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 static DEFINE_MUTEX(nfs_clid_init_mutex);
  73
  74 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  75 {
  76         struct nfs4_setclientid_res clid = {
  77                 .clientid = clp->cl_clientid,
  78                 .confirm = clp->cl_confirm,
  79         };
  80         unsigned short port;
  81         int status;
  82         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  83
  84         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  85                 goto do_confirm;
  86         port = nn->nfs_callback_tcpport;
  87         if (clp->cl_addr.ss_family == AF_INET6)
  88                 port = nn->nfs_callback_tcpport6;
  89
  90         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  91         if (status != 0)
  92                 goto out;
  93         clp->cl_clientid = clid.clientid;
  94         clp->cl_confirm = clid.confirm;
  95         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  96 do_confirm:
  97         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  98         if (status != 0)
  99                 goto out;
 100         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 101         nfs4_schedule_state_renewal(clp);
 102 out:
 103         return status;
 104 }
 105
 106 /**
 107  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 108  *
 109  * @clp: nfs_client under test
 110  * @result: OUT: found nfs_client, or clp
 111  * @cred: credential to use for trunking test
 112  *
 113  * Returns zero, a negative errno, or a negative NFS4ERR status.
 114  * If zero is returned, an nfs_client pointer is planted in
 115  * "result".
 116  *
 117  * Note: The returned client may not yet be marked ready.
 118  */
 119 int nfs40_discover_server_trunking(struct nfs_client *clp,
 120                                    struct nfs_client **result,
 121                                    struct rpc_cred *cred)
 122 {
 123         struct nfs4_setclientid_res clid = {
 124                 .clientid = clp->cl_clientid,
 125                 .confirm = clp->cl_confirm,
 126         };
 127         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 128         unsigned short port;
 129         int status;
 130
 131         port = nn->nfs_callback_tcpport;
 132         if (clp->cl_addr.ss_family == AF_INET6)
 133                 port = nn->nfs_callback_tcpport6;
 134
 135         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 136         if (status != 0)
 137                 goto out;
 138         clp->cl_clientid = clid.clientid;
 139         clp->cl_confirm = clid.confirm;
 140
 141         status = nfs40_walk_client_list(clp, result, cred);
 142         if (status == 0) {
 143                 /* Sustain the lease, even if it's empty.  If the clientid4
 144                  * goes stale it's of no use for trunking discovery. */
 145                 nfs4_schedule_state_renewal(*result);
 146         }
 147 out:
 148         return status;
 149 }
 150
 151 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 152 {
 153         struct rpc_cred *cred = NULL;
 154
 155         if (clp->cl_machine_cred != NULL)
 156                 cred = get_rpccred(clp->cl_machine_cred);
 157         return cred;
 158 }
 159
 160 static void nfs4_root_machine_cred(struct nfs_client *clp)
 161 {
 162         struct rpc_cred *cred, *new;
 163
 164         new = rpc_lookup_machine_cred(NULL);
 165         spin_lock(&clp->cl_lock);
 166         cred = clp->cl_machine_cred;
 167         clp->cl_machine_cred = new;
 168         spin_unlock(&clp->cl_lock);
 169         if (cred != NULL)
 170                 put_rpccred(cred);
 171 }
 172
 173 static struct rpc_cred *
 174 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 175 {
 176         struct rpc_cred *cred = NULL;
 177         struct nfs4_state_owner *sp;
 178         struct rb_node *pos;
 179
 180         for (pos = rb_first(&server->state_owners);
 181              pos != NULL;
 182              pos = rb_next(pos)) {
 183                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 184                 if (list_empty(&sp->so_states))
 185                         continue;
 186                 cred = get_rpccred(sp->so_cred);
 187                 break;
 188         }
 189         return cred;
 190 }
 191
 192 /**
 193  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 194  * @clp: client state handle
 195  *
 196  * Returns an rpc_cred with reference count bumped, or NULL.
 197  * Caller must hold clp->cl_lock.
 198  */
 199 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 200 {
 201         struct rpc_cred *cred = NULL;
 202         struct nfs_server *server;
 203
 204         /* Use machine credentials if available */
 205         cred = nfs4_get_machine_cred_locked(clp);
 206         if (cred != NULL)
 207                 goto out;
 208
 209         rcu_read_lock();
 210         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 211                 cred = nfs4_get_renew_cred_server_locked(server);
 212                 if (cred != NULL)
 213                         break;
 214         }
 215         rcu_read_unlock();
 216
 217 out:
 218         return cred;
 219 }
 220
 221 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 222 {
 223         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 224                 spin_lock(&tbl->slot_tbl_lock);
 225                 nfs41_wake_slot_table(tbl);
 226                 spin_unlock(&tbl->slot_tbl_lock);
 227         }
 228 }
 229
 230 static void nfs4_end_drain_session(struct nfs_client *clp)
 231 {
 232         struct nfs4_session *ses = clp->cl_session;
 233
 234         if (clp->cl_slot_tbl) {
 235                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 236                 return;
 237         }
 238
 239         if (ses != NULL) {
 240                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 241                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 242         }
 243 }
 244
 245 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 246 {
 247         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 248         spin_lock(&tbl->slot_tbl_lock);
 249         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 250                 reinit_completion(&tbl->complete);
 251                 spin_unlock(&tbl->slot_tbl_lock);
 252                 return wait_for_completion_interruptible(&tbl->complete);
 253         }
 254         spin_unlock(&tbl->slot_tbl_lock);
 255         return 0;
 256 }
 257
 258 static int nfs4_begin_drain_session(struct nfs_client *clp)
 259 {
 260         struct nfs4_session *ses = clp->cl_session;
 261         int ret = 0;
 262
 263         if (clp->cl_slot_tbl)
 264                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 265
 266         /* back channel */
 267         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 268         if (ret)
 269                 return ret;
 270         /* fore channel */
 271         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 272 }
 273
 274 #if defined(CONFIG_NFS_V4_1)
 275
 276 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 277 {
 278         int status;
 279         struct nfs_fsinfo fsinfo;
 280         unsigned long now;
 281
 282         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 283                 nfs4_schedule_state_renewal(clp);
 284                 return 0;
 285         }
 286
 287         now = jiffies;
 288         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 289         if (status == 0) {
 290                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
 291                 nfs4_schedule_state_renewal(clp);
 292         }
 293
 294         return status;
 295 }
 296
 297 static void nfs41_finish_session_reset(struct nfs_client *clp)
 298 {
 299         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 300         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 301         /* create_session negotiated new slot table */
 302         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 303         nfs41_setup_state_renewal(clp);
 304 }
 305
 306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 307 {
 308         int status;
 309
 310         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 311                 goto do_confirm;
 312         status = nfs4_proc_exchange_id(clp, cred);
 313         if (status != 0)
 314                 goto out;
 315         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 316 do_confirm:
 317         status = nfs4_proc_create_session(clp, cred);
 318         if (status != 0)
 319                 goto out;
 320         nfs41_finish_session_reset(clp);
 321         nfs_mark_client_ready(clp, NFS_CS_READY);
 322 out:
 323         return status;
 324 }
 325
 326 /**
 327  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 328  *
 329  * @clp: nfs_client under test
 330  * @result: OUT: found nfs_client, or clp
 331  * @cred: credential to use for trunking test
 332  *
 333  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 334  * If NFS4_OK is returned, an nfs_client pointer is planted in
 335  * "result".
 336  *
 337  * Note: The returned client may not yet be marked ready.
 338  */
 339 int nfs41_discover_server_trunking(struct nfs_client *clp,
 340                                    struct nfs_client **result,
 341                                    struct rpc_cred *cred)
 342 {
 343         int status;
 344
 345         status = nfs4_proc_exchange_id(clp, cred);
 346         if (status != NFS4_OK)
 347                 return status;
 348
 349         status = nfs41_walk_client_list(clp, result, cred);
 350         if (status < 0)
 351                 return status;
 352         if (clp != *result)
 353                 return 0;
 354
 355         /* Purge state if the client id was established in a prior instance */
 356         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
 357                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 358         else
 359                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 360         nfs4_schedule_state_manager(clp);
 361         status = nfs_wait_client_init_complete(clp);
 362         if (status < 0)
 363                 nfs_put_client(clp);
 364         return status;
 365 }
 366
 367 #endif /* CONFIG_NFS_V4_1 */
 368
 369 /**
 370  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 371  * @clp: client state handle
 372  *
 373  * Returns an rpc_cred with reference count bumped, or NULL.
 374  */
 375 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
 376 {
 377         struct rpc_cred *cred;
 378
 379         spin_lock(&clp->cl_lock);
 380         cred = nfs4_get_machine_cred_locked(clp);
 381         spin_unlock(&clp->cl_lock);
 382         return cred;
 383 }
 384
 385 static struct nfs4_state_owner *
 386 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 387 {
 388         struct rb_node **p = &server->state_owners.rb_node,
 389                        *parent = NULL;
 390         struct nfs4_state_owner *sp;
 391
 392         while (*p != NULL) {
 393                 parent = *p;
 394                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 395
 396                 if (cred < sp->so_cred)
 397                         p = &parent->rb_left;
 398                 else if (cred > sp->so_cred)
 399                         p = &parent->rb_right;
 400                 else {
 401                         if (!list_empty(&sp->so_lru))
 402                                 list_del_init(&sp->so_lru);
 403                         atomic_inc(&sp->so_count);
 404                         return sp;
 405                 }
 406         }
 407         return NULL;
 408 }
 409
 410 static struct nfs4_state_owner *
 411 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 412 {
 413         struct nfs_server *server = new->so_server;
 414         struct rb_node **p = &server->state_owners.rb_node,
 415                        *parent = NULL;
 416         struct nfs4_state_owner *sp;
 417         int err;
 418
 419         while (*p != NULL) {
 420                 parent = *p;
 421                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 422
 423                 if (new->so_cred < sp->so_cred)
 424                         p = &parent->rb_left;
 425                 else if (new->so_cred > sp->so_cred)
 426                         p = &parent->rb_right;
 427                 else {
 428                         if (!list_empty(&sp->so_lru))
 429                                 list_del_init(&sp->so_lru);
 430                         atomic_inc(&sp->so_count);
 431                         return sp;
 432                 }
 433         }
 434         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 435         if (err)
 436                 return ERR_PTR(err);
 437         rb_link_node(&new->so_server_node, parent, p);
 438         rb_insert_color(&new->so_server_node, &server->state_owners);
 439         return new;
 440 }
 441
 442 static void
 443 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 444 {
 445         struct nfs_server *server = sp->so_server;
 446
 447         if (!RB_EMPTY_NODE(&sp->so_server_node))
 448                 rb_erase(&sp->so_server_node, &server->state_owners);
 449         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 450 }
 451
 452 static void
 453 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 454 {
 455         sc->create_time = ktime_get();
 456         sc->flags = 0;
 457         sc->counter = 0;
 458         spin_lock_init(&sc->lock);
 459         INIT_LIST_HEAD(&sc->list);
 460         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 461 }
 462
 463 static void
 464 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 465 {
 466         rpc_destroy_wait_queue(&sc->wait);
 467 }
 468
 469 /*
 470  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 471  * create a new state_owner.
 472  *
 473  */
 474 static struct nfs4_state_owner *
 475 nfs4_alloc_state_owner(struct nfs_server *server,
 476                 struct rpc_cred *cred,
 477                 gfp_t gfp_flags)
 478 {
 479         struct nfs4_state_owner *sp;
 480
 481         sp = kzalloc(sizeof(*sp), gfp_flags);
 482         if (!sp)
 483                 return NULL;
 484         sp->so_server = server;
 485         sp->so_cred = get_rpccred(cred);
 486         spin_lock_init(&sp->so_lock);
 487         INIT_LIST_HEAD(&sp->so_states);
 488         nfs4_init_seqid_counter(&sp->so_seqid);
 489         atomic_set(&sp->so_count, 1);
 490         INIT_LIST_HEAD(&sp->so_lru);
 491         seqcount_init(&sp->so_reclaim_seqcount);
 492         mutex_init(&sp->so_delegreturn_mutex);
 493         return sp;
 494 }
 495
 496 static void
 497 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 498 {
 499         struct rb_node *rb_node = &sp->so_server_node;
 500
 501         if (!RB_EMPTY_NODE(rb_node)) {
 502                 struct nfs_server *server = sp->so_server;
 503                 struct nfs_client *clp = server->nfs_client;
 504
 505                 spin_lock(&clp->cl_lock);
 506                 if (!RB_EMPTY_NODE(rb_node)) {
 507                         rb_erase(rb_node, &server->state_owners);
 508                         RB_CLEAR_NODE(rb_node);
 509                 }
 510                 spin_unlock(&clp->cl_lock);
 511         }
 512 }
 513
 514 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 515 {
 516         nfs4_destroy_seqid_counter(&sp->so_seqid);
 517         put_rpccred(sp->so_cred);
 518         kfree(sp);
 519 }
 520
 521 static void nfs4_gc_state_owners(struct nfs_server *server)
 522 {
 523         struct nfs_client *clp = server->nfs_client;
 524         struct nfs4_state_owner *sp, *tmp;
 525         unsigned long time_min, time_max;
 526         LIST_HEAD(doomed);
 527
 528         spin_lock(&clp->cl_lock);
 529         time_max = jiffies;
 530         time_min = (long)time_max - (long)clp->cl_lease_time;
 531         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 532                 /* NB: LRU is sorted so that oldest is at the head */
 533                 if (time_in_range(sp->so_expires, time_min, time_max))
 534                         break;
 535                 list_move(&sp->so_lru, &doomed);
 536                 nfs4_remove_state_owner_locked(sp);
 537         }
 538         spin_unlock(&clp->cl_lock);
 539
 540         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 541                 list_del(&sp->so_lru);
 542                 nfs4_free_state_owner(sp);
 543         }
 544 }
 545
 546 /**
 547  * nfs4_get_state_owner - Look up a state owner given a credential
 548  * @server: nfs_server to search
 549  * @cred: RPC credential to match
 550  *
 551  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 552  */
 553 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 554                                               struct rpc_cred *cred,
 555                                               gfp_t gfp_flags)
 556 {
 557         struct nfs_client *clp = server->nfs_client;
 558         struct nfs4_state_owner *sp, *new;
 559
 560         spin_lock(&clp->cl_lock);
 561         sp = nfs4_find_state_owner_locked(server, cred);
 562         spin_unlock(&clp->cl_lock);
 563         if (sp != NULL)
 564                 goto out;
 565         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 566         if (new == NULL)
 567                 goto out;
 568         do {
 569                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 570                         break;
 571                 spin_lock(&clp->cl_lock);
 572                 sp = nfs4_insert_state_owner_locked(new);
 573                 spin_unlock(&clp->cl_lock);
 574         } while (sp == ERR_PTR(-EAGAIN));
 575         if (sp != new)
 576                 nfs4_free_state_owner(new);
 577 out:
 578         nfs4_gc_state_owners(server);
 579         return sp;
 580 }
 581
 582 /**
 583  * nfs4_put_state_owner - Release a nfs4_state_owner
 584  * @sp: state owner data to release
 585  *
 586  * Note that we keep released state owners on an LRU
 587  * list.
 588  * This caches valid state owners so that they can be
 589  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 590  * It also pins the uniquifier of dropped state owners for
 591  * a while, to ensure that those state owner names are
 592  * never reused.
 593  */
 594 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 595 {
 596         struct nfs_server *server = sp->so_server;
 597         struct nfs_client *clp = server->nfs_client;
 598
 599         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 600                 return;
 601
 602         sp->so_expires = jiffies;
 603         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 604         spin_unlock(&clp->cl_lock);
 605 }
 606
 607 /**
 608  * nfs4_purge_state_owners - Release all cached state owners
 609  * @server: nfs_server with cached state owners to release
 610  *
 611  * Called at umount time.  Remaining state owners will be on
 612  * the LRU with ref count of zero.
 613  */
 614 void nfs4_purge_state_owners(struct nfs_server *server)
 615 {
 616         struct nfs_client *clp = server->nfs_client;
 617         struct nfs4_state_owner *sp, *tmp;
 618         LIST_HEAD(doomed);
 619
 620         spin_lock(&clp->cl_lock);
 621         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 622                 list_move(&sp->so_lru, &doomed);
 623                 nfs4_remove_state_owner_locked(sp);
 624         }
 625         spin_unlock(&clp->cl_lock);
 626
 627         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 628                 list_del(&sp->so_lru);
 629                 nfs4_free_state_owner(sp);
 630         }
 631 }
 632
 633 static struct nfs4_state *
 634 nfs4_alloc_open_state(void)
 635 {
 636         struct nfs4_state *state;
 637
 638         state = kzalloc(sizeof(*state), GFP_NOFS);
 639         if (!state)
 640                 return NULL;
 641         atomic_set(&state->count, 1);
 642         INIT_LIST_HEAD(&state->lock_states);
 643         spin_lock_init(&state->state_lock);
 644         seqlock_init(&state->seqlock);
 645         return state;
 646 }
 647
 648 void
 649 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 650 {
 651         if (state->state == fmode)
 652                 return;
 653         /* NB! List reordering - see the reclaim code for why.  */
 654         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 655                 if (fmode & FMODE_WRITE)
 656                         list_move(&state->open_states, &state->owner->so_states);
 657                 else
 658                         list_move_tail(&state->open_states, &state->owner->so_states);
 659         }
 660         state->state = fmode;
 661 }
 662
 663 static struct nfs4_state *
 664 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 665 {
 666         struct nfs_inode *nfsi = NFS_I(inode);
 667         struct nfs4_state *state;
 668
 669         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 670                 if (state->owner != owner)
 671                         continue;
 672                 if (!nfs4_valid_open_stateid(state))
 673                         continue;
 674                 if (atomic_inc_not_zero(&state->count))
 675                         return state;
 676         }
 677         return NULL;
 678 }
 679
 680 static void
 681 nfs4_free_open_state(struct nfs4_state *state)
 682 {
 683         kfree(state);
 684 }
 685
 686 struct nfs4_state *
 687 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 688 {
 689         struct nfs4_state *state, *new;
 690         struct nfs_inode *nfsi = NFS_I(inode);
 691
 692         spin_lock(&inode->i_lock);
 693         state = __nfs4_find_state_byowner(inode, owner);
 694         spin_unlock(&inode->i_lock);
 695         if (state)
 696                 goto out;
 697         new = nfs4_alloc_open_state();
 698         spin_lock(&owner->so_lock);
 699         spin_lock(&inode->i_lock);
 700         state = __nfs4_find_state_byowner(inode, owner);
 701         if (state == NULL && new != NULL) {
 702                 state = new;
 703                 state->owner = owner;
 704                 atomic_inc(&owner->so_count);
 705                 list_add(&state->inode_states, &nfsi->open_states);
 706                 ihold(inode);
 707                 state->inode = inode;
 708                 spin_unlock(&inode->i_lock);
 709                 /* Note: The reclaim code dictates that we add stateless
 710                  * and read-only stateids to the end of the list */
 711                 list_add_tail(&state->open_states, &owner->so_states);
 712                 spin_unlock(&owner->so_lock);
 713         } else {
 714                 spin_unlock(&inode->i_lock);
 715                 spin_unlock(&owner->so_lock);
 716                 if (new)
 717                         nfs4_free_open_state(new);
 718         }
 719 out:
 720         return state;
 721 }
 722
 723 void nfs4_put_open_state(struct nfs4_state *state)
 724 {
 725         struct inode *inode = state->inode;
 726         struct nfs4_state_owner *owner = state->owner;
 727
 728         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 729                 return;
 730         spin_lock(&inode->i_lock);
 731         list_del(&state->inode_states);
 732         list_del(&state->open_states);
 733         spin_unlock(&inode->i_lock);
 734         spin_unlock(&owner->so_lock);
 735         iput(inode);
 736         nfs4_free_open_state(state);
 737         nfs4_put_state_owner(owner);
 738 }
 739
 740 /*
 741  * Close the current file.
 742  */
 743 static void __nfs4_close(struct nfs4_state *state,
 744                 fmode_t fmode, gfp_t gfp_mask, int wait)
 745 {
 746         struct nfs4_state_owner *owner = state->owner;
 747         int call_close = 0;
 748         fmode_t newstate;
 749
 750         atomic_inc(&owner->so_count);
 751         /* Protect against nfs4_find_state() */
 752         spin_lock(&owner->so_lock);
 753         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 754                 case FMODE_READ:
 755                         state->n_rdonly--;
 756                         break;
 757                 case FMODE_WRITE:
 758                         state->n_wronly--;
 759                         break;
 760                 case FMODE_READ|FMODE_WRITE:
 761                         state->n_rdwr--;
 762         }
 763         newstate = FMODE_READ|FMODE_WRITE;
 764         if (state->n_rdwr == 0) {
 765                 if (state->n_rdonly == 0) {
 766                         newstate &= ~FMODE_READ;
 767                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 768                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 769                 }
 770                 if (state->n_wronly == 0) {
 771                         newstate &= ~FMODE_WRITE;
 772                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 773                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 774                 }
 775                 if (newstate == 0)
 776                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 777         }
 778         nfs4_state_set_mode_locked(state, newstate);
 779         spin_unlock(&owner->so_lock);
 780
 781         if (!call_close) {
 782                 nfs4_put_open_state(state);
 783                 nfs4_put_state_owner(owner);
 784         } else
 785                 nfs4_do_close(state, gfp_mask, wait);
 786 }
 787
 788 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 789 {
 790         __nfs4_close(state, fmode, GFP_NOFS, 0);
 791 }
 792
 793 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 794 {
 795         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 796 }
 797
 798 /*
 799  * Search the state->lock_states for an existing lock_owner
 800  * that is compatible with current->files
 801  */
 802 static struct nfs4_lock_state *
 803 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 804 {
 805         struct nfs4_lock_state *pos;
 806         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 807                 if (pos->ls_owner != fl_owner)
 808                         continue;
 809                 atomic_inc(&pos->ls_count);
 810                 return pos;
 811         }
 812         return NULL;
 813 }
 814
 815 /*
 816  * Return a compatible lock_state. If no initialized lock_state structure
 817  * exists, return an uninitialized one.
 818  *
 819  */
 820 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 821 {
 822         struct nfs4_lock_state *lsp;
 823         struct nfs_server *server = state->owner->so_server;
 824
 825         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 826         if (lsp == NULL)
 827                 return NULL;
 828         nfs4_init_seqid_counter(&lsp->ls_seqid);
 829         atomic_set(&lsp->ls_count, 1);
 830         lsp->ls_state = state;
 831         lsp->ls_owner = fl_owner;
 832         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 833         if (lsp->ls_seqid.owner_id < 0)
 834                 goto out_free;
 835         INIT_LIST_HEAD(&lsp->ls_locks);
 836         return lsp;
 837 out_free:
 838         kfree(lsp);
 839         return NULL;
 840 }
 841
 842 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 843 {
 844         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 845         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 846         kfree(lsp);
 847 }
 848
 849 /*
 850  * Return a compatible lock_state. If no initialized lock_state structure
 851  * exists, return an uninitialized one.
 852  *
 853  */
 854 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 855 {
 856         struct nfs4_lock_state *lsp, *new = NULL;
 857
 858         for(;;) {
 859                 spin_lock(&state->state_lock);
 860                 lsp = __nfs4_find_lock_state(state, owner);
 861                 if (lsp != NULL)
 862                         break;
 863                 if (new != NULL) {
 864                         list_add(&new->ls_locks, &state->lock_states);
 865                         set_bit(LK_STATE_IN_USE, &state->flags);
 866                         lsp = new;
 867                         new = NULL;
 868                         break;
 869                 }
 870                 spin_unlock(&state->state_lock);
 871                 new = nfs4_alloc_lock_state(state, owner);
 872                 if (new == NULL)
 873                         return NULL;
 874         }
 875         spin_unlock(&state->state_lock);
 876         if (new != NULL)
 877                 nfs4_free_lock_state(state->owner->so_server, new);
 878         return lsp;
 879 }
 880
 881 /*
 882  * Release reference to lock_state, and free it if we see that
 883  * it is no longer in use
 884  */
 885 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 886 {
 887         struct nfs_server *server;
 888         struct nfs4_state *state;
 889
 890         if (lsp == NULL)
 891                 return;
 892         state = lsp->ls_state;
 893         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 894                 return;
 895         list_del(&lsp->ls_locks);
 896         if (list_empty(&state->lock_states))
 897                 clear_bit(LK_STATE_IN_USE, &state->flags);
 898         spin_unlock(&state->state_lock);
 899         server = state->owner->so_server;
 900         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 901                 struct nfs_client *clp = server->nfs_client;
 902
 903                 clp->cl_mvops->free_lock_state(server, lsp);
 904         } else
 905                 nfs4_free_lock_state(server, lsp);
 906 }
 907
 908 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 909 {
 910         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 911
 912         dst->fl_u.nfs4_fl.owner = lsp;
 913         atomic_inc(&lsp->ls_count);
 914 }
 915
 916 static void nfs4_fl_release_lock(struct file_lock *fl)
 917 {
 918         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 919 }
 920
 921 static const struct file_lock_operations nfs4_fl_lock_ops = {
 922         .fl_copy_lock = nfs4_fl_copy_lock,
 923         .fl_release_private = nfs4_fl_release_lock,
 924 };
 925
 926 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 927 {
 928         struct nfs4_lock_state *lsp;
 929
 930         if (fl->fl_ops != NULL)
 931                 return 0;
 932         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 933         if (lsp == NULL)
 934                 return -ENOMEM;
 935         fl->fl_u.nfs4_fl.owner = lsp;
 936         fl->fl_ops = &nfs4_fl_lock_ops;
 937         return 0;
 938 }
 939
 940 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 941                 struct nfs4_state *state,
 942                 const struct nfs_lockowner *lockowner)
 943 {
 944         struct nfs4_lock_state *lsp;
 945         fl_owner_t fl_owner;
 946         int ret = -ENOENT;
 947
 948
 949         if (lockowner == NULL)
 950                 goto out;
 951
 952         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 953                 goto out;
 954
 955         fl_owner = lockowner->l_owner;
 956         spin_lock(&state->state_lock);
 957         lsp = __nfs4_find_lock_state(state, fl_owner);
 958         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 959                 ret = -EIO;
 960         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
 961                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 962                 ret = 0;
 963         }
 964         spin_unlock(&state->state_lock);
 965         nfs4_put_lock_state(lsp);
 966 out:
 967         return ret;
 968 }
 969
 970 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 971 {
 972         const nfs4_stateid *src;
 973         int seq;
 974
 975         do {
 976                 src = &zero_stateid;
 977                 seq = read_seqbegin(&state->seqlock);
 978                 if (test_bit(NFS_OPEN_STATE, &state->flags))
 979                         src = &state->open_stateid;
 980                 nfs4_stateid_copy(dst, src);
 981         } while (read_seqretry(&state->seqlock, seq));
 982 }
 983
 984 /*
 985  * Byte-range lock aware utility to initialize the stateid of read/write
 986  * requests.
 987  */
 988 int nfs4_select_rw_stateid(struct nfs4_state *state,
 989                 fmode_t fmode, const struct nfs_lockowner *lockowner,
 990                 nfs4_stateid *dst, struct rpc_cred **cred)
 991 {
 992         int ret;
 993
 994         if (!nfs4_valid_open_stateid(state))
 995                 return -EIO;
 996         if (cred != NULL)
 997                 *cred = NULL;
 998         ret = nfs4_copy_lock_stateid(dst, state, lockowner);
 999         if (ret == -EIO)
1000                 /* A lost lock - don't even consider delegations */
1001                 goto out;
1002         /* returns true if delegation stateid found and copied */
1003         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1004                 ret = 0;
1005                 goto out;
1006         }
1007         if (ret != -ENOENT)
1008                 /* nfs4_copy_delegation_stateid() didn't over-write
1009                  * dst, so it still has the lock stateid which we now
1010                  * choose to use.
1011                  */
1012                 goto out;
1013         nfs4_copy_open_stateid(dst, state);
1014         ret = 0;
1015 out:
1016         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1017                 dst->seqid = 0;
1018         return ret;
1019 }
1020
1021 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1022 {
1023         struct nfs_seqid *new;
1024
1025         new = kmalloc(sizeof(*new), gfp_mask);
1026         if (new == NULL)
1027                 return ERR_PTR(-ENOMEM);
1028         new->sequence = counter;
1029         INIT_LIST_HEAD(&new->list);
1030         new->task = NULL;
1031         return new;
1032 }
1033
1034 void nfs_release_seqid(struct nfs_seqid *seqid)
1035 {
1036         struct nfs_seqid_counter *sequence;
1037
1038         if (seqid == NULL || list_empty(&seqid->list))
1039                 return;
1040         sequence = seqid->sequence;
1041         spin_lock(&sequence->lock);
1042         list_del_init(&seqid->list);
1043         if (!list_empty(&sequence->list)) {
1044                 struct nfs_seqid *next;
1045
1046                 next = list_first_entry(&sequence->list,
1047                                 struct nfs_seqid, list);
1048                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1049         }
1050         spin_unlock(&sequence->lock);
1051 }
1052
1053 void nfs_free_seqid(struct nfs_seqid *seqid)
1054 {
1055         nfs_release_seqid(seqid);
1056         kfree(seqid);
1057 }
1058
1059 /*
1060  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1061  * failed with a seqid incrementing error -
1062  * see comments nfs4.h:seqid_mutating_error()
1063  */
1064 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1065 {
1066         switch (status) {
1067                 case 0:
1068                         break;
1069                 case -NFS4ERR_BAD_SEQID:
1070                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1071                                 return;
1072                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1073                                         " sequence-id error on an"
1074                                         " unconfirmed sequence %p!\n",
1075                                         seqid->sequence);
1076                 case -NFS4ERR_STALE_CLIENTID:
1077                 case -NFS4ERR_STALE_STATEID:
1078                 case -NFS4ERR_BAD_STATEID:
1079                 case -NFS4ERR_BADXDR:
1080                 case -NFS4ERR_RESOURCE:
1081                 case -NFS4ERR_NOFILEHANDLE:
1082                         /* Non-seqid mutating errors */
1083                         return;
1084         };
1085         /*
1086          * Note: no locking needed as we are guaranteed to be first
1087          * on the sequence list
1088          */
1089         seqid->sequence->counter++;
1090 }
1091
1092 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1093 {
1094         struct nfs4_state_owner *sp;
1095
1096         if (seqid == NULL)
1097                 return;
1098
1099         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1100         if (status == -NFS4ERR_BAD_SEQID)
1101                 nfs4_drop_state_owner(sp);
1102         if (!nfs4_has_session(sp->so_server->nfs_client))
1103                 nfs_increment_seqid(status, seqid);
1104 }
1105
1106 /*
1107  * Increment the seqid if the LOCK/LOCKU succeeded, or
1108  * failed with a seqid incrementing error -
1109  * see comments nfs4.h:seqid_mutating_error()
1110  */
1111 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1112 {
1113         if (seqid != NULL)
1114                 nfs_increment_seqid(status, seqid);
1115 }
1116
1117 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1118 {
1119         struct nfs_seqid_counter *sequence;
1120         int status = 0;
1121
1122         if (seqid == NULL)
1123                 goto out;
1124         sequence = seqid->sequence;
1125         spin_lock(&sequence->lock);
1126         seqid->task = task;
1127         if (list_empty(&seqid->list))
1128                 list_add_tail(&seqid->list, &sequence->list);
1129         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1130                 goto unlock;
1131         rpc_sleep_on(&sequence->wait, task, NULL);
1132         status = -EAGAIN;
1133 unlock:
1134         spin_unlock(&sequence->lock);
1135 out:
1136         return status;
1137 }
1138
1139 static int nfs4_run_state_manager(void *);
1140
1141 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1142 {
1143         smp_mb__before_atomic();
1144         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1145         smp_mb__after_atomic();
1146         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1147         rpc_wake_up(&clp->cl_rpcwaitq);
1148 }
1149
1150 /*
1151  * Schedule the nfs_client asynchronous state management routine
1152  */
1153 void nfs4_schedule_state_manager(struct nfs_client *clp)
1154 {
1155         struct task_struct *task;
1156         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1157
1158         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1159                 return;
1160         __module_get(THIS_MODULE);
1161         atomic_inc(&clp->cl_count);
1162
1163         /* The rcu_read_lock() is not strictly necessary, as the state
1164          * manager is the only thread that ever changes the rpc_xprt
1165          * after it's initialized.  At this point, we're single threaded. */
1166         rcu_read_lock();
1167         snprintf(buf, sizeof(buf), "%s-manager",
1168                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1169         rcu_read_unlock();
1170         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1171         if (IS_ERR(task)) {
1172                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1173                         __func__, PTR_ERR(task));
1174                 nfs4_clear_state_manager_bit(clp);
1175                 nfs_put_client(clp);
1176                 module_put(THIS_MODULE);
1177         }
1178 }
1179
1180 /*
1181  * Schedule a lease recovery attempt
1182  */
1183 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1184 {
1185         if (!clp)
1186                 return;
1187         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1188                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1189         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1190                         clp->cl_hostname);
1191         nfs4_schedule_state_manager(clp);
1192 }
1193 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1194
1195 /**
1196  * nfs4_schedule_migration_recovery - trigger migration recovery
1197  *
1198  * @server: FSID that is migrating
1199  *
1200  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1201  * value is returned.
1202  */
1203 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1204 {
1205         struct nfs_client *clp = server->nfs_client;
1206
1207         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1208                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1209                                 clp->cl_hostname);
1210                 return -NFS4ERR_IO;
1211         }
1212
1213         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1214                 return -NFS4ERR_IO;
1215
1216         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1217                         __func__,
1218                         (unsigned long long)server->fsid.major,
1219                         (unsigned long long)server->fsid.minor,
1220                         clp->cl_hostname);
1221
1222         set_bit(NFS_MIG_IN_TRANSITION,
1223                         &((struct nfs_server *)server)->mig_status);
1224         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1225
1226         nfs4_schedule_state_manager(clp);
1227         return 0;
1228 }
1229 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1230
1231 /**
1232  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1233  *
1234  * @clp: server to check for moved leases
1235  *
1236  */
1237 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1238 {
1239         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1240                 __func__, clp->cl_clientid, clp->cl_hostname);
1241
1242         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1243         nfs4_schedule_state_manager(clp);
1244 }
1245 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1246
1247 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1248 {
1249         int res;
1250
1251         might_sleep();
1252
1253         atomic_inc(&clp->cl_count);
1254         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1255                                  nfs_wait_bit_killable, TASK_KILLABLE);
1256         if (res)
1257                 goto out;
1258         if (clp->cl_cons_state < 0)
1259                 res = clp->cl_cons_state;
1260 out:
1261         nfs_put_client(clp);
1262         return res;
1263 }
1264
1265 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1266 {
1267         unsigned int loop;
1268         int ret;
1269
1270         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1271                 ret = nfs4_wait_clnt_recover(clp);
1272                 if (ret != 0)
1273                         break;
1274                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1275                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1276                         break;
1277                 nfs4_schedule_state_manager(clp);
1278                 ret = -EIO;
1279         }
1280         return ret;
1281 }
1282
1283 /*
1284  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1285  * @clp: client to process
1286  *
1287  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1288  * resend of the SETCLIENTID and hence re-establish the
1289  * callback channel. Then return all existing delegations.
1290  */
1291 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1292 {
1293         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1294         nfs_expire_all_delegations(clp);
1295         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1296                         clp->cl_hostname);
1297 }
1298
1299 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1300 {
1301         nfs40_handle_cb_pathdown(clp);
1302         nfs4_schedule_state_manager(clp);
1303 }
1304
1305 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1306 {
1307
1308         if (!nfs4_valid_open_stateid(state))
1309                 return 0;
1310         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1311         /* Don't recover state that expired before the reboot */
1312         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1313                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1314                 return 0;
1315         }
1316         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1317         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1318         return 1;
1319 }
1320
1321 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1322 {
1323         if (!nfs4_valid_open_stateid(state))
1324                 return 0;
1325         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1326         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1327         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1328         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1329         return 1;
1330 }
1331
1332 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1333 {
1334         struct nfs_client *clp = server->nfs_client;
1335
1336         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1337                 return -EBADF;
1338         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1339                         clp->cl_hostname);
1340         nfs4_schedule_state_manager(clp);
1341         return 0;
1342 }
1343 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1344
1345 static struct nfs4_lock_state *
1346 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1347                 const nfs4_stateid *stateid)
1348 {
1349         struct nfs4_lock_state *pos;
1350
1351         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1352                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1353                         continue;
1354                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1355                         return pos;
1356         }
1357         return NULL;
1358 }
1359
1360 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1361                 const nfs4_stateid *stateid)
1362 {
1363         bool found = false;
1364
1365         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1366                 spin_lock(&state->state_lock);
1367                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1368                         found = true;
1369                 spin_unlock(&state->state_lock);
1370         }
1371         return found;
1372 }
1373
1374 void nfs_inode_find_state_and_recover(struct inode *inode,
1375                 const nfs4_stateid *stateid)
1376 {
1377         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1378         struct nfs_inode *nfsi = NFS_I(inode);
1379         struct nfs_open_context *ctx;
1380         struct nfs4_state *state;
1381         bool found = false;
1382
1383         spin_lock(&inode->i_lock);
1384         list_for_each_entry(ctx, &nfsi->open_files, list) {
1385                 state = ctx->state;
1386                 if (state == NULL)
1387                         continue;
1388                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1389                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1390                         found = true;
1391                         continue;
1392                 }
1393                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1394                     nfs4_state_mark_reclaim_nograce(clp, state))
1395                         found = true;
1396         }
1397         spin_unlock(&inode->i_lock);
1398
1399         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1400         if (found)
1401                 nfs4_schedule_state_manager(clp);
1402 }
1403
1404 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1405 {
1406         struct inode *inode = state->inode;
1407         struct nfs_inode *nfsi = NFS_I(inode);
1408         struct nfs_open_context *ctx;
1409
1410         spin_lock(&inode->i_lock);
1411         list_for_each_entry(ctx, &nfsi->open_files, list) {
1412                 if (ctx->state != state)
1413                         continue;
1414                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1415         }
1416         spin_unlock(&inode->i_lock);
1417 }
1418
1419 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1420 {
1421         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1422         nfs4_state_mark_open_context_bad(state);
1423 }
1424
1425
1426 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1427 {
1428         struct inode *inode = state->inode;
1429         struct nfs_inode *nfsi = NFS_I(inode);
1430         struct file_lock *fl;
1431         int status = 0;
1432         struct file_lock_context *flctx = inode->i_flctx;
1433         struct list_head *list;
1434
1435         if (flctx == NULL)
1436                 return 0;
1437
1438         list = &flctx->flc_posix;
1439
1440         /* Guard against delegation returns and new lock/unlock calls */
1441         down_write(&nfsi->rwsem);
1442         spin_lock(&flctx->flc_lock);
1443 restart:
1444         list_for_each_entry(fl, list, fl_list) {
1445                 if (nfs_file_open_context(fl->fl_file)->state != state)
1446                         continue;
1447                 spin_unlock(&flctx->flc_lock);
1448                 status = ops->recover_lock(state, fl);
1449                 switch (status) {
1450                 case 0:
1451                         break;
1452                 case -ESTALE:
1453                 case -NFS4ERR_ADMIN_REVOKED:
1454                 case -NFS4ERR_STALE_STATEID:
1455                 case -NFS4ERR_BAD_STATEID:
1456                 case -NFS4ERR_EXPIRED:
1457                 case -NFS4ERR_NO_GRACE:
1458                 case -NFS4ERR_STALE_CLIENTID:
1459                 case -NFS4ERR_BADSESSION:
1460                 case -NFS4ERR_BADSLOT:
1461                 case -NFS4ERR_BAD_HIGH_SLOT:
1462                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1463                         goto out;
1464                 default:
1465                         pr_err("NFS: %s: unhandled error %d\n",
1466                                         __func__, status);
1467                 case -ENOMEM:
1468                 case -NFS4ERR_DENIED:
1469                 case -NFS4ERR_RECLAIM_BAD:
1470                 case -NFS4ERR_RECLAIM_CONFLICT:
1471                         /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1472                         status = 0;
1473                 }
1474                 spin_lock(&flctx->flc_lock);
1475         }
1476         if (list == &flctx->flc_posix) {
1477                 list = &flctx->flc_flock;
1478                 goto restart;
1479         }
1480         spin_unlock(&flctx->flc_lock);
1481 out:
1482         up_write(&nfsi->rwsem);
1483         return status;
1484 }
1485
1486 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1487 {
1488         struct nfs4_state *state;
1489         struct nfs4_lock_state *lock;
1490         int status = 0;
1491
1492         /* Note: we rely on the sp->so_states list being ordered
1493          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1494          * states first.
1495          * This is needed to ensure that the server won't give us any
1496          * read delegations that we have to return if, say, we are
1497          * recovering after a network partition or a reboot from a
1498          * server that doesn't support a grace period.
1499          */
1500         spin_lock(&sp->so_lock);
1501         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1502 restart:
1503         list_for_each_entry(state, &sp->so_states, open_states) {
1504                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1505                         continue;
1506                 if (!nfs4_valid_open_stateid(state))
1507                         continue;
1508                 if (state->state == 0)
1509                         continue;
1510                 atomic_inc(&state->count);
1511                 spin_unlock(&sp->so_lock);
1512                 status = ops->recover_open(sp, state);
1513                 if (status >= 0) {
1514                         status = nfs4_reclaim_locks(state, ops);
1515                         if (status >= 0) {
1516                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1517                                         spin_lock(&state->state_lock);
1518                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1519                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1520                                                         pr_warn_ratelimited("NFS: "
1521                                                                             "%s: Lock reclaim "
1522                                                                             "failed!\n", __func__);
1523                                         }
1524                                         spin_unlock(&state->state_lock);
1525                                 }
1526                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1527                                         &state->flags);
1528                                 nfs4_put_open_state(state);
1529                                 spin_lock(&sp->so_lock);
1530                                 goto restart;
1531                         }
1532                 }
1533                 switch (status) {
1534                         default:
1535                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1536                                         __func__, status);
1537                         case -ENOENT:
1538                         case -ENOMEM:
1539                         case -EACCES:
1540                         case -EROFS:
1541                         case -EIO:
1542                         case -ESTALE:
1543                                 /* Open state on this file cannot be recovered */
1544                                 nfs4_state_mark_recovery_failed(state, status);
1545                                 break;
1546                         case -EAGAIN:
1547                                 ssleep(1);
1548                         case -NFS4ERR_ADMIN_REVOKED:
1549                         case -NFS4ERR_STALE_STATEID:
1550                         case -NFS4ERR_BAD_STATEID:
1551                         case -NFS4ERR_RECLAIM_BAD:
1552                         case -NFS4ERR_RECLAIM_CONFLICT:
1553                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1554                                 break;
1555                         case -NFS4ERR_EXPIRED:
1556                         case -NFS4ERR_NO_GRACE:
1557                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1558                         case -NFS4ERR_STALE_CLIENTID:
1559                         case -NFS4ERR_BADSESSION:
1560                         case -NFS4ERR_BADSLOT:
1561                         case -NFS4ERR_BAD_HIGH_SLOT:
1562                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1563                                 goto out_err;
1564                 }
1565                 nfs4_put_open_state(state);
1566                 spin_lock(&sp->so_lock);
1567                 goto restart;
1568         }
1569         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1570         spin_unlock(&sp->so_lock);
1571         return 0;
1572 out_err:
1573         nfs4_put_open_state(state);
1574         spin_lock(&sp->so_lock);
1575         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1576         spin_unlock(&sp->so_lock);
1577         return status;
1578 }
1579
1580 static void nfs4_clear_open_state(struct nfs4_state *state)
1581 {
1582         struct nfs4_lock_state *lock;
1583
1584         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1585         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1586         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1587         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1588         spin_lock(&state->state_lock);
1589         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1590                 lock->ls_seqid.flags = 0;
1591                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1592         }
1593         spin_unlock(&state->state_lock);
1594 }
1595
1596 static void nfs4_reset_seqids(struct nfs_server *server,
1597         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1598 {
1599         struct nfs_client *clp = server->nfs_client;
1600         struct nfs4_state_owner *sp;
1601         struct rb_node *pos;
1602         struct nfs4_state *state;
1603
1604         spin_lock(&clp->cl_lock);
1605         for (pos = rb_first(&server->state_owners);
1606              pos != NULL;
1607              pos = rb_next(pos)) {
1608                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1609                 sp->so_seqid.flags = 0;
1610                 spin_lock(&sp->so_lock);
1611                 list_for_each_entry(state, &sp->so_states, open_states) {
1612                         if (mark_reclaim(clp, state))
1613                                 nfs4_clear_open_state(state);
1614                 }
1615                 spin_unlock(&sp->so_lock);
1616         }
1617         spin_unlock(&clp->cl_lock);
1618 }
1619
1620 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1621         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1622 {
1623         struct nfs_server *server;
1624
1625         rcu_read_lock();
1626         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1627                 nfs4_reset_seqids(server, mark_reclaim);
1628         rcu_read_unlock();
1629 }
1630
1631 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1632 {
1633         /* Mark all delegations for reclaim */
1634         nfs_delegation_mark_reclaim(clp);
1635         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1636 }
1637
1638 static void nfs4_reclaim_complete(struct nfs_client *clp,
1639                                  const struct nfs4_state_recovery_ops *ops,
1640                                  struct rpc_cred *cred)
1641 {
1642         /* Notify the server we're done reclaiming our state */
1643         if (ops->reclaim_complete)
1644                 (void)ops->reclaim_complete(clp, cred);
1645 }
1646
1647 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1648 {
1649         struct nfs_client *clp = server->nfs_client;
1650         struct nfs4_state_owner *sp;
1651         struct rb_node *pos;
1652         struct nfs4_state *state;
1653
1654         spin_lock(&clp->cl_lock);
1655         for (pos = rb_first(&server->state_owners);
1656              pos != NULL;
1657              pos = rb_next(pos)) {
1658                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1659                 spin_lock(&sp->so_lock);
1660                 list_for_each_entry(state, &sp->so_states, open_states) {
1661                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1662                                                 &state->flags))
1663                                 continue;
1664                         nfs4_state_mark_reclaim_nograce(clp, state);
1665                 }
1666                 spin_unlock(&sp->so_lock);
1667         }
1668         spin_unlock(&clp->cl_lock);
1669 }
1670
1671 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1672 {
1673         struct nfs_server *server;
1674
1675         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1676                 return 0;
1677
1678         rcu_read_lock();
1679         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1680                 nfs4_clear_reclaim_server(server);
1681         rcu_read_unlock();
1682
1683         nfs_delegation_reap_unclaimed(clp);
1684         return 1;
1685 }
1686
1687 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1688 {
1689         const struct nfs4_state_recovery_ops *ops;
1690         struct rpc_cred *cred;
1691
1692         if (!nfs4_state_clear_reclaim_reboot(clp))
1693                 return;
1694         ops = clp->cl_mvops->reboot_recovery_ops;
1695         cred = nfs4_get_clid_cred(clp);
1696         nfs4_reclaim_complete(clp, ops, cred);
1697         put_rpccred(cred);
1698 }
1699
1700 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1701 {
1702         nfs_mark_test_expired_all_delegations(clp);
1703         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1704 }
1705
1706 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1707 {
1708         switch (error) {
1709                 case 0:
1710                         break;
1711                 case -NFS4ERR_CB_PATH_DOWN:
1712                         nfs40_handle_cb_pathdown(clp);
1713                         break;
1714                 case -NFS4ERR_NO_GRACE:
1715                         nfs4_state_end_reclaim_reboot(clp);
1716                         break;
1717                 case -NFS4ERR_STALE_CLIENTID:
1718                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1719                         nfs4_state_clear_reclaim_reboot(clp);
1720                         nfs4_state_start_reclaim_reboot(clp);
1721                         break;
1722                 case -NFS4ERR_EXPIRED:
1723                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1724                         nfs4_state_start_reclaim_nograce(clp);
1725                         break;
1726                 case -NFS4ERR_BADSESSION:
1727                 case -NFS4ERR_BADSLOT:
1728                 case -NFS4ERR_BAD_HIGH_SLOT:
1729                 case -NFS4ERR_DEADSESSION:
1730                 case -NFS4ERR_SEQ_FALSE_RETRY:
1731                 case -NFS4ERR_SEQ_MISORDERED:
1732                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1733                         /* Zero session reset errors */
1734                         break;
1735                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1736                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1737                         break;
1738                 default:
1739                         dprintk("%s: failed to handle error %d for server %s\n",
1740                                         __func__, error, clp->cl_hostname);
1741                         return error;
1742         }
1743         dprintk("%s: handled error %d for server %s\n", __func__, error,
1744                         clp->cl_hostname);
1745         return 0;
1746 }
1747
1748 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1749 {
1750         struct nfs4_state_owner *sp;
1751         struct nfs_server *server;
1752         struct rb_node *pos;
1753         int status = 0;
1754
1755 restart:
1756         rcu_read_lock();
1757         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1758                 nfs4_purge_state_owners(server);
1759                 spin_lock(&clp->cl_lock);
1760                 for (pos = rb_first(&server->state_owners);
1761                      pos != NULL;
1762                      pos = rb_next(pos)) {
1763                         sp = rb_entry(pos,
1764                                 struct nfs4_state_owner, so_server_node);
1765                         if (!test_and_clear_bit(ops->owner_flag_bit,
1766                                                         &sp->so_flags))
1767                                 continue;
1768                         if (!atomic_inc_not_zero(&sp->so_count))
1769                                 continue;
1770                         spin_unlock(&clp->cl_lock);
1771                         rcu_read_unlock();
1772
1773                         status = nfs4_reclaim_open_state(sp, ops);
1774                         if (status < 0) {
1775                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1776                                 nfs4_put_state_owner(sp);
1777                                 status = nfs4_recovery_handle_error(clp, status);
1778                                 return (status != 0) ? status : -EAGAIN;
1779                         }
1780
1781                         nfs4_put_state_owner(sp);
1782                         goto restart;
1783                 }
1784                 spin_unlock(&clp->cl_lock);
1785         }
1786         rcu_read_unlock();
1787         return 0;
1788 }
1789
1790 static int nfs4_check_lease(struct nfs_client *clp)
1791 {
1792         struct rpc_cred *cred;
1793         const struct nfs4_state_maintenance_ops *ops =
1794                 clp->cl_mvops->state_renewal_ops;
1795         int status;
1796
1797         /* Is the client already known to have an expired lease? */
1798         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1799                 return 0;
1800         spin_lock(&clp->cl_lock);
1801         cred = ops->get_state_renewal_cred_locked(clp);
1802         spin_unlock(&clp->cl_lock);
1803         if (cred == NULL) {
1804                 cred = nfs4_get_clid_cred(clp);
1805                 status = -ENOKEY;
1806                 if (cred == NULL)
1807                         goto out;
1808         }
1809         status = ops->renew_lease(clp, cred);
1810         put_rpccred(cred);
1811         if (status == -ETIMEDOUT) {
1812                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1813                 return 0;
1814         }
1815 out:
1816         return nfs4_recovery_handle_error(clp, status);
1817 }
1818
1819 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1820  * and for recoverable errors on EXCHANGE_ID for v4.1
1821  */
1822 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1823 {
1824         switch (status) {
1825         case -NFS4ERR_SEQ_MISORDERED:
1826                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1827                         return -ESERVERFAULT;
1828                 /* Lease confirmation error: retry after purging the lease */
1829                 ssleep(1);
1830                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1831                 break;
1832         case -NFS4ERR_STALE_CLIENTID:
1833                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1834                 nfs4_state_start_reclaim_reboot(clp);
1835                 break;
1836         case -NFS4ERR_CLID_INUSE:
1837                 pr_err("NFS: Server %s reports our clientid is in use\n",
1838                         clp->cl_hostname);
1839                 nfs_mark_client_ready(clp, -EPERM);
1840                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1841                 return -EPERM;
1842         case -EACCES:
1843         case -NFS4ERR_DELAY:
1844         case -ETIMEDOUT:
1845         case -EAGAIN:
1846                 ssleep(1);
1847                 break;
1848
1849         case -NFS4ERR_MINOR_VERS_MISMATCH:
1850                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1851                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1852                 dprintk("%s: exit with error %d for server %s\n",
1853                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1854                 return -EPROTONOSUPPORT;
1855         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1856                                  * in nfs4_exchange_id */
1857         default:
1858                 dprintk("%s: exit with error %d for server %s\n", __func__,
1859                                 status, clp->cl_hostname);
1860                 return status;
1861         }
1862         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1863         dprintk("%s: handled error %d for server %s\n", __func__, status,
1864                         clp->cl_hostname);
1865         return 0;
1866 }
1867
1868 static int nfs4_establish_lease(struct nfs_client *clp)
1869 {
1870         struct rpc_cred *cred;
1871         const struct nfs4_state_recovery_ops *ops =
1872                 clp->cl_mvops->reboot_recovery_ops;
1873         int status;
1874
1875         nfs4_begin_drain_session(clp);
1876         cred = nfs4_get_clid_cred(clp);
1877         if (cred == NULL)
1878                 return -ENOENT;
1879         status = ops->establish_clid(clp, cred);
1880         put_rpccred(cred);
1881         if (status != 0)
1882                 return status;
1883         pnfs_destroy_all_layouts(clp);
1884         return 0;
1885 }
1886
1887 /*
1888  * Returns zero or a negative errno.  NFS4ERR values are converted
1889  * to local errno values.
1890  */
1891 static int nfs4_reclaim_lease(struct nfs_client *clp)
1892 {
1893         int status;
1894
1895         status = nfs4_establish_lease(clp);
1896         if (status < 0)
1897                 return nfs4_handle_reclaim_lease_error(clp, status);
1898         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1899                 nfs4_state_start_reclaim_nograce(clp);
1900         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1901                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1902         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1903         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1904         return 0;
1905 }
1906
1907 static int nfs4_purge_lease(struct nfs_client *clp)
1908 {
1909         int status;
1910
1911         status = nfs4_establish_lease(clp);
1912         if (status < 0)
1913                 return nfs4_handle_reclaim_lease_error(clp, status);
1914         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1915         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1916         nfs4_state_start_reclaim_nograce(clp);
1917         return 0;
1918 }
1919
1920 /*
1921  * Try remote migration of one FSID from a source server to a
1922  * destination server.  The source server provides a list of
1923  * potential destinations.
1924  *
1925  * Returns zero or a negative NFS4ERR status code.
1926  */
1927 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1928 {
1929         struct nfs_client *clp = server->nfs_client;
1930         struct nfs4_fs_locations *locations = NULL;
1931         struct inode *inode;
1932         struct page *page;
1933         int status, result;
1934
1935         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1936                         (unsigned long long)server->fsid.major,
1937                         (unsigned long long)server->fsid.minor,
1938                         clp->cl_hostname);
1939
1940         result = 0;
1941         page = alloc_page(GFP_KERNEL);
1942         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1943         if (page == NULL || locations == NULL) {
1944                 dprintk("<-- %s: no memory\n", __func__);
1945                 goto out;
1946         }
1947
1948         inode = d_inode(server->super->s_root);
1949         result = nfs4_proc_get_locations(inode, locations, page, cred);
1950         if (result) {
1951                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1952                         __func__, result);
1953                 goto out;
1954         }
1955
1956         result = -NFS4ERR_NXIO;
1957         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1958                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1959                         __func__);
1960                 goto out;
1961         }
1962
1963         nfs4_begin_drain_session(clp);
1964
1965         status = nfs4_replace_transport(server, locations);
1966         if (status != 0) {
1967                 dprintk("<-- %s: failed to replace transport: %d\n",
1968                         __func__, status);
1969                 goto out;
1970         }
1971
1972         result = 0;
1973         dprintk("<-- %s: migration succeeded\n", __func__);
1974
1975 out:
1976         if (page != NULL)
1977                 __free_page(page);
1978         kfree(locations);
1979         if (result) {
1980                 pr_err("NFS: migration recovery failed (server %s)\n",
1981                                 clp->cl_hostname);
1982                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1983         }
1984         return result;
1985 }
1986
1987 /*
1988  * Returns zero or a negative NFS4ERR status code.
1989  */
1990 static int nfs4_handle_migration(struct nfs_client *clp)
1991 {
1992         const struct nfs4_state_maintenance_ops *ops =
1993                                 clp->cl_mvops->state_renewal_ops;
1994         struct nfs_server *server;
1995         struct rpc_cred *cred;
1996
1997         dprintk("%s: migration reported on \"%s\"\n", __func__,
1998                         clp->cl_hostname);
1999
2000         spin_lock(&clp->cl_lock);
2001         cred = ops->get_state_renewal_cred_locked(clp);
2002         spin_unlock(&clp->cl_lock);
2003         if (cred == NULL)
2004                 return -NFS4ERR_NOENT;
2005
2006         clp->cl_mig_gen++;
2007 restart:
2008         rcu_read_lock();
2009         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2010                 int status;
2011
2012                 if (server->mig_gen == clp->cl_mig_gen)
2013                         continue;
2014                 server->mig_gen = clp->cl_mig_gen;
2015
2016                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2017                                                 &server->mig_status))
2018                         continue;
2019
2020                 rcu_read_unlock();
2021                 status = nfs4_try_migration(server, cred);
2022                 if (status < 0) {
2023                         put_rpccred(cred);
2024                         return status;
2025                 }
2026                 goto restart;
2027         }
2028         rcu_read_unlock();
2029         put_rpccred(cred);
2030         return 0;
2031 }
2032
2033 /*
2034  * Test each nfs_server on the clp's cl_superblocks list to see
2035  * if it's moved to another server.  Stop when the server no longer
2036  * returns NFS4ERR_LEASE_MOVED.
2037  */
2038 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2039 {
2040         const struct nfs4_state_maintenance_ops *ops =
2041                                 clp->cl_mvops->state_renewal_ops;
2042         struct nfs_server *server;
2043         struct rpc_cred *cred;
2044
2045         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2046                         clp->cl_hostname);
2047
2048         spin_lock(&clp->cl_lock);
2049         cred = ops->get_state_renewal_cred_locked(clp);
2050         spin_unlock(&clp->cl_lock);
2051         if (cred == NULL)
2052                 return -NFS4ERR_NOENT;
2053
2054         clp->cl_mig_gen++;
2055 restart:
2056         rcu_read_lock();
2057         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2058                 struct inode *inode;
2059                 int status;
2060
2061                 if (server->mig_gen == clp->cl_mig_gen)
2062                         continue;
2063                 server->mig_gen = clp->cl_mig_gen;
2064
2065                 rcu_read_unlock();
2066
2067                 inode = d_inode(server->super->s_root);
2068                 status = nfs4_proc_fsid_present(inode, cred);
2069                 if (status != -NFS4ERR_MOVED)
2070                         goto restart;   /* wasn't this one */
2071                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2072                         goto restart;   /* there are more */
2073                 goto out;
2074         }
2075         rcu_read_unlock();
2076
2077 out:
2078         put_rpccred(cred);
2079         return 0;
2080 }
2081
2082 /**
2083  * nfs4_discover_server_trunking - Detect server IP address trunking
2084  *
2085  * @clp: nfs_client under test
2086  * @result: OUT: found nfs_client, or clp
2087  *
2088  * Returns zero or a negative errno.  If zero is returned,
2089  * an nfs_client pointer is planted in "result".
2090  *
2091  * Note: since we are invoked in process context, and
2092  * not from inside the state manager, we cannot use
2093  * nfs4_handle_reclaim_lease_error().
2094  */
2095 int nfs4_discover_server_trunking(struct nfs_client *clp,
2096                                   struct nfs_client **result)
2097 {
2098         const struct nfs4_state_recovery_ops *ops =
2099                                 clp->cl_mvops->reboot_recovery_ops;
2100         struct rpc_clnt *clnt;
2101         struct rpc_cred *cred;
2102         int i, status;
2103
2104         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2105
2106         clnt = clp->cl_rpcclient;
2107         i = 0;
2108
2109         mutex_lock(&nfs_clid_init_mutex);
2110 again:
2111         status  = -ENOENT;
2112         cred = nfs4_get_clid_cred(clp);
2113         if (cred == NULL)
2114                 goto out_unlock;
2115
2116         status = ops->detect_trunking(clp, result, cred);
2117         put_rpccred(cred);
2118         switch (status) {
2119         case 0:
2120                 break;
2121         case -ETIMEDOUT:
2122                 if (clnt->cl_softrtry)
2123                         break;
2124         case -NFS4ERR_DELAY:
2125         case -EAGAIN:
2126                 ssleep(1);
2127         case -NFS4ERR_STALE_CLIENTID:
2128                 dprintk("NFS: %s after status %d, retrying\n",
2129                         __func__, status);
2130                 goto again;
2131         case -EACCES:
2132                 if (i++ == 0) {
2133                         nfs4_root_machine_cred(clp);
2134                         goto again;
2135                 }
2136                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2137                         break;
2138         case -NFS4ERR_CLID_INUSE:
2139         case -NFS4ERR_WRONGSEC:
2140                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2141                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2142                         status = -EPERM;
2143                         break;
2144                 }
2145                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2146                 if (IS_ERR(clnt)) {
2147                         status = PTR_ERR(clnt);
2148                         break;
2149                 }
2150                 /* Note: this is safe because we haven't yet marked the
2151                  * client as ready, so we are the only user of
2152                  * clp->cl_rpcclient
2153                  */
2154                 clnt = xchg(&clp->cl_rpcclient, clnt);
2155                 rpc_shutdown_client(clnt);
2156                 clnt = clp->cl_rpcclient;
2157                 goto again;
2158
2159         case -NFS4ERR_MINOR_VERS_MISMATCH:
2160                 status = -EPROTONOSUPPORT;
2161                 break;
2162
2163         case -EKEYEXPIRED:
2164         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2165                                  * in nfs4_exchange_id */
2166                 status = -EKEYEXPIRED;
2167                 break;
2168         default:
2169                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2170                                 __func__, status);
2171                 status = -EIO;
2172         }
2173
2174 out_unlock:
2175         mutex_unlock(&nfs_clid_init_mutex);
2176         dprintk("NFS: %s: status = %d\n", __func__, status);
2177         return status;
2178 }
2179
2180 #ifdef CONFIG_NFS_V4_1
2181 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2182 {
2183         struct nfs_client *clp = session->clp;
2184
2185         switch (err) {
2186         default:
2187                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2188                 break;
2189         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2190                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2191         }
2192         nfs4_schedule_lease_recovery(clp);
2193 }
2194 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2195
2196 void nfs41_notify_server(struct nfs_client *clp)
2197 {
2198         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2199         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2200         nfs4_schedule_state_manager(clp);
2201 }
2202
2203 static void nfs4_reset_all_state(struct nfs_client *clp)
2204 {
2205         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2206                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2207                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2208                 nfs4_state_start_reclaim_nograce(clp);
2209                 dprintk("%s: scheduling reset of all state for server %s!\n",
2210                                 __func__, clp->cl_hostname);
2211                 nfs4_schedule_state_manager(clp);
2212         }
2213 }
2214
2215 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2216 {
2217         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2218                 nfs4_state_start_reclaim_reboot(clp);
2219                 dprintk("%s: server %s rebooted!\n", __func__,
2220                                 clp->cl_hostname);
2221                 nfs4_schedule_state_manager(clp);
2222         }
2223 }
2224
2225 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2226 {
2227         nfs4_reset_all_state(clp);
2228         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2229 }
2230
2231 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2232 {
2233         nfs4_state_start_reclaim_nograce(clp);
2234         nfs4_schedule_state_manager(clp);
2235
2236         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2237 }
2238
2239 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2240 {
2241         /* FIXME: For now, we destroy all layouts. */
2242         pnfs_destroy_all_layouts(clp);
2243         /* FIXME: For now, we test all delegations+open state+locks. */
2244         nfs41_handle_some_state_revoked(clp);
2245         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2246                         clp->cl_hostname);
2247 }
2248
2249 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2250 {
2251         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2252         nfs4_schedule_state_manager(clp);
2253
2254         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2255                         clp->cl_hostname);
2256 }
2257
2258 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2259 {
2260         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2261                 &clp->cl_state) == 0)
2262                 nfs4_schedule_state_manager(clp);
2263 }
2264
2265 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2266                 bool recovery)
2267 {
2268         if (!flags)
2269                 return;
2270
2271         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2272                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2273         /*
2274          * If we're called from the state manager thread, then assume we're
2275          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2276          * Those flags are expected to remain set until we're done
2277          * recovering (see RFC5661, section 18.46.3).
2278          */
2279         if (recovery)
2280                 goto out_recovery;
2281
2282         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2283                 nfs41_handle_server_reboot(clp);
2284         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2285                 nfs41_handle_all_state_revoked(clp);
2286         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2287                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2288                 nfs41_handle_some_state_revoked(clp);
2289         if (flags & SEQ4_STATUS_LEASE_MOVED)
2290                 nfs4_schedule_lease_moved_recovery(clp);
2291         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2292                 nfs41_handle_recallable_state_revoked(clp);
2293 out_recovery:
2294         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2295                 nfs41_handle_backchannel_fault(clp);
2296         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2297                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2298                 nfs41_handle_cb_path_down(clp);
2299 }
2300
2301 static int nfs4_reset_session(struct nfs_client *clp)
2302 {
2303         struct rpc_cred *cred;
2304         int status;
2305
2306         if (!nfs4_has_session(clp))
2307                 return 0;
2308         nfs4_begin_drain_session(clp);
2309         cred = nfs4_get_clid_cred(clp);
2310         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2311         switch (status) {
2312         case 0:
2313         case -NFS4ERR_BADSESSION:
2314         case -NFS4ERR_DEADSESSION:
2315                 break;
2316         case -NFS4ERR_BACK_CHAN_BUSY:
2317         case -NFS4ERR_DELAY:
2318                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2319                 status = 0;
2320                 ssleep(1);
2321                 goto out;
2322         default:
2323                 status = nfs4_recovery_handle_error(clp, status);
2324                 goto out;
2325         }
2326
2327         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2328         status = nfs4_proc_create_session(clp, cred);
2329         if (status) {
2330                 dprintk("%s: session reset failed with status %d for server %s!\n",
2331                         __func__, status, clp->cl_hostname);
2332                 status = nfs4_handle_reclaim_lease_error(clp, status);
2333                 goto out;
2334         }
2335         nfs41_finish_session_reset(clp);
2336         dprintk("%s: session reset was successful for server %s!\n",
2337                         __func__, clp->cl_hostname);
2338 out:
2339         if (cred)
2340                 put_rpccred(cred);
2341         return status;
2342 }
2343
2344 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2345 {
2346         struct rpc_cred *cred;
2347         int ret;
2348
2349         if (!nfs4_has_session(clp))
2350                 return 0;
2351         nfs4_begin_drain_session(clp);
2352         cred = nfs4_get_clid_cred(clp);
2353         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2354         if (cred)
2355                 put_rpccred(cred);
2356         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2357         switch (ret) {
2358         case 0:
2359                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2360                         __func__, clp->cl_hostname);
2361                 break;
2362         case -NFS4ERR_DELAY:
2363                 ssleep(1);
2364                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2365                 break;
2366         default:
2367                 return nfs4_recovery_handle_error(clp, ret);
2368         }
2369         return 0;
2370 }
2371 #else /* CONFIG_NFS_V4_1 */
2372 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2373
2374 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2375 {
2376         return 0;
2377 }
2378 #endif /* CONFIG_NFS_V4_1 */
2379
2380 static void nfs4_state_manager(struct nfs_client *clp)
2381 {
2382         int status = 0;
2383         const char *section = "", *section_sep = "";
2384
2385         /* Ensure exclusive access to NFSv4 state */
2386         do {
2387                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2388                         section = "purge state";
2389                         status = nfs4_purge_lease(clp);
2390                         if (status < 0)
2391                                 goto out_error;
2392                         continue;
2393                 }
2394
2395                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2396                         section = "lease expired";
2397                         /* We're going to have to re-establish a clientid */
2398                         status = nfs4_reclaim_lease(clp);
2399                         if (status < 0)
2400                                 goto out_error;
2401                         continue;
2402                 }
2403
2404                 /* Initialize or reset the session */
2405                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2406                         section = "reset session";
2407                         status = nfs4_reset_session(clp);
2408                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2409                                 continue;
2410                         if (status < 0)
2411                                 goto out_error;
2412                 }
2413
2414                 /* Send BIND_CONN_TO_SESSION */
2415                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2416                                 &clp->cl_state)) {
2417                         section = "bind conn to session";
2418                         status = nfs4_bind_conn_to_session(clp);
2419                         if (status < 0)
2420                                 goto out_error;
2421                         continue;
2422                 }
2423
2424                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2425                         section = "check lease";
2426                         status = nfs4_check_lease(clp);
2427                         if (status < 0)
2428                                 goto out_error;
2429                         continue;
2430                 }
2431
2432                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2433                         section = "migration";
2434                         status = nfs4_handle_migration(clp);
2435                         if (status < 0)
2436                                 goto out_error;
2437                 }
2438
2439                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2440                         section = "lease moved";
2441                         status = nfs4_handle_lease_moved(clp);
2442                         if (status < 0)
2443                                 goto out_error;
2444                 }
2445
2446                 /* First recover reboot state... */
2447                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2448                         section = "reclaim reboot";
2449                         status = nfs4_do_reclaim(clp,
2450                                 clp->cl_mvops->reboot_recovery_ops);
2451                         if (status == -EAGAIN)
2452                                 continue;
2453                         if (status < 0)
2454                                 goto out_error;
2455                         nfs4_state_end_reclaim_reboot(clp);
2456                 }
2457
2458                 /* Detect expired delegations... */
2459                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2460                         section = "detect expired delegations";
2461                         nfs_reap_expired_delegations(clp);
2462                         continue;
2463                 }
2464
2465                 /* Now recover expired state... */
2466                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2467                         section = "reclaim nograce";
2468                         status = nfs4_do_reclaim(clp,
2469                                 clp->cl_mvops->nograce_recovery_ops);
2470                         if (status == -EAGAIN)
2471                                 continue;
2472                         if (status < 0)
2473                                 goto out_error;
2474                 }
2475
2476                 nfs4_end_drain_session(clp);
2477                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2478                         nfs_client_return_marked_delegations(clp);
2479                         continue;
2480                 }
2481
2482                 nfs4_clear_state_manager_bit(clp);
2483                 /* Did we race with an attempt to give us more work? */
2484                 if (clp->cl_state == 0)
2485                         break;
2486                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2487                         break;
2488         } while (atomic_read(&clp->cl_count) > 1);
2489         return;
2490 out_error:
2491         if (strlen(section))
2492                 section_sep = ": ";
2493         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2494                         " with error %d\n", section_sep, section,
2495                         clp->cl_hostname, -status);
2496         ssleep(1);
2497         nfs4_end_drain_session(clp);
2498         nfs4_clear_state_manager_bit(clp);
2499 }
2500
2501 static int nfs4_run_state_manager(void *ptr)
2502 {
2503         struct nfs_client *clp = ptr;
2504
2505         allow_signal(SIGKILL);
2506         nfs4_state_manager(clp);
2507         nfs_put_client(clp);
2508         module_put_and_exit(0);
2509         return 0;
2510 }
2511
2512 /*
2513  * Local variables:
2514  *  c-basic-offset: 8
2515  * End:
2516  */