for_each_entry(pos, &state->lock_states, ls_locks) {
1430		if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1431			continue;
1432		if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1433			return pos;
1434	}
1435	return NULL;
1436}
1437
1438static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1439		const nfs4_stateid *stateid)
1440{
1441	bool found = false;
1442
1443	if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1444		spin_lock(&state->state_lock);
1445		if (nfs_state_find_lock_state_by_stateid(state, stateid))
1446			found = true;
1447		spin_unlock(&state->state_lock);
1448	}
1449	return found;
1450}
1451
1452void nfs_inode_find_state_and_recover(struct inode *inode,
1453		const nfs4_stateid *stateid)
1454{
1455	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1456	struct nfs_inode *nfsi = NFS_I(inode);
1457	struct nfs_open_context *ctx;
1458	struct nfs4_state *state;
1459	bool found = false;
1460
1461	rcu_read_lock();
1462	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1463		state = ctx->state;
1464		if (state == NULL)
1465			continue;
1466		if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1467		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1468			found = true;
1469			continue;
1470		}
1471		if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1472		    nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1473		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1474			found = true;
1475			continue;
1476		}
1477		if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1478		    nfs4_state_mark_reclaim_nograce(clp, state))
1479			found = true;
1480	}
1481	rcu_read_unlock();
1482
1483	nfs_inode_find_delegation_state_and_recover(inode, stateid);
1484	if (found)
1485		nfs4_schedule_state_manager(clp);
1486}
1487
1488static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1489{
1490	struct inode *inode = state->inode;
1491	struct nfs_inode *nfsi = NFS_I(inode);
1492	struct nfs_open_context *ctx;
1493
1494	rcu_read_lock();
1495	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1496		if (ctx->state != state)
1497			continue;
1498		set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1499		pr_warn("NFSv4: state recovery failed for open file %pd2, "
1500				"error = %d\n", ctx->dentry, err);
1501	}
1502	rcu_read_unlock();
1503}
1504
1505static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1506{
1507	set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1508	nfs4_state_mark_open_context_bad(state, error);
1509}
1510
1511
1512static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1513{
1514	struct inode *inode = state->inode;
1515	struct nfs_inode *nfsi = NFS_I(inode);
1516	struct file_lock *fl;
1517	struct nfs4_lock_state *lsp;
1518	int status = 0;
1519	struct file_lock_context *flctx = inode->i_flctx;
1520	struct list_head *list;
1521
1522	if (flctx == NULL)
1523		return 0;
1524
1525	list = &flctx->flc_posix;
1526
1527	/* Guard against delegation returns and new lock/unlock calls */
1528	down_write(&nfsi->rwsem);
1529	spin_lock(&flctx->flc_lock);
1530restart:
1531	list_for_each_entry(fl, list, fl_list) {
1532		if (nfs_file_open_context(fl->fl_file)->state != state)
1533			continue;
1534		spin_unlock(&flctx->flc_lock);
1535		status = ops->recover_lock(state, fl);
1536		switch (status) {
1537		case 0:
1538			break;
1539		case -ETIMEDOUT:
1540		case -ESTALE:
1541		case -NFS4ERR_ADMIN_REVOKED:
1542		case -NFS4ERR_STALE_STATEID:
1543		case -NFS4ERR_BAD_STATEID:
1544		case -NFS4ERR_EXPIRED:
1545		case -NFS4ERR_NO_GRACE:
1546		case -NFS4ERR_STALE_CLIENTID:
1547		case -NFS4ERR_BADSESSION:
1548		case -NFS4ERR_BADSLOT:
1549		case -NFS4ERR_BAD_HIGH_SLOT:
1550		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1551			goto out;
1552		default:
1553			pr_err("NFS: %s: unhandled error %d\n",
1554					__func__, status);
1555			fallthrough;
1556		case -ENOMEM:
1557		case -NFS4ERR_DENIED:
1558		case -NFS4ERR_RECLAIM_BAD:
1559		case -NFS4ERR_RECLAIM_CONFLICT:
1560			lsp = fl->fl_u.nfs4_fl.owner;
1561			if (lsp)
1562				set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1563			status = 0;
1564		}
1565		spin_lock(&flctx->flc_lock);
1566	}
1567	if (list == &flctx->flc_posix) {
1568		list = &flctx->flc_flock;
1569		goto restart;
1570	}
1571	spin_unlock(&flctx->flc_lock);
1572out:
1573	up_write(&nfsi->rwsem);
1574	return status;
1575}
1576
1577#ifdef CONFIG_NFS_V4_2
1578static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1579{
1580	struct nfs4_copy_state *copy;
1581
1582	if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1583		!test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1584		return;
1585
1586	spin_lock(&sp->so_server->nfs_client->cl_lock);
1587	list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1588		if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1589				!nfs4_stateid_match_other(&state->stateid,
1590				&copy->parent_dst_state->stateid)))
1591				continue;
1592		copy->flags = 1;
1593		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1594				&state->flags)) {
1595			clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1596			complete(&copy->completion);
1597		}
1598	}
1599	list_for_each_entry(copy, &sp->so_server->ss_copies, src_copies) {
1600		if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1601				!nfs4_stateid_match_other(&state->stateid,
1602				&copy->parent_src_state->stateid)))
1603				continue;
1604		copy->flags = 1;
1605		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1606				&state->flags))
1607			complete(&copy->completion);
1608	}
1609	spin_unlock(&sp->so_server->nfs_client->cl_lock);
1610}
1611#else /* !CONFIG_NFS_V4_2 */
1612static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1613					 struct nfs4_state *state)
1614{
1615}
1616#endif /* CONFIG_NFS_V4_2 */
1617
1618static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1619				     const struct nfs4_state_recovery_ops *ops)
1620{
1621	struct nfs4_lock_state *lock;
1622	int status;
1623
1624	status = ops->recover_open(sp, state);
1625	if (status < 0)
1626		return status;
1627
1628	status = nfs4_reclaim_locks(state, ops);
1629	if (status < 0)
1630		return status;
1631
1632	if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1633		spin_lock(&state->state_lock);
1634		list_for_each_entry(lock, &state->lock_states, ls_locks) {
1635			trace_nfs4_state_lock_reclaim(state, lock);
1636			if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1637				pr_warn_ratelimited("NFS: %s: Lock reclaim failed!\n", __func__);
1638		}
1639		spin_unlock(&state->state_lock);
1640	}
1641
1642	nfs42_complete_copies(sp, state);
1643	clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1644	return status;
1645}
1646
1647static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1648{
1649	struct nfs4_state *state;
1650	unsigned int loop = 0;
1651	int status = 0;
1652#ifdef CONFIG_NFS_V4_2
1653	bool found_ssc_copy_state = false;
1654#endif /* CONFIG_NFS_V4_2 */
1655
1656	/* Note: we rely on the sp->so_states list being ordered
1657	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1658	 * states first.
1659	 * This is needed to ensure that the server won't give us any
1660	 * read delegations that we have to return if, say, we are
1661	 * recovering after a network partition or a reboot from a
1662	 * server that doesn't support a grace period.
1663	 */
1664	spin_lock(&sp->so_lock);
1665	raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1666restart:
1667	list_for_each_entry(state, &sp->so_states, open_states) {
1668		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1669			continue;
1670		if (!nfs4_valid_open_stateid(state))
1671			continue;
1672		if (state->state == 0)
1673			continue;
1674#ifdef CONFIG_NFS_V4_2
1675		if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1676			nfs4_state_mark_recovery_failed(state, -EIO);
1677			found_ssc_copy_state = true;
1678			continue;
1679		}
1680#endif /* CONFIG_NFS_V4_2 */
1681		refcount_inc(&state->count);
1682		spin_unlock(&sp->so_lock);
1683		status = __nfs4_reclaim_open_state(sp, state, ops);
1684
1685		switch (status) {
1686		default:
1687			if (status >= 0) {
1688				loop = 0;
1689				break;
1690			}
1691			printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1692			fallthrough;
1693		case -ENOENT:
1694		case -ENOMEM:
1695		case -EACCES:
1696		case -EROFS:
1697		case -EIO:
1698		case -ESTALE:
1699			/* Open state on this file cannot be recovered */
1700			nfs4_state_mark_recovery_failed(state, status);
1701			break;
1702		case -EAGAIN:
1703			ssleep(1);
1704			if (loop++ < 10) {
1705				set_bit(ops->state_flag_bit, &state->flags);
1706				break;
1707			}
1708			fallthrough;
1709		case -NFS4ERR_ADMIN_REVOKED:
1710		case -NFS4ERR_STALE_STATEID:
1711		case -NFS4ERR_OLD_STATEID:
1712		case -NFS4ERR_BAD_STATEID:
1713		case -NFS4ERR_RECLAIM_BAD:
1714		case -NFS4ERR_RECLAIM_CONFLICT:
1715			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1716			break;
1717		case -NFS4ERR_EXPIRED:
1718		case -NFS4ERR_NO_GRACE:
1719			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1720			fallthrough;
1721		case -NFS4ERR_STALE_CLIENTID:
1722		case -NFS4ERR_BADSESSION:
1723		case -NFS4ERR_BADSLOT:
1724		case -NFS4ERR_BAD_HIGH_SLOT:
1725		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1726		case -ETIMEDOUT:
1727			goto out_err;
1728		}
1729		nfs4_put_open_state(state);
1730		spin_lock(&sp->so_lock);
1731		goto restart;
1732	}
1733	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1734	spin_unlock(&sp->so_lock);
1735#ifdef CONFIG_NFS_V4_2
1736	if (found_ssc_copy_state)
1737		return -EIO;
1738#endif /* CONFIG_NFS_V4_2 */
1739	return 0;
1740out_err:
1741	nfs4_put_open_state(state);
1742	spin_lock(&sp->so_lock);
1743	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1744	spin_unlock(&sp->so_lock);
1745	return status;
1746}
1747
1748static void nfs4_clear_open_state(struct nfs4_state *state)
1749{
1750	struct nfs4_lock_state *lock;
1751
1752	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1753	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1754	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1755	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1756	spin_lock(&state->state_lock);
1757	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1758		lock->ls_seqid.flags = 0;
1759		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1760	}
1761	spin_unlock(&state->state_lock);
1762}
1763
1764static void nfs4_reset_seqids(struct nfs_server *server,
1765	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1766{
1767	struct nfs_client *clp = server->nfs_client;
1768	struct nfs4_state_owner *sp;
1769	struct rb_node *pos;
1770	struct nfs4_state *state;
1771
1772	spin_lock(&clp->cl_lock);
1773	for (pos = rb_first(&server->state_owners);
1774	     pos != NULL;
1775	     pos = rb_next(pos)) {
1776		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1777		sp->so_seqid.flags = 0;
1778		spin_lock(&sp->so_lock);
1779		list_for_each_entry(state, &sp->so_states, open_states) {
1780			if (mark_reclaim(clp, state))
1781				nfs4_clear_open_state(state);
1782		}
1783		spin_unlock(&sp->so_lock);
1784	}
1785	spin_unlock(&clp->cl_lock);
1786}
1787
1788static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1789	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1790{
1791	struct nfs_server *server;
1792
1793	rcu_read_lock();
1794	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1795		nfs4_reset_seqids(server, mark_reclaim);
1796	rcu_read_unlock();
1797}
1798
1799static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1800{
1801	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1802	/* Mark all delegations for reclaim */
1803	nfs_delegation_mark_reclaim(clp);
1804	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1805}
1806
1807static int nfs4_reclaim_complete(struct nfs_client *clp,
1808				 const struct nfs4_state_recovery_ops *ops,
1809				 const struct cred *cred)
1810{
1811	/* Notify the server we're done reclaiming our state */
1812	if (ops->reclaim_complete)
1813		return ops->reclaim_complete(clp, cred);
1814	return 0;
1815}
1816
1817static void nfs4_clear_reclaim_server(struct nfs_server *server)
1818{
1819	struct nfs_client *clp = server->nfs_client;
1820	struct nfs4_state_owner *sp;
1821	struct rb_node *pos;
1822	struct nfs4_state *state;
1823
1824	spin_lock(&clp->cl_lock);
1825	for (pos = rb_first(&server->state_owners);
1826	     pos != NULL;
1827	     pos = rb_next(pos)) {
1828		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1829		spin_lock(&sp->so_lock);
1830		list_for_each_entry(state, &sp->so_states, open_states) {
1831			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1832						&state->flags))
1833				continue;
1834			nfs4_state_mark_reclaim_nograce(clp, state);
1835		}
1836		spin_unlock(&sp->so_lock);
1837	}
1838	spin_unlock(&clp->cl_lock);
1839}
1840
1841static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1842{
1843	struct nfs_server *server;
1844
1845	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1846		return 0;
1847
1848	rcu_read_lock();
1849	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1850		nfs4_clear_reclaim_server(server);
1851	rcu_read_unlock();
1852
1853	nfs_delegation_reap_unclaimed(clp);
1854	return 1;
1855}
1856
1857static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1858{
1859	const struct nfs4_state_recovery_ops *ops;
1860	const struct cred *cred;
1861	int err;
1862
1863	if (!nfs4_state_clear_reclaim_reboot(clp))
1864		return;
1865	ops = clp->cl_mvops->reboot_recovery_ops;
1866	cred = nfs4_get_clid_cred(clp);
1867	err = nfs4_reclaim_complete(clp, ops, cred);
1868	put_cred(cred);
1869	if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1870		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1871}
1872
1873static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1874{
1875	nfs_mark_test_expired_all_delegations(clp);
1876	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1877}
1878
1879static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1880{
1881	switch (error) {
1882	case 0:
1883		break;
1884	case -NFS4ERR_CB_PATH_DOWN:
1885		nfs40_handle_cb_pathdown(clp);
1886		break;
1887	case -NFS4ERR_NO_GRACE:
1888		nfs4_state_end_reclaim_reboot(clp);
1889		break;
1890	case -NFS4ERR_STALE_CLIENTID:
1891		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1892		nfs4_state_start_reclaim_reboot(clp);
1893		break;
1894	case -NFS4ERR_EXPIRED:
1895		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1896		nfs4_state_start_reclaim_nograce(clp);
1897		break;
1898	case -NFS4ERR_BADSESSION:
1899	case -NFS4ERR_BADSLOT:
1900	case -NFS4ERR_BAD_HIGH_SLOT:
1901	case -NFS4ERR_DEADSESSION:
1902	case -NFS4ERR_SEQ_FALSE_RETRY:
1903	case -NFS4ERR_SEQ_MISORDERED:
1904		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1905		/* Zero session reset errors */
1906		break;
1907	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1908		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1909		break;
1910	default:
1911		dprintk("%s: failed to handle error %d for server %s\n",
1912				__func__, error, clp->cl_hostname);
1913		return error;
1914	}
1915	dprintk("%s: handled error %d for server %s\n", __func__, error,
1916			clp->cl_hostname);
1917	return 0;
1918}
1919
1920static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1921{
1922	struct nfs4_state_owner *sp;
1923	struct nfs_server *server;
1924	struct rb_node *pos;
1925	LIST_HEAD(freeme);
1926	int status = 0;
1927
1928restart:
1929	rcu_read_lock();
1930	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1931		nfs4_purge_state_owners(server, &freeme);
1932		spin_lock(&clp->cl_lock);
1933		for (pos = rb_first(&server->state_owners);
1934		     pos != NULL;
1935		     pos = rb_next(pos)) {
1936			sp = rb_entry(pos,
1937				struct nfs4_state_owner, so_server_node);
1938			if (!test_and_clear_bit(ops->owner_flag_bit,
1939							&sp->so_flags))
1940				continue;
1941			if (!atomic_inc_not_zero(&sp->so_count))
1942				continue;
1943			spin_unlock(&clp->cl_lock);
1944			rcu_read_unlock();
1945
1946			status = nfs4_reclaim_open_state(sp, ops);
1947			if (status < 0) {
1948				set_bit(ops->owner_flag_bit, &sp->so_flags);
1949				nfs4_put_state_owner(sp);
1950				status = nfs4_recovery_handle_error(clp, status);
1951				return (status != 0) ? status : -EAGAIN;
1952			}
1953
1954			nfs4_put_state_owner(sp);
1955			goto restart;
1956		}
1957		spin_unlock(&clp->cl_lock);
1958	}
1959	rcu_read_unlock();
1960	nfs4_free_state_owners(&freeme);
1961	return 0;
1962}
1963
1964static int nfs4_check_lease(struct nfs_client *clp)
1965{
1966	const struct cred *cred;
1967	const struct nfs4_state_maintenance_ops *ops =
1968		clp->cl_mvops->state_renewal_ops;
1969	int status;
1970
1971	/* Is the client already known to have an expired lease? */
1972	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1973		return 0;
1974	cred = ops->get_state_renewal_cred(clp);
1975	if (cred == NULL) {
1976		cred = nfs4_get_clid_cred(clp);
1977		status = -ENOKEY;
1978		if (cred == NULL)
1979			goto out;
1980	}
1981	status = ops->renew_lease(clp, cred);
1982	put_cred(cred);
1983	if (status == -ETIMEDOUT) {
1984		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1985		return 0;
1986	}
1987out:
1988	return nfs4_recovery_handle_error(clp, status);
1989}
1990
1991/* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1992 * and for recoverable errors on EXCHANGE_ID for v4.1
1993 */
1994static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1995{
1996	switch (status) {
1997	case -NFS4ERR_SEQ_MISORDERED:
1998		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1999			return -ESERVERFAULT;
2000		/* Lease confirmation error: retry after purging the lease */
2001		ssleep(1);
2002		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2003		break;
2004	case -NFS4ERR_STALE_CLIENTID:
2005		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2006		nfs4_state_start_reclaim_reboot(clp);
2007		break;
2008	case -NFS4ERR_CLID_INUSE:
2009		pr_err("NFS: Server %s reports our clientid is in use\n",
2010			clp->cl_hostname);
2011		nfs_mark_client_ready(clp, -EPERM);
2012		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2013		return -EPERM;
2014	case -EACCES:
2015	case -NFS4ERR_DELAY:
2016	case -EAGAIN:
2017		ssleep(1);
2018		break;
2019
2020	case -NFS4ERR_MINOR_VERS_MISMATCH:
2021		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2022			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2023		dprintk("%s: exit with error %d for server %s\n",
2024				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
2025		return -EPROTONOSUPPORT;
2026	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2027				 * in nfs4_exchange_id */
2028	default:
2029		dprintk("%s: exit with error %d for server %s\n", __func__,
2030				status, clp->cl_hostname);
2031		return status;
2032	}
2033	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2034	dprintk("%s: handled error %d for server %s\n", __func__, status,
2035			clp->cl_hostname);
2036	return 0;
2037}
2038
2039static int nfs4_establish_lease(struct nfs_client *clp)
2040{
2041	const struct cred *cred;
2042	const struct nfs4_state_recovery_ops *ops =
2043		clp->cl_mvops->reboot_recovery_ops;
2044	int status;
2045
2046	status = nfs4_begin_drain_session(clp);
2047	if (status != 0)
2048		return status;
2049	cred = nfs4_get_clid_cred(clp);
2050	if (cred == NULL)
2051		return -ENOENT;
2052	status = ops->establish_clid(clp, cred);
2053	put_cred(cred);
2054	if (status != 0)
2055		return status;
2056	pnfs_destroy_all_layouts(clp);
2057	return 0;
2058}
2059
2060/*
2061 * Returns zero or a negative errno.  NFS4ERR values are converted
2062 * to local errno values.
2063 */
2064static int nfs4_reclaim_lease(struct nfs_client *clp)
2065{
2066	int status;
2067
2068	status = nfs4_establish_lease(clp);
2069	if (status < 0)
2070		return nfs4_handle_reclaim_lease_error(clp, status);
2071	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2072		nfs4_state_start_reclaim_nograce(clp);
2073	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2074		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2075	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2076	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2077	return 0;
2078}
2079
2080static int nfs4_purge_lease(struct nfs_client *clp)
2081{
2082	int status;
2083
2084	status = nfs4_establish_lease(clp);
2085	if (status < 0)
2086		return nfs4_handle_reclaim_lease_error(clp, status);
2087	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2088	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2089	nfs4_state_start_reclaim_nograce(clp);
2090	return 0;
2091}
2092
2093/*
2094 * Try remote migration of one FSID from a source server to a
2095 * destination server.  The source server provides a list of
2096 * potential destinations.
2097 *
2098 * Returns zero or a negative NFS4ERR status code.
2099 */
2100static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2101{
2102	struct nfs_client *clp = server->nfs_client;
2103	struct nfs4_fs_locations *locations = NULL;
2104	struct inode *inode;
2105	struct page *page;
2106	int status, result;
2107
2108	dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2109			(unsigned long long)server->fsid.major,
2110			(unsigned long long)server->fsid.minor,
2111			clp->cl_hostname);
2112
2113	result = 0;
2114	page = alloc_page(GFP_KERNEL);
2115	locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2116	if (page == NULL || locations == NULL) {
2117		dprintk("<-- %s: no memory\n", __func__);
2118		goto out;
2119	}
2120
2121	inode = d_inode(server->super->s_root);
2122	result = nfs4_proc_get_locations(inode, locations, page, cred);
2123	if (result) {
2124		dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2125			__func__, result);
2126		goto out;
2127	}
2128
2129	result = -NFS4ERR_NXIO;
2130	if (!locations->nlocations)
2131		goto out;
2132
2133	if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2134		dprintk("<-- %s: No fs_locations data, migration skipped\n",
2135			__func__);
2136		goto out;
2137	}
2138
2139	status = nfs4_begin_drain_session(clp);
2140	if (status != 0) {
2141		result = status;
2142		goto out;
2143	}
2144
2145	status = nfs4_replace_transport(server, locations);
2146	if (status != 0) {
2147		dprintk("<-- %s: failed to replace transport: %d\n",
2148			__func__, status);
2149		goto out;
2150	}
2151
2152	result = 0;
2153	dprintk("<-- %s: migration succeeded\n", __func__);
2154
2155out:
2156	if (page != NULL)
2157		__free_page(page);
2158	kfree(locations);
2159	if (result) {
2160		pr_err("NFS: migration recovery failed (server %s)\n",
2161				clp->cl_hostname);
2162		set_bit(NFS_MIG_FAILED, &server->mig_status);
2163	}
2164	return result;
2165}
2166
2167/*
2168 * Returns zero or a negative NFS4ERR status code.
2169 */
2170static int nfs4_handle_migration(struct nfs_client *clp)
2171{
2172	const struct nfs4_state_maintenance_ops *ops =
2173				clp->cl_mvops->state_renewal_ops;
2174	struct nfs_server *server;
2175	const struct cred *cred;
2176
2177	dprintk("%s: migration reported on \"%s\"\n", __func__,
2178			clp->cl_hostname);
2179
2180	cred = ops->get_state_renewal_cred(clp);
2181	if (cred == NULL)
2182		return -NFS4ERR_NOENT;
2183
2184	clp->cl_mig_gen++;
2185restart:
2186	rcu_read_lock();
2187	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2188		int status;
2189
2190		if (server->mig_gen == clp->cl_mig_gen)
2191			continue;
2192		server->mig_gen = clp->cl_mig_gen;
2193
2194		if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2195						&server->mig_status))
2196			continue;
2197
2198		rcu_read_unlock();
2199		status = nfs4_try_migration(server, cred);
2200		if (status < 0) {
2201			put_cred(cred);
2202			return status;
2203		}
2204		goto restart;
2205	}
2206	rcu_read_unlock();
2207	put_cred(cred);
2208	return 0;
2209}
2210
2211/*
2212 * Test each nfs_server on the clp's cl_superblocks list to see
2213 * if it's moved to another server.  Stop when the server no longer
2214 * returns NFS4ERR_LEASE_MOVED.
2215 */
2216static int nfs4_handle_lease_moved(struct nfs_client *clp)
2217{
2218	const struct nfs4_state_maintenance_ops *ops =
2219				clp->cl_mvops->state_renewal_ops;
2220	struct nfs_server *server;
2221	const struct cred *cred;
2222
2223	dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2224			clp->cl_hostname);
2225
2226	cred = ops->get_state_renewal_cred(clp);
2227	if (cred == NULL)
2228		return -NFS4ERR_NOENT;
2229
2230	clp->cl_mig_gen++;
2231restart:
2232	rcu_read_lock();
2233	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2234		struct inode *inode;
2235		int status;
2236
2237		if (server->mig_gen == clp->cl_mig_gen)
2238			continue;
2239		server->mig_gen = clp->cl_mig_gen;
2240
2241		rcu_read_unlock();
2242
2243		inode = d_inode(server->super->s_root);
2244		status = nfs4_proc_fsid_present(inode, cred);
2245		if (status != -NFS4ERR_MOVED)
2246			goto restart;	/* wasn't this one */
2247		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2248			goto restart;	/* there are more */
2249		goto out;
2250	}
2251	rcu_read_unlock();
2252
2253out:
2254	put_cred(cred);
2255	return 0;
2256}
2257
2258/**
2259 * nfs4_discover_server_trunking - Detect server IP address trunking
2260 *
2261 * @clp: nfs_client under test
2262 * @result: OUT: found nfs_client, or clp
2263 *
2264 * Returns zero or a negative errno.  If zero is returned,
2265 * an nfs_client pointer is planted in "result".
2266 *
2267 * Note: since we are invoked in process context, and
2268 * not from inside the state manager, we cannot use
2269 * nfs4_handle_reclaim_lease_error().
2270 */
2271int nfs4_discover_server_trunking(struct nfs_client *clp,
2272				  struct nfs_client **result)
2273{
2274	const struct nfs4_state_recovery_ops *ops =
2275				clp->cl_mvops->reboot_recovery_ops;
2276	struct rpc_clnt *clnt;
2277	const struct cred *cred;
2278	int i, status;
2279
2280	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2281
2282	clnt = clp->cl_rpcclient;
2283	i = 0;
2284
2285	mutex_lock(&nfs_clid_init_mutex);
2286again:
2287	status  = -ENOENT;
2288	cred = nfs4_get_clid_cred(clp);
2289	if (cred == NULL)
2290		goto out_unlock;
2291
2292	status = ops->detect_trunking(clp, result, cred);
2293	put_cred(cred);
2294	switch (status) {
2295	case 0:
2296	case -EINTR:
2297	case -ERESTARTSYS:
2298		break;
2299	case -ETIMEDOUT:
2300		if (clnt->cl_softrtry)
2301			break;
2302		fallthrough;
2303	case -NFS4ERR_DELAY:
2304	case -EAGAIN:
2305		ssleep(1);
2306		fallthrough;
2307	case -NFS4ERR_STALE_CLIENTID:
2308		dprintk("NFS: %s after status %d, retrying\n",
2309			__func__, status);
2310		goto again;
2311	case -EACCES:
2312		if (i++ == 0) {
2313			nfs4_root_machine_cred(clp);
2314			goto again;
2315		}
2316		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2317			break;
2318		fallthrough;
2319	case -NFS4ERR_CLID_INUSE:
2320	case -NFS4ERR_WRONGSEC:
2321		/* No point in retrying if we already used RPC_AUTH_UNIX */
2322		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2323			status = -EPERM;
2324			break;
2325		}
2326		clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2327		if (IS_ERR(clnt)) {
2328			status = PTR_ERR(clnt);
2329			break;
2330		}
2331		/* Note: this is safe because we haven't yet marked the
2332		 * client as ready, so we are the only user of
2333		 * clp->cl_rpcclient
2334		 */
2335		clnt = xchg(&clp->cl_rpcclient, clnt);
2336		rpc_shutdown_client(clnt);
2337		clnt = clp->cl_rpcclient;
2338		goto again;
2339
2340	case -NFS4ERR_MINOR_VERS_MISMATCH:
2341		status = -EPROTONOSUPPORT;
2342		break;
2343
2344	case -EKEYEXPIRED:
2345	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2346				 * in nfs4_exchange_id */
2347		status = -EKEYEXPIRED;
2348		break;
2349	default:
2350		pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2351				__func__, status);
2352		status = -EIO;
2353	}
2354
2355out_unlock:
2356	mutex_unlock(&nfs_clid_init_mutex);
2357	dprintk("NFS: %s: status = %d\n", __func__, status);
2358	return status;
2359}
2360
2361#ifdef CONFIG_NFS_V4_1
2362void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2363{
2364	struct nfs_client *clp = session->clp;
2365
2366	switch (err) {
2367	default:
2368		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2369		break;
2370	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2371		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2372	}
2373	nfs4_schedule_state_manager(clp);
2374}
2375EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2376
2377void nfs41_notify_server(struct nfs_client *clp)
2378{
2379	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
2380	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2381	nfs4_schedule_state_manager(clp);
2382}
2383
2384static void nfs4_reset_all_state(struct nfs_client *clp)
2385{
2386	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2387		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2388		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2389		nfs4_state_start_reclaim_nograce(clp);
2390		dprintk("%s: scheduling reset of all state for server %s!\n",
2391				__func__, clp->cl_hostname);
2392		nfs4_schedule_state_manager(clp);
2393	}
2394}
2395
2396static void nfs41_handle_server_reboot(struct nfs_client *clp)
2397{
2398	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2399		nfs4_state_start_reclaim_reboot(clp);
2400		dprintk("%s: server %s rebooted!\n", __func__,
2401				clp->cl_hostname);
2402		nfs4_schedule_state_manager(clp);
2403	}
2404}
2405
2406static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2407{
2408	nfs4_reset_all_state(clp);
2409	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2410}
2411
2412static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2413{
2414	nfs4_state_start_reclaim_nograce(clp);
2415	nfs4_schedule_state_manager(clp);
2416
2417	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2418}
2419
2420static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2421{
2422	/* FIXME: For now, we destroy all layouts. */
2423	pnfs_destroy_all_layouts(clp);
2424	nfs_test_expired_all_delegations(clp);
2425	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2426			clp->cl_hostname);
2427}
2428
2429static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2430{
2431	set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2432	nfs4_schedule_state_manager(clp);
2433
2434	dprintk("%s: server %s declared a backchannel fault\n", __func__,
2435			clp->cl_hostname);
2436}
2437
2438static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2439{
2440	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2441		&clp->cl_state) == 0)
2442		nfs4_schedule_state_manager(clp);
2443}
2444
2445void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2446		bool recovery)
2447{
2448	if (!flags)
2449		return;
2450
2451	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2452		__func__, clp->cl_hostname, clp->cl_clientid, flags);
2453	/*
2454	 * If we're called from the state manager thread, then assume we're
2455	 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2456	 * Those flags are expected to remain set until we're done
2457	 * recovering (see RFC5661, section 18.46.3).
2458	 */
2459	if (recovery)
2460		goto out_recovery;
2461
2462	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2463		nfs41_handle_server_reboot(clp);
2464	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2465		nfs41_handle_all_state_revoked(clp);
2466	if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2467			    SEQ4_STATUS_ADMIN_STATE_REVOKED))
2468		nfs41_handle_some_state_revoked(clp);
2469	if (flags & SEQ4_STATUS_LEASE_MOVED)
2470		nfs4_schedule_lease_moved_recovery(clp);
2471	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2472		nfs41_handle_recallable_state_revoked(clp);
2473out_recovery:
2474	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2475		nfs41_handle_backchannel_fault(clp);
2476	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2477				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2478		nfs41_handle_cb_path_down(clp);
2479}
2480
2481static int nfs4_reset_session(struct nfs_client *clp)
2482{
2483	const struct cred *cred;
2484	int status;
2485
2486	if (!nfs4_has_session(clp))
2487		return 0;
2488	status = nfs4_begin_drain_session(clp);
2489	if (status != 0)
2490		return status;
2491	cred = nfs4_get_clid_cred(clp);
2492	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2493	switch (status) {
2494	case 0:
2495	case -NFS4ERR_BADSESSION:
2496	case -NFS4ERR_DEADSESSION:
2497		break;
2498	case -NFS4ERR_BACK_CHAN_BUSY:
2499	case -NFS4ERR_DELAY:
2500		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2501		status = 0;
2502		ssleep(1);
2503		goto out;
2504	default:
2505		status = nfs4_recovery_handle_error(clp, status);
2506		goto out;
2507	}
2508
2509	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2510	status = nfs4_proc_create_session(clp, cred);
2511	if (status) {
2512		dprintk("%s: session reset failed with status %d for server %s!\n",
2513			__func__, status, clp->cl_hostname);
2514		status = nfs4_handle_reclaim_lease_error(clp, status);
2515		goto out;
2516	}
2517	nfs41_finish_session_reset(clp);
2518	dprintk("%s: session reset was successful for server %s!\n",
2519			__func__, clp->cl_hostname);
2520out:
2521	put_cred(cred);
2522	return status;
2523}
2524
2525static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2526{
2527	const struct cred *cred;
2528	int ret;
2529
2530	if (!nfs4_has_session(clp))
2531		return 0;
2532	ret = nfs4_begin_drain_session(clp);
2533	if (ret != 0)
2534		return ret;
2535	cred = nfs4_get_clid_cred(clp);
2536	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2537	put_cred(cred);
2538	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2539	switch (ret) {
2540	case 0:
2541		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2542			__func__, clp->cl_hostname);
2543		break;
2544	case -NFS4ERR_DELAY:
2545		ssleep(1);
2546		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2547		break;
2548	default:
2549		return nfs4_recovery_handle_error(clp, ret);
2550	}
2551	return 0;
2552}
2553
2554static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2555{
2556	int iomode = 0;
2557
2558	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &clp->cl_state))
2559		iomode += IOMODE_READ;
2560	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &clp->cl_state))
2561		iomode += IOMODE_RW;
2562	/* Note: IOMODE_READ + IOMODE_RW == IOMODE_ANY */
2563	if (iomode) {
2564		pnfs_layout_return_unused_byclid(clp, iomode);
2565		set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2566	}
2567}
2568#else /* CONFIG_NFS_V4_1 */
2569static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2570
2571static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2572{
2573	return 0;
2574}
2575
2576static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2577{
2578}
2579#endif /* CONFIG_NFS_V4_1 */
2580
2581static void nfs4_state_manager(struct nfs_client *clp)
2582{
2583	unsigned int memflags;
2584	int status = 0;
2585	const char *section = "", *section_sep = "";
2586
2587	/*
2588	 * State recovery can deadlock if the direct reclaim code tries
2589	 * start NFS writeback. So ensure memory allocations are all
2590	 * GFP_NOFS.
2591	 */
2592	memflags = memalloc_nofs_save();
2593
2594	/* Ensure exclusive access to NFSv4 state */
2595	do {
2596		trace_nfs4_state_mgr(clp);
2597		clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2598		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2599			section = "purge state";
2600			status = nfs4_purge_lease(clp);
2601			if (status < 0)
2602				goto out_error;
2603			continue;
2604		}
2605
2606		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2607			section = "lease expired";
2608			/* We're going to have to re-establish a clientid */
2609			status = nfs4_reclaim_lease(clp);
2610			if (status < 0)
2611				goto out_error;
2612			continue;
2613		}
2614
2615		/* Initialize or reset the session */
2616		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2617			section = "reset session";
2618			status = nfs4_reset_session(clp);
2619			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2620				continue;
2621			if (status < 0)
2622				goto out_error;
2623		}
2624
2625		/* Send BIND_CONN_TO_SESSION */
2626		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2627				&clp->cl_state)) {
2628			section = "bind conn to session";
2629			status = nfs4_bind_conn_to_session(clp);
2630			if (status < 0)
2631				goto out_error;
2632			continue;
2633		}
2634
2635		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2636			section = "check lease";
2637			status = nfs4_check_lease(clp);
2638			if (status < 0)
2639				goto out_error;
2640			continue;
2641		}
2642
2643		if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2644			section = "migration";
2645			status = nfs4_handle_migration(clp);
2646			if (status < 0)
2647				goto out_error;
2648		}
2649
2650		if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2651			section = "lease moved";
2652			status = nfs4_handle_lease_moved(clp);
2653			if (status < 0)
2654				goto out_error;
2655		}
2656
2657		/* First recover reboot state... */
2658		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2659			section = "reclaim reboot";
2660			status = nfs4_do_reclaim(clp,
2661				clp->cl_mvops->reboot_recovery_ops);
2662			if (status == -EAGAIN)
2663				continue;
2664			if (status < 0)
2665				goto out_error;
2666			nfs4_state_end_reclaim_reboot(clp);
2667			continue;
2668		}
2669
2670		/* Detect expired delegations... */
2671		if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2672			section = "detect expired delegations";
2673			nfs_reap_expired_delegations(clp);
2674			continue;
2675		}
2676
2677		/* Now recover expired state... */
2678		if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2679			section = "reclaim nograce";
2680			status = nfs4_do_reclaim(clp,
2681				clp->cl_mvops->nograce_recovery_ops);
2682			if (status == -EAGAIN)
2683				continue;
2684			if (status < 0)
2685				goto out_error;
2686			clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2687		}
2688
2689		memalloc_nofs_restore(memflags);
2690		nfs4_end_drain_session(clp);
2691		nfs4_clear_state_manager_bit(clp);
2692
2693		if (test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2694		    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING,
2695				      &clp->cl_state)) {
2696			memflags = memalloc_nofs_save();
2697			continue;
2698		}
2699
2700		if (!test_and_set_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state)) {
2701			if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2702				nfs_client_return_marked_delegations(clp);
2703				set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2704			}
2705			nfs4_layoutreturn_any_run(clp);
2706			clear_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state);
2707		}
2708
2709		return;
2710
2711	} while (refcount_read(&clp->cl_count) > 1 && !signalled());
2712	goto out_drain;
2713
2714out_error:
2715	if (strlen(section))
2716		section_sep = ": ";
2717	trace_nfs4_state_mgr_failed(clp, section, status);
2718	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2719			" with error %d\n", section_sep, section,
2720			clp->cl_hostname, -status);
2721	ssleep(1);
2722out_drain:
2723	memalloc_nofs_restore(memflags);
2724	nfs4_end_drain_session(clp);
2725	nfs4_clear_state_manager_bit(clp);
2726}
2727
2728static int nfs4_run_state_manager(void *ptr)
2729{
2730	struct nfs_client *clp = ptr;
2731	struct rpc_clnt *cl = clp->cl_rpcclient;
2732
2733	while (cl != cl->cl_parent)
2734		cl = cl->cl_parent;
2735
2736	allow_signal(SIGKILL);
2737again:
2738	nfs4_state_manager(clp);
2739
2740	if (test_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state) &&
2741	    !test_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state)) {
2742		wait_var_event_interruptible(&clp->cl_state,
2743					     test_bit(NFS4CLNT_RUN_MANAGER,
2744						      &clp->cl_state));
2745		if (!atomic_read(&cl->cl_swapper))
2746			clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2747		if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2748		    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2749			goto again;
2750		/* Either no longer a swapper, or were signalled */
2751		clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2752	}
2753
2754	if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2755	    test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2756	    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2757		goto again;
2758
2759	nfs_put_client(clp);
2760	module_put_and_exit(0);
2761	return 0;
2762}
2763
2764/*
2765 * Local variables:
2766 *  c-basic-offset: 8
2767 * End:
2768 */
2769