fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/kthread.h>
  46 #include <linux/module.h>
  47 #include <linux/random.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51 #include <linux/jiffies.h>
  52
  53 #include <linux/sunrpc/clnt.h>
  54
  55 #include "nfs4_fs.h"
  56 #include "callback.h"
  57 #include "delegation.h"
  58 #include "internal.h"
  59 #include "nfs4idmap.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid = {
  69         { .data = { 0 } },
  70         .type = NFS4_SPECIAL_STATEID_TYPE,
  71 };
  72 const nfs4_stateid invalid_stateid = {
  73         {
  74                 /* Funky initialiser keeps older gcc versions happy */
  75                 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
  76         },
  77         .type = NFS4_INVALID_STATEID_TYPE,
  78 };
  79
  80 const nfs4_stateid current_stateid = {
  81         {
  82                 /* Funky initialiser keeps older gcc versions happy */
  83                 .data = { 0x0, 0x0, 0x0, 0x1, 0 },
  84         },
  85         .type = NFS4_SPECIAL_STATEID_TYPE,
  86 };
  87
  88 static DEFINE_MUTEX(nfs_clid_init_mutex);
  89
  90 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  91 {
  92         struct nfs4_setclientid_res clid = {
  93                 .clientid = clp->cl_clientid,
  94                 .confirm = clp->cl_confirm,
  95         };
  96         unsigned short port;
  97         int status;
  98         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  99
 100         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 101                 goto do_confirm;
 102         port = nn->nfs_callback_tcpport;
 103         if (clp->cl_addr.ss_family == AF_INET6)
 104                 port = nn->nfs_callback_tcpport6;
 105
 106         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 107         if (status != 0)
 108                 goto out;
 109         clp->cl_clientid = clid.clientid;
 110         clp->cl_confirm = clid.confirm;
 111         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 112 do_confirm:
 113         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
 114         if (status != 0)
 115                 goto out;
 116         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 117         nfs4_schedule_state_renewal(clp);
 118 out:
 119         return status;
 120 }
 121
 122 /**
 123  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 124  *
 125  * @clp: nfs_client under test
 126  * @result: OUT: found nfs_client, or clp
 127  * @cred: credential to use for trunking test
 128  *
 129  * Returns zero, a negative errno, or a negative NFS4ERR status.
 130  * If zero is returned, an nfs_client pointer is planted in
 131  * "result".
 132  *
 133  * Note: The returned client may not yet be marked ready.
 134  */
 135 int nfs40_discover_server_trunking(struct nfs_client *clp,
 136                                    struct nfs_client **result,
 137                                    struct rpc_cred *cred)
 138 {
 139         struct nfs4_setclientid_res clid = {
 140                 .clientid = clp->cl_clientid,
 141                 .confirm = clp->cl_confirm,
 142         };
 143         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 144         unsigned short port;
 145         int status;
 146
 147         port = nn->nfs_callback_tcpport;
 148         if (clp->cl_addr.ss_family == AF_INET6)
 149                 port = nn->nfs_callback_tcpport6;
 150
 151         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 152         if (status != 0)
 153                 goto out;
 154         clp->cl_clientid = clid.clientid;
 155         clp->cl_confirm = clid.confirm;
 156
 157         status = nfs40_walk_client_list(clp, result, cred);
 158         if (status == 0) {
 159                 /* Sustain the lease, even if it's empty.  If the clientid4
 160                  * goes stale it's of no use for trunking discovery. */
 161                 nfs4_schedule_state_renewal(*result);
 162         }
 163 out:
 164         return status;
 165 }
 166
 167 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 168 {
 169         struct rpc_cred *cred = NULL;
 170
 171         if (clp->cl_machine_cred != NULL)
 172                 cred = get_rpccred(clp->cl_machine_cred);
 173         return cred;
 174 }
 175
 176 static void nfs4_root_machine_cred(struct nfs_client *clp)
 177 {
 178         struct rpc_cred *cred, *new;
 179
 180         new = rpc_lookup_machine_cred(NULL);
 181         spin_lock(&clp->cl_lock);
 182         cred = clp->cl_machine_cred;
 183         clp->cl_machine_cred = new;
 184         spin_unlock(&clp->cl_lock);
 185         if (cred != NULL)
 186                 put_rpccred(cred);
 187 }
 188
 189 static struct rpc_cred *
 190 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 191 {
 192         struct rpc_cred *cred = NULL;
 193         struct nfs4_state_owner *sp;
 194         struct rb_node *pos;
 195
 196         for (pos = rb_first(&server->state_owners);
 197              pos != NULL;
 198              pos = rb_next(pos)) {
 199                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 200                 if (list_empty(&sp->so_states))
 201                         continue;
 202                 cred = get_rpccred(sp->so_cred);
 203                 break;
 204         }
 205         return cred;
 206 }
 207
 208 /**
 209  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 210  * @clp: client state handle
 211  *
 212  * Returns an rpc_cred with reference count bumped, or NULL.
 213  * Caller must hold clp->cl_lock.
 214  */
 215 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 216 {
 217         struct rpc_cred *cred = NULL;
 218         struct nfs_server *server;
 219
 220         /* Use machine credentials if available */
 221         cred = nfs4_get_machine_cred_locked(clp);
 222         if (cred != NULL)
 223                 goto out;
 224
 225         rcu_read_lock();
 226         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 227                 cred = nfs4_get_renew_cred_server_locked(server);
 228                 if (cred != NULL)
 229                         break;
 230         }
 231         rcu_read_unlock();
 232
 233 out:
 234         return cred;
 235 }
 236
 237 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 238 {
 239         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 240                 spin_lock(&tbl->slot_tbl_lock);
 241                 nfs41_wake_slot_table(tbl);
 242                 spin_unlock(&tbl->slot_tbl_lock);
 243         }
 244 }
 245
 246 static void nfs4_end_drain_session(struct nfs_client *clp)
 247 {
 248         struct nfs4_session *ses = clp->cl_session;
 249
 250         if (clp->cl_slot_tbl) {
 251                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 252                 return;
 253         }
 254
 255         if (ses != NULL) {
 256                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 257                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 258         }
 259 }
 260
 261 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 262 {
 263         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 264         spin_lock(&tbl->slot_tbl_lock);
 265         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 266                 reinit_completion(&tbl->complete);
 267                 spin_unlock(&tbl->slot_tbl_lock);
 268                 return wait_for_completion_interruptible(&tbl->complete);
 269         }
 270         spin_unlock(&tbl->slot_tbl_lock);
 271         return 0;
 272 }
 273
 274 static int nfs4_begin_drain_session(struct nfs_client *clp)
 275 {
 276         struct nfs4_session *ses = clp->cl_session;
 277         int ret;
 278
 279         if (clp->cl_slot_tbl)
 280                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 281
 282         /* back channel */
 283         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 284         if (ret)
 285                 return ret;
 286         /* fore channel */
 287         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 288 }
 289
 290 #if defined(CONFIG_NFS_V4_1)
 291
 292 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 293 {
 294         int status;
 295         struct nfs_fsinfo fsinfo;
 296         unsigned long now;
 297
 298         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 299                 nfs4_schedule_state_renewal(clp);
 300                 return 0;
 301         }
 302
 303         now = jiffies;
 304         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 305         if (status == 0) {
 306                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
 307                 nfs4_schedule_state_renewal(clp);
 308         }
 309
 310         return status;
 311 }
 312
 313 static void nfs41_finish_session_reset(struct nfs_client *clp)
 314 {
 315         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 316         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 317         /* create_session negotiated new slot table */
 318         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 319         nfs41_setup_state_renewal(clp);
 320 }
 321
 322 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 323 {
 324         int status;
 325
 326         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 327                 goto do_confirm;
 328         status = nfs4_proc_exchange_id(clp, cred);
 329         if (status != 0)
 330                 goto out;
 331         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 332 do_confirm:
 333         status = nfs4_proc_create_session(clp, cred);
 334         if (status != 0)
 335                 goto out;
 336         nfs41_finish_session_reset(clp);
 337         nfs_mark_client_ready(clp, NFS_CS_READY);
 338 out:
 339         return status;
 340 }
 341
 342 /**
 343  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 344  *
 345  * @clp: nfs_client under test
 346  * @result: OUT: found nfs_client, or clp
 347  * @cred: credential to use for trunking test
 348  *
 349  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 350  * If NFS4_OK is returned, an nfs_client pointer is planted in
 351  * "result".
 352  *
 353  * Note: The returned client may not yet be marked ready.
 354  */
 355 int nfs41_discover_server_trunking(struct nfs_client *clp,
 356                                    struct nfs_client **result,
 357                                    struct rpc_cred *cred)
 358 {
 359         int status;
 360
 361         status = nfs4_proc_exchange_id(clp, cred);
 362         if (status != NFS4_OK)
 363                 return status;
 364
 365         status = nfs41_walk_client_list(clp, result, cred);
 366         if (status < 0)
 367                 return status;
 368         if (clp != *result)
 369                 return 0;
 370
 371         /*
 372          * Purge state if the client id was established in a prior
 373          * instance and the client id could not have arrived on the
 374          * server via Transparent State Migration.
 375          */
 376         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
 377                 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
 378                         set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 379                 else
 380                         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 381         }
 382         nfs4_schedule_state_manager(clp);
 383         status = nfs_wait_client_init_complete(clp);
 384         if (status < 0)
 385                 nfs_put_client(clp);
 386         return status;
 387 }
 388
 389 #endif /* CONFIG_NFS_V4_1 */
 390
 391 /**
 392  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 393  * @clp: client state handle
 394  *
 395  * Returns an rpc_cred with reference count bumped, or NULL.
 396  */
 397 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
 398 {
 399         struct rpc_cred *cred;
 400
 401         spin_lock(&clp->cl_lock);
 402         cred = nfs4_get_machine_cred_locked(clp);
 403         spin_unlock(&clp->cl_lock);
 404         return cred;
 405 }
 406
 407 static struct nfs4_state_owner *
 408 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 409 {
 410         struct rb_node **p = &server->state_owners.rb_node,
 411                        *parent = NULL;
 412         struct nfs4_state_owner *sp;
 413
 414         while (*p != NULL) {
 415                 parent = *p;
 416                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 417
 418                 if (cred < sp->so_cred)
 419                         p = &parent->rb_left;
 420                 else if (cred > sp->so_cred)
 421                         p = &parent->rb_right;
 422                 else {
 423                         if (!list_empty(&sp->so_lru))
 424                                 list_del_init(&sp->so_lru);
 425                         atomic_inc(&sp->so_count);
 426                         return sp;
 427                 }
 428         }
 429         return NULL;
 430 }
 431
 432 static struct nfs4_state_owner *
 433 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 434 {
 435         struct nfs_server *server = new->so_server;
 436         struct rb_node **p = &server->state_owners.rb_node,
 437                        *parent = NULL;
 438         struct nfs4_state_owner *sp;
 439
 440         while (*p != NULL) {
 441                 parent = *p;
 442                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 443
 444                 if (new->so_cred < sp->so_cred)
 445                         p = &parent->rb_left;
 446                 else if (new->so_cred > sp->so_cred)
 447                         p = &parent->rb_right;
 448                 else {
 449                         if (!list_empty(&sp->so_lru))
 450                                 list_del_init(&sp->so_lru);
 451                         atomic_inc(&sp->so_count);
 452                         return sp;
 453                 }
 454         }
 455         rb_link_node(&new->so_server_node, parent, p);
 456         rb_insert_color(&new->so_server_node, &server->state_owners);
 457         return new;
 458 }
 459
 460 static void
 461 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 462 {
 463         struct nfs_server *server = sp->so_server;
 464
 465         if (!RB_EMPTY_NODE(&sp->so_server_node))
 466                 rb_erase(&sp->so_server_node, &server->state_owners);
 467 }
 468
 469 static void
 470 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 471 {
 472         sc->create_time = ktime_get();
 473         sc->flags = 0;
 474         sc->counter = 0;
 475         spin_lock_init(&sc->lock);
 476         INIT_LIST_HEAD(&sc->list);
 477         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 478 }
 479
 480 static void
 481 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 482 {
 483         rpc_destroy_wait_queue(&sc->wait);
 484 }
 485
 486 /*
 487  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 488  * create a new state_owner.
 489  *
 490  */
 491 static struct nfs4_state_owner *
 492 nfs4_alloc_state_owner(struct nfs_server *server,
 493                 struct rpc_cred *cred,
 494                 gfp_t gfp_flags)
 495 {
 496         struct nfs4_state_owner *sp;
 497
 498         sp = kzalloc(sizeof(*sp), gfp_flags);
 499         if (!sp)
 500                 return NULL;
 501         sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
 502                                                 gfp_flags);
 503         if (sp->so_seqid.owner_id < 0) {
 504                 kfree(sp);
 505                 return NULL;
 506         }
 507         sp->so_server = server;
 508         sp->so_cred = get_rpccred(cred);
 509         spin_lock_init(&sp->so_lock);
 510         INIT_LIST_HEAD(&sp->so_states);
 511         nfs4_init_seqid_counter(&sp->so_seqid);
 512         atomic_set(&sp->so_count, 1);
 513         INIT_LIST_HEAD(&sp->so_lru);
 514         seqcount_init(&sp->so_reclaim_seqcount);
 515         mutex_init(&sp->so_delegreturn_mutex);
 516         return sp;
 517 }
 518
 519 static void
 520 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
 521 {
 522         /* This state_owner is no longer usable, but must
 523          * remain in place so that state recovery can find it
 524          * and the opens associated with it.
 525          * It may also be used for new 'open' request to
 526          * return a delegation to the server.
 527          * So update the 'create_time' so that it looks like
 528          * a new state_owner.  This will cause the server to
 529          * request an OPEN_CONFIRM to start a new sequence.
 530          */
 531         sp->so_seqid.create_time = ktime_get();
 532 }
 533
 534 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 535 {
 536         nfs4_destroy_seqid_counter(&sp->so_seqid);
 537         put_rpccred(sp->so_cred);
 538         ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
 539         kfree(sp);
 540 }
 541
 542 static void nfs4_gc_state_owners(struct nfs_server *server)
 543 {
 544         struct nfs_client *clp = server->nfs_client;
 545         struct nfs4_state_owner *sp, *tmp;
 546         unsigned long time_min, time_max;
 547         LIST_HEAD(doomed);
 548
 549         spin_lock(&clp->cl_lock);
 550         time_max = jiffies;
 551         time_min = (long)time_max - (long)clp->cl_lease_time;
 552         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 553                 /* NB: LRU is sorted so that oldest is at the head */
 554                 if (time_in_range(sp->so_expires, time_min, time_max))
 555                         break;
 556                 list_move(&sp->so_lru, &doomed);
 557                 nfs4_remove_state_owner_locked(sp);
 558         }
 559         spin_unlock(&clp->cl_lock);
 560
 561         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 562                 list_del(&sp->so_lru);
 563                 nfs4_free_state_owner(sp);
 564         }
 565 }
 566
 567 /**
 568  * nfs4_get_state_owner - Look up a state owner given a credential
 569  * @server: nfs_server to search
 570  * @cred: RPC credential to match
 571  *
 572  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 573  */
 574 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 575                                               struct rpc_cred *cred,
 576                                               gfp_t gfp_flags)
 577 {
 578         struct nfs_client *clp = server->nfs_client;
 579         struct nfs4_state_owner *sp, *new;
 580
 581         spin_lock(&clp->cl_lock);
 582         sp = nfs4_find_state_owner_locked(server, cred);
 583         spin_unlock(&clp->cl_lock);
 584         if (sp != NULL)
 585                 goto out;
 586         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 587         if (new == NULL)
 588                 goto out;
 589         spin_lock(&clp->cl_lock);
 590         sp = nfs4_insert_state_owner_locked(new);
 591         spin_unlock(&clp->cl_lock);
 592         if (sp != new)
 593                 nfs4_free_state_owner(new);
 594 out:
 595         nfs4_gc_state_owners(server);
 596         return sp;
 597 }
 598
 599 /**
 600  * nfs4_put_state_owner - Release a nfs4_state_owner
 601  * @sp: state owner data to release
 602  *
 603  * Note that we keep released state owners on an LRU
 604  * list.
 605  * This caches valid state owners so that they can be
 606  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 607  * It also pins the uniquifier of dropped state owners for
 608  * a while, to ensure that those state owner names are
 609  * never reused.
 610  */
 611 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 612 {
 613         struct nfs_server *server = sp->so_server;
 614         struct nfs_client *clp = server->nfs_client;
 615
 616         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 617                 return;
 618
 619         sp->so_expires = jiffies;
 620         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 621         spin_unlock(&clp->cl_lock);
 622 }
 623
 624 /**
 625  * nfs4_purge_state_owners - Release all cached state owners
 626  * @server: nfs_server with cached state owners to release
 627  *
 628  * Called at umount time.  Remaining state owners will be on
 629  * the LRU with ref count of zero.
 630  */
 631 void nfs4_purge_state_owners(struct nfs_server *server)
 632 {
 633         struct nfs_client *clp = server->nfs_client;
 634         struct nfs4_state_owner *sp, *tmp;
 635         LIST_HEAD(doomed);
 636
 637         spin_lock(&clp->cl_lock);
 638         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 639                 list_move(&sp->so_lru, &doomed);
 640                 nfs4_remove_state_owner_locked(sp);
 641         }
 642         spin_unlock(&clp->cl_lock);
 643
 644         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 645                 list_del(&sp->so_lru);
 646                 nfs4_free_state_owner(sp);
 647         }
 648 }
 649
 650 static struct nfs4_state *
 651 nfs4_alloc_open_state(void)
 652 {
 653         struct nfs4_state *state;
 654
 655         state = kzalloc(sizeof(*state), GFP_NOFS);
 656         if (!state)
 657                 return NULL;
 658         refcount_set(&state->count, 1);
 659         INIT_LIST_HEAD(&state->lock_states);
 660         spin_lock_init(&state->state_lock);
 661         seqlock_init(&state->seqlock);
 662         init_waitqueue_head(&state->waitq);
 663         return state;
 664 }
 665
 666 void
 667 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 668 {
 669         if (state->state == fmode)
 670                 return;
 671         /* NB! List reordering - see the reclaim code for why.  */
 672         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 673                 if (fmode & FMODE_WRITE)
 674                         list_move(&state->open_states, &state->owner->so_states);
 675                 else
 676                         list_move_tail(&state->open_states, &state->owner->so_states);
 677         }
 678         state->state = fmode;
 679 }
 680
 681 static struct nfs4_state *
 682 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 683 {
 684         struct nfs_inode *nfsi = NFS_I(inode);
 685         struct nfs4_state *state;
 686
 687         list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
 688                 if (state->owner != owner)
 689                         continue;
 690                 if (!nfs4_valid_open_stateid(state))
 691                         continue;
 692                 if (refcount_inc_not_zero(&state->count))
 693                         return state;
 694         }
 695         return NULL;
 696 }
 697
 698 static void
 699 nfs4_free_open_state(struct nfs4_state *state)
 700 {
 701         kfree_rcu(state, rcu_head);
 702 }
 703
 704 struct nfs4_state *
 705 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 706 {
 707         struct nfs4_state *state, *new;
 708         struct nfs_inode *nfsi = NFS_I(inode);
 709
 710         rcu_read_lock();
 711         state = __nfs4_find_state_byowner(inode, owner);
 712         rcu_read_unlock();
 713         if (state)
 714                 goto out;
 715         new = nfs4_alloc_open_state();
 716         spin_lock(&owner->so_lock);
 717         spin_lock(&inode->i_lock);
 718         state = __nfs4_find_state_byowner(inode, owner);
 719         if (state == NULL && new != NULL) {
 720                 state = new;
 721                 state->owner = owner;
 722                 atomic_inc(&owner->so_count);
 723                 list_add_rcu(&state->inode_states, &nfsi->open_states);
 724                 ihold(inode);
 725                 state->inode = inode;
 726                 spin_unlock(&inode->i_lock);
 727                 /* Note: The reclaim code dictates that we add stateless
 728                  * and read-only stateids to the end of the list */
 729                 list_add_tail(&state->open_states, &owner->so_states);
 730                 spin_unlock(&owner->so_lock);
 731         } else {
 732                 spin_unlock(&inode->i_lock);
 733                 spin_unlock(&owner->so_lock);
 734                 if (new)
 735                         nfs4_free_open_state(new);
 736         }
 737 out:
 738         return state;
 739 }
 740
 741 void nfs4_put_open_state(struct nfs4_state *state)
 742 {
 743         struct inode *inode = state->inode;
 744         struct nfs4_state_owner *owner = state->owner;
 745
 746         if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
 747                 return;
 748         spin_lock(&inode->i_lock);
 749         list_del_rcu(&state->inode_states);
 750         list_del(&state->open_states);
 751         spin_unlock(&inode->i_lock);
 752         spin_unlock(&owner->so_lock);
 753         iput(inode);
 754         nfs4_free_open_state(state);
 755         nfs4_put_state_owner(owner);
 756 }
 757
 758 /*
 759  * Close the current file.
 760  */
 761 static void __nfs4_close(struct nfs4_state *state,
 762                 fmode_t fmode, gfp_t gfp_mask, int wait)
 763 {
 764         struct nfs4_state_owner *owner = state->owner;
 765         int call_close = 0;
 766         fmode_t newstate;
 767
 768         atomic_inc(&owner->so_count);
 769         /* Protect against nfs4_find_state() */
 770         spin_lock(&owner->so_lock);
 771         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 772                 case FMODE_READ:
 773                         state->n_rdonly--;
 774                         break;
 775                 case FMODE_WRITE:
 776                         state->n_wronly--;
 777                         break;
 778                 case FMODE_READ|FMODE_WRITE:
 779                         state->n_rdwr--;
 780         }
 781         newstate = FMODE_READ|FMODE_WRITE;
 782         if (state->n_rdwr == 0) {
 783                 if (state->n_rdonly == 0) {
 784                         newstate &= ~FMODE_READ;
 785                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 786                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 787                 }
 788                 if (state->n_wronly == 0) {
 789                         newstate &= ~FMODE_WRITE;
 790                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 791                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 792                 }
 793                 if (newstate == 0)
 794                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 795         }
 796         nfs4_state_set_mode_locked(state, newstate);
 797         spin_unlock(&owner->so_lock);
 798
 799         if (!call_close) {
 800                 nfs4_put_open_state(state);
 801                 nfs4_put_state_owner(owner);
 802         } else
 803                 nfs4_do_close(state, gfp_mask, wait);
 804 }
 805
 806 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 807 {
 808         __nfs4_close(state, fmode, GFP_NOFS, 0);
 809 }
 810
 811 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 812 {
 813         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 814 }
 815
 816 /*
 817  * Search the state->lock_states for an existing lock_owner
 818  * that is compatible with either of the given owners.
 819  * If the second is non-zero, then the first refers to a Posix-lock
 820  * owner (current->files) and the second refers to a flock/OFD
 821  * owner (struct file*).  In that case, prefer a match for the first
 822  * owner.
 823  * If both sorts of locks are held on the one file we cannot know
 824  * which stateid was intended to be used, so a "correct" choice cannot
 825  * be made.  Failing that, a "consistent" choice is preferable.  The
 826  * consistent choice we make is to prefer the first owner, that of a
 827  * Posix lock.
 828  */
 829 static struct nfs4_lock_state *
 830 __nfs4_find_lock_state(struct nfs4_state *state,
 831                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
 832 {
 833         struct nfs4_lock_state *pos, *ret = NULL;
 834         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 835                 if (pos->ls_owner == fl_owner) {
 836                         ret = pos;
 837                         break;
 838                 }
 839                 if (pos->ls_owner == fl_owner2)
 840                         ret = pos;
 841         }
 842         if (ret)
 843                 refcount_inc(&ret->ls_count);
 844         return ret;
 845 }
 846
 847 /*
 848  * Return a compatible lock_state. If no initialized lock_state structure
 849  * exists, return an uninitialized one.
 850  *
 851  */
 852 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 853 {
 854         struct nfs4_lock_state *lsp;
 855         struct nfs_server *server = state->owner->so_server;
 856
 857         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 858         if (lsp == NULL)
 859                 return NULL;
 860         nfs4_init_seqid_counter(&lsp->ls_seqid);
 861         refcount_set(&lsp->ls_count, 1);
 862         lsp->ls_state = state;
 863         lsp->ls_owner = fl_owner;
 864         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 865         if (lsp->ls_seqid.owner_id < 0)
 866                 goto out_free;
 867         INIT_LIST_HEAD(&lsp->ls_locks);
 868         return lsp;
 869 out_free:
 870         kfree(lsp);
 871         return NULL;
 872 }
 873
 874 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 875 {
 876         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 877         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 878         kfree(lsp);
 879 }
 880
 881 /*
 882  * Return a compatible lock_state. If no initialized lock_state structure
 883  * exists, return an uninitialized one.
 884  *
 885  */
 886 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 887 {
 888         struct nfs4_lock_state *lsp, *new = NULL;
 889
 890         for(;;) {
 891                 spin_lock(&state->state_lock);
 892                 lsp = __nfs4_find_lock_state(state, owner, NULL);
 893                 if (lsp != NULL)
 894                         break;
 895                 if (new != NULL) {
 896                         list_add(&new->ls_locks, &state->lock_states);
 897                         set_bit(LK_STATE_IN_USE, &state->flags);
 898                         lsp = new;
 899                         new = NULL;
 900                         break;
 901                 }
 902                 spin_unlock(&state->state_lock);
 903                 new = nfs4_alloc_lock_state(state, owner);
 904                 if (new == NULL)
 905                         return NULL;
 906         }
 907         spin_unlock(&state->state_lock);
 908         if (new != NULL)
 909                 nfs4_free_lock_state(state->owner->so_server, new);
 910         return lsp;
 911 }
 912
 913 /*
 914  * Release reference to lock_state, and free it if we see that
 915  * it is no longer in use
 916  */
 917 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 918 {
 919         struct nfs_server *server;
 920         struct nfs4_state *state;
 921
 922         if (lsp == NULL)
 923                 return;
 924         state = lsp->ls_state;
 925         if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
 926                 return;
 927         list_del(&lsp->ls_locks);
 928         if (list_empty(&state->lock_states))
 929                 clear_bit(LK_STATE_IN_USE, &state->flags);
 930         spin_unlock(&state->state_lock);
 931         server = state->owner->so_server;
 932         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 933                 struct nfs_client *clp = server->nfs_client;
 934
 935                 clp->cl_mvops->free_lock_state(server, lsp);
 936         } else
 937                 nfs4_free_lock_state(server, lsp);
 938 }
 939
 940 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 941 {
 942         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 943
 944         dst->fl_u.nfs4_fl.owner = lsp;
 945         refcount_inc(&lsp->ls_count);
 946 }
 947
 948 static void nfs4_fl_release_lock(struct file_lock *fl)
 949 {
 950         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 951 }
 952
 953 static const struct file_lock_operations nfs4_fl_lock_ops = {
 954         .fl_copy_lock = nfs4_fl_copy_lock,
 955         .fl_release_private = nfs4_fl_release_lock,
 956 };
 957
 958 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 959 {
 960         struct nfs4_lock_state *lsp;
 961
 962         if (fl->fl_ops != NULL)
 963                 return 0;
 964         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 965         if (lsp == NULL)
 966                 return -ENOMEM;
 967         fl->fl_u.nfs4_fl.owner = lsp;
 968         fl->fl_ops = &nfs4_fl_lock_ops;
 969         return 0;
 970 }
 971
 972 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 973                 struct nfs4_state *state,
 974                 const struct nfs_lock_context *l_ctx)
 975 {
 976         struct nfs4_lock_state *lsp;
 977         fl_owner_t fl_owner, fl_flock_owner;
 978         int ret = -ENOENT;
 979
 980         if (l_ctx == NULL)
 981                 goto out;
 982
 983         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 984                 goto out;
 985
 986         fl_owner = l_ctx->lockowner;
 987         fl_flock_owner = l_ctx->open_context->flock_owner;
 988
 989         spin_lock(&state->state_lock);
 990         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
 991         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 992                 ret = -EIO;
 993         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
 994                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 995                 ret = 0;
 996         }
 997         spin_unlock(&state->state_lock);
 998         nfs4_put_lock_state(lsp);
 999 out:
1000         return ret;
1001 }
1002
1003 bool nfs4_refresh_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1004 {
1005         bool ret;
1006         int seq;
1007
1008         do {
1009                 ret = false;
1010                 seq = read_seqbegin(&state->seqlock);
1011                 if (nfs4_state_match_open_stateid_other(state, dst)) {
1012                         dst->seqid = state->open_stateid.seqid;
1013                         ret = true;
1014                 }
1015         } while (read_seqretry(&state->seqlock, seq));
1016         return ret;
1017 }
1018
1019 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1020 {
1021         bool ret;
1022         const nfs4_stateid *src;
1023         int seq;
1024
1025         do {
1026                 ret = false;
1027                 src = &zero_stateid;
1028                 seq = read_seqbegin(&state->seqlock);
1029                 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1030                         src = &state->open_stateid;
1031                         ret = true;
1032                 }
1033                 nfs4_stateid_copy(dst, src);
1034         } while (read_seqretry(&state->seqlock, seq));
1035         return ret;
1036 }
1037
1038 /*
1039  * Byte-range lock aware utility to initialize the stateid of read/write
1040  * requests.
1041  */
1042 int nfs4_select_rw_stateid(struct nfs4_state *state,
1043                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1044                 nfs4_stateid *dst, struct rpc_cred **cred)
1045 {
1046         int ret;
1047
1048         if (!nfs4_valid_open_stateid(state))
1049                 return -EIO;
1050         if (cred != NULL)
1051                 *cred = NULL;
1052         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1053         if (ret == -EIO)
1054                 /* A lost lock - don't even consider delegations */
1055                 goto out;
1056         /* returns true if delegation stateid found and copied */
1057         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1058                 ret = 0;
1059                 goto out;
1060         }
1061         if (ret != -ENOENT)
1062                 /* nfs4_copy_delegation_stateid() didn't over-write
1063                  * dst, so it still has the lock stateid which we now
1064                  * choose to use.
1065                  */
1066                 goto out;
1067         nfs4_copy_open_stateid(dst, state);
1068         ret = 0;
1069 out:
1070         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1071                 dst->seqid = 0;
1072         return ret;
1073 }
1074
1075 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1076 {
1077         struct nfs_seqid *new;
1078
1079         new = kmalloc(sizeof(*new), gfp_mask);
1080         if (new == NULL)
1081                 return ERR_PTR(-ENOMEM);
1082         new->sequence = counter;
1083         INIT_LIST_HEAD(&new->list);
1084         new->task = NULL;
1085         return new;
1086 }
1087
1088 void nfs_release_seqid(struct nfs_seqid *seqid)
1089 {
1090         struct nfs_seqid_counter *sequence;
1091
1092         if (seqid == NULL || list_empty(&seqid->list))
1093                 return;
1094         sequence = seqid->sequence;
1095         spin_lock(&sequence->lock);
1096         list_del_init(&seqid->list);
1097         if (!list_empty(&sequence->list)) {
1098                 struct nfs_seqid *next;
1099
1100                 next = list_first_entry(&sequence->list,
1101                                 struct nfs_seqid, list);
1102                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1103         }
1104         spin_unlock(&sequence->lock);
1105 }
1106
1107 void nfs_free_seqid(struct nfs_seqid *seqid)
1108 {
1109         nfs_release_seqid(seqid);
1110         kfree(seqid);
1111 }
1112
1113 /*
1114  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1115  * failed with a seqid incrementing error -
1116  * see comments nfs4.h:seqid_mutating_error()
1117  */
1118 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1119 {
1120         switch (status) {
1121                 case 0:
1122                         break;
1123                 case -NFS4ERR_BAD_SEQID:
1124                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1125                                 return;
1126                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1127                                         " sequence-id error on an"
1128                                         " unconfirmed sequence %p!\n",
1129                                         seqid->sequence);
1130                 case -NFS4ERR_STALE_CLIENTID:
1131                 case -NFS4ERR_STALE_STATEID:
1132                 case -NFS4ERR_BAD_STATEID:
1133                 case -NFS4ERR_BADXDR:
1134                 case -NFS4ERR_RESOURCE:
1135                 case -NFS4ERR_NOFILEHANDLE:
1136                 case -NFS4ERR_MOVED:
1137                         /* Non-seqid mutating errors */
1138                         return;
1139         };
1140         /*
1141          * Note: no locking needed as we are guaranteed to be first
1142          * on the sequence list
1143          */
1144         seqid->sequence->counter++;
1145 }
1146
1147 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1148 {
1149         struct nfs4_state_owner *sp;
1150
1151         if (seqid == NULL)
1152                 return;
1153
1154         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1155         if (status == -NFS4ERR_BAD_SEQID)
1156                 nfs4_reset_state_owner(sp);
1157         if (!nfs4_has_session(sp->so_server->nfs_client))
1158                 nfs_increment_seqid(status, seqid);
1159 }
1160
1161 /*
1162  * Increment the seqid if the LOCK/LOCKU succeeded, or
1163  * failed with a seqid incrementing error -
1164  * see comments nfs4.h:seqid_mutating_error()
1165  */
1166 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1167 {
1168         if (seqid != NULL)
1169                 nfs_increment_seqid(status, seqid);
1170 }
1171
1172 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1173 {
1174         struct nfs_seqid_counter *sequence;
1175         int status = 0;
1176
1177         if (seqid == NULL)
1178                 goto out;
1179         sequence = seqid->sequence;
1180         spin_lock(&sequence->lock);
1181         seqid->task = task;
1182         if (list_empty(&seqid->list))
1183                 list_add_tail(&seqid->list, &sequence->list);
1184         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1185                 goto unlock;
1186         rpc_sleep_on(&sequence->wait, task, NULL);
1187         status = -EAGAIN;
1188 unlock:
1189         spin_unlock(&sequence->lock);
1190 out:
1191         return status;
1192 }
1193
1194 static int nfs4_run_state_manager(void *);
1195
1196 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1197 {
1198         smp_mb__before_atomic();
1199         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1200         smp_mb__after_atomic();
1201         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1202         rpc_wake_up(&clp->cl_rpcwaitq);
1203 }
1204
1205 /*
1206  * Schedule the nfs_client asynchronous state management routine
1207  */
1208 void nfs4_schedule_state_manager(struct nfs_client *clp)
1209 {
1210         struct task_struct *task;
1211         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1212
1213         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1214                 return;
1215         __module_get(THIS_MODULE);
1216         refcount_inc(&clp->cl_count);
1217
1218         /* The rcu_read_lock() is not strictly necessary, as the state
1219          * manager is the only thread that ever changes the rpc_xprt
1220          * after it's initialized.  At this point, we're single threaded. */
1221         rcu_read_lock();
1222         snprintf(buf, sizeof(buf), "%s-manager",
1223                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1224         rcu_read_unlock();
1225         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1226         if (IS_ERR(task)) {
1227                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1228                         __func__, PTR_ERR(task));
1229                 nfs4_clear_state_manager_bit(clp);
1230                 nfs_put_client(clp);
1231                 module_put(THIS_MODULE);
1232         }
1233 }
1234
1235 /*
1236  * Schedule a lease recovery attempt
1237  */
1238 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1239 {
1240         if (!clp)
1241                 return;
1242         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1243                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1244         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1245                         clp->cl_hostname);
1246         nfs4_schedule_state_manager(clp);
1247 }
1248 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1249
1250 /**
1251  * nfs4_schedule_migration_recovery - trigger migration recovery
1252  *
1253  * @server: FSID that is migrating
1254  *
1255  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1256  * value is returned.
1257  */
1258 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1259 {
1260         struct nfs_client *clp = server->nfs_client;
1261
1262         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1263                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1264                                 clp->cl_hostname);
1265                 return -NFS4ERR_IO;
1266         }
1267
1268         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1269                 return -NFS4ERR_IO;
1270
1271         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1272                         __func__,
1273                         (unsigned long long)server->fsid.major,
1274                         (unsigned long long)server->fsid.minor,
1275                         clp->cl_hostname);
1276
1277         set_bit(NFS_MIG_IN_TRANSITION,
1278                         &((struct nfs_server *)server)->mig_status);
1279         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1280
1281         nfs4_schedule_state_manager(clp);
1282         return 0;
1283 }
1284 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1285
1286 /**
1287  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1288  *
1289  * @clp: server to check for moved leases
1290  *
1291  */
1292 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1293 {
1294         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1295                 __func__, clp->cl_clientid, clp->cl_hostname);
1296
1297         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1298         nfs4_schedule_state_manager(clp);
1299 }
1300 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1301
1302 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1303 {
1304         int res;
1305
1306         might_sleep();
1307
1308         refcount_inc(&clp->cl_count);
1309         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1310                                  nfs_wait_bit_killable, TASK_KILLABLE);
1311         if (res)
1312                 goto out;
1313         if (clp->cl_cons_state < 0)
1314                 res = clp->cl_cons_state;
1315 out:
1316         nfs_put_client(clp);
1317         return res;
1318 }
1319
1320 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1321 {
1322         unsigned int loop;
1323         int ret;
1324
1325         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1326                 ret = nfs4_wait_clnt_recover(clp);
1327                 if (ret != 0)
1328                         break;
1329                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1330                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1331                         break;
1332                 nfs4_schedule_state_manager(clp);
1333                 ret = -EIO;
1334         }
1335         return ret;
1336 }
1337
1338 /*
1339  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1340  * @clp: client to process
1341  *
1342  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1343  * resend of the SETCLIENTID and hence re-establish the
1344  * callback channel. Then return all existing delegations.
1345  */
1346 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1347 {
1348         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1349         nfs_expire_all_delegations(clp);
1350         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1351                         clp->cl_hostname);
1352 }
1353
1354 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1355 {
1356         nfs40_handle_cb_pathdown(clp);
1357         nfs4_schedule_state_manager(clp);
1358 }
1359
1360 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1361 {
1362
1363         if (!nfs4_valid_open_stateid(state))
1364                 return 0;
1365         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1366         /* Don't recover state that expired before the reboot */
1367         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1368                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1369                 return 0;
1370         }
1371         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1372         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1373         return 1;
1374 }
1375
1376 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1377 {
1378         if (!nfs4_valid_open_stateid(state))
1379                 return 0;
1380         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1381         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1382         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1383         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1384         return 1;
1385 }
1386
1387 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1388 {
1389         struct nfs_client *clp = server->nfs_client;
1390
1391         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1392                 return -EBADF;
1393         nfs_inode_find_delegation_state_and_recover(state->inode,
1394                         &state->stateid);
1395         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1396                         clp->cl_hostname);
1397         nfs4_schedule_state_manager(clp);
1398         return 0;
1399 }
1400 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1401
1402 static struct nfs4_lock_state *
1403 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1404                 const nfs4_stateid *stateid)
1405 {
1406         struct nfs4_lock_state *pos;
1407
1408         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1409                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1410                         continue;
1411                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1412                         return pos;
1413         }
1414         return NULL;
1415 }
1416
1417 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1418                 const nfs4_stateid *stateid)
1419 {
1420         bool found = false;
1421
1422         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1423                 spin_lock(&state->state_lock);
1424                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1425                         found = true;
1426                 spin_unlock(&state->state_lock);
1427         }
1428         return found;
1429 }
1430
1431 void nfs_inode_find_state_and_recover(struct inode *inode,
1432                 const nfs4_stateid *stateid)
1433 {
1434         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1435         struct nfs_inode *nfsi = NFS_I(inode);
1436         struct nfs_open_context *ctx;
1437         struct nfs4_state *state;
1438         bool found = false;
1439
1440         rcu_read_lock();
1441         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1442                 state = ctx->state;
1443                 if (state == NULL)
1444                         continue;
1445                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1446                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1447                         found = true;
1448                         continue;
1449                 }
1450                 if (nfs4_stateid_match_other(&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
1466 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
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         }
1478         rcu_read_unlock();
1479 }
1480
1481 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1482 {
1483         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1484         nfs4_state_mark_open_context_bad(state);
1485 }
1486
1487
1488 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1489 {
1490         struct inode *inode = state->inode;
1491         struct nfs_inode *nfsi = NFS_I(inode);
1492         struct file_lock *fl;
1493         struct nfs4_lock_state *lsp;
1494         int status = 0;
1495         struct file_lock_context *flctx = inode->i_flctx;
1496         struct list_head *list;
1497
1498         if (flctx == NULL)
1499                 return 0;
1500
1501         list = &flctx->flc_posix;
1502
1503         /* Guard against delegation returns and new lock/unlock calls */
1504         down_write(&nfsi->rwsem);
1505         spin_lock(&flctx->flc_lock);
1506 restart:
1507         list_for_each_entry(fl, list, fl_list) {
1508                 if (nfs_file_open_context(fl->fl_file)->state != state)
1509                         continue;
1510                 spin_unlock(&flctx->flc_lock);
1511                 status = ops->recover_lock(state, fl);
1512                 switch (status) {
1513                 case 0:
1514                         break;
1515                 case -ESTALE:
1516                 case -NFS4ERR_ADMIN_REVOKED:
1517                 case -NFS4ERR_STALE_STATEID:
1518                 case -NFS4ERR_BAD_STATEID:
1519                 case -NFS4ERR_EXPIRED:
1520                 case -NFS4ERR_NO_GRACE:
1521                 case -NFS4ERR_STALE_CLIENTID:
1522                 case -NFS4ERR_BADSESSION:
1523                 case -NFS4ERR_BADSLOT:
1524                 case -NFS4ERR_BAD_HIGH_SLOT:
1525                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1526                         goto out;
1527                 default:
1528                         pr_err("NFS: %s: unhandled error %d\n",
1529                                         __func__, status);
1530                         /* Fall through */
1531                 case -ENOMEM:
1532                 case -NFS4ERR_DENIED:
1533                 case -NFS4ERR_RECLAIM_BAD:
1534                 case -NFS4ERR_RECLAIM_CONFLICT:
1535                         lsp = fl->fl_u.nfs4_fl.owner;
1536                         if (lsp)
1537                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1538                         status = 0;
1539                 }
1540                 spin_lock(&flctx->flc_lock);
1541         }
1542         if (list == &flctx->flc_posix) {
1543                 list = &flctx->flc_flock;
1544                 goto restart;
1545         }
1546         spin_unlock(&flctx->flc_lock);
1547 out:
1548         up_write(&nfsi->rwsem);
1549         return status;
1550 }
1551
1552 #ifdef CONFIG_NFS_V4_2
1553 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1554 {
1555         struct nfs4_copy_state *copy;
1556
1557         if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags))
1558                 return;
1559
1560         spin_lock(&sp->so_server->nfs_client->cl_lock);
1561         list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1562                 if (nfs4_stateid_match_other(&state->stateid, &copy->parent_state->stateid))
1563                         continue;
1564                 copy->flags = 1;
1565                 complete(&copy->completion);
1566                 break;
1567         }
1568         spin_unlock(&sp->so_server->nfs_client->cl_lock);
1569 }
1570 #else /* !CONFIG_NFS_V4_2 */
1571 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1572                                          struct nfs4_state *state)
1573 {
1574 }
1575 #endif /* CONFIG_NFS_V4_2 */
1576
1577 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1578                                      const struct nfs4_state_recovery_ops *ops)
1579 {
1580         struct nfs4_lock_state *lock;
1581         int status;
1582
1583         status = ops->recover_open(sp, state);
1584         if (status < 0)
1585                 return status;
1586
1587         status = nfs4_reclaim_locks(state, ops);
1588         if (status < 0)
1589                 return status;
1590
1591         if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1592                 spin_lock(&state->state_lock);
1593                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1594                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1595                                 pr_warn_ratelimited("NFS: %s: Lock reclaim failed!\n", __func__);
1596                 }
1597                 spin_unlock(&state->state_lock);
1598         }
1599
1600         nfs42_complete_copies(sp, state);
1601         clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1602         return status;
1603 }
1604
1605 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1606 {
1607         struct nfs4_state *state;
1608         int status = 0;
1609
1610         /* Note: we rely on the sp->so_states list being ordered
1611          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1612          * states first.
1613          * This is needed to ensure that the server won't give us any
1614          * read delegations that we have to return if, say, we are
1615          * recovering after a network partition or a reboot from a
1616          * server that doesn't support a grace period.
1617          */
1618         spin_lock(&sp->so_lock);
1619         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1620 restart:
1621         list_for_each_entry(state, &sp->so_states, open_states) {
1622                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1623                         continue;
1624                 if (!nfs4_valid_open_stateid(state))
1625                         continue;
1626                 if (state->state == 0)
1627                         continue;
1628                 refcount_inc(&state->count);
1629                 spin_unlock(&sp->so_lock);
1630                 status = __nfs4_reclaim_open_state(sp, state, ops);
1631
1632                 switch (status) {
1633                 default:
1634                         if (status >= 0)
1635                                 break;
1636                         printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1637                         /* Fall through */
1638                 case -ENOENT:
1639                 case -ENOMEM:
1640                 case -EACCES:
1641                 case -EROFS:
1642                 case -EIO:
1643                 case -ESTALE:
1644                         /* Open state on this file cannot be recovered */
1645                         nfs4_state_mark_recovery_failed(state, status);
1646                         break;
1647                 case -EAGAIN:
1648                         ssleep(1);
1649                         /* Fall through */
1650                 case -NFS4ERR_ADMIN_REVOKED:
1651                 case -NFS4ERR_STALE_STATEID:
1652                 case -NFS4ERR_OLD_STATEID:
1653                 case -NFS4ERR_BAD_STATEID:
1654                 case -NFS4ERR_RECLAIM_BAD:
1655                 case -NFS4ERR_RECLAIM_CONFLICT:
1656                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1657                         break;
1658                 case -NFS4ERR_EXPIRED:
1659                 case -NFS4ERR_NO_GRACE:
1660                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1661                 case -NFS4ERR_STALE_CLIENTID:
1662                 case -NFS4ERR_BADSESSION:
1663                 case -NFS4ERR_BADSLOT:
1664                 case -NFS4ERR_BAD_HIGH_SLOT:
1665                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1666                         goto out_err;
1667                 }
1668                 nfs4_put_open_state(state);
1669                 spin_lock(&sp->so_lock);
1670                 goto restart;
1671         }
1672         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1673         spin_unlock(&sp->so_lock);
1674         return 0;
1675 out_err:
1676         nfs4_put_open_state(state);
1677         spin_lock(&sp->so_lock);
1678         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1679         spin_unlock(&sp->so_lock);
1680         return status;
1681 }
1682
1683 static void nfs4_clear_open_state(struct nfs4_state *state)
1684 {
1685         struct nfs4_lock_state *lock;
1686
1687         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1688         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1689         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1690         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1691         spin_lock(&state->state_lock);
1692         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1693                 lock->ls_seqid.flags = 0;
1694                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1695         }
1696         spin_unlock(&state->state_lock);
1697 }
1698
1699 static void nfs4_reset_seqids(struct nfs_server *server,
1700         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1701 {
1702         struct nfs_client *clp = server->nfs_client;
1703         struct nfs4_state_owner *sp;
1704         struct rb_node *pos;
1705         struct nfs4_state *state;
1706
1707         spin_lock(&clp->cl_lock);
1708         for (pos = rb_first(&server->state_owners);
1709              pos != NULL;
1710              pos = rb_next(pos)) {
1711                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1712                 sp->so_seqid.flags = 0;
1713                 spin_lock(&sp->so_lock);
1714                 list_for_each_entry(state, &sp->so_states, open_states) {
1715                         if (mark_reclaim(clp, state))
1716                                 nfs4_clear_open_state(state);
1717                 }
1718                 spin_unlock(&sp->so_lock);
1719         }
1720         spin_unlock(&clp->cl_lock);
1721 }
1722
1723 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1724         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1725 {
1726         struct nfs_server *server;
1727
1728         rcu_read_lock();
1729         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1730                 nfs4_reset_seqids(server, mark_reclaim);
1731         rcu_read_unlock();
1732 }
1733
1734 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1735 {
1736         /* Mark all delegations for reclaim */
1737         nfs_delegation_mark_reclaim(clp);
1738         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1739 }
1740
1741 static int nfs4_reclaim_complete(struct nfs_client *clp,
1742                                  const struct nfs4_state_recovery_ops *ops,
1743                                  struct rpc_cred *cred)
1744 {
1745         /* Notify the server we're done reclaiming our state */
1746         if (ops->reclaim_complete)
1747                 return ops->reclaim_complete(clp, cred);
1748         return 0;
1749 }
1750
1751 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1752 {
1753         struct nfs_client *clp = server->nfs_client;
1754         struct nfs4_state_owner *sp;
1755         struct rb_node *pos;
1756         struct nfs4_state *state;
1757
1758         spin_lock(&clp->cl_lock);
1759         for (pos = rb_first(&server->state_owners);
1760              pos != NULL;
1761              pos = rb_next(pos)) {
1762                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1763                 spin_lock(&sp->so_lock);
1764                 list_for_each_entry(state, &sp->so_states, open_states) {
1765                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1766                                                 &state->flags))
1767                                 continue;
1768                         nfs4_state_mark_reclaim_nograce(clp, state);
1769                 }
1770                 spin_unlock(&sp->so_lock);
1771         }
1772         spin_unlock(&clp->cl_lock);
1773 }
1774
1775 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1776 {
1777         struct nfs_server *server;
1778
1779         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1780                 return 0;
1781
1782         rcu_read_lock();
1783         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1784                 nfs4_clear_reclaim_server(server);
1785         rcu_read_unlock();
1786
1787         nfs_delegation_reap_unclaimed(clp);
1788         return 1;
1789 }
1790
1791 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1792 {
1793         const struct nfs4_state_recovery_ops *ops;
1794         struct rpc_cred *cred;
1795         int err;
1796
1797         if (!nfs4_state_clear_reclaim_reboot(clp))
1798                 return;
1799         ops = clp->cl_mvops->reboot_recovery_ops;
1800         cred = nfs4_get_clid_cred(clp);
1801         err = nfs4_reclaim_complete(clp, ops, cred);
1802         put_rpccred(cred);
1803         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1804                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1805 }
1806
1807 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1808 {
1809         nfs_mark_test_expired_all_delegations(clp);
1810         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1811 }
1812
1813 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1814 {
1815         switch (error) {
1816         case 0:
1817                 break;
1818         case -NFS4ERR_CB_PATH_DOWN:
1819                 nfs40_handle_cb_pathdown(clp);
1820                 break;
1821         case -NFS4ERR_NO_GRACE:
1822                 nfs4_state_end_reclaim_reboot(clp);
1823                 break;
1824         case -NFS4ERR_STALE_CLIENTID:
1825                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1826                 nfs4_state_start_reclaim_reboot(clp);
1827                 break;
1828         case -NFS4ERR_EXPIRED:
1829                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1830                 nfs4_state_start_reclaim_nograce(clp);
1831                 break;
1832         case -NFS4ERR_BADSESSION:
1833         case -NFS4ERR_BADSLOT:
1834         case -NFS4ERR_BAD_HIGH_SLOT:
1835         case -NFS4ERR_DEADSESSION:
1836         case -NFS4ERR_SEQ_FALSE_RETRY:
1837         case -NFS4ERR_SEQ_MISORDERED:
1838                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1839                 /* Zero session reset errors */
1840                 break;
1841         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1842                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1843                 break;
1844         default:
1845                 dprintk("%s: failed to handle error %d for server %s\n",
1846                                 __func__, error, clp->cl_hostname);
1847                 return error;
1848         }
1849         dprintk("%s: handled error %d for server %s\n", __func__, error,
1850                         clp->cl_hostname);
1851         return 0;
1852 }
1853
1854 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1855 {
1856         struct nfs4_state_owner *sp;
1857         struct nfs_server *server;
1858         struct rb_node *pos;
1859         int status = 0;
1860
1861 restart:
1862         rcu_read_lock();
1863         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1864                 nfs4_purge_state_owners(server);
1865                 spin_lock(&clp->cl_lock);
1866                 for (pos = rb_first(&server->state_owners);
1867                      pos != NULL;
1868                      pos = rb_next(pos)) {
1869                         sp = rb_entry(pos,
1870                                 struct nfs4_state_owner, so_server_node);
1871                         if (!test_and_clear_bit(ops->owner_flag_bit,
1872                                                         &sp->so_flags))
1873                                 continue;
1874                         if (!atomic_inc_not_zero(&sp->so_count))
1875                                 continue;
1876                         spin_unlock(&clp->cl_lock);
1877                         rcu_read_unlock();
1878
1879                         status = nfs4_reclaim_open_state(sp, ops);
1880                         if (status < 0) {
1881                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1882                                 nfs4_put_state_owner(sp);
1883                                 status = nfs4_recovery_handle_error(clp, status);
1884                                 return (status != 0) ? status : -EAGAIN;
1885                         }
1886
1887                         nfs4_put_state_owner(sp);
1888                         goto restart;
1889                 }
1890                 spin_unlock(&clp->cl_lock);
1891         }
1892         rcu_read_unlock();
1893         return 0;
1894 }
1895
1896 static int nfs4_check_lease(struct nfs_client *clp)
1897 {
1898         struct rpc_cred *cred;
1899         const struct nfs4_state_maintenance_ops *ops =
1900                 clp->cl_mvops->state_renewal_ops;
1901         int status;
1902
1903         /* Is the client already known to have an expired lease? */
1904         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1905                 return 0;
1906         spin_lock(&clp->cl_lock);
1907         cred = ops->get_state_renewal_cred_locked(clp);
1908         spin_unlock(&clp->cl_lock);
1909         if (cred == NULL) {
1910                 cred = nfs4_get_clid_cred(clp);
1911                 status = -ENOKEY;
1912                 if (cred == NULL)
1913                         goto out;
1914         }
1915         status = ops->renew_lease(clp, cred);
1916         put_rpccred(cred);
1917         if (status == -ETIMEDOUT) {
1918                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1919                 return 0;
1920         }
1921 out:
1922         return nfs4_recovery_handle_error(clp, status);
1923 }
1924
1925 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1926  * and for recoverable errors on EXCHANGE_ID for v4.1
1927  */
1928 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1929 {
1930         switch (status) {
1931         case -NFS4ERR_SEQ_MISORDERED:
1932                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1933                         return -ESERVERFAULT;
1934                 /* Lease confirmation error: retry after purging the lease */
1935                 ssleep(1);
1936                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1937                 break;
1938         case -NFS4ERR_STALE_CLIENTID:
1939                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1940                 nfs4_state_start_reclaim_reboot(clp);
1941                 break;
1942         case -NFS4ERR_CLID_INUSE:
1943                 pr_err("NFS: Server %s reports our clientid is in use\n",
1944                         clp->cl_hostname);
1945                 nfs_mark_client_ready(clp, -EPERM);
1946                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1947                 return -EPERM;
1948         case -EACCES:
1949         case -NFS4ERR_DELAY:
1950         case -ETIMEDOUT:
1951         case -EAGAIN:
1952                 ssleep(1);
1953                 break;
1954
1955         case -NFS4ERR_MINOR_VERS_MISMATCH:
1956                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1957                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1958                 dprintk("%s: exit with error %d for server %s\n",
1959                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1960                 return -EPROTONOSUPPORT;
1961         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1962                                  * in nfs4_exchange_id */
1963         default:
1964                 dprintk("%s: exit with error %d for server %s\n", __func__,
1965                                 status, clp->cl_hostname);
1966                 return status;
1967         }
1968         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1969         dprintk("%s: handled error %d for server %s\n", __func__, status,
1970                         clp->cl_hostname);
1971         return 0;
1972 }
1973
1974 static int nfs4_establish_lease(struct nfs_client *clp)
1975 {
1976         struct rpc_cred *cred;
1977         const struct nfs4_state_recovery_ops *ops =
1978                 clp->cl_mvops->reboot_recovery_ops;
1979         int status;
1980
1981         status = nfs4_begin_drain_session(clp);
1982         if (status != 0)
1983                 return status;
1984         cred = nfs4_get_clid_cred(clp);
1985         if (cred == NULL)
1986                 return -ENOENT;
1987         status = ops->establish_clid(clp, cred);
1988         put_rpccred(cred);
1989         if (status != 0)
1990                 return status;
1991         pnfs_destroy_all_layouts(clp);
1992         return 0;
1993 }
1994
1995 /*
1996  * Returns zero or a negative errno.  NFS4ERR values are converted
1997  * to local errno values.
1998  */
1999 static int nfs4_reclaim_lease(struct nfs_client *clp)
2000 {
2001         int status;
2002
2003         status = nfs4_establish_lease(clp);
2004         if (status < 0)
2005                 return nfs4_handle_reclaim_lease_error(clp, status);
2006         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2007                 nfs4_state_start_reclaim_nograce(clp);
2008         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2009                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2010         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2011         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2012         return 0;
2013 }
2014
2015 static int nfs4_purge_lease(struct nfs_client *clp)
2016 {
2017         int status;
2018
2019         status = nfs4_establish_lease(clp);
2020         if (status < 0)
2021                 return nfs4_handle_reclaim_lease_error(clp, status);
2022         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2023         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2024         nfs4_state_start_reclaim_nograce(clp);
2025         return 0;
2026 }
2027
2028 /*
2029  * Try remote migration of one FSID from a source server to a
2030  * destination server.  The source server provides a list of
2031  * potential destinations.
2032  *
2033  * Returns zero or a negative NFS4ERR status code.
2034  */
2035 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
2036 {
2037         struct nfs_client *clp = server->nfs_client;
2038         struct nfs4_fs_locations *locations = NULL;
2039         struct inode *inode;
2040         struct page *page;
2041         int status, result;
2042
2043         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2044                         (unsigned long long)server->fsid.major,
2045                         (unsigned long long)server->fsid.minor,
2046                         clp->cl_hostname);
2047
2048         result = 0;
2049         page = alloc_page(GFP_KERNEL);
2050         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2051         if (page == NULL || locations == NULL) {
2052                 dprintk("<-- %s: no memory\n", __func__);
2053                 goto out;
2054         }
2055
2056         inode = d_inode(server->super->s_root);
2057         result = nfs4_proc_get_locations(inode, locations, page, cred);
2058         if (result) {
2059                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2060                         __func__, result);
2061                 goto out;
2062         }
2063
2064         result = -NFS4ERR_NXIO;
2065         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2066                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2067                         __func__);
2068                 goto out;
2069         }
2070
2071         status = nfs4_begin_drain_session(clp);
2072         if (status != 0)
2073                 return status;
2074
2075         status = nfs4_replace_transport(server, locations);
2076         if (status != 0) {
2077                 dprintk("<-- %s: failed to replace transport: %d\n",
2078                         __func__, status);
2079                 goto out;
2080         }
2081
2082         result = 0;
2083         dprintk("<-- %s: migration succeeded\n", __func__);
2084
2085 out:
2086         if (page != NULL)
2087                 __free_page(page);
2088         kfree(locations);
2089         if (result) {
2090                 pr_err("NFS: migration recovery failed (server %s)\n",
2091                                 clp->cl_hostname);
2092                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2093         }
2094         return result;
2095 }
2096
2097 /*
2098  * Returns zero or a negative NFS4ERR status code.
2099  */
2100 static int nfs4_handle_migration(struct nfs_client *clp)
2101 {
2102         const struct nfs4_state_maintenance_ops *ops =
2103                                 clp->cl_mvops->state_renewal_ops;
2104         struct nfs_server *server;
2105         struct rpc_cred *cred;
2106
2107         dprintk("%s: migration reported on \"%s\"\n", __func__,
2108                         clp->cl_hostname);
2109
2110         spin_lock(&clp->cl_lock);
2111         cred = ops->get_state_renewal_cred_locked(clp);
2112         spin_unlock(&clp->cl_lock);
2113         if (cred == NULL)
2114                 return -NFS4ERR_NOENT;
2115
2116         clp->cl_mig_gen++;
2117 restart:
2118         rcu_read_lock();
2119         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2120                 int status;
2121
2122                 if (server->mig_gen == clp->cl_mig_gen)
2123                         continue;
2124                 server->mig_gen = clp->cl_mig_gen;
2125
2126                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2127                                                 &server->mig_status))
2128                         continue;
2129
2130                 rcu_read_unlock();
2131                 status = nfs4_try_migration(server, cred);
2132                 if (status < 0) {
2133                         put_rpccred(cred);
2134                         return status;
2135                 }
2136                 goto restart;
2137         }
2138         rcu_read_unlock();
2139         put_rpccred(cred);
2140         return 0;
2141 }
2142
2143 /*
2144  * Test each nfs_server on the clp's cl_superblocks list to see
2145  * if it's moved to another server.  Stop when the server no longer
2146  * returns NFS4ERR_LEASE_MOVED.
2147  */
2148 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2149 {
2150         const struct nfs4_state_maintenance_ops *ops =
2151                                 clp->cl_mvops->state_renewal_ops;
2152         struct nfs_server *server;
2153         struct rpc_cred *cred;
2154
2155         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2156                         clp->cl_hostname);
2157
2158         spin_lock(&clp->cl_lock);
2159         cred = ops->get_state_renewal_cred_locked(clp);
2160         spin_unlock(&clp->cl_lock);
2161         if (cred == NULL)
2162                 return -NFS4ERR_NOENT;
2163
2164         clp->cl_mig_gen++;
2165 restart:
2166         rcu_read_lock();
2167         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2168                 struct inode *inode;
2169                 int status;
2170
2171                 if (server->mig_gen == clp->cl_mig_gen)
2172                         continue;
2173                 server->mig_gen = clp->cl_mig_gen;
2174
2175                 rcu_read_unlock();
2176
2177                 inode = d_inode(server->super->s_root);
2178                 status = nfs4_proc_fsid_present(inode, cred);
2179                 if (status != -NFS4ERR_MOVED)
2180                         goto restart;   /* wasn't this one */
2181                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2182                         goto restart;   /* there are more */
2183                 goto out;
2184         }
2185         rcu_read_unlock();
2186
2187 out:
2188         put_rpccred(cred);
2189         return 0;
2190 }
2191
2192 /**
2193  * nfs4_discover_server_trunking - Detect server IP address trunking
2194  *
2195  * @clp: nfs_client under test
2196  * @result: OUT: found nfs_client, or clp
2197  *
2198  * Returns zero or a negative errno.  If zero is returned,
2199  * an nfs_client pointer is planted in "result".
2200  *
2201  * Note: since we are invoked in process context, and
2202  * not from inside the state manager, we cannot use
2203  * nfs4_handle_reclaim_lease_error().
2204  */
2205 int nfs4_discover_server_trunking(struct nfs_client *clp,
2206                                   struct nfs_client **result)
2207 {
2208         const struct nfs4_state_recovery_ops *ops =
2209                                 clp->cl_mvops->reboot_recovery_ops;
2210         struct rpc_clnt *clnt;
2211         struct rpc_cred *cred;
2212         int i, status;
2213
2214         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2215
2216         clnt = clp->cl_rpcclient;
2217         i = 0;
2218
2219         mutex_lock(&nfs_clid_init_mutex);
2220 again:
2221         status  = -ENOENT;
2222         cred = nfs4_get_clid_cred(clp);
2223         if (cred == NULL)
2224                 goto out_unlock;
2225
2226         status = ops->detect_trunking(clp, result, cred);
2227         put_rpccred(cred);
2228         switch (status) {
2229         case 0:
2230         case -EINTR:
2231         case -ERESTARTSYS:
2232                 break;
2233         case -ETIMEDOUT:
2234                 if (clnt->cl_softrtry)
2235                         break;
2236                 /* Fall through */
2237         case -NFS4ERR_DELAY:
2238         case -EAGAIN:
2239                 ssleep(1);
2240                 /* Fall through */
2241         case -NFS4ERR_STALE_CLIENTID:
2242                 dprintk("NFS: %s after status %d, retrying\n",
2243                         __func__, status);
2244                 goto again;
2245         case -EACCES:
2246                 if (i++ == 0) {
2247                         nfs4_root_machine_cred(clp);
2248                         goto again;
2249                 }
2250                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2251                         break;
2252                 /* Fall through */
2253         case -NFS4ERR_CLID_INUSE:
2254         case -NFS4ERR_WRONGSEC:
2255                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2256                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2257                         status = -EPERM;
2258                         break;
2259                 }
2260                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2261                 if (IS_ERR(clnt)) {
2262                         status = PTR_ERR(clnt);
2263                         break;
2264                 }
2265                 /* Note: this is safe because we haven't yet marked the
2266                  * client as ready, so we are the only user of
2267                  * clp->cl_rpcclient
2268                  */
2269                 clnt = xchg(&clp->cl_rpcclient, clnt);
2270                 rpc_shutdown_client(clnt);
2271                 clnt = clp->cl_rpcclient;
2272                 goto again;
2273
2274         case -NFS4ERR_MINOR_VERS_MISMATCH:
2275                 status = -EPROTONOSUPPORT;
2276                 break;
2277
2278         case -EKEYEXPIRED:
2279         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2280                                  * in nfs4_exchange_id */
2281                 status = -EKEYEXPIRED;
2282                 break;
2283         default:
2284                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2285                                 __func__, status);
2286                 status = -EIO;
2287         }
2288
2289 out_unlock:
2290         mutex_unlock(&nfs_clid_init_mutex);
2291         dprintk("NFS: %s: status = %d\n", __func__, status);
2292         return status;
2293 }
2294
2295 #ifdef CONFIG_NFS_V4_1
2296 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2297 {
2298         struct nfs_client *clp = session->clp;
2299
2300         switch (err) {
2301         default:
2302                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2303                 break;
2304         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2305                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2306         }
2307         nfs4_schedule_state_manager(clp);
2308 }
2309 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2310
2311 void nfs41_notify_server(struct nfs_client *clp)
2312 {
2313         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2314         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2315         nfs4_schedule_state_manager(clp);
2316 }
2317
2318 static void nfs4_reset_all_state(struct nfs_client *clp)
2319 {
2320         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2321                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2322                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2323                 nfs4_state_start_reclaim_nograce(clp);
2324                 dprintk("%s: scheduling reset of all state for server %s!\n",
2325                                 __func__, clp->cl_hostname);
2326                 nfs4_schedule_state_manager(clp);
2327         }
2328 }
2329
2330 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2331 {
2332         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2333                 nfs4_state_start_reclaim_reboot(clp);
2334                 dprintk("%s: server %s rebooted!\n", __func__,
2335                                 clp->cl_hostname);
2336                 nfs4_schedule_state_manager(clp);
2337         }
2338 }
2339
2340 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2341 {
2342         nfs4_reset_all_state(clp);
2343         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2344 }
2345
2346 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2347 {
2348         nfs4_state_start_reclaim_nograce(clp);
2349         nfs4_schedule_state_manager(clp);
2350
2351         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2352 }
2353
2354 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2355 {
2356         /* FIXME: For now, we destroy all layouts. */
2357         pnfs_destroy_all_layouts(clp);
2358         /* FIXME: For now, we test all delegations+open state+locks. */
2359         nfs41_handle_some_state_revoked(clp);
2360         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2361                         clp->cl_hostname);
2362 }
2363
2364 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2365 {
2366         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2367         nfs4_schedule_state_manager(clp);
2368
2369         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2370                         clp->cl_hostname);
2371 }
2372
2373 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2374 {
2375         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2376                 &clp->cl_state) == 0)
2377                 nfs4_schedule_state_manager(clp);
2378 }
2379
2380 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2381                 bool recovery)
2382 {
2383         if (!flags)
2384                 return;
2385
2386         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2387                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2388         /*
2389          * If we're called from the state manager thread, then assume we're
2390          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2391          * Those flags are expected to remain set until we're done
2392          * recovering (see RFC5661, section 18.46.3).
2393          */
2394         if (recovery)
2395                 goto out_recovery;
2396
2397         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2398                 nfs41_handle_server_reboot(clp);
2399         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2400                 nfs41_handle_all_state_revoked(clp);
2401         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2402                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2403                 nfs41_handle_some_state_revoked(clp);
2404         if (flags & SEQ4_STATUS_LEASE_MOVED)
2405                 nfs4_schedule_lease_moved_recovery(clp);
2406         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2407                 nfs41_handle_recallable_state_revoked(clp);
2408 out_recovery:
2409         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2410                 nfs41_handle_backchannel_fault(clp);
2411         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2412                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2413                 nfs41_handle_cb_path_down(clp);
2414 }
2415
2416 static int nfs4_reset_session(struct nfs_client *clp)
2417 {
2418         struct rpc_cred *cred;
2419         int status;
2420
2421         if (!nfs4_has_session(clp))
2422                 return 0;
2423         status = nfs4_begin_drain_session(clp);
2424         if (status != 0)
2425                 return status;
2426         cred = nfs4_get_clid_cred(clp);
2427         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2428         switch (status) {
2429         case 0:
2430         case -NFS4ERR_BADSESSION:
2431         case -NFS4ERR_DEADSESSION:
2432                 break;
2433         case -NFS4ERR_BACK_CHAN_BUSY:
2434         case -NFS4ERR_DELAY:
2435                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2436                 status = 0;
2437                 ssleep(1);
2438                 goto out;
2439         default:
2440                 status = nfs4_recovery_handle_error(clp, status);
2441                 goto out;
2442         }
2443
2444         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2445         status = nfs4_proc_create_session(clp, cred);
2446         if (status) {
2447                 dprintk("%s: session reset failed with status %d for server %s!\n",
2448                         __func__, status, clp->cl_hostname);
2449                 status = nfs4_handle_reclaim_lease_error(clp, status);
2450                 goto out;
2451         }
2452         nfs41_finish_session_reset(clp);
2453         dprintk("%s: session reset was successful for server %s!\n",
2454                         __func__, clp->cl_hostname);
2455 out:
2456         if (cred)
2457                 put_rpccred(cred);
2458         return status;
2459 }
2460
2461 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2462 {
2463         struct rpc_cred *cred;
2464         int ret;
2465
2466         if (!nfs4_has_session(clp))
2467                 return 0;
2468         ret = nfs4_begin_drain_session(clp);
2469         if (ret != 0)
2470                 return ret;
2471         cred = nfs4_get_clid_cred(clp);
2472         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2473         if (cred)
2474                 put_rpccred(cred);
2475         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2476         switch (ret) {
2477         case 0:
2478                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2479                         __func__, clp->cl_hostname);
2480                 break;
2481         case -NFS4ERR_DELAY:
2482                 ssleep(1);
2483                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2484                 break;
2485         default:
2486                 return nfs4_recovery_handle_error(clp, ret);
2487         }
2488         return 0;
2489 }
2490 #else /* CONFIG_NFS_V4_1 */
2491 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2492
2493 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2494 {
2495         return 0;
2496 }
2497 #endif /* CONFIG_NFS_V4_1 */
2498
2499 static void nfs4_state_manager(struct nfs_client *clp)
2500 {
2501         int status = 0;
2502         const char *section = "", *section_sep = "";
2503
2504         /* Ensure exclusive access to NFSv4 state */
2505         do {
2506                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2507                         section = "purge state";
2508                         status = nfs4_purge_lease(clp);
2509                         if (status < 0)
2510                                 goto out_error;
2511                         continue;
2512                 }
2513
2514                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2515                         section = "lease expired";
2516                         /* We're going to have to re-establish a clientid */
2517                         status = nfs4_reclaim_lease(clp);
2518                         if (status < 0)
2519                                 goto out_error;
2520                         continue;
2521                 }
2522
2523                 /* Initialize or reset the session */
2524                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2525                         section = "reset session";
2526                         status = nfs4_reset_session(clp);
2527                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2528                                 continue;
2529                         if (status < 0)
2530                                 goto out_error;
2531                 }
2532
2533                 /* Send BIND_CONN_TO_SESSION */
2534                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2535                                 &clp->cl_state)) {
2536                         section = "bind conn to session";
2537                         status = nfs4_bind_conn_to_session(clp);
2538                         if (status < 0)
2539                                 goto out_error;
2540                         continue;
2541                 }
2542
2543                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2544                         section = "check lease";
2545                         status = nfs4_check_lease(clp);
2546                         if (status < 0)
2547                                 goto out_error;
2548                         continue;
2549                 }
2550
2551                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2552                         section = "migration";
2553                         status = nfs4_handle_migration(clp);
2554                         if (status < 0)
2555                                 goto out_error;
2556                 }
2557
2558                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2559                         section = "lease moved";
2560                         status = nfs4_handle_lease_moved(clp);
2561                         if (status < 0)
2562                                 goto out_error;
2563                 }
2564
2565                 /* First recover reboot state... */
2566                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2567                         section = "reclaim reboot";
2568                         status = nfs4_do_reclaim(clp,
2569                                 clp->cl_mvops->reboot_recovery_ops);
2570                         if (status == -EAGAIN)
2571                                 continue;
2572                         if (status < 0)
2573                                 goto out_error;
2574                         nfs4_state_end_reclaim_reboot(clp);
2575                 }
2576
2577                 /* Detect expired delegations... */
2578                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2579                         section = "detect expired delegations";
2580                         nfs_reap_expired_delegations(clp);
2581                         continue;
2582                 }
2583
2584                 /* Now recover expired state... */
2585                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2586                         section = "reclaim nograce";
2587                         status = nfs4_do_reclaim(clp,
2588                                 clp->cl_mvops->nograce_recovery_ops);
2589                         if (status == -EAGAIN)
2590                                 continue;
2591                         if (status < 0)
2592                                 goto out_error;
2593                 }
2594
2595                 nfs4_end_drain_session(clp);
2596                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2597                         nfs_client_return_marked_delegations(clp);
2598                         continue;
2599                 }
2600
2601                 nfs4_clear_state_manager_bit(clp);
2602                 /* Did we race with an attempt to give us more work? */
2603                 if (clp->cl_state == 0)
2604                         return;
2605                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2606                         return;
2607         } while (refcount_read(&clp->cl_count) > 1 && !signalled());
2608         goto out_drain;
2609
2610 out_error:
2611         if (strlen(section))
2612                 section_sep = ": ";
2613         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2614                         " with error %d\n", section_sep, section,
2615                         clp->cl_hostname, -status);
2616         ssleep(1);
2617 out_drain:
2618         nfs4_end_drain_session(clp);
2619         nfs4_clear_state_manager_bit(clp);
2620 }
2621
2622 static int nfs4_run_state_manager(void *ptr)
2623 {
2624         struct nfs_client *clp = ptr;
2625
2626         allow_signal(SIGKILL);
2627         nfs4_state_manager(clp);
2628         nfs_put_client(clp);
2629         module_put_and_exit(0);
2630         return 0;
2631 }
2632
2633 /*
2634  * Local variables:
2635  *  c-basic-offset: 8
2636  * End:
2637  */