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
53#include <linux/sunrpc/clnt.h>
54
55#include "nfs4_fs.h"
56#include "callback.h"
57#include "delegation.h"
58#include "internal.h"
59#include "nfs4idmap.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
68const nfs4_stateid zero_stateid;
69static DEFINE_MUTEX(nfs_clid_init_mutex);
70
71int 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);
93do_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);
99out:
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 */
116int nfs40_discover_server_trunking(struct nfs_client *clp,
117				   struct nfs_client **result,
118				   struct rpc_cred *cred)
119{
120	struct nfs4_setclientid_res clid = {
121		.clientid = clp->cl_clientid,
122		.confirm = clp->cl_confirm,
123	};
124	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
125	unsigned short port;
126	int status;
127
128	port = nn->nfs_callback_tcpport;
129	if (clp->cl_addr.ss_family == AF_INET6)
130		port = nn->nfs_callback_tcpport6;
131
132	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
133	if (status != 0)
134		goto out;
135	clp->cl_clientid = clid.clientid;
136	clp->cl_confirm = clid.confirm;
137
138	status = nfs40_walk_client_list(clp, result, cred);
139	if (status == 0) {
140		/* Sustain the lease, even if it's empty.  If the clientid4
141		 * goes stale it's of no use for trunking discovery. */
142		nfs4_schedule_state_renewal(*result);
143	}
144out:
145	return status;
146}
147
148struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
149{
150	struct rpc_cred *cred = NULL;
151
152	if (clp->cl_machine_cred != NULL)
153		cred = get_rpccred(clp->cl_machine_cred);
154	return cred;
155}
156
157static void nfs4_root_machine_cred(struct nfs_client *clp)
158{
159	struct rpc_cred *cred, *new;
160
161	new = rpc_lookup_machine_cred(NULL);
162	spin_lock(&clp->cl_lock);
163	cred = clp->cl_machine_cred;
164	clp->cl_machine_cred = new;
165	spin_unlock(&clp->cl_lock);
166	if (cred != NULL)
167		put_rpccred(cred);
168}
169
170static struct rpc_cred *
171nfs4_get_renew_cred_server_locked(struct nfs_server *server)
172{
173	struct rpc_cred *cred = NULL;
174	struct nfs4_state_owner *sp;
175	struct rb_node *pos;
176
177	for (pos = rb_first(&server->state_owners);
178	     pos != NULL;
179	     pos = rb_next(pos)) {
180		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
181		if (list_empty(&sp->so_states))
182			continue;
183		cred = get_rpccred(sp->so_cred);
184		break;
185	}
186	return cred;
187}
188
189/**
190 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
191 * @clp: client state handle
192 *
193 * Returns an rpc_cred with reference count bumped, or NULL.
194 * Caller must hold clp->cl_lock.
195 */
196struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
197{
198	struct rpc_cred *cred = NULL;
199	struct nfs_server *server;
200
201	/* Use machine credentials if available */
202	cred = nfs4_get_machine_cred_locked(clp);
203	if (cred != NULL)
204		goto out;
205
206	rcu_read_lock();
207	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
208		cred = nfs4_get_renew_cred_server_locked(server);
209		if (cred != NULL)
210			break;
211	}
212	rcu_read_unlock();
213
214out:
215	return cred;
216}
217
218static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
219{
220	if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
221		spin_lock(&tbl->slot_tbl_lock);
222		nfs41_wake_slot_table(tbl);
223		spin_unlock(&tbl->slot_tbl_lock);
224	}
225}
226
227static void nfs4_end_drain_session(struct nfs_client *clp)
228{
229	struct nfs4_session *ses = clp->cl_session;
230
231	if (clp->cl_slot_tbl) {
232		nfs4_end_drain_slot_table(clp->cl_slot_tbl);
233		return;
234	}
235
236	if (ses != NULL) {
237		nfs4_end_drain_slot_table(&ses->bc_slot_table);
238		nfs4_end_drain_slot_table(&ses->fc_slot_table);
239	}
240}
241
242static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
243{
244	set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
245	spin_lock(&tbl->slot_tbl_lock);
246	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
247		reinit_completion(&tbl->complete);
248		spin_unlock(&tbl->slot_tbl_lock);
249		return wait_for_completion_interruptible(&tbl->complete);
250	}
251	spin_unlock(&tbl->slot_tbl_lock);
252	return 0;
253}
254
255static int nfs4_begin_drain_session(struct nfs_client *clp)
256{
257	struct nfs4_session *ses = clp->cl_session;
258	int ret = 0;
259
260	if (clp->cl_slot_tbl)
261		return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
262
263	/* back channel */
264	ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
265	if (ret)
266		return ret;
267	/* fore channel */
268	return nfs4_drain_slot_tbl(&ses->fc_slot_table);
269}
270
271#if defined(CONFIG_NFS_V4_1)
272
273static int nfs41_setup_state_renewal(struct nfs_client *clp)
274{
275	int status;
276	struct nfs_fsinfo fsinfo;
277
278	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
279		nfs4_schedule_state_renewal(clp);
280		return 0;
281	}
282
283	status = nfs4_proc_get_lease_time(clp, &fsinfo);
284	if (status == 0) {
285		/* Update lease time and schedule renewal */
286		spin_lock(&clp->cl_lock);
287		clp->cl_lease_time = fsinfo.lease_time * HZ;
288		clp->cl_last_renewal = jiffies;
289		spin_unlock(&clp->cl_lock);
290
291		nfs4_schedule_state_renewal(clp);
292	}
293
294	return status;
295}
296
297static void nfs41_finish_session_reset(struct nfs_client *clp)
298{
299	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301	/* create_session negotiated new slot table */
302	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303	nfs41_setup_state_renewal(clp);
304}
305
306int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
307{
308	int status;
309
310	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311		goto do_confirm;
312	status = nfs4_proc_exchange_id(clp, cred);
313	if (status != 0)
314		goto out;
315	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
316do_confirm:
317	status = nfs4_proc_create_session(clp, cred);
318	if (status != 0)
319		goto out;
320	nfs41_finish_session_reset(clp);
321	nfs_mark_client_ready(clp, NFS_CS_READY);
322out:
323	return status;
324}
325
326/**
327 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
328 *
329 * @clp: nfs_client under test
330 * @result: OUT: found nfs_client, or clp
331 * @cred: credential to use for trunking test
332 *
333 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
334 * If NFS4_OK is returned, an nfs_client pointer is planted in
335 * "result".
336 *
337 * Note: The returned client may not yet be marked ready.
338 */
339int nfs41_discover_server_trunking(struct nfs_client *clp,
340				   struct nfs_client **result,
341				   struct rpc_cred *cred)
342{
343	int status;
344
345	status = nfs4_proc_exchange_id(clp, cred);
346	if (status != NFS4_OK)
347		return status;
348
349	status = nfs41_walk_client_list(clp, result, cred);
350	if (status < 0)
351		return status;
352	if (clp != *result)
353		return 0;
354
355	/* Purge state if the client id was established in a prior instance */
356	if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
357		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
358	else
359		set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
360	nfs4_schedule_state_manager(clp);
361	status = nfs_wait_client_init_complete(clp);
362	if (status < 0)
363		nfs_put_client(clp);
364	return status;
365}
366
367#endif /* CONFIG_NFS_V4_1 */
368
369/**
370 * nfs4_get_clid_cred - Acquire credential for a setclientid operation
371 * @clp: client state handle
372 *
373 * Returns an rpc_cred with reference count bumped, or NULL.
374 */
375struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
376{
377	struct rpc_cred *cred;
378
379	spin_lock(&clp->cl_lock);
380	cred = nfs4_get_machine_cred_locked(clp);
381	spin_unlock(&clp->cl_lock);
382	return cred;
383}
384
385static struct nfs4_state_owner *
386nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
387{
388	struct rb_node **p = &server->state_owners.rb_node,
389		       *parent = NULL;
390	struct nfs4_state_owner *sp;
391
392	while (*p != NULL) {
393		parent = *p;
394		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
395
396		if (cred < sp->so_cred)
397			p = &parent->rb_left;
398		else if (cred > sp->so_cred)
399			p = &parent->rb_right;
400		else {
401			if (!list_empty(&sp->so_lru))
402				list_del_init(&sp->so_lru);
403			atomic_inc(&sp->so_count);
404			return sp;
405		}
406	}
407	return NULL;
408}
409
410static struct nfs4_state_owner *
411nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
412{
413	struct nfs_server *server = new->so_server;
414	struct rb_node **p = &server->state_owners.rb_node,
415		       *parent = NULL;
416	struct nfs4_state_owner *sp;
417	int err;
418
419	while (*p != NULL) {
420		parent = *p;
421		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
422
423		if (new->so_cred < sp->so_cred)
424			p = &parent->rb_left;
425		else if (new->so_cred > sp->so_cred)
426			p = &parent->rb_right;
427		else {
428			if (!list_empty(&sp->so_lru))
429				list_del_init(&sp->so_lru);
430			atomic_inc(&sp->so_count);
431			return sp;
432		}
433	}
434	err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
435	if (err)
436		return ERR_PTR(err);
437	rb_link_node(&new->so_server_node, parent, p);
438	rb_insert_color(&new->so_server_node, &server->state_owners);
439	return new;
440}
441
442static void
443nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
444{
445	struct nfs_server *server = sp->so_server;
446
447	if (!RB_EMPTY_NODE(&sp->so_server_node))
448		rb_erase(&sp->so_server_node, &server->state_owners);
449	ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
450}
451
452static void
453nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
454{
455	sc->create_time = ktime_get();
456	sc->flags = 0;
457	sc->counter = 0;
458	spin_lock_init(&sc->lock);
459	INIT_LIST_HEAD(&sc->list);
460	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
461}
462
463static void
464nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
465{
466	rpc_destroy_wait_queue(&sc->wait);
467}
468
469/*
470 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
471 * create a new state_owner.
472 *
473 */
474static struct nfs4_state_owner *
475nfs4_alloc_state_owner(struct nfs_server *server,
476		struct rpc_cred *cred,
477		gfp_t gfp_flags)
478{
479	struct nfs4_state_owner *sp;
480
481	sp = kzalloc(sizeof(*sp), gfp_flags);
482	if (!sp)
483		return NULL;
484	sp->so_server = server;
485	sp->so_cred = get_rpccred(cred);
486	spin_lock_init(&sp->so_lock);
487	INIT_LIST_HEAD(&sp->so_states);
488	nfs4_init_seqid_counter(&sp->so_seqid);
489	atomic_set(&sp->so_count, 1);
490	INIT_LIST_HEAD(&sp->so_lru);
491	seqcount_init(&sp->so_reclaim_seqcount);
492	mutex_init(&sp->so_delegreturn_mutex);
493	return sp;
494}
495
496static void
497nfs4_drop_state_owner(struct nfs4_state_owner *sp)
498{
499	struct rb_node *rb_node = &sp->so_server_node;
500
501	if (!RB_EMPTY_NODE(rb_node)) {
502		struct nfs_server *server = sp->so_server;
503		struct nfs_client *clp = server->nfs_client;
504
505		spin_lock(&clp->cl_lock);
506		if (!RB_EMPTY_NODE(rb_node)) {
507			rb_erase(rb_node, &server->state_owners);
508			RB_CLEAR_NODE(rb_node);
509		}
510		spin_unlock(&clp->cl_lock);
511	}
512}
513
514static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
515{
516	nfs4_destroy_seqid_counter(&sp->so_seqid);
517	put_rpccred(sp->so_cred);
518	kfree(sp);
519}
520
521static void nfs4_gc_state_owners(struct nfs_server *server)
522{
523	struct nfs_client *clp = server->nfs_client;
524	struct nfs4_state_owner *sp, *tmp;
525	unsigned long time_min, time_max;
526	LIST_HEAD(doomed);
527
528	spin_lock(&clp->cl_lock);
529	time_max = jiffies;
530	time_min = (long)time_max - (long)clp->cl_lease_time;
531	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
532		/* NB: LRU is sorted so that oldest is at the head */
533		if (time_in_range(sp->so_expires, time_min, time_max))
534			break;
535		list_move(&sp->so_lru, &doomed);
536		nfs4_remove_state_owner_locked(sp);
537	}
538	spin_unlock(&clp->cl_lock);
539
540	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
541		list_del(&sp->so_lru);
542		nfs4_free_state_owner(sp);
543	}
544}
545
546/**
547 * nfs4_get_state_owner - Look up a state owner given a credential
548 * @server: nfs_server to search
549 * @cred: RPC credential to match
550 *
551 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
552 */
553struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
554					      struct rpc_cred *cred,
555					      gfp_t gfp_flags)
556{
557	struct nfs_client *clp = server->nfs_client;
558	struct nfs4_state_owner *sp, *new;
559
560	spin_lock(&clp->cl_lock);
561	sp = nfs4_find_state_owner_locked(server, cred);
562	spin_unlock(&clp->cl_lock);
563	if (sp != NULL)
564		goto out;
565	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
566	if (new == NULL)
567		goto out;
568	do {
569		if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
570			break;
571		spin_lock(&clp->cl_lock);
572		sp = nfs4_insert_state_owner_locked(new);
573		spin_unlock(&clp->cl_lock);
574	} while (sp == ERR_PTR(-EAGAIN));
575	if (sp != new)
576		nfs4_free_state_owner(new);
577out:
578	nfs4_gc_state_owners(server);
579	return sp;
580}
581
582/**
583 * nfs4_put_state_owner - Release a nfs4_state_owner
584 * @sp: state owner data to release
585 *
586 * Note that we keep released state owners on an LRU
587 * list.
588 * This caches valid state owners so that they can be
589 * reused, to avoid the OPEN_CONFIRM on minor version 0.
590 * It also pins the uniquifier of dropped state owners for
591 * a while, to ensure that those state owner names are
592 * never reused.
593 */
594void nfs4_put_state_owner(struct nfs4_state_owner *sp)
595{
596	struct nfs_server *server = sp->so_server;
597	struct nfs_client *clp = server->nfs_client;
598
599	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
600		return;
601
602	sp->so_expires = jiffies;
603	list_add_tail(&sp->so_lru, &server->state_owners_lru);
604	spin_unlock(&clp->cl_lock);
605}
606
607/**
608 * nfs4_purge_state_owners - Release all cached state owners
609 * @server: nfs_server with cached state owners to release
610 *
611 * Called at umount time.  Remaining state owners will be on
612 * the LRU with ref count of zero.
613 */
614void nfs4_purge_state_owners(struct nfs_server *server)
615{
616	struct nfs_client *clp = server->nfs_client;
617	struct nfs4_state_owner *sp, *tmp;
618	LIST_HEAD(doomed);
619
620	spin_lock(&clp->cl_lock);
621	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
622		list_move(&sp->so_lru, &doomed);
623		nfs4_remove_state_owner_locked(sp);
624	}
625	spin_unlock(&clp->cl_lock);
626
627	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
628		list_del(&sp->so_lru);
629		nfs4_free_state_owner(sp);
630	}
631}
632
633static struct nfs4_state *
634nfs4_alloc_open_state(void)
635{
636	struct nfs4_state *state;
637
638	state = kzalloc(sizeof(*state), GFP_NOFS);
639	if (!state)
640		return NULL;
641	atomic_set(&state->count, 1);
642	INIT_LIST_HEAD(&state->lock_states);
643	spin_lock_init(&state->state_lock);
644	seqlock_init(&state->seqlock);
645	return state;
646}
647
648void
649nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
650{
651	if (state->state == fmode)
652		return;
653	/* NB! List reordering - see the reclaim code for why.  */
654	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
655		if (fmode & FMODE_WRITE)
656			list_move(&state->open_states, &state->owner->so_states);
657		else
658			list_move_tail(&state->open_states, &state->owner->so_states);
659	}
660	state->state = fmode;
661}
662
663static struct nfs4_state *
664__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
665{
666	struct nfs_inode *nfsi = NFS_I(inode);
667	struct nfs4_state *state;
668
669	list_for_each_entry(state, &nfsi->open_states, inode_states) {
670		if (state->owner != owner)
671			continue;
672		if (!nfs4_valid_open_stateid(state))
673			continue;
674		if (atomic_inc_not_zero(&state->count))
675			return state;
676	}
677	return NULL;
678}
679
680static void
681nfs4_free_open_state(struct nfs4_state *state)
682{
683	kfree(state);
684}
685
686struct nfs4_state *
687nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
688{
689	struct nfs4_state *state, *new;
690	struct nfs_inode *nfsi = NFS_I(inode);
691
692	spin_lock(&inode->i_lock);
693	state = __nfs4_find_state_byowner(inode, owner);
694	spin_unlock(&inode->i_lock);
695	if (state)
696		goto out;
697	new = nfs4_alloc_open_state();
698	spin_lock(&owner->so_lock);
699	spin_lock(&inode->i_lock);
700	state = __nfs4_find_state_byowner(inode, owner);
701	if (state == NULL && new != NULL) {
702		state = new;
703		state->owner = owner;
704		atomic_inc(&owner->so_count);
705		list_add(&state->inode_states, &nfsi->open_states);
706		ihold(inode);
707		state->inode = inode;
708		spin_unlock(&inode->i_lock);
709		/* Note: The reclaim code dictates that we add stateless
710		 * and read-only stateids to the end of the list */
711		list_add_tail(&state->open_states, &owner->so_states);
712		spin_unlock(&owner->so_lock);
713	} else {
714		spin_unlock(&inode->i_lock);
715		spin_unlock(&owner->so_lock);
716		if (new)
717			nfs4_free_open_state(new);
718	}
719out:
720	return state;
721}
722
723void nfs4_put_open_state(struct nfs4_state *state)
724{
725	struct inode *inode = state->inode;
726	struct nfs4_state_owner *owner = state->owner;
727
728	if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
729		return;
730	spin_lock(&inode->i_lock);
731	list_del(&state->inode_states);
732	list_del(&state->open_states);
733	spin_unlock(&inode->i_lock);
734	spin_unlock(&owner->so_lock);
735	iput(inode);
736	nfs4_free_open_state(state);
737	nfs4_put_state_owner(owner);
738}
739
740/*
741 * Close the current file.
742 */
743static void __nfs4_close(struct nfs4_state *state,
744		fmode_t fmode, gfp_t gfp_mask, int wait)
745{
746	struct nfs4_state_owner *owner = state->owner;
747	int call_close = 0;
748	fmode_t newstate;
749
750	atomic_inc(&owner->so_count);
751	/* Protect against nfs4_find_state() */
752	spin_lock(&owner->so_lock);
753	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
754		case FMODE_READ:
755			state->n_rdonly--;
756			break;
757		case FMODE_WRITE:
758			state->n_wronly--;
759			break;
760		case FMODE_READ|FMODE_WRITE:
761			state->n_rdwr--;
762	}
763	newstate = FMODE_READ|FMODE_WRITE;
764	if (state->n_rdwr == 0) {
765		if (state->n_rdonly == 0) {
766			newstate &= ~FMODE_READ;
767			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
768			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
769		}
770		if (state->n_wronly == 0) {
771			newstate &= ~FMODE_WRITE;
772			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
773			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
774		}
775		if (newstate == 0)
776			clear_bit(NFS_DELEGATED_STATE, &state->flags);
777	}
778	nfs4_state_set_mode_locked(state, newstate);
779	spin_unlock(&owner->so_lock);
780
781	if (!call_close) {
782		nfs4_put_open_state(state);
783		nfs4_put_state_owner(owner);
784	} else
785		nfs4_do_close(state, gfp_mask, wait);
786}
787
788void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
789{
790	__nfs4_close(state, fmode, GFP_NOFS, 0);
791}
792
793void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
794{
795	__nfs4_close(state, fmode, GFP_KERNEL, 1);
796}
797
798/*
799 * Search the state->lock_states for an existing lock_owner
800 * that is compatible with current->files
801 */
802static struct nfs4_lock_state *
803__nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
804{
805	struct nfs4_lock_state *pos;
806	list_for_each_entry(pos, &state->lock_states, ls_locks) {
807		if (pos->ls_owner != fl_owner)
808			continue;
809		atomic_inc(&pos->ls_count);
810		return pos;
811	}
812	return NULL;
813}
814
815/*
816 * Return a compatible lock_state. If no initialized lock_state structure
817 * exists, return an uninitialized one.
818 *
819 */
820static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
821{
822	struct nfs4_lock_state *lsp;
823	struct nfs_server *server = state->owner->so_server;
824
825	lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
826	if (lsp == NULL)
827		return NULL;
828	nfs4_init_seqid_counter(&lsp->ls_seqid);
829	atomic_set(&lsp->ls_count, 1);
830	lsp->ls_state = state;
831	lsp->ls_owner = fl_owner;
832	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
833	if (lsp->ls_seqid.owner_id < 0)
834		goto out_free;
835	INIT_LIST_HEAD(&lsp->ls_locks);
836	return lsp;
837out_free:
838	kfree(lsp);
839	return NULL;
840}
841
842void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
843{
844	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
845	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
846	kfree(lsp);
847}
848
849/*
850 * Return a compatible lock_state. If no initialized lock_state structure
851 * exists, return an uninitialized one.
852 *
853 */
854static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
855{
856	struct nfs4_lock_state *lsp, *new = NULL;
857
858	for(;;) {
859		spin_lock(&state->state_lock);
860		lsp = __nfs4_find_lock_state(state, owner);
861		if (lsp != NULL)
862			break;
863		if (new != NULL) {
864			list_add(&new->ls_locks, &state->lock_states);
865			set_bit(LK_STATE_IN_USE, &state->flags);
866			lsp = new;
867			new = NULL;
868			break;
869		}
870		spin_unlock(&state->state_lock);
871		new = nfs4_alloc_lock_state(state, owner);
872		if (new == NULL)
873			return NULL;
874	}
875	spin_unlock(&state->state_lock);
876	if (new != NULL)
877		nfs4_free_lock_state(state->owner->so_server, new);
878	return lsp;
879}
880
881/*
882 * Release reference to lock_state, and free it if we see that
883 * it is no longer in use
884 */
885void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
886{
887	struct nfs_server *server;
888	struct nfs4_state *state;
889
890	if (lsp == NULL)
891		return;
892	state = lsp->ls_state;
893	if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
894		return;
895	list_del(&lsp->ls_locks);
896	if (list_empty(&state->lock_states))
897		clear_bit(LK_STATE_IN_USE, &state->flags);
898	spin_unlock(&state->state_lock);
899	server = state->owner->so_server;
900	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
901		struct nfs_client *clp = server->nfs_client;
902
903		clp->cl_mvops->free_lock_state(server, lsp);
904	} else
905		nfs4_free_lock_state(server, lsp);
906}
907
908static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
909{
910	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
911
912	dst->fl_u.nfs4_fl.owner = lsp;
913	atomic_inc(&lsp->ls_count);
914}
915
916static void nfs4_fl_release_lock(struct file_lock *fl)
917{
918	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
919}
920
921static const struct file_lock_operations nfs4_fl_lock_ops = {
922	.fl_copy_lock = nfs4_fl_copy_lock,
923	.fl_release_private = nfs4_fl_release_lock,
924};
925
926int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
927{
928	struct nfs4_lock_state *lsp;
929
930	if (fl->fl_ops != NULL)
931		return 0;
932	lsp = nfs4_get_lock_state(state, fl->fl_owner);
933	if (lsp == NULL)
934		return -ENOMEM;
935	fl->fl_u.nfs4_fl.owner = lsp;
936	fl->fl_ops = &nfs4_fl_lock_ops;
937	return 0;
938}
939
940static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
941		struct nfs4_state *state,
942		const struct nfs_lockowner *lockowner)
943{
944	struct nfs4_lock_state *lsp;
945	fl_owner_t fl_owner;
946	int ret = -ENOENT;
947
948
949	if (lockowner == NULL)
950		goto out;
951
952	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
953		goto out;
954
955	fl_owner = lockowner->l_owner;
956	spin_lock(&state->state_lock);
957	lsp = __nfs4_find_lock_state(state, fl_owner);
958	if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
959		ret = -EIO;
960	else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
961		nfs4_stateid_copy(dst, &lsp->ls_stateid);
962		ret = 0;
963	}
964	spin_unlock(&state->state_lock);
965	nfs4_put_lock_state(lsp);
966out:
967	return ret;
968}
969
970static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
971{
972	const nfs4_stateid *src;
973	int seq;
974
975	do {
976		src = &zero_stateid;
977		seq = read_seqbegin(&state->seqlock);
978		if (test_bit(NFS_OPEN_STATE, &state->flags))
979			src = &state->open_stateid;
980		nfs4_stateid_copy(dst, src);
981	} while (read_seqretry(&state->seqlock, seq));
982}
983
984/*
985 * Byte-range lock aware utility to initialize the stateid of read/write
986 * requests.
987 */
988int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
989		fmode_t fmode, const struct nfs_lockowner *lockowner)
990{
991	int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
992	if (ret == -EIO)
993		/* A lost lock - don't even consider delegations */
994		goto out;
995	/* returns true if delegation stateid found and copied */
996	if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
997		ret = 0;
998		goto out;
999	}
1000	if (ret != -ENOENT)
1001		/* nfs4_copy_delegation_stateid() didn't over-write
1002		 * dst, so it still has the lock stateid which we now
1003		 * choose to use.
1004		 */
1005		goto out;
1006	nfs4_copy_open_stateid(dst, state);
1007	ret = 0;
1008out:
1009	if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1010		dst->seqid = 0;
1011	return ret;
1012}
1013
1014struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1015{
1016	struct nfs_seqid *new;
1017
1018	new = kmalloc(sizeof(*new), gfp_mask);
1019	if (new == NULL)
1020		return ERR_PTR(-ENOMEM);
1021	new->sequence = counter;
1022	INIT_LIST_HEAD(&new->list);
1023	new->task = NULL;
1024	return new;
1025}
1026
1027void nfs_release_seqid(struct nfs_seqid *seqid)
1028{
1029	struct nfs_seqid_counter *sequence;
1030
1031	if (seqid == NULL || list_empty(&seqid->list))
1032		return;
1033	sequence = seqid->sequence;
1034	spin_lock(&sequence->lock);
1035	list_del_init(&seqid->list);
1036	if (!list_empty(&sequence->list)) {
1037		struct nfs_seqid *next;
1038
1039		next = list_first_entry(&sequence->list,
1040				struct nfs_seqid, list);
1041		rpc_wake_up_queued_task(&sequence->wait, next->task);
1042	}
1043	spin_unlock(&sequence->lock);
1044}
1045
1046void nfs_free_seqid(struct nfs_seqid *seqid)
1047{
1048	nfs_release_seqid(seqid);
1049	kfree(seqid);
1050}
1051
1052/*
1053 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1054 * failed with a seqid incrementing error -
1055 * see comments nfs4.h:seqid_mutating_error()
1056 */
1057static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1058{
1059	switch (status) {
1060		case 0:
1061			break;
1062		case -NFS4ERR_BAD_SEQID:
1063			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1064				return;
1065			pr_warn_ratelimited("NFS: v4 server returned a bad"
1066					" sequence-id error on an"
1067					" unconfirmed sequence %p!\n",
1068					seqid->sequence);
1069		case -NFS4ERR_STALE_CLIENTID:
1070		case -NFS4ERR_STALE_STATEID:
1071		case -NFS4ERR_BAD_STATEID:
1072		case -NFS4ERR_BADXDR:
1073		case -NFS4ERR_RESOURCE:
1074		case -NFS4ERR_NOFILEHANDLE:
1075			/* Non-seqid mutating errors */
1076			return;
1077	};
1078	/*
1079	 * Note: no locking needed as we are guaranteed to be first
1080	 * on the sequence list
1081	 */
1082	seqid->sequence->counter++;
1083}
1084
1085void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1086{
1087	struct nfs4_state_owner *sp;
1088
1089	if (seqid == NULL)
1090		return;
1091
1092	sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1093	if (status == -NFS4ERR_BAD_SEQID)
1094		nfs4_drop_state_owner(sp);
1095	if (!nfs4_has_session(sp->so_server->nfs_client))
1096		nfs_increment_seqid(status, seqid);
1097}
1098
1099/*
1100 * Increment the seqid if the LOCK/LOCKU succeeded, or
1101 * failed with a seqid incrementing error -
1102 * see comments nfs4.h:seqid_mutating_error()
1103 */
1104void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1105{
1106	if (seqid != NULL)
1107		nfs_increment_seqid(status, seqid);
1108}
1109
1110int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1111{
1112	struct nfs_seqid_counter *sequence;
1113	int status = 0;
1114
1115	if (seqid == NULL)
1116		goto out;
1117	sequence = seqid->sequence;
1118	spin_lock(&sequence->lock);
1119	seqid->task = task;
1120	if (list_empty(&seqid->list))
1121		list_add_tail(&seqid->list, &sequence->list);
1122	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1123		goto unlock;
1124	rpc_sleep_on(&sequence->wait, task, NULL);
1125	status = -EAGAIN;
1126unlock:
1127	spin_unlock(&sequence->lock);
1128out:
1129	return status;
1130}
1131
1132static int nfs4_run_state_manager(void *);
1133
1134static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1135{
1136	smp_mb__before_atomic();
1137	clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1138	smp_mb__after_atomic();
1139	wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1140	rpc_wake_up(&clp->cl_rpcwaitq);
1141}
1142
1143/*
1144 * Schedule the nfs_client asynchronous state management routine
1145 */
1146void nfs4_schedule_state_manager(struct nfs_client *clp)
1147{
1148	struct task_struct *task;
1149	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1150
1151	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1152		return;
1153	__module_get(THIS_MODULE);
1154	atomic_inc(&clp->cl_count);
1155
1156	/* The rcu_read_lock() is not strictly necessary, as the state
1157	 * manager is the only thread that ever changes the rpc_xprt
1158	 * after it's initialized.  At this point, we're single threaded. */
1159	rcu_read_lock();
1160	snprintf(buf, sizeof(buf), "%s-manager",
1161			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1162	rcu_read_unlock();
1163	task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1164	if (IS_ERR(task)) {
1165		printk(KERN_ERR "%s: kthread_run: %ld\n",
1166			__func__, PTR_ERR(task));
1167		nfs4_clear_state_manager_bit(clp);
1168		nfs_put_client(clp);
1169		module_put(THIS_MODULE);
1170	}
1171}
1172
1173/*
1174 * Schedule a lease recovery attempt
1175 */
1176void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1177{
1178	if (!clp)
1179		return;
1180	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1181		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1182	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1183			clp->cl_hostname);
1184	nfs4_schedule_state_manager(clp);
1185}
1186EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1187
1188/**
1189 * nfs4_schedule_migration_recovery - trigger migration recovery
1190 *
1191 * @server: FSID that is migrating
1192 *
1193 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1194 * value is returned.
1195 */
1196int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1197{
1198	struct nfs_client *clp = server->nfs_client;
1199
1200	if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1201		pr_err("NFS: volatile file handles not supported (server %s)\n",
1202				clp->cl_hostname);
1203		return -NFS4ERR_IO;
1204	}
1205
1206	if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1207		return -NFS4ERR_IO;
1208
1209	dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1210			__func__,
1211			(unsigned long long)server->fsid.major,
1212			(unsigned long long)server->fsid.minor,
1213			clp->cl_hostname);
1214
1215	set_bit(NFS_MIG_IN_TRANSITION,
1216			&((struct nfs_server *)server)->mig_status);
1217	set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1218
1219	nfs4_schedule_state_manager(clp);
1220	return 0;
1221}
1222EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1223
1224/**
1225 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1226 *
1227 * @clp: server to check for moved leases
1228 *
1229 */
1230void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1231{
1232	dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1233		__func__, clp->cl_clientid, clp->cl_hostname);
1234
1235	set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1236	nfs4_schedule_state_manager(clp);
1237}
1238EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1239
1240int nfs4_wait_clnt_recover(struct nfs_client *clp)
1241{
1242	int res;
1243
1244	might_sleep();
1245
1246	atomic_inc(&clp->cl_count);
1247	res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1248				 nfs_wait_bit_killable, TASK_KILLABLE);
1249	if (res)
1250		goto out;
1251	if (clp->cl_cons_state < 0)
1252		res = clp->cl_cons_state;
1253out:
1254	nfs_put_client(clp);
1255	return res;
1256}
1257
1258int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1259{
1260	unsigned int loop;
1261	int ret;
1262
1263	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1264		ret = nfs4_wait_clnt_recover(clp);
1265		if (ret != 0)
1266			break;
1267		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1268		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1269			break;
1270		nfs4_schedule_state_manager(clp);
1271		ret = -EIO;
1272	}
1273	return ret;
1274}
1275
1276/*
1277 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1278 * @clp: client to process
1279 *
1280 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1281 * resend of the SETCLIENTID and hence re-establish the
1282 * callback channel. Then return all existing delegations.
1283 */
1284static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1285{
1286	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1287	nfs_expire_all_delegations(clp);
1288	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1289			clp->cl_hostname);
1290}
1291
1292void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1293{
1294	nfs40_handle_cb_pathdown(clp);
1295	nfs4_schedule_state_manager(clp);
1296}
1297
1298static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1299{
1300
1301	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1302	/* Don't recover state that expired before the reboot */
1303	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1304		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1305		return 0;
1306	}
1307	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1308	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1309	return 1;
1310}
1311
1312int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1313{
1314	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1315	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1316	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1317	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1318	return 1;
1319}
1320
1321int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1322{
1323	struct nfs_client *clp = server->nfs_client;
1324
1325	if (!nfs4_valid_open_stateid(state))
1326		return -EBADF;
1327	nfs4_state_mark_reclaim_nograce(clp, state);
1328	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1329			clp->cl_hostname);
1330	nfs4_schedule_state_manager(clp);
1331	return 0;
1332}
1333EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1334
1335void nfs_inode_find_state_and_recover(struct inode *inode,
1336		const nfs4_stateid *stateid)
1337{
1338	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1339	struct nfs_inode *nfsi = NFS_I(inode);
1340	struct nfs_open_context *ctx;
1341	struct nfs4_state *state;
1342	bool found = false;
1343
1344	spin_lock(&inode->i_lock);
1345	list_for_each_entry(ctx, &nfsi->open_files, list) {
1346		state = ctx->state;
1347		if (state == NULL)
1348			continue;
1349		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1350			continue;
1351		if (!nfs4_stateid_match(&state->stateid, stateid))
1352			continue;
1353		nfs4_state_mark_reclaim_nograce(clp, state);
1354		found = true;
1355	}
1356	spin_unlock(&inode->i_lock);
1357	if (found)
1358		nfs4_schedule_state_manager(clp);
1359}
1360
1361static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1362{
1363	struct inode *inode = state->inode;
1364	struct nfs_inode *nfsi = NFS_I(inode);
1365	struct nfs_open_context *ctx;
1366
1367	spin_lock(&inode->i_lock);
1368	list_for_each_entry(ctx, &nfsi->open_files, list) {
1369		if (ctx->state != state)
1370			continue;
1371		set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1372	}
1373	spin_unlock(&inode->i_lock);
1374}
1375
1376static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1377{
1378	set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1379	nfs4_state_mark_open_context_bad(state);
1380}
1381
1382
1383static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1384{
1385	struct inode *inode = state->inode;
1386	struct nfs_inode *nfsi = NFS_I(inode);
1387	struct file_lock *fl;
1388	int status = 0;
1389	struct file_lock_context *flctx = inode->i_flctx;
1390	struct list_head *list;
1391
1392	if (flctx == NULL)
1393		return 0;
1394
1395	list = &flctx->flc_posix;
1396
1397	/* Guard against delegation returns and new lock/unlock calls */
1398	down_write(&nfsi->rwsem);
1399	spin_lock(&flctx->flc_lock);
1400restart:
1401	list_for_each_entry(fl, list, fl_list) {
1402		if (nfs_file_open_context(fl->fl_file)->state != state)
1403			continue;
1404		spin_unlock(&flctx->flc_lock);
1405		status = ops->recover_lock(state, fl);
1406		switch (status) {
1407		case 0:
1408			break;
1409		case -ESTALE:
1410		case -NFS4ERR_ADMIN_REVOKED:
1411		case -NFS4ERR_STALE_STATEID:
1412		case -NFS4ERR_BAD_STATEID:
1413		case -NFS4ERR_EXPIRED:
1414		case -NFS4ERR_NO_GRACE:
1415		case -NFS4ERR_STALE_CLIENTID:
1416		case -NFS4ERR_BADSESSION:
1417		case -NFS4ERR_BADSLOT:
1418		case -NFS4ERR_BAD_HIGH_SLOT:
1419		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1420			goto out;
1421		default:
1422			pr_err("NFS: %s: unhandled error %d\n",
1423					__func__, status);
1424		case -ENOMEM:
1425		case -NFS4ERR_DENIED:
1426		case -NFS4ERR_RECLAIM_BAD:
1427		case -NFS4ERR_RECLAIM_CONFLICT:
1428			/* kill_proc(fl->fl_pid, SIGLOST, 1); */
1429			status = 0;
1430		}
1431		spin_lock(&flctx->flc_lock);
1432	}
1433	if (list == &flctx->flc_posix) {
1434		list = &flctx->flc_flock;
1435		goto restart;
1436	}
1437	spin_unlock(&flctx->flc_lock);
1438out:
1439	up_write(&nfsi->rwsem);
1440	return status;
1441}
1442
1443static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1444{
1445	struct nfs4_state *state;
1446	struct nfs4_lock_state *lock;
1447	int status = 0;
1448
1449	/* Note: we rely on the sp->so_states list being ordered
1450	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1451	 * states first.
1452	 * This is needed to ensure that the server won't give us any
1453	 * read delegations that we have to return if, say, we are
1454	 * recovering after a network partition or a reboot from a
1455	 * server that doesn't support a grace period.
1456	 */
1457	spin_lock(&sp->so_lock);
1458	raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1459restart:
1460	list_for_each_entry(state, &sp->so_states, open_states) {
1461		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1462			continue;
1463		if (!nfs4_valid_open_stateid(state))
1464			continue;
1465		if (state->state == 0)
1466			continue;
1467		atomic_inc(&state->count);
1468		spin_unlock(&sp->so_lock);
1469		status = ops->recover_open(sp, state);
1470		if (status >= 0) {
1471			status = nfs4_reclaim_locks(state, ops);
1472			if (status >= 0) {
1473				if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1474					spin_lock(&state->state_lock);
1475					list_for_each_entry(lock, &state->lock_states, ls_locks) {
1476						if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1477							pr_warn_ratelimited("NFS: "
1478									    "%s: Lock reclaim "
1479									    "failed!\n", __func__);
1480					}
1481					spin_unlock(&state->state_lock);
1482				}
1483				nfs4_put_open_state(state);
1484				clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1485					&state->flags);
1486				spin_lock(&sp->so_lock);
1487				goto restart;
1488			}
1489		}
1490		switch (status) {
1491			default:
1492				printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1493					__func__, status);
1494			case -ENOENT:
1495			case -ENOMEM:
1496			case -ESTALE:
1497				/* Open state on this file cannot be recovered */
1498				nfs4_state_mark_recovery_failed(state, status);
1499				break;
1500			case -EAGAIN:
1501				ssleep(1);
1502			case -NFS4ERR_ADMIN_REVOKED:
1503			case -NFS4ERR_STALE_STATEID:
1504			case -NFS4ERR_BAD_STATEID:
1505			case -NFS4ERR_RECLAIM_BAD:
1506			case -NFS4ERR_RECLAIM_CONFLICT:
1507				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1508				break;
1509			case -NFS4ERR_EXPIRED:
1510			case -NFS4ERR_NO_GRACE:
1511				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1512			case -NFS4ERR_STALE_CLIENTID:
1513			case -NFS4ERR_BADSESSION:
1514			case -NFS4ERR_BADSLOT:
1515			case -NFS4ERR_BAD_HIGH_SLOT:
1516			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1517				goto out_err;
1518		}
1519		nfs4_put_open_state(state);
1520		spin_lock(&sp->so_lock);
1521		goto restart;
1522	}
1523	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1524	spin_unlock(&sp->so_lock);
1525	return 0;
1526out_err:
1527	nfs4_put_open_state(state);
1528	spin_lock(&sp->so_lock);
1529	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1530	spin_unlock(&sp->so_lock);
1531	return status;
1532}
1533
1534static void nfs4_clear_open_state(struct nfs4_state *state)
1535{
1536	struct nfs4_lock_state *lock;
1537
1538	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1539	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1540	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1541	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1542	spin_lock(&state->state_lock);
1543	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1544		lock->ls_seqid.flags = 0;
1545		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1546	}
1547	spin_unlock(&state->state_lock);
1548}
1549
1550static void nfs4_reset_seqids(struct nfs_server *server,
1551	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1552{
1553	struct nfs_client *clp = server->nfs_client;
1554	struct nfs4_state_owner *sp;
1555	struct rb_node *pos;
1556	struct nfs4_state *state;
1557
1558	spin_lock(&clp->cl_lock);
1559	for (pos = rb_first(&server->state_owners);
1560	     pos != NULL;
1561	     pos = rb_next(pos)) {
1562		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1563		sp->so_seqid.flags = 0;
1564		spin_lock(&sp->so_lock);
1565		list_for_each_entry(state, &sp->so_states, open_states) {
1566			if (mark_reclaim(clp, state))
1567				nfs4_clear_open_state(state);
1568		}
1569		spin_unlock(&sp->so_lock);
1570	}
1571	spin_unlock(&clp->cl_lock);
1572}
1573
1574static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1575	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1576{
1577	struct nfs_server *server;
1578
1579	rcu_read_lock();
1580	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1581		nfs4_reset_seqids(server, mark_reclaim);
1582	rcu_read_unlock();
1583}
1584
1585static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1586{
1587	/* Mark all delegations for reclaim */
1588	nfs_delegation_mark_reclaim(clp);
1589	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1590}
1591
1592static void nfs4_reclaim_complete(struct nfs_client *clp,
1593				 const struct nfs4_state_recovery_ops *ops,
1594				 struct rpc_cred *cred)
1595{
1596	/* Notify the server we're done reclaiming our state */
1597	if (ops->reclaim_complete)
1598		(void)ops->reclaim_complete(clp, cred);
1599}
1600
1601static void nfs4_clear_reclaim_server(struct nfs_server *server)
1602{
1603	struct nfs_client *clp = server->nfs_client;
1604	struct nfs4_state_owner *sp;
1605	struct rb_node *pos;
1606	struct nfs4_state *state;
1607
1608	spin_lock(&clp->cl_lock);
1609	for (pos = rb_first(&server->state_owners);
1610	     pos != NULL;
1611	     pos = rb_next(pos)) {
1612		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1613		spin_lock(&sp->so_lock);
1614		list_for_each_entry(state, &sp->so_states, open_states) {
1615			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1616						&state->flags))
1617				continue;
1618			nfs4_state_mark_reclaim_nograce(clp, state);
1619		}
1620		spin_unlock(&sp->so_lock);
1621	}
1622	spin_unlock(&clp->cl_lock);
1623}
1624
1625static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1626{
1627	struct nfs_server *server;
1628
1629	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1630		return 0;
1631
1632	rcu_read_lock();
1633	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1634		nfs4_clear_reclaim_server(server);
1635	rcu_read_unlock();
1636
1637	nfs_delegation_reap_unclaimed(clp);
1638	return 1;
1639}
1640
1641static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1642{
1643	const struct nfs4_state_recovery_ops *ops;
1644	struct rpc_cred *cred;
1645
1646	if (!nfs4_state_clear_reclaim_reboot(clp))
1647		return;
1648	ops = clp->cl_mvops->reboot_recovery_ops;
1649	cred = nfs4_get_clid_cred(clp);
1650	nfs4_reclaim_complete(clp, ops, cred);
1651	put_rpccred(cred);
1652}
1653
1654static void nfs_delegation_clear_all(struct nfs_client *clp)
1655{
1656	nfs_delegation_mark_reclaim(clp);
1657	nfs_delegation_reap_unclaimed(clp);
1658}
1659
1660static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1661{
1662	nfs_delegation_clear_all(clp);
1663	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1664}
1665
1666static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1667{
1668	switch (error) {
1669		case 0:
1670			break;
1671		case -NFS4ERR_CB_PATH_DOWN:
1672			nfs40_handle_cb_pathdown(clp);
1673			break;
1674		case -NFS4ERR_NO_GRACE:
1675			nfs4_state_end_reclaim_reboot(clp);
1676			break;
1677		case -NFS4ERR_STALE_CLIENTID:
1678			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1679			nfs4_state_clear_reclaim_reboot(clp);
1680			nfs4_state_start_reclaim_reboot(clp);
1681			break;
1682		case -NFS4ERR_EXPIRED:
1683			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1684			nfs4_state_start_reclaim_nograce(clp);
1685			break;
1686		case -NFS4ERR_BADSESSION:
1687		case -NFS4ERR_BADSLOT:
1688		case -NFS4ERR_BAD_HIGH_SLOT:
1689		case -NFS4ERR_DEADSESSION:
1690		case -NFS4ERR_SEQ_FALSE_RETRY:
1691		case -NFS4ERR_SEQ_MISORDERED:
1692			set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1693			/* Zero session reset errors */
1694			break;
1695		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1696			set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1697			break;
1698		default:
1699			dprintk("%s: failed to handle error %d for server %s\n",
1700					__func__, error, clp->cl_hostname);
1701			return error;
1702	}
1703	dprintk("%s: handled error %d for server %s\n", __func__, error,
1704			clp->cl_hostname);
1705	return 0;
1706}
1707
1708static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1709{
1710	struct nfs4_state_owner *sp;
1711	struct nfs_server *server;
1712	struct rb_node *pos;
1713	int status = 0;
1714
1715restart:
1716	rcu_read_lock();
1717	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1718		nfs4_purge_state_owners(server);
1719		spin_lock(&clp->cl_lock);
1720		for (pos = rb_first(&server->state_owners);
1721		     pos != NULL;
1722		     pos = rb_next(pos)) {
1723			sp = rb_entry(pos,
1724				struct nfs4_state_owner, so_server_node);
1725			if (!test_and_clear_bit(ops->owner_flag_bit,
1726							&sp->so_flags))
1727				continue;
1728			if (!atomic_inc_not_zero(&sp->so_count))
1729				continue;
1730			spin_unlock(&clp->cl_lock);
1731			rcu_read_unlock();
1732
1733			status = nfs4_reclaim_open_state(sp, ops);
1734			if (status < 0) {
1735				set_bit(ops->owner_flag_bit, &sp->so_flags);
1736				nfs4_put_state_owner(sp);
1737				status = nfs4_recovery_handle_error(clp, status);
1738				return (status != 0) ? status : -EAGAIN;
1739			}
1740
1741			nfs4_put_state_owner(sp);
1742			goto restart;
1743		}
1744		spin_unlock(&clp->cl_lock);
1745	}
1746	rcu_read_unlock();
1747	return 0;
1748}
1749
1750static int nfs4_check_lease(struct nfs_client *clp)
1751{
1752	struct rpc_cred *cred;
1753	const struct nfs4_state_maintenance_ops *ops =
1754		clp->cl_mvops->state_renewal_ops;
1755	int status;
1756
1757	/* Is the client already known to have an expired lease? */
1758	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1759		return 0;
1760	spin_lock(&clp->cl_lock);
1761	cred = ops->get_state_renewal_cred_locked(clp);
1762	spin_unlock(&clp->cl_lock);
1763	if (cred == NULL) {
1764		cred = nfs4_get_clid_cred(clp);
1765		status = -ENOKEY;
1766		if (cred == NULL)
1767			goto out;
1768	}
1769	status = ops->renew_lease(clp, cred);
1770	put_rpccred(cred);
1771	if (status == -ETIMEDOUT) {
1772		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1773		return 0;
1774	}
1775out:
1776	return nfs4_recovery_handle_error(clp, status);
1777}
1778
1779/* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1780 * and for recoverable errors on EXCHANGE_ID for v4.1
1781 */
1782static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1783{
1784	switch (status) {
1785	case -NFS4ERR_SEQ_MISORDERED:
1786		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1787			return -ESERVERFAULT;
1788		/* Lease confirmation error: retry after purging the lease */
1789		ssleep(1);
1790		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1791		break;
1792	case -NFS4ERR_STALE_CLIENTID:
1793		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1794		nfs4_state_start_reclaim_reboot(clp);
1795		break;
1796	case -NFS4ERR_CLID_INUSE:
1797		pr_err("NFS: Server %s reports our clientid is in use\n",
1798			clp->cl_hostname);
1799		nfs_mark_client_ready(clp, -EPERM);
1800		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1801		return -EPERM;
1802	case -EACCES:
1803	case -NFS4ERR_DELAY:
1804	case -ETIMEDOUT:
1805	case -EAGAIN:
1806		ssleep(1);
1807		break;
1808
1809	case -NFS4ERR_MINOR_VERS_MISMATCH:
1810		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1811			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1812		dprintk("%s: exit with error %d for server %s\n",
1813				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
1814		return -EPROTONOSUPPORT;
1815	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1816				 * in nfs4_exchange_id */
1817	default:
1818		dprintk("%s: exit with error %d for server %s\n", __func__,
1819				status, clp->cl_hostname);
1820		return status;
1821	}
1822	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1823	dprintk("%s: handled error %d for server %s\n", __func__, status,
1824			clp->cl_hostname);
1825	return 0;
1826}
1827
1828static int nfs4_establish_lease(struct nfs_client *clp)
1829{
1830	struct rpc_cred *cred;
1831	const struct nfs4_state_recovery_ops *ops =
1832		clp->cl_mvops->reboot_recovery_ops;
1833	int status;
1834
1835	nfs4_begin_drain_session(clp);
1836	cred = nfs4_get_clid_cred(clp);
1837	if (cred == NULL)
1838		return -ENOENT;
1839	status = ops->establish_clid(clp, cred);
1840	put_rpccred(cred);
1841	if (status != 0)
1842		return status;
1843	pnfs_destroy_all_layouts(clp);
1844	return 0;
1845}
1846
1847/*
1848 * Returns zero or a negative errno.  NFS4ERR values are converted
1849 * to local errno values.
1850 */
1851static int nfs4_reclaim_lease(struct nfs_client *clp)
1852{
1853	int status;
1854
1855	status = nfs4_establish_lease(clp);
1856	if (status < 0)
1857		return nfs4_handle_reclaim_lease_error(clp, status);
1858	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1859		nfs4_state_start_reclaim_nograce(clp);
1860	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1861		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1862	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1863	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1864	return 0;
1865}
1866
1867static int nfs4_purge_lease(struct nfs_client *clp)
1868{
1869	int status;
1870
1871	status = nfs4_establish_lease(clp);
1872	if (status < 0)
1873		return nfs4_handle_reclaim_lease_error(clp, status);
1874	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1875	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1876	nfs4_state_start_reclaim_nograce(clp);
1877	return 0;
1878}
1879
1880/*
1881 * Try remote migration of one FSID from a source server to a
1882 * destination server.  The source server provides a list of
1883 * potential destinations.
1884 *
1885 * Returns zero or a negative NFS4ERR status code.
1886 */
1887static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1888{
1889	struct nfs_client *clp = server->nfs_client;
1890	struct nfs4_fs_locations *locations = NULL;
1891	struct inode *inode;
1892	struct page *page;
1893	int status, result;
1894
1895	dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1896			(unsigned long long)server->fsid.major,
1897			(unsigned long long)server->fsid.minor,
1898			clp->cl_hostname);
1899
1900	result = 0;
1901	page = alloc_page(GFP_KERNEL);
1902	locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1903	if (page == NULL || locations == NULL) {
1904		dprintk("<-- %s: no memory\n", __func__);
1905		goto out;
1906	}
1907
1908	inode = d_inode(server->super->s_root);
1909	result = nfs4_proc_get_locations(inode, locations, page, cred);
1910	if (result) {
1911		dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1912			__func__, result);
1913		goto out;
1914	}
1915
1916	result = -NFS4ERR_NXIO;
1917	if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1918		dprintk("<-- %s: No fs_locations data, migration skipped\n",
1919			__func__);
1920		goto out;
1921	}
1922
1923	nfs4_begin_drain_session(clp);
1924
1925	status = nfs4_replace_transport(server, locations);
1926	if (status != 0) {
1927		dprintk("<-- %s: failed to replace transport: %d\n",
1928			__func__, status);
1929		goto out;
1930	}
1931
1932	result = 0;
1933	dprintk("<-- %s: migration succeeded\n", __func__);
1934
1935out:
1936	if (page != NULL)
1937		__free_page(page);
1938	kfree(locations);
1939	if (result) {
1940		pr_err("NFS: migration recovery failed (server %s)\n",
1941				clp->cl_hostname);
1942		set_bit(NFS_MIG_FAILED, &server->mig_status);
1943	}
1944	return result;
1945}
1946
1947/*
1948 * Returns zero or a negative NFS4ERR status code.
1949 */
1950static int nfs4_handle_migration(struct nfs_client *clp)
1951{
1952	const struct nfs4_state_maintenance_ops *ops =
1953				clp->cl_mvops->state_renewal_ops;
1954	struct nfs_server *server;
1955	struct rpc_cred *cred;
1956
1957	dprintk("%s: migration reported on \"%s\"\n", __func__,
1958			clp->cl_hostname);
1959
1960	spin_lock(&clp->cl_lock);
1961	cred = ops->get_state_renewal_cred_locked(clp);
1962	spin_unlock(&clp->cl_lock);
1963	if (cred == NULL)
1964		return -NFS4ERR_NOENT;
1965
1966	clp->cl_mig_gen++;
1967restart:
1968	rcu_read_lock();
1969	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1970		int status;
1971
1972		if (server->mig_gen == clp->cl_mig_gen)
1973			continue;
1974		server->mig_gen = clp->cl_mig_gen;
1975
1976		if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1977						&server->mig_status))
1978			continue;
1979
1980		rcu_read_unlock();
1981		status = nfs4_try_migration(server, cred);
1982		if (status < 0) {
1983			put_rpccred(cred);
1984			return status;
1985		}
1986		goto restart;
1987	}
1988	rcu_read_unlock();
1989	put_rpccred(cred);
1990	return 0;
1991}
1992
1993/*
1994 * Test each nfs_server on the clp's cl_superblocks list to see
1995 * if it's moved to another server.  Stop when the server no longer
1996 * returns NFS4ERR_LEASE_MOVED.
1997 */
1998static int nfs4_handle_lease_moved(struct nfs_client *clp)
1999{
2000	const struct nfs4_state_maintenance_ops *ops =
2001				clp->cl_mvops->state_renewal_ops;
2002	struct nfs_server *server;
2003	struct rpc_cred *cred;
2004
2005	dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2006			clp->cl_hostname);
2007
2008	spin_lock(&clp->cl_lock);
2009	cred = ops->get_state_renewal_cred_locked(clp);
2010	spin_unlock(&clp->cl_lock);
2011	if (cred == NULL)
2012		return -NFS4ERR_NOENT;
2013
2014	clp->cl_mig_gen++;
2015restart:
2016	rcu_read_lock();
2017	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2018		struct inode *inode;
2019		int status;
2020
2021		if (server->mig_gen == clp->cl_mig_gen)
2022			continue;
2023		server->mig_gen = clp->cl_mig_gen;
2024
2025		rcu_read_unlock();
2026
2027		inode = d_inode(server->super->s_root);
2028		status = nfs4_proc_fsid_present(inode, cred);
2029		if (status != -NFS4ERR_MOVED)
2030			goto restart;	/* wasn't this one */
2031		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2032			goto restart;	/* there are more */
2033		goto out;
2034	}
2035	rcu_read_unlock();
2036
2037out:
2038	put_rpccred(cred);
2039	return 0;
2040}
2041
2042/**
2043 * nfs4_discover_server_trunking - Detect server IP address trunking
2044 *
2045 * @clp: nfs_client under test
2046 * @result: OUT: found nfs_client, or clp
2047 *
2048 * Returns zero or a negative errno.  If zero is returned,
2049 * an nfs_client pointer is planted in "result".
2050 *
2051 * Note: since we are invoked in process context, and
2052 * not from inside the state manager, we cannot use
2053 * nfs4_handle_reclaim_lease_error().
2054 */
2055int nfs4_discover_server_trunking(struct nfs_client *clp,
2056				  struct nfs_client **result)
2057{
2058	const struct nfs4_state_recovery_ops *ops =
2059				clp->cl_mvops->reboot_recovery_ops;
2060	struct rpc_clnt *clnt;
2061	struct rpc_cred *cred;
2062	int i, status;
2063
2064	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2065
2066	clnt = clp->cl_rpcclient;
2067	i = 0;
2068
2069	mutex_lock(&nfs_clid_init_mutex);
2070again:
2071	status  = -ENOENT;
2072	cred = nfs4_get_clid_cred(clp);
2073	if (cred == NULL)
2074		goto out_unlock;
2075
2076	status = ops->detect_trunking(clp, result, cred);
2077	put_rpccred(cred);
2078	switch (status) {
2079	case 0:
2080		break;
2081	case -ETIMEDOUT:
2082		if (clnt->cl_softrtry)
2083			break;
2084	case -NFS4ERR_DELAY:
2085	case -EAGAIN:
2086		ssleep(1);
2087	case -NFS4ERR_STALE_CLIENTID:
2088		dprintk("NFS: %s after status %d, retrying\n",
2089			__func__, status);
2090		goto again;
2091	case -EACCES:
2092		if (i++ == 0) {
2093			nfs4_root_machine_cred(clp);
2094			goto again;
2095		}
2096		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2097			break;
2098	case -NFS4ERR_CLID_INUSE:
2099	case -NFS4ERR_WRONGSEC:
2100		/* No point in retrying if we already used RPC_AUTH_UNIX */
2101		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2102			status = -EPERM;
2103			break;
2104		}
2105		clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2106		if (IS_ERR(clnt)) {
2107			status = PTR_ERR(clnt);
2108			break;
2109		}
2110		/* Note: this is safe because we haven't yet marked the
2111		 * client as ready, so we are the only user of
2112		 * clp->cl_rpcclient
2113		 */
2114		clnt = xchg(&clp->cl_rpcclient, clnt);
2115		rpc_shutdown_client(clnt);
2116		clnt = clp->cl_rpcclient;
2117		goto again;
2118
2119	case -NFS4ERR_MINOR_VERS_MISMATCH:
2120		status = -EPROTONOSUPPORT;
2121		break;
2122
2123	case -EKEYEXPIRED:
2124	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2125				 * in nfs4_exchange_id */
2126		status = -EKEYEXPIRED;
2127		break;
2128	default:
2129		pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2130				__func__, status);
2131		status = -EIO;
2132	}
2133
2134out_unlock:
2135	mutex_unlock(&nfs_clid_init_mutex);
2136	dprintk("NFS: %s: status = %d\n", __func__, status);
2137	return status;
2138}
2139
2140#ifdef CONFIG_NFS_V4_1
2141void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2142{
2143	struct nfs_client *clp = session->clp;
2144
2145	switch (err) {
2146	default:
2147		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2148		break;
2149	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2150		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2151	}
2152	nfs4_schedule_lease_recovery(clp);
2153}
2154EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2155
2156void nfs41_notify_server(struct nfs_client *clp)
2157{
2158	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
2159	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2160	nfs4_schedule_state_manager(clp);
2161}
2162
2163static void nfs4_reset_all_state(struct nfs_client *clp)
2164{
2165	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2166		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2167		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2168		nfs4_state_start_reclaim_nograce(clp);
2169		dprintk("%s: scheduling reset of all state for server %s!\n",
2170				__func__, clp->cl_hostname);
2171		nfs4_schedule_state_manager(clp);
2172	}
2173}
2174
2175static void nfs41_handle_server_reboot(struct nfs_client *clp)
2176{
2177	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2178		nfs4_state_start_reclaim_reboot(clp);
2179		dprintk("%s: server %s rebooted!\n", __func__,
2180				clp->cl_hostname);
2181		nfs4_schedule_state_manager(clp);
2182	}
2183}
2184
2185static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2186{
2187	nfs4_reset_all_state(clp);
2188	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2189}
2190
2191static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2192{
2193	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
2194	nfs4_schedule_state_manager(clp);
2195
2196	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2197}
2198
2199static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2200{
2201	/* FIXME: For now, we destroy all layouts. */
2202	pnfs_destroy_all_layouts(clp);
2203	/* FIXME: For now, we test all delegations+open state+locks. */
2204	nfs41_handle_some_state_revoked(clp);
2205	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2206			clp->cl_hostname);
2207}
2208
2209static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2210{
2211	set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2212	nfs4_schedule_state_manager(clp);
2213
2214	dprintk("%s: server %s declared a backchannel fault\n", __func__,
2215			clp->cl_hostname);
2216}
2217
2218static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2219{
2220	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2221		&clp->cl_state) == 0)
2222		nfs4_schedule_state_manager(clp);
2223}
2224
2225void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2226{
2227	if (!flags)
2228		return;
2229
2230	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2231		__func__, clp->cl_hostname, clp->cl_clientid, flags);
2232
2233	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2234		nfs41_handle_server_reboot(clp);
2235	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2236		nfs41_handle_all_state_revoked(clp);
2237	if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2238			    SEQ4_STATUS_ADMIN_STATE_REVOKED))
2239		nfs41_handle_some_state_revoked(clp);
2240	if (flags & SEQ4_STATUS_LEASE_MOVED)
2241		nfs4_schedule_lease_moved_recovery(clp);
2242	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2243		nfs41_handle_recallable_state_revoked(clp);
2244	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2245		nfs41_handle_backchannel_fault(clp);
2246	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2247				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2248		nfs41_handle_cb_path_down(clp);
2249}
2250
2251static int nfs4_reset_session(struct nfs_client *clp)
2252{
2253	struct rpc_cred *cred;
2254	int status;
2255
2256	if (!nfs4_has_session(clp))
2257		return 0;
2258	nfs4_begin_drain_session(clp);
2259	cred = nfs4_get_clid_cred(clp);
2260	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2261	switch (status) {
2262	case 0:
2263	case -NFS4ERR_BADSESSION:
2264	case -NFS4ERR_DEADSESSION:
2265		break;
2266	case -NFS4ERR_BACK_CHAN_BUSY:
2267	case -NFS4ERR_DELAY:
2268		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2269		status = 0;
2270		ssleep(1);
2271		goto out;
2272	default:
2273		status = nfs4_recovery_handle_error(clp, status);
2274		goto out;
2275	}
2276
2277	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2278	status = nfs4_proc_create_session(clp, cred);
2279	if (status) {
2280		dprintk("%s: session reset failed with status %d for server %s!\n",
2281			__func__, status, clp->cl_hostname);
2282		status = nfs4_handle_reclaim_lease_error(clp, status);
2283		goto out;
2284	}
2285	nfs41_finish_session_reset(clp);
2286	dprintk("%s: session reset was successful for server %s!\n",
2287			__func__, clp->cl_hostname);
2288out:
2289	if (cred)
2290		put_rpccred(cred);
2291	return status;
2292}
2293
2294static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2295{
2296	struct rpc_cred *cred;
2297	int ret;
2298
2299	if (!nfs4_has_session(clp))
2300		return 0;
2301	nfs4_begin_drain_session(clp);
2302	cred = nfs4_get_clid_cred(clp);
2303	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2304	if (cred)
2305		put_rpccred(cred);
2306	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2307	switch (ret) {
2308	case 0:
2309		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2310			__func__, clp->cl_hostname);
2311		break;
2312	case -NFS4ERR_DELAY:
2313		ssleep(1);
2314		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2315		break;
2316	default:
2317		return nfs4_recovery_handle_error(clp, ret);
2318	}
2319	return 0;
2320}
2321#else /* CONFIG_NFS_V4_1 */
2322static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2323
2324static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2325{
2326	return 0;
2327}
2328#endif /* CONFIG_NFS_V4_1 */
2329
2330static void nfs4_state_manager(struct nfs_client *clp)
2331{
2332	int status = 0;
2333	const char *section = "", *section_sep = "";
2334
2335	/* Ensure exclusive access to NFSv4 state */
2336	do {
2337		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2338			section = "purge state";
2339			status = nfs4_purge_lease(clp);
2340			if (status < 0)
2341				goto out_error;
2342			continue;
2343		}
2344
2345		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2346			section = "lease expired";
2347			/* We're going to have to re-establish a clientid */
2348			status = nfs4_reclaim_lease(clp);
2349			if (status < 0)
2350				goto out_error;
2351			continue;
2352		}
2353
2354		/* Initialize or reset the session */
2355		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2356			section = "reset session";
2357			status = nfs4_reset_session(clp);
2358			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2359				continue;
2360			if (status < 0)
2361				goto out_error;
2362		}
2363
2364		/* Send BIND_CONN_TO_SESSION */
2365		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2366				&clp->cl_state)) {
2367			section = "bind conn to session";
2368			status = nfs4_bind_conn_to_session(clp);
2369			if (status < 0)
2370				goto out_error;
2371			continue;
2372		}
2373
2374		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2375			section = "check lease";
2376			status = nfs4_check_lease(clp);
2377			if (status < 0)
2378				goto out_error;
2379			continue;
2380		}
2381
2382		if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2383			section = "migration";
2384			status = nfs4_handle_migration(clp);
2385			if (status < 0)
2386				goto out_error;
2387		}
2388
2389		if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2390			section = "lease moved";
2391			status = nfs4_handle_lease_moved(clp);
2392			if (status < 0)
2393				goto out_error;
2394		}
2395
2396		/* First recover reboot state... */
2397		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2398			section = "reclaim reboot";
2399			status = nfs4_do_reclaim(clp,
2400				clp->cl_mvops->reboot_recovery_ops);
2401			if (status == -EAGAIN)
2402				continue;
2403			if (status < 0)
2404				goto out_error;
2405			nfs4_state_end_reclaim_reboot(clp);
2406		}
2407
2408		/* Now recover expired state... */
2409		if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2410			section = "reclaim nograce";
2411			status = nfs4_do_reclaim(clp,
2412				clp->cl_mvops->nograce_recovery_ops);
2413			if (status == -EAGAIN)
2414				continue;
2415			if (status < 0)
2416				goto out_error;
2417		}
2418
2419		nfs4_end_drain_session(clp);
2420		if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2421			nfs_client_return_marked_delegations(clp);
2422			continue;
2423		}
2424
2425		nfs4_clear_state_manager_bit(clp);
2426		/* Did we race with an attempt to give us more work? */
2427		if (clp->cl_state == 0)
2428			break;
2429		if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2430			break;
2431	} while (atomic_read(&clp->cl_count) > 1);
2432	return;
2433out_error:
2434	if (strlen(section))
2435		section_sep = ": ";
2436	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2437			" with error %d\n", section_sep, section,
2438			clp->cl_hostname, -status);
2439	ssleep(1);
2440	nfs4_end_drain_session(clp);
2441	nfs4_clear_state_manager_bit(clp);
2442}
2443
2444static int nfs4_run_state_manager(void *ptr)
2445{
2446	struct nfs_client *clp = ptr;
2447
2448	allow_signal(SIGKILL);
2449	nfs4_state_manager(clp);
2450	nfs_put_client(clp);
2451	module_put_and_exit(0);
2452	return 0;
2453}
2454
2455/*
2456 * Local variables:
2457 *  c-basic-offset: 8
2458 * End:
2459 */
2460