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