Linux Audio

Check our new training course

Loading...
v6.13.7
   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 = atomic64_inc_return(&server->owner_ctr);
 505	sp->so_server = server;
 506	sp->so_cred = get_cred(cred);
 507	spin_lock_init(&sp->so_lock);
 508	INIT_LIST_HEAD(&sp->so_states);
 509	nfs4_init_seqid_counter(&sp->so_seqid);
 510	atomic_set(&sp->so_count, 1);
 511	INIT_LIST_HEAD(&sp->so_lru);
 512	mutex_init(&sp->so_delegreturn_mutex);
 513	return sp;
 514}
 515
 516static void
 517nfs4_reset_state_owner(struct nfs4_state_owner *sp)
 518{
 519	/* This state_owner is no longer usable, but must
 520	 * remain in place so that state recovery can find it
 521	 * and the opens associated with it.
 522	 * It may also be used for new 'open' request to
 523	 * return a delegation to the server.
 524	 * So update the 'create_time' so that it looks like
 525	 * a new state_owner.  This will cause the server to
 526	 * request an OPEN_CONFIRM to start a new sequence.
 527	 */
 528	sp->so_seqid.create_time = ktime_get();
 
 
 
 529}
 530
 531static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 532{
 533	nfs4_destroy_seqid_counter(&sp->so_seqid);
 534	put_cred(sp->so_cred);
 535	kfree(sp);
 536}
 537
 538static void nfs4_gc_state_owners(struct nfs_server *server)
 539{
 540	struct nfs_client *clp = server->nfs_client;
 541	struct nfs4_state_owner *sp, *tmp;
 542	unsigned long time_min, time_max;
 543	LIST_HEAD(doomed);
 544
 545	spin_lock(&clp->cl_lock);
 546	time_max = jiffies;
 547	time_min = (long)time_max - (long)clp->cl_lease_time;
 548	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 549		/* NB: LRU is sorted so that oldest is at the head */
 550		if (time_in_range(sp->so_expires, time_min, time_max))
 551			break;
 552		list_move(&sp->so_lru, &doomed);
 553		nfs4_remove_state_owner_locked(sp);
 554	}
 555	spin_unlock(&clp->cl_lock);
 556
 557	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 558		list_del(&sp->so_lru);
 559		nfs4_free_state_owner(sp);
 560	}
 561}
 562
 563/**
 564 * nfs4_get_state_owner - Look up a state owner given a credential
 565 * @server: nfs_server to search
 566 * @cred: RPC credential to match
 567 * @gfp_flags: allocation mode
 568 *
 569 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 570 */
 571struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 572					      const struct cred *cred,
 573					      gfp_t gfp_flags)
 574{
 575	struct nfs_client *clp = server->nfs_client;
 576	struct nfs4_state_owner *sp, *new;
 577
 578	spin_lock(&clp->cl_lock);
 579	sp = nfs4_find_state_owner_locked(server, cred);
 580	spin_unlock(&clp->cl_lock);
 581	if (sp != NULL)
 582		goto out;
 583	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 584	if (new == NULL)
 585		goto out;
 586	spin_lock(&clp->cl_lock);
 587	sp = nfs4_insert_state_owner_locked(new);
 588	spin_unlock(&clp->cl_lock);
 
 
 
 
 589	if (sp != new)
 590		nfs4_free_state_owner(new);
 591out:
 592	nfs4_gc_state_owners(server);
 593	return sp;
 594}
 595
 596/**
 597 * nfs4_put_state_owner - Release a nfs4_state_owner
 598 * @sp: state owner data to release
 599 *
 600 * Note that we keep released state owners on an LRU
 601 * list.
 602 * This caches valid state owners so that they can be
 603 * reused, to avoid the OPEN_CONFIRM on minor version 0.
 604 * It also pins the uniquifier of dropped state owners for
 605 * a while, to ensure that those state owner names are
 606 * never reused.
 607 */
 608void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 609{
 610	struct nfs_server *server = sp->so_server;
 611	struct nfs_client *clp = server->nfs_client;
 612
 613	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 614		return;
 615
 616	sp->so_expires = jiffies;
 617	list_add_tail(&sp->so_lru, &server->state_owners_lru);
 618	spin_unlock(&clp->cl_lock);
 619}
 620
 621/**
 622 * nfs4_purge_state_owners - Release all cached state owners
 623 * @server: nfs_server with cached state owners to release
 624 * @head: resulting list of state owners
 625 *
 626 * Called at umount time.  Remaining state owners will be on
 627 * the LRU with ref count of zero.
 628 * Note that the state owners are not freed, but are added
 629 * to the list @head, which can later be used as an argument
 630 * to nfs4_free_state_owners.
 631 */
 632void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
 633{
 634	struct nfs_client *clp = server->nfs_client;
 635	struct nfs4_state_owner *sp, *tmp;
 
 636
 637	spin_lock(&clp->cl_lock);
 638	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 639		list_move(&sp->so_lru, head);
 640		nfs4_remove_state_owner_locked(sp);
 641	}
 642	spin_unlock(&clp->cl_lock);
 643}
 644
 645/**
 646 * nfs4_free_state_owners - Release all cached state owners
 647 * @head: resulting list of state owners
 648 *
 649 * Frees a list of state owners that was generated by
 650 * nfs4_purge_state_owners
 651 */
 652void nfs4_free_state_owners(struct list_head *head)
 653{
 654	struct nfs4_state_owner *sp, *tmp;
 655
 656	list_for_each_entry_safe(sp, tmp, head, so_lru) {
 657		list_del(&sp->so_lru);
 658		nfs4_free_state_owner(sp);
 659	}
 660}
 661
 662static struct nfs4_state *
 663nfs4_alloc_open_state(void)
 664{
 665	struct nfs4_state *state;
 666
 667	state = kzalloc(sizeof(*state), GFP_KERNEL_ACCOUNT);
 668	if (!state)
 669		return NULL;
 670	refcount_set(&state->count, 1);
 671	INIT_LIST_HEAD(&state->lock_states);
 672	spin_lock_init(&state->state_lock);
 673	seqlock_init(&state->seqlock);
 674	init_waitqueue_head(&state->waitq);
 675	return state;
 676}
 677
 678void
 679nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 680{
 681	if (state->state == fmode)
 682		return;
 683	/* NB! List reordering - see the reclaim code for why.  */
 684	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 685		if (fmode & FMODE_WRITE)
 686			list_move(&state->open_states, &state->owner->so_states);
 687		else
 688			list_move_tail(&state->open_states, &state->owner->so_states);
 689	}
 690	state->state = fmode;
 691}
 692
 693static struct nfs4_state *
 694__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 695{
 696	struct nfs_inode *nfsi = NFS_I(inode);
 697	struct nfs4_state *state;
 698
 699	list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
 700		if (state->owner != owner)
 701			continue;
 702		if (!nfs4_valid_open_stateid(state))
 703			continue;
 704		if (refcount_inc_not_zero(&state->count))
 705			return state;
 706	}
 707	return NULL;
 708}
 709
 710static void
 711nfs4_free_open_state(struct nfs4_state *state)
 712{
 713	kfree_rcu(state, rcu_head);
 714}
 715
 716struct nfs4_state *
 717nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 718{
 719	struct nfs4_state *state, *new;
 720	struct nfs_inode *nfsi = NFS_I(inode);
 721
 722	rcu_read_lock();
 723	state = __nfs4_find_state_byowner(inode, owner);
 724	rcu_read_unlock();
 725	if (state)
 726		goto out;
 727	new = nfs4_alloc_open_state();
 728	spin_lock(&owner->so_lock);
 729	spin_lock(&inode->i_lock);
 730	state = __nfs4_find_state_byowner(inode, owner);
 731	if (state == NULL && new != NULL) {
 732		state = new;
 733		state->owner = owner;
 734		atomic_inc(&owner->so_count);
 
 735		ihold(inode);
 736		state->inode = inode;
 737		list_add_rcu(&state->inode_states, &nfsi->open_states);
 738		spin_unlock(&inode->i_lock);
 739		/* Note: The reclaim code dictates that we add stateless
 740		 * and read-only stateids to the end of the list */
 741		list_add_tail(&state->open_states, &owner->so_states);
 742		spin_unlock(&owner->so_lock);
 743	} else {
 744		spin_unlock(&inode->i_lock);
 745		spin_unlock(&owner->so_lock);
 746		if (new)
 747			nfs4_free_open_state(new);
 748	}
 749out:
 750	return state;
 751}
 752
 753void nfs4_put_open_state(struct nfs4_state *state)
 754{
 755	struct inode *inode = state->inode;
 756	struct nfs4_state_owner *owner = state->owner;
 757
 758	if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
 759		return;
 760	spin_lock(&inode->i_lock);
 761	list_del_rcu(&state->inode_states);
 762	list_del(&state->open_states);
 763	spin_unlock(&inode->i_lock);
 764	spin_unlock(&owner->so_lock);
 765	nfs4_inode_return_delegation_on_close(inode);
 766	iput(inode);
 767	nfs4_free_open_state(state);
 768	nfs4_put_state_owner(owner);
 769}
 770
 771/*
 772 * Close the current file.
 773 */
 774static void __nfs4_close(struct nfs4_state *state,
 775		fmode_t fmode, gfp_t gfp_mask, int wait)
 776{
 777	struct nfs4_state_owner *owner = state->owner;
 778	int call_close = 0;
 779	fmode_t newstate;
 780
 781	atomic_inc(&owner->so_count);
 782	/* Protect against nfs4_find_state() */
 783	spin_lock(&owner->so_lock);
 784	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 785		case FMODE_READ:
 786			state->n_rdonly--;
 787			break;
 788		case FMODE_WRITE:
 789			state->n_wronly--;
 790			break;
 791		case FMODE_READ|FMODE_WRITE:
 792			state->n_rdwr--;
 793	}
 794	newstate = FMODE_READ|FMODE_WRITE;
 795	if (state->n_rdwr == 0) {
 796		if (state->n_rdonly == 0) {
 797			newstate &= ~FMODE_READ;
 798			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 799			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 800		}
 801		if (state->n_wronly == 0) {
 802			newstate &= ~FMODE_WRITE;
 803			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 804			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 805		}
 806		if (newstate == 0)
 807			clear_bit(NFS_DELEGATED_STATE, &state->flags);
 808	}
 809	nfs4_state_set_mode_locked(state, newstate);
 810	spin_unlock(&owner->so_lock);
 811
 812	if (!call_close) {
 813		nfs4_put_open_state(state);
 814		nfs4_put_state_owner(owner);
 815	} else
 816		nfs4_do_close(state, gfp_mask, wait);
 
 
 
 817}
 818
 819void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 820{
 821	__nfs4_close(state, fmode, GFP_KERNEL, 0);
 822}
 823
 824void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 825{
 826	__nfs4_close(state, fmode, GFP_KERNEL, 1);
 827}
 828
 829/*
 830 * Search the state->lock_states for an existing lock_owner
 831 * that is compatible with either of the given owners.
 832 * If the second is non-zero, then the first refers to a Posix-lock
 833 * owner (current->files) and the second refers to a flock/OFD
 834 * owner (struct file*).  In that case, prefer a match for the first
 835 * owner.
 836 * If both sorts of locks are held on the one file we cannot know
 837 * which stateid was intended to be used, so a "correct" choice cannot
 838 * be made.  Failing that, a "consistent" choice is preferable.  The
 839 * consistent choice we make is to prefer the first owner, that of a
 840 * Posix lock.
 841 */
 842static struct nfs4_lock_state *
 843__nfs4_find_lock_state(struct nfs4_state *state,
 844		       fl_owner_t owner, fl_owner_t owner2)
 845{
 846	struct nfs4_lock_state *pos, *ret = NULL;
 847	list_for_each_entry(pos, &state->lock_states, ls_locks) {
 848		if (pos->ls_owner == owner) {
 849			ret = pos;
 
 
 
 
 850			break;
 
 
 
 851		}
 852		if (pos->ls_owner == owner2)
 853			ret = pos;
 854	}
 855	if (ret)
 856		refcount_inc(&ret->ls_count);
 857	return ret;
 858}
 859
 860/*
 861 * Return a compatible lock_state. If no initialized lock_state structure
 862 * exists, return an uninitialized one.
 863 *
 864 */
 865static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t owner)
 866{
 867	struct nfs4_lock_state *lsp;
 868	struct nfs_server *server = state->owner->so_server;
 869
 870	lsp = kzalloc(sizeof(*lsp), GFP_KERNEL_ACCOUNT);
 871	if (lsp == NULL)
 872		return NULL;
 873	nfs4_init_seqid_counter(&lsp->ls_seqid);
 874	refcount_set(&lsp->ls_count, 1);
 875	lsp->ls_state = state;
 876	lsp->ls_owner = owner;
 877	lsp->ls_seqid.owner_id = atomic64_inc_return(&server->owner_ctr);
 
 
 
 
 
 
 
 
 
 
 
 
 878	INIT_LIST_HEAD(&lsp->ls_locks);
 879	return lsp;
 
 
 
 880}
 881
 882void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 883{
 
 884	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 885	kfree(lsp);
 886}
 887
 888/*
 889 * Return a compatible lock_state. If no initialized lock_state structure
 890 * exists, return an uninitialized one.
 891 *
 892 */
 893static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 894{
 895	struct nfs4_lock_state *lsp, *new = NULL;
 896	
 897	for(;;) {
 898		spin_lock(&state->state_lock);
 899		lsp = __nfs4_find_lock_state(state, owner, NULL);
 900		if (lsp != NULL)
 901			break;
 902		if (new != NULL) {
 903			list_add(&new->ls_locks, &state->lock_states);
 904			set_bit(LK_STATE_IN_USE, &state->flags);
 905			lsp = new;
 906			new = NULL;
 907			break;
 908		}
 909		spin_unlock(&state->state_lock);
 910		new = nfs4_alloc_lock_state(state, owner);
 911		if (new == NULL)
 912			return NULL;
 913	}
 914	spin_unlock(&state->state_lock);
 915	if (new != NULL)
 916		nfs4_free_lock_state(state->owner->so_server, new);
 917	return lsp;
 918}
 919
 920/*
 921 * Release reference to lock_state, and free it if we see that
 922 * it is no longer in use
 923 */
 924void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 925{
 926	struct nfs_server *server;
 927	struct nfs4_state *state;
 928
 929	if (lsp == NULL)
 930		return;
 931	state = lsp->ls_state;
 932	if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
 933		return;
 934	list_del(&lsp->ls_locks);
 935	if (list_empty(&state->lock_states))
 936		clear_bit(LK_STATE_IN_USE, &state->flags);
 937	spin_unlock(&state->state_lock);
 938	server = state->owner->so_server;
 939	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 940		struct nfs_client *clp = server->nfs_client;
 941
 942		clp->cl_mvops->free_lock_state(server, lsp);
 943	} else
 944		nfs4_free_lock_state(server, lsp);
 945}
 946
 947static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 948{
 949	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 950
 951	dst->fl_u.nfs4_fl.owner = lsp;
 952	refcount_inc(&lsp->ls_count);
 953}
 954
 955static void nfs4_fl_release_lock(struct file_lock *fl)
 956{
 957	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 958}
 959
 960static const struct file_lock_operations nfs4_fl_lock_ops = {
 961	.fl_copy_lock = nfs4_fl_copy_lock,
 962	.fl_release_private = nfs4_fl_release_lock,
 963};
 964
 965int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 966{
 967	struct nfs4_lock_state *lsp;
 968
 969	if (fl->fl_ops != NULL)
 970		return 0;
 971	lsp = nfs4_get_lock_state(state, fl->c.flc_owner);
 
 
 
 
 
 
 972	if (lsp == NULL)
 973		return -ENOMEM;
 974	fl->fl_u.nfs4_fl.owner = lsp;
 975	fl->fl_ops = &nfs4_fl_lock_ops;
 976	return 0;
 977}
 978
 979static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 980		struct nfs4_state *state,
 981		const struct nfs_lock_context *l_ctx)
 982{
 983	struct nfs4_lock_state *lsp;
 984	fl_owner_t owner, fl_flock_owner;
 985	int ret = -ENOENT;
 986
 987	if (l_ctx == NULL)
 988		goto out;
 989
 990	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 991		goto out;
 992
 993	owner = l_ctx->lockowner;
 994	fl_flock_owner = l_ctx->open_context->flock_owner;
 995
 996	spin_lock(&state->state_lock);
 997	lsp = __nfs4_find_lock_state(state, owner, fl_flock_owner);
 998	if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 999		ret = -EIO;
1000	else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1001		nfs4_stateid_copy(dst, &lsp->ls_stateid);
1002		ret = 0;
1003	}
1004	spin_unlock(&state->state_lock);
1005	nfs4_put_lock_state(lsp);
1006out:
1007	return ret;
1008}
1009
1010bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1011{
1012	bool ret;
1013	const nfs4_stateid *src;
1014	int seq;
1015
1016	do {
1017		ret = false;
1018		src = &zero_stateid;
1019		seq = read_seqbegin(&state->seqlock);
1020		if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1021			src = &state->open_stateid;
1022			ret = true;
1023		}
1024		nfs4_stateid_copy(dst, src);
1025	} while (read_seqretry(&state->seqlock, seq));
1026	return ret;
1027}
1028
1029/*
1030 * Byte-range lock aware utility to initialize the stateid of read/write
1031 * requests.
1032 */
1033int nfs4_select_rw_stateid(struct nfs4_state *state,
1034		fmode_t fmode, const struct nfs_lock_context *l_ctx,
1035		nfs4_stateid *dst, const struct cred **cred)
1036{
1037	int ret;
1038
1039	if (!nfs4_valid_open_stateid(state))
1040		return -EIO;
1041	if (cred != NULL)
1042		*cred = NULL;
1043	ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1044	if (ret == -EIO)
1045		/* A lost lock - don't even consider delegations */
1046		goto out;
1047	/* returns true if delegation stateid found and copied */
1048	if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1049		ret = 0;
1050		goto out;
1051	}
1052	if (ret != -ENOENT)
1053		/* nfs4_copy_delegation_stateid() didn't over-write
1054		 * dst, so it still has the lock stateid which we now
1055		 * choose to use.
1056		 */
1057		goto out;
1058	ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1059out:
1060	if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1061		dst->seqid = 0;
1062	return ret;
1063}
1064
1065struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1066{
1067	struct nfs_seqid *new;
1068
1069	new = kmalloc(sizeof(*new), gfp_mask);
1070	if (new == NULL)
1071		return ERR_PTR(-ENOMEM);
1072	new->sequence = counter;
1073	INIT_LIST_HEAD(&new->list);
1074	new->task = NULL;
1075	return new;
1076}
1077
1078void nfs_release_seqid(struct nfs_seqid *seqid)
1079{
1080	struct nfs_seqid_counter *sequence;
1081
1082	if (seqid == NULL || list_empty(&seqid->list))
1083		return;
1084	sequence = seqid->sequence;
1085	spin_lock(&sequence->lock);
1086	if (list_is_first(&seqid->list, &sequence->list) &&
1087	    !list_is_singular(&sequence->list)) {
1088		struct nfs_seqid *next = list_next_entry(seqid, list);
 
 
 
1089		rpc_wake_up_queued_task(&sequence->wait, next->task);
1090	}
1091	list_del_init(&seqid->list);
1092	spin_unlock(&sequence->lock);
1093}
1094
1095void nfs_free_seqid(struct nfs_seqid *seqid)
1096{
1097	nfs_release_seqid(seqid);
1098	kfree(seqid);
1099}
1100
1101/*
1102 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1103 * failed with a seqid incrementing error -
1104 * see comments nfs4.h:seqid_mutating_error()
1105 */
1106static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1107{
 
1108	switch (status) {
1109		case 0:
1110			break;
1111		case -NFS4ERR_BAD_SEQID:
1112			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1113				return;
1114			pr_warn_ratelimited("NFS: v4 server returned a bad"
1115					" sequence-id error on an"
1116					" unconfirmed sequence %p!\n",
1117					seqid->sequence);
1118			return;
1119		case -NFS4ERR_STALE_CLIENTID:
1120		case -NFS4ERR_STALE_STATEID:
1121		case -NFS4ERR_BAD_STATEID:
1122		case -NFS4ERR_BADXDR:
1123		case -NFS4ERR_RESOURCE:
1124		case -NFS4ERR_NOFILEHANDLE:
1125		case -NFS4ERR_MOVED:
1126			/* Non-seqid mutating errors */
1127			return;
1128	}
1129	/*
1130	 * Note: no locking needed as we are guaranteed to be first
1131	 * on the sequence list
1132	 */
1133	seqid->sequence->counter++;
1134}
1135
1136void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1137{
1138	struct nfs4_state_owner *sp;
1139
1140	if (seqid == NULL)
1141		return;
1142
1143	sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1144	if (status == -NFS4ERR_BAD_SEQID)
1145		nfs4_reset_state_owner(sp);
1146	if (!nfs4_has_session(sp->so_server->nfs_client))
1147		nfs_increment_seqid(status, seqid);
1148}
1149
1150/*
1151 * Increment the seqid if the LOCK/LOCKU succeeded, or
1152 * failed with a seqid incrementing error -
1153 * see comments nfs4.h:seqid_mutating_error()
1154 */
1155void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1156{
1157	if (seqid != NULL)
1158		nfs_increment_seqid(status, seqid);
1159}
1160
1161int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1162{
1163	struct nfs_seqid_counter *sequence;
1164	int status = 0;
1165
1166	if (seqid == NULL)
1167		goto out;
1168	sequence = seqid->sequence;
1169	spin_lock(&sequence->lock);
1170	seqid->task = task;
1171	if (list_empty(&seqid->list))
1172		list_add_tail(&seqid->list, &sequence->list);
1173	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1174		goto unlock;
1175	rpc_sleep_on(&sequence->wait, task, NULL);
1176	status = -EAGAIN;
1177unlock:
1178	spin_unlock(&sequence->lock);
1179out:
1180	return status;
1181}
1182
1183static int nfs4_run_state_manager(void *);
1184
1185static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1186{
1187	clear_and_wake_up_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 
 
 
1188	rpc_wake_up(&clp->cl_rpcwaitq);
1189}
1190
1191/*
1192 * Schedule the nfs_client asynchronous state management routine
1193 */
1194void nfs4_schedule_state_manager(struct nfs_client *clp)
1195{
1196	struct task_struct *task;
1197	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1198	struct rpc_clnt *clnt = clp->cl_rpcclient;
1199	bool swapon = false;
1200
1201	if (clnt->cl_shutdown)
1202		return;
1203
1204	set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1205
1206	if (atomic_read(&clnt->cl_swapper)) {
1207		swapon = !test_and_set_bit(NFS4CLNT_MANAGER_AVAILABLE,
1208					   &clp->cl_state);
1209		if (!swapon) {
1210			wake_up_var(&clp->cl_state);
1211			return;
1212		}
1213	}
1214
1215	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1216		return;
1217
1218	__module_get(THIS_MODULE);
1219	refcount_inc(&clp->cl_count);
1220
1221	/* The rcu_read_lock() is not strictly necessary, as the state
1222	 * manager is the only thread that ever changes the rpc_xprt
1223	 * after it's initialized.  At this point, we're single threaded. */
1224	rcu_read_lock();
1225	snprintf(buf, sizeof(buf), "%s-manager",
1226			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1227	rcu_read_unlock();
1228	task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1229	if (IS_ERR(task)) {
1230		printk(KERN_ERR "%s: kthread_run: %ld\n",
1231			__func__, PTR_ERR(task));
1232		if (!nfs_client_init_is_complete(clp))
1233			nfs_mark_client_ready(clp, PTR_ERR(task));
1234		if (swapon)
1235			clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
1236		nfs4_clear_state_manager_bit(clp);
1237		nfs_put_client(clp);
1238		module_put(THIS_MODULE);
1239	}
1240}
1241
1242/*
1243 * Schedule a lease recovery attempt
1244 */
1245void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1246{
1247	if (!clp)
1248		return;
1249	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1250		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1251	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1252			clp->cl_hostname);
1253	nfs4_schedule_state_manager(clp);
1254}
1255EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1256
1257/**
1258 * nfs4_schedule_migration_recovery - trigger migration recovery
1259 *
1260 * @server: FSID that is migrating
1261 *
1262 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1263 * value is returned.
1264 */
1265int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1266{
1267	struct nfs_client *clp = server->nfs_client;
1268
1269	if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1270		pr_err("NFS: volatile file handles not supported (server %s)\n",
1271				clp->cl_hostname);
1272		return -NFS4ERR_IO;
1273	}
1274
1275	if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1276		return -NFS4ERR_IO;
1277
1278	dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1279			__func__,
1280			(unsigned long long)server->fsid.major,
1281			(unsigned long long)server->fsid.minor,
1282			clp->cl_hostname);
1283
1284	set_bit(NFS_MIG_IN_TRANSITION,
1285			&((struct nfs_server *)server)->mig_status);
1286	set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1287
1288	nfs4_schedule_state_manager(clp);
1289	return 0;
1290}
1291EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1292
1293/**
1294 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1295 *
1296 * @clp: server to check for moved leases
1297 *
1298 */
1299void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1300{
1301	dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1302		__func__, clp->cl_clientid, clp->cl_hostname);
1303
1304	set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1305	nfs4_schedule_state_manager(clp);
1306}
1307EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1308
1309int nfs4_wait_clnt_recover(struct nfs_client *clp)
1310{
1311	int res;
1312
1313	might_sleep();
1314
1315	refcount_inc(&clp->cl_count);
1316	res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1317				 nfs_wait_bit_killable,
1318				 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1319	if (res)
1320		goto out;
1321	if (clp->cl_cons_state < 0)
1322		res = clp->cl_cons_state;
1323out:
1324	nfs_put_client(clp);
1325	return res;
1326}
1327
1328int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1329{
1330	unsigned int loop;
1331	int ret;
1332
1333	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1334		ret = nfs4_wait_clnt_recover(clp);
1335		if (ret != 0)
1336			break;
1337		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1338		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1339			break;
1340		nfs4_schedule_state_manager(clp);
1341		ret = -EIO;
1342	}
1343	return ret;
1344}
1345
1346/*
1347 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1348 * @clp: client to process
1349 *
1350 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1351 * resend of the SETCLIENTID and hence re-establish the
1352 * callback channel. Then return all existing delegations.
1353 */
1354static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1355{
1356	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1357	nfs_expire_all_delegations(clp);
1358	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1359			clp->cl_hostname);
1360}
1361
1362void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1363{
1364	nfs40_handle_cb_pathdown(clp);
1365	nfs4_schedule_state_manager(clp);
1366}
1367
1368static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1369{
1370
1371	if (!nfs4_valid_open_stateid(state))
1372		return 0;
1373	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1374	/* Don't recover state that expired before the reboot */
1375	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1376		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1377		return 0;
1378	}
1379	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1380	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1381	return 1;
1382}
1383
1384int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1385{
1386	if (!nfs4_valid_open_stateid(state))
1387		return 0;
1388	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1389	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1390	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1391	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1392	return 1;
1393}
1394
1395int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1396{
1397	struct nfs_client *clp = server->nfs_client;
1398
1399	if (!nfs4_state_mark_reclaim_nograce(clp, state))
1400		return -EBADF;
1401	nfs_inode_find_delegation_state_and_recover(state->inode,
1402			&state->stateid);
1403	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1404			clp->cl_hostname);
1405	nfs4_schedule_state_manager(clp);
1406	return 0;
1407}
1408EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1409
1410static struct nfs4_lock_state *
1411nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1412		const nfs4_stateid *stateid)
1413{
1414	struct nfs4_lock_state *pos;
1415
1416	list_for_each_entry(pos, &state->lock_states, ls_locks) {
1417		if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1418			continue;
1419		if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1420			return pos;
1421	}
1422	return NULL;
1423}
1424
1425static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1426		const nfs4_stateid *stateid)
1427{
1428	bool found = false;
1429
1430	if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1431		spin_lock(&state->state_lock);
1432		if (nfs_state_find_lock_state_by_stateid(state, stateid))
1433			found = true;
1434		spin_unlock(&state->state_lock);
1435	}
1436	return found;
1437}
1438
1439void nfs_inode_find_state_and_recover(struct inode *inode,
1440		const nfs4_stateid *stateid)
1441{
1442	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1443	struct nfs_inode *nfsi = NFS_I(inode);
1444	struct nfs_open_context *ctx;
1445	struct nfs4_state *state;
1446	bool found = false;
1447
1448	rcu_read_lock();
1449	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1450		state = ctx->state;
1451		if (state == NULL)
1452			continue;
1453		if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1454		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1455			found = true;
1456			continue;
1457		}
1458		if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1459		    nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1460		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1461			found = true;
1462			continue;
1463		}
1464		if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1465		    nfs4_state_mark_reclaim_nograce(clp, state))
1466			found = true;
1467	}
1468	rcu_read_unlock();
1469
1470	nfs_inode_find_delegation_state_and_recover(inode, stateid);
1471	if (found)
1472		nfs4_schedule_state_manager(clp);
1473}
1474
1475static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1476{
1477	struct inode *inode = state->inode;
1478	struct nfs_inode *nfsi = NFS_I(inode);
1479	struct nfs_open_context *ctx;
1480
1481	rcu_read_lock();
1482	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1483		if (ctx->state != state)
1484			continue;
1485		set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1486		pr_warn("NFSv4: state recovery failed for open file %pd2, "
1487				"error = %d\n", ctx->dentry, err);
1488	}
1489	rcu_read_unlock();
1490}
1491
1492static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1493{
1494	set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1495	nfs4_state_mark_open_context_bad(state, error);
1496}
1497
1498
1499static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1500{
1501	struct inode *inode = state->inode;
1502	struct nfs_inode *nfsi = NFS_I(inode);
1503	struct file_lock *fl;
1504	struct nfs4_lock_state *lsp;
1505	int status = 0;
1506	struct file_lock_context *flctx = locks_inode_context(inode);
1507	struct list_head *list;
1508
1509	if (flctx == NULL)
1510		return 0;
1511
1512	list = &flctx->flc_posix;
1513
1514	/* Guard against delegation returns and new lock/unlock calls */
1515	down_write(&nfsi->rwsem);
1516	spin_lock(&flctx->flc_lock);
1517restart:
1518	for_each_file_lock(fl, list) {
1519		if (nfs_file_open_context(fl->c.flc_file)->state != state)
 
 
1520			continue;
1521		spin_unlock(&flctx->flc_lock);
1522		status = ops->recover_lock(state, fl);
1523		switch (status) {
1524		case 0:
1525			break;
1526		case -ETIMEDOUT:
1527		case -ESTALE:
1528		case -NFS4ERR_ADMIN_REVOKED:
1529		case -NFS4ERR_STALE_STATEID:
1530		case -NFS4ERR_BAD_STATEID:
1531		case -NFS4ERR_EXPIRED:
1532		case -NFS4ERR_NO_GRACE:
1533		case -NFS4ERR_STALE_CLIENTID:
1534		case -NFS4ERR_BADSESSION:
1535		case -NFS4ERR_BADSLOT:
1536		case -NFS4ERR_BAD_HIGH_SLOT:
1537		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1538			goto out;
1539		default:
1540			pr_err("NFS: %s: unhandled error %d\n",
1541					__func__, status);
1542			fallthrough;
1543		case -ENOMEM:
1544		case -NFS4ERR_DENIED:
1545		case -NFS4ERR_RECLAIM_BAD:
1546		case -NFS4ERR_RECLAIM_CONFLICT:
1547			lsp = fl->fl_u.nfs4_fl.owner;
1548			if (lsp)
1549				set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1550			status = 0;
1551		}
1552		spin_lock(&flctx->flc_lock);
1553	}
1554	if (list == &flctx->flc_posix) {
1555		list = &flctx->flc_flock;
1556		goto restart;
1557	}
1558	spin_unlock(&flctx->flc_lock);
1559out:
1560	up_write(&nfsi->rwsem);
1561	return status;
1562}
1563
1564#ifdef CONFIG_NFS_V4_2
1565static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1566{
1567	struct nfs4_copy_state *copy;
1568
1569	if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1570		!test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1571		return;
1572
1573	spin_lock(&sp->so_server->nfs_client->cl_lock);
1574	list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1575		if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1576				!nfs4_stateid_match_other(&state->stateid,
1577				&copy->parent_dst_state->stateid)))
1578				continue;
1579		copy->flags = 1;
1580		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1581				&state->flags)) {
1582			clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1583			complete(&copy->completion);
1584		}
1585	}
1586	list_for_each_entry(copy, &sp->so_server->ss_src_copies, src_copies) {
1587		if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1588				!nfs4_stateid_match_other(&state->stateid,
1589				&copy->parent_src_state->stateid)))
1590				continue;
1591		copy->flags = 1;
1592		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1593				&state->flags))
1594			complete(&copy->completion);
1595	}
1596	spin_unlock(&sp->so_server->nfs_client->cl_lock);
1597}
1598#else /* !CONFIG_NFS_V4_2 */
1599static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1600					 struct nfs4_state *state)
1601{
1602}
1603#endif /* CONFIG_NFS_V4_2 */
1604
1605static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1606				     const struct nfs4_state_recovery_ops *ops,
1607				     int *lost_locks)
1608{
1609	struct nfs4_lock_state *lock;
1610	int status;
1611
1612	status = ops->recover_open(sp, state);
1613	if (status < 0)
1614		return status;
1615
1616	status = nfs4_reclaim_locks(state, ops);
1617	if (status < 0)
1618		return status;
1619
1620	if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1621		spin_lock(&state->state_lock);
1622		list_for_each_entry(lock, &state->lock_states, ls_locks) {
1623			trace_nfs4_state_lock_reclaim(state, lock);
1624			if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags) &&
1625			    !test_bit(NFS_LOCK_UNLOCKING, &lock->ls_flags))
1626				*lost_locks += 1;
1627		}
1628		spin_unlock(&state->state_lock);
1629	}
1630
1631	nfs42_complete_copies(sp, state);
1632	clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1633	return status;
1634}
1635
1636static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp,
1637				   const struct nfs4_state_recovery_ops *ops,
1638				   int *lost_locks)
1639{
1640	struct nfs4_state *state;
1641	unsigned int loop = 0;
1642	int status = 0;
1643#ifdef CONFIG_NFS_V4_2
1644	bool found_ssc_copy_state = false;
1645#endif /* CONFIG_NFS_V4_2 */
1646
1647	/* Note: we rely on the sp->so_states list being ordered 
1648	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1649	 * states first.
1650	 * This is needed to ensure that the server won't give us any
1651	 * read delegations that we have to return if, say, we are
1652	 * recovering after a network partition or a reboot from a
1653	 * server that doesn't support a grace period.
1654	 */
1655	spin_lock(&sp->so_lock);
1656restart:
 
1657	list_for_each_entry(state, &sp->so_states, open_states) {
1658		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1659			continue;
1660		if (!nfs4_valid_open_stateid(state))
1661			continue;
1662		if (state->state == 0)
1663			continue;
1664#ifdef CONFIG_NFS_V4_2
1665		if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1666			nfs4_state_mark_recovery_failed(state, -EIO);
1667			found_ssc_copy_state = true;
1668			continue;
1669		}
1670#endif /* CONFIG_NFS_V4_2 */
1671		refcount_inc(&state->count);
1672		spin_unlock(&sp->so_lock);
1673		status = __nfs4_reclaim_open_state(sp, state, ops, lost_locks);
1674
1675		switch (status) {
1676		default:
1677			if (status >= 0) {
1678				loop = 0;
1679				break;
 
 
 
 
 
 
 
 
1680			}
1681			printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1682			fallthrough;
1683		case -ENOENT:
1684		case -ENOMEM:
1685		case -EACCES:
1686		case -EROFS:
1687		case -EIO:
1688		case -ESTALE:
1689			/* Open state on this file cannot be recovered */
1690			nfs4_state_mark_recovery_failed(state, status);
1691			break;
1692		case -EAGAIN:
1693			ssleep(1);
1694			if (loop++ < 10) {
1695				set_bit(ops->state_flag_bit, &state->flags);
 
1696				break;
1697			}
1698			fallthrough;
1699		case -NFS4ERR_ADMIN_REVOKED:
1700		case -NFS4ERR_STALE_STATEID:
1701		case -NFS4ERR_OLD_STATEID:
1702		case -NFS4ERR_BAD_STATEID:
1703		case -NFS4ERR_RECLAIM_BAD:
1704		case -NFS4ERR_RECLAIM_CONFLICT:
1705			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1706			break;
1707		case -NFS4ERR_EXPIRED:
1708		case -NFS4ERR_NO_GRACE:
1709			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1710			fallthrough;
1711		case -NFS4ERR_STALE_CLIENTID:
1712		case -NFS4ERR_BADSESSION:
1713		case -NFS4ERR_BADSLOT:
1714		case -NFS4ERR_BAD_HIGH_SLOT:
1715		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1716		case -ETIMEDOUT:
1717			goto out_err;
 
 
 
1718		}
1719		nfs4_put_open_state(state);
1720		spin_lock(&sp->so_lock);
1721		goto restart;
1722	}
1723	spin_unlock(&sp->so_lock);
1724#ifdef CONFIG_NFS_V4_2
1725	if (found_ssc_copy_state)
1726		return -EIO;
1727#endif /* CONFIG_NFS_V4_2 */
1728	return 0;
1729out_err:
1730	nfs4_put_open_state(state);
1731	spin_lock(&sp->so_lock);
1732	spin_unlock(&sp->so_lock);
1733	return status;
1734}
1735
1736static void nfs4_clear_open_state(struct nfs4_state *state)
1737{
1738	struct nfs4_lock_state *lock;
1739
1740	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1741	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1742	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1743	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1744	spin_lock(&state->state_lock);
1745	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1746		lock->ls_seqid.flags = 0;
1747		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1748	}
1749	spin_unlock(&state->state_lock);
1750}
1751
1752static void nfs4_reset_seqids(struct nfs_server *server,
1753	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1754{
1755	struct nfs_client *clp = server->nfs_client;
1756	struct nfs4_state_owner *sp;
1757	struct rb_node *pos;
1758	struct nfs4_state *state;
1759
1760	spin_lock(&clp->cl_lock);
1761	for (pos = rb_first(&server->state_owners);
1762	     pos != NULL;
1763	     pos = rb_next(pos)) {
1764		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1765		sp->so_seqid.flags = 0;
1766		spin_lock(&sp->so_lock);
1767		list_for_each_entry(state, &sp->so_states, open_states) {
1768			if (mark_reclaim(clp, state))
1769				nfs4_clear_open_state(state);
1770		}
1771		spin_unlock(&sp->so_lock);
1772	}
1773	spin_unlock(&clp->cl_lock);
1774}
1775
1776static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1777	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1778{
1779	struct nfs_server *server;
1780
1781	rcu_read_lock();
1782	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1783		nfs4_reset_seqids(server, mark_reclaim);
1784	rcu_read_unlock();
1785}
1786
1787static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1788{
1789	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1790	/* Mark all delegations for reclaim */
1791	nfs_delegation_mark_reclaim(clp);
1792	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1793}
1794
1795static int nfs4_reclaim_complete(struct nfs_client *clp,
1796				 const struct nfs4_state_recovery_ops *ops,
1797				 const struct cred *cred)
1798{
1799	/* Notify the server we're done reclaiming our state */
1800	if (ops->reclaim_complete)
1801		return ops->reclaim_complete(clp, cred);
1802	return 0;
1803}
1804
1805static void nfs4_clear_reclaim_server(struct nfs_server *server)
1806{
1807	struct nfs_client *clp = server->nfs_client;
1808	struct nfs4_state_owner *sp;
1809	struct rb_node *pos;
1810	struct nfs4_state *state;
1811
1812	spin_lock(&clp->cl_lock);
1813	for (pos = rb_first(&server->state_owners);
1814	     pos != NULL;
1815	     pos = rb_next(pos)) {
1816		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1817		spin_lock(&sp->so_lock);
1818		list_for_each_entry(state, &sp->so_states, open_states) {
1819			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1820						&state->flags))
1821				continue;
1822			nfs4_state_mark_reclaim_nograce(clp, state);
1823		}
1824		spin_unlock(&sp->so_lock);
1825	}
1826	spin_unlock(&clp->cl_lock);
1827}
1828
1829static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1830{
1831	struct nfs_server *server;
1832
1833	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1834		return 0;
1835
1836	rcu_read_lock();
1837	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1838		nfs4_clear_reclaim_server(server);
1839	rcu_read_unlock();
1840
1841	nfs_delegation_reap_unclaimed(clp);
1842	return 1;
1843}
1844
1845static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1846{
1847	const struct nfs4_state_recovery_ops *ops;
1848	const struct cred *cred;
1849	int err;
1850
1851	if (!nfs4_state_clear_reclaim_reboot(clp))
1852		return;
1853	pnfs_destroy_all_layouts(clp);
1854	ops = clp->cl_mvops->reboot_recovery_ops;
1855	cred = nfs4_get_clid_cred(clp);
1856	err = nfs4_reclaim_complete(clp, ops, cred);
1857	put_cred(cred);
1858	if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1859		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1860}
1861
1862static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1863{
1864	nfs_mark_test_expired_all_delegations(clp);
1865	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1866}
1867
 
 
 
 
 
 
 
 
1868static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1869{
1870	switch (error) {
1871	case 0:
1872		break;
1873	case -NFS4ERR_CB_PATH_DOWN:
1874		nfs40_handle_cb_pathdown(clp);
1875		break;
1876	case -NFS4ERR_NO_GRACE:
1877		nfs4_state_end_reclaim_reboot(clp);
1878		break;
1879	case -NFS4ERR_STALE_CLIENTID:
1880		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1881		nfs4_state_start_reclaim_reboot(clp);
1882		break;
1883	case -NFS4ERR_EXPIRED:
1884		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1885		nfs4_state_start_reclaim_nograce(clp);
1886		break;
1887	case -NFS4ERR_BADSESSION:
1888	case -NFS4ERR_BADSLOT:
1889	case -NFS4ERR_BAD_HIGH_SLOT:
1890	case -NFS4ERR_DEADSESSION:
1891	case -NFS4ERR_SEQ_FALSE_RETRY:
1892	case -NFS4ERR_SEQ_MISORDERED:
1893		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1894		/* Zero session reset errors */
1895		break;
1896	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1897		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1898		break;
1899	default:
1900		dprintk("%s: failed to handle error %d for server %s\n",
1901				__func__, error, clp->cl_hostname);
1902		return error;
 
 
 
 
 
 
1903	}
1904	dprintk("%s: handled error %d for server %s\n", __func__, error,
1905			clp->cl_hostname);
1906	return 0;
1907}
1908
1909static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1910{
1911	struct nfs4_state_owner *sp;
1912	struct nfs_server *server;
1913	struct rb_node *pos;
1914	LIST_HEAD(freeme);
1915	int lost_locks = 0;
1916	int status;
1917
1918	status = nfs4_begin_drain_session(clp);
1919	if (status < 0)
1920		return status;
1921restart:
1922	rcu_read_lock();
1923	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1924		nfs4_purge_state_owners(server, &freeme);
1925		spin_lock(&clp->cl_lock);
1926		for (pos = rb_first(&server->state_owners);
1927		     pos != NULL;
1928		     pos = rb_next(pos)) {
1929			sp = rb_entry(pos,
1930				struct nfs4_state_owner, so_server_node);
1931			if (!test_and_clear_bit(ops->owner_flag_bit,
1932							&sp->so_flags))
1933				continue;
1934			if (!atomic_inc_not_zero(&sp->so_count))
1935				continue;
1936			spin_unlock(&clp->cl_lock);
1937			rcu_read_unlock();
1938
1939			status = nfs4_reclaim_open_state(sp, ops, &lost_locks);
1940			if (status < 0) {
1941				if (lost_locks)
1942					pr_warn("NFS: %s: lost %d locks\n",
1943						clp->cl_hostname, lost_locks);
1944				set_bit(ops->owner_flag_bit, &sp->so_flags);
1945				nfs4_put_state_owner(sp);
1946				status = nfs4_recovery_handle_error(clp, status);
1947				nfs4_free_state_owners(&freeme);
1948				return (status != 0) ? status : -EAGAIN;
1949			}
1950
1951			nfs4_put_state_owner(sp);
1952			goto restart;
1953		}
1954		spin_unlock(&clp->cl_lock);
1955	}
1956	rcu_read_unlock();
1957	nfs4_free_state_owners(&freeme);
1958	if (lost_locks)
1959		pr_warn("NFS: %s: lost %d locks\n",
1960			clp->cl_hostname, lost_locks);
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 -ETIMEDOUT:
2015		if (clp->cl_cons_state == NFS_CS_SESSION_INITING) {
2016			nfs_mark_client_ready(clp, -EIO);
2017			return -EIO;
2018		}
2019		fallthrough;
2020	case -EACCES:
 
 
 
 
2021	case -NFS4ERR_DELAY:
 
2022	case -EAGAIN:
2023		ssleep(1);
2024		break;
2025
2026	case -NFS4ERR_MINOR_VERS_MISMATCH:
2027		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2028			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2029		dprintk("%s: exit with error %d for server %s\n",
2030				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
2031		return -EPROTONOSUPPORT;
2032	case -ENOSPC:
2033		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2034			nfs_mark_client_ready(clp, -EIO);
2035		return -EIO;
2036	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2037				 * in nfs4_exchange_id */
2038	default:
2039		dprintk("%s: exit with error %d for server %s\n", __func__,
2040				status, clp->cl_hostname);
2041		return status;
2042	}
2043	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2044	dprintk("%s: handled error %d for server %s\n", __func__, status,
2045			clp->cl_hostname);
2046	return 0;
2047}
2048
2049static int nfs4_establish_lease(struct nfs_client *clp)
2050{
2051	const struct cred *cred;
2052	const struct nfs4_state_recovery_ops *ops =
2053		clp->cl_mvops->reboot_recovery_ops;
2054	int status;
2055
2056	status = nfs4_begin_drain_session(clp);
2057	if (status != 0)
2058		return status;
2059	cred = nfs4_get_clid_cred(clp);
2060	if (cred == NULL)
2061		return -ENOENT;
2062	status = ops->establish_clid(clp, cred);
2063	put_cred(cred);
2064	if (status != 0)
2065		return status;
2066	return 0;
2067}
2068
2069/*
2070 * Returns zero or a negative errno.  NFS4ERR values are converted
2071 * to local errno values.
2072 */
2073static int nfs4_reclaim_lease(struct nfs_client *clp)
2074{
2075	int status;
2076
2077	status = nfs4_establish_lease(clp);
2078	if (status < 0)
2079		return nfs4_handle_reclaim_lease_error(clp, status);
2080	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2081		nfs4_state_start_reclaim_nograce(clp);
2082	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2083		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2084	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2085	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2086	return 0;
2087}
2088
2089static int nfs4_purge_lease(struct nfs_client *clp)
2090{
2091	int status;
2092
2093	status = nfs4_establish_lease(clp);
2094	if (status < 0)
2095		return nfs4_handle_reclaim_lease_error(clp, status);
2096	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2097	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2098	nfs4_state_start_reclaim_nograce(clp);
2099	return 0;
2100}
2101
2102/*
2103 * Try remote migration of one FSID from a source server to a
2104 * destination server.  The source server provides a list of
2105 * potential destinations.
2106 *
2107 * Returns zero or a negative NFS4ERR status code.
2108 */
2109static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2110{
2111	struct nfs_client *clp = server->nfs_client;
2112	struct nfs4_fs_locations *locations = NULL;
2113	struct nfs_fattr *fattr;
2114	struct inode *inode;
2115	struct page *page;
2116	int status, result;
2117
2118	dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2119			(unsigned long long)server->fsid.major,
2120			(unsigned long long)server->fsid.minor,
2121			clp->cl_hostname);
2122
2123	page = alloc_page(GFP_KERNEL);
2124	locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2125	fattr = nfs_alloc_fattr();
2126	if (page == NULL || locations == NULL || fattr == NULL) {
2127		dprintk("<-- %s: no memory\n", __func__);
2128		result = 0;
2129		goto out;
2130	}
2131
2132	locations->fattr = fattr;
2133	inode = d_inode(server->super->s_root);
2134	result = nfs4_proc_get_locations(server, NFS_FH(inode), locations,
2135					 page, cred);
2136	if (result) {
2137		dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2138			__func__, result);
2139		goto out;
2140	}
2141
2142	result = -NFS4ERR_NXIO;
2143	if (!locations->nlocations)
2144		goto out;
2145
2146	if (!(locations->fattr->valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2147		dprintk("<-- %s: No fs_locations data, migration skipped\n",
2148			__func__);
2149		goto out;
2150	}
2151
2152	status = nfs4_begin_drain_session(clp);
2153	if (status != 0) {
2154		result = status;
2155		goto out;
2156	}
2157
2158	status = nfs4_replace_transport(server, locations);
2159	if (status != 0) {
2160		dprintk("<-- %s: failed to replace transport: %d\n",
2161			__func__, status);
2162		goto out;
2163	}
2164
2165	result = 0;
2166	dprintk("<-- %s: migration succeeded\n", __func__);
2167
2168out:
2169	if (page != NULL)
2170		__free_page(page);
2171	if (locations != NULL)
2172		kfree(locations->fattr);
2173	kfree(locations);
2174	if (result) {
2175		pr_err("NFS: migration recovery failed (server %s)\n",
2176				clp->cl_hostname);
2177		set_bit(NFS_MIG_FAILED, &server->mig_status);
2178	}
2179	return result;
2180}
2181
2182/*
2183 * Returns zero or a negative NFS4ERR status code.
2184 */
2185static int nfs4_handle_migration(struct nfs_client *clp)
2186{
2187	const struct nfs4_state_maintenance_ops *ops =
2188				clp->cl_mvops->state_renewal_ops;
2189	struct nfs_server *server;
2190	const struct cred *cred;
2191
2192	dprintk("%s: migration reported on \"%s\"\n", __func__,
2193			clp->cl_hostname);
2194
2195	cred = ops->get_state_renewal_cred(clp);
2196	if (cred == NULL)
2197		return -NFS4ERR_NOENT;
2198
2199	clp->cl_mig_gen++;
2200restart:
2201	rcu_read_lock();
2202	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2203		int status;
2204
2205		if (server->mig_gen == clp->cl_mig_gen)
2206			continue;
2207		server->mig_gen = clp->cl_mig_gen;
2208
2209		if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2210						&server->mig_status))
2211			continue;
2212
2213		rcu_read_unlock();
2214		status = nfs4_try_migration(server, cred);
2215		if (status < 0) {
2216			put_cred(cred);
2217			return status;
2218		}
2219		goto restart;
2220	}
2221	rcu_read_unlock();
2222	put_cred(cred);
2223	return 0;
2224}
2225
2226/*
2227 * Test each nfs_server on the clp's cl_superblocks list to see
2228 * if it's moved to another server.  Stop when the server no longer
2229 * returns NFS4ERR_LEASE_MOVED.
2230 */
2231static int nfs4_handle_lease_moved(struct nfs_client *clp)
2232{
2233	const struct nfs4_state_maintenance_ops *ops =
2234				clp->cl_mvops->state_renewal_ops;
2235	struct nfs_server *server;
2236	const struct cred *cred;
2237
2238	dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2239			clp->cl_hostname);
2240
2241	cred = ops->get_state_renewal_cred(clp);
2242	if (cred == NULL)
2243		return -NFS4ERR_NOENT;
2244
2245	clp->cl_mig_gen++;
2246restart:
2247	rcu_read_lock();
2248	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2249		struct inode *inode;
2250		int status;
2251
2252		if (server->mig_gen == clp->cl_mig_gen)
2253			continue;
2254		server->mig_gen = clp->cl_mig_gen;
2255
2256		rcu_read_unlock();
2257
2258		inode = d_inode(server->super->s_root);
2259		status = nfs4_proc_fsid_present(inode, cred);
2260		if (status != -NFS4ERR_MOVED)
2261			goto restart;	/* wasn't this one */
2262		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2263			goto restart;	/* there are more */
2264		goto out;
2265	}
2266	rcu_read_unlock();
2267
2268out:
2269	put_cred(cred);
2270	return 0;
2271}
2272
2273/**
2274 * nfs4_discover_server_trunking - Detect server IP address trunking
2275 *
2276 * @clp: nfs_client under test
2277 * @result: OUT: found nfs_client, or clp
2278 *
2279 * Returns zero or a negative errno.  If zero is returned,
2280 * an nfs_client pointer is planted in "result".
2281 *
2282 * Note: since we are invoked in process context, and
2283 * not from inside the state manager, we cannot use
2284 * nfs4_handle_reclaim_lease_error().
2285 */
2286int nfs4_discover_server_trunking(struct nfs_client *clp,
2287				  struct nfs_client **result)
2288{
2289	const struct nfs4_state_recovery_ops *ops =
2290				clp->cl_mvops->reboot_recovery_ops;
2291	struct rpc_clnt *clnt;
2292	const struct cred *cred;
2293	int i, status;
2294
2295	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2296
2297	clnt = clp->cl_rpcclient;
2298	i = 0;
2299
2300	mutex_lock(&nfs_clid_init_mutex);
2301again:
2302	status  = -ENOENT;
2303	cred = nfs4_get_clid_cred(clp);
2304	if (cred == NULL)
2305		goto out_unlock;
2306
2307	status = ops->detect_trunking(clp, result, cred);
2308	put_cred(cred);
2309	switch (status) {
2310	case 0:
2311	case -EINTR:
2312	case -ERESTARTSYS:
2313		break;
2314	case -ETIMEDOUT:
2315		if (clnt->cl_softrtry)
2316			break;
2317		fallthrough;
2318	case -NFS4ERR_DELAY:
2319	case -EAGAIN:
2320		ssleep(1);
2321		fallthrough;
2322	case -NFS4ERR_STALE_CLIENTID:
2323		dprintk("NFS: %s after status %d, retrying\n",
2324			__func__, status);
2325		goto again;
2326	case -EACCES:
2327		if (i++ == 0) {
2328			nfs4_root_machine_cred(clp);
2329			goto again;
2330		}
2331		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2332			break;
2333		fallthrough;
2334	case -NFS4ERR_CLID_INUSE:
2335	case -NFS4ERR_WRONGSEC:
2336		/* No point in retrying if we already used RPC_AUTH_UNIX */
2337		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2338			status = -EPERM;
2339			break;
2340		}
2341		clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2342		if (IS_ERR(clnt)) {
2343			status = PTR_ERR(clnt);
2344			break;
2345		}
2346		/* Note: this is safe because we haven't yet marked the
2347		 * client as ready, so we are the only user of
2348		 * clp->cl_rpcclient
2349		 */
2350		clnt = xchg(&clp->cl_rpcclient, clnt);
2351		rpc_shutdown_client(clnt);
2352		clnt = clp->cl_rpcclient;
2353		goto again;
2354
2355	case -NFS4ERR_MINOR_VERS_MISMATCH:
2356		status = -EPROTONOSUPPORT;
2357		break;
2358
2359	case -EKEYEXPIRED:
2360	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2361				 * in nfs4_exchange_id */
2362		status = -EKEYEXPIRED;
2363		break;
2364	default:
2365		pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2366				__func__, status);
2367		status = -EIO;
2368	}
2369
2370out_unlock:
2371	mutex_unlock(&nfs_clid_init_mutex);
2372	dprintk("NFS: %s: status = %d\n", __func__, status);
2373	return status;
2374}
2375
2376#ifdef CONFIG_NFS_V4_1
2377void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2378{
2379	struct nfs_client *clp = session->clp;
2380
2381	switch (err) {
2382	default:
2383		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2384		break;
2385	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2386		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2387	}
2388	nfs4_schedule_state_manager(clp);
2389}
2390EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2391
2392void nfs41_notify_server(struct nfs_client *clp)
2393{
2394	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
2395	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
 
2396	nfs4_schedule_state_manager(clp);
2397}
2398
2399static void nfs4_reset_all_state(struct nfs_client *clp)
2400{
2401	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2402		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2403		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2404		nfs4_state_start_reclaim_nograce(clp);
2405		dprintk("%s: scheduling reset of all state for server %s!\n",
2406				__func__, clp->cl_hostname);
2407		nfs4_schedule_state_manager(clp);
2408	}
2409}
2410
2411static void nfs41_handle_server_reboot(struct nfs_client *clp)
2412{
2413	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2414		nfs4_state_start_reclaim_reboot(clp);
2415		dprintk("%s: server %s rebooted!\n", __func__,
2416				clp->cl_hostname);
2417		nfs4_schedule_state_manager(clp);
2418	}
2419}
2420
2421static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2422{
2423	nfs4_reset_all_state(clp);
2424	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2425}
2426
2427static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2428{
2429	nfs4_state_start_reclaim_nograce(clp);
2430	nfs4_schedule_state_manager(clp);
2431
2432	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2433}
2434
2435static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2436{
2437	/* FIXME: For now, we destroy all layouts. */
2438	pnfs_destroy_all_layouts(clp);
2439	nfs_test_expired_all_delegations(clp);
2440	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2441			clp->cl_hostname);
2442}
2443
2444static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2445{
2446	set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2447	nfs4_schedule_state_manager(clp);
2448
2449	dprintk("%s: server %s declared a backchannel fault\n", __func__,
2450			clp->cl_hostname);
2451}
2452
2453static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2454{
2455	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2456		&clp->cl_state) == 0)
2457		nfs4_schedule_state_manager(clp);
2458}
2459
2460void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2461		bool recovery)
2462{
2463	if (!flags)
2464		return;
2465
2466	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2467		__func__, clp->cl_hostname, clp->cl_clientid, flags);
2468	/*
2469	 * If we're called from the state manager thread, then assume we're
2470	 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2471	 * Those flags are expected to remain set until we're done
2472	 * recovering (see RFC5661, section 18.46.3).
2473	 */
2474	if (recovery)
2475		goto out_recovery;
2476
2477	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2478		nfs41_handle_server_reboot(clp);
2479	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2480		nfs41_handle_all_state_revoked(clp);
2481	if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2482			    SEQ4_STATUS_ADMIN_STATE_REVOKED))
2483		nfs41_handle_some_state_revoked(clp);
2484	if (flags & SEQ4_STATUS_LEASE_MOVED)
2485		nfs4_schedule_lease_moved_recovery(clp);
2486	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2487		nfs41_handle_recallable_state_revoked(clp);
2488out_recovery:
2489	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2490		nfs41_handle_backchannel_fault(clp);
2491	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2492				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2493		nfs41_handle_cb_path_down(clp);
2494}
2495
2496static int nfs4_reset_session(struct nfs_client *clp)
2497{
2498	const struct cred *cred;
2499	int status;
2500
2501	if (!nfs4_has_session(clp))
2502		return 0;
2503	status = nfs4_begin_drain_session(clp);
2504	if (status != 0)
2505		return status;
2506	cred = nfs4_get_clid_cred(clp);
2507	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2508	switch (status) {
2509	case 0:
2510	case -NFS4ERR_BADSESSION:
2511	case -NFS4ERR_DEADSESSION:
2512		break;
2513	case -NFS4ERR_BACK_CHAN_BUSY:
2514	case -NFS4ERR_DELAY:
2515		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2516		status = 0;
2517		ssleep(1);
2518		goto out;
2519	default:
2520		status = nfs4_recovery_handle_error(clp, status);
2521		goto out;
2522	}
2523
2524	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2525	status = nfs4_proc_create_session(clp, cred);
2526	if (status) {
2527		dprintk("%s: session reset failed with status %d for server %s!\n",
2528			__func__, status, clp->cl_hostname);
2529		status = nfs4_handle_reclaim_lease_error(clp, status);
2530		goto out;
2531	}
2532	nfs41_finish_session_reset(clp);
2533	dprintk("%s: session reset was successful for server %s!\n",
2534			__func__, clp->cl_hostname);
2535out:
2536	put_cred(cred);
 
2537	return status;
2538}
2539
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2540static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2541{
2542	const struct cred *cred;
2543	int ret;
2544
2545	if (!nfs4_has_session(clp))
2546		return 0;
2547	ret = nfs4_begin_drain_session(clp);
2548	if (ret != 0)
2549		return ret;
2550	cred = nfs4_get_clid_cred(clp);
2551	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2552	put_cred(cred);
 
2553	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2554	switch (ret) {
2555	case 0:
2556		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2557			__func__, clp->cl_hostname);
2558		break;
2559	case -NFS4ERR_DELAY:
2560		ssleep(1);
2561		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2562		break;
2563	default:
2564		return nfs4_recovery_handle_error(clp, ret);
2565	}
2566	return 0;
2567}
2568
2569static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2570{
2571	int iomode = 0;
2572
2573	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &clp->cl_state))
2574		iomode += IOMODE_READ;
2575	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &clp->cl_state))
2576		iomode += IOMODE_RW;
2577	/* Note: IOMODE_READ + IOMODE_RW == IOMODE_ANY */
2578	if (iomode) {
2579		pnfs_layout_return_unused_byclid(clp, iomode);
2580		set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2581	}
2582}
2583#else /* CONFIG_NFS_V4_1 */
2584static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
 
 
2585
2586static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2587{
2588	return 0;
2589}
2590
2591static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2592{
2593}
2594#endif /* CONFIG_NFS_V4_1 */
2595
2596static void nfs4_state_manager(struct nfs_client *clp)
2597{
2598	unsigned int memflags;
2599	int status = 0;
2600	const char *section = "", *section_sep = "";
2601
2602	/*
2603	 * State recovery can deadlock if the direct reclaim code tries
2604	 * start NFS writeback. So ensure memory allocations are all
2605	 * GFP_NOFS.
2606	 */
2607	memflags = memalloc_nofs_save();
2608
2609	/* Ensure exclusive access to NFSv4 state */
2610	do {
2611		trace_nfs4_state_mgr(clp);
2612		clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2613		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2614			section = "purge state";
2615			status = nfs4_purge_lease(clp);
2616			if (status < 0)
2617				goto out_error;
2618			continue;
 
2619		}
2620
2621		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2622			section = "lease expired";
2623			/* We're going to have to re-establish a clientid */
2624			status = nfs4_reclaim_lease(clp);
2625			if (status < 0)
2626				goto out_error;
2627			continue;
2628		}
2629
2630		/* Initialize or reset the session */
2631		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2632			section = "reset session";
2633			status = nfs4_reset_session(clp);
2634			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2635				continue;
2636			if (status < 0)
2637				goto out_error;
2638		}
2639
2640		/* Send BIND_CONN_TO_SESSION */
2641		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2642				&clp->cl_state)) {
2643			section = "bind conn to session";
2644			status = nfs4_bind_conn_to_session(clp);
2645			if (status < 0)
2646				goto out_error;
2647			continue;
2648		}
2649
2650		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2651			section = "check lease";
2652			status = nfs4_check_lease(clp);
2653			if (status < 0)
2654				goto out_error;
2655			continue;
 
2656		}
2657
2658		if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2659			section = "migration";
2660			status = nfs4_handle_migration(clp);
 
 
 
2661			if (status < 0)
2662				goto out_error;
2663		}
2664
2665		if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2666			section = "lease moved";
2667			status = nfs4_handle_lease_moved(clp);
 
2668			if (status < 0)
2669				goto out_error;
 
2670		}
2671
2672		/* First recover reboot state... */
2673		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2674			section = "reclaim reboot";
2675			status = nfs4_do_reclaim(clp,
2676				clp->cl_mvops->reboot_recovery_ops);
2677			if (status == 0)
2678				status = pnfs_layout_handle_reboot(clp);
2679			if (status == -EAGAIN)
2680				continue;
2681			if (status < 0)
2682				goto out_error;
2683			nfs4_state_end_reclaim_reboot(clp);
2684			continue;
2685		}
2686
2687		/* Detect expired delegations... */
2688		if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2689			section = "detect expired delegations";
2690			status = nfs4_begin_drain_session(clp);
2691			if (status < 0)
2692				goto out_error;
2693			nfs_reap_expired_delegations(clp);
2694			continue;
2695		}
2696
2697		/* Now recover expired state... */
2698		if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2699			section = "reclaim nograce";
2700			status = nfs4_do_reclaim(clp,
2701				clp->cl_mvops->nograce_recovery_ops);
2702			if (status == -EAGAIN)
 
 
2703				continue;
2704			if (status < 0)
2705				goto out_error;
2706			clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2707		}
2708
2709		memalloc_nofs_restore(memflags);
2710		nfs4_end_drain_session(clp);
2711		nfs4_clear_state_manager_bit(clp);
2712
2713		if (test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2714		    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING,
2715				      &clp->cl_state)) {
2716			memflags = memalloc_nofs_save();
2717			continue;
2718		}
2719
2720		if (!test_and_set_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state)) {
2721			if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2722				nfs_client_return_marked_delegations(clp);
2723				set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2724			}
2725			nfs4_layoutreturn_any_run(clp);
2726			clear_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state);
2727		}
2728
2729		return;
2730
2731	} while (refcount_read(&clp->cl_count) > 1 && !signalled());
2732	goto out_drain;
2733
 
 
 
 
 
 
 
 
2734out_error:
2735	if (strlen(section))
2736		section_sep = ": ";
2737	trace_nfs4_state_mgr_failed(clp, section, status);
2738	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2739			" with error %d\n", section_sep, section,
2740			clp->cl_hostname, -status);
2741	ssleep(1);
2742out_drain:
2743	memalloc_nofs_restore(memflags);
2744	nfs4_end_drain_session(clp);
2745	nfs4_clear_state_manager_bit(clp);
2746}
2747
2748static int nfs4_run_state_manager(void *ptr)
2749{
2750	struct nfs_client *clp = ptr;
2751	struct rpc_clnt *cl = clp->cl_rpcclient;
2752
2753	while (cl != cl->cl_parent)
2754		cl = cl->cl_parent;
2755
2756	allow_signal(SIGKILL);
2757again:
2758	nfs4_state_manager(clp);
2759
2760	if (test_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state) &&
2761	    !test_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state)) {
2762		wait_var_event_interruptible(&clp->cl_state,
2763					     test_bit(NFS4CLNT_RUN_MANAGER,
2764						      &clp->cl_state));
2765		if (!atomic_read(&cl->cl_swapper))
2766			clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2767		if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2768		    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2769			goto again;
2770		/* Either no longer a swapper, or were signalled */
2771		clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2772	}
2773
2774	if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2775	    test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2776	    !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2777		goto again;
2778
2779	nfs_put_client(clp);
2780	module_put_and_kthread_exit(0);
2781	return 0;
2782}
v3.5.6
   1/*
   2 *  fs/nfs/nfs4state.c
   3 *
   4 *  Client-side XDR for NFSv4.
   5 *
   6 *  Copyright (c) 2002 The Regents of the University of Michigan.
   7 *  All rights reserved.
   8 *
   9 *  Kendrick Smith <kmsmith@umich.edu>
  10 *
  11 *  Redistribution and use in source and binary forms, with or without
  12 *  modification, are permitted provided that the following conditions
  13 *  are met:
  14 *
  15 *  1. Redistributions of source code must retain the above copyright
  16 *     notice, this list of conditions and the following disclaimer.
  17 *  2. Redistributions in binary form must reproduce the above copyright
  18 *     notice, this list of conditions and the following disclaimer in the
  19 *     documentation and/or other materials provided with the distribution.
  20 *  3. Neither the name of the University nor the names of its
  21 *     contributors may be used to endorse or promote products derived
  22 *     from this software without specific prior written permission.
  23 *
  24 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35 *
  36 * Implementation of the NFSv4 state model.  For the time being,
  37 * this is minimal, but will be made much more complex in a
  38 * subsequent patch.
  39 */
  40
  41#include <linux/kernel.h>
  42#include <linux/slab.h>
  43#include <linux/fs.h>
  44#include <linux/nfs_fs.h>
  45#include <linux/nfs_idmap.h>
  46#include <linux/kthread.h>
  47#include <linux/module.h>
  48#include <linux/random.h>
  49#include <linux/ratelimit.h>
  50#include <linux/workqueue.h>
  51#include <linux/bitops.h>
  52#include <linux/jiffies.h>
 
 
 
  53
  54#include "nfs4_fs.h"
  55#include "callback.h"
  56#include "delegation.h"
  57#include "internal.h"
 
 
  58#include "pnfs.h"
 
 
  59
  60#define NFSDBG_FACILITY		NFSDBG_STATE
  61
  62#define OPENOWNER_POOL_SIZE	8
  63
  64const nfs4_stateid zero_stateid;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  65
  66static LIST_HEAD(nfs4_clientid_list);
 
 
 
 
 
 
 
 
 
 
 
 
  67
  68int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  69{
  70	struct nfs4_setclientid_res clid = {
  71		.clientid = clp->cl_clientid,
  72		.confirm = clp->cl_confirm,
  73	};
  74	unsigned short port;
  75	int status;
 
  76
  77	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  78		goto do_confirm;
  79	port = nfs_callback_tcpport;
  80	if (clp->cl_addr.ss_family == AF_INET6)
  81		port = nfs_callback_tcpport6;
  82
  83	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  84	if (status != 0)
  85		goto out;
  86	clp->cl_clientid = clid.clientid;
  87	clp->cl_confirm = clid.confirm;
  88	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  89do_confirm:
  90	status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  91	if (status != 0)
  92		goto out;
  93	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  94	nfs4_schedule_state_renewal(clp);
  95out:
  96	return status;
  97}
  98
  99struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 100{
 101	struct rpc_cred *cred = NULL;
 
 
 
 
 
 
 102
 103	if (clp->cl_machine_cred != NULL)
 104		cred = get_rpccred(clp->cl_machine_cred);
 105	return cred;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 106}
 107
 108static void nfs4_clear_machine_cred(struct nfs_client *clp)
 109{
 110	struct rpc_cred *cred;
 111
 112	spin_lock(&clp->cl_lock);
 113	cred = clp->cl_machine_cred;
 114	clp->cl_machine_cred = NULL;
 115	spin_unlock(&clp->cl_lock);
 116	if (cred != NULL)
 117		put_rpccred(cred);
 118}
 119
 120static struct rpc_cred *
 121nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 122{
 123	struct rpc_cred *cred = NULL;
 124	struct nfs4_state_owner *sp;
 125	struct rb_node *pos;
 126
 127	for (pos = rb_first(&server->state_owners);
 128	     pos != NULL;
 129	     pos = rb_next(pos)) {
 130		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 131		if (list_empty(&sp->so_states))
 132			continue;
 133		cred = get_rpccred(sp->so_cred);
 134		break;
 135	}
 136	return cred;
 137}
 138
 139/**
 140 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 141 * @clp: client state handle
 142 *
 143 * Returns an rpc_cred with reference count bumped, or NULL.
 144 * Caller must hold clp->cl_lock.
 145 */
 146struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 147{
 148	struct rpc_cred *cred = NULL;
 149	struct nfs_server *server;
 150
 151	/* Use machine credentials if available */
 152	cred = nfs4_get_machine_cred_locked(clp);
 153	if (cred != NULL)
 154		goto out;
 155
 
 156	rcu_read_lock();
 157	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 158		cred = nfs4_get_renew_cred_server_locked(server);
 159		if (cred != NULL)
 160			break;
 161	}
 162	rcu_read_unlock();
 
 163
 164out:
 165	return cred;
 166}
 167
 168#if defined(CONFIG_NFS_V4_1)
 169
 170static int nfs41_setup_state_renewal(struct nfs_client *clp)
 171{
 172	int status;
 173	struct nfs_fsinfo fsinfo;
 174
 175	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 176		nfs4_schedule_state_renewal(clp);
 177		return 0;
 178	}
 179
 180	status = nfs4_proc_get_lease_time(clp, &fsinfo);
 181	if (status == 0) {
 182		/* Update lease time and schedule renewal */
 183		spin_lock(&clp->cl_lock);
 184		clp->cl_lease_time = fsinfo.lease_time * HZ;
 185		clp->cl_last_renewal = jiffies;
 186		spin_unlock(&clp->cl_lock);
 187
 188		nfs4_schedule_state_renewal(clp);
 189	}
 190
 191	return status;
 192}
 193
 194/*
 195 * Back channel returns NFS4ERR_DELAY for new requests when
 196 * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 197 * is ended.
 198 */
 199static void nfs4_end_drain_session(struct nfs_client *clp)
 200{
 201	struct nfs4_session *ses = clp->cl_session;
 202	struct nfs4_slot_table *tbl;
 203	int max_slots;
 204
 205	if (ses == NULL)
 
 206		return;
 207	tbl = &ses->fc_slot_table;
 208	if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 209		spin_lock(&tbl->slot_tbl_lock);
 210		max_slots = tbl->max_slots;
 211		while (max_slots--) {
 212			if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
 213						nfs4_set_task_privileged,
 214						NULL) == NULL)
 215				break;
 216		}
 217		spin_unlock(&tbl->slot_tbl_lock);
 218	}
 219}
 220
 221static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 222{
 
 223	spin_lock(&tbl->slot_tbl_lock);
 224	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 225		INIT_COMPLETION(tbl->complete);
 226		spin_unlock(&tbl->slot_tbl_lock);
 227		return wait_for_completion_interruptible(&tbl->complete);
 228	}
 229	spin_unlock(&tbl->slot_tbl_lock);
 230	return 0;
 231}
 232
 233static int nfs4_begin_drain_session(struct nfs_client *clp)
 234{
 235	struct nfs4_session *ses = clp->cl_session;
 236	int ret = 0;
 
 
 
 237
 238	set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 239	/* back channel */
 240	ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 241	if (ret)
 242		return ret;
 243	/* fore channel */
 244	return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 245}
 246
 
 
 247static void nfs41_finish_session_reset(struct nfs_client *clp)
 248{
 249	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 250	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 251	/* create_session negotiated new slot table */
 252	clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
 253	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 254	nfs41_setup_state_renewal(clp);
 255}
 256
 257int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 258{
 259	int status;
 260
 261	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 262		goto do_confirm;
 263	nfs4_begin_drain_session(clp);
 264	status = nfs4_proc_exchange_id(clp, cred);
 265	if (status != 0)
 266		goto out;
 267	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 268do_confirm:
 269	status = nfs4_proc_create_session(clp, cred);
 270	if (status != 0)
 271		goto out;
 
 
 272	nfs41_finish_session_reset(clp);
 273	nfs_mark_client_ready(clp, NFS_CS_READY);
 274out:
 275	return status;
 276}
 277
 278struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 279{
 280	struct rpc_cred *cred;
 281
 282	spin_lock(&clp->cl_lock);
 283	cred = nfs4_get_machine_cred_locked(clp);
 284	spin_unlock(&clp->cl_lock);
 285	return cred;
 286}
 287
 288#endif /* CONFIG_NFS_V4_1 */
 
 
 
 
 289
 290static struct rpc_cred *
 291nfs4_get_setclientid_cred_server(struct nfs_server *server)
 292{
 293	struct nfs_client *clp = server->nfs_client;
 294	struct rpc_cred *cred = NULL;
 295	struct nfs4_state_owner *sp;
 296	struct rb_node *pos;
 297
 298	spin_lock(&clp->cl_lock);
 299	pos = rb_first(&server->state_owners);
 300	if (pos != NULL) {
 301		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 302		cred = get_rpccred(sp->so_cred);
 303	}
 304	spin_unlock(&clp->cl_lock);
 305	return cred;
 
 
 
 306}
 307
 
 
 308/**
 309 * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 310 * @clp: client state handle
 311 *
 312 * Returns an rpc_cred with reference count bumped, or NULL.
 313 */
 314struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 315{
 316	struct nfs_server *server;
 317	struct rpc_cred *cred;
 318
 319	spin_lock(&clp->cl_lock);
 320	cred = nfs4_get_machine_cred_locked(clp);
 321	spin_unlock(&clp->cl_lock);
 322	if (cred != NULL)
 323		goto out;
 324
 325	rcu_read_lock();
 326	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 327		cred = nfs4_get_setclientid_cred_server(server);
 328		if (cred != NULL)
 329			break;
 330	}
 331	rcu_read_unlock();
 332
 333out:
 334	return cred;
 335}
 336
 337static struct nfs4_state_owner *
 338nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 339{
 340	struct rb_node **p = &server->state_owners.rb_node,
 341		       *parent = NULL;
 342	struct nfs4_state_owner *sp;
 
 343
 344	while (*p != NULL) {
 345		parent = *p;
 346		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 
 347
 348		if (cred < sp->so_cred)
 349			p = &parent->rb_left;
 350		else if (cred > sp->so_cred)
 351			p = &parent->rb_right;
 352		else {
 353			if (!list_empty(&sp->so_lru))
 354				list_del_init(&sp->so_lru);
 355			atomic_inc(&sp->so_count);
 356			return sp;
 357		}
 358	}
 359	return NULL;
 360}
 361
 362static struct nfs4_state_owner *
 363nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 364{
 365	struct nfs_server *server = new->so_server;
 366	struct rb_node **p = &server->state_owners.rb_node,
 367		       *parent = NULL;
 368	struct nfs4_state_owner *sp;
 369	int err;
 370
 371	while (*p != NULL) {
 372		parent = *p;
 373		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 
 374
 375		if (new->so_cred < sp->so_cred)
 376			p = &parent->rb_left;
 377		else if (new->so_cred > sp->so_cred)
 378			p = &parent->rb_right;
 379		else {
 380			if (!list_empty(&sp->so_lru))
 381				list_del_init(&sp->so_lru);
 382			atomic_inc(&sp->so_count);
 383			return sp;
 384		}
 385	}
 386	err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 387	if (err)
 388		return ERR_PTR(err);
 389	rb_link_node(&new->so_server_node, parent, p);
 390	rb_insert_color(&new->so_server_node, &server->state_owners);
 391	return new;
 392}
 393
 394static void
 395nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 396{
 397	struct nfs_server *server = sp->so_server;
 398
 399	if (!RB_EMPTY_NODE(&sp->so_server_node))
 400		rb_erase(&sp->so_server_node, &server->state_owners);
 401	ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 402}
 403
 404static void
 405nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 406{
 407	sc->create_time = ktime_get();
 408	sc->flags = 0;
 409	sc->counter = 0;
 410	spin_lock_init(&sc->lock);
 411	INIT_LIST_HEAD(&sc->list);
 412	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 413}
 414
 415static void
 416nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 417{
 418	rpc_destroy_wait_queue(&sc->wait);
 419}
 420
 421/*
 422 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 423 * create a new state_owner.
 424 *
 425 */
 426static struct nfs4_state_owner *
 427nfs4_alloc_state_owner(struct nfs_server *server,
 428		struct rpc_cred *cred,
 429		gfp_t gfp_flags)
 430{
 431	struct nfs4_state_owner *sp;
 432
 433	sp = kzalloc(sizeof(*sp), gfp_flags);
 434	if (!sp)
 435		return NULL;
 
 436	sp->so_server = server;
 437	sp->so_cred = get_rpccred(cred);
 438	spin_lock_init(&sp->so_lock);
 439	INIT_LIST_HEAD(&sp->so_states);
 440	nfs4_init_seqid_counter(&sp->so_seqid);
 441	atomic_set(&sp->so_count, 1);
 442	INIT_LIST_HEAD(&sp->so_lru);
 
 443	return sp;
 444}
 445
 446static void
 447nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 448{
 449	struct rb_node *rb_node = &sp->so_server_node;
 450
 451	if (!RB_EMPTY_NODE(rb_node)) {
 452		struct nfs_server *server = sp->so_server;
 453		struct nfs_client *clp = server->nfs_client;
 454
 455		spin_lock(&clp->cl_lock);
 456		if (!RB_EMPTY_NODE(rb_node)) {
 457			rb_erase(rb_node, &server->state_owners);
 458			RB_CLEAR_NODE(rb_node);
 459		}
 460		spin_unlock(&clp->cl_lock);
 461	}
 462}
 463
 464static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 465{
 466	nfs4_destroy_seqid_counter(&sp->so_seqid);
 467	put_rpccred(sp->so_cred);
 468	kfree(sp);
 469}
 470
 471static void nfs4_gc_state_owners(struct nfs_server *server)
 472{
 473	struct nfs_client *clp = server->nfs_client;
 474	struct nfs4_state_owner *sp, *tmp;
 475	unsigned long time_min, time_max;
 476	LIST_HEAD(doomed);
 477
 478	spin_lock(&clp->cl_lock);
 479	time_max = jiffies;
 480	time_min = (long)time_max - (long)clp->cl_lease_time;
 481	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 482		/* NB: LRU is sorted so that oldest is at the head */
 483		if (time_in_range(sp->so_expires, time_min, time_max))
 484			break;
 485		list_move(&sp->so_lru, &doomed);
 486		nfs4_remove_state_owner_locked(sp);
 487	}
 488	spin_unlock(&clp->cl_lock);
 489
 490	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 491		list_del(&sp->so_lru);
 492		nfs4_free_state_owner(sp);
 493	}
 494}
 495
 496/**
 497 * nfs4_get_state_owner - Look up a state owner given a credential
 498 * @server: nfs_server to search
 499 * @cred: RPC credential to match
 
 500 *
 501 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 502 */
 503struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 504					      struct rpc_cred *cred,
 505					      gfp_t gfp_flags)
 506{
 507	struct nfs_client *clp = server->nfs_client;
 508	struct nfs4_state_owner *sp, *new;
 509
 510	spin_lock(&clp->cl_lock);
 511	sp = nfs4_find_state_owner_locked(server, cred);
 512	spin_unlock(&clp->cl_lock);
 513	if (sp != NULL)
 514		goto out;
 515	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 516	if (new == NULL)
 517		goto out;
 518	do {
 519		if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 520			break;
 521		spin_lock(&clp->cl_lock);
 522		sp = nfs4_insert_state_owner_locked(new);
 523		spin_unlock(&clp->cl_lock);
 524	} while (sp == ERR_PTR(-EAGAIN));
 525	if (sp != new)
 526		nfs4_free_state_owner(new);
 527out:
 528	nfs4_gc_state_owners(server);
 529	return sp;
 530}
 531
 532/**
 533 * nfs4_put_state_owner - Release a nfs4_state_owner
 534 * @sp: state owner data to release
 535 *
 536 * Note that we keep released state owners on an LRU
 537 * list.
 538 * This caches valid state owners so that they can be
 539 * reused, to avoid the OPEN_CONFIRM on minor version 0.
 540 * It also pins the uniquifier of dropped state owners for
 541 * a while, to ensure that those state owner names are
 542 * never reused.
 543 */
 544void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 545{
 546	struct nfs_server *server = sp->so_server;
 547	struct nfs_client *clp = server->nfs_client;
 548
 549	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 550		return;
 551
 552	sp->so_expires = jiffies;
 553	list_add_tail(&sp->so_lru, &server->state_owners_lru);
 554	spin_unlock(&clp->cl_lock);
 555}
 556
 557/**
 558 * nfs4_purge_state_owners - Release all cached state owners
 559 * @server: nfs_server with cached state owners to release
 
 560 *
 561 * Called at umount time.  Remaining state owners will be on
 562 * the LRU with ref count of zero.
 
 
 
 563 */
 564void nfs4_purge_state_owners(struct nfs_server *server)
 565{
 566	struct nfs_client *clp = server->nfs_client;
 567	struct nfs4_state_owner *sp, *tmp;
 568	LIST_HEAD(doomed);
 569
 570	spin_lock(&clp->cl_lock);
 571	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 572		list_move(&sp->so_lru, &doomed);
 573		nfs4_remove_state_owner_locked(sp);
 574	}
 575	spin_unlock(&clp->cl_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 576
 577	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 578		list_del(&sp->so_lru);
 579		nfs4_free_state_owner(sp);
 580	}
 581}
 582
 583static struct nfs4_state *
 584nfs4_alloc_open_state(void)
 585{
 586	struct nfs4_state *state;
 587
 588	state = kzalloc(sizeof(*state), GFP_NOFS);
 589	if (!state)
 590		return NULL;
 591	atomic_set(&state->count, 1);
 592	INIT_LIST_HEAD(&state->lock_states);
 593	spin_lock_init(&state->state_lock);
 594	seqlock_init(&state->seqlock);
 
 595	return state;
 596}
 597
 598void
 599nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 600{
 601	if (state->state == fmode)
 602		return;
 603	/* NB! List reordering - see the reclaim code for why.  */
 604	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 605		if (fmode & FMODE_WRITE)
 606			list_move(&state->open_states, &state->owner->so_states);
 607		else
 608			list_move_tail(&state->open_states, &state->owner->so_states);
 609	}
 610	state->state = fmode;
 611}
 612
 613static struct nfs4_state *
 614__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 615{
 616	struct nfs_inode *nfsi = NFS_I(inode);
 617	struct nfs4_state *state;
 618
 619	list_for_each_entry(state, &nfsi->open_states, inode_states) {
 620		if (state->owner != owner)
 621			continue;
 622		if (atomic_inc_not_zero(&state->count))
 
 
 623			return state;
 624	}
 625	return NULL;
 626}
 627
 628static void
 629nfs4_free_open_state(struct nfs4_state *state)
 630{
 631	kfree(state);
 632}
 633
 634struct nfs4_state *
 635nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 636{
 637	struct nfs4_state *state, *new;
 638	struct nfs_inode *nfsi = NFS_I(inode);
 639
 640	spin_lock(&inode->i_lock);
 641	state = __nfs4_find_state_byowner(inode, owner);
 642	spin_unlock(&inode->i_lock);
 643	if (state)
 644		goto out;
 645	new = nfs4_alloc_open_state();
 646	spin_lock(&owner->so_lock);
 647	spin_lock(&inode->i_lock);
 648	state = __nfs4_find_state_byowner(inode, owner);
 649	if (state == NULL && new != NULL) {
 650		state = new;
 651		state->owner = owner;
 652		atomic_inc(&owner->so_count);
 653		list_add(&state->inode_states, &nfsi->open_states);
 654		ihold(inode);
 655		state->inode = inode;
 
 656		spin_unlock(&inode->i_lock);
 657		/* Note: The reclaim code dictates that we add stateless
 658		 * and read-only stateids to the end of the list */
 659		list_add_tail(&state->open_states, &owner->so_states);
 660		spin_unlock(&owner->so_lock);
 661	} else {
 662		spin_unlock(&inode->i_lock);
 663		spin_unlock(&owner->so_lock);
 664		if (new)
 665			nfs4_free_open_state(new);
 666	}
 667out:
 668	return state;
 669}
 670
 671void nfs4_put_open_state(struct nfs4_state *state)
 672{
 673	struct inode *inode = state->inode;
 674	struct nfs4_state_owner *owner = state->owner;
 675
 676	if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 677		return;
 678	spin_lock(&inode->i_lock);
 679	list_del(&state->inode_states);
 680	list_del(&state->open_states);
 681	spin_unlock(&inode->i_lock);
 682	spin_unlock(&owner->so_lock);
 
 683	iput(inode);
 684	nfs4_free_open_state(state);
 685	nfs4_put_state_owner(owner);
 686}
 687
 688/*
 689 * Close the current file.
 690 */
 691static void __nfs4_close(struct nfs4_state *state,
 692		fmode_t fmode, gfp_t gfp_mask, int wait)
 693{
 694	struct nfs4_state_owner *owner = state->owner;
 695	int call_close = 0;
 696	fmode_t newstate;
 697
 698	atomic_inc(&owner->so_count);
 699	/* Protect against nfs4_find_state() */
 700	spin_lock(&owner->so_lock);
 701	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 702		case FMODE_READ:
 703			state->n_rdonly--;
 704			break;
 705		case FMODE_WRITE:
 706			state->n_wronly--;
 707			break;
 708		case FMODE_READ|FMODE_WRITE:
 709			state->n_rdwr--;
 710	}
 711	newstate = FMODE_READ|FMODE_WRITE;
 712	if (state->n_rdwr == 0) {
 713		if (state->n_rdonly == 0) {
 714			newstate &= ~FMODE_READ;
 715			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 716			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 717		}
 718		if (state->n_wronly == 0) {
 719			newstate &= ~FMODE_WRITE;
 720			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 721			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 722		}
 723		if (newstate == 0)
 724			clear_bit(NFS_DELEGATED_STATE, &state->flags);
 725	}
 726	nfs4_state_set_mode_locked(state, newstate);
 727	spin_unlock(&owner->so_lock);
 728
 729	if (!call_close) {
 730		nfs4_put_open_state(state);
 731		nfs4_put_state_owner(owner);
 732	} else {
 733		bool roc = pnfs_roc(state->inode);
 734
 735		nfs4_do_close(state, gfp_mask, wait, roc);
 736	}
 737}
 738
 739void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 740{
 741	__nfs4_close(state, fmode, GFP_NOFS, 0);
 742}
 743
 744void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 745{
 746	__nfs4_close(state, fmode, GFP_KERNEL, 1);
 747}
 748
 749/*
 750 * Search the state->lock_states for an existing lock_owner
 751 * that is compatible with current->files
 
 
 
 
 
 
 
 
 
 752 */
 753static struct nfs4_lock_state *
 754__nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 
 755{
 756	struct nfs4_lock_state *pos;
 757	list_for_each_entry(pos, &state->lock_states, ls_locks) {
 758		if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 759			continue;
 760		switch (pos->ls_owner.lo_type) {
 761		case NFS4_POSIX_LOCK_TYPE:
 762			if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 763				continue;
 764			break;
 765		case NFS4_FLOCK_LOCK_TYPE:
 766			if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 767				continue;
 768		}
 769		atomic_inc(&pos->ls_count);
 770		return pos;
 771	}
 772	return NULL;
 
 
 773}
 774
 775/*
 776 * Return a compatible lock_state. If no initialized lock_state structure
 777 * exists, return an uninitialized one.
 778 *
 779 */
 780static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 781{
 782	struct nfs4_lock_state *lsp;
 783	struct nfs_server *server = state->owner->so_server;
 784
 785	lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 786	if (lsp == NULL)
 787		return NULL;
 788	nfs4_init_seqid_counter(&lsp->ls_seqid);
 789	atomic_set(&lsp->ls_count, 1);
 790	lsp->ls_state = state;
 791	lsp->ls_owner.lo_type = type;
 792	switch (lsp->ls_owner.lo_type) {
 793	case NFS4_FLOCK_LOCK_TYPE:
 794		lsp->ls_owner.lo_u.flock_owner = fl_pid;
 795		break;
 796	case NFS4_POSIX_LOCK_TYPE:
 797		lsp->ls_owner.lo_u.posix_owner = fl_owner;
 798		break;
 799	default:
 800		goto out_free;
 801	}
 802	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 803	if (lsp->ls_seqid.owner_id < 0)
 804		goto out_free;
 805	INIT_LIST_HEAD(&lsp->ls_locks);
 806	return lsp;
 807out_free:
 808	kfree(lsp);
 809	return NULL;
 810}
 811
 812void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 813{
 814	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 815	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 816	kfree(lsp);
 817}
 818
 819/*
 820 * Return a compatible lock_state. If no initialized lock_state structure
 821 * exists, return an uninitialized one.
 822 *
 823 */
 824static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 825{
 826	struct nfs4_lock_state *lsp, *new = NULL;
 827	
 828	for(;;) {
 829		spin_lock(&state->state_lock);
 830		lsp = __nfs4_find_lock_state(state, owner, pid, type);
 831		if (lsp != NULL)
 832			break;
 833		if (new != NULL) {
 834			list_add(&new->ls_locks, &state->lock_states);
 835			set_bit(LK_STATE_IN_USE, &state->flags);
 836			lsp = new;
 837			new = NULL;
 838			break;
 839		}
 840		spin_unlock(&state->state_lock);
 841		new = nfs4_alloc_lock_state(state, owner, pid, type);
 842		if (new == NULL)
 843			return NULL;
 844	}
 845	spin_unlock(&state->state_lock);
 846	if (new != NULL)
 847		nfs4_free_lock_state(state->owner->so_server, new);
 848	return lsp;
 849}
 850
 851/*
 852 * Release reference to lock_state, and free it if we see that
 853 * it is no longer in use
 854 */
 855void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 856{
 
 857	struct nfs4_state *state;
 858
 859	if (lsp == NULL)
 860		return;
 861	state = lsp->ls_state;
 862	if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 863		return;
 864	list_del(&lsp->ls_locks);
 865	if (list_empty(&state->lock_states))
 866		clear_bit(LK_STATE_IN_USE, &state->flags);
 867	spin_unlock(&state->state_lock);
 868	if (lsp->ls_flags & NFS_LOCK_INITIALIZED) {
 869		if (nfs4_release_lockowner(lsp) == 0)
 870			return;
 871	}
 872	nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
 
 
 873}
 874
 875static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 876{
 877	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 878
 879	dst->fl_u.nfs4_fl.owner = lsp;
 880	atomic_inc(&lsp->ls_count);
 881}
 882
 883static void nfs4_fl_release_lock(struct file_lock *fl)
 884{
 885	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 886}
 887
 888static const struct file_lock_operations nfs4_fl_lock_ops = {
 889	.fl_copy_lock = nfs4_fl_copy_lock,
 890	.fl_release_private = nfs4_fl_release_lock,
 891};
 892
 893int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 894{
 895	struct nfs4_lock_state *lsp;
 896
 897	if (fl->fl_ops != NULL)
 898		return 0;
 899	if (fl->fl_flags & FL_POSIX)
 900		lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 901	else if (fl->fl_flags & FL_FLOCK)
 902		lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 903				NFS4_FLOCK_LOCK_TYPE);
 904	else
 905		return -EINVAL;
 906	if (lsp == NULL)
 907		return -ENOMEM;
 908	fl->fl_u.nfs4_fl.owner = lsp;
 909	fl->fl_ops = &nfs4_fl_lock_ops;
 910	return 0;
 911}
 912
 913static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 914		fl_owner_t fl_owner, pid_t fl_pid)
 
 915{
 916	struct nfs4_lock_state *lsp;
 917	bool ret = false;
 
 
 
 
 918
 919	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 920		goto out;
 921
 
 
 
 922	spin_lock(&state->state_lock);
 923	lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 924	if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0) {
 
 
 925		nfs4_stateid_copy(dst, &lsp->ls_stateid);
 926		ret = true;
 927	}
 928	spin_unlock(&state->state_lock);
 929	nfs4_put_lock_state(lsp);
 930out:
 931	return ret;
 932}
 933
 934static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 935{
 
 
 936	int seq;
 937
 938	do {
 
 
 939		seq = read_seqbegin(&state->seqlock);
 940		nfs4_stateid_copy(dst, &state->stateid);
 
 
 
 
 941	} while (read_seqretry(&state->seqlock, seq));
 
 942}
 943
 944/*
 945 * Byte-range lock aware utility to initialize the stateid of read/write
 946 * requests.
 947 */
 948void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 949		fmode_t fmode, fl_owner_t fl_owner, pid_t fl_pid)
 
 950{
 951	if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
 952		return;
 953	if (nfs4_copy_lock_stateid(dst, state, fl_owner, fl_pid))
 954		return;
 955	nfs4_copy_open_stateid(dst, state);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 956}
 957
 958struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 959{
 960	struct nfs_seqid *new;
 961
 962	new = kmalloc(sizeof(*new), gfp_mask);
 963	if (new != NULL) {
 964		new->sequence = counter;
 965		INIT_LIST_HEAD(&new->list);
 966		new->task = NULL;
 967	}
 968	return new;
 969}
 970
 971void nfs_release_seqid(struct nfs_seqid *seqid)
 972{
 973	struct nfs_seqid_counter *sequence;
 974
 975	if (list_empty(&seqid->list))
 976		return;
 977	sequence = seqid->sequence;
 978	spin_lock(&sequence->lock);
 979	list_del_init(&seqid->list);
 980	if (!list_empty(&sequence->list)) {
 981		struct nfs_seqid *next;
 982
 983		next = list_first_entry(&sequence->list,
 984				struct nfs_seqid, list);
 985		rpc_wake_up_queued_task(&sequence->wait, next->task);
 986	}
 
 987	spin_unlock(&sequence->lock);
 988}
 989
 990void nfs_free_seqid(struct nfs_seqid *seqid)
 991{
 992	nfs_release_seqid(seqid);
 993	kfree(seqid);
 994}
 995
 996/*
 997 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 998 * failed with a seqid incrementing error -
 999 * see comments nfs_fs.h:seqid_mutating_error()
1000 */
1001static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1002{
1003	BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
1004	switch (status) {
1005		case 0:
1006			break;
1007		case -NFS4ERR_BAD_SEQID:
1008			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1009				return;
1010			pr_warn_ratelimited("NFS: v4 server returned a bad"
1011					" sequence-id error on an"
1012					" unconfirmed sequence %p!\n",
1013					seqid->sequence);
 
1014		case -NFS4ERR_STALE_CLIENTID:
1015		case -NFS4ERR_STALE_STATEID:
1016		case -NFS4ERR_BAD_STATEID:
1017		case -NFS4ERR_BADXDR:
1018		case -NFS4ERR_RESOURCE:
1019		case -NFS4ERR_NOFILEHANDLE:
 
1020			/* Non-seqid mutating errors */
1021			return;
1022	};
1023	/*
1024	 * Note: no locking needed as we are guaranteed to be first
1025	 * on the sequence list
1026	 */
1027	seqid->sequence->counter++;
1028}
1029
1030void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1031{
1032	struct nfs4_state_owner *sp = container_of(seqid->sequence,
1033					struct nfs4_state_owner, so_seqid);
1034	struct nfs_server *server = sp->so_server;
 
1035
 
1036	if (status == -NFS4ERR_BAD_SEQID)
1037		nfs4_drop_state_owner(sp);
1038	if (!nfs4_has_session(server->nfs_client))
1039		nfs_increment_seqid(status, seqid);
1040}
1041
1042/*
1043 * Increment the seqid if the LOCK/LOCKU succeeded, or
1044 * failed with a seqid incrementing error -
1045 * see comments nfs_fs.h:seqid_mutating_error()
1046 */
1047void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1048{
1049	nfs_increment_seqid(status, seqid);
 
1050}
1051
1052int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1053{
1054	struct nfs_seqid_counter *sequence = seqid->sequence;
1055	int status = 0;
1056
 
 
 
1057	spin_lock(&sequence->lock);
1058	seqid->task = task;
1059	if (list_empty(&seqid->list))
1060		list_add_tail(&seqid->list, &sequence->list);
1061	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1062		goto unlock;
1063	rpc_sleep_on(&sequence->wait, task, NULL);
1064	status = -EAGAIN;
1065unlock:
1066	spin_unlock(&sequence->lock);
 
1067	return status;
1068}
1069
1070static int nfs4_run_state_manager(void *);
1071
1072static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1073{
1074	smp_mb__before_clear_bit();
1075	clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1076	smp_mb__after_clear_bit();
1077	wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1078	rpc_wake_up(&clp->cl_rpcwaitq);
1079}
1080
1081/*
1082 * Schedule the nfs_client asynchronous state management routine
1083 */
1084void nfs4_schedule_state_manager(struct nfs_client *clp)
1085{
1086	struct task_struct *task;
1087	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1088
1089	if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1090		return;
 
1091	__module_get(THIS_MODULE);
1092	atomic_inc(&clp->cl_count);
1093
1094	/* The rcu_read_lock() is not strictly necessary, as the state
1095	 * manager is the only thread that ever changes the rpc_xprt
1096	 * after it's initialized.  At this point, we're single threaded. */
1097	rcu_read_lock();
1098	snprintf(buf, sizeof(buf), "%s-manager",
1099			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1100	rcu_read_unlock();
1101	task = kthread_run(nfs4_run_state_manager, clp, buf);
1102	if (IS_ERR(task)) {
1103		printk(KERN_ERR "%s: kthread_run: %ld\n",
1104			__func__, PTR_ERR(task));
 
 
 
 
1105		nfs4_clear_state_manager_bit(clp);
1106		nfs_put_client(clp);
1107		module_put(THIS_MODULE);
1108	}
1109}
1110
1111/*
1112 * Schedule a lease recovery attempt
1113 */
1114void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1115{
1116	if (!clp)
1117		return;
1118	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1119		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1120	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1121			clp->cl_hostname);
1122	nfs4_schedule_state_manager(clp);
1123}
1124EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1125
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1126/*
1127 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1128 * @clp: client to process
1129 *
1130 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1131 * resend of the SETCLIENTID and hence re-establish the
1132 * callback channel. Then return all existing delegations.
1133 */
1134static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1135{
1136	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1137	nfs_expire_all_delegations(clp);
1138	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1139			clp->cl_hostname);
1140}
1141
1142void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1143{
1144	nfs40_handle_cb_pathdown(clp);
1145	nfs4_schedule_state_manager(clp);
1146}
1147
1148static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1149{
1150
 
 
1151	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1152	/* Don't recover state that expired before the reboot */
1153	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1154		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1155		return 0;
1156	}
1157	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1158	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1159	return 1;
1160}
1161
1162static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1163{
 
 
1164	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1165	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1166	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1167	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1168	return 1;
1169}
1170
1171void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1172{
1173	struct nfs_client *clp = server->nfs_client;
1174
1175	nfs4_state_mark_reclaim_nograce(clp, state);
 
 
 
1176	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1177			clp->cl_hostname);
1178	nfs4_schedule_state_manager(clp);
 
1179}
1180EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1181
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1182void nfs_inode_find_state_and_recover(struct inode *inode,
1183		const nfs4_stateid *stateid)
1184{
1185	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1186	struct nfs_inode *nfsi = NFS_I(inode);
1187	struct nfs_open_context *ctx;
1188	struct nfs4_state *state;
1189	bool found = false;
1190
1191	spin_lock(&inode->i_lock);
1192	list_for_each_entry(ctx, &nfsi->open_files, list) {
1193		state = ctx->state;
1194		if (state == NULL)
1195			continue;
1196		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
 
 
1197			continue;
1198		if (!nfs4_stateid_match(&state->stateid, stateid))
 
 
 
 
1199			continue;
1200		nfs4_state_mark_reclaim_nograce(clp, state);
1201		found = true;
 
 
1202	}
1203	spin_unlock(&inode->i_lock);
 
 
1204	if (found)
1205		nfs4_schedule_state_manager(clp);
1206}
1207
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1208
1209static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1210{
1211	struct inode *inode = state->inode;
1212	struct nfs_inode *nfsi = NFS_I(inode);
1213	struct file_lock *fl;
 
1214	int status = 0;
 
 
1215
1216	if (inode->i_flock == NULL)
1217		return 0;
1218
 
 
1219	/* Guard against delegation returns and new lock/unlock calls */
1220	down_write(&nfsi->rwsem);
1221	/* Protect inode->i_flock using the BKL */
1222	lock_flocks();
1223	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1224		if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1225			continue;
1226		if (nfs_file_open_context(fl->fl_file)->state != state)
1227			continue;
1228		unlock_flocks();
1229		status = ops->recover_lock(state, fl);
1230		switch (status) {
1231			case 0:
1232				break;
1233			case -ESTALE:
1234			case -NFS4ERR_ADMIN_REVOKED:
1235			case -NFS4ERR_STALE_STATEID:
1236			case -NFS4ERR_BAD_STATEID:
1237			case -NFS4ERR_EXPIRED:
1238			case -NFS4ERR_NO_GRACE:
1239			case -NFS4ERR_STALE_CLIENTID:
1240			case -NFS4ERR_BADSESSION:
1241			case -NFS4ERR_BADSLOT:
1242			case -NFS4ERR_BAD_HIGH_SLOT:
1243			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1244				goto out;
1245			default:
1246				printk(KERN_ERR "NFS: %s: unhandled error %d. "
1247					"Zeroing state\n", __func__, status);
1248			case -ENOMEM:
1249			case -NFS4ERR_DENIED:
1250			case -NFS4ERR_RECLAIM_BAD:
1251			case -NFS4ERR_RECLAIM_CONFLICT:
1252				/* kill_proc(fl->fl_pid, SIGLOST, 1); */
1253				status = 0;
 
 
 
 
1254		}
1255		lock_flocks();
1256	}
1257	unlock_flocks();
 
 
 
 
1258out:
1259	up_write(&nfsi->rwsem);
1260	return status;
1261}
1262
1263static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1264{
1265	struct nfs4_state *state;
1266	struct nfs4_lock_state *lock;
1267	int status = 0;
 
 
 
1268
1269	/* Note: we rely on the sp->so_states list being ordered 
1270	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1271	 * states first.
1272	 * This is needed to ensure that the server won't give us any
1273	 * read delegations that we have to return if, say, we are
1274	 * recovering after a network partition or a reboot from a
1275	 * server that doesn't support a grace period.
1276	 */
 
1277restart:
1278	spin_lock(&sp->so_lock);
1279	list_for_each_entry(state, &sp->so_states, open_states) {
1280		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1281			continue;
 
 
1282		if (state->state == 0)
1283			continue;
1284		atomic_inc(&state->count);
 
 
 
 
 
 
 
1285		spin_unlock(&sp->so_lock);
1286		status = ops->recover_open(sp, state);
1287		if (status >= 0) {
1288			status = nfs4_reclaim_locks(state, ops);
 
1289			if (status >= 0) {
1290				spin_lock(&state->state_lock);
1291				list_for_each_entry(lock, &state->lock_states, ls_locks) {
1292					if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1293						pr_warn_ratelimited("NFS: "
1294							"%s: Lock reclaim "
1295							"failed!\n", __func__);
1296				}
1297				spin_unlock(&state->state_lock);
1298				nfs4_put_open_state(state);
1299				goto restart;
1300			}
1301		}
1302		switch (status) {
1303			default:
1304				printk(KERN_ERR "NFS: %s: unhandled error %d. "
1305					"Zeroing state\n", __func__, status);
1306			case -ENOENT:
1307			case -ENOMEM:
1308			case -ESTALE:
1309				/*
1310				 * Open state on this file cannot be recovered
1311				 * All we can do is revert to using the zero stateid.
1312				 */
1313				memset(&state->stateid, 0,
1314					sizeof(state->stateid));
1315				/* Mark the file as being 'closed' */
1316				state->state = 0;
1317				break;
1318			case -EKEYEXPIRED:
1319				/*
1320				 * User RPCSEC_GSS context has expired.
1321				 * We cannot recover this stateid now, so
1322				 * skip it and allow recovery thread to
1323				 * proceed.
1324				 */
1325				break;
1326			case -NFS4ERR_ADMIN_REVOKED:
1327			case -NFS4ERR_STALE_STATEID:
1328			case -NFS4ERR_BAD_STATEID:
1329			case -NFS4ERR_RECLAIM_BAD:
1330			case -NFS4ERR_RECLAIM_CONFLICT:
1331				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1332				break;
1333			case -NFS4ERR_EXPIRED:
1334			case -NFS4ERR_NO_GRACE:
1335				nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1336			case -NFS4ERR_STALE_CLIENTID:
1337			case -NFS4ERR_BADSESSION:
1338			case -NFS4ERR_BADSLOT:
1339			case -NFS4ERR_BAD_HIGH_SLOT:
1340			case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1341				goto out_err;
1342		}
1343		nfs4_put_open_state(state);
 
1344		goto restart;
1345	}
1346	spin_unlock(&sp->so_lock);
 
 
 
 
1347	return 0;
1348out_err:
1349	nfs4_put_open_state(state);
 
 
1350	return status;
1351}
1352
1353static void nfs4_clear_open_state(struct nfs4_state *state)
1354{
1355	struct nfs4_lock_state *lock;
1356
1357	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1358	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1359	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1360	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1361	spin_lock(&state->state_lock);
1362	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1363		lock->ls_seqid.flags = 0;
1364		lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1365	}
1366	spin_unlock(&state->state_lock);
1367}
1368
1369static void nfs4_reset_seqids(struct nfs_server *server,
1370	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1371{
1372	struct nfs_client *clp = server->nfs_client;
1373	struct nfs4_state_owner *sp;
1374	struct rb_node *pos;
1375	struct nfs4_state *state;
1376
1377	spin_lock(&clp->cl_lock);
1378	for (pos = rb_first(&server->state_owners);
1379	     pos != NULL;
1380	     pos = rb_next(pos)) {
1381		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1382		sp->so_seqid.flags = 0;
1383		spin_lock(&sp->so_lock);
1384		list_for_each_entry(state, &sp->so_states, open_states) {
1385			if (mark_reclaim(clp, state))
1386				nfs4_clear_open_state(state);
1387		}
1388		spin_unlock(&sp->so_lock);
1389	}
1390	spin_unlock(&clp->cl_lock);
1391}
1392
1393static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1394	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1395{
1396	struct nfs_server *server;
1397
1398	rcu_read_lock();
1399	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1400		nfs4_reset_seqids(server, mark_reclaim);
1401	rcu_read_unlock();
1402}
1403
1404static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1405{
 
1406	/* Mark all delegations for reclaim */
1407	nfs_delegation_mark_reclaim(clp);
1408	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1409}
1410
1411static void nfs4_reclaim_complete(struct nfs_client *clp,
1412				 const struct nfs4_state_recovery_ops *ops)
 
1413{
1414	/* Notify the server we're done reclaiming our state */
1415	if (ops->reclaim_complete)
1416		(void)ops->reclaim_complete(clp);
 
1417}
1418
1419static void nfs4_clear_reclaim_server(struct nfs_server *server)
1420{
1421	struct nfs_client *clp = server->nfs_client;
1422	struct nfs4_state_owner *sp;
1423	struct rb_node *pos;
1424	struct nfs4_state *state;
1425
1426	spin_lock(&clp->cl_lock);
1427	for (pos = rb_first(&server->state_owners);
1428	     pos != NULL;
1429	     pos = rb_next(pos)) {
1430		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1431		spin_lock(&sp->so_lock);
1432		list_for_each_entry(state, &sp->so_states, open_states) {
1433			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1434						&state->flags))
1435				continue;
1436			nfs4_state_mark_reclaim_nograce(clp, state);
1437		}
1438		spin_unlock(&sp->so_lock);
1439	}
1440	spin_unlock(&clp->cl_lock);
1441}
1442
1443static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1444{
1445	struct nfs_server *server;
1446
1447	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1448		return 0;
1449
1450	rcu_read_lock();
1451	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1452		nfs4_clear_reclaim_server(server);
1453	rcu_read_unlock();
1454
1455	nfs_delegation_reap_unclaimed(clp);
1456	return 1;
1457}
1458
1459static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1460{
 
 
 
 
1461	if (!nfs4_state_clear_reclaim_reboot(clp))
1462		return;
1463	nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1464}
1465
1466static void nfs_delegation_clear_all(struct nfs_client *clp)
1467{
1468	nfs_delegation_mark_reclaim(clp);
1469	nfs_delegation_reap_unclaimed(clp);
1470}
1471
1472static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1473{
1474	nfs_delegation_clear_all(clp);
1475	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1476}
1477
1478static void nfs4_warn_keyexpired(const char *s)
1479{
1480	printk_ratelimited(KERN_WARNING "Error: state manager"
1481			" encountered RPCSEC_GSS session"
1482			" expired against NFSv4 server %s.\n",
1483			s);
1484}
1485
1486static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1487{
1488	switch (error) {
1489		case 0:
1490			break;
1491		case -NFS4ERR_CB_PATH_DOWN:
1492			nfs40_handle_cb_pathdown(clp);
1493			break;
1494		case -NFS4ERR_NO_GRACE:
1495			nfs4_state_end_reclaim_reboot(clp);
1496			break;
1497		case -NFS4ERR_STALE_CLIENTID:
1498		case -NFS4ERR_LEASE_MOVED:
1499			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1500			nfs4_state_clear_reclaim_reboot(clp);
1501			nfs4_state_start_reclaim_reboot(clp);
1502			break;
1503		case -NFS4ERR_EXPIRED:
1504			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1505			nfs4_state_start_reclaim_nograce(clp);
1506			break;
1507		case -NFS4ERR_BADSESSION:
1508		case -NFS4ERR_BADSLOT:
1509		case -NFS4ERR_BAD_HIGH_SLOT:
1510		case -NFS4ERR_DEADSESSION:
1511		case -NFS4ERR_SEQ_FALSE_RETRY:
1512		case -NFS4ERR_SEQ_MISORDERED:
1513			set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1514			/* Zero session reset errors */
1515			break;
1516		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1517			set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1518			break;
1519		case -EKEYEXPIRED:
1520			/* Nothing we can do */
1521			nfs4_warn_keyexpired(clp->cl_hostname);
1522			break;
1523		default:
1524			dprintk("%s: failed to handle error %d for server %s\n",
1525					__func__, error, clp->cl_hostname);
1526			return error;
1527	}
1528	dprintk("%s: handled error %d for server %s\n", __func__, error,
1529			clp->cl_hostname);
1530	return 0;
1531}
1532
1533static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1534{
1535	struct nfs4_state_owner *sp;
1536	struct nfs_server *server;
1537	struct rb_node *pos;
1538	int status = 0;
 
 
1539
 
 
 
1540restart:
1541	rcu_read_lock();
1542	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1543		nfs4_purge_state_owners(server);
1544		spin_lock(&clp->cl_lock);
1545		for (pos = rb_first(&server->state_owners);
1546		     pos != NULL;
1547		     pos = rb_next(pos)) {
1548			sp = rb_entry(pos,
1549				struct nfs4_state_owner, so_server_node);
1550			if (!test_and_clear_bit(ops->owner_flag_bit,
1551							&sp->so_flags))
1552				continue;
1553			atomic_inc(&sp->so_count);
 
1554			spin_unlock(&clp->cl_lock);
1555			rcu_read_unlock();
1556
1557			status = nfs4_reclaim_open_state(sp, ops);
1558			if (status < 0) {
 
 
 
1559				set_bit(ops->owner_flag_bit, &sp->so_flags);
1560				nfs4_put_state_owner(sp);
1561				return nfs4_recovery_handle_error(clp, status);
 
 
1562			}
1563
1564			nfs4_put_state_owner(sp);
1565			goto restart;
1566		}
1567		spin_unlock(&clp->cl_lock);
1568	}
1569	rcu_read_unlock();
1570	return status;
 
 
 
 
1571}
1572
1573static int nfs4_check_lease(struct nfs_client *clp)
1574{
1575	struct rpc_cred *cred;
1576	const struct nfs4_state_maintenance_ops *ops =
1577		clp->cl_mvops->state_renewal_ops;
1578	int status;
1579
1580	/* Is the client already known to have an expired lease? */
1581	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1582		return 0;
1583	spin_lock(&clp->cl_lock);
1584	cred = ops->get_state_renewal_cred_locked(clp);
1585	spin_unlock(&clp->cl_lock);
1586	if (cred == NULL) {
1587		cred = nfs4_get_setclientid_cred(clp);
1588		status = -ENOKEY;
1589		if (cred == NULL)
1590			goto out;
1591	}
1592	status = ops->renew_lease(clp, cred);
1593	put_rpccred(cred);
 
 
 
 
1594out:
1595	return nfs4_recovery_handle_error(clp, status);
1596}
1597
1598/* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1599 * on EXCHANGE_ID for v4.1
1600 */
1601static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1602{
1603	switch (status) {
1604	case -NFS4ERR_SEQ_MISORDERED:
1605		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1606			return -ESERVERFAULT;
1607		/* Lease confirmation error: retry after purging the lease */
1608		ssleep(1);
1609	case -NFS4ERR_CLID_INUSE:
 
1610	case -NFS4ERR_STALE_CLIENTID:
1611		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 
1612		break;
 
 
 
 
 
 
 
 
 
 
 
 
1613	case -EACCES:
1614		if (clp->cl_machine_cred == NULL)
1615			return -EACCES;
1616		/* Handle case where the user hasn't set up machine creds */
1617		nfs4_clear_machine_cred(clp);
1618	case -NFS4ERR_DELAY:
1619	case -ETIMEDOUT:
1620	case -EAGAIN:
1621		ssleep(1);
1622		break;
1623
1624	case -NFS4ERR_MINOR_VERS_MISMATCH:
1625		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1626			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1627		dprintk("%s: exit with error %d for server %s\n",
1628				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
1629		return -EPROTONOSUPPORT;
1630	case -EKEYEXPIRED:
1631		nfs4_warn_keyexpired(clp->cl_hostname);
 
 
1632	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1633				 * in nfs4_exchange_id */
1634	default:
1635		dprintk("%s: exit with error %d for server %s\n", __func__,
1636				status, clp->cl_hostname);
1637		return status;
1638	}
1639	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1640	dprintk("%s: handled error %d for server %s\n", __func__, status,
1641			clp->cl_hostname);
1642	return 0;
1643}
1644
1645static int nfs4_reclaim_lease(struct nfs_client *clp)
1646{
1647	struct rpc_cred *cred;
1648	const struct nfs4_state_recovery_ops *ops =
1649		clp->cl_mvops->reboot_recovery_ops;
1650	int status;
1651
1652	cred = ops->get_clid_cred(clp);
 
 
 
1653	if (cred == NULL)
1654		return -ENOENT;
1655	status = ops->establish_clid(clp, cred);
1656	put_rpccred(cred);
1657	if (status != 0)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1658		return nfs4_handle_reclaim_lease_error(clp, status);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1659	return 0;
1660}
1661
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1662#ifdef CONFIG_NFS_V4_1
1663void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1664{
1665	struct nfs_client *clp = session->clp;
1666
1667	switch (err) {
1668	default:
1669		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1670		break;
1671	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1672		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1673	}
1674	nfs4_schedule_lease_recovery(clp);
1675}
1676EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1677
1678void nfs41_handle_recall_slot(struct nfs_client *clp)
1679{
1680	set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1681	dprintk("%s: scheduling slot recall for server %s\n", __func__,
1682			clp->cl_hostname);
1683	nfs4_schedule_state_manager(clp);
1684}
1685
1686static void nfs4_reset_all_state(struct nfs_client *clp)
1687{
1688	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1689		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1690		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1691		nfs4_state_start_reclaim_nograce(clp);
1692		dprintk("%s: scheduling reset of all state for server %s!\n",
1693				__func__, clp->cl_hostname);
1694		nfs4_schedule_state_manager(clp);
1695	}
1696}
1697
1698static void nfs41_handle_server_reboot(struct nfs_client *clp)
1699{
1700	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1701		nfs4_state_start_reclaim_reboot(clp);
1702		dprintk("%s: server %s rebooted!\n", __func__,
1703				clp->cl_hostname);
1704		nfs4_schedule_state_manager(clp);
1705	}
1706}
1707
1708static void nfs41_handle_state_revoked(struct nfs_client *clp)
1709{
1710	nfs4_reset_all_state(clp);
1711	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1712}
1713
 
 
 
 
 
 
 
 
1714static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1715{
1716	/* This will need to handle layouts too */
1717	nfs_expire_all_delegations(clp);
 
1718	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1719			clp->cl_hostname);
1720}
1721
1722static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1723{
1724	nfs_expire_all_delegations(clp);
1725	if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1726		nfs4_schedule_state_manager(clp);
1727	dprintk("%s: server %s declared a backchannel fault\n", __func__,
1728			clp->cl_hostname);
1729}
1730
1731static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1732{
1733	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1734		&clp->cl_state) == 0)
1735		nfs4_schedule_state_manager(clp);
1736}
1737
1738void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
 
1739{
1740	if (!flags)
1741		return;
1742
1743	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
1744		__func__, clp->cl_hostname, clp->cl_clientid, flags);
 
 
 
 
 
 
 
 
1745
1746	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1747		nfs41_handle_server_reboot(clp);
1748	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1749			    SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1750			    SEQ4_STATUS_ADMIN_STATE_REVOKED |
1751			    SEQ4_STATUS_LEASE_MOVED))
1752		nfs41_handle_state_revoked(clp);
 
 
1753	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1754		nfs41_handle_recallable_state_revoked(clp);
 
1755	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
1756		nfs41_handle_backchannel_fault(clp);
1757	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1758				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1759		nfs41_handle_cb_path_down(clp);
1760}
1761
1762static int nfs4_reset_session(struct nfs_client *clp)
1763{
1764	struct rpc_cred *cred;
1765	int status;
1766
1767	nfs4_begin_drain_session(clp);
1768	cred = nfs4_get_exchange_id_cred(clp);
 
 
 
 
1769	status = nfs4_proc_destroy_session(clp->cl_session, cred);
1770	if (status && status != -NFS4ERR_BADSESSION &&
1771	    status != -NFS4ERR_DEADSESSION) {
 
 
 
 
 
 
 
 
 
 
1772		status = nfs4_recovery_handle_error(clp, status);
1773		goto out;
1774	}
1775
1776	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1777	status = nfs4_proc_create_session(clp, cred);
1778	if (status) {
1779		dprintk("%s: session reset failed with status %d for server %s!\n",
1780			__func__, status, clp->cl_hostname);
1781		status = nfs4_handle_reclaim_lease_error(clp, status);
1782		goto out;
1783	}
1784	nfs41_finish_session_reset(clp);
1785	dprintk("%s: session reset was successful for server %s!\n",
1786			__func__, clp->cl_hostname);
1787out:
1788	if (cred)
1789		put_rpccred(cred);
1790	return status;
1791}
1792
1793static int nfs4_recall_slot(struct nfs_client *clp)
1794{
1795	struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1796	struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1797	struct nfs4_slot *new, *old;
1798	int i;
1799
1800	nfs4_begin_drain_session(clp);
1801	new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1802		      GFP_NOFS);
1803        if (!new)
1804		return -ENOMEM;
1805
1806	spin_lock(&fc_tbl->slot_tbl_lock);
1807	for (i = 0; i < fc_tbl->target_max_slots; i++)
1808		new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1809	old = fc_tbl->slots;
1810	fc_tbl->slots = new;
1811	fc_tbl->max_slots = fc_tbl->target_max_slots;
1812	fc_tbl->target_max_slots = 0;
1813	fc_attrs->max_reqs = fc_tbl->max_slots;
1814	spin_unlock(&fc_tbl->slot_tbl_lock);
1815
1816	kfree(old);
1817	nfs4_end_drain_session(clp);
1818	return 0;
1819}
1820
1821static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1822{
1823	struct rpc_cred *cred;
1824	int ret;
1825
1826	nfs4_begin_drain_session(clp);
1827	cred = nfs4_get_exchange_id_cred(clp);
 
 
 
 
1828	ret = nfs4_proc_bind_conn_to_session(clp, cred);
1829	if (cred)
1830		put_rpccred(cred);
1831	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1832	switch (ret) {
1833	case 0:
1834		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
1835			__func__, clp->cl_hostname);
1836		break;
1837	case -NFS4ERR_DELAY:
1838		ssleep(1);
1839		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1840		break;
1841	default:
1842		return nfs4_recovery_handle_error(clp, ret);
1843	}
1844	return 0;
1845}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1846#else /* CONFIG_NFS_V4_1 */
1847static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1848static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1849static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1850
1851static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1852{
1853	return 0;
1854}
 
 
 
 
1855#endif /* CONFIG_NFS_V4_1 */
1856
1857static void nfs4_state_manager(struct nfs_client *clp)
1858{
 
1859	int status = 0;
 
 
 
 
 
 
 
 
1860
1861	/* Ensure exclusive access to NFSv4 state */
1862	do {
 
 
1863		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
1864			status = nfs4_reclaim_lease(clp);
 
1865			if (status < 0)
1866				goto out_error;
1867			clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1868			set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1869		}
1870
1871		if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
 
1872			/* We're going to have to re-establish a clientid */
1873			status = nfs4_reclaim_lease(clp);
1874			if (status < 0)
1875				goto out_error;
 
 
 
 
 
 
 
1876			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1877				continue;
1878			clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
 
 
1879
1880			if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1881					       &clp->cl_state))
1882				nfs4_state_start_reclaim_nograce(clp);
1883			else
1884				set_bit(NFS4CLNT_RECLAIM_REBOOT,
1885					&clp->cl_state);
1886
1887			pnfs_destroy_all_layouts(clp);
1888		}
1889
1890		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
 
1891			status = nfs4_check_lease(clp);
1892			if (status < 0)
1893				goto out_error;
1894			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1895				continue;
1896		}
1897
1898		/* Initialize or reset the session */
1899		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1900		   && nfs4_has_session(clp)) {
1901			status = nfs4_reset_session(clp);
1902			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1903				continue;
1904			if (status < 0)
1905				goto out_error;
1906		}
1907
1908		/* Send BIND_CONN_TO_SESSION */
1909		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1910				&clp->cl_state) && nfs4_has_session(clp)) {
1911			status = nfs4_bind_conn_to_session(clp);
1912			if (status < 0)
1913				goto out_error;
1914			continue;
1915		}
1916
1917		/* First recover reboot state... */
1918		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
 
1919			status = nfs4_do_reclaim(clp,
1920				clp->cl_mvops->reboot_recovery_ops);
1921			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1922			    test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
 
1923				continue;
 
 
1924			nfs4_state_end_reclaim_reboot(clp);
1925			if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1926				continue;
 
 
 
 
 
1927			if (status < 0)
1928				goto out_error;
 
 
1929		}
1930
1931		/* Now recover expired state... */
1932		if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
 
1933			status = nfs4_do_reclaim(clp,
1934				clp->cl_mvops->nograce_recovery_ops);
1935			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1936			    test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1937			    test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1938				continue;
1939			if (status < 0)
1940				goto out_error;
 
1941		}
1942
 
1943		nfs4_end_drain_session(clp);
1944		if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1945			nfs_client_return_marked_delegations(clp);
 
 
 
 
1946			continue;
1947		}
1948		/* Recall session slots */
1949		if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1950		   && nfs4_has_session(clp)) {
1951			status = nfs4_recall_slot(clp);
1952			if (status < 0)
1953				goto out_error;
1954			continue;
 
1955		}
1956
 
 
 
 
1957
1958		nfs4_clear_state_manager_bit(clp);
1959		/* Did we race with an attempt to give us more work? */
1960		if (clp->cl_state == 0)
1961			break;
1962		if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1963			break;
1964	} while (atomic_read(&clp->cl_count) > 1);
1965	return;
1966out_error:
1967	pr_warn_ratelimited("NFS: state manager failed on NFSv4 server %s"
1968			" with error %d\n", clp->cl_hostname, -status);
 
 
 
 
 
 
 
1969	nfs4_end_drain_session(clp);
1970	nfs4_clear_state_manager_bit(clp);
1971}
1972
1973static int nfs4_run_state_manager(void *ptr)
1974{
1975	struct nfs_client *clp = ptr;
 
 
 
 
1976
1977	allow_signal(SIGKILL);
 
1978	nfs4_state_manager(clp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1979	nfs_put_client(clp);
1980	module_put_and_exit(0);
1981	return 0;
1982}
1983
1984/*
1985 * Local variables:
1986 *  c-basic-offset: 8
1987 * End:
1988 */